A Study Of SAR Image Change Detection Based On Adaptive Weight Image Fusion And Clustering

SAR(Synthetic Aperture Radar, SAR) image change detection is to study the changes between two or more SAR images that taken from the same region in different time, and it can be applied in many aspects, such as society, economy, military and so on. In recent years, SAR image has increasingly become an important data source for change detection, because of its all-weather working ability and advantages of penetration and imaging large coverage area. In this thesis, we mainly focus on the generation of the difference image and analysis of the difference image and the other aspects. The author’s major contributions of the research are outlined as follows:1. A method of SAR image change detection based on adaptive weight for image fusion is proposed. A logarithmic ratio difference map and a direct ratio difference map are combined using the adaptive weights to form the fused image. Then, the k-means clustering is used to cluster the fused image into two clusters to obtain the final change detection result. The weights in this method can be adaptively adjusted according to different characteristics of difference map. In addition, the neighborhood gray information and spatial information is considered, which helps to improve the anti-noise performance and change detection accuracy.2. A means of SAR image change detection based on adaptive weight for image fusion and PCA feature extraction is proposed. According to adaptive weights, a direct ratio difference map and logarithmic ratio difference map are used to form the fused image. Then, we use PCA feature extraction method in the fused image to obtain feature vector according to each pixel, and constitute feature space matrix. We apply k-means clustering to the feature space matrix extracted with the method of PCA, the original data is used for feature extraction to obtain feature space matrix, then we utilize clustering method to cluster the image into two clusters. In this method, the nonlinear mapping can better distinguish and amplify the useful feature to achieve more accurate cluster. The error rate of change detection is reduced. Noise immunity is enhanced, the total number of errors is reduced and the accuracy of change detection is improved.3. A way of SAR image change detection based on fuzzy clustering with the method of Markov random field is proposed. The Markov random field model is applied to the fuzzy c-means clustering. This method improves the fitting accuracy of the proposed least square method of energy function in Markov random field and also improves the objective function of the fuzzy clustering. The regularized KL(Kullback-Leibler) information is added to the original objective function, making the clustering more accurate. Accordingly, the accuracy of the change detection is increased and the anti-noise capability is improved.This work was supported by the National Natural Science Foundation of China(No. 61003199) and the Fundamental Research Funds for the Central Universities(Nos. JB140216 and K5051202019).