Study Of Radar Automatic Target Recognition Base On High Range Resolution Profile

The high resolution range profile (HRRP) obtained from wideband radar reflectsthe distribution of the target scattering centers along the radar line of sight, whichcontains more information about the structure and shape of the target. Compared withtwo-dimensional or three-dimensional radar imagery, HRRP is easier to be captured.Besides, real-time recognition can be realized. Therefore, the target recognition basedon HRRP is currently an effective approach to identifying airplanes and otherhigh-speed moving targets.This dissertation which is supported by the pre-research projects of GeneralArmament Department of the Eleventh Five-Year Plan, studies feature extraction andselection, classifier design and out-of-database target rejection of HRRP radar automatictarget recognition system, based on the target aircraft recognition. The main content andinnovation is summarized as follows:1. To make full use of the information of target structure and shape included inHRRP, an adaptive difference operator is proposed to extract the target projection lengthalong the radar line of sight, and the targets are rough classified according to the lengthfeature.2. In view of the vague borderline in multi-target classification of classicalsubspace method, we studies a weighted one against one (WOAO) classified strategies.By set weight for each classifier, this method solves the vague classification problem ofthe norm one against one (OAO) classifier in the case of that multiple targets get thesame votes.3. From the perspective of information fusion, an identifying method with multiplecombinations of classifiers which integrate target length and subspace feature isproposed. The length feature is used to rough classify, and the classified results are usedto narrow the search scope of WOAO classifier, so as to improve the recognitionperformance.4. When classical Relief algorithm is adopted to select features for HRRP, iftargets' distribution is uneven on one feature, the evaluated result of this feature will be inaccurate. Thus, a modify Relief (MRe) feature selection algorithm is proposed, whichcan solve the influence of the targets' uneven distribution to the feature weight byincreasing the proportion of each pair of successfully classified samples to weightedaccumulation. In this case, the selected features will be more beneficial for targetclassification5. In order to solve the problem of the out-of-database target rejection in targetrecognition system, we study an improved generalized confidence function to realizereliable identification of out-of-database targets by setting the refuse threshold for everytarget.6. On the basis of biomimetic pattern recognition theory, a rejection andrecognition algorithm based on the chain coverage model is proposed. By means of thisalgorithm, compact and closed geometric boundary is constructed in the feature space oftarget sample. This boundary represents the distribution shape of the target sample. Theacceptance or rejection of the samples is accomplished by judging whether or notsamples are within the target coverage area.The performance of all above-mentioned algorithms are verified by experimentsbased on real measured data and simulated data.