The Remote Sensing Imagery Change Detection Using Pattern Recognition Knowledge

The remote sensing imagery change detection can be categorized into three classes according to the aims of the processing: the change detection of the specific targets, such as changes of the airports, the bridges, the harbors, the missile bases etc.; the change detection of the linear shape targets, such as changes of the roads, the airports, the buildings and the other linear targets whose outlines can be described by some lines; the change detection of large area targets, such as the changes of the cover of some region, the development of the cities, the disaster evaluation of the floods and so on. In this paper, the change detection methods using pattern recognition knowledge for the specific targets, the linear shape targets and the large area targets are discussed.The change detection based on the target detection is proposed to realize the specific targets change detection. The steps of this change detection method are: registering the multi-date image pairs, modeling the specific targets, detecting the specific targets, determining the positions of the specific targets, comparing the positions of the specific targets in the reference image and the detecting image respectively, reporting the change information. The model of the specific targets is a general model, which only contains the common features of the homogenous targets. In the experiments, the detection accuracy of airport model proposed by the paper reaches to 100%. After attaining the airports' positions, we compare the airports' positions of the reference image to that of the detecting image to detect the changes of the airports.The change detection based on the edge detection is proposed to execute the linear shape targets change detection. The steps of this change detection are: registering the multi-date image pairs, regularizing the image pairs, extracting the edges of the image pair, matching the edges of the edge-image pair, computing the edge difference imagepair, marking the difference edges in the copies of the image pair. In this part, a new edge operator - sine operator is presented. The detection performance, the localization performance and the single response to the single edge, the three criteria of the edge operator performance, are deduced. The sine operator has fine tolerance ability to noise and can detect the slight changes of the gray value. The theoretic results and experimental results demonstrate that the sine operator is an excellent edge operator.The change detection based on the clustering of the imagery is proposed to achieve change detection of the large area targets. The steps of this change detection method are: registering the multi-date image pairs, clustering the image pairs respectively, finding differences of the clustered-image pair, marking the differences using the striking colors in the copies of the image pair. The accuracy of the clustered-image is the primary problem in this change detection method, It is very important to find a robust image clustering method. In this paper, a new robust image features extracting method and a new image clustering method are proposed. The experimental results prove that the new method to extracting image features and clustering image has achieved good accuracy of the clustered-image.