Battlefield Passive Acoustic Target Recognition Method

Target identification using moving target's acoustic wave is a basic task of the acoustic detection system. The technique of target identification belongs to the research field of pattern recognition, and the key techniques are feature extraction and the design of classifiers.This thesis firstly studies two feature extraction methods of acoustic signals with modern signal processing techniques. According to the spectrum characteristics and the energies in different scales after the wavelet decomposition respectively, the algorithm of feature extraction of acoustic signals based on PCA and wavelet decomposition is presented. Features of four types of acoustic signals of battlefield target are extracted by this two means and the feature vectors are obtained in a low-dimension. K-nearest neighbor classifier and BP neural network classifier are designed for the acoustic target classification. These two classifiers and real acoustic signal data are used to do the experiments of target classification. Result shows that all techniques and classifiers presented here can stand for the difference among different acoustic targets well, and get satisfactory rate of correct recognition.The other part of the thesis studies the application of evidence reasoning on multi-target recognition. The energies in different scales after the wavelet decomposition respectively are applied to extract feature, and absorptive method based on Dempster's rule is used in feature fusion of multi-source information. The results of the simulations between absorptive method and nearest neighbor method are compared, which show the effectiveness of evidence reasoning used in multi-target recognition.