Support Vector Machine-based Sar Target Classification And Identification

Image interpretation, regarded as the important content of synthetic aperture radar (SAR), has widely draw attraction of research institution all over the world. Target classification and discrimination is the key process to implement image interpretation. Therefore, the speed and accuracy of target classification and discrimination are the crucial factors to obtain high-quality image interpretation. Because of low computational complexity and outstanding processing ability for high-dimensional sample of support vector machine (SVM) in target classification and discrimination, SAR image interpretation based on SVM has become the hot topic.The thesis focuses on the vital problem derived from SAR image target classification and discrimination based on SVM, which has the following research contents:1,Introduce the fundamental of SVM, analyze the strategies for non-linear classification and multi-kind target classification, and introduce SVM into SAR image target classification and discrimination.2,Aiming at the problem of high-accuracy SAR image segmentation, we research on the image segmentation algorithm based on Markov Random Field (MRF) including maximum a posteriori MRF segmentation system, optimization method of posteriori energy function, etc. The segmentation comparison is implemented through MSTAR tank data.3,Aiming at the optimization problem of posteriori energy function, we research on these optimization algorithm including ICM, Gibbs sampling, SA, MMD, and compare these algorithm by MSTAR tank data.4,Aiming at the problem of target azimuth estimation in SAR target classification and discrimination, we use a target azimuth estimation approach without azimuth grouping to improve the estimation accuracy, which is verified by MSTAR tank data.5,Aiming at the problem of SAR target classification and discrimination, we process SAR target classification and discrimination with SVM based on principle component characteristic. The experiment results demonstrate that the method can obtain high classification rate even no calibration for target standard position, but the classification rate declines along with the increasing of azimuth space.Besides, the thesis developed the monitoring experiment with SAR target classification and discrimination based on SVM by using the San Francisco Bay Area L-band fully-polarimetric SAR data collected by AIRSAR system. The experiment results show that SVM has the advantages of fast speed and high accuracy in multi-kind target classification.