Study Of Radar Target Recognition Technology Based On High Range Resolution Profile

A high-resolution range profile (HRRP) is the amplitude of coherent summationsof the complex time return from target scatterers in each range resolution cell, whichrepresents the projection of the complex returned from the target scatting centers ontothe line-of-sight (LOS). Since it contains the target structure signatures, radar HRRPtarget recognition has received intensive attention from the radar automatic targetrecognition (RATR) community. By considering the engineering background of theradar HRRP recognition, this dissertation studied the radar HRRP target recognitionfrom the feature extraction, classifier design and noise robust recognition methods.The dissertation is composed of seven chapters, which can be introduced asfollows:In Chapter1, we firstly illustrate the basic conception of HRRP, and then discussedthe fundamental theories of RATR and review some relative works in recent years.Finally, we introduced the main work of our dissertation.In Chapter2, we introduce the target structure signatures based on the scatteringcenter theory, and show some models which are often used in classification, such asadaptive Gaussian classifier (AGC), probabilistic principal component analysis (PPCA)and factor analysis (FA). Then, we extract the time domain feature and frequencydomain feature of HRRP for recognition. Finally, we discuss the sensitivity of radarHRRP recognition method under the condition of small sample or noisy environments.In Chapter3, a new truncated stick-breaking hidden Markov model (TSB-HMM)based on time domain feature is proposed to improve the performance of radarhigh-resolution range profile (HRRP) target recognition. Moreover, a hierarchicalclassification scheme based on TSB-HMM is employed, which utilizes both timedomain feature and power spectral density feature of HRRPs for hierarchicalrecognition. Experimental results based on measured data show that the TSB-HMM isan effective method for radar HRRP recognition, and the hierarchical classificationscheme can largely enhance the average recognition rate. Furthermore, the proposedmethod can obtain satisfactory recognition performances even with very limited trainingdata.In Chapter4, we utilize spectrogram feature of HRRP data to improve therecognition performance, of which the spectrogram is a2-dimensional feature providing the variation of frequency domain feature with time domain feature. And then, a newradar HRRP target recognition method is presented via a TSB-HMM. Moreover,multi-task learning (MTL) is employed, from which a full posterior distribution on thenumbers of states associated with the targets can be inferred and the target-dependentstates information are shared among multiple target-aspect frames of each target. Theframework of TSB-HMM allows efficient variational Bayesian (VB) inference, ofinterest for large-scale problem. Experimental results for measured data show that thespectrogram feature has significant advantages over the time domain sample in both therecognition and rejection performance, and MTL provides a better recognitionperformance.In Chapter5, to improve the recognition performance of statistical modeling basedradar HRRP recognition method under noisy environments, a novel radar HRRPrecognition method based on model modification is proposed. The model modificationmethod has two forms, the one revises the parameters based on the original Gaussianmodels, and the other revises the original Gaussian model and builds a non-Gaussianmodel. These proposed methods revise the original statistical models in the templatelibrary according to the estimated signal-to-noise ratio (SNR) of the noisy test HRRP,which make the statistical models trained by high SNR HRRP samples adaptivelymatch the noisy test HRRP. Therefore, this method achieves radar target robustrecognition. Moreover, we apply the proposed method to AGC and TSB-HMM asexample, and the modification formulas for parameters in these models are givenrespectively. Experimental results based on the measured data show that the proposedmethod can significantly improve the recognition rate and rejection performance withnoisy HRRP test samples, and when the SNR of test HRRP is not accurately estimated,this method still can obtain a good recognition and rejection performance.In Chapter6, we introduced the tied factor analysis (TFA) for HRRP recognition.We extract power spectrum feature of HRRP sample, briefly review the traditional FAmodel, and construct TFA model with power spectrum feature. Experimental results formeasured data show that TFA model has significant advantages over the FA model notonly in the recognition performance, but also the model robustness. These advantageslargely expand the practical applications of TFA.In Chapter7, we summerize the main results of the dissertation, and somerecommendations for future work are given.