Classification And Annotation Of High Resolution Synthetic Aperture Radar Images Based On Extended Supervised Topic Model

Synthetic Aperture Radar images have many innate shortcomings compared with optical images, such as the missing of the structural characteristics of targets, hard to understand the scene, difficult to find association between targets and scenes. Because of these shortcomings, SAR images are difficult to understand for the end users. So the interpretation of SAR images has always been the core foundational and common technical problem in the application of SAR images.Now SAR images are moving towards the direction of high resolution and multi-polarization. Facing new situations, such as the improvement of the distinguishing ability of SAR system, the introduction of polarization information, the increasing of handling objects, requirements of precise detection and massive data, it is necessary to develop a highly accurate, fast, highly automated and easy-to-manual operational interpretation technology for high resolution SAR images.This thesis is facing to the scene categorization and scene annotation problem in the interpretation of high resolution SAR images. For the domestic situation, this thesis is focusing on solving the problem of mapping the semantic gap and the problem of scene prior information acquisition in the interpretation of high resolution SAR images. In this thesis, we propose a high resolution SAR image classification and annotation method based on extended supervised topic model.Three different scenes of high resolution SAR images are used for the verification of proposed method. Through these experiments, the proposed method can finish the work of scene categorization and scene annotation for high resolution single polarization or full polarization SAR image at the same time accurately and conveniently, under the situation of a small training set, a small amount of manual operation, and a small amount of knowledge input.