Application of a Ground-based SAR System for - Sato Laboratory
Application of a Ground-based SAR System for - Sato Laboratory
Application of a Ground-based SAR System for - Sato Laboratory
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
<strong>Application</strong> <strong>of</strong> A <strong>Ground</strong>-Based Polarimetric<strong>SAR</strong> <strong>System</strong> <strong>for</strong> Environmental StudyA Dissertation PresentedbyZheng-Shu ZhoutoThe Graduate School <strong>of</strong> Engineeringin Partial Fulfillment <strong>of</strong> the Requirements <strong>for</strong> the Degree <strong>of</strong>Doctor <strong>of</strong> Engineeringin the Subject <strong>of</strong>Geoscience and TechnologySupervised byPr<strong>of</strong>. Motoyuki <strong>Sato</strong>Reviewed byPr<strong>of</strong>. Wolfgang-Martin BoernerTOHOKU UNIVERSITYAugust 2003
Abstractvarious scattering features <strong>of</strong> different components <strong>of</strong> trees, as well as during differentseasons.Finally, future perspectives on how to extend the system by implementation <strong>of</strong> recentadvances in polarimetric <strong>SAR</strong> are discussed.Key Words:Radar Polarimetry, <strong>Ground</strong>-<strong>based</strong> Synthetic Aperture Radar (GB-<strong>SAR</strong>), Radar Cross Section(RCS), Polarimetric Calibration, Diffraction Stacking, Short Time Fourier Trans<strong>for</strong>m (STFT),3-D Imaging, Tree Monitoring, Backscattering, Polarimetric Decomposition, EnvironmentalMonitoring.II
<strong>SAR</strong> 50MHz 20GHz 20m 1.5m III
Abstract - Japanese <strong>SAR</strong> IV
AcknowledgementsA special mention is reserved <strong>for</strong> Mr. Tadashi Hamasaki because <strong>of</strong> his friendly cooperationand assistance, especially in field experiments. I am grateful <strong>for</strong> my previous and presentcolleagues and friends in <strong>Sato</strong> <strong>Laboratory</strong>, such as Y. Nakashima, A. Yamamoto, M. Takeshita,K. Takasawa, Y. Komuro, T. Abe, K. Murakami, Lu Qi, N. Unumandakh, K. Takahashi, T.Koike, J. Zhao, H. Tajima, F. Mohammad, E. Igarashi, Y. Hamada, R. Tanaka, N. Ganchuluum,K. Iribe, K. Yoshida, K. Baker, J. McBride, K. Masuzawa, K. Watamura, and U. Ramdaras etal. They have made my studies an enjoyable and rewarding experience through these years.I would like to acknowledge ICF (International Communications Foundation sponsored byKDDI) <strong>for</strong> providing me research fellowship in the year <strong>of</strong> 2001.My parents and my elder aunt deserve very special thanks <strong>for</strong> their persistent support,understanding and encouragement throughout my life. In particular, I would like to thank twopersons that I love dearly, my wife Ping Shao and my daughter Duoduo, <strong>for</strong> theirnever-ending patience, understanding, love and support.My heartfelt thanks to all <strong>of</strong> you, again.Zheng-Shu ZhouAugust 08, 2003 in SendaiVI
CONTENTAbstractAcknowledgementsContentList <strong>of</strong> Figures, List <strong>of</strong> TablesChapter 1 Introduction 11.1 Recent Advancements <strong>of</strong> Polarimetric <strong>SAR</strong> Imaging ··························· 11.2 Current Studies <strong>of</strong> <strong>Ground</strong>-<strong>based</strong> <strong>SAR</strong> ··············································· 61.3 Research Objectives ··········································································· 81.4 Organization <strong>of</strong> the Dissertation ························································· 10Chapter 2 <strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> and Its Evaluation 132.1 <strong>System</strong> Principles ··············································································· 132.1.1 Remote Sensing and <strong>SAR</strong> Technology ·················································· 132.1.2 Radar Polarimetry ················································································· 172.1.3 Principle <strong>of</strong> <strong>Ground</strong>-<strong>based</strong> <strong>SAR</strong> <strong>System</strong> ··············································· 192.2 Simulation and Design <strong>of</strong> a <strong>Ground</strong>-<strong>based</strong> <strong>SAR</strong> <strong>System</strong> ··················· 202.2.1 Simulation <strong>of</strong> Azimuth Resolution ························································· 202.2.2 Design <strong>of</strong> a Broadband <strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> ········· 262.3 Polarimetric Per<strong>for</strong>mance Evaluation ················································· 312.3.1 Test Using a Metallic Sphere ································································· 312.3.2 Test Using Four Standard Reflectors ···················································· 362.4 Polarimetric Calibration ······································································ 382.4.1 RCS <strong>of</strong> Dihedral Corner Reflector ························································ 382.4.2 Polarimetric Calibration Using Dihedral Corner Reflectors ··················· 432.4.3 Calibration Results ················································································ 49VII
Content2.5 Summary ···························································································· 50Chapter 3 Test Experiments <strong>for</strong> Tree Monitoring 513.1 Description <strong>of</strong> Tree Scatterers ···························································· 513.2 Experimental Procedure and Data Collection ····································· 583.2.1 Measurements <strong>of</strong> Different Tree Type at Experimental Site A ··············· 583.2.2 Measurements <strong>of</strong> a Cherry Tree at Experimental Site B ······················· 613.3 Signal Verification ··············································································· 633.3.1 Signal Check <strong>for</strong> Four Standard Reflectors ·········································· 633.3.2 Signal Verification <strong>for</strong> Different Tree Types ··········································· 673.3.3 Signal Separated by Different Filters ···················································· 723.3.4 Signal Wave<strong>for</strong>m Comparison <strong>for</strong> a Cherry Tree ································ 773.4 Summary ···························································································· 80Chapter 4 Three Dimensional Image Reconstruction 814.1 Data Processing ················································································· 814.1.1 Signal Processing Flowchart ·································································· 814.1.2 Time Gating ·························································································· 824.1.3 Pulse Compression and Matched Filter ················································ 834.2 Image Reconstruction ········································································ 864.2.1 Algorithm <strong>of</strong> 3-D Imaging ······································································ 864.2.2 3-D Image Reconstruction ···································································· 874.3 Verification Using <strong>Ground</strong> Truths ························································ 974.3.1 Verification <strong>of</strong> Tree Different Trees ························································ 974.3.2 Verification <strong>of</strong> a Cherry Tree ································································ 1024.4 Summary ··························································································· 107Chapter 5 Data Interpretation with <strong>SAR</strong> Polarimetric Analysis 1095.1 Broadband Frequency Polarimetric Interpretation ····························· 1095.1.1 Enhancements <strong>of</strong> Amplitude ································································ 109VIII
Content5.1.2 Data Interpretation by Broadband Images············································· 1125.2 Single Frequency Polarimetric Interpretation ····································· 1175.2.1 STFT and Spatial Frequency Trans<strong>for</strong>mation········································ 1175.2.2 Polarimetric Target Descriptors ···························································· 1195.2.3 Single Frequency Scattering Matrices ················································· 1205.2.4 Polarimetric Power Density Images from Covariance Matrix ··············· 1225.2.5 Power Density Images from Pauli Covariance Matrix ·························· 1265.3 Entropy Based Polarimetric Target Decomposition ··························· 1295.3.1 Coherence Matrix ················································································ 1295.3.2 Polarimetric Entropy H and Angle Alpha ·············································· 1305.3.3 Decomposition <strong>of</strong> a Cherry Tree by the Eigenvector Based Method ··· 1325.4 Summary ··························································································· 139Chapter 6 Conclusions 141AppendicesA List <strong>of</strong> Acronyms ················································································ 145B RCS <strong>of</strong> Ideal Geometric Reflectors and Scattering Matrices··············· 147Bibliography 149Biography 159IX
LIST OF FIGURESFigure 1.01 Recent advances <strong>of</strong> polarimetric <strong>SAR</strong> imaging. ·············································· 5Figure 1.02 Research objectives <strong>of</strong> this dissertation. ························································· 9Figure 1.03 Structure <strong>of</strong> the dissertation. ·········································································· 10Figure 2.01 Microwave frequency bands used <strong>for</strong> imaging radar. ···································· 14Figure 2.02 Geometry <strong>of</strong> synthetic aperture radar. ··························································· 17Figure 2.03 Transmit / receive polarization signal succession in radar polarimetry.························································································································· 18Figure 2.04 Excitation pulse <strong>for</strong> simulation. ······································································ 22Figure 2.05 Simulated images <strong>of</strong> a point target at range <strong>of</strong> 10m: (a) horizontalslice with aperture length 5m, (b) horizontal slice with aperture length10m, (c) vertical slice with aperture length 5m and (d) vertical slicewith aperture length 10m. ··············································································· 23Figure 2.06 Simulated horizontal images <strong>of</strong> a point target at range 15m and20m : (a) with aperture 5m and range 15m, (b) with aperture 5m andrange 20m, (c) with aperture 10m and range 15m, (d) with aperture10m and range 20m, (e) with aperture 15m and range 15m, and (f)aperture 15m and range 20m. ········································································ 24Figure 2.07 Estimated resolutions by simulation. ····························································· 25Figure 2.08 <strong>System</strong> diagram <strong>of</strong> a broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong>. ··············· 27Figure 2.09 A dual polarized diagonal horn EMCO 3164-03: (a) front view, and(b) side view. ··································································································· 27Figure 2.10 Return loss and VSWR <strong>of</strong> a dual polarized diagonal horn: (a) Returnloss <strong>of</strong> EMCO 3164-03 #1043, (b) Return loss <strong>of</strong> EMCO 3164-03#1044, (c) VSWR <strong>of</strong> EMCO 3164-03 #1043 and (d) VSWR <strong>of</strong> EMCO3164-03 #1044. ······························································································ 28Figure 2.11 Radiation patterns <strong>of</strong> a dual polarized diagonal horn: (a) 1 GHz, (b) 2GHz, (c) 3 GHz, (d) 4 GHz and (e) 5 GHz. ····················································· 29Figure 2.12 A developed broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong> system. ················ 30Figure 2.13 Testing scenario <strong>of</strong> a metallic sphere. ··························································· 32X
List <strong>of</strong> FiguresFigure 2.14 Geometry <strong>of</strong> creeping wave <strong>for</strong> a sphere. ····················································· 33Figure 2.15 Reflection wave<strong>for</strong>ms from the metallic sphere. ············································ 34Figure 2.16 Theoretical normalized RCS curve and measured values <strong>of</strong> ametallic sphere: black solid line- theoretical curve, red dotted line -HH polarization, blue dashed line - VV polarization and greendash-dot line - VH polarization. ······································································ 35Figure 2.17 Four standard reflectors setups. ···································································· 36Figure 2.18 Reflection wave<strong>for</strong>ms <strong>of</strong> a dihedral corner reflector. ····································· 38Figure 2.19 Geometry <strong>of</strong> a dihedral corner reflector. ························································ 40Figure 2.20 Calculated RCS patterns <strong>of</strong> a dihedral corner reflector with sidelength 0.3 m: (a) 1 GHz, (b) 2 GHz, (c) 3 GHz, (d) 4 GHz and (e) 5GHz. ··············································································································· 42Figure 2.21 Backscattering model. ··················································································· 44Figure 2.22 Two calibrators: (a) vertical dihedral, (b) 45 o dihedral. ··································· 47Figure 2.23 Calibration coefficients <strong>of</strong> two cross-polarization terms. ································ 48Figure 3.01 Coordinate system <strong>of</strong> a measurement site: Exp#A. ······································ 52Figure 3.02 Leaves: (a) broad leaves <strong>of</strong> the tree T1: Japanese Zelkova, (b)needles <strong>of</strong> the tree T2: Japanese cedar and (c) leaves <strong>of</strong> the shorttree: Azalea and a plant: Japanese honeysuckle. ·········································· 53Figure 3.03 Target area in the first measurement: Exp#A-1. ············································ 53Figure 3.04 Target area in the second measurement: Exp#A-2. ······································ 54Figure 3.05 Target area in the third measurement: Exp#A-3. ··········································· 54Figure 3.06 Coordinate system <strong>for</strong> measurement site Exp#B, a cherry tree. ··················· 55Figure 3.07 Flowers and leaves <strong>of</strong> a Yoshino cherry tree. ················································ 56Figure 3.08 A Yoshino cherry tree in the first measurement: Exp#B-1. ···························· 56Figure 3.09 The cherry tree in the second measurement: Exp#B-2. ································ 57Figure 3.10 The cherry tree in the third measurement: Exp#B-3. ····································· 57Figure 3.11 Setup <strong>of</strong> four standard reflectors. ·································································· 64Figure 3.12 Reflected signals <strong>of</strong> four standard reflectors: (a) vertical dihedral, (b)45 o dihedral, (c) vertical wire, and (d) -45 o wire. ············································· 65Figure 3.13 Windowing functions. ···················································································· 66XI
List <strong>of</strong> FiguresFigure 3.14 Time domain signals in wiggle pr<strong>of</strong>iles: (a) broadband pass filteredand (b) low pass filtered. ················································································ 66Figure 3.15 Migrated images: (a) broadband pass filtered and (b) low passfiltered. ············································································································ 67Figure 3.16 Three observation points A(-3, 2), B(-0.3, 2) and C(4.5, 2). ·························· 68Figure 3.17 Wave<strong>for</strong>ms <strong>of</strong> point A -T1: (a) HH, (b) VH, and (c) VV. ·································· 69Figure 3.18 Wave<strong>for</strong>ms <strong>of</strong> point B -T2: (a) HH, (b) VH, and (c) VV. ································· 70Figure 3.19 Wave<strong>for</strong>ms <strong>of</strong> point C -T3: (a) HH, (b) VH, and (c) VV. ································· 71Figure 3.20 Four filters: (a) broadband filter, (b) low-pass, middle-pass andhigh-pass filters. ····························································································· 73Figure 3.21 HH signals filtered by different filters: (a) Exp#A-1, (b) Exp#A-2, and(c) Exp#A-3. ···································································································· 74Figure 3.22 VH signals filtered by different filters: (a) Exp#A-1, (b) Exp#A-2, and(c) Exp#A-3. ···································································································· 75Figure 3.23 VV signals filtered by different filters: (a) Exp#A-1, (b) Exp#A-2, and(c) Exp#A-3. ···································································································· 76Figure 3.24 Two observation points: M(-0.6, 2)m and N(2,1.6)m. ···································· 77Figure 3.25 Wave<strong>for</strong>ms <strong>of</strong> the cherry tree at point M (-0.6, 2) m: (a) HH, (b) VH,and (c) VV. ······································································································ 78Figure 3.26 Wave<strong>for</strong>ms <strong>of</strong> the cherry tree at point N (2, 1.6) m: (a) HH, (b) VH,and (c) VV. ······································································································ 79Figure 4.01 Signal processing flowchart. ·········································································· 82Figure 4.02 Algorithm <strong>of</strong> matched filtering. ······································································· 84Figure 4.03 Reference reflector <strong>of</strong> an aluminum plate: 1m x 2m. ····································· 85Figure 4.04 Pulse compression: (a) reference signal, (b) raw signal, and (c)compressed signal. ························································································· 85Figure 4.05 Geometry <strong>of</strong> image reconstruction by diffraction stacking withantenna directivity compensation. ·································································· 87Figure 4.06 Shape <strong>of</strong> a band-pass filter. ··········································································· 88Figure 4.07 3-D polarimetric images <strong>of</strong> trees in spring - Exp#A-1: (a) HH, (b) VH,and (c) VV. ······································································································ 90Figure 4.08 3-D polarimetric images <strong>of</strong> trees in summer - Exp#A-2: (a) HH, (b)VH, and (c) VV. ······························································································· 91XII
List <strong>of</strong> FiguresFigure 4.09 3-D polarimetric images <strong>of</strong> trees in autumn - Exp#A-3: (a) HH, (b) VH,and (c) VV. ······································································································ 92Figure 4.10 3-D images <strong>of</strong> a cherry tree in Exp#B-1: (a) HH, (b) VH, and (c) VV ············ 94Figure 4.11 3-D images <strong>of</strong> a cherry tree in Exp#B-2: (a) HH, (b) VH, and (c) VV. ············ 95Figure 4.12 3-D images <strong>of</strong> a cherry tree in Exp#B-3: (a) HH, (b) VH, and (c) VV ············ 96Figure 4.13 Experimental scenes <strong>of</strong> trees in different seasons: (a) Exp#A-1:spring, (b) Exp#A-2: summer, and (c) Exp#A-3: autumn. ······························· 98Figure 4.14 3-D HH images <strong>of</strong> trees in: (a) Exp#A-1: spring, (b) Exp#A-2:summer, and (c) Exp#A-3: autumn. ································································ 99Figure 4.15 3-D VH images <strong>of</strong> trees in: (a) Exp#A-1: spring, (b) Exp#A-2:summer, and (c) Exp#A-3: autumn. ······························································· 100Figure 4.16 3-D VV images <strong>of</strong> trees in: (a) Exp#A-1: spring, (b) Exp#A-2:summer, and (c) Exp#A-3: autumn. ································································ 101Figure 4.17 Experimental scenes <strong>of</strong> a cherry tree at different states: (a) Exp#B-1:buds, (b) Exp#B-2: flowers, and (c) Exp#B-3: leaves. ···································· 103Figure 4.18 3-D HH images <strong>of</strong> a cherry tree in: (a) Exp#B-1: buds, (b) Exp#B-2:flowers, and (c) Exp#B-3: leaves. ··································································· 104Figure 4.19 3-D VH images <strong>of</strong> a cherry tree in: (a) Exp#B-1: buds, (b) Exp#B-2:flowers, and (c) Exp#B-3: leaves. ··································································· 105Figure 4.20 3-D VV images <strong>of</strong> a cherry tree in: (a) Exp#B-1: buds, (b) Exp#B-2:flowers, and (c) Exp#B-3: leaves. ··································································· 106Figure 5.01 Functions <strong>for</strong> amplitude enhancement. ························································· 110Figure 5.02 Polarimetric color images (red-HH, green-VH, blue-VV) enhancedby: (a) original, (b) sin(x*pi/2), (c) sin(x 0.5 *pi/2) and (d) sin(x 0.3 *pi/2). ············· 111Figure 5.03 Corresponding areas <strong>of</strong> tress in: (a) spring, (b) summer, and (c)autumn. ··········································································································· 113Figure 5.04 Broadband vertical pr<strong>of</strong>iles <strong>of</strong> T1 at range <strong>of</strong> 12.9m red-HHgreen-VH blue-VV in: (a) spring, (b) summer, and (c) autumn. ······················ 114Figure 5.05 Broadband vertical pr<strong>of</strong>iles <strong>of</strong> T2 at range <strong>of</strong> 9.8m red-HH green-VHblue-VV in: (a) spring, (b) summer, and (c) autumn. ······································ 115Figure 5.06 Broadband vertical pr<strong>of</strong>iles <strong>of</strong> T3 at range <strong>of</strong> 8.3m red-HH green-VHblue-VV in: (a) spring, (b) summer, and (c) autumn. ······································ 116Figure 5.07 A time domain signal and the response spectrum distribution aftershort time Fourier trans<strong>for</strong>m. ·········································································· 118XIII
List <strong>of</strong> FiguresFigure 5.08 Different target polarimetric descriptors. ························································ 120Figure 5.09 Single-frequency power density images <strong>of</strong> cherry blossom andleaves red-|HH| 2 green-2|VH| 2 blue-|VV| 2 : (a) imaging area <strong>of</strong> flowers,(b) 1.5 GHz, (c) 2.5 GHz, (d) 3.5 GHz, (e) 4.5 GHz images <strong>of</strong> flowers,and (f) imaging area <strong>of</strong> leaves, (g) 1.5 GHz, (h) 2.5 GHz, (i) 3.5 GHz,(j) 4.5 GHz images <strong>of</strong> leaves. ········································································· 125Figure 5.10 Polarimetric power density images <strong>of</strong> trees in spring Exp#A-1 atheight <strong>of</strong> 2.5 m <strong>based</strong> on Equations (5.23)-(5.25) red-|HH+VV| 2green-2|VH| 2 blue-|HH-VV| 2 : (a) 1.5 GHz, (b) 2.5 GHz, (c) 3.5 GHzand (d) 4.5 GHz. ····························································································· 128Figure 5.11 Entropy distribution images <strong>of</strong> a cherry tree: (a) 1.5 GHz <strong>of</strong> Exp#B-1,(b) 4.5 GHz <strong>of</strong> Exp#B-1, (c) 1.5 GHz <strong>of</strong> Exp#B-2, (d) 4.5 GHz <strong>of</strong>Exp#B-2, (e) 1.5 GHz <strong>of</strong> Exp#B-3, and (f) 4.5 GHz <strong>of</strong> Exp#B-3. ···················· 133Figure 5.12 alpha distribution images <strong>of</strong> a cherry tree: (a) 1.5 GHz <strong>of</strong> Exp#B-1,(b) 4.5 GHz <strong>of</strong> Exp#B-1, (c) 1.5 GHz <strong>of</strong> Exp#B-2, (d) 4.5 GHz <strong>of</strong>Exp#B-2, (e) 1.5 GHz <strong>of</strong> Exp#B-3, and (f) 4.5 GHz <strong>of</strong> Exp#B-3. ···················· 134Figure 5.13 Two indicated analysis areas, blue square-buds/flowers/leaves, redbox-trunk: (a) Exp#B-1, (b) Exp#B-1, and (c) Exp#B-3. ································· 135Figure 5.14 Entropy-alpha distributions <strong>of</strong> a part <strong>of</strong> a cherry tree- the blue squareindicated in Figure 5.13: (a) 1.5 GHz <strong>of</strong> Exp#B-1, (b) 4.5 GHz <strong>of</strong>Exp#B-1, (c) 1.5 GHz <strong>of</strong> Exp#B-2, (d) 4.5 GHz <strong>of</strong> Exp#B-2, (e) 1.5GHz <strong>of</strong> Exp#B-3, and (f) 4.5 GHz <strong>of</strong> Exp#B-3. ··············································· 136Figure 5.15 Changes <strong>of</strong> mean value: (a) Entropy and (b) alpha in the twoanalysis areas. ································································································ 137XIV
LIST OF TABLESTable 1.01 Current studies on ground-<strong>based</strong> <strong>SAR</strong> ····························································· 8Table 2.01 Parameters used <strong>for</strong> simulation ······································································ 22Table 2.02 Specifications <strong>of</strong> a developed broadband ground-<strong>based</strong> polarimetric<strong>SAR</strong> system ···································································································· 31Table 2.03 Comparison <strong>of</strong> calibrated scattering matrices with measured scatteringmatrices ·········································································································· 49Table 2.04 Auto-calibrated scattering matrices. ································································ 50Table 3.01 Condition and target situations <strong>for</strong> measurements <strong>of</strong> different tree types························································································································· 58Table 3.02 Parameters <strong>for</strong> measurements <strong>of</strong> different tree types ····································· 59Table 3.03 Data collection <strong>of</strong> measurements <strong>of</strong> different tree types ································· 60Table 3.04 Condition and target situations <strong>for</strong> measurements <strong>of</strong> a cherry tree ················ 61Table 3.05 Parameters <strong>for</strong> measurements <strong>of</strong> a Yoshino cherry tree ································· 62Table 3.06 Data <strong>for</strong>mation <strong>of</strong> measurements <strong>of</strong> a Yoshino cherry tree ····························· 63Table 5.01 Schematic representation <strong>of</strong> the entropy H interpretation ······························· 131Table 5.02 Schematic representation <strong>of</strong> the alpha-angle interpretation ···························· 132Table 5.03 Comparison <strong>of</strong> mean value <strong>of</strong> Entropy and alpha in two analysis areas:red face fonts <strong>for</strong> trunk, blue face fonts <strong>for</strong> buds/flower/leaves ······················ 137Table B.01 Radar cross section <strong>of</strong> some ideal geometric scatterers ································ 147Table B.02 Scattering matrices <strong>of</strong> orientation independent scatterers ······························ 148Table B.03 Scattering matrices <strong>of</strong> scatterers with orientated dependence ······················· 148XV
Chapter 1INTRODUCTION1.1 Recent Advancements <strong>of</strong> Polarimetric <strong>SAR</strong> ImagingPolarimetry deals with the full vector nature <strong>of</strong> polarized electromagnetic waves throughoutthe frequency spectrum from ultra-low-frequencies (ULF) to above the far-ultra-violet (FUV)[1, 2, 3]. Where there are abrupt or gradual changes in the index <strong>of</strong> refraction (or permittivity,magnetic permeability, and conductivity), the polarization state <strong>of</strong> a narrow-band(single-frequency) wave is trans<strong>for</strong>med, and the electromagnetic vector wave is re-polarized.When the wave passes through a medium <strong>of</strong> changing index <strong>of</strong> refraction or when it strikes anobject such as a radar target and/or a scattering surface and it is reflected; then, characteristicin<strong>for</strong>mation about the reflectivity, shape and orientation <strong>of</strong> the reflecting body can be obtainedby implementing “polarization control” [1, 3, 4, 5]. The time-dependent behavior <strong>of</strong> theelectric field vector, in general describing an ellipse in a plane transverse to propagation, playsan essential role in the interaction <strong>of</strong> electromagnetic waves with material bodies, and thepropagation medium [5, 6, 7, 8]. Whereas, this polarization trans<strong>for</strong>mation behavior isdenoted as polarimetry in radar, and synthetic aperture radar (<strong>SAR</strong>) sensing and imaging [1,3, 11, 12] - using the ancient Greek meaning <strong>of</strong> “measuring orientation and object shape”.Thus, polarimetry is concerned with the control <strong>of</strong> the coherent polarization properties <strong>of</strong> theoptical wave and microwaves [1, 3, 4, 12].Very decisive progress was made in advancing fundamental polarimetric interferometric <strong>SAR</strong>and algorithm developments during the past decade [9, 10-16], which was <strong>based</strong> on theunderlying accomplishments <strong>of</strong> fully polarimetric <strong>SAR</strong> (POL<strong>SAR</strong>) [2, 11, 14] and differential<strong>SAR</strong> interferometry [4, 11, 14] and its current merger gradually [14, 15, 16]. This wasaccomplished with the aid <strong>of</strong> airborne & shuttle plat<strong>for</strong>ms supporting single-to-multi-band,multi-modal polarimetric <strong>SAR</strong> and also some polarimetric interferometric <strong>SAR</strong>(POL-IN-<strong>SAR</strong>) sensor systems, which will be compared and assessed with the aim <strong>of</strong>establishing the now not completed but required missions such as polarimetric tomographicand holographic imaging. There are hitherto the following tendencies <strong>of</strong> development <strong>of</strong>1
