Application of a Ground-based SAR System for - Sato Laboratory

Application of A Ground-Based PolarimetricSAR System for Environmental StudyA Dissertation PresentedbyZheng-Shu ZhoutoThe Graduate School of Engineeringin Partial Fulfillment of the Requirements for the Degree ofDoctor of Engineeringin the Subject ofGeoscience and TechnologySupervised byProf. Motoyuki SatoReviewed byProf. Wolfgang-Martin BoernerTOHOKU UNIVERSITYAugust 2003

Abstractvarious scattering features of different components of trees, as well as during differentseasons.Finally, future perspectives on how to extend the system by implementation of recentadvances in polarimetric SAR are discussed.Key Words:Radar Polarimetry, Ground-based Synthetic Aperture Radar (GB-SAR), Radar Cross Section(RCS), Polarimetric Calibration, Diffraction Stacking, Short Time Fourier Transform (STFT),3-D Imaging, Tree Monitoring, Backscattering, Polarimetric Decomposition, EnvironmentalMonitoring.II

SAR 50MHz 20GHz 20m 1.5m III

Abstract - Japanese SAR IV

AcknowledgementsA special mention is reserved for Mr. Tadashi Hamasaki because of his friendly cooperationand assistance, especially in field experiments. I am grateful for my previous and presentcolleagues and friends in Sato Laboratory, 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) for providing me research fellowship in the year of 2001.My parents and my elder aunt deserve very special thanks for 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, for theirnever-ending patience, understanding, love and support.My heartfelt thanks to all of you, again.Zheng-Shu ZhouAugust 08, 2003 in SendaiVI

CONTENTAbstractAcknowledgementsContentList of Figures, List of TablesChapter 1 Introduction 11.1 Recent Advancements of Polarimetric SAR Imaging ··························· 11.2 Current Studies of Ground-based SAR ··············································· 61.3 Research Objectives ··········································································· 81.4 Organization of the Dissertation ························································· 10Chapter 2 Ground-based Polarimetric SAR System and Its Evaluation 132.1 System Principles ··············································································· 132.1.1 Remote Sensing and SAR Technology ·················································· 132.1.2 Radar Polarimetry ················································································· 172.1.3 Principle of Ground-based SAR System ··············································· 192.2 Simulation and Design of a Ground-based SAR System ··················· 202.2.1 Simulation of Azimuth Resolution ························································· 202.2.2 Design of a Broadband Ground-based Polarimetric SAR System ········· 262.3 Polarimetric Performance 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 of 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 for Tree Monitoring 513.1 Description of Tree Scatterers ···························································· 513.2 Experimental Procedure and Data Collection ····································· 583.2.1 Measurements of Different Tree Type at Experimental Site A ··············· 583.2.2 Measurements of a Cherry Tree at Experimental Site B ······················· 613.3 Signal Verification ··············································································· 633.3.1 Signal Check for Four Standard Reflectors ·········································· 633.3.2 Signal Verification for Different Tree Types ··········································· 673.3.3 Signal Separated by Different Filters ···················································· 723.3.4 Signal Waveform Comparison for 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 of 3-D Imaging ······································································ 864.2.2 3-D Image Reconstruction ···································································· 874.3 Verification Using Ground Truths ························································ 974.3.1 Verification of Tree Different Trees ························································ 974.3.2 Verification of a Cherry Tree ································································ 1024.4 Summary ··························································································· 107Chapter 5 Data Interpretation with SAR Polarimetric Analysis 1095.1 Broadband Frequency Polarimetric Interpretation ····························· 1095.1.1 Enhancements of Amplitude ································································ 109VIII

Content5.1.2 Data Interpretation by Broadband Images············································· 1125.2 Single Frequency Polarimetric Interpretation ····································· 1175.2.1 STFT and Spatial Frequency Transformation········································ 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 of a Cherry Tree by the Eigenvector Based Method ··· 1325.4 Summary ··························································································· 139Chapter 6 Conclusions 141AppendicesA List of Acronyms ················································································ 145B RCS of Ideal Geometric Reflectors and Scattering Matrices··············· 147Bibliography 149Biography 159IX

LIST OF FIGURESFigure 1.01 Recent advances of polarimetric SAR imaging. ·············································· 5Figure 1.02 Research objectives of this dissertation. ························································· 9Figure 1.03 Structure of the dissertation. ·········································································· 10Figure 2.01 Microwave frequency bands used for imaging radar. ···································· 14Figure 2.02 Geometry of synthetic aperture radar. ··························································· 17Figure 2.03 Transmit / receive polarization signal succession in radar polarimetry.························································································································· 18Figure 2.04 Excitation pulse for simulation. ······································································ 22Figure 2.05 Simulated images of a point target at range of 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 of 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 System diagram of a broadband ground-based polarimetric SAR. ··············· 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 of a dual polarized diagonal horn: (a) Returnloss of EMCO 3164-03 #1043, (b) Return loss of EMCO 3164-03#1044, (c) VSWR of EMCO 3164-03 #1043 and (d) VSWR of EMCO3164-03 #1044. ······························································································ 28Figure 2.11 Radiation patterns of 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-based polarimetric SAR system. ················ 30Figure 2.13 Testing scenario of a metallic sphere. ··························································· 32X

List of FiguresFigure 2.14 Geometry of creeping wave for a sphere. ····················································· 33Figure 2.15 Reflection waveforms from the metallic sphere. ············································ 34Figure 2.16 Theoretical normalized RCS curve and measured values of 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 waveforms of a dihedral corner reflector. ····································· 38Figure 2.19 Geometry of a dihedral corner reflector. ························································ 40Figure 2.20 Calculated RCS patterns of 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 of two cross-polarization terms. ································ 48Figure 3.01 Coordinate system of a measurement site: Exp#A. ······································ 52Figure 3.02 Leaves: (a) broad leaves of the tree T1: Japanese Zelkova, (b)needles of the tree T2: Japanese cedar and (c) leaves of 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 for measurement site Exp#B, a cherry tree. ··················· 55Figure 3.07 Flowers and leaves of 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 of four standard reflectors. ·································································· 64Figure 3.12 Reflected signals of 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 of FiguresFigure 3.14 Time domain signals in wiggle profiles: (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 Waveforms of point A -T1: (a) HH, (b) VH, and (c) VV. ·································· 69Figure 3.18 Waveforms of point B -T2: (a) HH, (b) VH, and (c) VV. ································· 70Figure 3.19 Waveforms of 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 Waveforms of the cherry tree at point M (-0.6, 2) m: (a) HH, (b) VH,and (c) VV. ······································································································ 78Figure 3.26 Waveforms of 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 of matched filtering. ······································································· 84Figure 4.03 Reference reflector of 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 of image reconstruction by diffraction stacking withantenna directivity compensation. ·································································· 87Figure 4.06 Shape of a band-pass filter. ··········································································· 88Figure 4.07 3-D polarimetric images of trees in spring - Exp#A-1: (a) HH, (b) VH,and (c) VV. ······································································································ 90Figure 4.08 3-D polarimetric images of trees in summer - Exp#A-2: (a) HH, (b)VH, and (c) VV. ······························································································· 91XII

List of FiguresFigure 4.09 3-D polarimetric images of trees in autumn - Exp#A-3: (a) HH, (b) VH,and (c) VV. ······································································································ 92Figure 4.10 3-D images of a cherry tree in Exp#B-1: (a) HH, (b) VH, and (c) VV ············ 94Figure 4.11 3-D images of a cherry tree in Exp#B-2: (a) HH, (b) VH, and (c) VV. ············ 95Figure 4.12 3-D images of a cherry tree in Exp#B-3: (a) HH, (b) VH, and (c) VV ············ 96Figure 4.13 Experimental scenes of 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 of 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 of 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 of trees in: (a) Exp#A-1: spring, (b) Exp#A-2:summer, and (c) Exp#A-3: autumn. ································································ 101Figure 4.17 Experimental scenes of 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 of 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 of 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 of a cherry tree in: (a) Exp#B-1: buds, (b) Exp#B-2:flowers, and (c) Exp#B-3: leaves. ··································································· 106Figure 5.01 Functions for 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 of tress in: (a) spring, (b) summer, and (c)autumn. ··········································································································· 113Figure 5.04 Broadband vertical profiles of T1 at range of 12.9m red-HHgreen-VH blue-VV in: (a) spring, (b) summer, and (c) autumn. ······················ 114Figure 5.05 Broadband vertical profiles of T2 at range of 9.8m red-HH green-VHblue-VV in: (a) spring, (b) summer, and (c) autumn. ······································ 115Figure 5.06 Broadband vertical profiles of T3 at range of 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 transform. ·········································································· 118XIII

List of FiguresFigure 5.08 Different target polarimetric descriptors. ························································ 120Figure 5.09 Single-frequency power density images of cherry blossom andleaves red-|HH| 2 green-2|VH| 2 blue-|VV| 2 : (a) imaging area of flowers,(b) 1.5 GHz, (c) 2.5 GHz, (d) 3.5 GHz, (e) 4.5 GHz images of flowers,and (f) imaging area of leaves, (g) 1.5 GHz, (h) 2.5 GHz, (i) 3.5 GHz,(j) 4.5 GHz images of leaves. ········································································· 125Figure 5.10 Polarimetric power density images of trees in spring Exp#A-1 atheight of 2.5 m based 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 of a cherry tree: (a) 1.5 GHz of Exp#B-1,(b) 4.5 GHz of Exp#B-1, (c) 1.5 GHz of Exp#B-2, (d) 4.5 GHz ofExp#B-2, (e) 1.5 GHz of Exp#B-3, and (f) 4.5 GHz of Exp#B-3. ···················· 133Figure 5.12 alpha distribution images of a cherry tree: (a) 1.5 GHz of Exp#B-1,(b) 4.5 GHz of Exp#B-1, (c) 1.5 GHz of Exp#B-2, (d) 4.5 GHz ofExp#B-2, (e) 1.5 GHz of Exp#B-3, and (f) 4.5 GHz of 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 of a part of a cherry tree- the blue squareindicated in Figure 5.13: (a) 1.5 GHz of Exp#B-1, (b) 4.5 GHz ofExp#B-1, (c) 1.5 GHz of Exp#B-2, (d) 4.5 GHz of Exp#B-2, (e) 1.5GHz of Exp#B-3, and (f) 4.5 GHz of Exp#B-3. ··············································· 136Figure 5.15 Changes of mean value: (a) Entropy and (b) alpha in the twoanalysis areas. ································································································ 137XIV

LIST OF TABLESTable 1.01 Current studies on ground-based SAR ····························································· 8Table 2.01 Parameters used for simulation ······································································ 22Table 2.02 Specifications of a developed broadband ground-based polarimetricSAR system ···································································································· 31Table 2.03 Comparison of calibrated scattering matrices with measured scatteringmatrices ·········································································································· 49Table 2.04 Auto-calibrated scattering matrices. ································································ 50Table 3.01 Condition and target situations for measurements of different tree types························································································································· 58Table 3.02 Parameters for measurements of different tree types ····································· 59Table 3.03 Data collection of measurements of different tree types ································· 60Table 3.04 Condition and target situations for measurements of a cherry tree ················ 61Table 3.05 Parameters for measurements of a Yoshino cherry tree ································· 62Table 3.06 Data formation of measurements of a Yoshino cherry tree ····························· 63Table 5.01 Schematic representation of the entropy H interpretation ······························· 131Table 5.02 Schematic representation of the alpha-angle interpretation ···························· 132Table 5.03 Comparison of mean value of Entropy and alpha in two analysis areas:red face fonts for trunk, blue face fonts for buds/flower/leaves ······················ 137Table B.01 Radar cross section of some ideal geometric scatterers ································ 147Table B.02 Scattering matrices of orientation independent scatterers ······························ 148Table B.03 Scattering matrices of scatterers with orientated dependence ······················· 148XV

Chapter 1INTRODUCTION1.1 Recent Advancements of Polarimetric SAR ImagingPolarimetry deals with the full vector nature of 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 of refraction (or permittivity,magnetic permeability, and conductivity), the polarization state of a narrow-band(single-frequency) wave is transformed, and the electromagnetic vector wave is re-polarized.When the wave passes through a medium of changing index of refraction or when it strikes anobject such as a radar target and/or a scattering surface and it is reflected; then, characteristicinformation about the reflectivity, shape and orientation of the reflecting body can be obtainedby implementing “polarization control” [1, 3, 4, 5]. The time-dependent behavior of theelectric field vector, in general describing an ellipse in a plane transverse to propagation, playsan essential role in the interaction of electromagnetic waves with material bodies, and thepropagation medium [5, 6, 7, 8]. Whereas, this polarization transformation behavior isdenoted as polarimetry in radar, and synthetic aperture radar (SAR) sensing and imaging [1,3, 11, 12] - using the ancient Greek meaning of “measuring orientation and object shape”.Thus, polarimetry is concerned with the control of the coherent polarization properties of theoptical wave and microwaves [1, 3, 4, 12].Very decisive progress was made in advancing fundamental polarimetric interferometric SARand algorithm developments during the past decade [9, 10-16], which was based on theunderlying accomplishments of fully polarimetric SAR (POLSAR) [2, 11, 14] and differentialSAR interferometry [4, 11, 14] and its current merger gradually [14, 15, 16]. This wasaccomplished with the aid of airborne & shuttle platforms supporting single-to-multi-band,multi-modal polarimetric SAR and also some polarimetric interferometric SAR(POL-IN-SAR) sensor systems, which will be compared and assessed with the aim ofestablishing the now not completed but required missions such as polarimetric tomographicand holographic imaging. There are hitherto the following tendencies of development of1

