Studies On Change Detection Of Multispectral Remote Sensing Images Based On Spectral Reflectance Value

Change detection of remote sensing images means to detect the change information by comparing the remote sensing images which is acquired at the same area with different time. At present, the research on change detection of remote sensing images is becomed a hotspot, and widely used in many areas, such as disaster monitering and assessment, vegetation cover, urban management planing, agricultural investigation, meteorological monitoring and so on. Rich spectrum information in the multispectral remote sensing images increases the credibility to detect the change, thus the application of change detection of multispectral remote sensing images is increasing.This thesis revolves around how to effectively build the difference image, use the relevance and difference between each band to detect change information in difference image, which completes the following two aspects:1. Proposed a change detection algorithm in multispectral remote sensing images based on change neighborhood graph and neighborhood probability fusion. First, the change magnitude of spectral reflectance, the change angle mapping of spectral reflectance and the change neighborhood graph are computed. Then the spectral clustering(SC) algorithm is used to get the initial change detection result, respectively. Finally, the initial change detection results are fused with neighborhood probability fusion method, then the final change detection result is obtained. This method has been validated by testing on three real multispectral remote sensing images that the segmentation accuracy is improved, the number of missed pixels and false pixels in final change detection is reduced.2. Proposed a change detection algorithm in multispectral remote sensing images based on affinity propagation(AP) clustering algorithm. At first, the initial bands of both temporal are used to produce new bands according to affinity propagation clustering algorithm. Then, the new bands are used to get difference map by calculating the merged class difference square value of each pixel. The class marking map is generated by K means-based affinity propagation clustering algorithm upon the difference map, and the k-means clustering algorithm upon the class marking map is used to get the final change detection result. This method has been validated by testing on three real multispectral remote sensing images.