Chapter 1polarimetric <strong>SAR</strong> imaging.Extension <strong>of</strong> Time DomainFor airborne or space-borne remote sensing <strong>of</strong> earth’s surface, the all-weather day- andnight-time capability <strong>of</strong> electromagnetic radiation at microwave frequencies has attractedproponents over a number <strong>of</strong> years. Recently, the research activity in the microwave area hasaccelerated considerably, as evidenced by the number <strong>of</strong> papers and even special issuesdevoted to the topic [9-18].Expansion <strong>of</strong> Frequency DomainBoth optical [7, 12] and radar [1, 3, 12] imaging have matured considerably, and the benefits<strong>of</strong> using one imaging modality over the other are discussed frequently [4, 14]. For example,hyper-spectral optical radiometric imaging in the range <strong>of</strong> FIR-VIS-FUV [15] is considered tobecome the exclusive remote sensing system <strong>of</strong> the twenty-first century, and thought to besuperior to ultra-wide-band (UWB) microwave <strong>SAR</strong> imaging in the range <strong>of</strong>HF-UHF-SHF-EHF [15]. Even, it was argued that UWB <strong>SAR</strong> imaging is superfluous andcould be scrapped altogether because <strong>of</strong> the exorbitant costs in developing this abstract rather‘invisible’ remote sensing technology [11, 12, 14]. In either case, the inherent electromagneticvector wave interaction processes are subjected to Maxwell’s equations; and constrained bythe carrier frequency and bandwidth, the amplitude, phase and polarization [5, 6, 13].Especially those depend on the dispersive and polarization-dependent material constituents <strong>of</strong>the propagation medium as well as <strong>of</strong> the illuminated scattering surface, its geometry andstructure, and its voluminous vegetative over-burden as well as its composite geologicalunder-burden. However, in order to identify parameters describing voluminous scatteringscenarios beyond the skin depth <strong>of</strong> the vegetation canopy, the entire amenable air/space-bornefrequency regime from MF (100 KHz) to FUV (10 PHz) needs to be implemented [14, 15] inremote sensing. This implies that we require both radar and optical imaging together withfull scattering matrix acquisition capabilities - in order to recover fully the intricate scatteringmechanisms [5, 8, 21] and bio-mass assessment tasks.There<strong>for</strong>e, every possible ef<strong>for</strong>t should be made to expand and to extend but not to give up theexisting, insufficient availability <strong>of</strong> free scientific remote sensing spectral windows, whichmust absolutely be spread with deca-logarithmic periodicity throughout the pertinent2
Introductionfrequency bands <strong>of</strong> about 1 MHz to 300 GHz.Expansion <strong>of</strong> Spatial RegionThe development <strong>of</strong> radar polarimetry is advancing rapidly. Whereas with radar polarimetrythe textural fine-structure, target orientation, symmetries and material constituents can berecovered with considerable improvement above that <strong>of</strong> standard “amplitude-only” radar;with radar interferometry the spatial (in depth) structure can be explored. Because theoperation <strong>of</strong> airborne test-beds is extremely expensive, aircraft plat<strong>for</strong>ms are not suited <strong>for</strong>routine monitoring missions, those are better accomplished with the use <strong>of</strong> drones (UAV),which can in addition be operated at much greater heights [4, 14]. Such unmanned aerialvehicles (drones) were hitherto developed <strong>for</strong> defense applications, however currently lackingthe sophistication <strong>for</strong> implementing advanced <strong>for</strong>efront polarimetric interferometric <strong>SAR</strong>technology. This shortcoming will be thoroughly scrutinized resulting in the finding that wedo now need to develop most rapidly also polarimetric interferometric <strong>SAR</strong> drone-plat<strong>for</strong>mtechnology especially <strong>for</strong> environmental stress-change monitoring subject to severeoperational constraints due to adverse unsafe flight conditions with a great variance <strong>of</strong>applications beginning with flood, bush/<strong>for</strong>est-fire to tectonic-stress (earth-quake to volcaniceruptions) <strong>for</strong> real-short-time hazard mitigation. However, <strong>for</strong> routine global monitoringpurposes <strong>of</strong> the terrestrial covers neither airborne sensor implementation - aircraft and/ordrones - are sufficient; and there-<strong>for</strong>e multi-modal and multi-band space-borne polarimetricinterferometric <strong>SAR</strong> space-shuttle and satellite sensor technology needs to be furtheradvanced at a much more rapid pace. The existing ENVISAT with the <strong>for</strong>thcomingALOS-PAL<strong>SAR</strong>, RADARSAT-2 and the TERRA<strong>SAR</strong> will be compared in [14],demonstrating that at this phase <strong>of</strong> development the fully polarimetric andpolarimetric-interferometric <strong>SAR</strong> modes <strong>of</strong> operation must be treated as preliminaryalgorithm verification support, and at this phase <strong>of</strong> development are still not to be viewed asroutine modes. The same considerations apply to the near future implementation <strong>of</strong> anysatellite-cluster bi/multi-static space-borne tomographic imaging modes as described in [23],which must however be developed concurrently in collaboration <strong>of</strong> all major national or jointcontinental ef<strong>for</strong>ts in order to reduce proliferation <strong>of</strong> space-plat<strong>for</strong>ms and <strong>for</strong> cost-cuttingreasons. Prioritization <strong>of</strong> developmental stages need to be assessed according to applications,and will have to differ <strong>for</strong> air-borne to space-borne sensors with the aim <strong>of</strong> developing apermanently orbiting fleet <strong>of</strong> equidistantly space-distributed satellites – similar to the GPSconfiguration, however each equipped with the identical set <strong>of</strong> multi-band polarimetric <strong>SAR</strong>3
Chapter 1sensors as proposed in [4].Progress <strong>of</strong> TechniquesThere is currently widespread interest in the development <strong>of</strong> radar sensors <strong>for</strong> the detection <strong>of</strong>surface and buried targets and the remote sensing <strong>of</strong> land, sea and ice surfaces.Inpolarimetric interferometric <strong>SAR</strong> imaging, it is possible to recover such co-registered texturaland spatial in<strong>for</strong>mation from POL-IN-<strong>SAR</strong> digital image data sets simultaneously, includingthe extraction <strong>of</strong> digital elevation maps (DEM) from either polarimetric (scattering matrix) orinterferometric (single plat<strong>for</strong>m: dual antenna) <strong>SAR</strong> systems [19, 20]. SimultaneousPOL-IN-<strong>SAR</strong> <strong>of</strong>fers the additional benefit <strong>of</strong> obtaining co-registered textural plus spatialthree-dimensional POL-IN-DEM in<strong>for</strong>mation, which when applied to repeat-passimage-overlay interferometry provides differential background validation, stress assessmentand environmental stress-change in<strong>for</strong>mation with high accuracy. Then, by either designingmultiple dual-polarization antenna POL-IN-<strong>SAR</strong> systems or by applying advancedPOL-IN-<strong>SAR</strong> image compression techniques will result in polarimetric tomographic(multi-interferometric) <strong>SAR</strong> (POL-TOMO-<strong>SAR</strong>) imaging as was shown by Reigber [23, 24].This is <strong>of</strong> direct relevance to wide-area, dynamic battle-space surveillance and local to globalenvironmental background validation, stress assessment and stress-change monitoring <strong>of</strong> theterrestrial and planetary covers. While several airborne systems can now provide diversityover all three <strong>of</strong> these, it is the combination <strong>of</strong> polarimetry with interferometry at a singlewavelength that <strong>for</strong>ms the central focus <strong>of</strong> future challenges in developing new and originaldata processing. The main reason <strong>for</strong> this is the imminent launch <strong>of</strong> a series <strong>of</strong> advancedsatellite radar systems such as PAL<strong>SAR</strong>, an L-band <strong>SAR</strong> sensor on board the NASDA ALOSsatellite (in September <strong>of</strong> 2004) and RADARSAT II [4], a C-band polarimetric sensor (inspring <strong>of</strong> 2005). These are typical innovations <strong>of</strong> a new generation <strong>of</strong> radars with the potential<strong>for</strong> providing data from various combinations <strong>of</strong> polarimetry and interferometry.Advance Data ProcessingAn important feature <strong>of</strong> electromagnetic radiation is its state <strong>of</strong> polarization and a wide range<strong>of</strong> classification algorithms and inversion techniques have recently been developed <strong>based</strong> onthe trans<strong>for</strong>mation <strong>of</strong> polarization state by scattering objects [9, 10, 16]. There are threeprimary ways in which multi-parameter radar measurements can be made: multi-frequency,single or multi-baseline interferometry and multi-polarization. Recent progress in polarimetric4
Introductionand interferometric <strong>SAR</strong> data processing covers advances in classification <strong>of</strong> polarimetric<strong>SAR</strong> data and it is addressing the important topic <strong>of</strong> quantitative data inversion <strong>of</strong> radar databy considering applications in <strong>for</strong>est height mapping using polarimetric interferometry <strong>SAR</strong>data processing [20 ,21].There<strong>for</strong>e, the combination <strong>of</strong> spatial in<strong>for</strong>mation, spectral in<strong>for</strong>mation and polarizationin<strong>for</strong>mation with multi-frequency and broadband sensors defines the recent advancesaccomplished in polarimetric <strong>SAR</strong> systems development. The relation can be described inFigure 1.01. <strong>SAR</strong> is a well-known technique used <strong>for</strong> airborne or space borne remote sensing.It can usefully be exploited in ground-<strong>based</strong> radar imaging, we call it ground-<strong>based</strong> <strong>SAR</strong>. It isa new application <strong>of</strong> the conventional <strong>SAR</strong> expansion in the spatial domain.Spatial In<strong>for</strong>mationRadar Imaging <strong>System</strong>sImaging RadarImaging RadarSpectrometersSpectral In<strong>for</strong>mationRadar SpectrometersMulti-FrequencyImaging RadarPolarimetersPolarimetersMulti-FrequencyRadar PolarimetersPolarization In<strong>for</strong>mationRadar PolarimetersFigure 1.01 Recent advances <strong>of</strong> polarimetric <strong>SAR</strong> imaging (according to van Zyl).By combining polarimetric and interferometric <strong>SAR</strong> techniques, it is today possible to recoversuch co-registered textural plus spatial properties simultaneously. This has importantconsequences <strong>for</strong> the design <strong>of</strong> future space and airborne sensors <strong>for</strong> carbon sequestration andclimate change studies.It has now been shown that the accelerated advancement <strong>of</strong> POL<strong>SAR</strong> techniques is <strong>of</strong> directrelevance and <strong>of</strong> utmost priority to local-to-global environmental ground-truth measurementand validation, stress assessment, and stress-change monitoring <strong>of</strong> the terrestrial and planetarycovers [3, 7, 12]. POL<strong>SAR</strong> remote sensing <strong>of</strong>fers an efficient and reliable means <strong>of</strong> collecting5
Chapter 1the in<strong>for</strong>mation required in order to extract the biophysical and geophysical parameters aboutthe Earth’s surface; and it has found successful application in crop monitoring and damageassessment, in <strong>for</strong>estry clear cut mapping, de<strong>for</strong>estation and burn mapping, in land surfacestructure (geology) land cover (biomass) and land use, in hydrology (soil moisture, flooddelineation), in sea ice monitoring, in oceans and coastal monitoring (oil spill detection), etc[10, 17, 22].Today, it can be said that there is more and more a rapidly increasing interest in the use <strong>of</strong>radar polarimetry and interferometry <strong>for</strong> radar remote sensing; and wave polarizationutilization is today <strong>of</strong> fundamental importance in the in<strong>for</strong>mation retrieval problem <strong>of</strong>microwave imaging and vector (polarization) inverse scattering.1.2 Current Studies <strong>of</strong> <strong>Ground</strong>-<strong>based</strong> <strong>SAR</strong>Synthetic aperture radar is usually used <strong>for</strong> airborne or space borne remote sensing. <strong>SAR</strong> canalso advantageously be exploited in a ground-<strong>based</strong> radar imaging system. We call it<strong>Ground</strong>-<strong>based</strong> <strong>SAR</strong>.A few design options [25, 33, 35, 41] <strong>of</strong> <strong>Ground</strong>-<strong>based</strong> <strong>SAR</strong> systems were proposed asmonitoring tools <strong>for</strong> agriculture, terrain mapping, a variety <strong>of</strong> large man-made structures, aswell as environmental studies and ground surface de<strong>for</strong>mation detection, includingde<strong>for</strong>mation maps <strong>of</strong> static displacements <strong>of</strong> a variety <strong>of</strong> structures like concrete girders in acontrolled environment, building models, dams, and bridges, and so on.A ground-<strong>based</strong> microwave operation [25] was developed <strong>for</strong> the Canadian Center <strong>of</strong> RemoteSensing by the University <strong>of</strong> Saskatchewan. It is a three-band, ground-<strong>based</strong> scatterometersystem. Part <strong>of</strong> this activity had been motivated by the synthetic aperture microwave satelliteRADARSAT-I. Hence, they acquired a three-band, ground-<strong>based</strong> scatterometer system, whichwas operated under contract by an interdisciplinary <strong>Ground</strong> Microwave Operations (GMO)team at the University <strong>of</strong> Saskatchewan including the Departments <strong>of</strong> Agriculture and SpacePhysics. The research was directed toward a general understanding <strong>of</strong> the microwaveinteraction with surface targets, with particular emphasis on crops and soils, in keeping withRADARSAT-I objectives. In the earlier 1990’s, they had achieved a few <strong>of</strong> applications usingthis system [26-32]. Of paramount importance is the well documented result <strong>of</strong> the radar6
Introductioncross section (RCS) <strong>of</strong> plants on the diurnal soil-root-plant fluid exchange cycle, whichdeserves subtle additional analyses <strong>for</strong> other climatologic regions hitherto not executed.Pr<strong>of</strong>essor F. T. Ulaby et al. developed a bistatic measurement facility (BMF) bistatic-radarfacility <strong>for</strong> detecting a target obscured by foliage [33, 34] first at Kansas University and thenat the University <strong>of</strong> Michigan. A fully polarimetric bistatic-radar facility was constructed atthe University <strong>of</strong> Michigan, to serve as a research tool <strong>for</strong> improved understanding <strong>of</strong> thenature <strong>of</strong> bistatic scattering <strong>for</strong> point and distributed targets. The facility was capable <strong>of</strong>operation at 10, 35, and 94 GHz. To meet both the size and design constraints both a hornantenna operating in the far-field, and a parabolic-dish antenna operating in a near-fieldfocused mode, are utilized. A newly developed bistatic-calibration technique, using a flatmetal plate, was used to calibrate the facility. Validation results, using a hemisphere over aconducting metal plate, show that the facility is capable <strong>of</strong> characterizing the radar crosssection <strong>of</strong> a point target to within ±1 dB in magnitude and ±5° in polarization phase differenceover a wide range <strong>of</strong> bistatic angles. Sample data <strong>for</strong> a point target and a distributed target arepresented.M. Pieraccini et al. at University <strong>of</strong> Florence Italy, proposed a ground-<strong>based</strong> interferometric<strong>SAR</strong> technique <strong>for</strong> terrain mapping [36]. It was <strong>based</strong> on a coherent continuous-wavestepped-frequency (CW-SF) radar moved along a linear horizontal rail. It works bysynthesizing microwave holographic images taken from different view angles to obtainelevation maps by phase comparison. The focusing algorithm <strong>for</strong> imaging syntheticholograms and digital elevation models was able to correct topographic distortion and phasewrap through an iterative multi-baseline procedure. Then they also proposed an innovativesurvey radar technique <strong>based</strong> on microwave holographic images <strong>for</strong> dynamic testing <strong>of</strong> largestructures taking care <strong>of</strong> both vibration amplitude pattern and frequency. Theoreticalbackground was provided, verified and experimental results obtained during a dynamic teston concrete and masonry buildings were achieved [35, 37, 38].D. Tarchi et al. at Joint Research Centre, Space <strong>Application</strong> Institute (JRC-SAI) described aninnovative application <strong>of</strong> radar interferometric techniques aimed to monitor structuralde<strong>for</strong>mations <strong>of</strong> buildings <strong>for</strong> the cultural heritage survey [40]. The proposed application was<strong>based</strong> on the use <strong>of</strong> ground-<strong>based</strong> instrumentation able to operate as interferometric <strong>SAR</strong>.Preliminary experimental results on a model <strong>of</strong> an historical building encourage thedevelopment <strong>of</strong> this technique <strong>for</strong> use in architectural heritage surveys.7
Chapter 1Table 1.01 Current studies <strong>of</strong> ground-<strong>based</strong> <strong>SAR</strong>GroupF. Ulaby et al.M. Pieraccini et al.D. Tarchi et al.Michigan Univ.Florence Univ./JRCJRC SAI<strong>System</strong> BMF&moblie backscatter system <strong>Ground</strong>-<strong>based</strong> <strong>SAR</strong> <strong>SAR</strong> systemType Polarimetry Interferometry InterferometryFrequency 35 GHz 5.7 GHz / 4.5 GHz 10-18 GHzBandwidth0.6 GHz / 1 GHzMovement Spherical scan & 1D scan 1D scan/+spherical scan 1D scanCondition Indoor & outdoor Outdoor Indoor<strong>Application</strong>Detecting a target obscured byTerrain mapping &Structural changes <strong>of</strong>foliagebuilding monitoringartifical targetMost <strong>of</strong> these applications [35-44] <strong>of</strong> ground-<strong>based</strong> <strong>SAR</strong> were <strong>based</strong> on interferometrictechniques and/or were developed only <strong>for</strong> narrow bandwidths summarized in Table 1.01.1.3 Research ObjectivesSynthetic aperture radar using airborne and space borne plat<strong>for</strong>ms has attracted interest <strong>for</strong>many years. Polarimetric <strong>SAR</strong> interferometry is a microwave imaging technique <strong>for</strong>extracting geophysical and volumetric vegetation parameters from <strong>SAR</strong> images, and itsusefulness in terrain classification and surface change detection has been demonstrated [12].Typical modern methods <strong>of</strong> <strong>SAR</strong> polarimetry are radar target decomposition and classification,topographic mapping, and more recently, monitoring <strong>of</strong> ground displacements usingdifferential interferometric <strong>SAR</strong> concepts [2, 22]. <strong>SAR</strong> polarimetry can also be exploited inground-<strong>based</strong> radar imaging systems. The precise monitoring <strong>of</strong> the de<strong>for</strong>mation <strong>of</strong> groundsurfaces is important in many applications. For instance, the ability to monitor volcanoactivity by observing the surface bulging be<strong>for</strong>e the eruption would be advantageous <strong>for</strong>disaster prevention. For these and other applications, ground-<strong>based</strong> broadband <strong>SAR</strong> <strong>of</strong>fers a8
Introductionparticularly useful technique. Moreover, it can be applied to detect the subsurface structureand surface movements under cloudy and/or dusty conditions, or even inside a building,situations in which other optical imaging and ranging techniques, such as laser range findersand GPS, cannot be used as was assessed in [14, 17].There<strong>for</strong>e, we planed to develop a broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong> system <strong>for</strong>monitoring <strong>of</strong> different environmental features, <strong>for</strong> example, different types <strong>of</strong> vegetationsuch as trees. For application <strong>of</strong> this broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong> system, thefollowing research objectives had to be met:(a)(b)(c)(d)(e)(f)Development <strong>of</strong> a broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong> system <strong>based</strong> oncomputer simulation results;<strong>System</strong> evaluation with polarimetric calibration;Testing experiments <strong>for</strong> trees in natural condition;<strong>Ground</strong>-<strong>based</strong> polarimetric <strong>SAR</strong> data processing;Polarimetric analysis & interpretation <strong>of</strong> broadband ground-<strong>based</strong> <strong>SAR</strong> data; andPotential application <strong>of</strong> ground truth <strong>for</strong> airborne or space-borne <strong>SAR</strong>.Design & Development <strong>of</strong> <strong>System</strong><strong>System</strong> Evaluation with CalibrationNatural Site Testing<strong>Ground</strong>-<strong>based</strong> <strong>SAR</strong> Data ProcessingAnalysis & InterpretationPotential <strong>Application</strong>Figure 1.02 Research objectives <strong>of</strong> this dissertation9
Chapter 11.4 Organization <strong>of</strong> the DissertationThe dissertation on the “<strong>Application</strong> <strong>of</strong> a <strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> <strong>for</strong>Environmental Studies” covers ground-<strong>based</strong> <strong>SAR</strong> working principles, polarimetriccalibration, three-dimensional (3-D) image reconstruction and polarimetric <strong>SAR</strong> dataprocessing as well as interpretation <strong>for</strong> environmental monitoring. It is divided into sixchapters:Chapter 1 Introduction.Chapter 2 <strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> and Its Evaluation.Chapter 3 Test Experiments <strong>for</strong> Tree Monitoring.Chapter 4 3D Image Reconstruction.Chapter 5 Data Interpretation with Polarimetric Analysis.Chapter 6 Conclusions.Following flowchart shows the relations <strong>of</strong> six chapters.Chapter 1IntroductionChapter 2Development & Evaluation <strong>of</strong> <strong>System</strong>Chapter 3Data AcquisitionChapter 43D ImagingChapter 5Polarimetric Analysis & InterpretationChapter 6ConclusionsFigure 1.03 Structure <strong>of</strong> the dissertation10
IntroductionChapter 1 gives the introduction including recent advances <strong>of</strong> polarimetric <strong>SAR</strong> imaging, asummary on current studies <strong>of</strong> ground-<strong>based</strong> <strong>SAR</strong>, and research purposes <strong>of</strong> this dissertation.Design and features evaluation <strong>of</strong> a broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong> system aredescribed in Chapter 2. In this chapter, computer simulation is used <strong>for</strong> determination <strong>of</strong>system specifications and measurement parameters <strong>based</strong> on working principles <strong>of</strong>ground-<strong>based</strong> <strong>SAR</strong>. Implementing polarimetric calibration and tests, the per<strong>for</strong>mances <strong>of</strong> thedeveloped system are also demonstrated. Chapter 3 describes the determination and choice <strong>of</strong>test targets, test procedures <strong>of</strong> natural environmental scatterers, and data acquisition. Bysimple signal processing, some signal wave<strong>for</strong>ms are plotted <strong>for</strong> verification includingdescriptions <strong>of</strong> the testing situation. In Chapter 4, main data processing and 3-D imagingalgorithms are presented. 3-D images <strong>of</strong> two experimental sites with different conditions areverified with ground truths. Different polarimetric descriptors <strong>for</strong> acquired data are discussedin Chapter 5. Data interpretations are carried out by polarimetric analysis from broadbandcontiguous frequencies images to multi-discrete-frequency images. The eigenvector-<strong>based</strong>polarimetric decomposition is discussed in this chapter. The last Chapter 6 provides theconclusions <strong>of</strong> the dissertation, including suggestions <strong>for</strong> ongoing future studies.There are two appendices that are added <strong>for</strong> supporting the text. Appendix A lists and expandsthe acronyms that are used throughout the dissertation. Appendix B provides the RCS <strong>of</strong> a few<strong>of</strong> the common ideal geometric reflectors implemented as test targets and their scatteringmatrices. At the end <strong>of</strong> this dissertation is a list <strong>of</strong> references, which explore each topic inmore detail.11
Chapter 112
Chapter 2GROUND-BASED POLARIMETRIC <strong>SAR</strong> SYSTEMAND ITS EVALUATIONFirst, principles and resolution <strong>of</strong> a ground-<strong>based</strong> <strong>SAR</strong> will be introduced from conventional<strong>SAR</strong> in this chapter. Based on some computer simulation results, a broadband ground-<strong>based</strong>polarimetric <strong>SAR</strong> system is designed and developed. Procedures <strong>of</strong> per<strong>for</strong>mance evaluationand polarimetric calibration are introduced. <strong>System</strong> specifications and calibration results arepresented.2.1 <strong>System</strong> Principles2.1.1 Remote Sensing and Synthetic Aperture Radar TechnologyRemote SensingRemote sensing is the science (and to some extent, art) <strong>of</strong> acquiring in<strong>for</strong>mation about theEarth's surface without actually being in contact with it. This is done by sensing and recordingreflected or emitted energy and processing, analyzing, and applying that in<strong>for</strong>mation [12, 14,49].In much <strong>of</strong> remote sensing, the process involves an interaction between incident radiation andthe targets <strong>of</strong> interest. This is exemplified by the use <strong>of</strong> imaging systems where manyelements are involved. However, the remote sensing technology also involves the sensing <strong>of</strong>emitted energy and the use <strong>of</strong> non-imaging sensors.There are two main categories <strong>of</strong> remote sensing imaging systems: passive and active [12].While passive systems make use <strong>of</strong> naturally emitted, reflected or scattered radiation fromsurfaces <strong>of</strong> targets, active systems are equipped with a transmitting unit and receive the signal13
Chapter 2backscattered or reflected from the illuminated terrain. An important class <strong>of</strong> active imagingsensors are radar systems operating in the microwave region <strong>of</strong> the electromagnetic spectrum.The active operating mode makes these sensors to be independent from external illuminationsources and additionally the fact <strong>of</strong> operating at the microwave region reduces drastically theimpact <strong>of</strong> clouds, fog and rain on the obtained images, different from the millimetrewavelength regime and higher. Thus, active radar imaging systems operated at microwavefrequencies allow widely day and night all-weather imaging, an important requirement <strong>for</strong>continuous global monitoring; and because at microwave frequencies the operatingwavelengths are <strong>of</strong> the order <strong>of</strong> effective length <strong>of</strong> isolated (point) and/or distributedscattering objects within the vegetated layer <strong>of</strong> the earth, resulting in improved resonancebehaviour <strong>for</strong> detection and identification.Imaging RadarAn imaging radar generates surface images that are at first glance very similar to the morefamiliar images produced by instruments that operate in the visible or infrared parts <strong>of</strong> theelectromagnetic spectrum. In fact, the image characteristics <strong>of</strong> radar are also quite differentfrom that <strong>of</strong> visible and infrared images due to a quite different scattering and diffractionmechanism <strong>for</strong> generating the images.Because <strong>of</strong> the way in which microwaves interact with the atmosphere and the ground, only aselect few frequency bands are useful <strong>for</strong> imaging. These are shown in Figure 2.01. Thewavelength affects the penetration depth and also the size <strong>of</strong> a target necessary <strong>for</strong> a signal toreturn to the radar.mcmmmλ 330332.01.030157.53.752.51.671.117.54.0BandIGP LS C XKu KKa VW1.53.01.02.04.08.01.21.82.74.07.5f (Hz)10 8 10 9 10 1010 11Figure 2.01 Microwave frequency bands used <strong>for</strong> imaging radar.14
<strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> and Its EvaluationImaging radar system generally provides a two-dimensional image <strong>of</strong> the radar reflectivity <strong>of</strong>a scene by illuminating it with microwave pulses and receiving the scattered field. There aretwo possible operation scenarios <strong>for</strong> such radar systems. The first one is to use the samesystem <strong>for</strong> transmitting and receiving. In this case receiver and transmitter are located at thesame position and there<strong>for</strong>e this configuration is known as monostatic configuration. It is theclassical operation scenario <strong>for</strong> space and airborne radar systems. In the second scenarioreceiver and transmitter are spatially separated from each other by using one active (i.e.transmitting) system to illuminate the scene and one or more passive (i.e. receiving only)systems <strong>for</strong> receiving the scattered field. Such bi/multi-static configurations are up to nowmostly used <strong>for</strong> laboratory and military radar measurements but may become a prospectivealternative <strong>for</strong> cost-effective future space borne multi-static implementations such as theCart-Wheel concept [4]. In our case, the monostatic radar system is employed also <strong>for</strong> thebroadband ground-<strong>based</strong> polarimetric <strong>SAR</strong> application.In a simplified description, a monostatic radar system consists <strong>of</strong> a pulsed microwavetransmitter, an antenna which is used <strong>for</strong> transmission and reception, and a receiver unit. It ismounted on a moving plat<strong>for</strong>m such as a ground-<strong>based</strong> positioner, an airplane, a drone, thespace-shuttle or a satellite and operated in a side-looking geometry. Accordingly, the antennabeam is directed slant-wise towards the targets and – in the most conventional implementation- orthogonal to the moving direction, which is referred to as range or cross-track direction.The transmitter pulse is directionally radiated by the antenna towards the target area. Thebackscattered transmitter pulse signal is received by the receiver-antenna and the receiver unit.The plat<strong>for</strong>m motion in the moving direction provides the scanning in the direction <strong>of</strong> thesensor trajectory, which is referred to as azimuth or along-track direction [47, 48].Resolutions and Synthetic Aperture RadarOne <strong>of</strong> the most important quality criteria <strong>of</strong> an imaging sensor is its spatial resolution. It is ameasure <strong>of</strong> how close two point-like objects can be located to each other in order to still beseparated in the image. For radar imaging sensors, the spatial resolution is given <strong>for</strong> range andazimuth separately. In range direction it is inversely proportional to the bandwidth <strong>of</strong> thetransmitted signal [22, 47]. A wider bandwidth results into a higher range resolution. Theconstraints on the achievable range resolution are mainly given by the bandwidth limitation <strong>of</strong>the antennas. Nowadays, the achievable range resolutions are about 1 to 10 meters <strong>for</strong>space-borne systems while airborne systems are capable to achieve resolutions on the order <strong>of</strong>ten centimetres and ground-<strong>based</strong> systems provide range resolution about several centimetres15
