Font Size: a A A

Research On Radar HRRP Target Recognition Technology Based On Feature Extraction

Posted on:2020-03-02Degree:MasterType:Thesis
Country:ChinaCandidate:W C ChenFull Text:PDF
GTID:2428330590472228Subject:Measuring and Testing Technology and Instruments
Abstract/Summary:PDF Full Text Request
Due to Radar High Resolution Range Profile(HRRP)'s advantages of easy acquisition,small data size and processing simplisity,Radar Automatic Target Recognition(RATR)technology based on HRRP has become a hot research field of radar signal processing.Effective achievement of Radar Automatic Target Recognition with HRRP is of great theoretical significance and practical engineering application value for maintaining the integrity of national territorial sovereignty,promoting the recognitive ability and intelligent degree of radar systems.This thesis focuses on the problem of feature extraction in Radar Target Recognition technology based on HRRP,and the main research contents are as follows:(1)Sensitivity elimination method of HRRP is studied,and frame center extraction method based on Frame Maximum Likelihood Profile(FMLP)is proposed.The method selects the original HRRP signal which has the greatest similarity with other HRRP signals in each azimuth frame as the frame center of the corresponding azimuth frame.Compared with the commonly used frame center of mean vector,Frame Maximum Likelihood Profile retains original details of HRRP to maximum extent.(2)For different targets with large differences in structure and size,on the basis of extracting geometrical features such as equivalent target length and number of equivalent strong scattering points,Radar Target Recognition is effectively realized by using Frame Maximum Likelihood Profile as frame center.Experimental results show that compared with frame center of mean vector,Frame Maximum Likelihood Profile has better recognition performance in Radar Target Recognition based on geometric features than frame center of mean vector.At the same time,a Radar HRRP Target Recognition method based on Multi-SVM Adaptive Weighting Fusion Decision is proposed,which effectively realizes decision-level fusion recognition of multiple geometric structure features.(3)A Radar HRRP Target Recognition method based on Selected Principal Component Analysis(SPCA)is proposed.Inspired by the idea of improving the separability of reconstructed signals,on the basis of Radar HRRP Target Recognition method based on Principal Component Analysis(PCA),Fisher criterion is used to select the principal component with the largest separability between different targets,with which testing HRRP signal is reconstructed for recognition in the algorithm.Experimental results show that compared with Radar HRRP Target Recognition method based on Kernel Principal Component Analysis(KPCA),the proposed method has the same good recognition performance under the circumstance of reducing algorithm complexity and improving algorithm real-time performance.(4)A Radar HRRP Target Recognition method based on Stacked Frame Maximum Likelihood Profile-Trajectory Similarity Auto-encoder(Stacked FMLP-TASE)is proposed.Based on the characteristics that Deep Auto-encoder(DAE)can adaptively extract the deep information of signals,the algorithm constrains errors between reconstructed spectrum of all training HRRP samples and spectrum of Frame Maximum Likelihood Profile in each azimuth frame to be minimal.At the same time,considering the trajectory continuity characteristics of training HRRP samples in each azimuth frame,errors between reconstructed spectrums of the adjacent training HRRP samples is further constrained to be minimal.Experimental results show that the separability of deep features of different targets extracted by Stacked FMLP-TASE is significantly enhanced,which further improves the performance of Radar HRRP Target Recognition based on Deep Auto-encoders.
Keywords/Search Tags:High Resolution Range Profile (HRRP), Radar Automatic Target Recognition (RATR), Frame Maximum Likelihood Profile (FMLP), Geometric Structure Feature, Selected Principal Component Analysis(SPCA), Deep Auto-encoder
PDF Full Text Request
Related items