Classification of vessel targets using wavelet statistical features

In this paper, a new algorithm of wavelet band-based feature extraction scheme is developed for usage in classifying underwater targets from the acoustic non-stationary signals. Based on the advantage of the wavelet transform (WT) in non-stationary signal processing, the algorithm extracts statistical features of the sequential data in each discrete wavelet frequency modulation band of the vessel radiated signals. Using a person-by-person optimization (PBPO) approach to select the target separability features from the line combination of the band-sequence zero cross density (BZD) features, the band-sequence variation degree (BVD) features and the band-sequence maxima density (BMD) features to create the final classification feature vectors. Theory analysis using distance function of the final feature vectors show that the optimal selected feature vector is effective. Experiment using two different type targets sea trial data show that this feature extraction scheme in underwater target classification is feasible and the recognition rate reaches 93.3%.