The Research On Automatic Acquirement Of Target's ROI From SAR Imagery

The topic of this thesis is to deeply investigate the automatic techniques of acquiring target's ROI (Region-of-interests) from SAR images rapidly and effectively, which is among the demands of SAR image interpretation applications. Aiming at developing the practical techniques of automatically acquiring target's ROI, some key techniques such as statistical modeling of SAR clutter images, automatic target detection and discrimination from SAR images are systematically studied by theoretic analysis and experimental validation with lots of real high-resolution SAR data.In chapter 1 the research background and significance of automatically acquiring target's ROI from SAR images are summarized. Then, the progress of this research is reviewed briefly and the currently existing problems are pointed out. Finally, the main work and the innovations of this thesis are summarized.Chapter 2 deals with the problem of statistical modeling of clutter in SAR images. Firstly, the relevant techniques of statistical modeling of clutter in SAR images are reviewed in detail. Secondly, based on this review, the multiplicative noise model (product model) for SAR images is analyzed comprehensively to reach some significant conclusions for the application and development of statistical modeling of clutter in SAR images. The clutter is analyzed comprehensively with the real SAR data characterizing different terrain categories. A conclusion is reached that the G°distribution is most appropriate to describe the statistical property of clutter and is relatively accurate for statistically modeling the homogeneous, heterogeneous and extremely heterogeneous clutter regions.Aiming at building the practical automatic target detection process, the problem of automatic target detection in SAR images is studied in chapter 3. Based on an extensive survey of the existing studies and the conclusion of chapter 2, we focus on the CFAR detection algorithms. An intelligent and fast CFAR algorithm based on automatic censoring is proposed for target detection in SAR images. This algorithm avoids the limitation of the conventional CFAR algorithms and can intelligently decide the clutter environment of detection. By introducing the G°distribution as the statistical model of clutter in this algorithm, the uniform CFAR detection is established under different clutter environments to enhance the automatic degree of target detection. Based on the theoretical analysis of the computation cost of the intelligent CFAR algorithm, the corresponding fast algorithm of the intelligent CFAR algorithm is also proposed to improve the practicality of target detection.In chapter 4, in order to conduct the powerfully practical automatic target discrimination process, the problem of automatic target discrimination in SAR images is studied. Based on the extensive summarization of the existing studies, a series of studies including feature extraction and selection of target discrimination, the design of discriminator, and so on are done. A new scheme of target discrimination in SAR images, consisted of frames, models and algorithms, is proposed. Under such a scheme, a global frame, combining orderly the algorithm based on feature extraction and that based on knowledge, is then proposed. Moreover, in the method of target discrimination based on feature extraction, a "loose-coupling" model is given. The existing features are chosen and three new features about the contrast are given under the "loose-coupling". Meanwhile, an algorithm of feature selection based on Genetic Algorithm is also modified to solve the problem that the existing algorithm can not evaluate the goodness-of-features comprehensively. The weighted quadratic distance discriminator is designed to improve the performance of target discrimination. Finally, a method based on the knowledge of target groups to remove clutter false alarms is also given.Chapter 5 concludes the research of this thesis. Some problems and interesting area for future research are pointed out.