Minehunting with multi-layer perceptrons

Abstract

The authors describe the use of multilayer perceptrons to solve the problem of distinguishing mine-like objects from clutter. Three increasingly sophisticated and effective approaches were applied against difficult side scan sonar imagery containing a highly cluttered and variable environment. Performances of the three approaches are compared using receiver operating curves (ROCs). Comparisons show that one can achieve a detection rate of 0.97 for a 0.01 false alarm rate. A subset of the networks have been demonstrated on special purpose hardware to run in real time.<>