Multi-aspect acoustic classification of buried objects

In 2001, broadband echoes from objects buried 1-3 feet in a sand bottom were collected with the prototype of a Buried Object Scanning Sonar (BOSS), currently being developed by Florida Atlantic University and EdgeTech Inc. The object field consisted of one mine-like object as well as numerous non-targets, including scuba tanks, a concrete block, an aluminum sphere and an ordnance shell. The BOSS system was suspended in the water from a cable and translated across the object field. The task of detecting and classifying target objects is very challenging using only single-aspect acoustic returns. Features were extracted from these acoustic returns and then applied to various back-propagation neural network (BPNN) classifiers. A non-linear decision-level fusion was then applied to perform multi-aspect classification. The results are presented in terms of the receiver operating characteristics (ROC) curves and confusion matrices for the data collected by the BOSS system.