Research On Noise Robust Narrowband Target Classification Method

The micro-Doppler effect can reflect the motion characteristics and geometric information of the micro-moving parts of the target,which provides a new idea for narrow-band radar target classification.However,due to the weak echo of micro-moving parts,the use of micro-Doppler effect for target classification requires a high SNR.However,in practice,the target to be classified is usually far away from the radar,and its echo SNR is relatively low,which will cause the classification performance to be seriously degraded.In response to this problem,this paper carried out research on robust noise narrow-band target classification,the main research content is summarized as follows:1.First,an aircraft rotor echo signal model is established,and based on this model,the differences in the micro-Doppler modulation characteristics of helicopters,propeller aircraft,and jets are analyzed from multiple angles.Then the effects of blade angle,pitch angle,PRF,dwell time and SNR on the echo signal are analyzed.2.The noise robust classification method combining SNR matching and echo enhancement is studied.This method is based on the idea of matching with noise.In the training stage,by adding noise to the training samples with high SNR,a training sample set with different SNR levels is obtained.Based on this,the classification under different SNR is trained and stored Classifier template;in the testing stage,classify by estimating the SNR of the test sample and selecting the classifier that matches it.Simulation experiment results prove that this method can effectively improve the noise robustness of classification on the basis of echo enhancement.3.A robust noise classification method based on convolutional neural network is studied.This method uses the powerful nonlinear mapping ability of convolutional neural networks to extract robust classification features of noise from the echo amplitude information.The results of simulation experiments show that the extracted features are less affected by noise and have certain noise robustness.Considering that in addition to the echo amplitude information,the phase also contains classification information,so a two-branch network is designed to reasonably utilize the phase information.Simulation experiment results prove that the two-branch network can further improve the noise robustness of classification.4.In order to make full use of the SNR information of samples in the classification process to improve the performance of classification,a robust noise classification network assisted by the SNR information was designed.The network uses the SNR information of the sample as an additional input to participate in the learning process of the convolution kernel and the weight of the fully connected layer.The network can adaptively adjust network parameters according to the change of the SNR and learn the current SNR.The characteristics of better separability under conditions.The results of simulation experiments prove that the features of the SNR information assisted network further improve the classification performance based on the features of the convolutional neural network.