Research On Denoising Method Of Underwate Rtarget’s Acoustic Signal Based On Compressed Sensing

Recent years,with the continuous development of the marine industry,the exploitation and utilization of underwater resources are also increasing.Underwater target recognition and classification is one of the most important research fields.The underwater signal recognition and classification technology not only plays an important role in the cause of national defense,but also has commercial value in civilian area,therefore underwater target recognition is widely concerned.In this paper,the ship’s radiation signal extracted by passive sonar is the research object,and the denoising and feature extraction technology of the signal are studied.The main contributions are as follows:Firstly,this paper analysis the basic principle of compressed sensing,meanwhile,compressed sensing is introduced to ship signal denoising.The clean acoustic signal is approximate sparse in discrete cosine transform domain.By using this point,signal and noise are separated to achieve the purpose of denoising.In this paper,the threshold is introduced into the sparse transform of compressed sensing algorithm,meanwhile,different threshold functions are constructed based on different measurement matrix,the coefficient of large absolute value will be retained and the absolute value of small is infinitely close to zero.In this way to enhance signal sparsity,guarantee signal reconstruction performance and improve denoising efficiency.Secondly,due to the non stability,nonlinear and non Gauss features of the ship’s radiation signal,the single feature analysis cannot be efficiently identify targets,resulting in low classification accuracy.In the paper,the time domain,frequency domain and time-frequency domain feature extraction method is proposed to extract comprehensive features,which can obtain complete feature matrix,and then,aiming at the problems of the poor interpretability and low accuracy caused by high dimension,PCA is applied to the matrix to reduce the dimension of the feature,preserving most of the information of the feature and reduce the dimension,so as to avoid overfitting in the classifier.Finally,using the least squares support vector machine as the classifier to classify the signals after the feature extraction,so as to realize the effective recognition of the ship’s radiant signal.The data measured by passive sonar is used to denoising,feature extraction and recognition in the experiment simulation.The results show that the effect of denoising with improved compress sensing is obvious,furthermore,the comprehensive feature extraction based on PCA and LSSVM has better recognition results before reducing the dimension.