Radar Target Recognition Based On Sparse Learning

The radar target recognition techonology is a kind of non-contact detection techonology, which launch and receive electromagnetic wave based on fixed or moving target. With the development of the international situation, radar target recognition is more and more get the favour of researchers around the world. With the application of high resolution radar system, it is possible to make sure the more detailed target geometry information and detail information can be obtained. The two systems of high resolution radar(HRR) and Synthetic Aperture Radar(SAR), whose radar echo signal and SAR image as a typical high resolution radar signal also become the research hot spot in the field of target recognition. In this paper, on the basis of sparse learning theory, radar target recognition is studied based on HRRP target and SAR image target, and the main work is summarized as follows:1. Research on Sparse learning theory. Firstly, this paper is studied in three typical sparse modeling methods, three kinds of classical thinning algorithm, as well as the application of sparse learning. Then, study the HRRP target and SAR image sparse representation methods, and analyses its sparse, respectively.2. An algorithm of SAR image denoising based on Bayesian model and Shearlet transform is presented in this paper. The algorithm takes advantage of both the spatial correlation of sparse coefficient and through the bayesian model to obtain the dynamic noise threshold. And the proposed algorithm can filter speckle, as well as can restrain the image edge information better. Firstly, SAR image which through logarithm transformation, transformation to Shearlet domain. Then, according to the statistical features of the sparse coefficient, this paper use bayesian model to modeling of noise detection. Finally, the noise pixels of the SAR image are smoothing by the weighted adaptive algorithm shrinkage. The experimental results verify the feasibility and effectiveness of the proposed method based on MSTAR database.3. A Dynamic Sparese K-SVD(DSK-SVD) dictionary learning algorithm is presented. The highlight the advantages of the proposed algorithm are able to dynamically calculating the sparsity of sparse coding, and the dictionary atoms are updated in parallel. Firstly, this paper use the mutual-coherence of dictionary matrix to define the sparsity of sparse coding and dynamically control the sparsity of the sparse coefficient, in the process of sparse coding. Then, update the dictionary atoms and sparse coefficient using the rule of parallel atomic updates. The experimental results verify the feasibility and effectiveness of the proposed method based on MSTAR database and HRRP database.4. A new method is proposed for the radar target fusion recognition based on D-S evidence iterative discount theory. The algorithm use DSK-SVD algorithm study the characteristics of the training samples. And the reconstruction error of test sample is used to define the basic probability assignment(BPA) function. Firstly, using the the the confusion matrix and BPA function to compute discount factor. Then, each of iteration is used to modify the source of evidence, until the conflict coefficient is less than the given threshold. Finally, fusion recognition is achieved by using the revised evidence. Compared with other typical fusion recognition method, the proposed method under the condition of small sample can keep better recognition performance.