Implementation of an Intelligent Target Classifier with Bicoherence Feature Set
This paper examines the feasibility of bispectral analysing of acoustic signals emanated from underwater targets, for the purpose of classification. Higher order analysis, especially bispectral analysis has been widely used to analyse signals when non-Gaussianity and non-linearity are involved. Bicoherence, which is a normalized form of bispectrum, has been used to extract source specific features, which is finally fed to a neural network classifier. Vector quantization has been used to reduce the dimensionality of the feature set, thereby reducing computational costs. Simulations were carried out with linear, tan and log-sigmoid transfer functions and also with different code book sizes. It is found that the bicoherence feature set can provide acceptable levels of classification accuracy with a properly trained neural network classifier.