SAR Image Change Detection Based On Spatial Correlation

Change detection in Synthetic Aperture Radar (SAR) image is to analyze thechanged information of two SAR images which are obtained in the same area anddifferent time. With the high and new technology development and SAR image dataaccumulation, SAR image change detection is widely used in earthquake disasterassessment, oil spills on the sea surface, forest fire, urban development and militaryreconnaissance.Considering the shortcomings of the existing methods and the characteristics ofSAR image, this paper completed the following three works.(1) We propose a two-step change detection method for SAR image, including thesubtraction image segmentation and improved the post classification comparison. Weremove the unchanged region in the subtraction image by a threshold value. Theremaining points are contained in many regions, we remove the small regions whichcaused by speckle noise to get a rough change map. In the second step, the Bayesianclassifier is applied to each of the original images to get the classification maps, wherethe samples for training are denoted by the rough change map obtained in the first step.By combining the two classification maps, an approximate difference map is got.Finally, the rough change map and the approximate difference map are fused to get asynthesized map. The experimental results have demonstrated the effectiveness of theproposed method.(2) We propose a SAR image change detection method based on adaptiveneighborhood. Firstly, find the related areas of all the pixels in the two original imagesuse an adaptive neighborhood method. Use the related area replace the33neighborhood area which used in the mean ratio method to get an adaptiveneighborhood mean ratio map. The classic spectral clustering method is applied to theadaptive neighborhood mean ratio map to get the initial change detection result. Usingthe spatial correlation optimize the initial change detection result, the change detectionresult is got. The experimental results have demonstrated the effectiveness of theproposed method.(3) A novel method for change detection in SAR image by Global Contrast basedSalient Region Detection (GCSRD) is proposed. This method improve the correlationwhich used in the salient region detection based histogram to get a new method tocalculate the salience. Calculate the salient map on the mean ratio image of the twooriginal SAR images with the new salience. Finally, the classic spectral clustering method is applied to the new salient map, and gets the change detection result. Theexperimental results have demonstrated the effectiveness of the proposed method.This work was supported by the National Natural Science Foundation of China (No.61272282), the Program for New Century Excellent Talents in University (No.NCET-13-0948), and the Fundamental Research Funds for the Central Universities(grant K50511020011).