Chapter 1polarimetric SAR imaging.Extension of Time DomainFor airborne or space-borne remote sensing of earth’s surface, the all-weather day- andnight-time capability of electromagnetic radiation at microwave frequencies has attractedproponents over a number of years. Recently, the research activity in the microwave area hasaccelerated considerably, as evidenced by the number of papers and even special issuesdevoted to the topic [9-18].Expansion of Frequency DomainBoth optical [7, 12] and radar [1, 3, 12] imaging have matured considerably, and the benefitsof using one imaging modality over the other are discussed frequently [4, 14]. For example,hyper-spectral optical radiometric imaging in the range of FIR-VIS-FUV [15] is considered tobecome the exclusive remote sensing system of the twenty-first century, and thought to besuperior to ultra-wide-band (UWB) microwave SAR imaging in the range ofHF-UHF-SHF-EHF [15]. Even, it was argued that UWB SAR imaging is superfluous andcould be scrapped altogether because of 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 ofthe propagation medium as well as of 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 of 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.Therefore, every possible effort should be made to expand and to extend but not to give up theexisting, insufficient availability of free scientific remote sensing spectral windows, whichmust absolutely be spread with deca-logarithmic periodicity throughout the pertinent2

Introductionfrequency bands of about 1 MHz to 300 GHz.Expansion of Spatial RegionThe development of 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 of standard “amplitude-only” radar;with radar interferometry the spatial (in depth) structure can be explored. Because theoperation of airborne test-beds is extremely expensive, aircraft platforms are not suited forroutine monitoring missions, those are better accomplished with the use of drones (UAV),which can in addition be operated at much greater heights [4, 14]. Such unmanned aerialvehicles (drones) were hitherto developed for defense applications, however currently lackingthe sophistication for implementing advanced forefront polarimetric interferometric SARtechnology. This shortcoming will be thoroughly scrutinized resulting in the finding that wedo now need to develop most rapidly also polarimetric interferometric SAR drone-platformtechnology especially for environmental stress-change monitoring subject to severeoperational constraints due to adverse unsafe flight conditions with a great variance ofapplications beginning with flood, bush/forest-fire to tectonic-stress (earth-quake to volcaniceruptions) for real-short-time hazard mitigation. However, for routine global monitoringpurposes of the terrestrial covers neither airborne sensor implementation - aircraft and/ordrones - are sufficient; and there-fore multi-modal and multi-band space-borne polarimetricinterferometric SAR space-shuttle and satellite sensor technology needs to be furtheradvanced at a much more rapid pace. The existing ENVISAT with the forthcomingALOS-PALSAR, RADARSAT-2 and the TERRASAR will be compared in [14],demonstrating that at this phase of development the fully polarimetric andpolarimetric-interferometric SAR modes of operation must be treated as preliminaryalgorithm verification support, and at this phase of development are still not to be viewed asroutine modes. The same considerations apply to the near future implementation of anysatellite-cluster bi/multi-static space-borne tomographic imaging modes as described in [23],which must however be developed concurrently in collaboration of all major national or jointcontinental efforts in order to reduce proliferation of space-platforms and for cost-cuttingreasons. Prioritization of developmental stages need to be assessed according to applications,and will have to differ for air-borne to space-borne sensors with the aim of developing apermanently orbiting fleet of equidistantly space-distributed satellites – similar to the GPSconfiguration, however each equipped with the identical set of multi-band polarimetric SAR3

Chapter 1sensors as proposed in [4].Progress of TechniquesThere is currently widespread interest in the development of radar sensors for the detection ofsurface and buried targets and the remote sensing of land, sea and ice surfaces.Inpolarimetric interferometric SAR imaging, it is possible to recover such co-registered texturaland spatial information from POL-IN-SAR digital image data sets simultaneously, includingthe extraction of digital elevation maps (DEM) from either polarimetric (scattering matrix) orinterferometric (single platform: dual antenna) SAR systems [19, 20]. SimultaneousPOL-IN-SAR offers the additional benefit of obtaining co-registered textural plus spatialthree-dimensional POL-IN-DEM information, which when applied to repeat-passimage-overlay interferometry provides differential background validation, stress assessmentand environmental stress-change information with high accuracy. Then, by either designingmultiple dual-polarization antenna POL-IN-SAR systems or by applying advancedPOL-IN-SAR image compression techniques will result in polarimetric tomographic(multi-interferometric) SAR (POL-TOMO-SAR) imaging as was shown by Reigber [23, 24].This is of direct relevance to wide-area, dynamic battle-space surveillance and local to globalenvironmental background validation, stress assessment and stress-change monitoring of theterrestrial and planetary covers. While several airborne systems can now provide diversityover all three of these, it is the combination of polarimetry with interferometry at a singlewavelength that forms the central focus of future challenges in developing new and originaldata processing. The main reason for this is the imminent launch of a series of advancedsatellite radar systems such as PALSAR, an L-band SAR sensor on board the NASDA ALOSsatellite (in September of 2004) and RADARSAT II [4], a C-band polarimetric sensor (inspring of 2005). These are typical innovations of a new generation of radars with the potentialfor providing data from various combinations of polarimetry and interferometry.Advance Data ProcessingAn important feature of electromagnetic radiation is its state of polarization and a wide rangeof classification algorithms and inversion techniques have recently been developed based onthe transformation of 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 SAR data processing covers advances in classification of polarimetricSAR data and it is addressing the important topic of quantitative data inversion of radar databy considering applications in forest height mapping using polarimetric interferometry SARdata processing [20 ,21].Therefore, the combination of spatial information, spectral information and polarizationinformation with multi-frequency and broadband sensors defines the recent advancesaccomplished in polarimetric SAR systems development. The relation can be described inFigure 1.01. SAR is a well-known technique used for airborne or space borne remote sensing.It can usefully be exploited in ground-based radar imaging, we call it ground-based SAR. It isa new application of the conventional SAR expansion in the spatial domain.Spatial InformationRadar Imaging SystemsImaging RadarImaging RadarSpectrometersSpectral InformationRadar SpectrometersMulti-FrequencyImaging RadarPolarimetersPolarimetersMulti-FrequencyRadar PolarimetersPolarization InformationRadar PolarimetersFigure 1.01 Recent advances of polarimetric SAR imaging (according to van Zyl).By combining polarimetric and interferometric SAR techniques, it is today possible to recoversuch co-registered textural plus spatial properties simultaneously. This has importantconsequences for the design of future space and airborne sensors for carbon sequestration andclimate change studies.It has now been shown that the accelerated advancement of POLSAR techniques is of directrelevance and of utmost priority to local-to-global environmental ground-truth measurementand validation, stress assessment, and stress-change monitoring of the terrestrial and planetarycovers [3, 7, 12]. POLSAR remote sensing offers an efficient and reliable means of collecting5

Chapter 1the information 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 forestry clear cut mapping, deforestation 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 ofradar polarimetry and interferometry for radar remote sensing; and wave polarizationutilization is today of fundamental importance in the information retrieval problem ofmicrowave imaging and vector (polarization) inverse scattering.1.2 Current Studies of Ground-based SARSynthetic aperture radar is usually used for airborne or space borne remote sensing. SAR canalso advantageously be exploited in a ground-based radar imaging system. We call itGround-based SAR.A few design options [25, 33, 35, 41] of Ground-based SAR systems were proposed asmonitoring tools for agriculture, terrain mapping, a variety of large man-made structures, aswell as environmental studies and ground surface deformation detection, includingdeformation maps of static displacements of a variety of structures like concrete girders in acontrolled environment, building models, dams, and bridges, and so on.A ground-based microwave operation [25] was developed for the Canadian Center of RemoteSensing by the University of Saskatchewan. It is a three-band, ground-based scatterometersystem. Part of this activity had been motivated by the synthetic aperture microwave satelliteRADARSAT-I. Hence, they acquired a three-band, ground-based scatterometer system, whichwas operated under contract by an interdisciplinary Ground Microwave Operations (GMO)team at the University of Saskatchewan including the Departments of Agriculture and SpacePhysics. The research was directed toward a general understanding of 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 of applications usingthis system [26-32]. Of paramount importance is the well documented result of the radar6

Introductioncross section (RCS) of plants on the diurnal soil-root-plant fluid exchange cycle, whichdeserves subtle additional analyses for other climatologic regions hitherto not executed.Professor F. T. Ulaby et al. developed a bistatic measurement facility (BMF) bistatic-radarfacility for detecting a target obscured by foliage [33, 34] first at Kansas University and thenat the University of Michigan. A fully polarimetric bistatic-radar facility was constructed atthe University of Michigan, to serve as a research tool for improved understanding of thenature of bistatic scattering for point and distributed targets. The facility was capable ofoperation 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 of characterizing the radar crosssection of a point target to within ±1 dB in magnitude and ±5° in polarization phase differenceover a wide range of bistatic angles. Sample data for a point target and a distributed target arepresented.M. Pieraccini et al. at University of Florence Italy, proposed a ground-based interferometricSAR technique for terrain mapping [36]. It was based 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 for 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 based on microwave holographic images for dynamic testing of largestructures taking care of 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 Application Institute (JRC-SAI) described aninnovative application of radar interferometric techniques aimed to monitor structuraldeformations of buildings for the cultural heritage survey [40]. The proposed application wasbased on the use of ground-based instrumentation able to operate as interferometric SAR.Preliminary experimental results on a model of an historical building encourage thedevelopment of this technique for use in architectural heritage surveys.7

Chapter 1Table 1.01 Current studies of ground-based SARGroupF. Ulaby et al.M. Pieraccini et al.D. Tarchi et al.Michigan Univ.Florence Univ./JRCJRC SAISystem BMF&moblie backscatter system Ground-based SAR SAR 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 IndoorApplicationDetecting a target obscured byTerrain mapping &Structural changes offoliagebuilding monitoringartifical targetMost of these applications [35-44] of ground-based SAR were based on interferometrictechniques and/or were developed only for narrow bandwidths summarized in Table 1.01.1.3 Research ObjectivesSynthetic aperture radar using airborne and space borne platforms has attracted interest formany years. Polarimetric SAR interferometry is a microwave imaging technique forextracting geophysical and volumetric vegetation parameters from SAR images, and itsusefulness in terrain classification and surface change detection has been demonstrated [12].Typical modern methods of SAR polarimetry are radar target decomposition and classification,topographic mapping, and more recently, monitoring of ground displacements usingdifferential interferometric SAR concepts [2, 22]. SAR polarimetry can also be exploited inground-based radar imaging systems. The precise monitoring of the deformation of groundsurfaces is important in many applications. For instance, the ability to monitor volcanoactivity by observing the surface bulging before the eruption would be advantageous fordisaster prevention. For these and other applications, ground-based broadband SAR offers 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].Therefore, we planed to develop a broadband ground-based polarimetric SAR system formonitoring of different environmental features, for example, different types of vegetationsuch as trees. For application of this broadband ground-based polarimetric SAR system, thefollowing research objectives had to be met:(a)(b)(c)(d)(e)(f)Development of a broadband ground-based polarimetric SAR system based oncomputer simulation results;System evaluation with polarimetric calibration;Testing experiments for trees in natural condition;Ground-based polarimetric SAR data processing;Polarimetric analysis & interpretation of broadband ground-based SAR data; andPotential application of ground truth for airborne or space-borne SAR.Design & Development of SystemSystem Evaluation with CalibrationNatural Site TestingGround-based SAR Data ProcessingAnalysis & InterpretationPotential ApplicationFigure 1.02 Research objectives of this dissertation9

Chapter 11.4 Organization of the DissertationThe dissertation on the “Application of a Ground-based Polarimetric SAR System forEnvironmental Studies” covers ground-based SAR working principles, polarimetriccalibration, three-dimensional (3-D) image reconstruction and polarimetric SAR dataprocessing as well as interpretation for environmental monitoring. It is divided into sixchapters:Chapter 1 Introduction.Chapter 2 Ground-based Polarimetric SAR System and Its Evaluation.Chapter 3 Test Experiments for Tree Monitoring.Chapter 4 3D Image Reconstruction.Chapter 5 Data Interpretation with Polarimetric Analysis.Chapter 6 Conclusions.Following flowchart shows the relations of six chapters.Chapter 1IntroductionChapter 2Development & Evaluation of SystemChapter 3Data AcquisitionChapter 43D ImagingChapter 5Polarimetric Analysis & InterpretationChapter 6ConclusionsFigure 1.03 Structure of the dissertation10

