Buried Object Classification using a Sediment Volume Imaging SAS and Electromagnetic Gradiometer

Abstract

To advance naval capabilities in identifying buried mines and unexploded ordnance, hybrid systems that fuse data from disparate sensors are being developed. This paper describes preliminary results of a classification engine that combines target features and classification parameters from a synthetic aperture Buried Object Scanning Sonar (BOSS) and an electromagnetic Real-time Tracking Gradiometer (RTG). The target characteristics that generate signals of interest for these sensors (acoustic backscatter and induced changes in local magnetic field) are sufficiently diverse that optimal combination should effectively increase the probability of correct target classification and reduce false alarm rates. Geometric and backscatter intensity features automatically extracted from three-dimensional acoustic imagery are combined with magnetic moment and associated parameters in a joint-Gaussian Bayesian classifier (JBC), which makes mine-like/non-mine-like decisions for each contact. The fused acoustic-magnetic classifier was evaluated using a combination of sea-trial and synthetic data sets. Nine data runs were processed to yield acoustic and magnetic features, supplemented by the synthetic data. An initially large variety of feature types were down-selected by a training process to a critical subset. With this limited dataset, initial results show probabilities of false classification (Pfc) from 1.6% to 6.3% when at high probability of correct classification (Pcc).