Chapter 2[4, 46].More important is the quest <strong>for</strong> achieving reasonably high resolution in the azimuth direction.The first generation <strong>of</strong> radar imaging system was the so-called real aperture radar,characterised by an azimuth resolution depending on the operating frequency, the size <strong>of</strong> thereal antenna, and the distance between sensor and the object to be imaged. Due to technicaland operational constrains on both, the size <strong>of</strong> the used antenna and the transmitted frequency,the achievable resolution was poor and varied inside the image [12, 49].The solution to these limiting problems was given by the next generation <strong>of</strong> radar sensorsoperating according to the signal processing concept <strong>of</strong> a synthetic antenna or a syntheticaperture. This radar imaging system is called synthetic aperture radar (<strong>SAR</strong>). <strong>SAR</strong> systemsare active remote sensing sensors. It generates a reflectivity map <strong>of</strong> an illuminated areathrough transmission and reception <strong>of</strong> electromagnetic energy in the microwave region.Thebasic idea <strong>of</strong> this concept is to simulate a very long antenna by moving a small real antennaalong the moving direction. The coherent integration <strong>of</strong> the received signals along the movingtrack allows synthesising a long virtual-antenna and leads to images with high azimuth spatialresolution independent <strong>of</strong> the operating frequency and the distance to the scene [47]. Figure2.02 shows geometry <strong>of</strong> <strong>SAR</strong>. The spatial resolutions <strong>of</strong> <strong>SAR</strong> can be obtained by thefollowing equations.Drgcτc= = (2.01)2 2 ⋅ BλLsar≈θ⋅ R= R(2.02)Lλθsar= (2.03)2L sarDazL= θsar⋅ R= (2.04)2whereDrgis the range resolution, c is the speed <strong>of</strong> light, B is the radar bandwidth,Lsaristhe synthetic aperture length <strong>of</strong> <strong>SAR</strong>, θ is the angular resolution <strong>of</strong> antenna beam, R is therange distance, λ is the wavelength <strong>of</strong> the centre frequency, L is the antenna physical size inthe azimuth direction, θsaris the angular resolution <strong>of</strong> synthetic aperture length, and Dazisthe azimuth resolution <strong>of</strong> <strong>SAR</strong>. The maximum length <strong>for</strong> the synthetic aperture is the length<strong>of</strong> the moving path from which a scatterer is illuminated. The angular resolution <strong>of</strong> a synthetic16
<strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> and Its Evaluationaperture <strong>of</strong> the length Lsaris given by the diffraction limit (due to diffraction effects on itsaperture). The factor 2 is the result <strong>of</strong> the synthetic aperture <strong>for</strong>mation [47].We can find that both range and azimuth resolutions are independent <strong>of</strong> distance to target. For<strong>SAR</strong>, the appropriate coherent integration <strong>of</strong> the signals received from a scatterer by an array<strong>of</strong> antenna elements is used in order to synthesize a large antenna with a very narrow beam.The constrains on the azimuth resolution <strong>of</strong> <strong>SAR</strong> systems are given by practical limitations onthe transmitted power, and data rate, leading to resolutions <strong>of</strong> several meters <strong>for</strong> space-borne<strong>SAR</strong>, on the order <strong>of</strong> one meter or better <strong>for</strong> airborne radars and on tens <strong>of</strong> centimetres or less<strong>for</strong> ground-<strong>based</strong> sensors [46].LsarLAzimuthRangeθRFigure 2.02 Geometry <strong>of</strong> synthetic aperture radar.2.1.2 Radar PolarimetryMany radar systems are designed to transmit microwave radiation <strong>of</strong> orthogonal polarizationpairs, <strong>for</strong> example, that is either horizontally polarized (H) or vertically polarized (V). Atransmitted wave <strong>of</strong> either polarization can generate a backscattered wave with a variety <strong>of</strong>polarizations. It is the analysis <strong>of</strong> these pairs <strong>of</strong> orthogonal transmitting and receivingpolarization combinations that constitutes the science <strong>of</strong> radar polarimetry [50, 51].Any polarization on either transmission or reception can be synthesized by using H and Vcomponents with a well-defined relationship between them shown in Figure 2.03. For reasons<strong>based</strong> on the design <strong>of</strong> polarization switches, and so on, systems that transmit and receive17
Chapter 2both <strong>of</strong> these linear polarizations are commonly used. With these radars, there can be fourcombinations <strong>of</strong> transmit and receive polarizations:(a)(b)(c)(d)HH – <strong>for</strong> horizontal transmit and horizontal receiveVV – <strong>for</strong> vertical transmit and vertical receiveHV – <strong>for</strong> vertical transmit and horizontal receive, andVH – <strong>for</strong> horizontal transmit and vertical receive.The first two polarization combinations are referred to as “co-polarized” because thetransmitting and receiving polarizations are the same. The last two combinations are referredto as “cross-polarized” because the transmitting and receiving polarizations are orthogonal toone another.TXYR XR YSXXSYXSXYSYYFigure 2.03 Transmit / receive polarimzation signal succession in radar polarimetry.The primary description <strong>of</strong> how a radar target or surface feature scatters electromagneticenergy is expressed in terms <strong>of</strong> the scattering matrix. A polarimetric radar can be used todetermine the target response or scattering matrix using two orthogonal polarizations,typically linear H and linear V, as well as X polarization and Y polarization on each <strong>of</strong>transmitter and receiver, respectively. If a scattering matrix is known, the response <strong>of</strong> thetarget to any combination <strong>of</strong> incident and received polarizations can be computed by equation(2.05). [50, 52, 53]si⎡E ⎤ 1 ⎡SxxSxy⎤⎡E ⎤⎢ ⎥ =s ⎢ ⎥⎢ ⎥i e⎢⎣Ey⎥⎦ 4π r ⎢⎣SyxSyy⎥⎦⎢⎣Ey⎥⎦x x − jkr(2.05)18
<strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> and Its EvaluationThis is referred to as polarization synthesis, and illustrates the power and flexibility <strong>of</strong> a fullypolarimetric radar system.Both the wavelength and polarization affect how a radar system “observes” the elements inthe scene. There<strong>for</strong>e, radar imagery collected using different polarization and wavelengthcombinations may provide different and complementary in<strong>for</strong>mation. Furthermore, whenthree scattering matrix returns are combined in a color composite, the in<strong>for</strong>mation is presentedin a way that an image interpreter can infer more in<strong>for</strong>mation <strong>of</strong> the surface characteristics.2.1.3 Principle <strong>of</strong> <strong>Ground</strong>-<strong>based</strong> <strong>SAR</strong> <strong>System</strong>Synthetic aperture radar using space borne and airborne data has been a very interestingtechnique in the domain <strong>of</strong> remote sensing <strong>for</strong> many years. It can also be exploited in thedesign <strong>of</strong> ground-<strong>based</strong> radar imaging systems. The highly precise monitoring <strong>of</strong> thede<strong>for</strong>mation <strong>of</strong> ground surfaces is very important in many applications. For instance,monitoring <strong>of</strong> volcano activities by observing the surface bulging be<strong>for</strong>e the eruption wouldbe advantageous <strong>for</strong> disaster prevention due to volcano eruptions, as those are still notstraight-<strong>for</strong>wardly predictable by other methods currently in use. For these applications,ground-<strong>based</strong> broadband <strong>SAR</strong> has a wide range <strong>of</strong> applications and advantages due to theper<strong>for</strong>mance <strong>of</strong> broadband electromagnetic wave interrogation. Moreover, it can be applied todetect the subsurface structure and surface movements even under cloudy and dustyconditions, or also inside buildings, in which cases other optical imaging techniques, such aslaser distance meters, and also GPS cannot be used.Hence, a broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong> technique is employed <strong>for</strong> environmentalstudies by detecting vegetation changes <strong>of</strong> various kinds <strong>of</strong> vegetation cover due to seasonalvariations by our research group [17, 46].In a ground-<strong>based</strong> arrangement, the synthesis aperture can be obtained by scanning theantenna on a linear horizontal mechanical guide and/or a vertical guide. The syntheticaperture is realized by scanning the antenna system on a horizontal rail and movingadditionally along a vertical bar. The scan along the rail provides image azimuth spatialresolution; the vertical scan contributes to vertical spatial resolution. The range resolution isprovided by the broadband <strong>SAR</strong> synthesis algorithm <strong>of</strong> radar signals and it depends on theradar bandwidth [47]. As the radar image is obtained through synthesis and sampling19
Chapter 2techniques, its characteristics are constrained by the radar measuring parameters: 1)bandwidth; 2) frequency step; 3) scan aperture length; and 4) scan interval. There<strong>for</strong>e, we canassume the three spatial resolutions: the azimuth resolution Dx; the rabge resolution Dy; and,the vertical resolution Dz, which are given by the following equations [54].c ⎛Lx⎞Dx = ⎜ ⎟ + R2Lx ⋅ f ⎝ 2 ⎠22(2.06)cDy = cosϕ(2.07)2Bc ⎛Lz⎞Dz = ⎜ ⎟ + R2Lz ⋅ f ⎝ 2 ⎠22(2.08)where c is the speed <strong>of</strong> light, Lx is the total horizontal scan aperture <strong>of</strong> the antenna, Lz is thevertical scan aperture length, ϕ is the antenna elevation angle, R is the range distance, f isthe radar center frequency, B is the radar bandwidth. For the broadband radar system case, 4GHz <strong>of</strong> bandwidth could be used <strong>for</strong> achieving range resolution <strong>of</strong> 0.04m, which is generallysatisfactory in practical application. The azimuth resolution will be 0.11m with azimuthaperture <strong>of</strong> 10m and range distance <strong>of</strong> 20m.2.2 Simulation and Design <strong>of</strong> a <strong>Ground</strong>-<strong>based</strong> <strong>SAR</strong> <strong>System</strong>2.2.1 Simulation <strong>of</strong> Azimuth ResolutionThe principle <strong>of</strong> <strong>SAR</strong> shows that the azimuth resolution <strong>of</strong> <strong>SAR</strong> is independent <strong>of</strong> the rangedistance. It is theoretically true when the <strong>SAR</strong> data were acquired along a very long surveyline. This ideal situation can easily be satisfied <strong>for</strong> space and air borne <strong>SAR</strong> system. But it isnot easy achieved <strong>for</strong> the ground-<strong>based</strong> <strong>SAR</strong> system. If the scanning aperture length is limited,the <strong>SAR</strong> resolution can be a function <strong>of</strong> a range distance. The azimuth resolution <strong>of</strong>reconstructed images depends on the synthetic aperture length, media conductivity and20
<strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> and Its Evaluationpermittivity, and antenna directivity. This section discusses the azimuth resolution in terms <strong>of</strong>these controlling parameters <strong>based</strong> on computer simulation results in order to achieve the bestper<strong>for</strong>mance <strong>of</strong> the ground-<strong>based</strong> <strong>SAR</strong> system.Ray Tracing MethodRay tracing is a well-known technique in simulating the behavior <strong>of</strong> wave propagation in ahomogeneous space [55]. There are several variations <strong>of</strong> the algorithm, which are not allcovered here. In the basic algorithm the transmitting source emits microwave rays, which arethen reflected at surfaces according to specular reflections and the receiver keeps track onwhich rays have returned to it by recording the reflections. The specular reflection rule ismost common, in which the incident angle <strong>of</strong> an incoming ray is the same as the incidentangle <strong>of</strong> the outgoing ray. More advanced rules which include <strong>for</strong> example some diffusionalgorithm have also been studied [56]. In practice a sphere is in most cases the best choice,since it provides an omni-directional sensitivity pattern and it is easy to implement. A typicalgoal is to have a uni<strong>for</strong>m distribution <strong>of</strong> rays over a sphere. Of course, it can be simplified asa point target. The ray tracing method has anadvantage over physical optics, no restriction isrequired concerning the distance between theradiating source and the reflecting interfaces,or between the interfaces in the case <strong>of</strong> multi-reflection environments.Excitation Pulse and Generated Diffraction DataWe used the ray tracing method to simulate the measured data set <strong>of</strong> a ground-<strong>based</strong> <strong>SAR</strong>within limited aperture length and range distance, and reconstruct the image <strong>of</strong> the target todetermine the estimated resolution <strong>of</strong> the system.For our study case, a single pulse is recovered from a broadband frequency spectrum byinverse Fourier Trans<strong>for</strong>m. The excitation pulse is shown in Figure 2.04, which is used as asource signal to propagate to the target. We assume that the main targets are located in therange <strong>of</strong> 20 meters. So we select the range distance from 10m, 15m and 20m. The scanningaperture lengths are 2m, 5m, 10m and 15m, respectively. The parameters are shown in Table2.01.21
Chapter 2Excitation pulseFigure 2.04 Excitation pulse <strong>for</strong> simulation.Table 2.01 Parameters used <strong>for</strong> simulationTargetA pointTime step0.0488 nsRange distance10, 15, 20 mPulse duration0.35 nsScan aperture2, 5, 10, 15, 20 mImage pixel0.05 x 0.05 mScan interval0.1 mImage ReconstructionUsing the generated datasets <strong>for</strong> 12 cases,a focusing algorithm <strong>of</strong> diffraction stacking is22
<strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> and Its Evaluationused <strong>for</strong> image reconstruction [46]. This algorithm will be discussed in detail in Chapter 4.Some reconstructed images are shown in Figure 2.05 and Figure 2.06, respectively.Aperture Length: 5 mAperture Length: 10 m0.6m0.6m 0.6mVertical Slice0.6mHorizontal Slice(a)(b)(c)(d)Figure 2.05 Simulated images <strong>of</strong> a point target at range <strong>of</strong> 10m: (a) horizontal slice withaperture length 5m, (b) horizontal slice with aperture length 10m, (c) vertical slice withaperture length 5m and (d) vertical slice with aperture length 10m.In above figure, we can find that the azimuth resolution is about 0.15m <strong>for</strong> an aperture length<strong>of</strong> 5m; and 0.1m <strong>for</strong> an aperture length <strong>of</strong> 10m while the range distance is the same, 10m. Thevertical resolution is poor in Figure 2.05(c) and 2.05(d) due to shorter scanning lengths alongvertical direction, which we do not further consider because <strong>of</strong> the limited aperture length invertical direction.23
Chapter 2Range 15 mRange 20 m2 m in Range2 m in Azimuth3 52 m in Azimuth2 m in Range2 m in RangeAperture length : 5m2 m in Range(a)(b)Aperture length : 10m2 m in Azimuth(c)2 m in Azimuth(d)Aperture length : 15m2 m in Range2 m in Azimuth(e)2 m in AzimuthFigure 2.06 Simulated horizontal images <strong>of</strong> a point target at range 15m and 20m with: (a)aperture 5m and range 15m, (b) aperture 5m and range 20m, (c) aperture 10m and range15m, (d) aperture 10m and range 20m , (e) aperture 15m and range 15m, and (f) aperture2 m in Range(f)24
<strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> and Its Evaluation15m and range 20m.In Figure 2.06, the azimuth resolutions are about 0.3m <strong>for</strong> aperture length <strong>of</strong> 5m, 0.12m <strong>for</strong>aperture length <strong>of</strong> 10m, and 0.1m <strong>for</strong> scanning aperture 15m while the range distance is 15m.From Figure 2.06(b), 2.06(d) and 2.06(f), the azimuth resolutions are about 0.5m <strong>for</strong> aperturelength <strong>of</strong> 5m, 0.15m <strong>for</strong> aperture length <strong>of</strong> 10m, and 0.12m <strong>for</strong> scanning aperture 15m whilethe range distance is 20m, respectively.According to the estimated resolution <strong>for</strong> each case, curves <strong>for</strong> the relation between scanningaperture length and estimated resolution <strong>for</strong> different range distances are drawn in Figure2.07.Figure 2.07 Estimated resolutions by simulation.The resolution will become worse rapidly when the scanning aperture length becomes shorterthan 5m. If distance <strong>of</strong> target to the antenna is less than 20m, the scanning aperture lengthmust remain at more than 10m <strong>for</strong> an azimuth resolution <strong>of</strong> about 0.15m which is satisfactory<strong>for</strong> tree monitoring. Of course, the scanning aperture length should be extended in case abetter resolution is necessary. In fact, with increase <strong>of</strong> range distance, the scanning apertureshould be expanded synchronously, to retain a good resolution. Additionally, we found that25
Chapter 2the resolution is not sensitive with the change <strong>of</strong> the scanning interval in azimuth direction.From the simulation results, we found the estimated resolution value is consistent with thecalculated value given by Equation (2.06).2.2.2 Design <strong>of</strong> a Broadband <strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong><strong>System</strong> DescriptionBased on <strong>SAR</strong> principles and above simulation results, we have extended those earlierapproaches and designed an in-house laboratory broadband, ground-<strong>based</strong>, fully polarimetric<strong>SAR</strong> system <strong>for</strong> environmental studies. The ground-<strong>based</strong> polarimetric broadband <strong>SAR</strong>technique is employed <strong>for</strong> environmental studies by detecting changes in various kinds <strong>of</strong>vegetation cover due to seasonal variations by our research group.The radar system consisted <strong>of</strong> a vector network analyzer [Agilent HP8720ES], a diagonal dualpolarized broadband horn antenna [ETS-EMCO model 3164-03], an antenna positioner unit[DEVICE Inc.], and a PC-<strong>based</strong> control unit. The network analyzer, operated in a steppedfrequency continuous-wave mode [58, 59], was used to generate the transmitting signal and todetect scattered signals both in amplitude and phase. The synthetic aperture is realized byscanning the antennas on a horizontal rail and moving along a vertical post. The horizontaland vertical scanning aperture widths determine the horizontal and vertical resolutions. Therange resolution depends on the radar frequency and bandwidth [47]. Figure 2.08 shows anillustration <strong>of</strong> a broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong> system.26
<strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> and Its EvaluationDual polarizeddiagonal horn antennaVectornetworkanalyzerPositioner controllerPositionerNECPCFigure 2.08 <strong>System</strong> diagram <strong>of</strong> a broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong>.A dual polarized diagonal horn is used <strong>for</strong> the broadband ground-<strong>based</strong> polarimetric radarsystem shown in Figure 2.09.(a)(b)Figure 2.09 A dual polarized diagonal horn EMCO 3164-03 # 1043: (a) front view, and (b)side view.27
Chapter 2With the aid <strong>of</strong> a testing measurement in the Anechoic Chamber, the return loss and voltagestanding wave ratio (VSWR) <strong>of</strong> the dual polarized diagonal horn antenna EMCO 3164-03 arepresented in Figure 2.10. From this figure, the working frequency region can be determined. Itcan be operated effectively from 0.8 GHz to 6.3 GHz. By comparing the two antennas: #1043and #1044, we found the antenna <strong>of</strong> # 1044 has a better polarized balance, showing that it canbe used <strong>for</strong> the polarimetric measurements and it is used <strong>for</strong> the polarimetric measurement.Then the radiation patterns are calculated and plotted in Figure 2.11, where #1043 istransmitter and # 1044 is receiver. The antenna beam changes gradually from 60 degree with1GHz to 20 degree with 5GHz. The antenna directivity will be used <strong>for</strong> antenna patterncompensation when migration is carried out.(a)(b)(c)(d)Figure 2.10 Return loss and VSWR <strong>of</strong> a dual polarized diagonal horn: (a) Return loss <strong>of</strong>EMCO 3164-03 #1043, (b) Return loss <strong>of</strong> EMCO 3164-03 #1044, (c) VSWR <strong>of</strong> EMCO3164-03 #1043 and (d) VSWR <strong>of</strong> EMCO 3164-03 #1044.28
<strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> and Its Evaluation(a)(b)(c)(d)(e)Figure 2.11 Radiation patterns <strong>of</strong> a dual polarized diagonal horn: (a) 1 GHz, (b) 2 GHz, (c) 3GHz, (d) 4 GHz and (e) 5 GHz.29
Chapter 2<strong>System</strong> SpecificationFigure 2.12 shows a photograph <strong>of</strong> the broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong> system.This system may be operated at frequencies between 400 MHz and 6 GHz, with a scanningaperture <strong>of</strong> 20 m in the horizontal and 1.5 m in the vertical range. The working frequencyrange can extend from 50 MHz up to 20 GHz in case double-ridged waveguide broadbandhorn antennas [Antenna Giken Corp.: 3M-00124B] are employed. The measurement and dataacquisition were accomplished by a specially designed control program. The broadbandpolarimetric radar images can be reconstructed from the scattering data by wave migrationalgorithms as discussed in [17]. Because a three-dimensional reflectivity image can be <strong>for</strong>medby synthesizing the two-dimensional aperture, a diffraction stacking algorithm was applied toreconstruct the radar image from the scattering data. Based on computer simulation results,some experimental parameters are determined in advance <strong>for</strong> antenna calibration and systemvalidation shown in Table 2.02 [46].Figure 2.12 A developed broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong> system.30
<strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> and Its EvaluationTable 2.02 Specifications <strong>of</strong> a developed broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong> systemVector network analyzerAntennaorPositionerAperture lengthHP 8720ES / HP 8753EDual polarized diagonal horn:ETS-EMCO 3164-03 # 1044Double-ridged waveguide horn:Antenna Giken. #3M-00124BDEVICE Inc. DX3151BV1/OHorizontalVertical0.05 ~20.05 GHz0.5 ~6.3 GHz0.1~12 GHzAccuracy 0.1 mm20 m1.5 m2.3 Polarimetric Per<strong>for</strong>mance EvaluationFor evaluating the polarimetric ground-<strong>based</strong> <strong>SAR</strong> system, we have per<strong>for</strong>med severalmeasurements to test the dual polarization broadband diagonal horn antenna using thestandard reflectors.2.3.1 Test Using a Metallic SphereFirstly, we used a metallic sphere with radius 0.075 m as a calibrator to test the radar system.The measurement was carried out in outdoor condition and some absorbing materials wereused <strong>for</strong> reducing reflection from target stand. Figure 2.13 shows a test scene <strong>for</strong> a metallicsphere.31
Chapter 2Figure 2.13 Testing scenario <strong>of</strong> a metallic sphere.If the range distance is not sufficiently long enough <strong>for</strong> the dual polarized diagonal broadbandhorn antenna [ETS 3164-03] <strong>for</strong> such a wide frequency range, the signal returned from thecalibration target was hardly distinguishable from the direct coupling wave [61]. Otherproblems were the long range separation <strong>of</strong> the calibration target and its relative electricdistance above the ground. Since the calibration target (sphere) was relatively very close tothe ground-surface, the returned signals could not easily be separated.This fact can be described as follow. We can simply calculate the arrival time differencebetween the direct reflection from the target and the wave reflected from the ground and thenre-reflected by the target.ttarget2r= (2.09)ctg+t2⎛r⎞ +2h +2 ⎜ ⎟ r⎝r⎠= (2.10)c32
<strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> and Its Evaluation2⎛r⎞ +2h −2 ⎜ ⎟ r⎝r⎠∆ t = tg+t− ttarget= (2.11)cwheret t arg et: arrival time <strong>of</strong> the reflection from targettg + t : arrival time <strong>of</strong> ground reflectionr: distance between antenna and targeth: height <strong>of</strong> antenna and target from groundc: light speed in airThe time-elapse <strong>of</strong> reflections from the front <strong>of</strong> the target versus the backscatter from thetarget (the first-round creeping-wave) in Figure 2.14, <strong>for</strong> example, <strong>for</strong> a metallic sphere withradius a = 0.075 m, is approximately given as [64]2a+π aδt= = 1.286ns(2.12)caaFigure 2.14 Geometry <strong>of</strong> creeping wave <strong>for</strong> a sphereHence, the arrival time difference between the direct reflection from the target and the wavereflected from the ground and then re-reflected by the target ∆ t must be greater thanδ t , thetime-elapse <strong>of</strong> reflections from the front <strong>of</strong> the target versus that <strong>of</strong> the backscatter from thetarget (the creeping wave). Otherwise, the reflections <strong>of</strong> the target and the ground scatteringare so close that we can not separate them clearly. In this case, we selected the height and therange as h = 2.4 m, r = 8 m. Hence, ∆ t = 4.432ns, and <strong>for</strong> this choice <strong>of</strong> parameters, we caneasily distinguish the wave<strong>for</strong>ms associated with the reflections <strong>of</strong> the target and the groundin Figure 2.15, which shows the returned signal <strong>of</strong> the metallic sphere in Figure 2.13.33
Chapter 2Reflection <strong>of</strong> sphereamplitudetime [ns]Figure 2.15 Reflection wave<strong>for</strong>ms from the metallic sphere.The radar cross section (RCS) [61, 62] <strong>of</strong> an object exposed to a radar is a fictitious area thatdescribes the intensity <strong>of</strong> the wave reflected back to the radar, like the effective area <strong>of</strong> anantenna. The backscattering RCS <strong>of</strong> a metallic sphere can be calculated from the Mie seriesand is given by the following <strong>for</strong>mulas [64].22∞ nσ = λ / π ∑ ( − 1) ( n+ 0.5)( bn−an)(2.13)n=1whereabnnj ( ka)=h ( ka)n(1)nkaj ( ka) − nj ( ka)=kah ( ka) nh ( ka)n−1n(1) (1)n−1−n(2.14)andk = 2 π / λh ( x) = j ( x) + jy ( x)(1)n n n(2.15)a : the radius <strong>of</strong> sphere34
<strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> and Its Evaluationλ : the wavelength <strong>of</strong> radar waveh(1) ( x ): the spherical Hankel function <strong>of</strong> the first kindnjn ( x ): the spherical Bessel function <strong>of</strong> the first kindyn( x ): the spherical Bessel function <strong>of</strong> the second kindThe normalized radar cross section, NRCS, is defined as the backscatter factor <strong>of</strong> radar targetand can be expressed as [61, 64]NRCS = σ / πa2∞22n(2 / ka) ∑( 1) ( n 0.5)( bnan)n=1= − + −(2.16)Substituting the radius <strong>of</strong> the metallic sphere <strong>of</strong> radius a = 0.075 m, we obtain the theoreticalNRCS value shown as the solid line in Figure 2.16. The measured data <strong>of</strong> HH, VV and VHcomponents, compensated by the gain factor <strong>of</strong> the antenna used in the system, are alsoplotted in same figure synchronously.Figure 2.16 Theoretical normalized RCS curve and measured values <strong>of</strong> a metallic sphere:black solid line- theoretical curve, red dotted line- HH polarization, blue dashed line- VVpolarization and green dash-dot line- VH polarization.For the sphere case, we find that the wave<strong>for</strong>ms <strong>of</strong> both the HH and VV reflected signals in35
Chapter 2Figure 2.15 have very similar amplitude and the same phase. We also observe that there arethe same number <strong>of</strong> peaks and that the resonance points <strong>of</strong> the measured curves coincide withthose <strong>of</strong> the theoretical value shown in Figure 2.16. Hence, the measured backscattering data<strong>of</strong> co-polarization components are consistent and very similar to the theoretical value. Thecross-polarization component is about 12 dB lower than co-polarization components.2.3.2 Test Using Four Standard ReflectorsBased on the developed broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong> system, we carried outsome experiments with a few <strong>of</strong> the standard reflectors to demonstrate the polarimetricper<strong>for</strong>mance <strong>of</strong> this system. We used a vertical dihedral, a 45-degree dihedral, a vertical wireand a –45-degree wire as the calibration targets shown in Figure 2.17.Figure 2.17 Four standard reflectors setups.36
<strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> and Its EvaluationAccording to the reflecting signals <strong>of</strong> each standard reflector <strong>for</strong> HH, VH and VV scatteringmatrix elements, we calculated the associated scattering matrix at frequency 2 GHz <strong>for</strong> eachreflector as given below.Vertical dihedral:Sm1oo⎡− j113.7 j129.40.9813 e 0.0365e⎤= ⎢ ⎥o (2.17)j129.4⎢⎣0.0365 e 1 ⎥⎦45-degree dihedral:Sm2o⎡j120.40.2977 e 1 ⎤= ⎢ ⎥o (2.18)j112.3⎢⎣1 0.2891e⎥⎦Vertical wire:Sm3oo⎡j167.7 − j153.70.2635 e 0.1345e⎤= ⎢ ⎥o (2.19)− j153.7⎢⎣0.1345 e 1 ⎥⎦-45-degree wire:Sm4o⎡− j93.11 0.5747e⎤= ⎢ ⎥o o (2.20)− j93.1 − j172.0⎢⎣0.5747 e 0.8770e⎥⎦The theoretical scattering matrices <strong>of</strong> the above four reflectors are known and given byS1⎡−1 0 ⎤ ⎡0 1⎤⎡0 0⎤1 ⎡1 -1⎤= ⎢0 1 ⎥⎣ ⎦ , S = 2 ⎢1 0 ⎥⎣ ⎦ , S = 3 ⎢0 1 ⎥⎣ ⎦ , and S = 4 2 ⎢-1 1 ⎥⎣⎦ , respectively.Compared with the theoretical results, good agreement on magnitude <strong>of</strong> the scatteringmatrixes can be found. We can also observe the consistent relation <strong>for</strong> most test targets except<strong>for</strong> the result <strong>of</strong> -45 degree wire is worse due to the strong interfering effect <strong>of</strong> the target stand.Moreover, from the HH and VV reflection wave<strong>for</strong>ms <strong>of</strong> the vertical dihedral in Figure 2.18,we observe the same magnitude and inverse phases. This behavior demonstrates the goodpolarimetric per<strong>for</strong>mance <strong>of</strong> this system <strong>for</strong> both narrow and broadband measurements.37
Chapter 2Figure 2.18 Reflection wave<strong>for</strong>ms <strong>of</strong> a dihedral corner reflector.2.4 Polarimetric CalibrationIn the previous section, we found that the difference between measured value and theoreticalvalue and polarization imbalance existed. Hence, polarimetric calibration is necessary <strong>for</strong> thebroadband ground-<strong>based</strong> polarimetric <strong>SAR</strong> system.2.4.1 Radar Cross Section <strong>of</strong> Dihedral Corner ReflectorThe radar cross section (RCS) <strong>of</strong> a target is the projected area that would intercept thetransmitted signal and reflect isotropically ally an amount that produces the returned signal at38
<strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> and Its Evaluationthe receiver. In other words, radar cross section provides an indication <strong>of</strong> how well a giventarget reflects radar energy [63]. With these ideas in mind, it is not surprising that the physicalarea <strong>of</strong> a target is normally greater than the radar cross section because some <strong>of</strong> the incidentenergy is scattered and absorbed by the target.The radar cross section <strong>of</strong> a target is not constant with operating frequency. There are threebroad regions <strong>of</strong> interest with respect to physical target size, operating frequency and resultingradar cross section. These regions are [60]:(a)(b)(c)Raleigh region. If the target is considerably smaller than the wavelength <strong>of</strong> the radarsystem, the target is said to be in the Raleigh region. If the target is in the Raleighregion, the radar cross section <strong>of</strong> the target tends to be smaller than the target'sphysical size.Resonance region. If the target is <strong>of</strong> similar dimension to that <strong>of</strong> the wavelength, thetarget is said to be in the resonance region. In the resonance region, the radar crosssection <strong>of</strong> the target may vary a great deal but tends to be larger than the physicalsize <strong>of</strong> the target.Optical region. The optical region occurs when the target is much larger than theoperating wavelength <strong>of</strong> the radar. This is quite <strong>of</strong>ten the case with operational radarsystems whose wavelengths are normally in the order <strong>of</strong> centimeters in length. Whenoperating in this region, the radar cross section <strong>of</strong> the target is similar to its physicalsize.The implication <strong>of</strong> these three regions is that the operating wavelength should not be selectedin total isolation from target considerations.As one <strong>of</strong> the most popular radar targets, the dihedral corner reflector has been discussed byEugene F. Knott on its RCS evaluation and reduction [65], and also by T. Griesser and C. A.Balanis et al. [66-69]. It is known that it can produce strong back scattering in a wide region,especially when the dihedral angle is 90 ° . An effective, although approximate method is to usegeometrical optics to find the surface field distributions due to the reflected rays and to theuse physical optics approximation to calculated the far-scattered fields due to thesedistributions. In this method, the diffraction terms will be neglected <strong>for</strong> simplicity.The dihedral geometry is shown in Figure 2.19. Faces A and B have dimensions a and b, andthe length perpendicular to the plane <strong>of</strong> the figure will be designated l. The angle between thetwo faces is 2β and the angle <strong>of</strong> arrival <strong>of</strong> an incident plane wave is ϕ , measured from the39
Chapter 2plane bisecting the dihedral angle. There are 4 distinct contributions to be considered; two arethe direct reflection (single-bounce) from faces A and B, a third is the double-bouncecontribution <strong>of</strong>f face A to face B and then back to the radar, and the fourth is a double-bouncecontribution <strong>of</strong>f face B to face A and then to the radar. Using single and double subscripts todenote the four, the radar cross section is given by:2λ2σ = Sa + Sb + Sab + Sba(2.21)πwhere Sa, Sb, Sab, and S baare the scattering coefficients <strong>of</strong> face A, face B,double-bounce from A to B and double-bounce from B to A, respectively.For aspect angles −β ≤ϕ ≤ β , both faces A and B are fully illuminated by the incident waveand then have:S =− jka(/ l λ)sin( β + ϕ)eaS =−jkb(/ l λ)sin( β −ϕ)ebcos( β ϕ) sin( ka cos( β ϕ))ka cos( β + ϕ)− jka + +cos( β ϕ) sin( kbcos( β ϕ))kb cos( β −ϕ)− jkb − −(2.22)(2.23)where λ is wavelength <strong>of</strong> incident wave and k = 2 π / λ is the wavenumber.a’ b’aββbface Aface BϕIncidentdirectionFigure 2.19 Geometry <strong>of</strong> a dihedral corner reflector according to [65].For incident electric polarizations parallel and perpendicular to the dihedral seam (E and Hpolarizations, respectively), the physical optics <strong>for</strong>mulas <strong>for</strong> Saand S bdo not change, butthe double-bounce contributions S aband S bachange from one polarization to the other.Hence, the equations S aand S bhold <strong>for</strong> either E or H polarizations, and it has beenassumed that the receiver and the transmitter are co-polarized.40
<strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> and Its EvaluationThe double-bounce contributions can be estimated by finding the surface illumination <strong>of</strong> oneface to a geometrical reflection <strong>of</strong> the incident wave <strong>of</strong>f the other. For double-bounce, theeffective face widths are used instead <strong>of</strong> the true widths. It is possible under certain conditions<strong>of</strong> arrival <strong>of</strong> the incident wave that a face may not be illuminated at all by a wave reflectedfrom the other face. There<strong>for</strong>e the effective face widths are taken as:⎧⎪0ϕ ≤ −α⎪a'= ⎨ a−α ≤ ϕ ≤ γ −α⎪ sin( β − ϕ)⎪ bϕ ≥ γ −α⎪⎩ sin(3 β − ϕ)(2.24)⎧ sin( β + ϕ)⎪aϕ ≤ γ − βsin(3 β + ϕ)⎪b ' = ⎨ bλ − β ≤ ϕ ≤ α⎪ 0ϕ ≥ α⎪⎪⎩(2.25)wherebsin(2 β )α = π − 3 β and tanγ= (2.26)a − bcos(2 β )Again assuming that the incident and transmitted polarizations are coplanar, and then obtain:S =− jkb'( l/ λ)sin(3 β + ϕ)eabS =−jka'( l/ λ)sin(3 β −ϕ)eba'cos(2 β)cos( β ϕ) sin( kb 'cos(2 β )cos( β ϕ))kb 'cos(2 β)cos( β + ϕ)− jkb+ +'cos(2 β)cos( β ϕ) sin( ka 'cos(2 β )cos( β ϕ))ka 'cos(2 β)cos( β −ϕ)− jka− −(2.27)(2.28)<strong>for</strong> E polarization. The results <strong>for</strong> H polarization may be obtained by replacing sin(3 β ± ϕ)by −sin( β m ϕ).The total contribution is the sum <strong>of</strong>Sa,Sb,Saband S ba. In our experiment, a = b = l =0.3 m and 2β = 90o , and then we can calculate the RCS <strong>of</strong> dihedral with different frequencyshown in Figure 2.20.41
Chapter 2(a)(b)(c)(d)(e)Figure 2.20 Calculated RCS patterns <strong>of</strong> a dihedral corner reflector with side length 0.3 m: (a)1 GHz, (b) 2 GHz, (c) 3 GHz, (d) 4 GHz and (e) 5 GHz.42
<strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> and Its Evaluation2.4.2 Polarimetric Calibration Using Dihedral Corner ReflectorsThere are number <strong>of</strong> radar polarimetric calibration algorithms and methods proposed in[70-76]. Moreover, polarimetric <strong>SAR</strong> calibration was discussed by A. Freeman et al. [77-82].We introduce a modified polarimetric calibration method using dihedral corner reflectors.Backscattering ModelGenerally, the RCS <strong>of</strong> a target is polarization dependent. For linear polarizations, a 2 x 2scattering matrix is employed <strong>for</strong> full polarimetric description. The generalized targetscattering matrix is given by [51][ S]⎡ShhShv⎤= ⎢SvhS ⎥⎣ vv ⎦(2.29)In measuring a target scattering matrix, the received signals are determined not only by thedesired target scattering matrix but also by the transmission and reception properties <strong>of</strong> themeasurement system. A monostatic backscatter model is shown in Figure 2.21. The radarpolarimetry backscattering matrix equation is given by:jϕ[ ] [ ] [ ][ ][ ]⎡ ⎤ = + +m⎣S ⎦ I N Ae R S T(2.30)where [S m ] is defined the measured data.[I] is the transmitter and receiver antenna coupling.[N] is the background noise and ground clutter.jAe ϕ is a complex coefficient.[R] is the normalized receiving channel factor.[S] is the target theoretical scattering matrix.[T] is the normalized transmitting channel factor.43
Chapter 2T X R XT XS mITRSR XNFigure 2.21 Backscattering modelCalibration Algorithm and ProcedureFor a monostatic RCS measurement, the receive distortion matrix should be the transpose <strong>of</strong>the transmit one without considering the four measurement channel frequency responseswhich may be nonreciprocal due to the active microwave components or the data acquisitionprocedure. The calibration model separates the antenna polarization effects from the possiblynonreciprocal channel behaviour, which leads to four independent channel parameters. As aresult, no assumption on the transmit/receive channels has to be made and only six instead <strong>of</strong>eight parameters are needed to model a general radar system according to [83].A modified polarimetric RCS calibration technique using two orientations <strong>of</strong> the dihedralcorner reflectors as calibration targets is introduced [83]. This method is valid <strong>for</strong> anymonostatic or quasi-monostatic radar system. There are these aspects to be considered:(a)(b)(c)Scattering measurement <strong>of</strong> two orientations <strong>of</strong> calibration dihedral corner reflectorsare needed;The radiation polarization effect and the four measurement channels are separated inthe model;The frequency responses <strong>of</strong> four measurement channels are modeled as mutuallyindependent and thus, no special care has to be taken <strong>for</strong> signal paths and44
<strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> and Its Evaluationnonreciprocal properties in a measurement system;(d)(e)The calibration procedure is simplified by taking advantage <strong>of</strong> small cross-polarizedterms, and no complicated matrix algebra is involved;The calibration results are valid <strong>for</strong> wideband applications because no assumptionsare made relative to the theoretical solution <strong>for</strong> the calibration dihedral.Here, let us per<strong>for</strong>m the derivation <strong>of</strong> the calibration procedure. For a monostatic backscattercase, we have[ S] [ S] T= (2.31)Substituting [S m ]-[I]-[N] by [M] into Equation (2.30), we havejϕ[ ] [ ] [ ] [ ][ ][ ]m⎡⎣S ⎤− ⎦ I − N = M = Ae R S T(2.32)where [M] indicates the matrix after subtracting background noise and ground clutters andremoving direct coupling by time gating.Since system is reciprocal due to monostatic case, we have[ R] [ T ] T= (2.33)and[ R][ T ]⎡1δ y⎤= ⎢δ x 1 ⎥⎣ ⎦⎡1δ x⎤= ⎢δ y 1 ⎥⎣ ⎦(2.34)(2.35)where δ x and δ y are the normalized cross-polarization components, which account <strong>for</strong>the cross-polarization errors in the target illumination.Equation (2.32) can be rewritten as⎡Mhh Mhv ⎤ 1 SjhhSϕ ⎡ δ y⎤⎡ hv ⎤⎡1δ x⎤⎢ AeMvhM⎥ = ⎢vvδx 1⎥⎢ Svh S⎥⎢ vvδy1⎥⎣ ⎦ ⎣ ⎦⎣ ⎦⎣ ⎦(2.36)45
Chapter 2and⎡MhhChh MhvChv ⎤ ⎡1 δ y⎤⎡Shh Shv⎤⎡1δ x⎤⎢MvhCvh MvvC ⎥=⎢vvδx1 ⎥⎢ Svh S⎥⎢ vvδy1⎥⎣ ⎦ ⎣ ⎦⎣ ⎦⎣ ⎦(2.37)where [C] is reciprocal <strong>of</strong>jAe ϕFrom (2.37), we obtain( δ δ δ δ )( δ δ δ δ )( δ δ δ δ )( δ δ δ δ )−1hh=hh hh+vh+hv+vvM C S yS S y yS y−1hv=hv hh+vh+hv+vvM C S x yS x S yS−1vh=vh hh+vh+hv+vvM C xS S xS y S y−1vv=vv hh+hv+vh+vvM C xS x xS S x S(2.38)We know the theoretical scattering matrices⎡SS(1) (1)(1) hh hv⎡⎣S⎤⎦ = ⎢(1) (1)⎥⎢SvhSvv⎥⎣⎤⎦<strong>for</strong> 0 degree dihedral (2.39)and⎡SS(2) (2)(2) hh hv⎡⎣S⎤⎦ = ⎢(2) (2)⎥⎢SvhSvv⎥⎣⎤⎦<strong>for</strong> 45 degree dihedral (2.40)Hence, two standard dihedral corner reflectors are used <strong>for</strong> calibration measurements asshown in Figure 2.22. Two faces <strong>of</strong> dihedral are square and all side lengths are 0.3 m.46
<strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> and Its Evaluation(a)(b)Figure 2.22 Two calibrators: (a) vertical dihedral, (b) 45 o dihedral.By assumption <strong>of</strong> twosystem, we haveδ s terms neglected and <strong>of</strong> the reciprocity theorem <strong>for</strong> a monostaticSS= S = 0(2.41)(1) (1)hv vh≈ S(2.42)(2) (2)hv vhSubstituting both <strong>of</strong> measured and theoretical values <strong>of</strong> the 0-degree dihedral, we obtainChhS= (2.43)M(1)hh(1)hhCvvS= (2.44)M(1)vv(1)vvAnd using values from 45 degree dihedral, we haveC M = S + δyS + δyS(2.45)(2) (2) (2) (2)hh hh hh hv vhC M = δxS + δxS + S(2.46)(2) (2) (2) (2)vv vv hv vh vvThen, we can obtain the solutions <strong>of</strong> all variables as calibration coefficients. They areEquations (2.43), (2.44), (2.47), (2.48), (2.49), and (2.50).47
Chapter 2C M − Sδ y = (2.47)(2) (2)hh hh hh(2)2SvhC M − Sδ x = (2.48)(2) (2)vv vv vv(2)2Svh(2) −1(2) (2) (2)vh vhδhh hvδvvC = M ( xS + S + yS )(2.49)(2) −1(2) (2) (2)hv hvδhh vhδvvC = M ( xS + S + yS )(2.50)There<strong>for</strong>e, scattering matrix <strong>of</strong> an arbitrary target can be calculated by ( 2.51):[ S]−1 −1hh hv ⎡1 δy⎤ hh hh hv hv ⎡1δx⎤⎡S S ⎤ ⎡M C M C ⎤= ⎢Svh S⎥ = ⎢vvδx 1⎥ ⎢MvhCvh MvvC ⎥⎢ vvδy1⎥⎣ ⎦ ⎣ ⎦ ⎣ ⎦⎣ ⎦(2.51)Two terms <strong>of</strong> cross-polarization errors are shown in Figure 2.23.Figure 2.23 Calibration coefficients <strong>of</strong> two cross-polarization terms.48
<strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> and Its Evaluation2.4.3 Calibration ResultsFrom the theoretical value <strong>of</strong> two calibrators and measured data, we can completely solve theequation group (2.38) to recover the calibration coefficients (2.43)-(2.50). In fact, themeasured data are picked up by time gating from the real measured signals <strong>of</strong> calibratorssubtracted by the background signals. Using the calibration coefficients, the calibratedscattering matrix can be calculated with the aid <strong>of</strong> this equation (2.51). Table 2.03 gives themeasured scattering matrices <strong>of</strong> three standard reflectors and their calibrated scatteringmatrices <strong>of</strong> 2 GHz. We can find the improvements after calibration, especially in the phaseterm. With the use <strong>of</strong> the calibration coefficients, the auto-calibrated results <strong>of</strong> 2 GHz arepresented in Table 2.04 resulting in observable improvements. Although neglecting effects <strong>of</strong>noise in the calibration measurement, the improvements due to calibration have beenachieved.Table 2.03 Comparison <strong>of</strong> calibrated scattering matrices with measured scattering matrices.Measured scattering matrixCalibrated scattering matrixVerticaldihedralSm1oo⎡−j113.7 j129.40.9813 e0.0365e⎤= ⎢ ⎥oj129.4⎢⎣0.0365 e1⎥⎦Sc m1oo⎡−j179.11 −j101.541.0146 e0.0657e⎤= ⎢ ⎥o−j102.39⎢⎣0.0657 e1⎥⎦45 o dihedralSphereSSm2m3o⎡j120.40.2977 e1⎤= ⎢ ⎥oj112.3⎢⎣1 0.2891e⎥⎦o⎡j112.831 0.2331e⎤= ⎢ ⎥oj91.76 j106.83⎢0.0677 e 0.8252e − o⎣⎥⎦SS⎡⎤= ⎢ ⎥⎢⎣0.9904 e0.1285e⎥⎦o−j103.82c 0.1381 e1m 2oo−j 1.43 −j43.20o⎡−j23.111 0.2454e⎤= ⎢ ⎥⎢⎣0.0806 e0.9345e⎥⎦c m 3ooj 32.18 −j21.9649
Chapter 2Table 2.04 Auto-calibrated scattering matrices.Measured scattering matrixCalibrated scattering matrixVerticaldihedralSm1oo⎡−j114.8 −j115.60.9709 e0.0173e⎤= ⎢ ⎥o−j106.4⎢⎣0.0166 e1⎥⎦Sc m1oo⎡−j180.0 −j109.01.0004 e0.0186e⎤= ⎢ ⎥o−j116.6⎢⎣0.0182 e1⎥⎦45 o dihedralSm2o⎡−j91.10.0850 e1⎤= ⎢ ⎥ooj1.4 −j166.7⎢⎣1.0095 e0.0175e⎥⎦Sc m2o⎡j144.50.0001 e1⎤= ⎢ ⎥⎣1 0⎦2.5 SummaryBased on conventional <strong>SAR</strong> principle, a broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong> system isdeveloped. First, the principle <strong>of</strong> ground-<strong>based</strong> <strong>SAR</strong> was discussed. Using ray tracing method,ground-<strong>based</strong> <strong>SAR</strong> data acquisition and imaging were simulated. The estimated azimuthresolutions agreed with the theoretical values. There<strong>for</strong>e, a network analyzer <strong>based</strong>ground-<strong>based</strong> <strong>SAR</strong> system was assembled. Testing results with several standard reflectorsshowed the satisfactory per<strong>for</strong>mance <strong>of</strong> the developed system. Applying two orientationdihedral corner reflectors, a polarimetric calibration method was introduced. In the same time,Per<strong>for</strong>mance <strong>of</strong> a dual polarized diagonal horn antenna used in this system and backscatteringfeatures <strong>of</strong> a metal sphere and dihedral corner reflector were discussed. After calibration,the scattering matrices <strong>of</strong> standard reflectors showed satisfactory agreements with theirtheoretical values. Results demonstrated the system has very good polarimetric and broadbandper<strong>for</strong>mances.50
Chapter 3TEST EXPERIMENTS FOR TREE MONITORINGThis chapter describes two experimental sites and data acquisition procedures by using thedeveloped system. There are tree different types <strong>of</strong> trees in the experimental site, Exp#A, anddata acquisition <strong>for</strong> trees were carried out in spring, summer and late autumn, respectively. Inanother experimental site, Exp#B, data were acquired <strong>for</strong> a cherry tree continually in differentgrowth states: i) being full <strong>of</strong> buds just be<strong>for</strong>e blooming, ii) blooming cherry blossoms, andiii) covered with lush leaves. Then, signal check and verification were per<strong>for</strong>med to showhigh data quality in the third section. It is a basis <strong>for</strong> further data processing andinterpretation.3.1 Description <strong>of</strong> Tree ScatterersBased on the developed broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong> system, we planed to usethis system <strong>for</strong> real measurements in environmental studies [84]. First <strong>of</strong> all, we selected treesas our research objects. The reasons are as follow:(a)(b)(c)(d)Trees are some <strong>of</strong> the most important factors <strong>of</strong> environmental studies, which werementioned in [85-87].Different types <strong>of</strong> trees have different shapes to <strong>for</strong>m different scatterers, <strong>for</strong> example,deciduous tree, evergreen tree, broad-leaved tree and conifer, etc.Trees also change their shapes with different seasons, <strong>for</strong> example, bare trunk andbranches in winter and early spring <strong>for</strong> most trees and plants, being in buds in spring,blooming flowers and growth <strong>of</strong> leaves in late spring and summer, and leaves falling<strong>of</strong>f in autumn.Water content <strong>of</strong> tree and moisture <strong>of</strong> surrounding air also vary with seasons, and51
Chapter 3may be even diurnally during a full day-and-night cycle. That causes differentpermittivity, conductivity and permeability affecting the scattering from thesevegetation scatterers. More so, alive and dead trees display different features too.(e)Description <strong>of</strong> selected trees and ground truth validation.Hence, we selected two experimental sites with different types <strong>of</strong> trees used <strong>for</strong> the broadbandground-<strong>based</strong> polarimetric <strong>SAR</strong> measurements. The experimental sites are both located at theKawauchi Campus, Tohoku University. The experimental site with three different trees iscalled Exp#A. The configuration <strong>of</strong> the target area and the coordinate system are shown inFigure 3.01. The main targets are denoted by T1, T2 and T3, and represented three differentkinds <strong>of</strong> trees. Tree T1 is a Japanese Zelkova (scientific name: Zelkova serrata (Thunb.)Makino), which is a deciduous tree that has no leaf in spring, exuberant broad leaves duringearly summer up to mid-autumn, after then leaves fallen <strong>of</strong>f. Tree T2 is a Japanese cedar(scientific name: Cryptomeria japonica (L. fil.) D. Don), which is an evergreen with needles.It almost does not change from spring to winter. Tree T3 is a kind shrub <strong>of</strong> the Azalea genus(scientific name: Rhododendronquinquefolium Bisset et Moore) which are ashort bush surrounded by some plants:Japanese honeysuckle (scientific name:Lonicera japonica Thunb.). From spring tosummer, they have very dense foliagewhile there are some stems and branchesafter autumn.T1 (-2.85 12.9)Diameter: 0.31YT3 (4.5 8.5)T2 (-0.5 9.8)Diameter: 0.4X-6 m 06 mFigure 3.01 Coordinate system <strong>of</strong> a measurement site: Exp#A.Figure 3.02 shows the leaves <strong>of</strong> three trees. Leaves in Figure 3.02(a) are <strong>for</strong> the deciduoustree T1. The size <strong>of</strong> one leaf is about 10 cm in length and 5 cm in width. Figure 3.02(b) showsthe needled twigs <strong>of</strong> the conifer tree T2. On each stem with length <strong>of</strong> 15 cm, there are number<strong>of</strong> needles with length <strong>of</strong> 1 to 2 cm. There are two main types <strong>of</strong> leaves shown in Figure3.02(c). The dimensions <strong>of</strong> the elliptically shaped leaves are from 2 cm to 6 cm. We havecarried out three measurements at the same exact position <strong>for</strong> the three different trees in latespring (April 19, 2002), in early summer (May 28, 2002) and in late autumn (November 11,2002), respectively. The experimental scenarios are shown in Figure 3.03 <strong>for</strong> the experiment52
Test Experiments <strong>for</strong> Tree MonitoringExp#A-1, Figure 3.04 <strong>for</strong> the experiment Exp#A-2 and Figure 3.05 <strong>for</strong> the ExperimentExp#A-3, respectively. There were a few fresh leaves during the first measurement in spring.Very significant growth in leaves and branches was observed in the second measurement insummer, while the third measurement was at a time when the leaves had fallen <strong>of</strong>f.(a) (b) (c)Figure 3.02 Leaves: (a) broad leaves <strong>of</strong> the tree T1: Japanese Zelkova, (b) needles <strong>of</strong> the treeT2: Japanese cedar and (c) leaves <strong>of</strong> the short tree: Azalea and a plant: Japanesehoneysuckle.Figure 3.03 Target area in the first measurement: Exp#A-1.53
Chapter 3Figure 3.04 Target area in the second measurement: Exp#A-2.Figure 3.05 Target area in the third measurement: Exp#A-3.Another experimental site is called Exp#B. The target is a Yoshino cherry tree (scientificname: Prunus x yedoensis Matsumura), one well-known kind <strong>of</strong> tree in Japan. Cherry treeswill be in buds at mid-spring. Very flourishing flowers are usually blooming in middle Aprilin Sendai, Miyagi-ken <strong>of</strong> the northeast area <strong>of</strong> Japan. The period <strong>of</strong> flowers, known as“hanami”, continues about 7 to 10 days. After then, the flower petals fall <strong>of</strong>f quickly. Soon54
Test Experiments <strong>for</strong> Tree Monitoringthereafter, fresh leaves appear at the bare stems gradually from the beginning <strong>of</strong> May. Verylush leaves cover the whole tree after another 10 days. The coordinate system and location <strong>of</strong>the cherry tree trunk are shown in Figure 3.06. The green color indicates the covering area <strong>of</strong>branches with foliage <strong>of</strong> the cherry tree. The origin locates the center <strong>of</strong> the scanning apertureon the rails. The horizontal axis is X-axis <strong>for</strong> azimuth direction, the Y-axis <strong>for</strong> the rangedirection and the vertical Z-axis <strong>for</strong> elevation, respectively.The dimensions and shapes <strong>of</strong> cherry flowers and leaves are shown in Figure 3.07(a) and (b).The flower petal is about 1 cm in dimension and leaf is about 7 cm in length and 4 cm inwidth. Three measurements were repeated on April 02, 2003 with number <strong>of</strong> buds just be<strong>for</strong>ethe flowers bloom; on April 14, 2003 with many blooming flowers but no leaves; and, on May27, 2003 with substantial growth <strong>of</strong> leaves, respectively. Figure 3.08, Figure 3.09 and Figure3.10 show the experimental scenes, respectively.YCherry(2.1, 15.2 )-6m 06mXFigure 3.06 Coordinate system <strong>for</strong> measurement site Exp#B, a cherry tree.55
Chapter 3(a)(b)Figure 3.07 Flowers and leaves <strong>of</strong> a Yoshino cherry tree.Figure 3.08 A Yoshino cherry tree in the first measurement: Exp#B-1.56
Test Experiments <strong>for</strong> Tree MonitoringFigure 3.09 The cherry tree in the second measurement: Exp#B-2.Figure 3.10 The cherry tree in the third measurement: Exp#B-3.57
Chapter 33.2 Experimental Procedures and Data Collection3.2.1 Measurements <strong>of</strong> Different Tree Types at the Experimental Site A:Exp#AUsing the developed ground-<strong>based</strong> polarimetric <strong>SAR</strong> system, we carried out measurements onseasonal variation <strong>of</strong> tree conditions. The three measurements were carried out on April 19,2002, on May 28, 2002 and on November 11, 2002, respectively. Table 3.01 shows themeasurement conditions and situations <strong>of</strong> targets. Luckily due to protective requirements <strong>of</strong>equipment operation, we successfully found good weather days to per<strong>for</strong>m measurements.Table 3.01 Condition and target situations <strong>for</strong> measurements <strong>of</strong> different tree typesMeasurement numberExp#A-1Exp#A-2Exp#A-3SeasonSpringSummerAutumnWeatherFineFine & light windySunny / cloudy &windyTree 1Japanese ZelkovaBare twigs with budFlourishing largeleavesLeaves almost fall <strong>of</strong>fTree 2Japanese cedarEvergreenEvergreenEvergreenTree 3Azalea&HoneysuckleBush with fresh leavesWith dense plantsWith a few leavesTable 3.02 <strong>of</strong>fers the measurement parameters <strong>for</strong> the monitoring <strong>of</strong> three different trees. Weused the frequency range from 1 GHz to 5 GHz according to the antenna per<strong>for</strong>manceverification in Chapter 2. It almost covers the L-band to C-band. The dual polarized diagonalhorn antenna looks <strong>for</strong>ward to the targets. It possesses the capability <strong>for</strong> measuring fullypolarimetric data sets. In the vertical 2-D scanning aperture, there were 121 measuring pointsfrom -6 meters to 6 meters in the azimuth (X) direction and 16 measuring points from 0.9meter to 2.4 meters in the vertical (Z) direction according to the coordinate system shown inFigure 3.01.58
Test Experiments <strong>for</strong> Tree MonitoringTable 3.02 Parameters <strong>for</strong> measurements <strong>of</strong> different tree typesExperiment Exp#A-1 Exp#A-2 Exp#A-3Date April 19, 2002 May 28, 2002 November 11, 2002Target Three types <strong>of</strong> trees Three types <strong>of</strong> trees Three types <strong>of</strong> treeKawauchi campusKawachi campusKawauchi campusLocationTohoku UniversityTohoku UniversityTohoku UniversityVNA HP 8753E HP 8753E HP 8753EStart frequency [GHz] 1 1 1Stop frequency [GHz] 5 5 5Number <strong>of</strong> points 801 801 1601IF [Hz] 100 100 100Sweep time [sec.] 10 10 25Power [dBm] 10 10 10Antenna look-<strong>for</strong>wardelevation angle [degree] 0 0 0Scan aperture [m] 12 x 1.5 12 x 1.5 12 x 1.5Lowest scan line aboveground [m] 0.9 0.9 0.9Scan interval [m] 0.1 x 0.1 0.1 x 0.1 0.1 x 0.1Compared with the first two experiments Exp#A-1 and Exp#A-2, there were a few changeswith the third experiment <strong>of</strong> Exp#A-3. The leaves <strong>of</strong> the broadleaf tree were falling during thisseason <strong>for</strong> Exp#A-3, but flowers were going to bloom <strong>for</strong> the first round experiment <strong>of</strong>Exp#A-1 on April 19, 2002 and leaves were very exuberant <strong>for</strong> the second round experimentExp#A-2 on May 28, 2002. Most <strong>of</strong> the measuring parameters <strong>of</strong> the third experiment werethe same as the ones <strong>for</strong> the previous measurements but the number <strong>of</strong> point were doubled and59
Chapter 3the sweep time increased in order to obtain a longer detection distance. It can also improve thesignal quality by separating the negative time wave<strong>for</strong>m that is caused by the IFFT to the farend part <strong>of</strong> the time axis. Moreover, the measurement control and data acquisition programshad been modified efficaciously. Though the sweep time increased from 10 seconds to 25seconds, the total measurement time was the same as <strong>for</strong> the <strong>for</strong>mer ones. Table 3.03 lists thedata collection <strong>of</strong> three measurements <strong>for</strong> different trees. The three scattering matrix elementdata <strong>of</strong> HH, VH and VV were acquired simultaneously. A total <strong>of</strong> 1936 points (121 x 16)data <strong>for</strong> each polarization were recorded. There are 31 KB <strong>for</strong> HH and VV data, and 33 KB<strong>for</strong> VH data, total <strong>of</strong> 178 MB data <strong>for</strong> Exp#A-1 and Exp#A-2. There are 62 KB <strong>for</strong> HH andVV data, and 66 KB <strong>for</strong> VH data, total <strong>of</strong> 360 MB <strong>for</strong> Exp#A-3. Each data set has 801 pairnumbers <strong>for</strong> Exp#A-1 and Exp#A-2, and 1601 pair numbers <strong>for</strong> Exp#A-3, where each pairincludes the real part and imaginary parts <strong>for</strong> a frequency.Table 3.03 Data collection <strong>of</strong> measurements <strong>of</strong> different tree typesMeasurementExp#A-1Exp#A-2Exp#A-3PolarizationHH, VH, VVHH, VH, VVHH, VH, VVData <strong>for</strong>matFrequency domainwith real part andimaginary partFrequency domainwith real part andimaginary partFrequency domainwith real part andimaginary partStart frequency [GHz]111Stop frequency [GHz]555Number <strong>of</strong> points8018011601Data size <strong>of</strong> HH [KB]313162Data size <strong>of</strong> VH [KB]333366Data size <strong>of</strong> VV [KB]313162Measurement points121 x 16121 x 16121 x 16Total data size [MB]17817836060
