Change Detection Of Multi-Temporal Remote Sening Images Based On Wavelet Analysis And Clustering

Remote sensing science and technology as an advanced tool that broad view of landscapesand observed field for people. The unparalleled superiority is showed to natural resourcemanagers and researchers in government agencies, conservation organizations, and industry. Bydetecting the change of remote sensing images acquired in the area but different time, allowsresource managers to monitor landscape dynamics over large areas.Some key change detection techniques of multi-temporal remote sensing images areresearched in this paper. We have finished the following aspects work:1．The general introduction of the various factors is considered by the remote sensing changedetection and the processing flow. The existing change detection methods are also morecomprehensive and summarized comments. And, three kinds of building the different image aredescribed in detail.2．A change detection method of remote sensing images based on algebra and expectationmaximization threshold are introduced. In order to overcome the difficult of establish changethresholds in difference image, extraction of change region are treated as statistical decisionproblem. Under the hypothesis that pixels in the difference image are independent of one anotherand can be modeled by mixture Gaussian distributions, using the Expectation-Maximization(EM)algorithm to estimate the values of model parameters,then the Bayes rule for minimum error isapplied in order to select decision threshold.3．A change detection method of remote sensing images based on PCA (principal componentanalysis) and k-means clustering is proposed. As the top method considered the pixels are allindependence and the contextual information is ignored, the result was not good enough. The contextual method that difference image is partitioned into h hno overlapping blocks. Thenthe eigenvector space is created by applying the PCA on these blocks collected from the entiredifference image. The feature vector at each pixel position is computed by projecting the localchange of the pixel values onto the eigenvector space. The binary k-means clustering is thenemployed on feature vectors to compute the final change detection result.4． A change detection method of remote sensing images based on wavelet fusion andk-means clustering is proposed. Considering the advantages of image differencing and log-ratiooperator, a method constructs difference image by wavelet fusion the results of differencingoperation and log-ration operation. Then, by taking into account the spatial-Neighborhoodinformation, a change detection algorithm based k-means clustering is developed to obtainquantitative detection results.5． A change detection method of remote sensing images based on undecimated discrete wavelettransform (UDWT) and kernel k-means clustering is proposed. The difference image is decomposed usingUDWT to obtain multi-scale representation. By taking advantages of kernel method, multi-scale feature vectoris clustered the two groups of pixels belonging to the ‘change’ and ‘no change’ classes. What’s more, somenonlinear nature of the change is extracted. In order to overcome bad initializations, EM algorithm isused to estimate initialization parameters on the original resolution of the difference image.The experiments bring out the efficiency and reliability of the proposed technique to interpretevery change.