IntroductionChapter 1 gives the introduction including recent advances of polarimetric SAR imaging, asummary on current studies of ground-based SAR, and research purposes of this dissertation.Design and features evaluation of a broadband ground-based polarimetric SAR system aredescribed in Chapter 2. In this chapter, computer simulation is used for determination ofsystem specifications and measurement parameters based on working principles ofground-based SAR. Implementing polarimetric calibration and tests, the performances of thedeveloped system are also demonstrated. Chapter 3 describes the determination and choice oftest targets, test procedures of natural environmental scatterers, and data acquisition. Bysimple signal processing, some signal waveforms are plotted for verification includingdescriptions of the testing situation. In Chapter 4, main data processing and 3-D imagingalgorithms are presented. 3-D images of two experimental sites with different conditions areverified with ground truths. Different polarimetric descriptors for 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-basedpolarimetric decomposition is discussed in this chapter. The last Chapter 6 provides theconclusions of the dissertation, including suggestions for ongoing future studies.There are two appendices that are added for supporting the text. Appendix A lists and expandsthe acronyms that are used throughout the dissertation. Appendix B provides the RCS of a fewof the common ideal geometric reflectors implemented as test targets and their scatteringmatrices. At the end of this dissertation is a list of references, which explore each topic inmore detail.11

Chapter 112

Chapter 2GROUND-BASED POLARIMETRIC SAR SYSTEMAND ITS EVALUATIONFirst, principles and resolution of a ground-based SAR will be introduced from conventionalSAR in this chapter. Based on some computer simulation results, a broadband ground-basedpolarimetric SAR system is designed and developed. Procedures of performance evaluationand polarimetric calibration are introduced. System specifications and calibration results arepresented.2.1 System Principles2.1.1 Remote Sensing and Synthetic Aperture Radar TechnologyRemote SensingRemote sensing is the science (and to some extent, art) of acquiring information 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 information [12, 14,49].In much of remote sensing, the process involves an interaction between incident radiation andthe targets of interest. This is exemplified by the use of imaging systems where manyelements are involved. However, the remote sensing technology also involves the sensing ofemitted energy and the use of non-imaging sensors.There are two main categories of remote sensing imaging systems: passive and active [12].While passive systems make use of naturally emitted, reflected or scattered radiation fromsurfaces of targets, active systems are equipped with a transmitting unit and receive the signal13

Chapter 2backscattered or reflected from the illuminated terrain. An important class of active imagingsensors are radar systems operating in the microwave region of the electromagnetic spectrum.The active operating mode makes these sensors to be independent from external illuminationsources and additionally the fact of operating at the microwave region reduces drastically theimpact of 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 forcontinuous global monitoring; and because at microwave frequencies the operatingwavelengths are of the order of effective length of isolated (point) and/or distributedscattering objects within the vegetated layer of the earth, resulting in improved resonancebehaviour for 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 of theelectromagnetic spectrum. In fact, the image characteristics of radar are also quite differentfrom that of visible and infrared images due to a quite different scattering and diffractionmechanism for generating the images.Because of the way in which microwaves interact with the atmosphere and the ground, only aselect few frequency bands are useful for imaging. These are shown in Figure 2.01. Thewavelength affects the penetration depth and also the size of a target necessary for 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 for imaging radar.14

Ground-based Polarimetric SAR System and Its EvaluationImaging radar system generally provides a two-dimensional image of the radar reflectivity ofa scene by illuminating it with microwave pulses and receiving the scattered field. There aretwo possible operation scenarios for such radar systems. The first one is to use the samesystem for transmitting and receiving. In this case receiver and transmitter are located at thesame position and therefore this configuration is known as monostatic configuration. It is theclassical operation scenario for 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 for receiving the scattered field. Such bi/multi-static configurations are up to nowmostly used for laboratory and military radar measurements but may become a prospectivealternative for cost-effective future space borne multi-static implementations such as theCart-Wheel concept [4]. In our case, the monostatic radar system is employed also for thebroadband ground-based polarimetric SAR application.In a simplified description, a monostatic radar system consists of a pulsed microwavetransmitter, an antenna which is used for transmission and reception, and a receiver unit. It ismounted on a moving platform such as a ground-based 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 platform motion in the moving direction provides the scanning in the direction of thesensor trajectory, which is referred to as azimuth or along-track direction [47, 48].Resolutions and Synthetic Aperture RadarOne of the most important quality criteria of an imaging sensor is its spatial resolution. It is ameasure of 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 for range andazimuth separately. In range direction it is inversely proportional to the bandwidth of 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 ofthe antennas. Nowadays, the achievable range resolutions are about 1 to 10 meters forspace-borne systems while airborne systems are capable to achieve resolutions on the order often centimetres and ground-based systems provide range resolution about several centimetres15

Chapter 2[4, 46].More important is the quest for achieving reasonably high resolution in the azimuth direction.The first generation of radar imaging system was the so-called real aperture radar,characterised by an azimuth resolution depending on the operating frequency, the size of thereal antenna, and the distance between sensor and the object to be imaged. Due to technicaland operational constrains on both, the size of 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 of radar sensorsoperating according to the signal processing concept of a synthetic antenna or a syntheticaperture. This radar imaging system is called synthetic aperture radar (SAR). SAR systemsare active remote sensing sensors. It generates a reflectivity map of an illuminated areathrough transmission and reception of electromagnetic energy in the microwave region.Thebasic idea of this concept is to simulate a very long antenna by moving a small real antennaalong the moving direction. The coherent integration of the received signals along the movingtrack allows synthesising a long virtual-antenna and leads to images with high azimuth spatialresolution independent of the operating frequency and the distance to the scene [47]. Figure2.02 shows geometry of SAR. The spatial resolutions of SAR 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 of light, B is the radar bandwidth,Lsaristhe synthetic aperture length of SAR, θ is the angular resolution of antenna beam, R is therange distance, λ is the wavelength of the centre frequency, L is the antenna physical size inthe azimuth direction, θsaris the angular resolution of synthetic aperture length, and Dazisthe azimuth resolution of SAR. The maximum length for the synthetic aperture is the lengthof the moving path from which a scatterer is illuminated. The angular resolution of a synthetic16

Ground-based Polarimetric SAR System and Its Evaluationaperture of the length Lsaris given by the diffraction limit (due to diffraction effects on itsaperture). The factor 2 is the result of the synthetic aperture formation [47].We can find that both range and azimuth resolutions are independent of distance to target. ForSAR, the appropriate coherent integration of the signals received from a scatterer by an arrayof antenna elements is used in order to synthesize a large antenna with a very narrow beam.The constrains on the azimuth resolution of SAR systems are given by practical limitations onthe transmitted power, and data rate, leading to resolutions of several meters for space-borneSAR, on the order of one meter or better for airborne radars and on tens of centimetres or lessfor ground-based sensors [46].LsarLAzimuthRangeθRFigure 2.02 Geometry of synthetic aperture radar.2.1.2 Radar PolarimetryMany radar systems are designed to transmit microwave radiation of orthogonal polarizationpairs, for example, that is either horizontally polarized (H) or vertically polarized (V). Atransmitted wave of either polarization can generate a backscattered wave with a variety ofpolarizations. It is the analysis of these pairs of orthogonal transmitting and receivingpolarization combinations that constitutes the science of 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 reasonsbased on the design of polarization switches, and so on, systems that transmit and receive17

Chapter 2both of these linear polarizations are commonly used. With these radars, there can be fourcombinations of transmit and receive polarizations:(a)(b)(c)(d)HH – for horizontal transmit and horizontal receiveVV – for vertical transmit and vertical receiveHV – for vertical transmit and horizontal receive, andVH – for 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 of how a radar target or surface feature scatters electromagneticenergy is expressed in terms of 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 oftransmitter and receiver, respectively. If a scattering matrix is known, the response of thetarget to any combination of 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

Ground-based Polarimetric SAR System and Its EvaluationThis is referred to as polarization synthesis, and illustrates the power and flexibility of a fullypolarimetric radar system.Both the wavelength and polarization affect how a radar system “observes” the elements inthe scene. Therefore, radar imagery collected using different polarization and wavelengthcombinations may provide different and complementary information. Furthermore, whenthree scattering matrix returns are combined in a color composite, the information is presentedin a way that an image interpreter can infer more information of the surface characteristics.2.1.3 Principle of Ground-based SAR SystemSynthetic aperture radar using space borne and airborne data has been a very interestingtechnique in the domain of remote sensing for many years. It can also be exploited in thedesign of ground-based radar imaging systems. The highly precise monitoring of thedeformation of ground surfaces is very important in many applications. For instance,monitoring of volcano activities by observing the surface bulging before the eruption wouldbe advantageous for disaster prevention due to volcano eruptions, as those are still notstraight-forwardly predictable by other methods currently in use. For these applications,ground-based broadband SAR has a wide range of applications and advantages due to theperformance of 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-based polarimetric SAR technique is employed for environmentalstudies by detecting vegetation changes of various kinds of vegetation cover due to seasonalvariations by our research group [17, 46].In a ground-based 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 SAR synthesis algorithm of 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. Therefore, 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 of light, Lx is the total horizontal scan aperture of 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 of bandwidth could be used for achieving range resolution of 0.04m, which is generallysatisfactory in practical application. The azimuth resolution will be 0.11m with azimuthaperture of 10m and range distance of 20m.2.2 Simulation and Design of a Ground-based SAR System2.2.1 Simulation of Azimuth ResolutionThe principle of SAR shows that the azimuth resolution of SAR is independent of the rangedistance. It is theoretically true when the SAR data were acquired along a very long surveyline. This ideal situation can easily be satisfied for space and air borne SAR system. But it isnot easy achieved for the ground-based SAR system. If the scanning aperture length is limited,the SAR resolution can be a function of a range distance. The azimuth resolution ofreconstructed images depends on the synthetic aperture length, media conductivity and20

Ground-based Polarimetric SAR System and Its Evaluationpermittivity, and antenna directivity. This section discusses the azimuth resolution in terms ofthese controlling parameters based on computer simulation results in order to achieve the bestperformance of the ground-based SAR system.Ray Tracing MethodRay tracing is a well-known technique in simulating the behavior of wave propagation in ahomogeneous space [55]. There are several variations of 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 of an incoming ray is the same as the incidentangle of the outgoing ray. More advanced rules which include for 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 uniform distribution of 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 of multi-reflection environments.Excitation Pulse and Generated Diffraction DataWe used the ray tracing method to simulate the measured data set of a ground-based SARwithin limited aperture length and range distance, and reconstruct the image of the target todetermine the estimated resolution of the system.For our study case, a single pulse is recovered from a broadband frequency spectrum byinverse Fourier Transform. 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 of 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 for simulation.Table 2.01 Parameters used for 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 for 12 cases,a focusing algorithm of diffraction stacking is22

Ground-based Polarimetric SAR System and Its Evaluationused for 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 of a point target at range of 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 for an aperture lengthof 5m; and 0.1m for an aperture length of 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 of 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 of 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

Ground-based Polarimetric SAR System and Its Evaluation15m and range 20m.In Figure 2.06, the azimuth resolutions are about 0.3m for aperture length of 5m, 0.12m foraperture length of 10m, and 0.1m for 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 for aperturelength of 5m, 0.15m for aperture length of 10m, and 0.12m for scanning aperture 15m whilethe range distance is 20m, respectively.According to the estimated resolution for each case, curves for the relation between scanningaperture length and estimated resolution for 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 of target to the antenna is less than 20m, the scanning aperture lengthmust remain at more than 10m for an azimuth resolution of about 0.15m which is satisfactoryfor tree monitoring. Of course, the scanning aperture length should be extended in case abetter resolution is necessary. In fact, with increase of 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 of 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 of a Broadband Ground-based Polarimetric SAR SystemSystem DescriptionBased on SAR principles and above simulation results, we have extended those earlierapproaches and designed an in-house laboratory broadband, ground-based, fully polarimetricSAR system for environmental studies. The ground-based polarimetric broadband SARtechnique is employed for environmental studies by detecting changes in various kinds ofvegetation cover due to seasonal variations by our research group.The radar system consisted of 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-based 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 of a broadband ground-based polarimetric SAR system.26

Ground-based Polarimetric SAR System and Its EvaluationDual polarizeddiagonal horn antennaVectornetworkanalyzerPositioner controllerPositionerNECPCFigure 2.08 System diagram of a broadband ground-based polarimetric SAR.A dual polarized diagonal horn is used for the broadband ground-based 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 of a testing measurement in the Anechoic Chamber, the return loss and voltagestanding wave ratio (VSWR) of 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 of # 1044 has a better polarized balance, showing that it canbe used for the polarimetric measurements and it is used for 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 for antenna patterncompensation when migration is carried out.(a)(b)(c)(d)Figure 2.10 Return loss and VSWR of a dual polarized diagonal horn: (a) Return loss ofEMCO 3164-03 #1043, (b) Return loss of EMCO 3164-03 #1044, (c) VSWR of EMCO3164-03 #1043 and (d) VSWR of EMCO 3164-03 #1044.28

Ground-based Polarimetric SAR System and Its Evaluation(a)(b)(c)(d)(e)Figure 2.11 Radiation patterns of a dual polarized diagonal horn: (a) 1 GHz, (b) 2 GHz, (c) 3GHz, (d) 4 GHz and (e) 5 GHz.29

Chapter 2System SpecificationFigure 2.12 shows a photograph of the broadband ground-based polarimetric SAR system.This system may be operated at frequencies between 400 MHz and 6 GHz, with a scanningaperture of 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 formedby 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 for antenna calibration and systemvalidation shown in Table 2.02 [46].Figure 2.12 A developed broadband ground-based polarimetric SAR system.30

