Studies On Radar Target Recognition Based On Nonlinear Approaches

It is known to all that radar target recognition (RTR) is very valuable for military application. With the development of radar technology and the complexity of the environments of radar targets, nonlinear approaches are meaningful for RTR. This dissertation researches RTR using nonlinear approaches.In the first chapter, the previous research approaches of RTR are reviewed. The nonlinear approaches used in RTR field are stressed. The research background and content of this dissertation are introduced finally.In chapter 2, a method is presented which gets radar target high dimensional structure information from one dimensional range profiles based on Independent Component Analysis (ICA). One dimensional range profiles has shortcoming in RTR for its limitation of target information and sensitive of target azimuth. In order to get more sufficient and stable characteristics, getting target 2D and 3D structure information from multi-azimuth range profiles is a valid way. On base of this idea, an approach based on ICA is presented. The new approach can get stable target structure information and needn' t the information of target azimuth or the visual angles between the range profiles. The approach computation is not enormous, which in favor of the realization on real-time.Angular glint is a characteristic of Complex target, which can reflect the target structure. Researching of target angular glint can help suppressing it and classifying target. Nonlinear dynamics characteristics of target glint serials are detect by method of surrogate data in chapter 3. And then, an algorithm of suppressing angular glint is presented, which based on results weighted chaotic prediction. Simulations show the new algorithm is super to that of distances weighted chaotic prediction. In reconstructed phase space of angular glint serials, the fractal features of strange attractions are calculated as target nonlinear characteristics used in targets classify. Simulations show the method is valid.Finding optimum feature extracting algorithm and classes' centres in feature space are two important problems. In chapter 4, two algorithms for selecting the optimum classes' centres in Fisher' s criterion based on LDA and NLDA are researched. The first algorithm calculates the optimum classes' centres of LDA by Genetic Algorithm. The second algorithm calculates the optimum classes' centres of NLDA by method of grads descending. Both of them are applied in one dimension range profiles' feature extracting and recognition, and improve the target recognition.Summary of this dissertation is made and the problems needed further research are pointed out in chapter 5.