Methods Of Feature Extraction From Envelop Of Ship Radiated Noise

Automatic passive underwater target classification becomes more and more important. The key step of classification is the extraction of target features. The research work in this thesis focus on the applications of wavelet transform in underwater targets, which include denoising by wavelet transform, feature extraction , model recognition and neural network. Through the computer simulation, the experimental results are obtained.The main contents of this thesis are as follows:(l)The statistic properties of one dimensional signal and noise based on wavelet transform are studied. The application that wavelet transform is used for noise reduction is based on this work.(2)The denoising principle and method based on wavelet transform are applied in underwater target signal. It has been proved through computer simulation that the effect of noise reduction is much better by using multiwavelets than single wavelet. (3)Wavelet theory is applied to envelope analysis and wavelet envelope method is discussed in detail. The Gaussian combination wavelet filters fitting for underwater targets are designed.(4)An efficient feature extraction method based on fast wavelet transform is presented, which divides the matrix of computed wavelet coefficients into clusters equal to rowvectors. The features are eventually calculated by the Euclidean norms of the clusters. The combination of features extracted from wave structures, wavelet coefficients and frequency spectrum get better classification result than one kind of feature.(5)The underwater target classifier is designed by using artificial neural network, which is tested with practical noise samples. The results show that the presented features and corresponding extraction methods are satisfactory.