Research On PCA And KPCA Self-Fusion Based MSTAR SAR Automatic Target Recognition Algorithm

Synthetic aperture radar (SAR) is a kind of high-resolution active coherent imaging radar, it has the advantages of working all-weather, all-time with strong penetrating power, etc., and has been widely used in military and civilian fields. SAR images automatic target recognition process simulates the process of human optesthesia, analysis and classification, using a computer to complete the SAR image feature extraction and classification, so as to achieve the purpose of access to information of some fields. MSTAR is a classic data set for SAR recognition algorithms testing.In this paper, we propose a principal component analysis (PCA) and kernel principal component analysis (KPCA) self-fusion based MSTAR SAR automatic target recognition algorithm. This algorithm firstly extracts linear feature using PCA and nonlinear feature using KPCA of the original samples, then utilizes the weighted maximum margin criterion (WMMC) to calculate the fusion coefficients, and fuse the extracted features according to them. Then genetic algorithm is used to select the most useful features which will be used for classification in the next step from the fused features. In the simulation, respectively using the nearest neighbor classifier and the linear regression classifier to distinguish the features extracted from SAR images, comparing and analysing the simulation results.Compared with the traditional feature fusion algorithm with preset coefficient, the proposed algorithm calculate the fusion coefficients through the established optimal equation, ensuring that the interval between different classes within the fusion space is maximum, and to some extent, improving the recognition rate of the targets. In the experiment, we compare the target recognition rate of using PCA, KPCA and the proposed self-fusion algorithm, results show that the proposed PCA and KPCA self-fusion based MSTAR SAR automatic target recognition algorithm has a higher recognition rate. The self-fusion algorithm is reasonably designed, time complexity can be accepted. However, this paper does not involve the choice of kernel function and its parameters, in the simulation, we just manually select the optimal kernel function and its parameters for recognition by testing different kernel functions and parameters, this problem will be the focus on in the future.