Study Of Radar High Resolution Range Profile Target Recognition Based On Dictionary Learning

High resolution range profile(HRRP)has the characteristics of accurately reflecting the physical structure information of the target.Therefore,the radar target recognition technology based on HRRP becomes an important branch of radar automatic target recognition(RATR).In recent years,with the rise of dictionary learning theory,it has been widely used in pattern recognition and signal processing.In this thesis,we aim at radar target recognition technology based on HRRP,and study the division of target angle domain,the feature extraction and classifier design based on dictionary learning,nonlinear feature extraction and data dimensionality reduction.The main content is summarized as follows:(1)This thesis introduces the basic theory of HRRP,discusses the research status in recent years,and analyzes the key technologies and common solutions to radar target recognition based on HRRP.(2)In the process of extracting HRRP features and classify targets by using dictionary learning algorithm,due to the fact that HRRP attitude sensitivity leads to excessive redundancy of the dictionary,the maxium probability difference framentation segment algorithm based on probabilistic principal component analysis model is proposed to divide the target angle domain.The power spectrum of HRRP with obvious attitude changes is adaptively selected to form an initial dictionary,which effectively reduces the redundancy of dictionaries.At the same time,a statistical dictionary learning(SLC-KSVD)algorithm is proposed,and the sparse similarity error is introduced to optimize the traditional dictionary learning criterion to get the optimal dictionary and the optimal classifier.The measured data of radar show that the proposed method realizes the accurate identification of HRRP with low signal-to-noise ratio.(3)The methods of non-linear feature extraction and data dimensionality reduction are discussed.Two types of nuclear discriminant analysis methods are compared and analyzed: the discriminant analysis based on global mean and the discriminant analysis based on local mean.Inspired by the above methods,this thesis presents a method based on kernel joint discriminant analysis(KJDA),and uses the nearest neighbor classifier to make classification decisions.The experimental results based on HRRP data of radar fully verify that KJDA has the advantage of enhancing the data separability while reducing the data dimension.