Research On Active Target Echo Classification Based On PNN And CNN

Underwater active target echo classification has always been one of the important research directions in the field of active sonar signal processing.Accurate and efficient underwater active target echo classification is of great significance in both military and civilian fields.The underwater active target echo contains much information reflecting the characteristics of the underwater target,which is the basis of the active target echo classification.However,due to the complexity and variability of the underwater environment,the active target echo is susceptible to various underwater noises and interferences,which leads to the poor classification performance of the traditional underwater active target echo classification method.In this dissertation,the characteristics of underwater active target echoes are fully analyzed.Based on the velocity and scale of the underwater target,different feature extraction methods are used to extract the characteristics of the echo signal in the time domain and the time-frequency domain,respectively.Combined with two different types of artificial neural networks,two classification methods of the underwater active target echo are proposed,and the main contributions are listed as follows:1.The common active sonar transmitting signals are introduced and their properties are compared.The classification process of active sonar target echo signals is introduced.Moreover,the characteristics of the geometric acoustic scattering of the underwater active target are analyzed and the highlight model theory of the underwater target echo signal is introduced.The principle of the matched filter is introduced and the performance of the traditional underwater active target echo classification method based on the matched filter is evaluated.The classification results of the numerical simulation data and the measured data show that this method is susceptible to noise,and the classification accuracy is greatly reduced when the SNR is low.Moreover,the robustness of this method is poor.2.Utilizing the characteristics in the time domain of underwater active target echoes,an active target echo classification method based on the Orthogonal Matching Pursuit(OMP)algorithm and the Probabilistic Neural Network(PNN)is proposed.Firstly,based on the highlight model theory of the underwater target echo signal,a fixed signal dictionary suitable for the sparse decomposition of underwater active target echoes is constructed in this classification method.Then,the OMP algorithm is used for the sparse decomposition of underwater active target echoes to obtain the corresponding sparse coefficients and the PNN is used as a classifier to classify underwater active target echoes.The performance of this classification method is verified by the numerical simulation data and the measured data,respectively.In the case of different SNR values and different classification criteria,the classification accuracy of this method is higher than that of the underwater active target echo classification method based on the matched filter.This classification method shows strong anti-noise performance and robustness.In addition,the classification results in the case of different classification criteria show that this method is more suitable for the classification according to the velocity of the underwater target corresponding to the target echo.3.Utilizing the characteristics in the time-frequency domain of underwater active target echoes,an active target echo classification method based on the Short Time Fourier Transform(STFT)and the Convolutional Neural Networks(CNN)is proposed.Utilizing the STFT,this classification method generates time-frequency images of underwater active target echoes in this classification method.The time-frequency line spectrum detection and extraction algorithm based on the birth and death process is used to extract time-frequency line spectrums of time-frequency images.Then,time-frequency characteristic images for the underwater active target echo classification are generated and the CNN is used as a classifier to classify underwater active target echoes.The numerical simulation data and the measured data verify the performance of the proposed classification method.Specifically,the classification accuracy,the anti-noise performance and the robustness of this classification method are superior to the traditional active target echo classification method based on the matched filter.In the case of classifying target echoes according to velocity features of targets corresponding to target echoes,the classification accuracy of this method is slightly lower than that of the active target echo classification method based on the OMP algorithm and the PNN.However,the classification accuracies are relatively higher in the case of classifying target echoes according to scale and overall features of targets corresponding to target echoes.Therefore,the proposed method is more suitable for the classification according to the scale of the underwater target corresponding to the target echo.