SAR Image Change Detection Based On Threshold Segmentation And Fuzzy Clustering

Synthetic Aperture Radar(SAR)is a radar sensor that produces high-resolution images in a variety of complex climates.SAR has the advantage of being unaffected by weather,light and time,so SAR images are playing an increasingly important role in military and civilian applications.SAR image change detection is a method to research the change information generated at the same place and at different times.As one of the key technologies for SAR image analysis,it has great value in various fields.However,due to some of the multiplicative speckle noise inherent in SAR images.These noises often affect the results of change detection,increasing the difficulty of change detection.This thesis mainly researches the problem of low accuracy caused by noise pollution,poor robustness to noise and timeliness of the change detection process in the process of SAR image change detection.The main contents are as follows:A SAR image change detection method based on iterative logarithmic mean threshold is proposed.First,a neighborhood log-ratio method with weights is used to generate the difference map,and the change region and the unchanging region are divided for the difference map.Then,the weight matrix when constructing the difference map is modified by the classification information of one pixel and its neighboring pixels.The adjusted weight matrix will use the neighborhood information of the pixel point in a targeted manner,so that the difference map effect is gradually improved.A better difference map will produce a better detection map,iteratively,and use a better change detection map to help adjust the weight matrix.We believe that the best change detection map is when the classification of each pixel is consistent with the classification of its neighborhood.Therefore,the result with the least number of inconsistencies between the classification of each pixel and the classification of its neighborhood is selected as the result map of the change detection.This method constructs a better difference map by iterating,which improves the robustness to noise.A SAR image change detection method based on the log-mean threshold of key point selection is proposed.First,the neighborhood log-ratio method and the log-ratio method are used to generate the difference map,and the two difference maps are combined by a combination strategy.So that the newly generated difference map can retain good edge information,and can reduce the sensitivity to noise.At the same time,the newly generated difference map does not change the mean information in the original image,which is beneficial to the final threshold segmentation.Secondly,the key points are extracted from the newly generated difference map by the method of scale-invariant feature transformation.And select candidate regions by key points.That can further narrow the selection range of the change region,improve the detection accuracy,and reduce the interference caused by noise.Finally,a log-mean threshold is used in the selected candidate regions to divide the changed regions and the unchanged regions.The method divides the key areas by key points,reduces the amount of data in the process of change detection,and uses two methods in the process of constructing the difference map.Therefore,more accurate results can be obtained in a shorter time and are not sensitive to noise.A fuzzy C-means SAR image change detection method based on Mean Shift preclassification is proposed.Firstly,the original image is pre-classified based on Mean Shift clustering.Mean Shift is a clustering method with non-parametric density estimation,which can effectively maintain the edge information of the object and also smooth the pixel intensity of the same type of object.To reduce the impact of noise on change detection.Then,the difference graph is generated by the log-ratio operator and classified into a changed region,an uncertain region,and an unchanged region.The image information of the unchanged region is adjusted to reduce the influence of different noise levels.After the adjustment,the Mean Shift is used to pre-classify and generate the difference map.Finally,the improved FCM algorithm is used to classify the difference map to generate a change detection result map.The improved FCM algorithm make use of spatial information and intensity information in the image to further improve the robustness of the method to noise.This method uses Mean Shift pre-classification to solve the effect of noise on the change detection results when the two images have different noise levels.