Change detection in remote sensing images finds extensive applications, such as military target surveillance, urban planning, environmental monitoring and natural disaster detection. Compared to low- and median-resolution remote sensing images, high-resolution remote sensing images contain more details about the earth surface and enable detecting more subtle land use and land-cover changes. However, due to the complexity of high-resolution remote sensing images, the change detection techniques in the literature are immature, and mostly insufficient in terms of accuracy, speed and versatility. In consequence, it is very urgent to develop methods and algorithms for high-resolution remote sensing image change detection. Based on a comprehensive survey of the state of the art of high-resolution image change detection, this thesis proposes two methods taking into account the characteristics of high-resolution remote sensing images and the merits of sparse representation. Specifically, A new method based on discriminative dictionary learning is proposed for high-resolution image change detection. In this method, the feature representation and the classifier are learned simultaneously, so as to improve the discrimination capability. Experimental results on the QuickBird image datasets confirm the effectiveness of the proposed method. A hierarchical clustering method based on sparse representation is proposed for high resolution image change detection. It learns hierarchical dictionary to model the multimodal distribution of change features, and the sampleto- class is measured by the sparse representation error. Compared to the traditional clustering-based change detection methods, the proposed approach is novel in dealing with the multimodal distribution.
修改评论