Target detection using a neural network based passive sonar system

A neural-network (NN)-based system for the passive detection of targetlike signals in underwater acoustic fields is being developed. The input to the NN is an intensity modulated signal, which is a measure of the power of the received signal plus noise at different frequencies as time varies. Thus, a two-dimensional array (image) is to be examined to reach a decision. It is assumed that the target emits a sinusoidal signal at a fixed frequency f/sub 0/. If the target moves with a constant speed with respect to the receiver, the received signal frequency will be (1+ delta ) f/sub 0/, where delta is the Doppler shift. The received two-dimensional image is first thresholded to obtain a binary (0 or 1) image. The first stage of the proposed system consists of an autoassociative memory (ASM) whose function is to eliminate the noise and reconstruct the received signal. The output of the ASM is input to the second stage of the system, which consists of a multilayer perceptron (MLP) classifier trained using the backpropagation algorithm. The MLP outputs a decision regarding the presence or absence of the targets. Results of an initial experimental study are reported. A promising classification accuracy of 97% for targets and 100% for no-targets has been obtained.<>