Test Experiments <strong>for</strong> Tree Monitoring3.2.2 Measurements <strong>of</strong> a Cherry Tree at the Experimental Site B: Exp#BUsing the ground-<strong>based</strong> polarimetric broadband <strong>SAR</strong> system, we have carried out threemeasurements <strong>for</strong> a cherry tree. The three measurements were produced on April 02, 2003just be<strong>for</strong>e the flowers bloomed, on April 14, 2003 with very blooming flowers but no leaves,and on May 27, 2003 with all flowers gone and exuberant leaves, respectively. Experimentalconditions and target situation <strong>for</strong> the cherry tree is shown in Table 3.04.Table 3.04 Condition and target situations <strong>for</strong> measurements <strong>of</strong> a cherry treeMeasurement numberExp#B-1Exp#B-2Exp#B-3SeasonSpringLate springEarly summerWeatherFineFineFine & slightly windyYoshino cherry treeBare stems with budsBlooming flowersExuberant leavesIn Table 3.05, the measurement parameters <strong>for</strong> the cherry tree experiments are noted. We usedthe same frequency range same as <strong>for</strong> the <strong>for</strong>mer measurements. The antenna was set at alook-up elevation angle <strong>of</strong> 10 degrees in order to reduce the reflection <strong>of</strong> ground clutter. Thevector network analyzer HP 8720ES was used, which has a wideband frequency range andhigh accuracy. It possesses the capability <strong>for</strong> measuring data <strong>for</strong> the same target at higherfrequency bands. For 2-D scanning, there were 121 measuring points from -6 meters to 6meters in the azimuth (X) direction and 16 measuring points from 0.5 meter to 2.0 meters inthe vertical (Z) direction according to the coordinate system shown in Figure 3.06.61
Chapter 3Table 3.05 Parameters <strong>for</strong> measurements <strong>of</strong> a Yoshino cherry treeExperiment Exp#B-1 Exp#B-2 Exp#B-3Date April 2, 2003 April 14, 2003 May 27, 2003Target A cherry tree A cherry tree A cherry treeKawauchi campusKawachi campusKawauchi campusLocationTohoku UniversityTohoku UniversityTohoku UniversityVNA HP 8720ES HP 8720ES HP 8720ESStart frequency [GHz] 1 1 1Stop frequency [GHz] 5 5 5Number <strong>of</strong> points 1601 1601 1601IF [Hz] 100 100 100Sweep time [sec.] 25 25 25Power [dBm] 5 5 5Antenna look-upelevation angle [degree] 10 10 10Scan aperture [m] 12 x 1.5 12 x 1.5 12 x 1.5Lowest scan line aboveground [m] 0.5 0.5 0.5Scan interval [m] 0.1 x 0.1 0.1 x 0.1 0.1 x 0.1Table 3.06 lists the data <strong>for</strong>mation <strong>for</strong> three measurements <strong>for</strong> a cherry tree. Threepolarimetric scattering matrix element data sets, those <strong>of</strong> HH, VH and VV, were acquiredsimultaneously. A total <strong>of</strong> 1936 data points (121 x 16) <strong>for</strong> each polarimetric element wererecorded. There are 62 KB <strong>for</strong> the HH and VV data sets, and 66 KB <strong>for</strong> the VH data set, atotal <strong>of</strong> 360 MB <strong>for</strong> all three measurements. Each data set has 1601 pair numbers whichincludes the real part and imaginary parts <strong>for</strong> each frequency.62
Test Experiments <strong>for</strong> Tree MonitoringTable 3.06 Data <strong>for</strong>mation <strong>of</strong> measurements <strong>of</strong> a Yoshino cherry treeMeasurementExp#B-1Exp#B-2Exp#B-3PolarizationHH, VH, VVHH, VH, VVHH, VH, VVData <strong>for</strong>matFrequency domainwith real part andimaginary partFrequency domainwith real part andimaginary partFrequency domainwith real part andimaginary partStart frequency [GHz]111Stop frequency [GHz]555Number <strong>of</strong> points160116011601Data size <strong>of</strong> HH [KB]626262Data size <strong>of</strong> VH [KB]666666Data size <strong>of</strong> VV [KB]626262Measurement points121 x 16121 x 16121 x 16Total data size [MB]3563603583.3 Signal Verification3.3.1 Signal Check <strong>for</strong> Four Standard ReflectorsOn May 30, 2002, we had per<strong>for</strong>med the measurement with standard reflectors that were justfollowing the second round experiment on different trees. In this measurement, we placedfour standard reflectors into the target area, shown in Figure 2.17. There were two horizontalsurvey lines scanned by the ground-<strong>based</strong> polarimetric <strong>SAR</strong> system. The purpose <strong>of</strong> themeasurement was to demonstrate the polarimetric per<strong>for</strong>mance <strong>of</strong> the system and check thedata quality. From left to right in Figure 2.17, the four standard reflectors are a -45 degreemetal wire, a vertical metal wire, a 45 degree dihedral corner reflector and a vertical dihedral63
Chapter 3corner reflector. Their configurations are shown in Figure 3.11. The length and diameter <strong>of</strong> the–45 degree metal wire are 2515 mm and 12 mm, and <strong>for</strong> the vertical wire a length <strong>of</strong> 2513mm and a diameter <strong>of</strong> 10 mm, respectively. Two faces <strong>of</strong> both the vertical dihedral and theinclined dihedral are squares. The dimensions <strong>of</strong> the side length are 300 mm. All reflectorswere set at a height <strong>of</strong> 1.5 m.T1 (-2.85 12.9)Diameter: 0.31Y45_dihedralT2 (-0.5 9.8)Diameter: 0.4(1.4 10.2)T3 (4.5 8.5)-45_wire(-2.65 8.88) V_wire(0.55 7.75)V_dihedral(3 6.7)X-6m 06mFigure 3.11 Setup <strong>of</strong> four standard reflectors.The raw data <strong>of</strong> height 1.5 m were processed by band-pass filtering. A 4096 points IFFT wasemployed <strong>for</strong> trans<strong>for</strong>ming the data set to time domain data set. We picked signals <strong>of</strong> fourobservation points which are corresponding to four standard reflectors and show thecorresponding wave<strong>for</strong>ms in Figure 3.12. The black circles indicate the reflections from eachreflector which was identified by the observation position and the signal travel time. Itdemonstrated the accuracy <strong>of</strong> the data sets recorded by the developed system.64
Test Experiments <strong>for</strong> Tree Monitoring(a)(b)(c)(d)Figure 3.12 Reflected signals <strong>of</strong> four standard reflectors: (a) vertical dihedral, (b) 45 odihedral, (c) vertical wire, and (d) -45 o wire.65
Chapter 3There are two windowing functions, one band-pass filter and a low-pass filter, respectively,shown in Figure 3.13. We use these two filters to process one line <strong>of</strong> VV raw data at height1.5 m. Time domain datasets were obtained by IFFT. The band-pass filtered data andlow-pass filtered data are shown in Figure 3.14. We can clearly observe the strongly reflectingsignals from the vertical dihedral reflector, the vertical wire and the trunks <strong>of</strong> tree T2 and treeT3 in both <strong>of</strong> the band-pass filtered wiggle pr<strong>of</strong>ile and the low-pass filtered wiggle pr<strong>of</strong>ile,respectively. The hyperbolas show the diffraction signals <strong>for</strong> the targets identified above.Figure 3.13 Windowing functions.(a)(b)Figure 3.14 Time domain signals in wiggle pr<strong>of</strong>iles: (a) broadband pass filtered and (b) lowpass filtered.66
Test Experiments <strong>for</strong> Tree MonitoringDiffraction stacking has been used to reconstruct the radar image. Both <strong>of</strong> the band-passfiltered migrated images and the low-pass filtered migrated images are shown in Figure 3.15.Targets can be found in both sub-figures. But in low-pass filtered migrated images <strong>of</strong> Figure3.15(b), the trunk <strong>of</strong> tree T2, the vertical wire and vertical dihedral corner reflector can beidentified clearly due to other high frequency sensitive scatterers being removed by low-passfiltering. At the same time, the other two inclined standard reflectors do not appear in the VVimages. This observation agrees with the fundamental polarization theorems. Hence, thedifferent polarimetric scattering per<strong>for</strong>mances <strong>of</strong> standard reflectors had been demonstrated.(a)(b)Figure 3.15 Migrated images: (a) broadband-pass filtered and (b) low-pass filtered.3.3.2 Signal Verification <strong>for</strong> Different Tree TypesFigure 3.16 shows the configuration <strong>of</strong> experimental site Exp#A again. There are threeobservation points: A, B and C noted in the rectangular coordinate system. The coordinates<strong>of</strong> A, B and C are (-3, 2) m, (-0.3, 2) m and (4.5, 2) m, respectively. Figure 3.17, Figure 3.18and Figure 3.19 show the signal wave<strong>for</strong>ms corresponding to point A, B and C. In each figure,signals <strong>of</strong> the HH, VH and VV <strong>of</strong> three experiments are drawn simultaneously. We canobserve that most wave<strong>for</strong>ms are similar but slight differences exist among differentpolarimetric scattering matrix elements and experiments. Based on those phenomena, which67
Chapter 3can be explained, the high quality <strong>of</strong> acquired data is demonstrated.We picked signals <strong>of</strong> three observation points A, B and C indicated in Figure 3.16.T1 (-2.85 12.9)Diameter: 0.31T2 (-0.5 9.8)Diameter: 0.4YT3 (4.5 8.5)A B CX-6 m 06 mFigure 3.16 Three observation points A(-3, 2), B(-0.3, 2) and C(4.5, 2).68
Test Experiments <strong>for</strong> Tree Monitoring(a)(b)(c)Figure 3.17 Wave<strong>for</strong>ms <strong>of</strong> point A -T1: (a) HH, (b) VH, and (c) VV.69
Chapter 3(a)(b)(c)Figure 3.18 Wave<strong>for</strong>ms <strong>of</strong> point B -T2: (a) HH, (b) VH, and (c) VV.70
Test Experiments <strong>for</strong> Tree Monitoring(a)(b)(c)Figure 3.19 Wave<strong>for</strong>ms <strong>of</strong> point C -T3: (a) HH, (b) VH, and (c) VV.71
Chapter 33.3.3 Signals Separated by Different FiltersBecause there were different situations with T1 during three measurements, we are focusingon point A to analyze the raw signal by different frequency bandwidth filters. The windowingfunctions <strong>of</strong> filters are shown in Figure 3.20. Filter BP is a broadband-pass filter withfrequency region covering almost the entire measured frequency band from 1.2 GHz to 4.8GHz. The LP is a low-band-pass filter from 1 GHz to 2 GHz; the bandwidth <strong>of</strong> MP is from 2GHz to 4 GHz; the HP is a high pass filter from 4 GHz to 5 GHz. All <strong>of</strong> them take the <strong>for</strong>ms<strong>of</strong>2F( f) = sin ( fπ/ B)(3.01)where B is the bandwidth.Figure 3.21, Figure 3.22 and Figure 3.23 show the signal wave<strong>for</strong>ms corresponding to point A<strong>of</strong> HH, VH and VV, respectively. In each sub-figure, wave<strong>for</strong>ms filtered by BP, LP, MP andHP <strong>for</strong> three experiments are plotted simultaneously. We can find that most <strong>of</strong> the wave<strong>for</strong>msare similar, and only slight differences exist among different polarimetric returns andexperiments in the sub-figure processed <strong>for</strong> BP. But other wave<strong>for</strong>ms divided by differentfrequency bands show quit varying shapes. For instance, in sub-figures (a), (b) and (c) <strong>of</strong>Figure 3.21 and Figure 3.23, there are two main reflecting areas between 60 ns –90 ns. Theone that appeared around 68 ns shows that the reflection <strong>for</strong> low frequency is stronger that <strong>for</strong>high frequency. The other one appeared around 85 ns shows a kind <strong>of</strong> high frequencydependence. We can assume that the low frequency dependent part is the reflection from thethick tree T2 and the high frequency dependent component comes from the thinner trunk andbranches <strong>of</strong> T1. Moreover, Figure 3.22(b) displays greater reflection after frequency partition,especially at higher frequencies. We can assume the reason is that there were a number <strong>of</strong>exuberant leaves and branches in experiment Exp#A-2.According to these phenomena, it can be demonstrated that the data sets which have beenacquired not only display polarimetric in<strong>for</strong>mation but also contain in<strong>for</strong>mation on frequencydependence, which depicts additional characteristic in <strong>for</strong>mation on the scattering sizes. Thereis the potential to obtain polarimetric images with different frequency divisions <strong>for</strong> a larger set<strong>of</strong> different plants <strong>of</strong> varying shapes and sizes and <strong>for</strong> variant seasons.72
Test Experiments <strong>for</strong> Tree Monitoring(a)Figure 3.20 Four filters: (a) broadband filter, (b) low-pass, middle-pass and high-pass filters.(b)73
Chapter 3(a)(b)(c)Figure 3.21 HH signals filtered by different filters: (a) Exp#A-1, (b) Exp#A-2, and (c)Exp#A-3.74
Test Experiments <strong>for</strong> Tree Monitoring(a)(b)(c)Figure 3.22 VH signals filtered by different filters: (a) Exp#A-1, (b) Exp#A-2, and (c)Exp#A-3.75
Chapter 3(a)(b)(c)Figure 3.23 VV signals filtered by different filter: (a) Exp#A-1, (b) Exp#A-2, and (c) Exp#A-3.76
Test Experiments <strong>for</strong> Tree Monitoring3.3.4 Signal Wave<strong>for</strong>m Comparison <strong>for</strong> a Cherry TreeUsing the ground-<strong>based</strong> polarimetric broadband <strong>SAR</strong> system, we have carried out threemeasurements <strong>for</strong> a cherry tree. We have chosen data sets <strong>for</strong> several measuring points fromthree measurements to show the data quality.Two observation points were selected in Figure 3.24. At point M, the antenna looked into thefront <strong>of</strong> the upper part with some horizontal branches. Point N was at the front <strong>of</strong> the trunk <strong>of</strong>the cherry tree. The coordinate system is illustrated in Figure 3.06. By using signal processingmethods, signal wave<strong>for</strong>ms <strong>of</strong> observation point M are shown in Figure 3.25. Fromtime-domain signal wave<strong>for</strong>ms, we can observe that slight differences appear in HH and VHsignals. For VV signal wave<strong>for</strong>ms, there are no obvious changes, and the reflections from thetrunk that appeared at 104 ns were identified. We could not find the vertical trunk thatappeared in HH signals. But at point N, the big reflections <strong>of</strong> the trunk appeared at 102 ns inboth HH and VV signals due to the fact that the observation point N is on precisely at thefront <strong>of</strong> the trunk. Changes appeared also in VH signal wave<strong>for</strong>ms.Point MPoint NFigure 3.24 Two observation points: M(-0.6, 2)m and N(2,1.6)m.77
Chapter 3(a)(b)(c)TrunkFigure 3.25 Wave<strong>for</strong>ms <strong>of</strong> the cherry tree at point M (-0.6, 2) m: (a) HH, (b) VH, and (c) VV.78
Test Experiments <strong>for</strong> Tree Monitoring(a)Trunk(b)(c)TrunkFigure 3.26 Wave<strong>for</strong>ms <strong>of</strong> the cherry tree at point N (2, 1.6) m: (a) HH, (b) VH, and (c) VV.79
Chapter 33.4 SummaryTarget determination and procedures <strong>of</strong> measurements on two experimental sites weredescribed in this chapter.Using the developed broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong>system, continued data acquisitions were carried out <strong>for</strong> three different types <strong>of</strong> trees in threeseasons, and a cherry tree during different growth states at two experimental sites. For eachmeasurement, HH, VH and VV polarization data were acquired simultaneously. With few <strong>of</strong>signal verifications, good data quality has been demonstrated. Acquired data not only includein<strong>for</strong>mation about polarimetric per<strong>for</strong>mances <strong>for</strong> the targets but also imply frequencydependence <strong>of</strong> size-dependent scattering mechanisms by signal comparison, which cannoteasily be demonstrated by means <strong>of</strong> straight-<strong>for</strong>ward implementation <strong>of</strong> either <strong>SAR</strong>polarimetry, <strong>SAR</strong> interferometry and/or polarimetric <strong>SAR</strong> interferometry, and thus addconsiderably to multimodal <strong>SAR</strong> image analysis.80
Chapter 4THREE DIMENSIONAL IMAGE RECONSTRUCTIONThe content <strong>of</strong> this chapter is to introduce the signal processing approach <strong>for</strong> the broadbandground-<strong>based</strong> polarimetric <strong>SAR</strong> data analysis. Following a signal processing flowchart,several advanced signal processing algorithms and their advantages are described, <strong>for</strong> instance,time gating, matched filtering, migration <strong>of</strong> diffraction stacking. Then 3-D spatial domain dataare synthesized by migration and 3-D images are reconstructed. Using the ground-truths, 3-Dimages <strong>for</strong> different seasons are compared and demonstrated using the correspondingalgorithms and image data sets.4.1 Data Processing4.1.1 Signal Processing FlowchartFor the acquired broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong> data, a series <strong>of</strong> different signalprocessing methods are applied. The processing flow chart is shown in Figure 4.01. Due todifferent data structure, ground-<strong>based</strong> <strong>SAR</strong> signal processing is not the same as <strong>for</strong> airborne<strong>SAR</strong> data processing [88, 91]. First, the acquired frequency domain data are filtered by thebroadband-pass filter; then trans<strong>for</strong>med to the time domain by IFFT [89]. By selecting theinteresting range at time axis, time gating is employed to the time domain data; thenback-trans<strong>for</strong>med to the frequency domain by FFT. The frequency domain data are calibratedby radar system calibration coefficients. Thereafter, the calibrated data are compressed bymatched filtering, in which the reference signal was the reflection from an aluminum platemeasured by the same radar system and parameters. After using the IFFT again, the timedomain data are obtained. Finally, diffraction stacking is applied to migrate the time domaindata while antenna radiation directivity compensation is done synchronously andsimultaneously; and the special domain data sets are obtained.81
Chapter 4Based on the spatial domain data, different polarimetric target descriptors and interpretationtechniques are applied <strong>for</strong> ground-<strong>based</strong> <strong>SAR</strong> data sets. These will be discussed later on indetail.Acquired frequency domain dataFiltered frequency domain dataTime domain dataTime domain data after time gatingFrequency domain dataFrequency domain data after matched filterTime domain dataSpatial domain databroadband pass filteringIFFTtime gating <strong>for</strong> removingantenna couplingFFTcalibration &matched filteringIFFTmigration with antennadirectivity compensation3D imageSpatial-frequency domain dataSingle-frequency Multi-frequency RGB imageSTFT or FFTPolarimetric analysisFigure 4.01 Signal processing flowchart.4.1.2 Time GatingWhen seismic researchers desire to window data, they <strong>of</strong>ten want to do time gating. Timegating is a very useful spatial filtering process <strong>for</strong> reducing antenna direct coupling and noisebeyond the region <strong>of</strong> interest. It uses a windowing function to filter a time domain signal and82
Three Dimensional Image Reconstructionpicks up the signal <strong>of</strong> interest only. The basic <strong>for</strong>mulation <strong>of</strong> time gating is shown in Equation(4.01).() () ( )f t = f t ⋅ w ζ(4.01)0f0() t is a original time domain signal, and w( ζ ) is a windowing function. f () t is thegated signal.An option <strong>for</strong> windowing (time gating) looks like this:w( ζ )2⎧ 1⎛tmin−ζ⎞− ⎜ ⎟2 ρ⎪ ⎝ ⎠e ζ < tmin⎪⎪= ⎨⎪1tmin≤ζ≤t2⎪ 1⎛ζ−tmax⎞− ⎜ ⎟⎪ 2⎝ρ ⎠⎩eζ > tmaxmax(4.02)where:tmin is the min time to pass.tmin is the max time to pass.ρ is the slope coefficient <strong>of</strong> window shape.During the data processing procedure <strong>of</strong> broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong>, timegating is used in several places, <strong>for</strong> instance, during calibration, signal preprocess and pulsecompression.4.1.3 Pulse Compression and Matched FilterThe matched filter is employed to carry out the pulse compression. The concept <strong>of</strong> a matchedfilter is a very general concept, common to many aspects <strong>of</strong> radar and other types <strong>of</strong> signalanalysis [91]. A filter matched, as defined below, to a given input signal can be shown to be an83
Chapter 4optimum filter <strong>for</strong> signal reception when the received signal is corrupted by additive whiteGaussian noise [92]. That means matched filtering extracts a known signal from a noisychannel. The noisy signal is cross-correlated with the signal that is being detected. Thiscross-correlation produces a zero-phase wavelet (autocorrelation) with its central peak at thezero-time <strong>of</strong> the wavelet, and with a maximum signal to noise ratio (SNR).A filter matched to an input signal si( t ) with spectrum Si( f ) is defined in terms <strong>of</strong> thematched-filter transfer function H(f) and the corresponding impulse response function h(t) asfollows.H( f ) KS ( f ) e= (4.03)* − j 2πftiandht () = Ks( τ − t)(4.04)iWhere K is the fixed constant <strong>of</strong> net gain through the filter and τ is the fixed component <strong>of</strong>delay through the filter and H(f) is the Fourier trans<strong>for</strong>m <strong>of</strong> h(t). The asterisk refers to thecomplex conjugate <strong>for</strong>m. The algorithm <strong>of</strong> the matched filter is shown in Figure 4.02.Raw signalTime domainfs( t )Frequency domainFs( ω )Reference signalf ( t ) F ( ω)refrefFs∗( ω) ⋅ F ( ω)refPulse compressionf ( t ) F ( ω )Figure 4.02 Algorithm <strong>of</strong> matched filtering.84
Three Dimensional Image ReconstructionMatched-filter processing <strong>of</strong> a target echo signal can be thought <strong>of</strong> as a coherent summation<strong>of</strong> the reflected signal energy from each <strong>of</strong> the target’s reflection points. The processedresponse <strong>for</strong> the entire target is the phasor summation <strong>of</strong> the individual matched-filterresponses <strong>for</strong> all <strong>of</strong> the target’s reflection points, which are generally spread in range over thetarget’s range extent. Figure 4.04 shows an example <strong>of</strong> pulse compression by the matchedfilter approach, where the reference signal was the reflection from an aluminum platemeasured by the same radar system and parameters shown in Figure 4.03.Figure 4.03 Reference reflector <strong>of</strong> an aluminum plate: 1m x 2m.(a)amplitude(b)(c)time [ns]Figure 4.04 Pulse compression: (a) reference signal, (b) raw signal, and (c) compressedsignal.85
Chapter 44.2 Image Reconstruction4.2.1 Algorithm <strong>of</strong> 3-D ImagingA three-dimensional <strong>SAR</strong> image can be reconstructed by synthesizing the data set acquiredwith the two-dimensional aperture. Migrations are <strong>of</strong>ten used <strong>for</strong> image reconstruction bywave field extrapolation, which were discussed by many authors in [93-103]. The goal <strong>of</strong>migration is to make the stacked section appear similar to the real location. A migrationmethod <strong>of</strong> diffraction stacking has been employed to reconstruct the image [46, 93]. It is avery convenient and traditional method, which is commonly used <strong>for</strong> seismic data processing.This approach is an integral solution <strong>of</strong> the wave equation. It can extrapolate wave fields <strong>of</strong> anobservation aperture to larger 3-D space. Diffraction stacking migration produces onemigrated sample (a pixel) at a time by computing a diffraction shape <strong>for</strong> a scatter point at thatlocation, summing and weighting the input energy along a diffraction path, and placing thesummed energy at the scatter point location on the migrated section. The process is repeated<strong>for</strong> all migrated samples. During summation, the amplitudes <strong>of</strong> the input data are weighted.This migration algorithm is to focus the time domain scattering signal on the target in spaceby computing (4.05).( ) ( τ ) ( θ)P x, y, z = ∫∫ f , xm, ym, zm |y m = 0A dzmdxm(4.05)where P( x, y,z ) is the imaging area.( , , , )f τ x y z defines the acquired data in time domain.m m m( x − x) + ( y − y) + ( z − z) + ( x − x) + ( y − y) + ( z −z)2 2 2 2 2 2t t t r r rτ = (4.06)vv is the electromagnetic wave propagating speed in media.( , , )Tx x y z is the transmitting antenna position.t t t86
( , , )Rx x y z is the receiving antenna position.r r rThree Dimensional Image ReconstructionA( θ ) is the antenna directivity compensation coefficient.θ is the compensation angle where max value is taken a half <strong>of</strong> antenna beam width.These variables are described in Figure 4.05. In this algorithm, antenna radiation patterncompensation is included. It is used to weight the time domain data. It makes the datasynthesis consistent with the real conditions <strong>of</strong> the radar system measurement.ZP(x, y, z)YRx(x r y r z r )Tx(x t y t z t )0ScanningapertureXFigure 4.05 Geometry <strong>of</strong> image reconstruction by diffraction stacking with antenna directivitycompensation.4.2.2 3-D Image Reconstruction3-D Imaging <strong>of</strong> Three Different TreesBased on measurements at experimental site Exp#A, we have obtained three data sets87
Chapter 4including HH, VH and VV polarization data. Following the signal processing flowchart, eachpolarization data set is processed separately. First, acquired frequency domain data are filteredby a broadband-pass filter which is shown in Figure 4.06. The low frequency is 1.2 GHZ topass and the high frequency is 4.8 GHz to pass in order to keep the broadband in<strong>for</strong>mationand also smoothen the data to reduce the resonances <strong>of</strong> the time domain signal caused byinverse Fourier trans<strong>for</strong>ming. After processing by inverse Fourier trans<strong>for</strong>m, time gating isemployed to the time domain data. Due to the fact that the main targets are here located inrange <strong>of</strong> 6 meters to 14 meters, the min time <strong>of</strong> 30 ns and max time <strong>of</strong> 120 ns are selected <strong>for</strong>time gating. It can remove the antenna direct coupling, near range ground reflections andeffects from far range reflections.Figure 4.06 Shape <strong>of</strong> a band-pass filter.Then time domain data are back-trans<strong>for</strong>med to the frequency domain by FFT. The frequencydomain data are calibrated by radar system calibration coefficients which were calculatedduring system per<strong>for</strong>mance evolution in Chapter 2. After then, the calibrated data arecompressed by a matched filter, in which the reference signal was the reflection from analuminum plate measured by the same radar system and parameters shown in Figure 4.03.After using the inverse FFT again, the time domain data are obtained. Migration <strong>of</strong> diffraction88
Three Dimensional Image Reconstructionstacking is applied to migrate the time domain data while antenna radiation directivitycompensation is done simultaneously, <strong>for</strong> which compensation angle is 30 degrees; and thespecial domain data are obtained. Here the spatial domain resolutions are set as 0.05 m x 0.05m x 0.05 m <strong>for</strong> diffraction stacking calculations. The spatial region <strong>of</strong> the reconstructedimages is –6 m to 6 m in azimuth (X) direction, 5 m to 15 m in range (Y) direction and 0 m to6 m in vertical (Z) direction according to the coordinate system in Figure 3.01.There<strong>for</strong>e, using the 3-D spatial domain data, the 3-D visualization images are obtained.Figure 4.07, Figure 4.08 and Figure 4.09 give the 3-D reconstructed images <strong>of</strong> the HH, VHand VV polarimetric returns <strong>for</strong> three experiments, respectively.Analyzing the three polarimetric images <strong>of</strong> each measurement, we can find differences amongthe different polarimetric returns. The HH image shows the reflections from trunks, somehorizontal branches and ground clutter, the VH image indicates strong reflections from leavesand some slant branches, and the VV image shows reflections from vertical trunks andbranches.Those rectangle box and circles marked in 3-D images show good polarimetriccorrespondences with components <strong>of</strong> trees. Blue rectangle boxes and circles show the image<strong>of</strong> tree T1, yellow rectangle boxes and circle show the image <strong>of</strong> tree T2, and red circles showthe image <strong>of</strong> tree T3 in Figure 4.07, Figure 4.08 and Figure 4.09, respectively.89
Chapter 4height [m]range [m]azimuth(a) amplitude: 5.5x10 -3height [m]range [m]azimuth(b) amplitude: 1.2x10 -3height [m]range [m]azimuth(c) amplitude: 5.5x10 -3Figure 4.07 3-D polarimetric images <strong>of</strong> trees in spring - Exp#A-1: (a) HH, (b) VH, and (c) VV.90
Three Dimensional Image Reconstructionheight [m]range [m]azimuth(a) amplitude: 5.5x10 -3height [m]range [m]azimuth(b) amplitude: 1.2x10 -3height [m]range [m]azimuth(c) amplitude: 5.5x10 -3Figure 4.08 3-D polarimetric images <strong>of</strong> trees in summer-Exp#A-2: (a) HH, (b) VH and (c) VV.91
Chapter 4height [m]range [m]azimuth(a) amplitude: 5.5x10 -3height [m]range [m]azimuth(b) amplitude: 1.2x10 -3height [m]range [m]azimuth(c) amplitude: 5.5x10 -3Figure 4.09 3-D polarimetric images <strong>of</strong> trees in autumn-Exp#A-3: (a) HH, (b) VH and (c) VV.92
Three Dimensional Image Reconstruction3-D Imaging <strong>of</strong> a Cherry TreeVery similar data processing was done to acquire data <strong>for</strong> a cherry tree at experimental siteExp#B. For a clear understanding, the description <strong>of</strong> the signal processing procedures isrepeated by emphasizing the different terms. Using the HH, VH and VV polarimetric data sets,each polarimetric data set is processing separately. Firstly, acquired frequency domain dataare filtered by the same broadband-pass filter shown in Figure 4.06. The low frequency <strong>of</strong> 1.2GHZ and upper frequency <strong>of</strong> 4.8 GHz are chosen <strong>for</strong> smoothing the data in order to reducethe resonances <strong>of</strong> time domain signal caused by inverse Fourier trans<strong>for</strong>ming. Afterprocessing by inverse Fourier trans<strong>for</strong>ming, time gating is employed to the time domain data.Due to fact that the cherry tree is located at a range <strong>of</strong> 10 meters to 20 meters, hence, theminimum time <strong>of</strong> 60 ns and maximum time <strong>of</strong> 150 ns are used <strong>for</strong> time gating. It can removethe antenna direct coupling, near range ground reflections and effects from far rangereflections.Then time domain data are back-trans<strong>for</strong>med to frequency domain by FFT. The frequencydomain data are calibrated by radar system calibration coefficients. Thereafter, the calibrateddata are compressed by a matched filter, in which the reference signal was the reflection froman aluminum plate, which was set apart from antenna by about 15 meters measured by thesame radar system and parameters.After using the inverse FFT again, the time domain data are obtained. Due to 15 degreeslook-up <strong>for</strong> the antenna setup, the slant range signals must be converted to ground range data.Then diffraction stacking is applied to focus the time domain diffraction signals to the scatterposition with antenna radiation directivity compensation. Here the spatial domain resolutionsare the same with 0.05 m x 0.05 m x 0.05 m <strong>for</strong> diffraction stacking calculations. The spatialregion <strong>of</strong> the reconstructed images is –6 m to 6 m in azimuth (X) direction, 10 m to 20 m inrange (Y) direction and 0 m to 8 m in vertical (Z) direction according to the coordinate systemin Figure 3.07.Now, the 3-D visualized images are obtained by 3-D spatial domain data. Figure 4.10, Figure4.11 and Figure 4.12 show the 3-D reconstructed images <strong>of</strong> the HH, VH and VV polarimetricreturns <strong>for</strong> three experiments, respectively.93
