20 - NEURAL NETWORKS FOR SONAR SIGNAL PROCESSING

Abstract

This chapter presents a macroscopic view of all the areas of sonar signal processing with neural networks by outlining a generic multi-stage sonar signal processing system and then discussing the neural networks that can be used at each stage of the system. The chapter focuses on those portions of sonar signal processing where neural networks have direct applicability. A generic sonar signal processing system is broken into four stages: (1) noise cancellation, (2) feature extraction, (3) detection/classification, and (4) post-processing/display. Neural networks can be applied to the first three stages and some earlier signal collection efforts. Neural networks have been applied to bearing estimation or source localization, determining the direction of the source that is radiating the sonar signal. Accurately estimating the bearing of a signal can improve the beam-forming effort and aide in tracking the source. Two neural network techniques—ADALINE Finite Impulse Response (FIR) filters and back-propagation FIR filters—have been proposed to perform noise cancellation.