The objective of this project was to demonstrate that an innovative automatic processing system, which implements matched filtering and target attribute screening, will improve buried unexploded ordnance (UXO) detection and false alarm rejection relative to procedures currently used in the field. The innovation lies in the implementation of optimal linear filtering in a threshold-based UXO detection processor and the application of Boolean logic to target attribute screening to rationalize the UXO/clutter discrimination process.
Current UXO detection practice involves experienced operators selecting anomalies from displays of survey data. There are always buried UXO targets which are too small or are buried too deeply to be detected with current systems. Studies of visual signal detection, notably in the medical imaging community, consistently show that human operators do not perform as well as the optimal linear processor. In the optimum linear detection processor, a threshold is applied to the output of a prewhitening matched filter which compensates for any background correlation structure and then correlates the data with the expected signal. It is optimal in that it maximizes the output signal-to-noise ratio. Optimal linear filtering has not been used for UXO detection.
In this project, a Matched Filter AutoProcessor (MFAP) was developed that implements an optimal linear filter in a threshold-based UXO detection processor. In order to discriminate between UXO and the false alarms due to metallic clutter, screening procedures were implemented based on target attributes (e.g., size, depth, orientation) that were determined from their magnetic signatures. The individual screening criteria compare the various target attribute values with threshold levels set by the user.
The MFAP was tested on magnetometer data collected during the 2000 UXO Detection/Discrimination Advanced Technology Demonstration at Jefferson Proving Ground (JPG) in Madison, Indiana. Results were compared with electromagnetic induction (EMI) systems and magnetometer data from the various JPG demonstrators as well as the ground truth of the JPG Technology Demonstration to indicate the relative performance of the MFAP. The MFAP did a good job of isolating potential targets in the magnetically noisy and active areas encountered at JPG. The filter output peaked over potential targets as expected, reducing uncertainty in their exact locations. Performance was better for targets larger than 20 mm and was affected by gaps in the data.
The detection and classification modules developed through this effort can be added to existing detection and analysis systems. The matched filter reduces uncertainty in the location of a target as well as improves detection of targets in magnetically noisy areas. As such, these modules will serve as a testbed for exploring site-specific optimization of threshold settings and decision rules for UXO detection and false alarm rejection. The target selection routine should provide more automatic and efficient target selection with lower expenses for larger surveys. Future versions of the MFAP can be modified to analyze electromagnetic data in addition to passive magnetic data. (Project Completed - 2003)