The objective of this project was to adapt and extend existing strategies for discrimination of unexploded ordnance (UXO) to emerging next generation electromagnetic induction (EMI) sensors and to test discrimination performance at increasingly complex and cluttered sites. Specific technical objectives were to (1) develop robust inversion strategies to extract polarization tensor parameters from both production-type and next generation EMI sensor data that are applicable when the spatial signatures of adjacent anomalies overlap; (2) further develop and test a suite of discrimination strategies (e.g., statistical classification of polarization tensor parameters, library-based matching, statistical classification of data-based features) at increasingly complex and cluttered sites for a wide range of EM sensors (production and next generation); and (3) develop tools and expertise to decide on the optimum discrimination strategy to apply at a newly encountered site.

Through SERDP and ESTCP sponsorship, Sky Research, Inc. and the University of British Columbia-Geophysical Inversion Facility have developed and tested numerous discrimination methodologies using both production-type data (Geonics EM-61 and EM-63, total-field magnetic) and next generation sensor data (e.g., BUD, TEMTADS, MPV, MetalMapper). These signal processing algorithms and strategies were adapted and extended to support the next series of ESTCP's Classification Pilot Program. The project team incorporated algorithmic and methodological advances made in SERDP projects MR-1573 (magnetic soils), MR-1629 (robust statistics), and MR-1637 (model selection).

The project team tested several UXO classification algorithms and data processing concepts that were developed in the SERDP program. Many of the processing techniques they use for production remediation work today were tested and validated in this project using ESTCP Live-Site demonstration data. Various classification approaches were tested, including library based, rule based and machine learning classification. Multi-stage classification strategies that take into account the robustness with which target parameters could be estimated were used. Fully automated and analyst guided techniques were applied. Data simulations and synthetic seeding were used to establish thresholds for various decision metrics and cluster analysis methods for identifying training data were developed. Data from the Former Camp Butner and Camp Beale were processed. At Camp Butner, the Geonics EM61, Geometrics MetalMapper Classic and Time-domain Electromagnetic Multi-sensor Towed Array Detection System (TEMTADS) 5x5 data were processed. The limitations of the older Geonics EM61 technology were confirmed and the MetalMapper and TEMTADS advanced classification sensors performed well. At Camp Beale, the MetalMapper Classic was again deployed, along with newly developed man portable advanced geophysical classification sensors. The MetalMapper Classic was again able to find all UXO efficiently. All the man portable systems - i.e. the Man Portable Vector (MPV) sensor, the Berkeley Unexploded Ordnance Discriminator (BUD), and TEMTADS 2x2 - showed similar performance, with 100% of UXO found with few non-UXO digs required. For both the Camp Butner and Camp Beale demonstrations, it was important to understand how data quality impacts the accuracy with which dipole parameters can be extracted for a target, and the importance in selecting a classification strategy that is consistent with the polarizability accuracy and separation in model space between targets of interest and scrap. In addition to testing processing methods, they began the development of an API that ports dipole solvers to C++. They demonstrated the efficiency and accuracy of the API solvers within Oasis Montaj using MetalMapper cued interrogation data from Camp Butner.

No single discrimination strategy or sensor technology will be universally applicable to the wide variety of UXO-contaminated sites within the United States. Each site typically comprises a different mix of ordnance and clutter and has additional site-specific attributes that influence technology deployment, including cultural infrastructure, vegetation, topography, and geological background. This project provides the Department of Defense (DoD) with flexible, reliable, robust, efficient, and effective signal processing algorithms and strategies that can be adapted to (virtually) any site conditions or EMI sensor technology. The final outcomes are techniques and algorithms for reliable discrimination of UXO applicable to a wide variety of sites and validated discrimination strategies together with all the resources (software, trained staff, etc.) to immediately apply the technology to other DoD sites.