Research On Radar Target Recognition Based On High Resolution Range Profile

Detection and ranging are the two basic functions of radar. However, they have been far from enough to meet the growing demands of the modern radar to acquire more and more target information. In many military and civilian applications, it’s not only necessary to detect the target, but also desirable to distinguish the type or identification of the target. Consequently, radar target recognition has grown to be one of the important research fields in modern radar signal and information processing. The wide band of radar signal provides high resolution along range dimension and makes the high resolution imaging of a target to be possible. Particularly, HRRPs reflect the radial distribution of scattering centers of a target and are easy to acquire and process. For this reason, it has been said to provide a radar target recognition method which is of huge potentials. In this dissertation, focusing on such key issues as robust feature extraction, multiple features combination and fusion, and system design, we study in depth the topic of HRRP based radar target recognition both theoretically and experimentally. Below are the primary work and contributions of this dissertation.(1) Two classical manifold learning algorithms including neighborhood preserving projection(NPP) and local tangent space alignment(LTSA), are studied, which are proved to be able to loosen HRRP’s sensitivity to the attitudes of targets. Based on the idea of NPP and LTSA, an enhanced neighborhood preserving projection(ENPP) and a linear discriminant local tangent space alignment(LDLTSA), and their corresponding kernelized versions, enhanced kernel neighborhood preserving projection(EKNPP) and kernel discriminant local tangent space alignment(KDLTSA), are proposed respectively, for RRP based radar target recognition. Experimental results show their effectiveness and priority, compared with their counterparts.(2) With very limited training samples, the performances of many classical subspace learning methods degrade. For this problem, the subspace learning methods based on the point-to-space measurement are studied, and two novel algorithms, neighborhood feature space discriminant analysis I and II( NFSDA-I and NFSDA-II), are proposed for HRRP based radar target recognition. Experimental results indicate the proposed methods bear more discriminative power and can achieve better recognition performance, compared with their counterparts.(3) Some potential geometrical and structural features of HRRP are analyzed physically. By using statistical method, eight different features are extracted from HRRP, which can represent different geometrical information of a target. Moreover, based on the idea of multiple features combination, eight combined features are obtained by selecting different single features and combing them together. Experimental results demonstrate the effectiveness of some extracted features, such as entropy and irregularity. The excellent performance of multiple feature combination is also shown.(4) Some research achievements in auditory recognition on spectral envelope are introduced in HRRP based radar target recognition. Nine features are extracted from the spectral envelope of a HRRP, and are combined together according to the correlation relationship among them, so as to obtain another 21 combined features. Experimental results show that spectral envelope and some of its attributes are effective to deal with the problem of HRRP based radar target recognition, thus deserving further and deep study. Moreover, the method of combining multiple spectral envelope features indicates its prominent benefits.(5) The data fusion framework for radar target recognition and some mainstream fusion methods are studied. A HRRP based radar target recognition scheme based on multiple features fusion is presented, where four different kinds of features are extracted from HRRP, two different classifiers are utilized for classification, and the Dempster-Shafe theory is employed for data fusion and final decision-making. Experimental results demonstrate the effectiveness of the presented scheme and indicate the promising performance of the Dempster-Shafe theory for radar target recognition.(6) A radar target recognition system which is based on a S-band wide-band digital array radar(WDAR) system with 16 elements linear array is analyzed. The signal and information processing subsystem of the WDAR system is constructed based on OpenVPX bus, and an open software architecture for target recognition is built to accommodate the high-speed serial bus.