Some key techniques related to change detection of VHR images are studied in depth in the dissertation. The main contributions of this dissertation include the following issues: (1)The difficulties in change detection of VHR images are analyzed from the viewpoint of the imaging principle, based on which a new change detection framework is presented. (2)A new approach based on SIFT and the distance between the generalized tight pair-wise prototypes is proposed for VHR remote sensing image registration. The proposed method is verified to be effective in increasing the number of matches and the number of correct matches. The relation between different matching methods based on SIFT is established by the concept of the generalized tight pair-wise prototypes. (3)A hybrid approach is proposed for remote sensing image registration. By combining the merits of the feature-based and area-based approaches, the proposed hybrid method can achieve a better balance among accuracy, robustness, efficiency and automation. (4)A novel object-level approach based on multiscale fusion is presented for change detection of VHR images. Object-level change detection is helpful for improving the discriminability between the changed class and the unchanged class, and multiscale fusion is beneficial for mitigating the dependance of the change and the object on the scale. (5)A novel multiscale change detection approach is proposed based on scale propagation. By taking advantages of coarse-to-fine strategy, the different statistical distributions of change features at different scales are captured. The proposed approach is high in accuracy, efficient in computation, robust to mis-registration and view-angle variation. (6) A fast object-level approach is presented for change detection of VHR images based on the progressive transductive SVM. The computation efficiency is improved significantly by taking advantages of fast multi-temporal segmentation and object-specific change feature classification, and object-level change detection is implemented automatically by utilizing the progressive transductive SVM. (7)To confirm the effectiveness of the approaches proposed in this dissertation, a prototype system is developed for change detection of VHR images.
修改评论