Chapter 4height [m]range [m]azimuth [m](a) amplitude: 6.5x10 -3height [m]range [m]azimuth [m](b) amplitude: 1.8x10 -3height [m]range [m]azimuth [m](c) amplitude: 6.5x10 -3Figure 4.10 3-D images <strong>of</strong> a cherry tree in Exp#B-1: (a) HH, (b) VH, and (c) VV.94
Three Dimensional Image Reconstructionheight [m]range [m]azimuth [m](a) amplitude: 6.5x10 -3height [m]range [m]azimuth [m](b) amplitude: 1.8x10 -3height [m]range [m]azimuth [m](c) amplitude: 6.5x10 -3Figure 4.11 3-D images <strong>of</strong> a cherry tree in Exp#B-2: (a) HH, (b) VH, and (c) VV.95
Chapter 4height [m]range [m]azimuth [m](a) amplitude: 6.5x10 -3height [m]range [m]azimuth [m](b) amplitude: 1.8x10 -3height [m]range [m]azimuth [m](c) amplitude: 6.5x10 -3Figure 4.12 3-D images <strong>of</strong> a cherry tree in Exp#B-3: (a) HH, (b) VH, and (c) VV.96
Three Dimensional Image Reconstruction4.3 Verification Using <strong>Ground</strong> Truths4.3.1 Verification <strong>of</strong> Three Different TreesNow we put the same polarization images together <strong>for</strong> the three measurements. 3-D HHimages <strong>of</strong> trees are shown in Figure 4.14, and VH images in Figure 4.15, VV images in Figure4.16. There are only a few <strong>of</strong> differences among 3 HH images. This indicates that the maintargets have not suffered many changes. But <strong>for</strong> the VH images, the obvious difference exists.Especially in summer, there are strong reflections from close-up regions around T3, and als<strong>of</strong>or T1 and T2. We can assume that those changes are caused by volume scattering resultingfrom spots with a high number <strong>of</strong> inclined twigs and branches. The extent <strong>of</strong> the foliagevolumes differs with varying seasons. We can find the difference between images <strong>of</strong> Exp#A-1and Exp#A-2, then we can assume that the water content inside these scatterers are differentbetween spring and autumn although the volumes <strong>of</strong> targets are very similar. In VV images,there are not so many differences because there were not obvious changes from spring toautumn <strong>for</strong> trunks.Comparing the corresponding photographs in Figure 4.13, we find that the reflections fromthe trunks and the leaves are different and demonstrate seasonal variation. There are strongerreflections in the images <strong>of</strong> Exp#A-2 due to the lush growth <strong>of</strong> branches and leaves in summer.Moreover, the clearer ground surface reflection can readily be seen in Figure 4.14(c), Figure4.15(c) and Figure 4.16(c) in autumn, which agrees well with the real situation. Thesedifferences also exist between Figure 4.14(a) and Figure 4.14(c), and between Figure 4.16(a)and Figure 4.16(c). Hence, we can assume that the main reflections were caused by thevolume scattering due to the different volume distribution <strong>of</strong> leaves, twigs and small branches<strong>of</strong> different shape and different electric properties during the different seasons. Although theshapes <strong>of</strong> the trees were very similar in late spring and late autumn, the water content <strong>of</strong> thetrees in late spring was higher than that in autumn. This fact caused the slight differencesbetween the reconstructed polarimetric images in late spring and late autumn.97
Chapter 4On April 19, 2002T1T2T3(a) Exp#A-1On May 28, 2002T1T2T3(b) Exp#A-2On Nov. 11, 2002T1T2T3(c) Exp#A-3Figure 4.13 Experimental scenes <strong>of</strong> trees in different seasons: (a) Exp#A-1: spring, (b)Exp#A-2: summer, and (c) Exp#A-3: autumn.98
Three Dimensional Image Reconstructionheight [m]range [m]azimuth(a) amplitude: 5.5x10 -3height [m]range [m]azimuth(b) amplitude: 5.5x10 -3height [m]range [m]azimuth(c) amplitude: 5.5x10 -3Figure 4.14 3-D HH images <strong>of</strong> trees in: (a) Exp#A-1: spring, (b) Exp#A-2: summer, and (c)Exp#A-3: autumn.99
Chapter 4height [m]range [m]azimuth(a) amplitude: 1.2x10 -3height [m]range [m]azimuth(b) amplitude: 1.2x10 -3height [m]range [m]azimuth(c) amplitude: 1.2x10 -3Figure 4.15 3-D VH images <strong>of</strong> trees in: (a) Exp#A-1: spring, (b) Exp#A-2: summer, and (c)Exp#A-3: autumn.100
Three Dimensional Image Reconstructionheight [m]range [m]azimuth(a) amplitude: 5.5x10 -3height [m]range [m]azimuth(b) amplitude: 5.5x10 -3height [m]range [m]azimuth(c) amplitude: 5.5x10 -3Figure 4.16 3-D VV images <strong>of</strong> trees in: (a) Exp#A-1: spring, (b) Exp#A-2: summer, and (c)Exp#A-3: autumn.101
Chapter 4The different scattering per<strong>for</strong>mances <strong>of</strong> the different trees in the different seasons can beobserved clearly, and more measurements <strong>for</strong> these and other vegetation structures would bedesirable in order to more clearly distinguish the associated scattering mechanisms. Todemonstrate the detailed features <strong>of</strong> vegetations by broadband ground-<strong>based</strong> <strong>SAR</strong> system, theadditional data processing algorithms have to be used to expose the data we have acquired inChapter 5.4.3.2 Verification <strong>of</strong> a Cherry TreeSame polarization images <strong>of</strong> the cherry tree are put together again. Figure 4.18 shows the 3-Dreconstructed images <strong>of</strong> HH <strong>of</strong> the cherry tree. The main corresponding parts <strong>of</strong> threemeasurements are very similar. But the different reflections from the trunk and the branchesare observed during different conditions. Comparing the corresponding measurement scenesin Figure 4.17, larger reflection areas appear in Figure 4.18(b) image and Figure 4.18(c)image.Figure 4.19 shows the 3-D reconstructed images <strong>of</strong> VH components. Comparing thecorresponding photographs again, the differences among three measurements are obvious. Forinstance, because there were many blooming flowers, larger reflection areas appear in Figure4.19(b) and more strong reflection areas appears in Figure 4.19(c). Moreover, there weremany tall grown grasses in the second and third measurements. Hence, we can find enhancedreflections in Figure 4.19(b) and Figure 4.19(c), too. We can assume that the main reflectionswere caused by the volume scattering due to the different voluminous distribution, differentshape and different water content <strong>of</strong> cherry flowers and lush leaves during the differentmeasurements. We also found the volume <strong>of</strong> the whole tree was larger than in the <strong>for</strong>mer twocases and the water moisture content was higher two.Very similar images <strong>for</strong> VV are shown in Figure 4.20. There were not many differencesobserved <strong>for</strong> the vertical polarimetric returns <strong>of</strong> the cherry tree in three quite differentconditions.102
Three Dimensional Image Reconstruction(a) Exp#B-1(b) Exp#B-2(c) Exp#B-3Figure 4.17 Experimental scenes <strong>of</strong> a cherry tree at different states: (a) Exp#B-1: buds, (b)Exp#B-2: flowers, and (c) Exp#B-3: leaves.103
Chapter 4height [m]range [m]azimuth [m](a) amplitude: 6.5x10 -3height [m]range [m]azimuth [m](b) amplitude: 6.5x10 -3height [m]range [m]azimuth [m](c) amplitude: 6.5x10 -3Figure 4.18 3-D HH images <strong>of</strong> a cherry tree in: (a) Exp#B-1: buds, (b) Exp#B-2: flowers, and(c) Exp#B-3: leaves.104
Three Dimensional Image Reconstructionheight [m]range [m]azimuth [m](a) amplitude: 1.8x10 -3height [m]range [m]azimuth [m](b) amplitude: 1.8x10 -3height [m]range [m]azimuth [m](c) amplitude: 1.8x10 -3Figure 4.19 3-D VH images <strong>of</strong> a cherry tree in: (a) Exp#B-1: buds, (b) Exp#B-2: flowers, and(c) Exp#B-3: leaves.105
Chapter 4height [m]range [m]azimuth [m](a) amplitude: 6.5x10 -3height [m]range [m]azimuth [m](b) amplitude: 6.5x10 -3height [m]range [m]azimuth [m](c) amplitude: 6.5x10 -3Figure 4.20 3-D VV images <strong>of</strong> a cherry tree in: (a) Exp#B-1: buds, (b) Exp#B-2: flowers, and(c) Exp#B-3: leaves.106
Three Dimensional Image ReconstructionWe find that the polarimetric broadband <strong>SAR</strong> technique provides a great amount <strong>of</strong>in<strong>for</strong>mation about vegetation scatter, and that the different scattering features <strong>of</strong> the cherrytree during the different seasons can be detected rather well by the polarimetric measurement.4.4 SummaryThis chapter was to review signal processing methods <strong>for</strong> broadband ground-<strong>based</strong>polarimetric <strong>SAR</strong> data. Algorithms <strong>of</strong> time gating, matched filtering and diffraction stackingwere described. It was demonstrated that time gating could remove antenna direct couplingand reduce reflections <strong>of</strong> ground clutter. Migration by diffraction stacking with antennadirectivity compensation is a very important technique <strong>for</strong> synthesizing 2-D scanning data. Italso enables easily to extrapolate wave fields in a limited aperture to a large 3-D space andprovides excellent reconstructed 3-D images.Reconstructed 3-D images showed satisfactory effects to monitoring trees by the developedbroadband ground-<strong>based</strong> polarimetric <strong>SAR</strong> system. They showed good consistency with theground truths. Radar polarimetry provided valuable scattering in<strong>for</strong>mation about targets sothat the different components <strong>of</strong> trees in different periods could be distinguished [104].107
Chapter 4108
Chapter 5DATA INTERPRETATION WITH<strong>SAR</strong> POLARIMETRIC ANALYSISWith radar polarimetry, the textural fine-structure, target-orientation and shape, symmetriesand material constituents <strong>of</strong> the Earth surface can be recovered with considerableimprovements above that <strong>of</strong> standard amplitude-only radar. By combining differentpolarisation states, it has been shown that important structural in<strong>for</strong>mation relating tovegetation height and density can be obtained [2].In this chapter, several target polarimetric description techniques are used <strong>for</strong> interpreting theground-<strong>based</strong> <strong>SAR</strong> data. Broadband and multi-frequency polarimetric <strong>SAR</strong> data hold thepromise <strong>of</strong> providing valuable in<strong>for</strong>mation about distributed targets. Many differentapproaches to model the expected responses have been published, and several papers havedescribed the phenomenology <strong>of</strong> multi-frequency polarimetric data. There are differentways to view polarimetric images <strong>of</strong> multi-parameter <strong>SAR</strong> data, including polarimetric colorimages in different bases, as well as broadband images.5.1 Broadband Frequency Polarimetric Interpretations5.1.1 Enhancements <strong>of</strong> AmplitudeApplying the previous processed spatial domain data <strong>of</strong> Exp#A-1 <strong>of</strong> April 19, 2002,polarimetric images by were created directly. The image pixel size is 0.05 x 0.05 x 0.05 m.The spatial region <strong>of</strong> the reconstructed images is –6m to 6m in azimuth (X) direction, 5m to15m in range (Y) direction and 0m to 6m in vertical (Z) direction according to a coordinatesystem shown in Figure 3.01. When an imaging region <strong>of</strong> interest was decided, the spatialdomain data <strong>of</strong> HH, VH and VV were normalized by the maximum value among whole data.The polarimetric images <strong>of</strong> vertical slices are shown continuously with increasing rangedistance. Red color indicates the HH polarimetric return; green color indicates the VH109
Chapter 5polarimetric return; and blue shows the VV component. From the contiguous images, we canobserve the tree T3 appears first at range <strong>of</strong> 7 m. A lot <strong>of</strong> green color indicates the leaves,twigs and slant branches. Soon the leaves <strong>of</strong> tree T2 come in the picture followed by the trunkat range <strong>of</strong> 9.8 m, which causes the strongest reflections (white, the combination <strong>of</strong> HH, VHand VV). Then tree T3 disappears from the image, and instead is replaced by the inclined bigbranch <strong>of</strong> tree T1 at about 11 m. After then, there is the main trunk <strong>of</strong> tree T3 in the low part<strong>of</strong> the image at 13 m. From these contiguous images, we could roughly find the main parts <strong>of</strong>trees and their locations only. The sensitivity <strong>for</strong> those particular cases was not good. In thoseimages, red and blue colors are dominating and green color was hardly observed due to thesmall VH polarimetric return.In order to improve the sensitivity <strong>of</strong> the polarimetric images, we tried to use an enhancementfunction to adjust the values <strong>of</strong> the different polarimetric components [105, 106]. Figure 5.01shows four enhancement functions which are tested, where x is the original normalized valueand y is the enhanced value.Enhanced valueOriginal normalized valueFigure 5.01 Functions <strong>for</strong> amplitude enhancement.Firstly, constant value 2 was multiplied to VH component to enhance cross polarizationterm according to the span invariance [107]. Using four enhancement functions to the sameprocessed data set as were used above, we plot the polarimetric images <strong>of</strong> the vertical slice atthe range distance <strong>of</strong> 9.8 m in Figure 5.02. The red color shows the reflection from trunks,110
Data Interpretation with <strong>SAR</strong> Polarimetric Analysissome horizontal branches and ground clutter, green indicates strong reflections from leaves,some inclined branches and ground clutter, and blue illustrates the reflections from trunks andvertical branches.Polarimetric Image red: HH green: VH blue: VVPolarimetric Image red: HH green: VH blue: VV(a)Polarimetric Image red: HH green: VH blue: VV(b)Polarimetric Image red: HH green: VH blue: VV(c)(d)Figure 5.02 Polarimetric color images (red-HH, green-VH, blue-VV) enhanced by: (a)⎛π⎞original, (b)sin ⎜ x⎟⎝ 2 ⎠ , (c) ⎛π0.5 ⎞sin ⎜ x ⎟⎝ 2 ⎠ , and (d) ⎛π0.3 ⎞sin ⎜ x ⎟⎝ 2 ⎠ .Comparing the images <strong>of</strong> Figures 5.02(a), 5.02(b), 5.02(c) and 5.02(d), the varying sensitivitycan be observed. In Figure 5.02(a), the image <strong>of</strong> trunk <strong>of</strong> tree T2 appears only. It is the realsituation since we did not change the normalized values. From Figures 5.02(b), only a fewspots <strong>of</strong> green part appear on the upper center. However, we can observe a green area inFigure 5.02(c), which corresponds to the reflections <strong>of</strong> leaves, twigs and branches <strong>of</strong> tree T2and reflection from tree T3. But in Figure 5.02(d), we cannot identify the different parts again⎛π0.5 ⎞due to too many specks. It was over enhanced. Hence, the function y = sin ⎜ x ⎟ seems to⎝ 2 ⎠be the best one <strong>for</strong> enhancing polarimetric data and improving image sensitivity.111
Chapter 55.1.2 Data Interpretation by Broadband ImagesApplying the migrated spatial domain data sets acquired in spring, summer and autumn, wetake absolute values <strong>of</strong> whole data. Then the normalized data sets are produced by themaximum value among the whole migrated data values in each measurement. VH data are⎛π0.5 ⎞adjusted by multiplying by 2 . The entire data are enhanced by function y = sin ⎜ x ⎟⎝ 2 ⎠shown in Figure 5.01. Then VH data are reorganized by threshold (the value greater than 0.18kept, otherwise being zero). The vertical images <strong>of</strong> three measurements <strong>of</strong> Exp#A at range12.9 m, 9.8 m and 8.3 m are shown in Figure 5.04, Figure 5.05, and Figure 5.06, respectively.Comparing with the corresponding photographs shown in Figure 5.03, the reflections from thetrunks, branches and the leaves are different with seasonal variations shown in Figure 5.04,Figure 5.05, and Figure 5.06. From the images <strong>of</strong> Figure 5.04(a), (b) and (c), the differentphenomena can be observed. The strong reflection (red, blue) in the left lower part (yellowdotted line rectangle) shows scattering from the trunk <strong>of</strong> tree T1. On the upper part indicatedby a yellow dashed line square, the blue shows the vertical branch. The reflections at twosides <strong>of</strong> trunk <strong>of</strong> tree T1 indicate the inclined branches. Obviously, the green part in Figure5.04(b) is stronger than others due to the many leaves and the growing branches <strong>of</strong> tree T1 insummer (yellow dash-dotted line ellipse). We can find the corresponding components fromthe photographs in Figure 5.03. In Figure 5.05, the strong reflection (red, blue and whitecolor) in the center (white solid line rectangle) shows scattering from the trunk <strong>of</strong> tree T2. Theupper part (white solid line ellipse) indicates the leaves <strong>of</strong> tree T2. They are almost the samedue to the evergreen tree T2. The lower parts in Figure 5.05(b) show greater cross polarizationdue to the exuberant short tree (shrub) T3 and ground plants in summer. In Figure 5.06, a part<strong>of</strong> leaves and branches <strong>of</strong> tree T2 (indicated by small blue solid line circle and small bluesolid line ellipse) and difference <strong>of</strong> tree T3 (larger blue solid line circle) in different seasonscan be found. For instance, because there were many plants (shown in the yellowdash-dot-lined ellipse) around the tree T3, the large area <strong>of</strong> reflection appeared at thecorresponding position in Figure 5.06(b). There are stronger reflection in the image <strong>of</strong>Exp#A-2 due to the exuberant branches and leaves. Of course, some ground clutter alsoaffects the cross polarization component, which shows green color in the low parts <strong>of</strong> each <strong>of</strong>the images. Hence, we can assume that the main reason is volume scattering due to differentvolumes within different seasons. Although the shapes <strong>of</strong> trees were very similar, the water112
Data Interpretation with <strong>SAR</strong> Polarimetric Analysiscontent <strong>of</strong> trees in spring was higher than in autumn due to the observed different reflections.SpringT1 T2 T3(a)SummerT1T2T3(b)AutumnT1 T2 T3(c)Figure 5.03 Corresponding areas <strong>of</strong> tress in: (a) spring, (b) summer, and (c) autumn.113
Chapter 5Polarimetric Image in Spring red: HH green: VH blue: VV(a)Polarimetric Image in Summer red: HH green: VH blue: VV(b)Polarimetric Image in Autumn red: HH green: VH blue: VV(c)Figure 5.04 Broadband vertical pr<strong>of</strong>iles <strong>of</strong> T1 at range <strong>of</strong> 12.9m red-HH green-VH blue-VVin: (a) spring, (b) summer, and (c) autumn.114
Data Interpretation with <strong>SAR</strong> Polarimetric AnalysisPolarimetric Image in Spring red: HH green: VH blue: VV(a)Polarimetric Image in Summer red: HH green: VH blue: VV(b)Polarimetric Image in Autumn red: HH green: VH blue: VV(c)Figure 5.05 Broadband vertical pr<strong>of</strong>iles <strong>of</strong> T2 at range <strong>of</strong> 9.8m red-HH green-VH blue-VV in:(a) spring, (b) summer, and (c) autumn.115
Chapter 5Polarimetric Image in Spring red: HH green: VH blue: VV(a)Polarimetric Image in Summer red: HH green: VH blue: VV(b)Polarimetric Image in Autumn red: HH green: VH blue: VV(c)Figure 5.06 Broadband vertical pr<strong>of</strong>iles <strong>of</strong> T3 at range <strong>of</strong> 8.3m red-HH green-VH blue-VV in:(a) spring, (b) summer, and (c) autumn.116
Data Interpretation with <strong>SAR</strong> Polarimetric AnalysisThere<strong>for</strong>e, using the broadband polarimetric image description method, different components<strong>of</strong> trees and their different scattering per<strong>for</strong>mances in the different seasons can be detected bythe polarimetric measurement, generally.5.2 Single Frequency Polarimetric Interpretation5.2.1 Short Time Fourier Trans<strong>for</strong>m and Spatial-Frequency Trans<strong>for</strong>mationSTFTThe short time Fourier trans<strong>for</strong>m (STFT), also referred to as the Windowed FourierTrans<strong>for</strong>m, was developed in 1946 by Dennis Gabor as an attempt to overcome the lack <strong>of</strong>in<strong>for</strong>mation on time locality in Fourier analysis [108-110]. It is a compromise between thetime and the frequency view <strong>of</strong> a signal, and is thereby better suited <strong>for</strong> analysis <strong>of</strong>non-stationary signals than the ordinary Fourier Trans<strong>for</strong>m is. STFT computes atime-frequency distribution <strong>of</strong> an input signal x within a given frequency interval as asequence <strong>of</strong> short-time spectra <strong>of</strong> windowed signal segments with constant length.In STFT the signal is analyzed section by section. A localized time window function γ ( t − τ )is slid over the signal, and Fourier analysis is per<strong>for</strong>med to the signal segment in the window.The STFT is expressed as [110]:orj2( , τ) ∞−= ( ) γ ( −τ) ∫ (5.01)π ftX f x t t e dt−∞( ) ( )∞− j2 π fτ ' * ' j2 π f τ ''∫ (5.02)X( f, τ ) = e X f Γ f − f e df−∞The STFT provides some in<strong>for</strong>mation on both time locality and frequency spectra, but theprecision <strong>of</strong> the in<strong>for</strong>mation is limited and it is determined by the size <strong>of</strong> the window. Thereasons <strong>for</strong> this limitation may be reduced to the Heisenberg uncertainty principle [109]. It117
Chapter 5'can be expressed by the uncertainty relationship δt⋅ δ f = C . Here C indicates a constant. Asmall window leads to high precision in time but poor frequency resolution. The frequencyresolution <strong>of</strong> the STFT can be improved by using a larger window, however this leads to lowprecision in time and problems with non-stationary data.The disadvantage <strong>of</strong> the STFT is that all frequencies <strong>of</strong> the signal are analyzed using the sameresolution. One specific size <strong>of</strong> window is chosen as a compromise between requirements <strong>of</strong>time and frequency resolution, and this window is then used to analyze every frequency <strong>of</strong> thesignal.It is displayed as time (x-axis) versus frequency (y-axis) with the power or amplitude spectralvalues given on density image shown in Figure 5.07.Figure 5.07 A time domain signal and the response spectrum distribution after short timeFourier trans<strong>for</strong>m.118
Data Interpretation with <strong>SAR</strong> Polarimetric AnalysisSpatial-Frequency Trans<strong>for</strong>mFrom Figure 5.07, we know that it includes both 3-D spatial in<strong>for</strong>mation and frequencyspectrum in<strong>for</strong>mation if a signal is processed by STFT. The cost is poor frequency resolutionand longer computer time while spatial resolution is precise. If we want to obtain precisefrequency domain resolution and do not care about spatial resolution in range direction, wecan use a long time window to cover the one spatial dimension entirely or do not use the timewindow to trans<strong>for</strong>m spatial domain data to wave-number domain by Fourier trans<strong>for</strong>mation.We have the following trans<strong>for</strong>m relation.p( zxy , , ) ⇒ Pzxk ( , , y)(5.03)where p(, zxy , ) is the 3-D spatial domain data and Pzxk (, , ) is the spatial-wavenumberdomain data.yky2π 2π f 2πf= = = (5.04)λ v c / ε µy y r rwhere λyis the wavelength along y-direction, v yis the y-direction velocity <strong>of</strong> theelectromagnetic wave in media, f is the frequency, c is the speed <strong>of</strong> light, εris therelative permittivity and µ is the relative permeability. In air ( ε = 1 and µ = 1), we haverc⋅k yf = (5.05)2πThere<strong>for</strong>e, we use this equation to obtain frequency in<strong>for</strong>mation.rr5.2.2 Polarimetric Target DescriptorsFor deterministic scatterers, deterministic scattering or completely polarized scattering can becompletely described by the coherent Sinclair matrix. But <strong>for</strong> partial scatterers, randomscattering or partially polarized scattering, they can not be described by the Sinclair matrix.The statistical descriptions are necessary such as using the Kennaugh or the Covariancematrices [107]. Hence, the Coherency matrix and the Covariance Matrix may usefully beemployed as target polarimetric descriptors. Figure 5.08 shows different target polarimetric119
Chapter 5descriptors.Sinclair Matrix[ ]S = ⎡ ⎣ ⎢Kennaugh MatrixS HHS HV⎤1S ⎥=⎛⎜T ⎡ ⊗*VH S VV ⎦2 ⎝ ⎣⎢⎤⎦⎥[ K] [ V] [ S] [ S] [ V]⎞⎠⎟Different Target Polarimetric Descriptorsk =12Scattering Vector k[ S + S S − S 2S] THHVVHHVVHVScattering Vector ΩΩ =[ S 2S S ] THHHVVVCoherency Matrix [T]* T[ T ] = k ⋅ kCovariance Matrix [C]T*[ C] = ΩΩFigure 5.08 Different target polarimetric descriptors.For the monostatic case, [T] and [C] have the same eigenvalues. Both contain the samein<strong>for</strong>mation about polarimetric scattering amplitudes, phase angles and correlations. [T] iscloser related to physical and geometrical properties <strong>of</strong> the scattering process, and thus allowsa better and direct physical interpretation. [C] is directly related to the system measurable. [T]is directly related to the Kennaugh matrix [ 5, 111, 112] and the Huynen parameters[112-114].5.2.3 Single Frequency Scattering MatricesThe migrated data <strong>for</strong> Exp#A-1, Exp#A-2 and Exp#A-3 <strong>of</strong> three different tree measurementswere used to be trans<strong>for</strong>med to spatial frequency domain by short time Fourier trans<strong>for</strong>m120
Data Interpretation with <strong>SAR</strong> Polarimetric Analysis(5.01). Three observation points from the target area are selected. They are Point A(-2.8 12.92.5)m, Point B (-0.3, 9.8, 2.5)m and Point C(5, 8.8, 2.5)m, which correspond to T1, T2 and T3coordinated in Figure 3.01, respectively. At each point, the data contain frequency in<strong>for</strong>mationand value. Applying 3 GHz data, the scattering matrixes are calculated as follows.Point A in Exp#A-1:SMA1oo⎡j55.93 − j117.58e0.0199e⎤= ⎢ ⎥(5.06)o o− j117.58 − j56.58⎢⎣0.0199 e 0.5229e⎥⎦Point A in Exp#A-2:SMA2oo⎡j52.24 − j123.23e0.4803e⎤= ⎢ ⎥(5.07)o o− j123.23 j54.29⎢⎣0.4803 e 1.5813e⎥⎦Point A in Exp#A-3:SMA3oo⎡j122.77 j49.63e0.1329e⎤= ⎢ ⎥(5.08)o oj49.63 j54.11⎢⎣0.1329 e 1.945e⎥⎦Point B in Exp#A-1:SMB1oo⎡− j146.31 j34.42e0.6493e⎤= ⎢ ⎥(5.09)o oj34.42 j35.45⎢⎣0.6493 e 0.7574e⎥⎦Point B in Exp#A-2:SMB2oo⎡j34.67 j34.36e0.4718e⎤= ⎢ ⎥(5.10)o oj34.36 j34.40⎢⎣0.4718 e 1.3186e⎥⎦Point B in Exp#A-3:SMB3oo⎡j34.86 j35.47e0.6198e⎤= ⎢ ⎥(5.11)o oj35.47 j38.56⎢⎣0.6198 e 1.2708e⎥⎦121
Chapter 5Point C in Exp#A-1:SMC1oo⎡− j152.92 − j153.19e1.4826e⎤= ⎢ ⎥(5.12)o o− j153.19 − j150.39⎢⎣1.4826 e 1.0974e⎥⎦Point C in Exp#A-2:SMC 2oo⎡j26.49 − j155.75e1.1656e⎤= ⎢ ⎥(5.13)o o− j155.75 j25.21⎢⎣1.1656 e 1.0230e⎥⎦Point C in Exp#A-3:SMC3oo⎡− j154.0 − j152.51e0.4570e⎤= ⎢ ⎥(5.14)o o− j152.51 j25.88⎢⎣0.4570 e 1.0769e⎥⎦For Point A, the HH and VV are dominated due to horizontal scatterers, ground clutter andvertical branches and the trunk <strong>of</strong> tree T1.Moreove, values <strong>of</strong> the cross polarization terms inspring and autumn are smaller than the ones <strong>for</strong> summer. The fact may be that there weremany larger leaves in summer to cause stronger VH reflection.We can also find that the co-polarization terms <strong>of</strong> point B are dominant but the VH returns arealmost alike and stronger than those at point A. It is due to the fact that point B is located at anevergreen tree. The conifer needles and branches cause large VH reflection.For point C, VH terms are stronger than HH and VV terms in spring (5.12) and in summer(5.13) due to number leaves and plants. When leaves fall <strong>of</strong>f, the VH term became smaller inEquation (5.14).5.2.4 Polarimetric Power Density Images from Covariance MatrixElectromagnetic wave propagation is a vector phenomenon, i.e. all electromagnetic waves canbe expressed as complex vectors. Plane electromagnetic waves can be represented bytwo-dimensional complex vectors. This is also the case <strong>for</strong> spherical waves when the122
Data Interpretation with <strong>SAR</strong> Polarimetric Analysisobservation point is sufficiently far removed from the source <strong>of</strong> the spherical wave.There<strong>for</strong>e, if one observes a wave transmitted by a radar antenna when the wave is a largedistance from the antenna (in the far-field <strong>of</strong> the antenna), the radiated electromagnetic wavecan be adequately described by a two-dimensional complex vector. If this radiated wave isnow scattered by an object, and one observes this wave in the far-field <strong>of</strong> the scatterer, thescattered wave can again be adequately described by a two-dimensional vector. In thisabstract way, one can consider the scatterer as a mathematical operator which takes onetwo-dimensional complex vector (the wave impinging upon the object) and changes that intoanother two-dimensional vector (the scattered wave). Mathematically, there<strong>for</strong>e, a scatterercan be characterized by a complex 2x2 scattering matrix [111]. However, this matrix is afunction <strong>of</strong> the radar frequency, and the viewing geometry.This coherent Sinclair scattering matrix S is a function <strong>of</strong> frequency and relative viewingangleS⎡ShhShv⎤= ⎢SvhS ⎥⎣ vv ⎦(5.15)For backscattering in reciprocal media, S hv= S vh, the span and determinantal invariancesbecome2 2 2{ ( )} { hh2hv vv }span S hv = vh = S + S + S(5.16)and2{ S( hv vh)} ShhSvv Shvdet = = − ( )(5.17)Whereas in radar polarimetry the use <strong>of</strong> the associated Mueller matrix M, expressed in terms<strong>of</strong> the <strong>for</strong>ward propagation coherent Jones matrix J is used extensively in polarimetricapplications; the Kennaugh matrix was found to be less expedient <strong>for</strong> analyzing polarimetric<strong>SAR</strong> image properties whereas it is <strong>of</strong> great utility in polarimetric radar meteorology <strong>for</strong>which the covariance matrix representation is less useful. Instead, the covariance matrix orcoherency matrices were found to be more expedient as reviewed in Cloude and Pottier [115].The process <strong>of</strong> averaging has to be accomplished using the coherency matrices, or thecovariance matrices. The covariance matrix is <strong>based</strong> on the components <strong>of</strong> the scatteringmatrix (5.15). In the linear basis, the scatterer is represented by its Lexicographic scatteringvector as:123
