| This thesis revolves around how to effectively segment images, how to calculate the texture features and clustered into two groups, which proposed the following two algorithms:(1)Aiming at the problems that remote sensing image change detection has strong correlation with segmentation method and scale, this paper proposes a new object level image change detection method based on multi-scale segmentation and fusion. First of all, an improved minimum spanning tree segmentation algorithm was adopted to segment two images with multi-scale together, so as to ensure the space position of segmentation image spot is exactly the same. Then spectral and texture features was used to calculate the different index of corresponding spots in multi-scale, and finally the fusion of different scale was applied to detect the image change. Experiments showed that the proposed method obtained better detection result and accuracy.(2)To obtain complementary information of difference image and gain a better change detection result, this paper proposes an unsupervised change detection algorithm. The technique is based on minimum spanning tree(MST) clustering. First of all, a normalized neighborhood ratio approach is adopted to obtain the difference image of two images acquired in the same geographical area at different moments. Then divide pixels into changed pixels, unchanged pixels and uncertain pixels according to gray difference histogram. Aiming at uncertain pixels, use minimum spanning tree algorithm twice to clustering: texture features and minimum spanning tree algorithm are used for first clustering, then calculate every clustering’s average weight, and regarded first clustering results as nodes to calculate the minimum spanning tree again. Finally, on the basis of optimal objective function, the uncertain pixels belong to the two clusters: changed pixels and unchanged pixels.The proposed algorithm can obtain higher detection accuracy and shorter running time. |
PDF Full Text Request