Study on the feature extraction and classification of underwater target radiated noise based on bispectrum

Abstract

The radiated noise of underwater targets apparently consists of non-Gaussian ingredients. In the paper, based on plenty of radiated noise data, non-Gaussian feature of target signals is studied via high-order cumulates. Then, from Bispectrum estimation and WALSH dimensionality reduction, 65 dimensional Bispectrum feature from different kind of targets is extracted. The results show that the approach can efficiently classify underwater targets, and the colored Gaussian noise is restrained. The ration of recognition can arrive 92% toward three different kinds of underwater targets.