SAR Image Change Detection Based On Fuzzy Theory

Because synthetic aperture radar(SAR)is not affected by light conditions and clouds,SAR images are widely used in our lives.The SAR image change detection refers to the comparative analysis of SAR images covering the same area at different times,and the change information of the ground objects is acquired according to the differences between the images.With the development of satellite technology and SAR sensor technology,SAR image change detection has played an irreplaceable role in many fields such as national production and life and military defense construction.This paper considers the fuzziness of the SAR image and the uncertainty in the process of change detection.So I applied fuzzy theory to the change detection of SAR image and completed the following work :(1)When constructing the difference diagram using the ratio method,the number of pixels of “new class” or “disappearance class” may be “submerged” at the end of the “unchanged” pixel statistics histogram,resulting in missed detection.This paper presents a method of constructing difference diagram with improved logarithm ratio operators.This method merges "new class" and "disappearance class" into "change class",so that the difference map contains only "change class" and "unchanged class".The improved method can increase the number of pixels in the change area,reduce the missed detection rate,and then improve the detection accuracy.(2)After constructing the difference diagram,it is necessary to split it to obtain the change detection result map.In the traditional MRF segmentation method,each pixel is assigned a hard label in the iterative process,which results in loss of information and accumulation of errors.This paper proposes a method of SAR image change detection based on fuzzy MRF.This method uses the fuzzy membership function to describe the label of the pixel,which can reduce the impact of the hard decision caused by the MRF method in the iterative process.The experimental results show that the fuzzy MRF algorithm can effectively remove the "pseudo-change" of dark features such as rivers and roads.It can improve the accuracy of change detection.(3)This paper proposes a multi-feature fuzzy fusion SAR image change detection method.At first,we calculate the structural similarities of gray co-occurrence matrix texture feature,texture feature based on Gabor filter and texture feature based on statistical characteristics.Then we use Sigmoid membership function to describe the uncertainty near the structural similarity threshold.Finally,the change region is obtained by fuzzy fusion.The experimental results show that the accuracy of the detection results after fuzzy fusion is higher than that of the single feature detection,and the detection results after the fusion of Gabor features and statistical texture features are the best.