Studies On Clustering Algorithm For Multispectral Remote Sensing Images Change Detection

Change detection of remote sensing images means to detect the change information to compare the remote sensing images which are acquired, and extract the useful information at the same area with different time by using all kinds of detecting methods. With the rapid development of remote sensing technology, remote sensing image change detection has been widely used in monitoring of vegetation cover, changing of climate and environmental, disaster damage assessment and monitoring, urban construction planning,military deployment and other fields.In this thesis, taking into consideration each band of the multispectral image between the relevant information and spatial structure, the corresponding change detection methods of multispectral images are studied. Our works completed the following two aspects:1. Proposed a change detection algorithm for the multispectral remote sensing images based on dimension reduction by logistic regression fitting and improved Data Streaming with Affinity Propagation clustering algorithm based on similarity. First, an improved neighborhood filtering pre-processing method is proposed. Then the multispectral images after preprocessing are performed, the single image after dimension reduction is obtained by logistic regression. Finally, the local space structures of multispectral remote sensing images are considered, the change detection results of multispectral images are obtained by using the Data Streaming with Affinity Propagation clustering algorithm of the improved density similarity. The improving of change detection accuracy of our method has been demonstrated by comparing with other corresponding methods on three real multispectral remote sensing images.2. A change detection algorithm for the multispectral remote sensing images based on improved water index and Bayesian Network clustering method is proposed. At first, according to the spectral reflectance of different ground objects are difference in the multispectral remote sensing images, an improved water index method is proposed in order to enhance the information of the water area and suppress of other ground object information, Finally, the conditional probability of the various bands is computed by the mutual information, and the Bayesian Network is used to differentiate the change and non-change classes. The effective of this method has been demonstrated by comparing with other corresponding methods on three real multispectral remote sensing images.