Radar Target Distance As Identification Method

With urgent requirement for radar to get more information from the battlefield, radar Automatic Target Recognition (ATR) has emerged to fulfill this task. Recently, the research of ATR is rapidly developed in virtue of high resolution radar. High range resolution profile (HRRP) reflects the target structure signature along the radar's line of sight, which can be easily acquired and processed. Thereby radar target recognition using HRRP has received increasing attention. Based on this research, new ideas and main work are arranged as follows:1. On the basis of scatter-center model, HRRP of radar target and its character are studied. In order to overcome the shift influence on HRRP, zero phase method is adopted to align HRRP, which retains the shape feature of HRRP. Experiment results demonstrate: Compared with DFT method, the profile processed by zeros phase method performs more poorly when using nearest center (NC) method for classifying, which disagrees with the ordinary suggestion.2. Analyses the traditional subspace learning methods. Based on the basic criteria of SKM (Supervised Kampong Measure), MSKM (Modified Supervised Kampong Measure) method is proposed to modify the criterion matrix in a weighted manner, which eliminates the neighboring inhomogeneous sample points from overlapping in the lower dimensional subspace, meanwhile, ensures each data point distributes closely to its corresponding center, which facilitates target recognition using NC method.3. Applies Canonical Correlation Analysis (CCA) method and its kernel extension (KCCA) to radar target recognition. KCCA often suffers from singular matrix problem when solving the eigen-equation. To avoid this problem while loss as less identifying information as possible, this dissertation develops the Dual space KCCA (Dual-KCCA) method, which extracts the sample features respectively from two different feature subspaces, main space and supplementary space namely. Then combaine the feature vectors in a weighted manner for target recognition.4. Gives brief introduction of classics manifold leaning method. Discusses the supervised linear graph embedding method, which is called Local Discriminant Embedding (LDE). Following the same principle of LDE, Modified LDE (MLDE) method is proposed to improve the identification ability by modifying the similarity matrix of LDE. The project direction of MLDE focuses on maintaining the identifiability of those neighboring inhomogeneous sample points, which space a certain distance originally. Equally, if the neighboring homogeneous sample points are distant from each other, the project direction will also emphasize on gathering corresponding lower dimensional points closer. Finally, the kernel pattern of MLDE—KMLDE (Kernel MLDE) is developed to make HRRP feature vector much more identifiable in the noneliner high dimensional space.