Research On Radar Automatic Target Recognition Based On High Resolution Range Profile

With the development of advanced modem weapon systems and radar electronic technology,people’s demand for radar is not only satisfied with the ability to detect and locate targets.Radar target recognition,as one of the important development directions of modern radar technology,is of great significance to the future weapon system.Among them,the radar target High Range Resolution Profile(HRRP)recognition is the most widely studied.Radar High Range Resolution Profile can provide the information needed for target recognition and can reflect the geometric shape and structural characteristics of the target.Target recognition using High Range Resolution Profile has always been the focus of research in this field.This dissertation combines the relevant theoretical frameworks and engineering techoiques from the aspects of physical characteristics analysis,data pre-processing,feature recognition and classification of High Range Resolution Profile,and carries out research on radar High Range Resolution Profile.The main work is as follows:From the perspective of High Range Resolution Profile physical properties,the basic principles of target recognition are used to analyze the influence of sensitivity issues on High Range Resolution Profile target recognition.The basic ideas and common methods of feature extraction and target classification are discussed.In this paper,the data preprocessing is studied for the sensitive characteristics of High Range Resolution Profile.The target echo region is extracted by first rough screening and then precise screening.Three kinds of distance alignment methods for solving translational sensitivities are analyzed,and the advantages and disadvantages are discussed by comparing the experimental data.The effects and effects of the normalized amplitude are discussed,and the data preprocessing effect is verified experimentally.A target recognition method combined with Generalized Regression Neural Network(GRNN)is proposed.This paper aims at the artificially adjusted parameter spread smoothing factor for the network model in the traditional method,and uses the K-Folder Cross Validation method to train the neural network,and uses the Mean Square Error(MSE)according to the minimum mean square error(MSE).The cycle discriminant method enables the network to automatically filter out the optimum values of the spread factor for the sample parameters and the optimal input and output values for the training samples during the identification process,and continuously corrects the parameter values in the subsequent online identification.The method overcomes the arbitrariness of the spread selection and reduces the influence of human factors on the prediction result.Simulation results show that the target recognition rate and stability based on this method model have been improved.This paper discusses the problem of refusal of radar target in the process of real target recognition.In order to deal with the problem of lacking out-of-the-box training samples in object rejection,a method of artificially generating out-of-library target samples is studied.Two rejection criteria evaluation criteria based on the receiver operating characteristic curve are given.A confidence assessment method based on non-parametric estimation was designed and applied to the experiment of rejecting out-of-library targets for radar High Range Resolution Profile target recognition.