Research On The Bottom Target Feature Extraction And Recognition

Because of the low target echo signal reverberation ratio when detecting bottom smalltarget with active sonar, it is difficult to extract the target echo feature for identifying, thusthis paper proposed the method of extraction of feature of target echo based ontime-frequency filtering. The basic idea of this paper is: firstly, doing XWVD transform withtarget echo and transmit copy signal to obtain a two-dimensional time-frequency domainsignal, then on the time-frequency domain for ridgelet transform and threshold denoising onit. Secondly, put the signal which had threshold denoising to do ridgelet inverse transform toobtain the time-frequency domain signal, then do Hough transform on it, and in Hough planefilter noise through hard denoising. Finally, getting the feature vector of this paper for targetrecognition through projected onto one-dimensional.In the practical issues of the active target recognition problem, the choice of a suitableclassifier is essential. Support vector machine is a method of pattern recognition, which needa small sample of the data for learning and training, and it has good generalization ability,thus in this paper, choose support vector machine as a classifier for target recognition system,put extract the target feature into support vector machine for learning and identification.Finally, in order to improve the support vector machine algorithm efficiency, the paperresearch the application of least squares support vector machine (LS-SVM) in bottom targetof the target recognition, through simulation and practical experiments show that the trainingand testing time of the LS-SVM great less than standard SVM.This paper through simulation and actual experimental data processing, achieved quitesatisfactory results, and verified the validity of the time-frequency filtering method forextraction of the target features, so as to provide new ideas and technology base for theclassification of bottom target.