Ground-based Polarimetric SAR System and Its EvaluationTable 2.02 Specifications of a developed broadband ground-based polarimetric SAR 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 Performance EvaluationFor evaluating the polarimetric ground-based SAR system, we have performed 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 for reducing reflection from target stand. Figure 2.13 shows a test scene for a metallicsphere.31

Chapter 2Figure 2.13 Testing scenario of a metallic sphere.If the range distance is not sufficiently long enough for the dual polarized diagonal broadbandhorn antenna [ETS 3164-03] for 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 of 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

Ground-based Polarimetric SAR System and Its Evaluation2⎛r⎞ +2h −2 ⎜ ⎟ r⎝r⎠∆ t = tg+t− ttarget= (2.11)cwheret t arg et: arrival time of the reflection from targettg + t : arrival time of ground reflectionr: distance between antenna and targeth: height of antenna and target from groundc: light speed in airThe time-elapse of reflections from the front of the target versus the backscatter from thetarget (the first-round creeping-wave) in Figure 2.14, for example, for a metallic sphere withradius a = 0.075 m, is approximately given as [64]2a+π aδt= = 1.286ns(2.12)caaFigure 2.14 Geometry of creeping wave for 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 of reflections from the front of the target versus that of the backscatter from thetarget (the creeping wave). Otherwise, the reflections of 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 for this choice of parameters, we caneasily distinguish the waveforms associated with the reflections of the target and the groundin Figure 2.15, which shows the returned signal of the metallic sphere in Figure 2.13.33

Chapter 2Reflection of sphereamplitudetime [ns]Figure 2.15 Reflection waveforms from the metallic sphere.The radar cross section (RCS) [61, 62] of an object exposed to a radar is a fictitious area thatdescribes the intensity of the wave reflected back to the radar, like the effective area of anantenna. The backscattering RCS of a metallic sphere can be calculated from the Mie seriesand is given by the following formulas [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 of sphere34

Ground-based Polarimetric SAR System and Its Evaluationλ : the wavelength of radar waveh(1) ( x ): the spherical Hankel function of the first kindnjn ( x ): the spherical Bessel function of the first kindyn( x ): the spherical Bessel function of the second kindThe normalized radar cross section, NRCS, is defined as the backscatter factor of 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 of the metallic sphere of radius a = 0.075 m, we obtain the theoreticalNRCS value shown as the solid line in Figure 2.16. The measured data of HH, VV and VHcomponents, compensated by the gain factor of the antenna used in the system, are alsoplotted in same figure synchronously.Figure 2.16 Theoretical normalized RCS curve and measured values of 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 waveforms of 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 of peaks and that the resonance points of the measured curves coincide withthose of the theoretical value shown in Figure 2.16. Hence, the measured backscattering dataof 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-based polarimetric SAR system, we carried outsome experiments with a few of the standard reflectors to demonstrate the polarimetricperformance of 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

Ground-based Polarimetric SAR System and Its EvaluationAccording to the reflecting signals of each standard reflector for HH, VH and VV scatteringmatrix elements, we calculated the associated scattering matrix at frequency 2 GHz for 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 of 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 of the scatteringmatrixes can be found. We can also observe the consistent relation for most test targets exceptfor the result of -45 degree wire is worse due to the strong interfering effect of the target stand.Moreover, from the HH and VV reflection waveforms of the vertical dihedral in Figure 2.18,we observe the same magnitude and inverse phases. This behavior demonstrates the goodpolarimetric performance of this system for both narrow and broadband measurements.37

Chapter 2Figure 2.18 Reflection waveforms of 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 for thebroadband ground-based polarimetric SAR system.2.4.1 Radar Cross Section of Dihedral Corner ReflectorThe radar cross section (RCS) of a target is the projected area that would intercept thetransmitted signal and reflect isotropically ally an amount that produces the returned signal at38

Ground-based Polarimetric SAR System and Its Evaluationthe receiver. In other words, radar cross section provides an indication of how well a giventarget reflects radar energy [63]. With these ideas in mind, it is not surprising that the physicalarea of a target is normally greater than the radar cross section because some of the incidentenergy is scattered and absorbed by the target.The radar cross section of a target is not constant with operating frequency. There are threebroad regions of 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 of the radarsystem, the target is said to be in the Raleigh region. If the target is in the Raleighregion, the radar cross section of the target tends to be smaller than the target'sphysical size.Resonance region. If the target is of similar dimension to that of the wavelength, thetarget is said to be in the resonance region. In the resonance region, the radar crosssection of the target may vary a great deal but tends to be larger than the physicalsize of the target.Optical region. The optical region occurs when the target is much larger than theoperating wavelength of the radar. This is quite often the case with operational radarsystems whose wavelengths are normally in the order of centimeters in length. Whenoperating in this region, the radar cross section of the target is similar to its physicalsize.The implication of these three regions is that the operating wavelength should not be selectedin total isolation from target considerations.As one of 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 for 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 of the figure will be designated l. The angle between thetwo faces is 2β and the angle of arrival of 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 off face A to face B and then back to the radar, and the fourth is a double-bouncecontribution off 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 of 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 of incident wave and k = 2 π / λ is the wavenumber.a’ b’aββbface Aface BϕIncidentdirectionFigure 2.19 Geometry of 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 formulas for 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 for either E or H polarizations, and it has beenassumed that the receiver and the transmitter are co-polarized.40

Ground-based Polarimetric SAR System and Its EvaluationThe double-bounce contributions can be estimated by finding the surface illumination of oneface to a geometrical reflection of the incident wave off the other. For double-bounce, theeffective face widths are used instead of the true widths. It is possible under certain conditionsof arrival of the incident wave that a face may not be illuminated at all by a wave reflectedfrom the other face. Therefore 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)for E polarization. The results for H polarization may be obtained by replacing sin(3 β ± ϕ)by −sin( β m ϕ).The total contribution is the sum ofSa,Sb,Saband S ba. In our experiment, a = b = l =0.3 m and 2β = 90o , and then we can calculate the RCS of dihedral with different frequencyshown in Figure 2.20.41

Chapter 2(a)(b)(c)(d)(e)Figure 2.20 Calculated RCS patterns of 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

Ground-based Polarimetric SAR System and Its Evaluation2.4.2 Polarimetric Calibration Using Dihedral Corner ReflectorsThere are number of radar polarimetric calibration algorithms and methods proposed in[70-76]. Moreover, polarimetric SAR 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 of a target is polarization dependent. For linear polarizations, a 2 x 2scattering matrix is employed for 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 of 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 ofthe 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 ofeight parameters are needed to model a general radar system according to [83].A modified polarimetric RCS calibration technique using two orientations of the dihedralcorner reflectors as calibration targets is introduced [83]. This method is valid for anymonostatic or quasi-monostatic radar system. There are these aspects to be considered:(a)(b)(c)Scattering measurement of two orientations of calibration dihedral corner reflectorsare needed;The radiation polarization effect and the four measurement channels are separated inthe model;The frequency responses of four measurement channels are modeled as mutuallyindependent and thus, no special care has to be taken for signal paths and44

Ground-based Polarimetric SAR System and Its Evaluationnonreciprocal properties in a measurement system;(d)(e)The calibration procedure is simplified by taking advantage of small cross-polarizedterms, and no complicated matrix algebra is involved;The calibration results are valid for wideband applications because no assumptionsare made relative to the theoretical solution for the calibration dihedral.Here, let us perform the derivation of 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 forthe 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 ofjAe 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⎥⎣⎤⎦for 0 degree dihedral (2.39)and⎡SS(2) (2)(2) hh hv⎡⎣S⎤⎦ = ⎢(2) (2)⎥⎢SvhSvv⎥⎣⎤⎦for 45 degree dihedral (2.40)Hence, two standard dihedral corner reflectors are used for calibration measurements asshown in Figure 2.22. Two faces of dihedral are square and all side lengths are 0.3 m.46

Ground-based Polarimetric SAR System and Its Evaluation(a)(b)Figure 2.22 Two calibrators: (a) vertical dihedral, (b) 45 o dihedral.By assumption of twosystem, we haveδ s terms neglected and of the reciprocity theorem for a monostaticSS= S = 0(2.41)(1) (1)hv vh≈ S(2.42)(2) (2)hv vhSubstituting both of measured and theoretical values of 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 of 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)Therefore, scattering matrix of 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 of cross-polarization errors are shown in Figure 2.23.Figure 2.23 Calibration coefficients of two cross-polarization terms.48

Ground-based Polarimetric SAR System and Its Evaluation2.4.3 Calibration ResultsFrom the theoretical value of 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 of calibratorssubtracted by the background signals. Using the calibration coefficients, the calibratedscattering matrix can be calculated with the aid of this equation (2.51). Table 2.03 gives themeasured scattering matrices of three standard reflectors and their calibrated scatteringmatrices of 2 GHz. We can find the improvements after calibration, especially in the phaseterm. With the use of the calibration coefficients, the auto-calibrated results of 2 GHz arepresented in Table 2.04 resulting in observable improvements. Although neglecting effects ofnoise in the calibration measurement, the improvements due to calibration have beenachieved.Table 2.03 Comparison of 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 SAR principle, a broadband ground-based polarimetric SAR system isdeveloped. First, the principle of ground-based SAR was discussed. Using ray tracing method,ground-based SAR data acquisition and imaging were simulated. The estimated azimuthresolutions agreed with the theoretical values. Therefore, a network analyzer basedground-based SAR system was assembled. Testing results with several standard reflectorsshowed the satisfactory performance of the developed system. Applying two orientationdihedral corner reflectors, a polarimetric calibration method was introduced. In the same time,Performance of a dual polarized diagonal horn antenna used in this system and backscatteringfeatures of a metal sphere and dihedral corner reflector were discussed. After calibration,the scattering matrices of standard reflectors showed satisfactory agreements with theirtheoretical values. Results demonstrated the system has very good polarimetric and broadbandperformances.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 of trees in the experimental site, Exp#A, anddata acquisition for trees were carried out in spring, summer and late autumn, respectively. Inanother experimental site, Exp#B, data were acquired for a cherry tree continually in differentgrowth states: i) being full of buds just before blooming, ii) blooming cherry blossoms, andiii) covered with lush leaves. Then, signal check and verification were performed to showhigh data quality in the third section. It is a basis for further data processing andinterpretation.3.1 Description of Tree ScatterersBased on the developed broadband ground-based polarimetric SAR system, we planed to usethis system for real measurements in environmental studies [84]. First of all, we selected treesas our research objects. The reasons are as follow:(a)(b)(c)(d)Trees are some of the most important factors of environmental studies, which werementioned in [85-87].Different types of trees have different shapes to form different scatterers, for example,deciduous tree, evergreen tree, broad-leaved tree and conifer, etc.Trees also change their shapes with different seasons, for example, bare trunk andbranches in winter and early spring for most trees and plants, being in buds in spring,blooming flowers and growth of leaves in late spring and summer, and leaves fallingoff in autumn.Water content of tree and moisture of 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 of selected trees and ground truth validation.Hence, we selected two experimental sites with different types of trees used for the broadbandground-based polarimetric SAR measurements. The experimental sites are both located at theKawauchi Campus, Tohoku University. The experimental site with three different trees iscalled Exp#A. The configuration of 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 of 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 off. 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 of 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 of a measurement site: Exp#A.Figure 3.02 shows the leaves of three trees. Leaves in Figure 3.02(a) are for the deciduoustree T1. The size of one leaf is about 10 cm in length and 5 cm in width. Figure 3.02(b) showsthe needled twigs of the conifer tree T2. On each stem with length of 15 cm, there are numberof needles with length of 1 to 2 cm. There are two main types of leaves shown in Figure3.02(c). The dimensions of the elliptically shaped leaves are from 2 cm to 6 cm. We havecarried out three measurements at the same exact position for 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 for the experiment52

Test Experiments for Tree MonitoringExp#A-1, Figure 3.04 for the experiment Exp#A-2 and Figure 3.05 for 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 off.(a) (b) (c)Figure 3.02 Leaves: (a) broad leaves of the tree T1: Japanese Zelkova, (b) needles of the treeT2: Japanese cedar and (c) leaves of 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 of tree in Japan. Cherry treeswill be in buds at mid-spring. Very flourishing flowers are usually blooming in middle Aprilin Sendai, Miyagi-ken of the northeast area of Japan. The period of flowers, known as“hanami”, continues about 7 to 10 days. After then, the flower petals fall off quickly. Soon54

Test Experiments for Tree Monitoringthereafter, fresh leaves appear at the bare stems gradually from the beginning of May. Verylush leaves cover the whole tree after another 10 days. The coordinate system and location ofthe cherry tree trunk are shown in Figure 3.06. The green color indicates the covering area ofbranches with foliage of the cherry tree. The origin locates the center of the scanning apertureon the rails. The horizontal axis is X-axis for azimuth direction, the Y-axis for the rangedirection and the vertical Z-axis for elevation, respectively.The dimensions and shapes of 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 of buds just beforethe flowers bloom; on April 14, 2003 with many blooming flowers but no leaves; and, on May27, 2003 with substantial growth of 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 for measurement site Exp#B, a cherry tree.55

