The Department of Defense has tens of millions of acres of land and more than 1,000 sites at which buried unexploded ordnance (UXO) is a potential hazard to humans and to the environment. At these sites, the U.S. Army Corps of Engineers estimates that 70% of remediation costs are typically consumed by excavation of non-ordnance items (i.e., false alarms). Technical advances are needed to increase the efficiency of characterizing and remediating these lands. The objective of this project was to employ a new, broadband, fully polarimetric radar system to aid in discriminating subsurface ordnance from non-ordnance items.

The system demonstrated in this project employs a combination of traditional UXO detection methods and broadband radar technology. Once "hot spots" have been identified at a site by the lower frequency systems, more complete information is obtained at the higher radar frequency band being demonstrated. The radar system, developed by the Ohio State University Electroscience Laboratory, consists of an orthogonal set of hornfed dipole antennas, controlled by a network analyzer over the ultra-wideband 20 MHz to 420 MHz. The nature of the feed permits a reliable input without distortion by ground loading. Rotation of the opposing dipole configuration produces complete polarization information, with sensitivity to target orientation, while providing orientation-independent characterization as well. Recording of the cross-polarized reflections avoids signal saturation by reflections from the ground surface or from other interfaces that are not of interest. In addition to providing orientation data, the radar's ultra-wideband spectrum facilitates inference of target depth and length. New special-purpose algorithms also were tested to infer target aspect ratio, gross shape, and some morphological detail, such as the presence of fins.

The ground penetrating radar (GPR) system exhibited modest discrimination capability in the baseline performance analyses. Despite considerable variation of relevant environmental factors (e.g., soil), the ultimate receiver operating characteristic (ROC) curve performance pattern was similar for the different sites. Before various improvements were applied, an approximately 55% probability of detection (Pd) of UXO-like targets was obtained with only a 10% probability of false alarm (Pfa); however, progress along the curve to 90% Pd was achieved slowly, only after an 80% Pfa was reached. Various processing improvements were tested by which additional external or "prior" information was worked into the GPR classification system. In the cases considered, use of prior depth estimates decreased the Pfa by about 15% to 30% without significantly affecting the Pd, provided that the depth information was accurate.

This project evaluated the utility of radar for UXO discrimination under different site conditions. Despite potential for discriminating subsurface target characteristics, investigators determined that it is unlikely that GPR will gain a prominent position for the detection aspect of surveying, except in distinct circumstances, such as magnetic but relatively nonconductive soil/rock or small, shallow, widely dispersed metallic clutter items. (Project Completed - 2005)