Wednesday (Group 2) - SERDP-ESTCP - Strategic Environmental ...

Munitions Management (MM)
Poster Number 41 – Wednesday
Ground Based Detection and Discrimination — EMI & Magnetometers / Modeling & Processing
STATISTICAL SIGNAL PROCESSING FOR UXO DISCRIMINATION FOR
NEXT-GENERATION SENSOR DATA
DR. LESLIE M. COLLINS
Duke University
ECE Department
Room 130 Hudson Hall, Box 90291
Durham, NC 27708-0291
(919) 660-5260
lcollins@ee.duke.edu
CO-PERFORMERS: Dr. Stacy Tantum, Dr. Chandra Throckmorton, Dr. Jeremiah Remus, and
Dr. Lawrence Carin (Duke University); Dr. David Wright (USGS);
Dr. Erika Gasperikova (LBL)
U
ntil recently, detection algorithms could not reliably distinguish between buried UXO and
clutter, leading to many false alarms. Over the last several years modern geophysical
techniques have been developed that merge more sophisticated sensors, underlying physical
models, and statistical signal processing algorithms. These new approaches have dramatically
reduced false alarm rates, although for the most part they have been applied to data collected at
sites with relatively benign topology. To address these problems, SERDP and ESTCP have been
supporting efforts to develop a new generation of UXO sensors that will produce data streams of
multi-axis vector or gradiometric measurements. The focus of the research that we will present
here is on development of new physics-based signal processing approaches applicable to the
problem in which vector data is available from such sensors.
Specifically, we will present modeling and processing results obtained using state of the art
multi-axis sensors developed by LBL and USGS. First, we demonstrate that utilization of the
phenomenological models developed during this program for data inversion results in improved
discrimination performance over inversion strategies that use simplified models. We also
consider the impact of relaxing the assumption of a symmetric object in the inversion process,
and demonstrate improved classification results. We carefully consider options for the inversion
process, and demonstrate that careful data selection can impact performance quite significantly.
In addition, we also report on new classifier work. Results are presented for test stand data from
the ALLTEM system and Camp Sibert data for the BUD system. Both calibration and blind
results are presented for the BUD system.
This research was supported by the Strategic Environmental Research and Development
Program (SERDP) under Project MM-1442.
G-9