Research On Feature Extraction Of Noises Radiated From Underwater Targets

Recognition technique on underwater targets is an important research in the underwater acoustic signal processing field, but also one of the difficulties in the field. Feature extraction is the key link of the targets recognition, on account of the sea environment complexity and the underwater acoustic channel particularity, it's a difficult problem to extract a kind of target essence feature and distant detection effective feature. This thesis studies four methods of the target feature information extraction, and gives the corresponding results of the simulation and ship radiated noises.Firstly, the thesis introduces the mechanism of the ship radiated noises, and carries on the simulation to its component. On the basis of the above study , the thesis works over the extraction of the characteristics by the cycle map line spectral, and extracts the continuous spectrum component and the linear spectrum component from the simulation targets and the objectives from SongHua Lake under the condition of the three different situations, and makes comparison with the similar targets under the different conditions.Secondly, this thesis studies the energy feature extraction algorithm based on the wavelet packet decomposition, and extracts the energy characteristics from the simulation targets and the data comes from the Lake , and in the most obvious characteristics of the first and second band, the results are analyzed and compared, the advantages and disadvantages of this method are pointed out.Thirdly, the box feature extraction algorithm is introduced based on the fractal theory, and the box feature is extracted .the stylebook from the lake is analyzed, and the differences between the similar target are compared.Finally, the feature extraction based on the high-order statistical analysis is analyzed aiming at the misses of the traditional radial integral bi-spectrum and the circumference integral Bi-spectrum and the shortcomings of the re-use of the Bi-spectrum. The improved method is presented . the bi-spectrum features from the lake data are extracted , the integral bi-spectrum features, gradient and peaks are improved, the character differences of the similar target under the different conditions are compared.