Sonar target recognition using radial basis function networks

Abstract

The authors consider the problem of active sonar target classification based on the targets' material composition using a radial basis function (RBF) network. Sonar target responses were measured under controlled laboratory conditions in a laboratory tank. Spherical targets of different material composition were used. An important task in the design of RBF networks is the appropriate choice of the RBF centers. They propose a Karhunen-Loeve (KL) expansion based approach for centre selection. Results of the classification performance of the RBF network trained using the KL expansion based training procedure are provided.<>