Study On Feature Extraction And Classification Based On Signal Of Dragging Fishery Detector

The study of this dissertation is a part of the international scientific and technological cooperation project" Fishery Detector of Dragging Style ". Feature extraction and classification of under-water target are the points of the study .The study belongs to the passive sonar target recognition,which distinguish the category of the under-water target by eradiate noise. According to the principle of mode recognition, the paper can be divided into three parts: feature extraction, classification and experimental study.The process of feature extraction is to transform the eradiate noise signal to feature space and extract the feature vectors that reflect the category of the input sample. The extracted features are the input modes to the classifier. Three methods of eradiate noise's feature extraction were selected. The methods include Lofar Spectrum, Wavelet Packet Decomposing,Higher-Order Statistics. In those methods, the Lofar Spectrum bases on time-frequency field; the Wavelet Packet Decomposing can extract the energy features in different frequency bands; the Higher-Order Statistics can extract the non-Gaussian features of underwater targets. The eradiate noise's different sides' features were extracted and strong elements were offered for the next process: classifying.The classifier can decide the category of current input sample according to its pattern. The classifier this dissertation introduced is the neural network classifier.It can improve the classifier's performance.In the paper, test data was analyzed and different targets' features were extracted using the three methods above .At last ,the extracted features are passed into the Probability Neural Network to recognize. The ratio of recognition was given and make a conclusion that the feature extraction arithmetic and classifier arithmetic are efficient.