Chapter 3(a)(b)Figure 3.07 Flowers and leaves of a Yoshino cherry tree.Figure 3.08 A Yoshino cherry tree in the first measurement: Exp#B-1.56

Test Experiments for 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 of Different Tree Types at the Experimental Site A:Exp#AUsing the developed ground-based polarimetric SAR system, we carried out measurements onseasonal variation of 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 of targets. Luckily due to protective requirements ofequipment operation, we successfully found good weather days to perform measurements.Table 3.01 Condition and target situations for measurements of different tree typesMeasurement numberExp#A-1Exp#A-2Exp#A-3SeasonSpringSummerAutumnWeatherFineFine & light windySunny / cloudy &windyTree 1Japanese ZelkovaBare twigs with budFlourishing largeleavesLeaves almost fall offTree 2Japanese cedarEvergreenEvergreenEvergreenTree 3Azalea&HoneysuckleBush with fresh leavesWith dense plantsWith a few leavesTable 3.02 offers the measurement parameters for the monitoring of three different trees. Weused the frequency range from 1 GHz to 5 GHz according to the antenna performanceverification in Chapter 2. It almost covers the L-band to C-band. The dual polarized diagonalhorn antenna looks forward to the targets. It possesses the capability for 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 for Tree MonitoringTable 3.02 Parameters for measurements of different tree typesExperiment Exp#A-1 Exp#A-2 Exp#A-3Date April 19, 2002 May 28, 2002 November 11, 2002Target Three types of trees Three types of trees Three types of treeKawauchi campusKawachi campusKawauchi campusLocationTohoku UniversityTohoku UniversityTohoku UniversityVNA HP 8753E HP 8753E HP 8753EStart frequency [GHz] 1 1 1Stop frequency [GHz] 5 5 5Number of points 801 801 1601IF [Hz] 100 100 100Sweep time [sec.] 10 10 25Power [dBm] 10 10 10Antenna look-forwardelevation 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 of Exp#A-3. The leaves of the broadleaf tree were falling during thisseason for Exp#A-3, but flowers were going to bloom for the first round experiment ofExp#A-1 on April 19, 2002 and leaves were very exuberant for the second round experimentExp#A-2 on May 28, 2002. Most of the measuring parameters of the third experiment werethe same as the ones for the previous measurements but the number of 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 waveform that is caused by the IFFT to the farend part of 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 for the former ones. Table 3.03 lists thedata collection of three measurements for different trees. The three scattering matrix elementdata of HH, VH and VV were acquired simultaneously. A total of 1936 points (121 x 16)data for each polarization were recorded. There are 31 KB for HH and VV data, and 33 KBfor VH data, total of 178 MB data for Exp#A-1 and Exp#A-2. There are 62 KB for HH andVV data, and 66 KB for VH data, total of 360 MB for Exp#A-3. Each data set has 801 pairnumbers for Exp#A-1 and Exp#A-2, and 1601 pair numbers for Exp#A-3, where each pairincludes the real part and imaginary parts for a frequency.Table 3.03 Data collection of measurements of different tree typesMeasurementExp#A-1Exp#A-2Exp#A-3PolarizationHH, VH, VVHH, VH, VVHH, VH, VVData formatFrequency domainwith real part andimaginary partFrequency domainwith real part andimaginary partFrequency domainwith real part andimaginary partStart frequency [GHz]111Stop frequency [GHz]555Number of points8018011601Data size of HH [KB]313162Data size of VH [KB]333366Data size of VV [KB]313162Measurement points121 x 16121 x 16121 x 16Total data size [MB]17817836060

Test Experiments for Tree Monitoring3.2.2 Measurements of a Cherry Tree at the Experimental Site B: Exp#BUsing the ground-based polarimetric broadband SAR system, we have carried out threemeasurements for a cherry tree. The three measurements were produced on April 02, 2003just before 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 for the cherry tree is shown in Table 3.04.Table 3.04 Condition and target situations for measurements of 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 for the cherry tree experiments are noted. We usedthe same frequency range same as for the former measurements. The antenna was set at alook-up elevation angle of 10 degrees in order to reduce the reflection of ground clutter. Thevector network analyzer HP 8720ES was used, which has a wideband frequency range andhigh accuracy. It possesses the capability for measuring data for 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 for measurements of 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 of 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 formation for three measurements for a cherry tree. Threepolarimetric scattering matrix element data sets, those of HH, VH and VV, were acquiredsimultaneously. A total of 1936 data points (121 x 16) for each polarimetric element wererecorded. There are 62 KB for the HH and VV data sets, and 66 KB for the VH data set, atotal of 360 MB for all three measurements. Each data set has 1601 pair numbers whichincludes the real part and imaginary parts for each frequency.62

Test Experiments for Tree MonitoringTable 3.06 Data formation of measurements of a Yoshino cherry treeMeasurementExp#B-1Exp#B-2Exp#B-3PolarizationHH, VH, VVHH, VH, VVHH, VH, VVData formatFrequency domainwith real part andimaginary partFrequency domainwith real part andimaginary partFrequency domainwith real part andimaginary partStart frequency [GHz]111Stop frequency [GHz]555Number of points160116011601Data size of HH [KB]626262Data size of VH [KB]666666Data size of VV [KB]626262Measurement points121 x 16121 x 16121 x 16Total data size [MB]3563603583.3 Signal Verification3.3.1 Signal Check for Four Standard ReflectorsOn May 30, 2002, we had performed 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-based polarimetric SAR system. The purpose of themeasurement was to demonstrate the polarimetric performance of 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 of the–45 degree metal wire are 2515 mm and 12 mm, and for the vertical wire a length of 2513mm and a diameter of 10 mm, respectively. Two faces of both the vertical dihedral and theinclined dihedral are squares. The dimensions of the side length are 300 mm. All reflectorswere set at a height of 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 of four standard reflectors.The raw data of height 1.5 m were processed by band-pass filtering. A 4096 points IFFT wasemployed for transforming the data set to time domain data set. We picked signals of fourobservation points which are corresponding to four standard reflectors and show thecorresponding waveforms 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 of the data sets recorded by the developed system.64

Test Experiments for Tree Monitoring(a)(b)(c)(d)Figure 3.12 Reflected signals of 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 of 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 of tree T2 and treeT3 in both of the band-pass filtered wiggle profile and the low-pass filtered wiggle profile,respectively. The hyperbolas show the diffraction signals for the targets identified above.Figure 3.13 Windowing functions.(a)(b)Figure 3.14 Time domain signals in wiggle profiles: (a) broadband pass filtered and (b) lowpass filtered.66

Test Experiments for Tree MonitoringDiffraction stacking has been used to reconstruct the radar image. Both of 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 of Figure3.15(b), the trunk of 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 performances of 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 for Different Tree TypesFigure 3.16 shows the configuration of experimental site Exp#A again. There are threeobservation points: A, B and C noted in the rectangular coordinate system. The coordinatesof 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 waveforms corresponding to point A, B and C. In each figure,signals of the HH, VH and VV of three experiments are drawn simultaneously. We canobserve that most waveforms 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 of acquired data is demonstrated.We picked signals of 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 for Tree Monitoring(a)(b)(c)Figure 3.17 Waveforms of point A -T1: (a) HH, (b) VH, and (c) VV.69

Chapter 3(a)(b)(c)Figure 3.18 Waveforms of point B -T2: (a) HH, (b) VH, and (c) VV.70

Test Experiments for Tree Monitoring(a)(b)(c)Figure 3.19 Waveforms of 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 of 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 of MP is from 2GHz to 4 GHz; the HP is a high pass filter from 4 GHz to 5 GHz. All of them take the formsof2F( f) = sin ( fπ/ B)(3.01)where B is the bandwidth.Figure 3.21, Figure 3.22 and Figure 3.23 show the signal waveforms corresponding to point Aof HH, VH and VV, respectively. In each sub-figure, waveforms filtered by BP, LP, MP andHP for three experiments are plotted simultaneously. We can find that most of the waveformsare similar, and only slight differences exist among different polarimetric returns andexperiments in the sub-figure processed for BP. But other waveforms divided by differentfrequency bands show quit varying shapes. For instance, in sub-figures (a), (b) and (c) ofFigure 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 for low frequency is stronger that forhigh frequency. The other one appeared around 85 ns shows a kind of 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 of 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 ofexuberant 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 information but also contain information on frequencydependence, which depicts additional characteristic in formation on the scattering sizes. Thereis the potential to obtain polarimetric images with different frequency divisions for a larger setof different plants of varying shapes and sizes and for variant seasons.72

Test Experiments for 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 for 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 for Tree Monitoring3.3.4 Signal Waveform Comparison for a Cherry TreeUsing the ground-based polarimetric broadband SAR system, we have carried out threemeasurements for a cherry tree. We have chosen data sets for 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 of the upper part with some horizontal branches. Point N was at the front of the trunk ofthe cherry tree. The coordinate system is illustrated in Figure 3.06. By using signal processingmethods, signal waveforms of observation point M are shown in Figure 3.25. Fromtime-domain signal waveforms, we can observe that slight differences appear in HH and VHsignals. For VV signal waveforms, 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 of 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 of the trunk. Changes appeared also in VH signal waveforms.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 Waveforms of the cherry tree at point M (-0.6, 2) m: (a) HH, (b) VH, and (c) VV.78

Test Experiments for Tree Monitoring(a)Trunk(b)(c)TrunkFigure 3.26 Waveforms of 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 of measurements on two experimental sites weredescribed in this chapter.Using the developed broadband ground-based polarimetric SARsystem, continued data acquisitions were carried out for three different types of 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 ofsignal verifications, good data quality has been demonstrated. Acquired data not only includeinformation about polarimetric performances for the targets but also imply frequencydependence of size-dependent scattering mechanisms by signal comparison, which cannoteasily be demonstrated by means of straight-forward implementation of either SARpolarimetry, SAR interferometry and/or polarimetric SAR interferometry, and thus addconsiderably to multimodal SAR image analysis.80

Chapter 4THREE DIMENSIONAL IMAGE RECONSTRUCTIONThe content of this chapter is to introduce the signal processing approach for the broadbandground-based polarimetric SAR data analysis. Following a signal processing flowchart,several advanced signal processing algorithms and their advantages are described, for instance,time gating, matched filtering, migration of diffraction stacking. Then 3-D spatial domain dataare synthesized by migration and 3-D images are reconstructed. Using the ground-truths, 3-Dimages for 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-based polarimetric SAR data, a series of different signalprocessing methods are applied. The processing flow chart is shown in Figure 4.01. Due todifferent data structure, ground-based SAR signal processing is not the same as for airborneSAR data processing [88, 91]. First, the acquired frequency domain data are filtered by thebroadband-pass filter; then transformed to the time domain by IFFT [89]. By selecting theinteresting range at time axis, time gating is employed to the time domain data; thenback-transformed 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 for ground-based SAR 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 for 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 often want to do time gating. Timegating is a very useful spatial filtering process for reducing antenna direct coupling and noisebeyond the region of interest. It uses a windowing function to filter a time domain signal and82

Three Dimensional Image Reconstructionpicks up the signal of interest only. The basic formulation of 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 for 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 of window shape.During the data processing procedure of broadband ground-based polarimetric SAR, timegating is used in several places, for 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 of a matchedfilter is a very general concept, common to many aspects of radar and other types of signalanalysis [91]. A filter matched, as defined below, to a given input signal can be shown to be an83

Chapter 4optimum filter for 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 of 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 of 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 of net gain through the filter and τ is the fixed component ofdelay through the filter and H(f) is the Fourier transform of h(t). The asterisk refers to thecomplex conjugate form. The algorithm of 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 of matched filtering.84

Three Dimensional Image ReconstructionMatched-filter processing of a target echo signal can be thought of as a coherent summationof the reflected signal energy from each of the target’s reflection points. The processedresponse for the entire target is the phasor summation of the individual matched-filterresponses for all of the target’s reflection points, which are generally spread in range over thetarget’s range extent. Figure 4.04 shows an example of 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 of 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 of 3-D ImagingA three-dimensional SAR image can be reconstructed by synthesizing the data set acquiredwith the two-dimensional aperture. Migrations are often used for image reconstruction bywave field extrapolation, which were discussed by many authors in [93-103]. The goal ofmigration is to make the stacked section appear similar to the real location. A migrationmethod of diffraction stacking has been employed to reconstruct the image [46, 93]. It is avery convenient and traditional method, which is commonly used for seismic data processing.This approach is an integral solution of the wave equation. It can extrapolate wave fields of anobservation aperture to larger 3-D space. Diffraction stacking migration produces onemigrated sample (a pixel) at a time by computing a diffraction shape for 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 repeatedfor all migrated samples. During summation, the amplitudes of 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 of 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 of 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 of image reconstruction by diffraction stacking with antenna directivitycompensation.4.2.2 3-D Image Reconstruction3-D Imaging of 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 informationand also smoothen the data to reduce the resonances of the time domain signal caused byinverse Fourier transforming. After processing by inverse Fourier transform, time gating isemployed to the time domain data. Due to the fact that the main targets are here located inrange of 6 meters to 14 meters, the min time of 30 ns and max time of 120 ns are selected fortime gating. It can remove the antenna direct coupling, near range ground reflections andeffects from far range reflections.Figure 4.06 Shape of a band-pass filter.Then time domain data are back-transformed to the frequency domain by FFT. The frequencydomain data are calibrated by radar system calibration coefficients which were calculatedduring system performance 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 of diffraction88

