LGP Discrimination and Residual Risk Analysis: Camp Sibert and Camp San Luis Obispo
Abstract
This report describes a 2-year unexploded ordnance (UXO) classification demonstrating the application of the Linear Genetic Programming (LGP) Discrimination Process(TM) to the problem of UXO discrimination and residual risk analysis. In support of project objectives, we analyzed multisensor electromagnetic and magnetic data acquired at two live sites. The objective of this project was to discriminate a variety of potentially hazardous munitions from items that may be safely left in the ground. At former Camp San Louis Obispo (SLO) the targets of interest (TOI) included 60 mm mortars, 81 mm mortars, 2.36-inch rockets, and 4.2- inch mortars. At former Camp Sibert, the lone TOI was a 4.2-inch mortar. The LGP Discrimination Process(TM) begins with the digital geophysical mapping (DGM) from a site suspected of containing UXO. It then (1) extracts attributes from the DGM near potential targets that may be UXO, (2) uses LGP and the attributes to rank the potential targets in their order of likelihood of being UXO, and (3) applies statistical residual risk analysis to determine which of the ranked targets may be safely left in the ground as Not-UXO. The attributes extracted for each target are analyzed by information-theoretic and statistical methods to reduce the attribute set to a handful of highly predictive attributes. Then, LGP is used to rank the "blind" targets as either UXO or Not-UXO using a small "training" set of targets for which ground truth was provided. Finally, statistical residual risk analysis is applied to the rankings and to the training ground truth to determine the stop-digging cutoff.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 2010
- Accession Number
- ADA539958
Entities
Organizations
- Environmental Security Technology Certification Program