Recognition Of Radar Target Based On The High-resolution One-dimensional Distance

Radar automatic target recognition which is an important part of modern radar has been used widely in the military and civilian fields. High resolution range profiles (HRRP) of radar target contains lots of recognition information of the target. HRRP could be obtained more easily than two dimensions profile. Radar automatic target recognition based on high resolution range profile is studied in this dissertation. The main work is summarized as follows:1. Feature fusion method based on posterior probability is discussed. Target recognition methods based on HRRP which fusing principal component analysis (PCA) and linear discriminant analysis (LDA)'s feature which are extracted from HRRP are proposed and recognition rate is improved.2. Based on UDP (Unsupervised Discriminant Projection) which is a method based on manifold learning, Advanced UDP (AUDP) based on HRRP is proposed and it is expanded to KAUDP with Kernel function. This advanced method process an equal objective function of UDP using simultaneous diagonalization. The features'variance is normalized after the process of diagonalization. The experiment show AUDP has a better recognition rate than UDP.3. KMFA (Kernel Marginal Fisher Analysis) which is a method based on manifold learning is studied and the characteristic that KMFA's intraclass compactness matrix has a null space is found. A null space KMFA method based on HRRP which use intraclass compactness matrix's null space is proposed. The feature is extracted from the null space which has discriminative information of intraclass compactness matrix in this method. The experiment results show this method has good performance.4. Class Information Incorporated principal component analysis (CIPCA) is introduced to target recognition based on HRRP. This method adds samples'information to the PCA and the experiment show CIPCA has a better recognition rate than PCA. In order to improve the performance of CIPCA, the CIPCA is expanded with Kernel function and it is called as KCIPCA.5. Different classifiers are compared using CIPCA and other feature separately. The experiment results show that the performance of different classifier using CIPCA and LDA's feature are different. Feature which want to show the best performance must select appropriate classifier.