Three Dimensional Image Reconstructionstacking is applied to migrate the time domain data while antenna radiation directivitycompensation is done simultaneously, for 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 for diffraction stacking calculations. The spatial region of 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.Therefore, 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 of the HH, VHand VV polarimetric returns for three experiments, respectively.Analyzing the three polarimetric images of 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 of trees. Blue rectangle boxes and circles show the imageof tree T1, yellow rectangle boxes and circle show the image of tree T2, and red circles showthe image of 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 of 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 of 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 of trees in autumn-Exp#A-3: (a) HH, (b) VH and (c) VV.92

Three Dimensional Image Reconstruction3-D Imaging of a Cherry TreeVery similar data processing was done to acquire data for a cherry tree at experimental siteExp#B. For a clear understanding, the description of 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 of 1.2GHZ and upper frequency of 4.8 GHz are chosen for smoothing the data in order to reducethe resonances of time domain signal caused by inverse Fourier transforming. Afterprocessing by inverse Fourier transforming, time gating is employed to the time domain data.Due to fact that the cherry tree is located at a range of 10 meters to 20 meters, hence, theminimum time of 60 ns and maximum time of 150 ns are used for time gating. It can removethe antenna direct coupling, near range ground reflections and effects from far rangereflections.Then time domain data are back-transformed 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 for 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 for diffraction stacking calculations. The spatialregion of 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 of the HH, VH and VV polarimetricreturns for 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 of 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 of 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 of a cherry tree in Exp#B-3: (a) HH, (b) VH, and (c) VV.96

Three Dimensional Image Reconstruction4.3 Verification Using Ground Truths4.3.1 Verification of Three Different TreesNow we put the same polarization images together for the three measurements. 3-D HHimages of trees are shown in Figure 4.14, and VH images in Figure 4.15, VV images in Figure4.16. There are only a few of differences among 3 HH images. This indicates that the maintargets have not suffered many changes. But for the VH images, the obvious difference exists.Especially in summer, there are strong reflections from close-up regions around T3, and alsofor T1 and T2. We can assume that those changes are caused by volume scattering resultingfrom spots with a high number of inclined twigs and branches. The extent of the foliagevolumes differs with varying seasons. We can find the difference between images of 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 of targets are very similar. In VV images,there are not so many differences because there were not obvious changes from spring toautumn for 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 of Exp#A-2 due to the lush growth of 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 of leaves, twigs and small branchesof different shape and different electric properties during the different seasons. Although theshapes of the trees were very similar in late spring and late autumn, the water content of 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 of 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 of 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 of 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 of trees in: (a) Exp#A-1: spring, (b) Exp#A-2: summer, and (c)Exp#A-3: autumn.101

Chapter 4The different scattering performances of the different trees in the different seasons can beobserved clearly, and more measurements for these and other vegetation structures would bedesirable in order to more clearly distinguish the associated scattering mechanisms. Todemonstrate the detailed features of vegetations by broadband ground-based SAR system, theadditional data processing algorithms have to be used to expose the data we have acquired inChapter 5.4.3.2 Verification of a Cherry TreeSame polarization images of the cherry tree are put together again. Figure 4.18 shows the 3-Dreconstructed images of HH of the cherry tree. The main corresponding parts of 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 of 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 of cherry flowers and lush leaves during the differentmeasurements. We also found the volume of the whole tree was larger than in the former twocases and the water moisture content was higher two.Very similar images for VV are shown in Figure 4.20. There were not many differencesobserved for the vertical polarimetric returns of 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 of 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 of 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 of 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 of 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 SAR technique provides a great amount ofinformation about vegetation scatter, and that the different scattering features of 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 for broadband ground-basedpolarimetric SAR data. Algorithms of time gating, matched filtering and diffraction stackingwere described. It was demonstrated that time gating could remove antenna direct couplingand reduce reflections of ground clutter. Migration by diffraction stacking with antennadirectivity compensation is a very important technique for 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-based polarimetric SAR system. They showed good consistency with theground truths. Radar polarimetry provided valuable scattering information about targets sothat the different components of trees in different periods could be distinguished [104].107

Chapter 4108

Chapter 5DATA INTERPRETATION WITHSAR POLARIMETRIC ANALYSISWith radar polarimetry, the textural fine-structure, target-orientation and shape, symmetriesand material constituents of the Earth surface can be recovered with considerableimprovements above that of standard amplitude-only radar. By combining differentpolarisation states, it has been shown that important structural information relating tovegetation height and density can be obtained [2].In this chapter, several target polarimetric description techniques are used for interpreting theground-based SAR data. Broadband and multi-frequency polarimetric SAR data hold thepromise of providing valuable information about distributed targets. Many differentapproaches to model the expected responses have been published, and several papers havedescribed the phenomenology of multi-frequency polarimetric data. There are differentways to view polarimetric images of multi-parameter SAR data, including polarimetric colorimages in different bases, as well as broadband images.5.1 Broadband Frequency Polarimetric Interpretations5.1.1 Enhancements of AmplitudeApplying the previous processed spatial domain data of Exp#A-1 of 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 of 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 of interest was decided, the spatialdomain data of HH, VH and VV were normalized by the maximum value among whole data.The polarimetric images of 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 of 7 m. A lot of green color indicates the leaves,twigs and slant branches. Soon the leaves of tree T2 come in the picture followed by the trunkat range of 9.8 m, which causes the strongest reflections (white, the combination of HH, VHand VV). Then tree T3 disappears from the image, and instead is replaced by the inclined bigbranch of tree T1 at about 11 m. After then, there is the main trunk of tree T3 in the low partof the image at 13 m. From these contiguous images, we could roughly find the main parts oftrees and their locations only. The sensitivity for 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 of the polarimetric images, we tried to use an enhancementfunction to adjust the values of 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 for 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 of the vertical slice atthe range distance of 9.8 m in Figure 5.02. The red color shows the reflection from trunks,110

Data Interpretation with SAR 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 of 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 of trunk of tree T2 appears only. It is the realsituation since we did not change the normalized values. From Figures 5.02(b), only a fewspots of green part appear on the upper center. However, we can observe a green area inFigure 5.02(c), which corresponds to the reflections of leaves, twigs and branches of 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 for 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 of 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 of three measurements of 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 of 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 of tree T1. On the upper part indicatedby a yellow dashed line square, the blue shows the vertical branch. The reflections at twosides of trunk of 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 of 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 of tree T2. Theupper part (white solid line ellipse) indicates the leaves of 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 partof leaves and branches of tree T2 (indicated by small blue solid line circle and small bluesolid line ellipse) and difference of 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 of reflection appeared at thecorresponding position in Figure 5.06(b). There are stronger reflection in the image ofExp#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 of each ofthe images. Hence, we can assume that the main reason is volume scattering due to differentvolumes within different seasons. Although the shapes of trees were very similar, the water112

Data Interpretation with SAR Polarimetric Analysiscontent of 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 of 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 profiles of T1 at range of 12.9m red-HH green-VH blue-VVin: (a) spring, (b) summer, and (c) autumn.114

Data Interpretation with SAR 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 profiles of T2 at range of 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 profiles of T3 at range of 8.3m red-HH green-VH blue-VV in:(a) spring, (b) summer, and (c) autumn.116

Data Interpretation with SAR Polarimetric AnalysisTherefore, using the broadband polarimetric image description method, different componentsof trees and their different scattering performances in the different seasons can be detected bythe polarimetric measurement, generally.5.2 Single Frequency Polarimetric Interpretation5.2.1 Short Time Fourier Transform and Spatial-Frequency TransformationSTFTThe short time Fourier transform (STFT), also referred to as the Windowed FourierTransform, was developed in 1946 by Dennis Gabor as an attempt to overcome the lack ofinformation on time locality in Fourier analysis [108-110]. It is a compromise between thetime and the frequency view of a signal, and is thereby better suited for analysis ofnon-stationary signals than the ordinary Fourier Transform is. STFT computes atime-frequency distribution of an input signal x within a given frequency interval as asequence of short-time spectra of 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 performed 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 information on both time locality and frequency spectra, but theprecision of the information is limited and it is determined by the size of the window. Thereasons for 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 of 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 of the STFT is that all frequencies of the signal are analyzed using the sameresolution. One specific size of window is chosen as a compromise between requirements oftime and frequency resolution, and this window is then used to analyze every frequency of 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 transform.118

Data Interpretation with SAR Polarimetric AnalysisSpatial-Frequency TransformFrom Figure 5.07, we know that it includes both 3-D spatial information and frequencyspectrum information 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 transform spatial domain data to wave-number domain by Fourier transformation.We have the following transform 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 of theelectromagnetic wave in media, f is the frequency, c is the speed of light, εris therelative permittivity and µ is the relative permeability. In air ( ε = 1 and µ = 1), we haverc⋅k yf = (5.05)2πTherefore, we use this equation to obtain frequency information.rr5.2.2 Polarimetric Target DescriptorsFor deterministic scatterers, deterministic scattering or completely polarized scattering can becompletely described by the coherent Sinclair matrix. But for 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 sameinformation about polarimetric scattering amplitudes, phase angles and correlations. [T] iscloser related to physical and geometrical properties of 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 for Exp#A-1, Exp#A-2 and Exp#A-3 of three different tree measurementswere used to be transformed to spatial frequency domain by short time Fourier transform120

Data Interpretation with SAR 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 informationand 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 of tree T1.Moreove, values of the cross polarization terms inspring and autumn are smaller than the ones for 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 of 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 off, 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 for spherical waves when the122

Data Interpretation with SAR Polarimetric Analysisobservation point is sufficiently far removed from the source of the spherical wave.Therefore, if one observes a wave transmitted by a radar antenna when the wave is a largedistance from the antenna (in the far-field of 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 of 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, therefore, a scatterercan be characterized by a complex 2x2 scattering matrix [111]. However, this matrix is afunction of the radar frequency, and the viewing geometry.This coherent Sinclair scattering matrix S is a function of 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 of the associated Mueller matrix M, expressed in termsof the forward propagation coherent Jones matrix J is used extensively in polarimetricapplications; the Kennaugh matrix was found to be less expedient for analyzing polarimetricSAR image properties whereas it is of great utility in polarimetric radar meteorology forwhich 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 of averaging has to be accomplished using the coherency matrices, or thecovariance matrices. The covariance matrix is based on the components of 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 transformation. Based on equation (5.19), wecalculate covariance matrices by using the spatial frequency domain data. Therefore, thepolarimetric power density images based on diagonal terms of the covariance matrix of (5.19)for Exp#B-2 data (cherry flowers) and Exp#B-3 (cherry leaves) are shown in Figure 5.09. Thescene of cherry flowers is shown in Figure 5.09(a). Power density images of cherry flowerswith frequencies of 1.5 GHz, 2.5 GHz, 3.5 GHz and 4.5 GHz are shown on the left side ofFigure 5.09. The scene and images of cherry leaves are shown on the right side of 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 of 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 of flowers and leaves, we can find that a little more green color appear incherry tree leaves images of higher frequencies ( 3.5 GHz and 4.5 GHz). The reason may beassumed that the volume of the cherry tree leaves is greater than that of 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 SAR 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 of cherry blossom and leaves red-|HH| 2green-2|VH| 2 blue-|VV| 2 : (a) imaging area of flowers, (b) 1.5 GHz, (c) 2.5 GHz, (d) 3.5 GHz,(e) 4.5 GHz images of flowers, and (f) imaging area of leaves, (g) 1.5 GHz, (h) 2.5 GHz, (i)3.5 GHz, (j) 4.5 GHz images of 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) for symmetric matrix case as.1Tk = [ Shh + Svv Shh − Svv 2Svh](5.20)2Another way to visualize the polarimetric information is to transform to the Pauli spin matrixbasis, first introduced in this context by Cloude [116]. In the basis of 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 ofthe first term of 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 of 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 of the Pauli covariance matrix are just the magnitudes of thethree elements of the target vector squared. Therefore, if we display the three diagonalelements of the Pauli covariance matrix as red for the first term, green for the second term andblue for 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 of the diagonal terms of 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 SAR 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 of 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 of 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. Therefore, one has to be careful when interpreting the colorsin terms of scattering mechanisms. A more correct interpretation is that red colors indicatescattering more indicative of single reflections than double reflections, with the reverse truefor 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 of Exp#A-1 for different trees by STFT. Hence, the polarimetric power density imagesbased on Pauli covariance matrix for Exp#A-1 are shown in Figure 5.10. Here data atfrequencies of 1.5 GHz, 2.5 GHz, 3.5 GHz and 4.5 GHz are displayed. We show singlefrequency images of different types of trees in spring. For lower frequency, the main parts oftrees 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 of 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 of trees in spring Exp#A-1 at height of 2.5 mbased 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 SAR Polarimetric Analysis5.3 Entropy Based Polarimetric Target DecompositionIn this section, the main polarimetric parameters which can be used for the extraction ofphysical information about distributed scatterers from polarimetric coherence matrix data areintroduced and discussed.There are many polarimetric target decomposition methods [117-127] based on the covariancematrix and the coherency matrix. Here, we show one of the currently most usefuleigenvector-based decomposition; it is the polarimetric H/alpha decomposition developed byS.R. Cloude & E. Pottier [123].Based on these eigenvalues and eigenvectors of a coherency matrix, they defined three usefulamong five parameters that characterize the scattering property of 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 of 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)for 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 transformation of the form [115, 116, 122, 126]. Theidea of the eigenvector approach is to use the diagonalisation of the coherence matrix [T] of adistributed scatterer, which is in general of rank 3, in order to decompose it into thenon-coherent sum of three independent coherence matrices [T i ].5.3.2 Polarimetric Entropy H and Angle αCloude defined a parameter, polarimetric entropy H, to account for randomness in distributedtargets’ scattering mechanisms. First, the probabilities of 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 of polarimetric entropy is described here. For a deterministic scattererwith entropy H = 0, the anisotropy becomes zero, as well as in the case of a pure randomscatterer with entropy H = 1. For scatterers characterised by intermediate entropy values, ahigh anisotropy indicates the presence of only one strong secondary scattering process, whilea low anisotropy indicates the appearance of two equally strong scattering processes.Schematic representation of the entropy H interpretation is listed in Table 5.01.130

