Change Detections Algorithm In Multi-temporal Satellite Images

Satellite images change detection technique is an integrated technology thataccording to the remote sensing images of the same region but not the same phase,and using image processing methods to detect the changes of the surface features ofthe region, and by extracted change information through a comprehensive analysis ofmulti-temporal remote sensing images, finally achieve the remote sensing monitoring.Now, the remote sensing image change detection technology has become a hot issuefor the research of the satellite images, and it has been used in many fields, such as:the assessment of the environmental hazard the monitor of the forest resourceschanges, the monitor of the land resources changes, the prospect of the agriculturalresources, the dynamic monitor of military targets and other fields.According to the difference of analysis levels the algorithm of the changedetection can be divided to three categories: the pixel level, feature level and targetlevel; According to whether be classified the first it can be classified to two types:classification comparison method and the direct comparison method; According to theprocess of classification it can be classified to oversight and supervision two types.Two algorithms in this paper belong to unsupervised change detection algorithms inpixel level.An dual-threshold change detection algorithm in multi-temporal satellite imagesbased on nonsubsampled contourlet transform is proposed in chapter four. Thealgorithm use the NSCT extracted the multi-scale and multi-direction texturecorresponding to each pixel in ratio logarithm images, reconstructed the ratiologarithmic image, then used the expectation maximization (EM) algorithmprocessed the reconstructed images to produce dual-threshold, distinguished by thecriterion of dual-threshold to get no change image and different changes (changes inregional enhanced class and change the image of the region weakened class), and, ultimately, get the change detection image. After that,processed the reconstructedimage using the K-means clustering algorithm,conversed the problem ofchanges detection in regional to the problem of classification between the two types,then got area with changes and no change by the threshold criterion, finally, gotdetection images with changes in regional.In the fourth chapter, proposed a new remote sensing image algorithm which wasa pixel-level, unsupervised, multi-temporal detection algorithm combined thenonsubsampled Contourlet transform (NSCT) and k-means clustering (the K-MEANS)algorithm.The algorithm combined the neighborhood information of each pixel,usingnonsubsampled Contourlet transform to extract the features of the vector of theratio logarithm image, and reconstructed the ratio logarithm image.The experimental results show that compare with the comparison algorithmPCA-KMEANS,the proposed NSCT and the EM algorithm has a higher detectionaccuracy and lower missing rate against Gaussian noise and speckle noise. Comparewith the comparison PCA-KMEANS, the NSCT and the K-MEANS algorithmproposed in chapter4th has a higher noise immunity and detection accuracy with theGaussian noise and speckle noise.