Target detection using a neural network based passive sonar system

摘要

A neural-network (NN)-based system for the passive detection ofntargetlike signals in underwater acoustic fields is being developed. Theninput to the NN is an intensity modulated signal, which is a measure ofnthe power of the received signal plus noise at different frequencies asntime varies. Thus, a two-dimensional array (image) is to be examined tonreach a decision. It is assumed that the target emits a sinusoidalnsignal at a fixed frequency f₀. If the target movesnwith a constant speed with respect to the receiver, the received signalnfrequency will be (1+δ) f₀, where δ isnthe Doppler shift. The received two-dimensional image is firstnthresholded to obtain a binary (0 or 1) image. The first stage of thenproposed system consists of an autoassociative memory (ASM) whosenfunction is to eliminate the noise and reconstruct the received signal.nThe output of the ASM is input to the second stage of the system, whichnconsists of a multilayer perceptron (MLP) classifier trained using thenbackpropagation algorithm. The MLP outputs a decision regarding thenpresence or absence of the targets. Results of an initial experimentalnstudy are reported. A promising classification accuracy of 97% forntargets and 100% for no-targets has been obtained