Use of principal component images for classification of the EM response of unexploded ordnance

Abstract

The location and identification of unexploded ordnance (UXO) is a major challenge for environmental rehabilitation of former military firing ranges and bombing target areas. EM methods are in widespread use for the location of metal objects, however the presence of large quantities of scrap metal from successful detonation of munitions makes discrimination between munitions and scrap and munitions of various sizes a necessity in order for efficient location, digging and removal of UXOs to proceed. Several recent papers show that detailed three-component EM measurements followed by inversion to dipole moments of an EM target is effective in characterising a target, however such techniques require precise data, usually from stationary data acquisition. EM data acquired from a moving ground platform for UXO detection is typically high in motion-induced noise which limits the usefulness of decay-curve analysis in target characterization. We use a data set from the Australian Air Force Newholme UXO Test Range, Armidale, NSW, and show that a) images of time-window data are unsuccessful in discriminating between different types of munitions, b) adaptive decay index methods are ineffective due to high noise levels on the observed decay curves, and c) principal-component transforms of the data are successful in differentiating between different types of munitions. The method provides a tool for initial classification and prioritization of anomalies from a surveyed area, thus facilitating preparation of an efficient program of follow-up surveys and site clean-up.