Target detection using a three-layered neural network trained by supervised back-propagation

Abstract

In any sonar system a detection process has to be performed at

the processor output to decide whether or not a particular signal is present in the water. In the particular case of an active sonar employing coherent processing the requirement is to examine the output of the matched filter and decide whether an output signifying the presence of a target echo is present or not. In the present study a neural network has been trained and then applied to this problem. Its performance has been evaluated by examining the statistics of the probability of detection and probability of false alarm using unfamiliar but synthesized data. A preliminary investigation of the effect of varying some of the network parameters has been performed.