Chapter 5⎡ Shh⎤⎢ ⎥Ω= ⎢ 2Svh⎥⎢ S ⎥⎣vv⎦(5.18)The2 , on the S vhterm in (5.18) is to ensure consistency in the span (total power)computation. In (5.18) we have limited our focus to the backscatter (or monostatic) case only.The Covariance matrix is defined as2⎡* *Shh 2 ShhSvh ShhS⎤vv⎢⎥* T ⎢* 2* ⎥Cvh = ΩΩ = ⎢ 2 SvhShh 2 Svh 2SvhSvv⎥⎢⎥* *2⎢ SvvShh 2 SvvSvh Svv⎥⎣⎦(5.19)The superscript * denotes the complex conjugate. The superscript T means transpose.Complex spatial frequency domain data are produced from migrated spatial domain data byFFT, which is the so-called fast Fourier trans<strong>for</strong>mation. Based on equation (5.19), wecalculate covariance matrices by using the spatial frequency domain data. There<strong>for</strong>e, thepolarimetric power density images <strong>based</strong> on diagonal terms <strong>of</strong> the covariance matrix <strong>of</strong> (5.19)<strong>for</strong> Exp#B-2 data (cherry flowers) and Exp#B-3 (cherry leaves) are shown in Figure 5.09. Thescene <strong>of</strong> cherry flowers is shown in Figure 5.09(a). Power density images <strong>of</strong> cherry flowerswith frequencies <strong>of</strong> 1.5 GHz, 2.5 GHz, 3.5 GHz and 4.5 GHz are shown on the left side <strong>of</strong>Figure 5.09. The scene and images <strong>of</strong> cherry leaves are shown on the right side <strong>of</strong> Figure 5.09.In lower frequencies images (1.5 GHz and 2.5 GHz), the red color and blue color(co-polarization components) are dominant. The reason is that the lower frequencyelectromagnetic wave can propagate into the canopy <strong>of</strong> the cherry tree and reflect backagainst large horizontal and vertical branches. With frequency increased, the green colorappears more and more, and becomes dominant. The fact is the higher frequencyelectromagnetic wave return from the smaller scatterers like inclined twigs, flowers or leaves.Comparing images <strong>of</strong> flowers and leaves, we can find that a little more green color appear incherry tree leaves images <strong>of</strong> higher frequencies ( 3.5 GHz and 4.5 GHz). The reason may beassumed that the volume <strong>of</strong> the cherry tree leaves is greater than that <strong>of</strong> cherry flowers andthere are larger scattering surface areas with randomly oriented leaves. Actually, thedifference is very small. If up to 10 GHz electromagnetic wave is used, the difference shouldbe greater.124
Data Interpretation with <strong>SAR</strong> Polarimetric Analysis(a) Flowers(f) Leaves(b) 1.5 GHz(g) 1.5 GHz(c) 2.5 GHz(h) 2.5 GHz(d) 3.5 GHz(i) 3.5 GHz(e) 4.5 GHz(j) 4.5 GHzFigure 5.09 Single-frequency power density images <strong>of</strong> cherry blossom and leaves red-|HH| 2green-2|VH| 2 blue-|VV| 2 : (a) imaging area <strong>of</strong> flowers, (b) 1.5 GHz, (c) 2.5 GHz, (d) 3.5 GHz,(e) 4.5 GHz images <strong>of</strong> flowers, and (f) imaging area <strong>of</strong> leaves, (g) 1.5 GHz, (h) 2.5 GHz, (i)3.5 GHz, (j) 4.5 GHz images <strong>of</strong> leaves.125
Chapter 55.2.5 Power Density Images from Pauli Covariance MatrixPauli Spin Matrix BasisA general Pauli scattering vector k is defined from (5.15) <strong>for</strong> symmetric matrix case as.1Tk = [ Shh + Svv Shh − Svv 2Svh](5.20)2Another way to visualize the polarimetric in<strong>for</strong>mation is to trans<strong>for</strong>m to the Pauli spin matrixbasis, first introduced in this context by Cloude [116]. In the basis <strong>of</strong> the backscatter case, thePauli spin target scattering vector can be rewritten from (5.15)1Tks = [ Shh + Svv 2 Svh Shh − Svv](5.21)2The subscript s is to distinguish the vector from the general Pauli target scattering vector k.For single bounce scattering from a sphere, Shhand S vvare in phase and the magnitude <strong>of</strong>the first term <strong>of</strong> the Pauli spin scattering vector is much larger than the other two. Forscattering from a metallic dihedral corner reflector, however, one expects a 180 degree phasedifference between Shhand S vv. In that case, the magnitude <strong>of</strong> the third term in the Paulispin scattering vector will be much larger that the other two.Then Pauli covariance matrix is defined ass*TT ksks= ⋅ (5.22)Note that the diagonal terms <strong>of</strong> the Pauli covariance matrix are just the magnitudes <strong>of</strong> thethree elements <strong>of</strong> the target vector squared. There<strong>for</strong>e, if we display the three diagonalelements <strong>of</strong> the Pauli covariance matrix as red <strong>for</strong> the first term, green <strong>for</strong> the second term andblue <strong>for</strong> the third term, we can interpret red areas as having increased single reflectionscattering, blue areas as mostly double reflection scattering, and green areas as diffusescattering.If one has to be somewhat careful to simply equate a blue color with double reflections,careful consideration <strong>of</strong> the diagonal terms <strong>of</strong> the Pauli covariance matrix shows that the first,second and third terms ares 1 2 1 2 1 2 *T11= Shh + Svv = Shh + Svv +R e( ShhSvv)(5.23)2 2 2126
Data Interpretation with <strong>SAR</strong> Polarimetric Analysiss2T22 2 S vh= (5.24)ands 1 2 1 2 1 2 *T33= Shh − Svv = Shh + Svv −R e( ShhSvv)(5.25)2 2 2From these equations, it is clear that as long as the real part <strong>of</strong> the cross-correlation betweenShhand S vvis positive, (which means that the co-polarized phase difference is less than 90degrees) the first term will be larger than the third, i.e. the image will exhibit a red color.When the real part <strong>of</strong> the cross-correlation between Shhand S vvis negative, (which meansthe co-polarized phase difference is greater than 90 degrees) the reverse will be true and theimage will exhibit a blue color. There<strong>for</strong>e, one has to be careful when interpreting the colorsin terms <strong>of</strong> scattering mechanisms. A more correct interpretation is that red colors indicatescattering more indicative <strong>of</strong> single reflections than double reflections, with the reverse true<strong>for</strong> blue colors. This becomes especially important when two areas have phase differencejust less than or just more than 90 degrees.Moreover, complex frequency domain data are produced from the migrated spatial domaindata <strong>of</strong> Exp#A-1 <strong>for</strong> different trees by STFT. Hence, the polarimetric power density images<strong>based</strong> on Pauli covariance matrix <strong>for</strong> Exp#A-1 are shown in Figure 5.10. Here data atfrequencies <strong>of</strong> 1.5 GHz, 2.5 GHz, 3.5 GHz and 4.5 GHz are displayed. We show singlefrequency images <strong>of</strong> different types <strong>of</strong> trees in spring. For lower frequency, the main parts <strong>of</strong>trees can be detected. With frequency increased, the green color area becomes large andconfuse scattering term dominates. It caused by leaves and small branches. Small scatterershave shown strong backscattering features with high frequency.127
Chapter 5Image at 1.5 GHz |HH+VV| 2 |VH| 2 |HH-VV| 2 Image at 2.5 GHz |HH+VV| 2 |VH| 2 |HH-VV| 2(a)T1 (-2.85 12.9)Diameter: 0.31T2 (-0.5 9.8)Diameter: 0.4YT3 (4.5 8.5)(b)Rail <strong>of</strong> antennapositioner-6m 06mXImage at 3.5 GHz |HH+VV| 2 |VH| 2 |HH-VV| 2 Image at 4.5 GHz |HH+VV| 2 |VH| 2 |HH-VV| 2(c)(d)Figure 5.10 Polarimetric power density images <strong>of</strong> trees in spring Exp#A-1 at height <strong>of</strong> 2.5 m<strong>based</strong> on Equations (5.23)-(5.25) red-|HH+VV| 2 green-2|VH| 2 blue-|HH-VV| 2 : (a) 1.5 GHz,(b) 2.5 GHz, (c) 3.5 GHz and (d) 4.5 GHz.128
Data Interpretation with <strong>SAR</strong> Polarimetric Analysis5.3 Entropy Based Polarimetric Target DecompositionIn this section, the main polarimetric parameters which can be used <strong>for</strong> the extraction <strong>of</strong>physical in<strong>for</strong>mation about distributed scatterers from polarimetric coherence matrix data areintroduced and discussed.There are many polarimetric target decomposition methods [117-127] <strong>based</strong> on the covariancematrix and the coherency matrix. Here, we show one <strong>of</strong> the currently most usefuleigenvector-<strong>based</strong> decomposition; it is the polarimetric H/alpha decomposition developed byS.R. Cloude & E. Pottier [123].Based on these eigenvalues and eigenvectors <strong>of</strong> a coherency matrix, they defined three usefulamong five parameters that characterize the scattering property <strong>of</strong> the medium: entropy H,anisotropy A and angle α . The other angles β and γ defined in this decompositionhave found few applications so far and await further analyses.5.3.1 Coherence MatrixFrom the Pauli scattering vector k (5.20), the coherency matrix can be rewritten as1N1NNN* T[ T] = ∑ki⋅ ki = ∑ [ Ti](5.26)i= 1 i=1and in terms <strong>of</strong> the eigenvectors and eigenvalues as⎡λ0 0 ⎤⎢ ⎥⎢ ⎥⎢⎣0 0 λ ⎥3 ⎦11 * T−[ T] = [ U ][ Σ ][ U ] = [ u u u ] 0 λ 0 [ u u u ]3 3 1 2 3 2 1 2 3(5.27)where [ U3]is an orthogonal eigenvector matrix representation129
Chapter 5⎡cos α1 cos α2 cosα3⎤⎢⎥= ⎢ ⎥⎢iγ1 iγ2iγ3sinα1sin β1e sinα2sin β2e sinα3sinβ3e⎥⎣⎦iδ3[ ] sin cos 1 iδsin cos 2iδU α β e α β e sinα cos β e3 1 1 2 2 3 3(5.28)<strong>for</strong> which the parametersαi, βi, δi, and γiare explained in [123].Theλiin [ Σ ] are real nonnegative eigenvalues.Since the coherence matrix [T] is hermitian positive semi-definite, it can always bediagonalized by an unitary similarity trans<strong>for</strong>mation <strong>of</strong> the <strong>for</strong>m [115, 116, 122, 126]. Theidea <strong>of</strong> the eigenvector approach is to use the diagonalisation <strong>of</strong> the coherence matrix [T] <strong>of</strong> adistributed scatterer, which is in general <strong>of</strong> rank 3, in order to decompose it into thenon-coherent sum <strong>of</strong> three independent coherence matrices [T i ].5.3.2 Polarimetric Entropy H and Angle αCloude defined a parameter, polarimetric entropy H, to account <strong>for</strong> randomness in distributedtargets’ scattering mechanisms. First, the probabilities <strong>of</strong> λiare obtained from (5.27).Pi=3λ∑k = 1iλk(5.29)The polarimetric entropy H can be calculated from equation (5.28).3∑ ilog3i(5.30)i=1H = −P PThe physical meaning <strong>of</strong> polarimetric entropy is described here. For a deterministic scattererwith entropy H = 0, the anisotropy becomes zero, as well as in the case <strong>of</strong> a pure randomscatterer with entropy H = 1. For scatterers characterised by intermediate entropy values, ahigh anisotropy indicates the presence <strong>of</strong> only one strong secondary scattering process, whilea low anisotropy indicates the appearance <strong>of</strong> two equally strong scattering processes.Schematic representation <strong>of</strong> the entropy H interpretation is listed in Table 5.01.130
Data Interpretation with <strong>SAR</strong> Polarimetric AnalysisTable 5.01 Schematic representation <strong>of</strong> the entropy H interpretationH value0
Chapter 5Table 5.02 Schematic representation <strong>of</strong> the alpha-angle interpretationalpha value 0 o0 o < α
Data Interpretation with <strong>SAR</strong> Polarimetric Analysis(a)Entropy(b)Entropy(c)Entropy(d)Entropy(e)Entropy(f)EntropyFigure 5.11 Entropy distribution images <strong>of</strong> a cherry tree: (a) 1.5 GHz <strong>of</strong> Exp#B-1, (b) 4.5GHz <strong>of</strong> Exp#B-1, (c) 1.5 GHz <strong>of</strong> Exp#B-2, (d) 4.5 GHz <strong>of</strong> Exp#B-2, (e) 1.5 GHz <strong>of</strong> Exp#B-3,and (f) 4.5 GHz <strong>of</strong> Exp#B-3.133
Chapter 5(a)alpha [deg](b)alpha [deg](c)alpha [deg](d)alpha [deg](e)alpha [deg](f)alpha [deg]Figure 5.12 alpha distribution images <strong>of</strong> a cherry tree: (a) 1.5 GHz <strong>of</strong> Exp#B-1, (b) 4.5 GHz<strong>of</strong> Exp#B-1, (c) 1.5 GHz <strong>of</strong> Exp#B-2, (d) 4.5 GHz <strong>of</strong> Exp#B-2, (e) 1.5 GHz <strong>of</strong> Exp#B-3, and (f)4.5 GHz <strong>of</strong> Exp#B-3.134
Data Interpretation with <strong>SAR</strong> Polarimetric AnalysisFor precisely detecting small changes <strong>of</strong> a cherry tree during different periods, two analysisareas with three measurements <strong>for</strong> a cherry tree, which are indicated by a dotted line squareand a solid line rectangle, are shown in Figure 5.13. There are quite different situations (budsand branches, flowers and leaves) during three measurements inside the blue square; and thereis the trunk in a red rectangle area.Exp#B-1 on April 2, 2003 Exp#B-2 on April 14, 2003 Exp#B-3 on May 27, 2003(a)(b)(c)Figure 5.13 Two indicated analysis areas, blue square-buds/flowers/leaves, red box-trunk: (a)Exp#B-1, (b) Exp#B-1, and (c) Exp#B-3.Entropy-alpha distributions <strong>of</strong> the upper square area with frequencies 1.5 GHz and 4.5 GHzare shown in Figure 5.14. From this figure, we can find the lower frequency (1.5 GHz)distribution, in Figure 5.14(a); whereas Figure 5.14 (c) and Figure 5.14 (e), show higherentropy values - concentrated between 0.6 and 0.9, shown in red color circles. But <strong>for</strong> higherfrequency (4.5 GHz) distributions, in Figure 5.14 (b), Figure 5.14 (d) and Figure 5.14 (f), theconcentrated entropy values center between 0.4 to 0.8 and alpha values are also smaller thanthose <strong>for</strong> the lower frequency. They indicate that some branches, flowers and leaves appear asrandom scatterers with larger entropy <strong>for</strong> the lower frequency. But the same targets showfeatures <strong>of</strong> close to stationary scattering behavior with smaller entropy <strong>for</strong> higher frequencies.Comparing the three higher frequency distributions in Figure 5.14 (b), Figure 5.14 (d) andFigure 5.14 (f), we can find the alpha values shift down a little (about 5 degrees) - step bystep. That demonstrates that the small branches in the first measurement (Exp#B-1) showdipole scattering characteristics. Then, they gradually display surface scattering features withflowers (in Exp#B-2) and leaves (in Exp#B-3) increased.135
Chapter 5(a)(b)(c)(d)(e)(f)Figure 5.14 Entropy-alpha distributions <strong>of</strong> a part <strong>of</strong> a cherry tree- the blue square indicatedin Figure 5.13: (a) 1.5 GHz <strong>of</strong> Exp#B-1, (b) 4.5 GHz <strong>of</strong> Exp#B-1, (c) 1.5 GHz <strong>of</strong> Exp#B-2,(d) 4.5 GHz <strong>of</strong> Exp#B-2, (e) 1.5 GHz <strong>of</strong> Exp#B-3, and (f) 4.5 GHz <strong>of</strong> Exp#B-3.136
Data Interpretation with <strong>SAR</strong> Polarimetric AnalysisNow the average entropy values and alpha values are calculated <strong>for</strong> two analysis areas withthree measurements, respectively. Mean values <strong>of</strong> Entropy and alpha <strong>for</strong> two indicatedanalysis areas are listed in Table 5.03. Differences <strong>of</strong> entropy and alpha are shown in Figure5.15 (a) and Figure 5.15 (b).Table 5.03 Comparison <strong>of</strong> mean value <strong>of</strong> Entropy and alpha in two analysis areas: red facefonts <strong>for</strong> trunk, blue face fonts <strong>for</strong> buds/flower/leavesExp#B-1Exp#B-2Exp#B-3on April 2, 2003on April 14, 2003on May 27, 2003alpha54.1254.4455.831.5 GHzalpha56.1058.0457.84entropy0.54350.56760.5447entropy0.56690.60690.6007alpha52.3852.0853.604.5 GHzalpha58.8257.3654.11entropy0.23850.25870.3217entropy0.39880.42470.4643(a)(b)Figure 5.15 Changes <strong>of</strong> mean value: (a) Entropy and (b) alpha in the two analysis areas.137
Chapter 5We can recover these aspects and reasons:(a)(b)(c)(d)Entropies <strong>of</strong> lower frequency (1.5 GHz) almost did not change in Figure 5.15(a). Butentropies <strong>of</strong> higher frequency (4.5 GHz) increased slightly along three measurements.That means the scatterers are more sensitive at higher frequencies than <strong>for</strong> lowerfrequencies during changing growth conditions.Entropies at lower frequencies (1.5 GHz) are greater than the ones <strong>of</strong> higherfrequencies (4.5 GHz) among three measurements. The fact is that scatterers appearas random scatterers with larger entropy <strong>for</strong> the lower frequency and show features<strong>of</strong> close to stationary scattering behavior with smaller entropy <strong>for</strong> higher frequencies.For high frequency, the entropy <strong>of</strong> the trunk is lower than the value <strong>for</strong>branches/flowers/leaves due to their stationary scattering features.alpha values <strong>of</strong> both low and high frequencies <strong>for</strong> trunk do not change so much dueto a close dipole scattering feature.(e) alpha values with high frequency in areas with branches shift down by about 5degrees. The fact is that the flowers, especially the number <strong>of</strong> leaves on brancheshave larger reflection surfaces than the bare small branches and tiny flower-stemsdisplay.At low frequencies, EM waves propagate into the canopy <strong>of</strong> a cherry tree and backscatter alittle from some <strong>of</strong> the larger branches inside the tree canopy. It shows larger entropy due torandom scattering <strong>of</strong> a mixture <strong>of</strong> scatterers.At high frequencies, the electromagnetic wave scatters from the canopy <strong>of</strong> the tree. For thesecond measurement (cherry tree in bloom) and the third measurement (leaves), the entropyvalues are almost the same as <strong>for</strong> the first one (buds), but the alpha values have becomesmaller. The fact is the flower petals and leaves on branches have larger scattering surfaces ascompared to the bare small branches at the canopy <strong>of</strong> the cherry tree.138
Data Interpretation with <strong>SAR</strong> Polarimetric Analysis5.4 SummaryIn this chapter, we have introduced a number <strong>of</strong> different analysis tools <strong>for</strong> viewing andinterpreting broadband and single-frequency polarimetric data. Various color images weredescribed and their interpretations were discussed. Scattering mechanisms <strong>of</strong> differentcomponents <strong>of</strong> vegetation, e.g. trees, were demonstrated by polarimetric analysis techniques.We also discussed different measurements to interpret the amount <strong>of</strong> diffuse scattering, <strong>for</strong>instance, buds, flowers and leaves,and showed that they all provide similar in<strong>for</strong>mationand different features by polarimetric analysis. Finally, it showed that the alpha/entropydecomposition provides a convenient way to recover more in<strong>for</strong>mation about vegetationscattering.It showed that the developed broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong> system providedhigh qualitative data <strong>for</strong> trees. Slight differences <strong>of</strong> trees could be detected by polarimetricanalysis and interpretations <strong>for</strong> the acquired data in dependence <strong>of</strong> tree type and seasonalgrowth patterns.139
Chapter 5140
Chapter 6CONCLUSIONSIn this dissertation, the development <strong>of</strong> a broadband ground-<strong>based</strong> <strong>SAR</strong> system and itsapplication to environmental studies are described. We have used a newly developedground-<strong>based</strong> polarimetric broadband <strong>SAR</strong> system to monitor seasonal variations in trees.In Chapter 2, <strong>based</strong> on the conventional <strong>SAR</strong> <strong>for</strong>mulation, the principle <strong>of</strong> a ground-<strong>based</strong><strong>SAR</strong> was derived. By computer simulation, the estimated resolution was derived that wasconsistent with theoretical resolution. Including the description <strong>of</strong> the pertinent measurementequipments, specifications <strong>of</strong> a broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong> system werepresented. Testing results showed satisfactory polarimetric per<strong>for</strong>mances <strong>of</strong> the developedsystem. Further a modified two-way oriented dihedral <strong>based</strong> polarimetric calibration wasintroduced and its use was verified. Theoretical RCS calculations <strong>for</strong> a metallic sphere and adihedral corner reflector were also per<strong>for</strong>med. Improvement <strong>of</strong> the specific polarimetriccharacteristics <strong>for</strong> the broadband ground-<strong>based</strong> <strong>SAR</strong> system was determined by polarimetriccalibration results.Due to the fact that trees are some <strong>of</strong> the most important vegetation scatterers inenvironmental studies, we implemented the developed <strong>SAR</strong> system <strong>for</strong> real measurements <strong>for</strong>the monitoring <strong>of</strong> different tree types and during different seasons. Target and distributedscatterer determination and procedures <strong>of</strong> measurements <strong>for</strong> two experimental sites aredescribed in the Chapter 3. Using the developed broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong>system, continued data acquisitions were carried out <strong>for</strong> three different types <strong>of</strong> trees in threeseasons, and <strong>for</strong> a cherry tree during different situations at two experimental sites. For eachmeasurement, HH, VH and VV polarimetric scattering data were acquired simultaneously.Signal verification showed that good quality data has been acquired by the ground-<strong>based</strong> <strong>SAR</strong>system. It demonstrated that the acquired data not only included in<strong>for</strong>mation about targetpolarimetric per<strong>for</strong>mances but also frequency regime dependence <strong>of</strong> differently sizedcomponents <strong>of</strong> trees by signal comparison.Chapter 4 was to review signal processing methods <strong>for</strong> broadband ground-<strong>based</strong> polarimetric141
Chapter 6<strong>SAR</strong> data. Algorithms <strong>of</strong> time gating, matched filtering and diffraction stacking were veryeffective techniques <strong>for</strong> ground-<strong>based</strong> <strong>SAR</strong> data processing, especially the time gating methodand diffraction stacking migration. The advantages have been discussed in Sections 4.1 and4.2. They made up <strong>for</strong> the drawbacks <strong>of</strong> the ground-<strong>based</strong> imaging system, <strong>for</strong> instance,restraining antenna direct coupling <strong>of</strong> monostatic measurement and reflection <strong>of</strong> groundclutter, and extrapolating the imaging space by alternate diffraction stacking. Antennadirectivity compensation was also realized easily. Reconstructed 3-D images <strong>of</strong> HH, VH andVV polarization components showed good consistency with ground-truths and polarimetrytheory. Hence, radar polarimetry provided valuable scattering in<strong>for</strong>mation about targets,particularly the scattering features <strong>of</strong> trees in different seasons, so that the differentcomponents <strong>of</strong> trees in different periods could be distinguished.In Chapter 5, different analysis tools <strong>for</strong> viewing and interpreting broadband polarimetric<strong>SAR</strong> data were reviewed. Various color-coded polarimetric images were described and theirinterpretations were discussed. Broadband polarimetric images were first displayed in thisdissertation. Those could indicate scattering mechanisms <strong>for</strong> vegetation. Scatteringmechanisms <strong>of</strong> different components <strong>of</strong> vegetation, e.g. trees, were demonstrated bypolarimetric analysis techniques. We also discussed, using different data acquired <strong>for</strong> differentcases to interpret the amount <strong>of</strong> diffuse scattering, <strong>for</strong> instance, <strong>for</strong> buds, flowers and leaves. Itshowed that they all provide similar in<strong>for</strong>mation but different features could be identified bypolarimetric analysis. Moreover, it showed that the alpha/entropy decomposition provides aconvenient way to obtain more in<strong>for</strong>mation on vegetation scattering as functions <strong>of</strong> speciesand seasonal plus diurnal variations.It was demonstrated that the developed broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong> systemprovided high qualitative data <strong>for</strong> three different kinds <strong>of</strong> trees. Slight differences <strong>of</strong> specifictree features could be detected by polarimetric analysis and interpretations <strong>of</strong> acquiredground-truth data. The test results showed that the polarimetric per<strong>for</strong>mance <strong>of</strong> the <strong>SAR</strong>system is satisfactory and could be implemental <strong>for</strong> an improved understanding <strong>of</strong> theunderlying scattering mechanisms <strong>for</strong> widely distributed vegetation scatter.The most important achievements <strong>of</strong> this dissertation are the resolution estimation bysimulation <strong>of</strong> a ground-<strong>based</strong> <strong>SAR</strong>, system per<strong>for</strong>mance improvement by dihedral <strong>based</strong>polarimetric calibration, and a data comparative processing analysis with polarimetricinterpretation <strong>for</strong> broadband ground-<strong>based</strong> <strong>SAR</strong> data.It was demonstrated that the broadband ground-<strong>based</strong> polarimetric <strong>SAR</strong> system presentsadvantages <strong>for</strong> vegetation monitoring as well as environmental studies <strong>for</strong> a great variety <strong>of</strong>142
Conclusionsvegetation structures. More measurements <strong>for</strong> the vegetation structures would be desirable inorder to more clearly distinguish the associated scattering mechanisms, <strong>for</strong> instance, study onmicrowave vegetation scattering analyses, especially the diurnal (day-cycle) variations, thedependence on the relative observer versus sun plus scattering-patch locations, plus theinterlaced dependence on the diurnal vegetation-phases will be very interesting. At the sametime, a ground-<strong>based</strong> <strong>SAR</strong> system also can be used as ground truth demonstration tool <strong>for</strong>airborne <strong>SAR</strong> and space-borne <strong>SAR</strong> in a great variety <strong>of</strong> applications.143
Chapter 6144
Appendix ALIST OF ACRONYM3-D Three-DimensionalBMFCW-SFDEMFFTFUVGB-<strong>SAR</strong>GMOGPSIFFTIn<strong>SAR</strong>MPVNRCSPOL-IN-<strong>SAR</strong>POL-<strong>SAR</strong>POL-TOMO-<strong>SAR</strong>RAMRCS<strong>SAR</strong>SNRSTFTUAVBistatic Measurement FacilityContinuous-Wave Step-FrequencyDigital Elevation MapsFast Fourier Trans<strong>for</strong>mationFar Ultra Violet<strong>Ground</strong>-Based <strong>SAR</strong><strong>Ground</strong> Microwave OperationsGlobal Positioning <strong>System</strong>Inverse Fast Fourier Trans<strong>for</strong>mationInterferometric <strong>SAR</strong>Minimally Piloted VehicleNormalized Radar Cross SectionPolarimetric Interferometic <strong>SAR</strong>Polarimetric <strong>SAR</strong>Polarimetric Tomography <strong>SAR</strong>Radar Absorbent MaterialRadar Cross SectionSynthetic Aperture RadarSignal to Noise RatioShort Time Fourier Trans<strong>for</strong>mUnmanned Aerial Vehicle145
Appendix AULFVSWRVISVSWRUltra-Low-FrequenciesVoltage Standing Wave RadioVisible lightVoltage Standing Wave Radio146
Appendix BRCS OF IDEAL GEOMETRIC SCATTERERSAND THEIR SCATTERING MATRICESTable B.01 lists theoretical cross section <strong>of</strong> some ideal geometric shapes, which could at anytarget viewing angle approximate certain portions <strong>of</strong> the target’s physical features. For thephysical dimension a, a sphere produces the least RCS <strong>of</strong> any <strong>of</strong> the shapes, and its RCS isindependent <strong>of</strong> wavelength <strong>for</strong> a >> λ , which is called the optical region. The RCS <strong>of</strong> all <strong>of</strong>the other shapes can be seen to be wavelength-dependent, increasing in RCS as wavelengthdecreases.Table B.01 Radar cross section <strong>of</strong> some ideal geometric scatterersScattererDimensionDimensiondependenceFrequencydependenceCrosssectionalareaMaximumRCSSphereradius rr 2f 0π r2π r2Cylinderl x rl 2 , rf 12rl2 π rlλ2Flat platea x aa 4f 2a24 π a2λ4Dihedralcornera, a, aa 4f 22a28 πa2λ4Squaretrihedrala, a, aa 4f 23a2212 πa2λ4For measurements <strong>of</strong> radar system polarimetric calibration and <strong>SAR</strong> image calibration, weneed to know the scattering matrices <strong>of</strong> some standard scatterers. Table B.02 gives scatteringmatrices <strong>of</strong> some scatterers which have independence with scatterer orientation. Table B.03shows scattering matrices <strong>of</strong> wire and dihedral have orientated dependence.147
Appendix BTable B.02 Scattering matrices <strong>of</strong> orientation independent scatterersScattererSpherePlateTrihedralcornerRight helixLeft helixScatteringmatrix⎡1 0⎤⎢0 1⎥⎣ ⎦⎡1 0⎤⎢0 1⎥⎣ ⎦⎡1 0⎤⎢0 1⎥⎣ ⎦1 ⎡1− j ⎤2⎢− j −1⎥⎣ ⎦1 ⎡12⎢⎣jj ⎤−1⎥⎦Table B.03 Scattering matrices <strong>of</strong> scatterers with orientated dependenceOrientation(clockwise)Vertical 0 o45 oHorizontal 90 o-45 oWire⎡0 0⎤⎢0 1⎥⎣ ⎦1 ⎡1 1⎤2⎢1 1⎥⎣ ⎦⎡1 0 ⎤⎢0 0⎥⎣ ⎦1 ⎡1 −1⎤2⎢−1 1⎥⎣ ⎦Dihedralcorner⎡−1 0⎤⎢0 1⎥⎣ ⎦⎡0 1⎤⎢1 0⎥⎣ ⎦⎡1 0 ⎤⎢0 −1⎥⎣ ⎦⎡0 −1⎤⎢−1 0⎥⎣ ⎦148
BIBLIOGRAPHY[1] Boerner, W.-M., H. Mott, E. Lüneburg, C. Livingston, B. Brisco, R. J. Brown and J. S.Paterson with contributions by S.R. Cloude, E. Krogager, J. S. Lee, D. L. Schuler, J. J. van Zyl,D. Randall, P. Budkewitsch and E. Pottier, 1998, “Polarimetry in Radar Remote Sensing: Basicand Applied Concepts,” Chapter 5 in F.M. Henderson, and A.J. Lewis, (eds.), Principles and<strong>Application</strong>s <strong>of</strong> Imaging Radar, vol. 2 <strong>of</strong> Manual <strong>of</strong> Remote Sensing, (ed. R.A. Reyerson), ThirdEdition, John Willey & Sons, New York, 940 p, ISBN: 0-471-29406-3. pp. 271-358.[2] Cloude, S. R., and K. P. Papathanassiou, “Polarimetric <strong>SAR</strong> Interferometry,” IEEE Trans.Geosci. Remote Sensing, vol. 36. no. 5, pp 1551-1565, September 1998.[3] Boerner, W.-M., “Polarimetry in Remote Sensing and Imaging <strong>of</strong> Terrestrial and PlanetaryEnvironments,” Proceedings <strong>of</strong> Third International Workshop on Radar Polarimetry (JIPR-3,95), IRESTE, Univ.-Nantes, France, 1995, pp. 1-38.[4] Boerner, W.-M., “Recent advances in extra-wide-band polarimetry, interferometry andpolarimetric interferometry in synthetic aperture remote sensing, and its applications,” IEEProc.-Radar Sonar Navigation, Special Issue <strong>of</strong> the EU<strong>SAR</strong>-02, vol. 150, no. 3, June 2003, pp.113-125.[5] Kennaugh, E. M., Polarization Properties <strong>of</strong> Radar Reflections, M. Sc. Thesis, Ohio StateUniversity, Columbus, March 1952.[6] Boerner, W.-M., et al., “Polarization Dependence in Electromagnetic Inverse Problems,” IEEETrans. Antenna Propagat., vol. 29, pp. 262-274, March 1981.[7] Goldstein, D. H. and R. A. Chipman, “Optical Polarization: Measurement, Analysis, andRemote Sensing,” Proc. SPIE-3121, 1997.[8] Boerner, W.-M., et al., (Guest Eds.), Special Issue on Inverse Methods in Electromagnetics,IEEE Trans. Antenna Propagat., vol. 29, no. 2, March 1981.[9] Pottier, E., S. R. Cloude and W.-M. Boerner, “Recent Development <strong>of</strong> Data Processing inPolarimetric and Interferometric <strong>SAR</strong>; Invited Paper, Radio Science Bulletin,” (ed. R.W. Stone),Special Issue on URSI-GA-03; Maastricht; NL, RSB-304, March 2003, pp. 48-59, ISSN1024-4530.[10] Lee, J.-S., M. R. Grunes., E. Pottier, “Quantitative Comparison <strong>of</strong> Classification Capability:Fully Polarimetric Versus Dual and Single-Polarization <strong>SAR</strong>,” IEEE Trans. Geosci. RemoteSensing, vol. 39, no. 11, pp. 2343-2351, November 2001.149