Data Interpretation with SAR Polarimetric AnalysisTable 5.01 Schematic representation of the entropy H interpretationH value0

Chapter 5Table 5.02 Schematic representation of the alpha-angle interpretationalpha value 0 o0 o < α

Data Interpretation with SAR Polarimetric Analysis(a)Entropy(b)Entropy(c)Entropy(d)Entropy(e)Entropy(f)EntropyFigure 5.11 Entropy distribution images of a cherry tree: (a) 1.5 GHz of Exp#B-1, (b) 4.5GHz of Exp#B-1, (c) 1.5 GHz of Exp#B-2, (d) 4.5 GHz of Exp#B-2, (e) 1.5 GHz of Exp#B-3,and (f) 4.5 GHz of 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 of a cherry tree: (a) 1.5 GHz of Exp#B-1, (b) 4.5 GHzof Exp#B-1, (c) 1.5 GHz of Exp#B-2, (d) 4.5 GHz of Exp#B-2, (e) 1.5 GHz of Exp#B-3, and (f)4.5 GHz of Exp#B-3.134

Data Interpretation with SAR Polarimetric AnalysisFor precisely detecting small changes of a cherry tree during different periods, two analysisareas with three measurements for 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 of 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 for 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 for the lower frequency. They indicate that some branches, flowers and leaves appear asrandom scatterers with larger entropy for the lower frequency. But the same targets showfeatures of close to stationary scattering behavior with smaller entropy for 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 of a part of a cherry tree- the blue square indicatedin Figure 5.13: (a) 1.5 GHz of Exp#B-1, (b) 4.5 GHz of Exp#B-1, (c) 1.5 GHz of Exp#B-2,(d) 4.5 GHz of Exp#B-2, (e) 1.5 GHz of Exp#B-3, and (f) 4.5 GHz of Exp#B-3.136

Data Interpretation with SAR Polarimetric AnalysisNow the average entropy values and alpha values are calculated for two analysis areas withthree measurements, respectively. Mean values of Entropy and alpha for two indicatedanalysis areas are listed in Table 5.03. Differences of entropy and alpha are shown in Figure5.15 (a) and Figure 5.15 (b).Table 5.03 Comparison of mean value of Entropy and alpha in two analysis areas: red facefonts for trunk, blue face fonts for 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 of 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 of lower frequency (1.5 GHz) almost did not change in Figure 5.15(a). Butentropies of higher frequency (4.5 GHz) increased slightly along three measurements.That means the scatterers are more sensitive at higher frequencies than for lowerfrequencies during changing growth conditions.Entropies at lower frequencies (1.5 GHz) are greater than the ones of higherfrequencies (4.5 GHz) among three measurements. The fact is that scatterers appearas random scatterers with larger entropy for the lower frequency and show featuresof close to stationary scattering behavior with smaller entropy for higher frequencies.For high frequency, the entropy of the trunk is lower than the value forbranches/flowers/leaves due to their stationary scattering features.alpha values of both low and high frequencies for 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 of leaves on brancheshave larger reflection surfaces than the bare small branches and tiny flower-stemsdisplay.At low frequencies, EM waves propagate into the canopy of a cherry tree and backscatter alittle from some of the larger branches inside the tree canopy. It shows larger entropy due torandom scattering of a mixture of scatterers.At high frequencies, the electromagnetic wave scatters from the canopy of the tree. For thesecond measurement (cherry tree in bloom) and the third measurement (leaves), the entropyvalues are almost the same as for 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 of the cherry tree.138

Data Interpretation with SAR Polarimetric Analysis5.4 SummaryIn this chapter, we have introduced a number of different analysis tools for viewing andinterpreting broadband and single-frequency polarimetric data. Various color images weredescribed and their interpretations were discussed. Scattering mechanisms of differentcomponents of vegetation, e.g. trees, were demonstrated by polarimetric analysis techniques.We also discussed different measurements to interpret the amount of diffuse scattering, forinstance, buds, flowers and leaves,and showed that they all provide similar informationand different features by polarimetric analysis. Finally, it showed that the alpha/entropydecomposition provides a convenient way to recover more information about vegetationscattering.It showed that the developed broadband ground-based polarimetric SAR system providedhigh qualitative data for trees. Slight differences of trees could be detected by polarimetricanalysis and interpretations for the acquired data in dependence of tree type and seasonalgrowth patterns.139

Chapter 5140

Chapter 6CONCLUSIONSIn this dissertation, the development of a broadband ground-based SAR system and itsapplication to environmental studies are described. We have used a newly developedground-based polarimetric broadband SAR system to monitor seasonal variations in trees.In Chapter 2, based on the conventional SAR formulation, the principle of a ground-basedSAR was derived. By computer simulation, the estimated resolution was derived that wasconsistent with theoretical resolution. Including the description of the pertinent measurementequipments, specifications of a broadband ground-based polarimetric SAR system werepresented. Testing results showed satisfactory polarimetric performances of the developedsystem. Further a modified two-way oriented dihedral based polarimetric calibration wasintroduced and its use was verified. Theoretical RCS calculations for a metallic sphere and adihedral corner reflector were also performed. Improvement of the specific polarimetriccharacteristics for the broadband ground-based SAR system was determined by polarimetriccalibration results.Due to the fact that trees are some of the most important vegetation scatterers inenvironmental studies, we implemented the developed SAR system for real measurements forthe monitoring of different tree types and during different seasons. Target and distributedscatterer determination and procedures of measurements for two experimental sites aredescribed in the Chapter 3. Using the developed broadband ground-based polarimetric SARsystem, continued data acquisitions were carried out for three different types of trees in threeseasons, and for 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-based SARsystem. It demonstrated that the acquired data not only included information about targetpolarimetric performances but also frequency regime dependence of differently sizedcomponents of trees by signal comparison.Chapter 4 was to review signal processing methods for broadband ground-based polarimetric141

Chapter 6SAR data. Algorithms of time gating, matched filtering and diffraction stacking were veryeffective techniques for ground-based SAR data processing, especially the time gating methodand diffraction stacking migration. The advantages have been discussed in Sections 4.1 and4.2. They made up for the drawbacks of the ground-based imaging system, for instance,restraining antenna direct coupling of monostatic measurement and reflection of groundclutter, and extrapolating the imaging space by alternate diffraction stacking. Antennadirectivity compensation was also realized easily. Reconstructed 3-D images of HH, VH andVV polarization components showed good consistency with ground-truths and polarimetrytheory. Hence, radar polarimetry provided valuable scattering information about targets,particularly the scattering features of trees in different seasons, so that the differentcomponents of trees in different periods could be distinguished.In Chapter 5, different analysis tools for viewing and interpreting broadband polarimetricSAR 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 for vegetation. Scatteringmechanisms of different components of vegetation, e.g. trees, were demonstrated bypolarimetric analysis techniques. We also discussed, using different data acquired for differentcases to interpret the amount of diffuse scattering, for instance, for buds, flowers and leaves. Itshowed that they all provide similar information but different features could be identified bypolarimetric analysis. Moreover, it showed that the alpha/entropy decomposition provides aconvenient way to obtain more information on vegetation scattering as functions of speciesand seasonal plus diurnal variations.It was demonstrated that the developed broadband ground-based polarimetric SAR systemprovided high qualitative data for three different kinds of trees. Slight differences of specifictree features could be detected by polarimetric analysis and interpretations of acquiredground-truth data. The test results showed that the polarimetric performance of the SARsystem is satisfactory and could be implemental for an improved understanding of theunderlying scattering mechanisms for widely distributed vegetation scatter.The most important achievements of this dissertation are the resolution estimation bysimulation of a ground-based SAR, system performance improvement by dihedral basedpolarimetric calibration, and a data comparative processing analysis with polarimetricinterpretation for broadband ground-based SAR data.It was demonstrated that the broadband ground-based polarimetric SAR system presentsadvantages for vegetation monitoring as well as environmental studies for a great variety of142

Conclusionsvegetation structures. More measurements for the vegetation structures would be desirable inorder to more clearly distinguish the associated scattering mechanisms, for 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-based SAR system also can be used as ground truth demonstration tool forairborne SAR and space-borne SAR in a great variety of applications.143

Chapter 6144

Appendix ALIST OF ACRONYM3-D Three-DimensionalBMFCW-SFDEMFFTFUVGB-SARGMOGPSIFFTInSARMPVNRCSPOL-IN-SARPOL-SARPOL-TOMO-SARRAMRCSSARSNRSTFTUAVBistatic Measurement FacilityContinuous-Wave Step-FrequencyDigital Elevation MapsFast Fourier TransformationFar Ultra VioletGround-Based SARGround Microwave OperationsGlobal Positioning SystemInverse Fast Fourier TransformationInterferometric SARMinimally Piloted VehicleNormalized Radar Cross SectionPolarimetric Interferometic SARPolarimetric SARPolarimetric Tomography SARRadar Absorbent MaterialRadar Cross SectionSynthetic Aperture RadarSignal to Noise RatioShort Time Fourier TransformUnmanned 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 of some ideal geometric shapes, which could at anytarget viewing angle approximate certain portions of the target’s physical features. For thephysical dimension a, a sphere produces the least RCS of any of the shapes, and its RCS isindependent of wavelength for a >> λ , which is called the optical region. The RCS of all ofthe other shapes can be seen to be wavelength-dependent, increasing in RCS as wavelengthdecreases.Table B.01 Radar cross section of 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 of radar system polarimetric calibration and SAR image calibration, weneed to know the scattering matrices of some standard scatterers. Table B.02 gives scatteringmatrices of some scatterers which have independence with scatterer orientation. Table B.03shows scattering matrices of wire and dihedral have orientated dependence.147

Appendix BTable B.02 Scattering matrices of 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 of 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 andApplications of Imaging Radar, vol. 2 of Manual of 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 SAR 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 of Terrestrial and PlanetaryEnvironments,” Proceedings of 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 of the EUSAR-02, vol. 150, no. 3, June 2003, pp.113-125.[5] Kennaugh, E. M., Polarization Properties of 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 of Data Processing inPolarimetric and Interferometric SAR; 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 of Classification Capability:Fully Polarimetric Versus Dual and Single-Polarization SAR,” 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 of 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 RadarSystems, 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 of 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 of Speckle Filtering in the Context of Estimation Theory,” IEEE Trans.Geosci. Remote Sensing, vol. 40, no. 11, pp. 2392-2404, November 2002.[17] Zhou, Z.-S., K. Takasawa and M. Sato, “Interferometric Polarimetric Synthetic Aperture RadarSystem,” in Proc. SPIE, vol. 4548, Oct. 2001, pp. 18-23.[18] Gough, P. T. and D. W. Hawkins, “Unified Framework for Modern Synthetic Aperture ImagingAlgorithms,” Journal of Imaging Systems and Technology, vol. 8, pp. 343 - 358, John Wiley &Sons, Inc., 1997.[19] Hajnsek, I., Inversion of Surface Parameters Using Polarimetric SAR, Doctoral Thesis, FSU,Jena, October 2001.[20] Lee, J.-S., D. L. Schuler and T. L. Ainsworth, “Polarimetric SAR Data Compensation forTerrain 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 of Polarimetric SAR ImagesThat Preserves Scattering Mechanisms,”Proc. EUSAR2002, June 2002.[22] Papathanassiou, K. P., Polarimetric SAR Interferometry, Ph. D. Thesis, Tech. Univ. Graz, 1999.[23] Reigber, A., Airborne Polarimetric SAR Tomography, Doctoral Thesis, University of Stuttgart,October, 2001.[24] Reigber, A., and A. Moreira, “First demonstration of airborne SAR tomography usingmultibaseline L-band data,” IEEE Trans. Geosci. Remote Sensing, vol. 38 no. 5/1, pp. 2142-2152, Sept. 2000.150

