In 2003, the Defense Science Board observed: “The problem is that instruments that can detect the buried unexploded ordnance (UXO) also detect numerous scrap metal objects and other artifacts, which leads to an enormous amount of expensive digging. Typically, 100 holes may be dug before a real UXO is unearthed! The Task Force assessment was that much of this wasteful digging can be eliminated by the use of more advanced technology instruments that exploit modern digital processing and advanced multi-mode sensors to achieve an improved level of discrimination of scrap from UXO.” Since 2003, significant progress has been made in UXO classification technology. To date, testing of these approaches has been primarily limited to test sites with only limited application at live sites. Acceptance of classification technologies requires demonstration of system capabilities at real UXO sites under real world conditions. Any attempt to declare detected anomalies to be harmless and requiring no further investigation will require demonstration to regulators of not only individual technologies, but an entire decision making process.

At the Spencer Range Tennessee site covered by this report, the objective was to discriminate targets of interest (TOI) (including 37-millimeter [mm], 60-mm, 75-mm, 105-mm, 155-mm targets and small and medium industry standard objects [ISOs]) from non-hazardous shrapnel, range and cultural debris. Researchers describe the performance of classification techniques that utilized: (1) full coverage, dynamically acquired survey data collected with both the Time Domain Electromagnetic Towed Array Detection System (TEMTADS) 2x2 and the MetalMapper advanced electromagnetic induction (EMI) sensors; and (2) static, cued interrogation style data acquired with MetalMapper and TEMTADS 2x2.

Technology Description

The Advanced Geophysical Classification (AGC) techniques applied to the Spencer Range data use dipole model-based features extracted from multi-static, multi-channel EMI data. Single source and multi-source dipole inversions were used to estimate target location, orientation, and principal polarizabilities. From the extracted feature vectors, prioritized dig-lists were created for: (1) the MetalMapper deployed in a dynamic, full coverage mode; (2) the MetalMapper deployed in a static, cued mode (for 2 unique datasets over identical targets collected by Naeva Geophysics Inc. [NAEVA] and URS Corporation [URS]); (3) the TEMTADS 2x2 deployed in a dynamic, full coverage mode; and (4) TEMTADS 2x2 deployed in a static, cued mode,

Anomalies were prioritized based on a match of estimated principal polarizabilities to a polarizability library of known TOIs, polarizability magnitude, and the rate of decay of polarizabilities. For each dataset, a reference library of polarizabilities was constructed from test pit measurements, site specific training data, and polarizabilities derived from data acquired at previous Environmental Security Technology Certification Program (ESTCP) demonstration sites. Since dynamic data have higher noise levels than cued data, a more conservative classification algorithm - the Combined Classifier Ranking (CCR) algorithm - was used. The MetalMapper (NAEVA) cued data were processed automated classification that combines multiple ranking rules (e.g. size, decay, etc.) and library matching. Classification parameters were determined by DigZilla, with the only input by the analyst being the reference library of ordnance polarizabilities.

All model fits and classification analysis were performed using a classification software suite (UXOLab) that was jointly developed by the University of British Columbia – Geophysical Inversion Facility and Black Tusk Geophysics (BTG).

Demonstration Results

Performance metrics defined by the ESTCP program office were calculated for each data set.  Application of dipole based classification was successful for all data sets, with 100% of TOI being identified and marked for excavation in each case. The reduction of clutter digs while retaining all TOI was greater than 80% in all cases. Not surprisingly, the data sets processed with the more conservative CCR algorithm had a higher percentage of false alarms than the data sets processed with a more aggressive approach. The number of "can't analyze" anomalies met the success criteria for all data sets except the MetalMapper (URS) dynamic data acquired in the Dynamic area. For the dynamically acquired data, reliable target parameters were estimated for only 92.1% due to the lack of data coverage at the edges of the survey area.

For the cued data sets with inertial measurement unit (IMU) information and the dynamically acquired data, the target location estimate error had a standard deviation of less than 10 centimeters (cm). Data acquired in the Open area and Dynamic area had depth estimate errors with a standard deviation of less than 10 cm. The data acquired in the Treed area by the TEMTADS 2x2 had a depth estimate error standard deviation of 13 cm, which did not meet the success criteria of having a depth estimate error with a standard deviation of less than 10 cm.  The survey conditions in the Treed area may have resulted in variation of the ground clearance height of the instrument, whereas researchers assumed a fixed ground clearance height for all anomalies.

Implementation Issues

A major implementation issue with AGC technology is providing easy to use software tools so that non-expert staff can obtain good classification performance. This project included a technology transfer and training component. A member of Shaw Environmental’ s production team attended a one week training session in Vancouver, British Columbia, Canada with BTG algorithm and software developers. The training session included an overview of UXO inversion and classification theory and software routines. The Shaw geophysicist was responsible for executing all parts of the classification workflow: from data and inversion quality control (QC), training data selection, to dig list creation and submittal. The dig list submitted by the Shaw geophysicist successfully identified all TOI and greatly reduced the number of non-TOI digs.