Radar Target Recognition Based On Ultra-Wideband Signal Features And Deep Learning

Ultra-wideband(UWB)radar has a wide range of applications,such as military warhead recognition,aircraft identification,civil vehicle identification,pedestrian identification,home alarm,disaster relief detection and so on.Because ultra-wideband radar belongs to the radar of the new system,the target echo carries more complex electromagnetic spectrum information,so it can be extracted by a variety of signal analysis means more effective target identification features than the traditional system,so as to make use of the advantages of the ultra-wideband system as much as possible.This paper mainly studies the time domain high resolution range profile(HRRP)and frequency domain electromagnetic scattering characteristics of ultra-wideband radar echo generation joint identification method and unknown category target identification method in general scene,the specific work is as follows:1.Firstly,in order to expand the feature space of target samples and improve the accuracy of target recognition based on UWB features,the parameters obtained from the geometrical theory of diffraction(GTD)of target are taken as the parametric features of target recognition,and the HRRP of target in time domain is taken as the time domain features.In order to solve the problem that there is no cross domain feature selection method based on convolutional neural network(CNN)with automatic feature extraction,a parameterized feature selection method of radar target based on CNN back propagation algorithm is proposed by combining with deep learning network,and the importance evaluation values of several different frequency domain parameterized auxiliary features for target classification results are calculated.Secondly,aiming at the problem of how to use time-frequency features for joint recognition more effectively,a joint learning target recognition algorithm based on deep learning is designed.Compared with the classical pattern recognition algorithms such as support vector machine(SVM),convolution neural network,multi-layer artificial neural network(DNN)and so on,the recognition accuracy of the algorithm is compared under different SNR conditions.The experimental results show that the target recognition accuracy of the time-frequency fusion algorithm is higher and the anti noise performance of the algorithm is stronger in the case of lower SNR.2.To solve the problem that the system can't recognize the target correctly in some target recognition scenarios,this paper proposes an unknown class target recognition algorithm based on self encoder and anomaly detection model.This algorithm is different from the common unknown class detection algorithms based on convolutional neural network,and has low dependence on large-scale data.The algorithm assumes that the known class samples and unknown class samples in the sample set conform to the characteristics of Gaussian distribution,and based on this assumption,the threshold of unknown class target recognition is calculated,and the unknown class target recognition task is finally completed.The experimental results show that the algorithm has higher detection accuracy of unknown categories and shorter average running time compared with the common unknown categories algorithm under the condition of higher SNR.