Bibliography[25] Sofko, G. j., J. A. Koehler, M. J. McKibben, A. G. Wacker, M. R. Hinds, R. J. Brown and B.Brisco, “Ground Microwave Operations,” Canadian Journal of Remote Sensing, vol. 15, no. 1,May 1989, pp. 14-27. [also see: M. R. Hinds, Polarization Characteristics of Crops, M. Sc.Thesis, University of Saskatchewan, Saskatoon, Canada, May 1989.][26] Boisvert, J. B., Q. H. J. Gwyn, B. Brisco, D. J. Major and R. J. Brown, “Evaluation of SoilMoisture Estimation Techniques and Microwave Penetration Depth for Radar Applications,”Canadian Journal of 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 ofMulti-Frequency and Multi-Polarized Radar Backscatter to Post-Harvest Crop Residue,”Canadian Journal of 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 of 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 Ground-Based ScatterometerObservations of Crops in Western Canada,” Canadian Journal of 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 ofRemote Sensing, vol. 20, no. 1, pp. 71-77, January-February 1994.[31] Brisco, B., R. J. Brown, J. A. Koehler, G. J. Sofko and M. J. McKibben, “The Diurnal Pattern ofMicriwave 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 of GrainStubble Fields,” Int. J. Remote Sensing, vol. 12, no. 11, pp 2283-2298, 1991.[33] Ulaby, F. T., A. Y. Nashashibi, Y. Du, “Radar Systems for Backscatter and Bistatic Scatteringfrom Terrain,” ISAP 2002, [CD-ROM], Japan, 2002.[34] Hauck, B., F. T. Ulaby, R. DeRoo., “Polarimetric Bistatic-Measurement Facility for 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 for Masonry Investigation,” GPR2002, [CD-ROM], May 2002.[36] Pieraccini, M., G. Luzi, C. Atzeni, “Terrain Mapping by Ground-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. Noferini and C. Atzeni, “A Microwave Radar techniquefor Dynamic Testing of 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 of Structural Changes by RadarInterferometry,” Research in Nondestructive Evaluation. vol. 9, pp. 213-225, 1997.[39] Tarchi, D., E. Ohlmer, A. Sieber, “Monitoring of 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 of Buildings Using aGround-Based SAR: Application 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 Ground-BasedParasitic SAR Experiment,” IEEE Trans. Geosci. Remote Sensing, vol. 38, no. 5, pp. 2132-2141,September 2000.[42] Kant, Y., K. V. S. Badarinath, “Ground-Based Method for Measuring Thermal Infrared EffectiveEmissivities: Implications and Perspectives on the Measurement of Land Surface temperaturefrom Satellite Data,” Int. J. Remote Sensing, vol. 23, no. 11, pp. 2179-2191, 2002.[43] Saatchi, S. S., M. Moghaddam, “Estimation of Crown and Stem Water Content and BiomassForest Using Polarimetric SAR 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 SAR Interferometry,” IEEE Trans. Geosci. Remote Sensing, vol.38, no. 2, pp. 671-684, March 2000.[45] Krogager, E., Aspects of Polarimetric Radar Imaging, Ph.D. thesis, Technical University ofDenmark (TUD), Electromagnetics Institute, Lyngby, DK, March 1993.[46] Zhou, Z.-S., and M. Sato, “Development and performance evaluation of a ground-based SARsystem,” 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 Applications,” 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 Applications, Artech HouseNorwood USA, 1994.[50] Lüneburg, E., “Principles of Radar Polarimetry,” IEICE Trans. on the Electronic Theory, vol.E78-C, no. 10, pp. 1339-1345, 1995.[51] Lüneburg, E., “Radar polarimetry: A Revision of 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 for Geo-science Applications, ArtechHouse, Inc., Norwood, MA, 1990.[53] van Zyl, J. J. On the Importance of Polarization in Radar Scattering Problems, Ph.D. thesis,California Institute of Technology, Pasadena, CA, December 1985.[54] Rudolf, H., D. Tarchi, and A. J. Sieber, “Combination of Linear and Circular SAR for 3-DFeatures,” Proc of IGARSS 1997, Singapore 3-8 August 1997, pp. 1551-1553.[55] Kulowski, A., “Algorithmic representation of the ray tracing technique,” Applied Acoustics, vol.18, no. 6, pp.449-469, 1985.[56] Kulowski, A., “Error investigation for the ray tracing technique,” Applied Acoustics, vol. 15, no.4. pp. 263-274, 1982.[57] Mott, H., Antennas for 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 of Dihedral Corners,” IEEE Trans. Antenna Propagat., vol. 25,pp. 406-409, May 1977.[66] Griesser, T., and C. A. Balanis, “Backscatter Analysis of Dihedral Corner Reflectors UsingPhysical Optics and the Physical Theory of 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 for Lossy Dihedral CornerReflectors,” Proc. IEEE., vol. 77, no. 5, May 1989, pp. 806-814.[69] Anderson, W. C., “Consequences of Nonorthogonality on the Scattering Propertires of 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 for 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 Correctionfor 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 for 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 of Polarimetric Radar Systemswith 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 SAR Calibration Using Target Reciprocity,” J.Electromagnetic Waves and Applications., vol. 5, no. 7, pp. 735-751, 1991.[76] Chen, T.-J., T.-H. Chu and F.-C. Chen, “A New Calibration Algorithm of Wide-BandPolarimatric Measurement System,” IEEE Trans. Antenna Propagat., vol. 39, pp. 1188-1192,August 1991.[77] Freeman, A., “SAR 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 of Stokes andScattering Matrix Format Polarimetric SAR 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 for Polarimetric SARData,” 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 for SyntheticAperture Radar Systems,” IEEE Trans. Geosci. Remote Sensing, vol. 26, no. 6, pp.753-763,November, 1988.[82] Ulander, L. M. H., “Accuracy of Using Point Targets for SAR 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. Sato, “Development of a Ground-Based PolarimetricBroadband SAR System for Non-Invasive Ground-Truth Validation in Vegetation Monitoring,”IEEE Transactions on Geosci. Remote Sensing, submitted.[85] Imhoff, M. L., S. Carson, and P. Johnson, “A Low Frequency Radar Sensor for 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] Imhoff, M. L., “A Theoretical Analysis of the Affect Forest Structure on Synthetic ApertureRadar Backscatter and the Remote Sensing of 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 Transform and its Applications, Sec. Rev. Ed., McGraw Hill,New York, 1986 (Third Revised Edition: 1999).[90] Margrave, G. F., “Theory of Nonstationary Linear Filtering in the Fourier Domain withApplication 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 of Exploration Geophysicists, 1987, pp.403-409.[94] Stolt, R. H., “Migration by Fourier Transform,” Geophysics, vol. 43, no. 1, pp.23-48, Feb. 1978.[95] Chun, J. H. and C.A. Jacewitz, “Foundamentals of Frequency domain Migration,” Geophysics,vol. 46, no. 5, pp.717-733, May, 1981.155

Bibliography[96] Fortuny, J and A. J. Sieber, “Fast Algorithm for 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 of 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 of Synthetic Aperture Radar(SAR) Algorithms in the Time-Space Domain,” IEEE Trans. Geosci. Remote Sensing, vol. 36,no. 6, pp.1995-1999, November, 1998.[99] Cafforio, C., C. Prati, F. Rocca, “SAR 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 Algorithmfor Ground-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 for 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 for Reconstructionfrom Projections”, IEEE Trans. Antenna Propagat., vol. 26, no. 2, pp. 274-279, 1978.[104] Zhou, Z.-S., and M. Sato, “Ground-Based Polarimetric SAR Systems for 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 for 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 of Optimization Processing for 2x2Sinclair, 2x2 Graves, 3x3 Covariance, and 4x4 Mueller (Symmetric) Matrices in CoherentRadar Polarimetry and its Application 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 Transformations,” Chapter 12 in TheTransforms and Applications 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 of Kennaugh’s target characteristicpolarization state theory using the polarization transformation ratio formalism for 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 of RadarPolarimetry (Invited Review): The Target Characteristic Polarization State theory of 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 ofRandom Radar Targets,” pp.27.1 - 27.12, in Proc. NATO-AGARD-EPP Symposium onTarget and Clutter Scattering and Their Effects on Military Radar Performance, Ottawa,Canada, May 6 - 10, 1991.[114] Yang, J., Y. Yamaguchi, H. Yamada, M. Sengoku, S.-M. Lin, “Stable Decomposition of MuellerMatrix,” IEICE Trans. Comm., vol. E81-B(6), pp. 1261-1268, June 1998[115] Cloude, S. R., E. Pottier, “A Review of 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 of 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 of the Analysis of Multi-frequency Polarimetric SAR Data,”Proceedings of 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 of 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 oftarget 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 of the Sphere, Di-plane and Helix Decomposition,”Proc. of 3rd International Workshop on Radar Polarimetry, IRESTE, University of Nantes,France, April 1995, pp. 106-114.[121] Krogager, E, and W.-M. Boerner, “On the Importance of Utilizing Polarimetric Information 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 for Land Applications ofPolarimetric SAR,” 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 of Terrain Surface Azimuthal/ Range Slopes using Polarimetric Decomposition of POLSAR Data,” Proc. IGARSS 1999.Hamburg, 28 June - 2 July 1999.[125] Treuhaft, R. N., and S. R. Cloude, “The Structure of 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, AIRSAR-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 of Birth: September 22, 1968Nationality: ChinaMarital Status: MarriedSex: MalePlace of Birth: Suizhou, Hubei, ChinaHealth: ExcellentHobby: Bridge, Football, Tour etc.Education:2000.04-2003.09 Graduate School of Engineering, Tohoku University, Doctor Course,Doctoral Dissertation: Application of a Ground-based Polarimetric SAR System forEnvironmental Study.1994.09-1997.06 Graduate School of Computer Science, Central China Normal University(CCNU), Master Course, Master Dissertation: Study of HFC Computer Networkand Its Digital Modulation Technology.1986.09-1990.06 Physics Department of Central China Normal University, Bachelor ofScience with a Thesis: Explosion Mechanism of Supernova SN1987A and ItsNeutrino Procedure.1983.09-1986.07 No. 2 Senior High School of Huangshi, Hubei Province.1980.09-1983.07 No. 1 Junior Middle School of Daye Iron Mine of Wuhan Steel Corp.1978.09-1980.07 No. 1 Primary School of Daye Iron Mine of Wuhan Steel Corp.1975.09-1978.07 Hongfeng Primary School of Shuanghe, Suizhou, Hubei Province.Experience:1999.12-2000.03 Center for Northeast Asian Studies, Tohoku University, Japan, Researcher1998.05-1999.08 Department of Information Management, CCNU, China, Lecturer1997.10-1998.04 School of Information and Communications, University of North London,UK, Visiting Lecturer1990.07-1997.09 Department of Information Management, CCNU, China, AssistantResearch Interests:Radar polarimetry, Radar remote sensing, Ground-based SAR data processing, Digital signalprocessing, Analog communication and digital communication159

BiographyPublications:Zheng-Shu Zhou, Tadashi Hamasaki and Motoyuki Sato, “Polarimetric Applications of ABroadband Ground-based SAR System to Tree Monitoring,” IEICE Tech. Rep., vol. 103,no.265, pp. 1-6, August 2003.Tadashi Hamasaki, Zheng-Shu Zhou and Motoyuki Sato, “Development of a Ground-basedInterferometric SAR System,” IEICE Conference on AP, Sendai, Aug. 25, 2003 (inJapanese).Zheng-Shu Zhou, Wolfgang-Martin Boerner and Motoyuki Sato, “Development of aGround-Based Polarimetric Broadband SAR System for Non-Invasive Ground-TruthValidation in Vegetation Monitoring,” IEEE Transactions on Geoscience and RemoteSensing, submitted.Zheng-Shu Zhou and Motoyuki Sato, “Ground-Based Polarimetric SAR Systems forEnvironment Study”, IEEE AP-S 2003 Digest, vol. 1, pp.202-205, Columbus, June 2003.Zheng-Shu Zhou. Tadashi Hamasaki and Motoyuki Sato, “Development of BroadbandGround-based Polarimetric SAR Systems for Environment Study,” IEICE GeneralConference 2003, [CD-ROM], SB-1-10, Sendai, March 2003.Zheng-Shu Zhou, Kenji Takasawa and Motoyuki Sato, “Interferometric PolarimetricSynthetic Aperture Radar System,” in Proc. SPIE, vol. 4548, pp. 18-23, Oct. 2001.Zheng-Shu Zhou and Motoyuki Sato, “Development and Performance Evaluation of aGround-based SAR System,” IEICE Tech. Rep. vol. 102, no. 84, pp. 31-36, 2002.Zheng-Shu Zhou and Motoyuki Sato, “Shallow Subsurface GPR Survey in Sendai-Jo Castle,”Proc. of 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 Information Highway and Its Main Technologies,” in Reformand Explore, China Archives Press, August 1997, pp. 189-192 (in Chinese).Study Report:Zheng-Shu Zhou, “Distributed Database System -- a Core Module for Master Level Courseon Information Management”, submitted to British Council, UK, May 1998.Debao Xiao, and Zheng-Shu Zhou, “Analysis, Design and Application of MAN Based onHFC,” submitted to Science and Technology Committee of Wuhan, China, December 1996(in Chinese).160