Research Of Highlight Feature Extraction And Recognition Based On Time-Frequency Domain Filtering

In the reverberation background,when the active sonar detects the underwater target,the reverberation interference increases the difficulty of extracting the target highlight feature from the target echo signal.When the signal-to-reverberation ratio is low and the transmitted signal is a chirp signal,the characteristics of the target highlights echo for such non-stationary signals cannot be fully expressed from the time domain or the frequency domain analysis.In order to solve the problem of extracting the target highlights in the reverberation background,this paper proposes to suppress the reverberation interference by time-frequency filtering in the time-frequency domain.According to the correlation between the target echo and the reverberation in the time-frequency domain,the adaptive kernel time-frequency analysis method is used to transform the target echo signal into the time-frequency domain for analysis.The target echo and reverberation are divided into the sparse matrix and the low rank matrix by the low rank matrix recovery method,thereby separating the target echo and the reverberation,and reducing the interference of the reverberation on the echo signal.The Hough transform is used to extract the bright peaks in the echoes for the sparse matrix,and the highlight features of the target are obtained.In addition,this paper also studies a linear line of energy concentration in the time-frequency domain according to the transmitted chirp signal,and the energy distribution of reverberation and noise in the time-frequency domain is disorganized.The frequency modulation slope is known by the transmitted signal,and the time-frequency domain signal is projected to the frequency axis according to this slope,and a threshold is set on the projection domain for filtering.According to the obtained time-frequency domain signal,a Hough transform is used to extract a highlight peak in the echo.Simulation and experiments show that the low-rank matrix recovery method and the projection filtering method can further separate the signal and reverberation in the time-frequency domain,and obtain the target highlight features that are more easily identified.Due to the non-cooperativeness of the target,the target feature obtained by the active sonar is unlikely to have a large sample size.The support vector machine is suitable for the learning training of such small sample data,so the support vector machine is used as the classifier.The parameters of the support vector machine have an important influence on the recognition effect.However,the setting of its parameters is often empirical.Therefore,the particle swarm optimization algorithm is used to optimize the penalty factors C and parameters ? in the support vector machine.The simulation and experimental data show that the particle swarm optimization The support vector machine can efficiently find the optimal parameters to achieve the optimal recognition rate.The experimental data is the echo data of the Songhua River under the ice.Based on the recognition problem of active sonar target echo signal feature extraction,the support vector machine based on particle swarm optimization is studied.The target echo intensity feature obtained by active sonar is unlikely to be large,and the support vector machine is particularly suitable for the learning training of such small sample data.However,because the parameters of the support vector machine are fixed,the parameters of the optimal solution cannot be changed automatically for different problems.Therefore,the penalty factors and parameters of the particle swarm optimization support vector machine are adopted.The simulation and experiment show that the particle swarm optimization support vector machine can Find the optimal value.