Research On Active Sonar Target Classification And Recognition Based On Sparse Representation

Active sonar target classification and identification has the advantages of good exploration and strong decision-making efficiency,which is of great importance significance both in the military field of protecting national security and the ocean resources development for civilian.The active sonar target echo signal is the basis of active sonar target classification and recognition,which has a large amount of information reflecting the essential characteristics.But noises and reverberations are received by active sonar in the complex and changeable underwater environment.It makes the extraction of effective features from the target become very difficult.In order to address this problem,this thesis focuses on classification and recognition algorithm under noise and reverberation dominant environments.The main works are as follows:Firstly,combining the sparse coefficient with the Support Vector Machine,the active sonar target classification recognition method based on sparse coefficient feature is proposed.It's based on the good noise processing ability of sparse representation theory and sparse coefficient contains the essential information of the target.Then,applicated on the background of noise.It shows that good noise suppression ability of this method from the result of changed signal-to-noise ratio and test sample size.Comparing the results of spectrum feature and dual-spectrum feature under the same conditions,it proved this method has good classification and recognition performance.Secondly,combining the ideas of dictionary learning,sparse reconstruction and matching degree,the active sonar target classification recognition method with dictionary learning is proposed.The dictionary is obtained by the training of each target signal,after that,reconstructing test signal.Then,the classification strategy is made up of the matching degree of each reconstruction signal and the test signal.Last,proving the noise suppression ability from the result of changed signal-to-noise ratio and test sample size under the background of noise too,and verified that this method has an advantage in solving the problem under low SNR,and improves the performance of classification by compared the results of several algorithms,such as spectrum feature combine with support vector machine,K-near neighbor and sparse coefficient.Finally,combining the active sonar target classification recognition method with dictionary learning and Fractional Fourier Transform,the active sonar target classification recognition method based on reverb suppression is proposed.Among them,pre-processed the signal by the FRFT,which has good suppressed reverb signal ability.Then,applicating in the reverb background,compared the results of the active sonar target classification recognition method with dictionary learning under the same conditions.Proving that the method has a good ability to suppression reverb,and improved the classification recognition performance.