Acoustic transient analysis using wavelet decomposition

Abstract

The authors demonstrate the use of wavelet decomposition in extracting relevant information from passive acoustic signals. These decompositions were used in generating features for classifiers which were applied against the standard data set of transients obtained from NUSC. Complete separation of four classes, i.e., three transients and a quiet ocean background, was obtained using two classification approaches: one based on a quadratic Bayesian classifier and the other based on a multilayer perceptron. The authors describe the wavelet-based features and the classifier design and provide class scatter diagrams.<>