Bibliography[11] Mott, H. and W.-M. Boerner, editors, “Wideband Interferometric Sensing and ImagingPolarimetry,” SPIE's Annual Mtg., Polarimetry and Spectrometry in Remote Sensing ConferenceSeries, 1997 July 27- Aug 01, San Diego Convention Center, Proc. SPIE, vol. 3120, 1997.[12] Ulaby, F. T., R. K. Moore, and A. K. Fung, Microwave Remote Sensing, vols. 1- 3, AddisonWesley, Reading, MA, 1981.[13] Boerner, W.-M., “Use <strong>of</strong> Polarization in Electromagnetic Inverse Scattering,” Radio Science, vol.16, no. 6, (Special Issue: 1980 Munich Symposium on EM Waves) , pp. 1037-1045, 1981.[14] Boerner, W.-M., Y. Yamaguchi, “Extra Wideband Polarimetry, Interferometry and PolarimetricInterferometry in Synthetic Aperture Remote Sensing,” Special Issue on Advances in Radar<strong>System</strong>s, IEICE Trans. Commun., vol. E83-B (9) , pp 1906-1915, Sept. 2000.[15] Boerner, W.-M., S. R. Cloude, and A. Moreira, “User collision in sharing <strong>of</strong> electromagneticspectrum: Frequency allocation, RF interference reduction and RF security threat mitigation inradio propagation and passive and active remote sensing,” URSI-F Open Symp., Session 3AP-1,2002 Feb. 14, Garmisch-Partenkirchen, Germany.[16] Touzi, R., “A Review <strong>of</strong> Speckle Filtering in the Context <strong>of</strong> Estimation Theory,” IEEE Trans.Geosci. Remote Sensing, vol. 40, no. 11, pp. 2392-2404, November 2002.[17] Zhou, Z.-S., K. Takasawa and M. <strong>Sato</strong>, “Interferometric Polarimetric Synthetic Aperture Radar<strong>System</strong>,” in Proc. SPIE, vol. 4548, Oct. 2001, pp. 18-23.[18] Gough, P. T. and D. W. Hawkins, “Unified Framework <strong>for</strong> Modern Synthetic Aperture ImagingAlgorithms,” Journal <strong>of</strong> Imaging <strong>System</strong>s and Technology, vol. 8, pp. 343 - 358, John Wiley &Sons, Inc., 1997.[19] Hajnsek, I., Inversion <strong>of</strong> Surface Parameters Using Polarimetric <strong>SAR</strong>, Doctoral Thesis, FSU,Jena, October 2001.[20] Lee, J.-S., D. L. Schuler and T. L. Ainsworth, “Polarimetric <strong>SAR</strong> Data Compensation <strong>for</strong>Terrain Azimuth Slope Variation,” IEEE Trans. Geosci. Remote Sensing, vol. 38, no. 5, pp.2153-2163, September 2000.[21] Lee, J.-S., M. R. Grunes, E. Pottier, L. Ferro-Famil, “Segmentation <strong>of</strong> Polarimetric <strong>SAR</strong> ImagesThat Preserves Scattering Mechanisms,”Proc. EU<strong>SAR</strong>2002, June 2002.[22] Papathanassiou, K. P., Polarimetric <strong>SAR</strong> Interferometry, Ph. D. Thesis, Tech. Univ. Graz, 1999.[23] Reigber, A., Airborne Polarimetric <strong>SAR</strong> Tomography, Doctoral Thesis, University <strong>of</strong> Stuttgart,October, 2001.[24] Reigber, A., and A. Moreira, “First demonstration <strong>of</strong> airborne <strong>SAR</strong> tomography usingmultibaseline L-band data,” IEEE Trans. Geosci. Remote Sensing, vol. 38 no. 5/1, pp. 2142-2152, Sept. 2000.150
Bibliography[25] S<strong>of</strong>ko, G. j., J. A. Koehler, M. J. McKibben, A. G. Wacker, M. R. Hinds, R. J. Brown and B.Brisco, “<strong>Ground</strong> Microwave Operations,” Canadian Journal <strong>of</strong> Remote Sensing, vol. 15, no. 1,May 1989, pp. 14-27. [also see: M. R. Hinds, Polarization Characteristics <strong>of</strong> Crops, M. Sc.Thesis, University <strong>of</strong> Saskatchewan, Saskatoon, Canada, May 1989.][26] Boisvert, J. B., Q. H. J. Gwyn, B. Brisco, D. J. Major and R. J. Brown, “Evaluation <strong>of</strong> SoilMoisture Estimation Techniques and Microwave Penetration Depth <strong>for</strong> Radar <strong>Application</strong>s,”Canadian Journal <strong>of</strong> Remote Sensing, vol. 21, no. 2, pp. 110-123, June 1995.[27] McNairn, H., C. Duguary, J. Boisvert, E. Huffman and B. Brisco, “Defining the Sensitivity <strong>of</strong>Multi-Frequency and Multi-Polarized Radar Backscatter to Post-Harvest Crop Residue,”Canadian Journal <strong>of</strong> Remote Sensing, vol. 27, no. 3, pp. 247-263, June 2001.[28] Major, D. J., F. J. Larney, B. Brisco, C. W. Lindwall and R. J. Brown, “Tillage Effects on RadarBackscatter in Southern Alberta,” Canadian Journal <strong>of</strong> Remote Sensing, vol. 19, no. 2, pp.170-176, April-May 1993.[29] Brisco, B., R. J. Brown, J. G. Gairns and B. Snider, “Temporal <strong>Ground</strong>-Based ScatterometerObservations <strong>of</strong> Crops in Western Canada,” Canadian Journal <strong>of</strong> Remote Sensing, vol. 18, no. 1,pp. 14-21, January 1992.[30] Major, D. J., A. M. Smith, M. J. Hill, W. D. Willms, B. Brisco and R. J. Brown, “RadarBackscatter and Visible Infrared Reflectance from Short-Grass Prairie,” Canadian Journal <strong>of</strong>Remote Sensing, vol. 20, no. 1, pp. 71-77, January-February 1994.[31] Brisco, B., R. J. Brown, J. A. Koehler, G. J. S<strong>of</strong>ko and M. J. McKibben, “The Diurnal Pattern <strong>of</strong>Micriwave Backscattering by Wheat,” Remote Sens. Environ., vol. 34, pp. 37-47, 1990.[32] Brisco, B., R. J. Brown, “Tillage Effects on the Radar Backscattering Coefficient <strong>of</strong> GrainStubble Fields,” Int. J. Remote Sensing, vol. 12, no. 11, pp 2283-2298, 1991.[33] Ulaby, F. T., A. Y. Nashashibi, Y. Du, “Radar <strong>System</strong>s <strong>for</strong> Backscatter and Bistatic Scatteringfrom Terrain,” ISAP 2002, [CD-ROM], Japan, 2002.[34] Hauck, B., F. T. Ulaby, R. DeRoo., “Polarimetric Bistatic-Measurement Facility <strong>for</strong> Point andDistributed Targets”, IEEE Antennas and Propagation Magazine, vol. 40, no. 1, pp. 31-41,February 1998.[35] Pieraccini, M., G. Luzi, “A High Frequency Penetrating Radar <strong>for</strong> Masonry Investigation,” GPR2002, [CD-ROM], May 2002.[36] Pieraccini, M., G. Luzi, C. Atzeni, “Terrain Mapping by <strong>Ground</strong>-Based Interferometric Radar,”IEEE Trans. Geosci. Remote Sensing, vol. 39, no. 10, pp. 2176-2181, October 2001.[37] Pieraccini, M., G. Luzi, D. Mecatti, L. N<strong>of</strong>erini and C. Atzeni, “A Microwave Radar technique<strong>for</strong> Dynamic Testing <strong>of</strong> Large Structures,” IEEE Trans. Microwave Theory Tech., vol. 51, no. 5,pp. 1603-1609, May 2003.151
Bibliography[38] Tarchi, D., E. Ohlmer and A. Sieber, “Monitoring <strong>of</strong> Structural Changes by RadarInterferometry,” Research in Nondestructive Evaluation. vol. 9, pp. 213-225, 1997.[39] Tarchi, D., E. Ohlmer, A. Sieber, “Monitoring <strong>of</strong> Structural Changes by Radar Interferometry,”Res Nondestr Eval, no. 9, pp 213-225, 1997.[40] Tarchi, D., H. Rudolf, M. Pieraccini, C. Atzeni, “Remote Monitoring <strong>of</strong> Buildings Using a<strong>Ground</strong>-Based <strong>SAR</strong>: <strong>Application</strong> to Cultural Heritage Survey,” Int. J. Remote Sensing, vol. 21,no. 18, pp 3545-3551, 2000.[41] Cazzani, L., C. Colesanti, D. Leva, G. Nesti, C. Prati, F. Rocca, D. Tarchi, “A <strong>Ground</strong>-BasedParasitic <strong>SAR</strong> Experiment,” IEEE Trans. Geosci. Remote Sensing, vol. 38, no. 5, pp. 2132-2141,September 2000.[42] Kant, Y., K. V. S. Badarinath, “<strong>Ground</strong>-Based Method <strong>for</strong> Measuring Thermal Infrared EffectiveEmissivities: Implications and Perspectives on the Measurement <strong>of</strong> Land Surface temperaturefrom Satellite Data,” Int. J. Remote Sensing, vol. 23, no. 11, pp. 2179-2191, 2002.[43] Saatchi, S. S., M. Moghaddam, “Estimation <strong>of</strong> Crown and Stem Water Content and BiomassForest Using Polarimetric <strong>SAR</strong> Imagery,” IEEE Trans. Geosci. Remote Sensing, vol. 38, no. 2,pp. 697-709, March 2000.[44] Sagues, L., J. M. Lopez-Sanchez, J. Fortuny, X. Fabregas, A. Broquetas, A. Sieber, “IndoorExperiments on Polarimetric <strong>SAR</strong> Interferometry,” IEEE Trans. Geosci. Remote Sensing, vol.38, no. 2, pp. 671-684, March 2000.[45] Krogager, E., Aspects <strong>of</strong> Polarimetric Radar Imaging, Ph.D. thesis, Technical University <strong>of</strong>Denmark (TUD), Electromagnetics Institute, Lyngby, DK, March 1993.[46] Zhou, Z.-S., and M. <strong>Sato</strong>, “Development and per<strong>for</strong>mance evaluation <strong>of</strong> a ground-<strong>based</strong> <strong>SAR</strong>system,” IEICE Tech. Rep. vol. 102, no. 84, pp. 31-36, 2002[47] Brown, W. M., “Synthetic Aperture Radar,” IEEE Trans.Aerosp. Electron. Syst, vol. AES-3, no.2, pp. 217-229, March 1967.[48] Cloude, S. R., “Polarimetry in Wave Scattering <strong>Application</strong>s,” Chapter 1.6.2 in SCATTERING,Eds. R Pike, P Sabatier, Academic Press, December 1999.[49] Fung, A. K., Microwave Scattering and Emission Models and their <strong>Application</strong>s, Artech HouseNorwood USA, 1994.[50] Lüneburg, E., “Principles <strong>of</strong> Radar Polarimetry,” IEICE Trans. on the Electronic Theory, vol.E78-C, no. 10, pp. 1339-1345, 1995.[51] Lüneburg, E., “Radar polarimetry: A Revision <strong>of</strong> Basic Concepts,” in Direct and InverseElectromagnetic Scattering, H. Serbest and S. Cloude, eds., Pittman Research Notes inMathematics Series 361, Addison Wesley Longman, Harlow, U.K., 1996, pp. 257 – 275.152
Bibliography[52] Ulaby, F. T. and C. Elachi, Editors, Radio Polarimetry <strong>for</strong> Geo-science <strong>Application</strong>s, ArtechHouse, Inc., Norwood, MA, 1990.[53] van Zyl, J. J. On the Importance <strong>of</strong> Polarization in Radar Scattering Problems, Ph.D. thesis,Cali<strong>for</strong>nia Institute <strong>of</strong> Technology, Pasadena, CA, December 1985.[54] Rudolf, H., D. Tarchi, and A. J. Sieber, “Combination <strong>of</strong> Linear and Circular <strong>SAR</strong> <strong>for</strong> 3-DFeatures,” Proc <strong>of</strong> IGARSS 1997, Singapore 3-8 August 1997, pp. 1551-1553.[55] Kulowski, A., “Algorithmic representation <strong>of</strong> the ray tracing technique,” Applied Acoustics, vol.18, no. 6, pp.449-469, 1985.[56] Kulowski, A., “Error investigation <strong>for</strong> the ray tracing technique,” Applied Acoustics, vol. 15, no.4. pp. 263-274, 1982.[57] Mott, H., Antennas <strong>for</strong> Radar and Communications: A Polarimetric Approach, John Wiley &Sons, New York, 1992.[58] Yamaguchi, Y., T. Nishikawa, M. Sengoku, W-M. Boerner and H. J. Eom, “Fundamental Studyon Synthetic Aperture FM-CW Radar Polarimetry,” IEEE Trans. Comm., vol. S77-B, no. 1, pp.73-80, January 1994.[59] Yamaguchi, Y., T. Nishikawa, M. Sengoku and W.-M. Boerner, “Two-Dimensional and FullPolarimetric Imaging by a Synthetic Aperture FM-CW Radar,” IEEE Trans. Geosci. RemoteSensing, vol. 33, no. 2, pp. 421-427, March 1995.[60] Taylor, J. D., Introduction to ultra-wideband radar systems. Boca Raton Florida: CRC Press,1995. pp. 466-468.[61] Kouyoumjian, R. G., and L. Perters, “Range requirements in radar cross-section measurements,”Proc. IEEE, vol. 53, Aug. 1965, pp. 920-928.[62] Blacksmith, JR. P., R. E. Hiatt, and R. B. Mack, “Introduction to Radar Cross-SectionMeasurements,” Proc. IEEE, vol. 53, August 1965, pp. 901-919.[63] Dybdal, R. B., “Radar Cross Section Measurements,” Proc. IEEE, vol. 75, no. 4, April 1987, pp.498-516.[64] Knott, E. F., Radar Cross Section Measurements. New York: Van Nostrand Reinhold, 1993. pp.52-195.[65] Knott, E. F., “RCS Reducation <strong>of</strong> Dihedral Corners,” IEEE Trans. Antenna Propagat., vol. 25,pp. 406-409, May 1977.[66] Griesser, T., and C. A. Balanis, “Backscatter Analysis <strong>of</strong> Dihedral Corner Reflectors UsingPhysical Optics and the Physical Theory <strong>of</strong> Diffraction,” IEEE Trans. Antenna Propagat., vol.AP-35, no. 10, pp. 1137-1147, October 1987.153
Bibliography[67] Griesser, T., and C. A. Balanis, “Dihedral Corner Reflector Backscatter Using Higher OrderRefelctions and Diffractions,” IEEE Trans. Antenna Propagat., vol. AP-35, no. 11, pp.1235-1247, November 1987.[68] Griesser, T., C. A. Balanis, and K. Liu, “RCS Analysis and Reduction <strong>for</strong> Lossy Dihedral CornerReflectors,” Proc. IEEE., vol. 77, no. 5, May 1989, pp. 806-814.[69] Anderson, W. C., “Consequences <strong>of</strong> Nonorthogonality on the Scattering Propertires <strong>of</strong> DihedralReflectors,” IEEE Trans. Antenna Propagat., vol. AP-35, no. 10, pp. 1154-1159, October 1987.[70] Wiesbeck, W., and S. Riegger, “A Complete Error Model <strong>for</strong> Free Space PolarimetricMeasurements,” IEEE Trans. Antenna Propagat., vol. 39, pp. 1105-1111, August 1991.[71] Wiesbeck, W., and D. Kähny, “Single Reference, Three Target Calibration and Error Correction<strong>for</strong> Monostatic, Polarimetric Free Space Measurements,” Proc. IEEE, vol. 79, no. 10, October1991, pp. 1551-1558.[72] Wang, S.-Y., and S.-K. Jeng, “A Compact RCS Formula <strong>for</strong> a Dihedral Corner Reflector arAribitrary Aspect Angles,” IEEE Trans. Antenna Propagat., vol. 46, pp. 1112-1113, July 1998.[73] Sarabandi, K., F. T. Ulaby and M. A. Tassoudji, “Calibration <strong>of</strong> Polarimetric Radar <strong>System</strong>swith Good Polarization Isolation,” IEEE Trans. Geosci. Remote Sensing, vol. 28, no. 1, pp.70-75, January, 1990.[74] Riegger, S., and W. Wiesbeck, “Wide-band Polarimetry and Complex Radar Cross SectionSignatures,” Proc. IEEE, vol. 77, no. 5, May 1989, pp.649-658.[75] Klein, J. D., and A. Freeman, “Quadpolarization <strong>SAR</strong> Calibration Using Target Reciprocity,” J.Electromagnetic Waves and <strong>Application</strong>s., vol. 5, no. 7, pp. 735-751, 1991.[76] Chen, T.-J., T.-H. Chu and F.-C. Chen, “A New Calibration Algorithm <strong>of</strong> Wide-BandPolarimatric Measurement <strong>System</strong>,” IEEE Trans. Antenna Propagat., vol. 39, pp. 1188-1192,August 1991.[77] Freeman, A., “<strong>SAR</strong> Calibration: An Overview,” IEEE Trans. Geosci. Remote Sensing, vol. 30,no. 6, pp.1107-1121, November 1992.[78] Freeman, A., J. J. van Zyl, J. D. Klein, H. A. Zebker, and Y. Shen, “Calibration <strong>of</strong> Stokes andScattering Matrix Format Polarimetric <strong>SAR</strong> Data,” IEEE Trans. Geosci. Remote Sensing, vol.30, no. 3, pp.531-539, May 1992.[79] Freeman, A. and S. T. Durden, “A Three-Component Scattering Model <strong>for</strong> Polarimetric <strong>SAR</strong>Data,” IEEE Trans. Geosci. Remote Sensing, vol. 36(3), pp. 963-973, 1998.[80] Gray, A. L., and P. W. Vachon, C. E. Livingstone, and T. I. Lukowski, “Synthetic Aperture RadarCalibration Using Reference Reflectors,” IEEE Trans. Geosci. Remote Sensing, vol. 28, no. 3,pp. 374-383, May 1990.154
Bibliography[81] Larson, R. W., P. L. Jackson and E. S. Kasischke, “A Digital Calibration Method <strong>for</strong> SyntheticAperture Radar <strong>System</strong>s,” IEEE Trans. Geosci. Remote Sensing, vol. 26, no. 6, pp.753-763,November, 1988.[82] Ulander, L. M. H., “Accuracy <strong>of</strong> Using Point Targets <strong>for</strong> <strong>SAR</strong> Calibration,” IEEE Trans. Aerosp.Electron. Syst, vol. 27, no. 1, pp. 139-148, January 1991.[83] Gau, J.-R. J., and W. D. Burnside, “New polarimetric calibration technique using a singlecalibration dihedral,” IEE Proc. Microw. Antennas Propag., vol. 142, no. 1, February 1995, pp.19-25.[84] Zhou, Z.-S., W.-M. Boerner and M. <strong>Sato</strong>, “Development <strong>of</strong> a <strong>Ground</strong>-Based PolarimetricBroadband <strong>SAR</strong> <strong>System</strong> <strong>for</strong> Non-Invasive <strong>Ground</strong>-Truth Validation in Vegetation Monitoring,”IEEE Transactions on Geosci. Remote Sensing, submitted.[85] Imh<strong>of</strong>f, M. L., S. Carson, and P. Johnson, “A Low Frequency Radar Sensor <strong>for</strong> VegetationBiomass Measurement,” IEEE Trans. Geosci. Remote Sensing, vol. 36, no. 6, pp. 1988-1991,1998.[86] Chiu, T., K. Sarabandi, “Electromagnetic Scattering from Short Branching Vegetation,” IEEETrans. Geosci. Remote Sensing, vol. 38, no. 2, pp. 911-925, March 2000.[87] Imh<strong>of</strong>f, M. L., “A Theoretical Analysis <strong>of</strong> the Affect Forest Structure on Synthetic ApertureRadar Backscatter and the Remote Sensing <strong>of</strong> Biomass,” IEEE Trans. Geosci. Remote Sensing,vol. 33, no.2, pp. 341-352, March 1995.[88] Soumekh, M., Synthetic Aperture Radar Signal Processing with MATLAB Algorithms, NewYork: John Willey & Sons, 1999.[89] Bracewell, R. N., The Fourier Trans<strong>for</strong>m and its <strong>Application</strong>s, Sec. Rev. Ed., McGraw Hill,New York, 1986 (Third Revised Edition: 1999).[90] Margrave, G. F., “Theory <strong>of</strong> Nonstationary Linear Filtering in the Fourier Domain with<strong>Application</strong> to Time-Variant Filtering,” Geophysics, vol. 63, no. 1, pp.244-259, Jan.-Feb., 1998.[91] Curlander, J. C., and R. N. McDonough, Synthetic aperture radar systems & signal processing.New York: John Wiley & Sons, 1991, pp. 126-152.[92] Wehner, D. R., High-Resolution Radar. 2nd ed., Norwood: Artech House, 1995, pp. 72-75.[93] Yilmaz, Ő., Seismic Data Processing. Tulsa: Society <strong>of</strong> Exploration Geophysicists, 1987, pp.403-409.[94] Stolt, R. H., “Migration by Fourier Trans<strong>for</strong>m,” Geophysics, vol. 43, no. 1, pp.23-48, Feb. 1978.[95] Chun, J. H. and C.A. Jacewitz, “Foundamentals <strong>of</strong> Frequency domain Migration,” Geophysics,vol. 46, no. 5, pp.717-733, May, 1981.155
Bibliography[96] Fortuny, J and A. J. Sieber, “Fast Algorithm <strong>for</strong> Near-Field Synthetic Aperture RadarProcessor,” IEEE Trans. Geosci. Remote Sensing, vol. 42, no. 10, pp.1458-1460, Oct. 1994.[97] Fortuny-Guash, J and J. M. Lopez-Sanchez, “Extension <strong>of</strong> 3-D Range Migration Algorithm toCylindrical and Spherical Scanning Geometries,” IEEE Trans. Geosci. Remote Sensing, vol. 49,no. 10, pp.1434-1444, May 2001.[98] Gunawarfena, A., and D. Longstaff, “Wave Equation Formulation <strong>of</strong> Synthetic Aperture Radar(<strong>SAR</strong>) Algorithms in the Time-Space Domain,” IEEE Trans. Geosci. Remote Sensing, vol. 36,no. 6, pp.1995-1999, November, 1998.[99] Caf<strong>for</strong>io, C., C. Prati, F. Rocca, “<strong>SAR</strong> data focusing using seismic migration techniques,” IEEETrans. Aerosp. Electron. Syst, vol. 27, no. 2, pp. 194-207, March 1991.[100] Leuschen, C. J and R. G. Plumb, “A Matched-Filter-Based Reverse-Time Migration Algorithm<strong>for</strong> <strong>Ground</strong>-Penetrating Radar Data,” IEEE Trans. Geosci. Remote Sensing, vol. 39, no. 5,pp.929-936, May 1996.[101] Lopez-Sanchez, J. M and J. F. Fortuny-Guasch, “3-D Radar Imaging Using Range migrationTechniques,” IEEE Trans. Geosci. Remote Sensing, vol. 48, no. 5, pp.728-737, May 2000.[102] Boag, A., “A Fast Multilevel Domain Decomposition Algorithm <strong>for</strong> Radar Imaging,” IEEEIEEE Trans. Antenna Propagat., vol. 49, no. 4, pp.666-671, April 2001.[103] Das, Y. and W.-M. Boerner, “On Radar Target Estimation Using Algorithms <strong>for</strong> Reconstructionfrom Projections”, IEEE Trans. Antenna Propagat., vol. 26, no. 2, pp. 274-279, 1978.[104] Zhou, Z.-S., and M. <strong>Sato</strong>, “<strong>Ground</strong>-Based Polarimetric <strong>SAR</strong> <strong>System</strong>s <strong>for</strong> Environment Study,”IEEE AP-S 2003 Digest, vol. 1, Columbus, June 2003, pp.202-205.[105] Yamaguchi, Y., Y. Takayanagi, W.-M. Boerner, H. J. Eom and M. Sengoku, “PolarimetricEnhancement in Radar Channel Imagery,” IEICE Trans. Communications, vol. E78-B, no. 1, pp.45-51, Jan 1996.[106] Yang, J., Y. Yamaguchi, H. Yamada, M. Sengoku, S.-M. Lin, “Optimal problem <strong>for</strong> contrastenhancement in polarimetric radar remote sensing,” IEICE Trans. Commun., vol.E82-B, no.1,pp.174-183, Jan. 1999.[107] Boerner, W.-M., C. L. Liu, and Zhang, “Comparison <strong>of</strong> Optimization Processing <strong>for</strong> 2x2Sinclair, 2x2 Graves, 3x3 Covariance, and 4x4 Mueller (Symmetric) Matrices in CoherentRadar Polarimetry and its <strong>Application</strong> to Target Versus Background Discrimination inMicrowave Remote Sensing,” EARSeL Advances in Remote Sensing, vol. 2, no. 1, pp. 55-82,1993.[108] Misti, M., Y. Misti, G. Oppenheim, and J.M. Poggi. Wavelet Toolbox User's Guide. Math WorksInc.Massachusetts, 1996.156
Bibliography[109] Strang, G. and T. Nguyen. Wavelets and Filter Banks. Wellesley-Cambridge Press, 1996.[110] Boudreaux-Bartels, G. F., “Mixed Time-frequency Signal Trans<strong>for</strong>mations,” Chapter 12 in TheTrans<strong>for</strong>ms and <strong>Application</strong>s Handbook, 2nd ed., A. D. Poularikas Ed, Boca Raton, Florida:CRC Press LLC, 2000, pp. 12.6–12.8.[111] Agrawal, A. B. and W.-M. Boerner, “Redevelopment <strong>of</strong> Kennaugh’s target characteristicpolarization state theory using the polarization trans<strong>for</strong>mation ratio <strong>for</strong>malism <strong>for</strong> the coherentcase,” IEEE Trans. Geosci. Remote Sensing, vol. 27, no.1, pp. 2-14, 1989.[112] Boerner, W.-M., W.-L. Yan, A.-Q. Xi and Y. Yamaguchi, “On the Principles <strong>of</strong> RadarPolarimetry (Invited Review): The Target Characteristic Polarization State theory <strong>of</strong> Kennaugh,Huynen's Polarization Fork Concept, and Its Extension to the Partially Polarized Case,” Proc.IEEE, Special Issue on Electromagnetic Theory, vol. 79, no. 10, Oct. 1991, pp. 1538-1550.[113] Lüneburg, E., V. Ziegler, A. Schroth, and K. Tragl, “Polarimetric Covariance Matrix Analysis <strong>of</strong>Random Radar Targets,” pp.27.1 - 27.12, in Proc. NATO-AGARD-EPP Symposium onTarget and Clutter Scattering and Their Effects on Military Radar Per<strong>for</strong>mance, Ottawa,Canada, May 6 - 10, 1991.[114] Yang, J., Y. Yamaguchi, H. Yamada, M. Sengoku, S.-M. Lin, “Stable Decomposition <strong>of</strong> MuellerMatrix,” IEICE Trans. Comm., vol. E81-B(6), pp. 1261-1268, June 1998[115] Cloude, S. R., E. Pottier, “A Review <strong>of</strong> Target Decomposition Theorems in Radar Polarimetry,”IEEE Trans. Geosci. Remote Sensing, vol. 34, no. 2, pp. 498-518, March 1996.[116] Cloude, S. R., “Uniqueness <strong>of</strong> target decomposition theorems in radar polarimetry,” In Directand Inverse Methods in Radar Polarimetry: Part 1, W.-M. Boerner (Ed.), pp. 267-296, KluwerAcademic Publishers, Dordrecht, 1992.[117] van Zyl, J. J., “An Overview <strong>of</strong> the Analysis <strong>of</strong> Multi-frequency Polarimetric <strong>SAR</strong> Data,”Proceedings <strong>of</strong> the US-AU PACRIM Significant Results Workshop, MHPCC, Kihei, Maui, HI,August 1999, pp.24 - 26.[118] Kong, J.A., S. H. Yueh, R.T. Shin, J.J. van Zyl, “Classification <strong>of</strong> Earth Terrain usingPolarimetric Synthetic Aperture Radar Images,” Chapter 6 in PIER Volume 3, ed. J.A. Kong,Elsevier, 1990.[119] Moriyama, T., M. Nakamura, Y. Yamaguchi, H. Yamada, W. -M. Boerner, “Classification <strong>of</strong>target buried in the underground by radar polarimetry,” J-IEICE Trans. Commun., vol.E82-B,no.6, pp.951-957, June 1999[120] Krogager, E., and Z. H. Czyz, “Properties <strong>of</strong> the Sphere, Di-plane and Helix Decomposition,”Proc. <strong>of</strong> 3rd International Workshop on Radar Polarimetry, IRESTE, University <strong>of</strong> Nantes,France, April 1995, pp. 106-114.[121] Krogager, E, and W.-M. Boerner, “On the Importance <strong>of</strong> Utilizing Polarimetric In<strong>for</strong>mation in157
BibliographyRadar Imaging Classification,” AGARD Proc. 582-17, April 1996, pp. 1-13.[122] Cloude, S. R., “Radar Target Decomposition Theorems,” Inst. Elect. Eng. Electron. Lett., vol. 21,no. 1, pp.22-24, January 1985.[123] Cloude, S. R, E. Pottier, “An Entropy Based Classification Scheme <strong>for</strong> Land <strong>Application</strong>s <strong>of</strong>Polarimetric <strong>SAR</strong>,” IEEE Trans. Geosci. Remote Sensing, vol. 35, no. 1, pp. 68-78, January1997.[124] Pottier, E., D.L. Schuler, J.-S. Lee, and T. Ainsworth, “Estimation <strong>of</strong> Terrain Surface Azimuthal/ Range Slopes using Polarimetric Decomposition <strong>of</strong> POL<strong>SAR</strong> Data,” Proc. IGARSS 1999.Hamburg, 28 June - 2 July 1999.[125] Treuhaft, R. N., and S. R. Cloude, “The Structure <strong>of</strong> Oriented Vegetation from PolarimetricEntropy,” IEEE Trans. Geosci. Remote Sensing, vol. 37, no. 5/2, pp. 2620-2624, September1999.[126] Cloude, S. R., E. Pottier, W.-M. Boerner, “Unsupervised Image Classification using theEntropy/Alpha/Anisotropy Method in Radar Polarimetry,” NASA-JPL, AIR<strong>SAR</strong>-02 Workshop,Double Tree Hotel, Pasadena, CA, March 2002.[127] Lee, J.-S., M. R. Grunes, T. L. Ainsworth, L. J. Du, D. L. Schuler, S. R. Cloude, “UnsupervisedClassification using Polarimetric Decomposition and the Complex Wishart Distribution,” IEEETrans. Geosci. Remote Sensing, vol. 37/1, no. 5, pp. 2249-2259, September 1999.[128] Boerner, W.-M. and H. Überall, “Advanced Research Short Course, Vector Inverse Methods inRadar Target/Clutter Imaging,” Proc. ARSC, Co-Ed., Ecole Superieur d'Electricité,Gif-sur-Yvette, France, Sept. 1-4, 1986, Springer-Verlag, Heidelberg, 1995, 428p.(ISBN3-540-57791-2).158
BIOGRAPHYName: Zheng-Shu ZHOUDate <strong>of</strong> Birth: September 22, 1968Nationality: ChinaMarital Status: MarriedSex: MalePlace <strong>of</strong> Birth: Suizhou, Hubei, ChinaHealth: ExcellentHobby: Bridge, Football, Tour etc.Education:2000.04-2003.09 Graduate School <strong>of</strong> Engineering, Tohoku University, Doctor Course,Doctoral Dissertation: <strong>Application</strong> <strong>of</strong> a <strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong> <strong>for</strong>Environmental Study.1994.09-1997.06 Graduate School <strong>of</strong> Computer Science, Central China Normal University(CCNU), Master Course, Master Dissertation: Study <strong>of</strong> HFC Computer Networkand Its Digital Modulation Technology.1986.09-1990.06 Physics Department <strong>of</strong> Central China Normal University, Bachelor <strong>of</strong>Science with a Thesis: Explosion Mechanism <strong>of</strong> Supernova SN1987A and ItsNeutrino Procedure.1983.09-1986.07 No. 2 Senior High School <strong>of</strong> Huangshi, Hubei Province.1980.09-1983.07 No. 1 Junior Middle School <strong>of</strong> Daye Iron Mine <strong>of</strong> Wuhan Steel Corp.1978.09-1980.07 No. 1 Primary School <strong>of</strong> Daye Iron Mine <strong>of</strong> Wuhan Steel Corp.1975.09-1978.07 Hongfeng Primary School <strong>of</strong> Shuanghe, Suizhou, Hubei Province.Experience:1999.12-2000.03 Center <strong>for</strong> Northeast Asian Studies, Tohoku University, Japan, Researcher1998.05-1999.08 Department <strong>of</strong> In<strong>for</strong>mation Management, CCNU, China, Lecturer1997.10-1998.04 School <strong>of</strong> In<strong>for</strong>mation and Communications, University <strong>of</strong> North London,UK, Visiting Lecturer1990.07-1997.09 Department <strong>of</strong> In<strong>for</strong>mation Management, CCNU, China, AssistantResearch Interests:Radar polarimetry, Radar remote sensing, <strong>Ground</strong>-<strong>based</strong> <strong>SAR</strong> data processing, Digital signalprocessing, Analog communication and digital communication159
BiographyPublications:Zheng-Shu Zhou, Tadashi Hamasaki and Motoyuki <strong>Sato</strong>, “Polarimetric <strong>Application</strong>s <strong>of</strong> ABroadband <strong>Ground</strong>-<strong>based</strong> <strong>SAR</strong> <strong>System</strong> to Tree Monitoring,” IEICE Tech. Rep., vol. 103,no.265, pp. 1-6, August 2003.Tadashi Hamasaki, Zheng-Shu Zhou and Motoyuki <strong>Sato</strong>, “Development <strong>of</strong> a <strong>Ground</strong>-<strong>based</strong>Interferometric <strong>SAR</strong> <strong>System</strong>,” IEICE Conference on AP, Sendai, Aug. 25, 2003 (inJapanese).Zheng-Shu Zhou, Wolfgang-Martin Boerner and Motoyuki <strong>Sato</strong>, “Development <strong>of</strong> a<strong>Ground</strong>-Based Polarimetric Broadband <strong>SAR</strong> <strong>System</strong> <strong>for</strong> Non-Invasive <strong>Ground</strong>-TruthValidation in Vegetation Monitoring,” IEEE Transactions on Geoscience and RemoteSensing, submitted.Zheng-Shu Zhou and Motoyuki <strong>Sato</strong>, “<strong>Ground</strong>-Based Polarimetric <strong>SAR</strong> <strong>System</strong>s <strong>for</strong>Environment Study”, IEEE AP-S 2003 Digest, vol. 1, pp.202-205, Columbus, June 2003.Zheng-Shu Zhou. Tadashi Hamasaki and Motoyuki <strong>Sato</strong>, “Development <strong>of</strong> Broadband<strong>Ground</strong>-<strong>based</strong> Polarimetric <strong>SAR</strong> <strong>System</strong>s <strong>for</strong> Environment Study,” IEICE GeneralConference 2003, [CD-ROM], SB-1-10, Sendai, March 2003.Zheng-Shu Zhou, Kenji Takasawa and Motoyuki <strong>Sato</strong>, “Interferometric PolarimetricSynthetic Aperture Radar <strong>System</strong>,” in Proc. SPIE, vol. 4548, pp. 18-23, Oct. 2001.Zheng-Shu Zhou and Motoyuki <strong>Sato</strong>, “Development and Per<strong>for</strong>mance Evaluation <strong>of</strong> a<strong>Ground</strong>-<strong>based</strong> <strong>SAR</strong> <strong>System</strong>,” IEICE Tech. Rep. vol. 102, no. 84, pp. 31-36, 2002.Zheng-Shu Zhou and Motoyuki <strong>Sato</strong>, “Shallow Subsurface GPR Survey in Sendai-Jo Castle,”Proc. <strong>of</strong> International Conference on GPR in Archaeology, pp. 36-37, Nara, February 2001.Guidun Pan and Zheng-Shu Zhou, “INTERNET’s New High-speed Channel – the CATVDigital Networks,” CCF Transactions on Computer Research and Development, SciencePress, vol. 35, no. 6, pp. 538-542, June 1998 (in Chinese).Zheng-Shu Zhou, “Discuss on In<strong>for</strong>mation Highway and Its Main Technologies,” in Re<strong>for</strong>mand Explore, China Archives Press, August 1997, pp. 189-192 (in Chinese).Study Report:Zheng-Shu Zhou, “Distributed Database <strong>System</strong> -- a Core Module <strong>for</strong> Master Level Courseon In<strong>for</strong>mation Management”, submitted to British Council, UK, May 1998.Debao Xiao, and Zheng-Shu Zhou, “Analysis, Design and <strong>Application</strong> <strong>of</strong> MAN Based onHFC,” submitted to Science and Technology Committee <strong>of</strong> Wuhan, China, December 1996(in Chinese).160