Change Detection In SAR Images Based On Denoising And Elite Genetic Fuzzy Clustering

Change detection in images is to analysis two images of the same area obtained at different time, and then get the change information. Now, Synthetic Aperture Radar(SAR) shows growing importance in many areas. SAR images have the advantage of high resolution, are available at all weather conditions and at any time, and can also be conveniently obtaint at different time in the same area. Therefore the change detection in SAR images has played a critical role in the change detection in remote sensing images. When applying the change detection technology in SAR images, images of the same area at different time will be firstly compared to obtain the difference image, then the SAR images will be classified into changed area and unchanged area using the gray values of the difference image. The change detection of SAR images has been widely used in environmental monitoring, disaster estimation, deforestation monitoring, crop growth status monitoring and so on. In this thesis, our work mainly focuses on sovling problems of the existing change detection in SAR images, such as the time-consuming drawback and the low accuracy, the main works are as follow:1. Change detection in SAR images based on histogram and Elite Genetic Clustering Algorithm is proposed here. First, the proposed method uses the operation of the histogram to reduce the number of samples of the images, so that the processing time of image could be effectively reduced. Then, the fuzzy c-means(FCM) clustering algorithm is used to initial the population of genetic algorithm and to calculate the fitness values. In the evolutionary process, an elitist strategy is used to find the optimal which is set as the initial clustering center of FCM. Therefore, the proposed method could combine the local searching ability of FCM with the global searching ability of genetic algorithm, and consequently promote the convergence speed and the accuracy of the algorithm.2. Change detection in SAR image based on Mean Shift and Elite Genetic Clustering Algorithm is proposed. The main problems in SAR image processing is denoiseing. To solve this problem, we use the Mean Shift to eliminate the noise in difference images, and thus effectively reduce the impact of noise in the change detection results. Experiments are taken on real SAR images to test the proposed method, and comparisons are made with several classical algorithms. The experiment results havefully demonstrated the validity of the proposed method in SAR image change detection.3. Change detection in SAR image based on the improved non-local means is proposed. Since the Gauss kernel weighted Euclidean distance in non-local means is not the most effective in the extraction of image features and the robustness to speckle noise, and meanwhile, the edge information of an image could not be handled well by utilizing an exponential function to calculate weights. Thus, the improved non-local means is proposed by introducing the Euclidean distance of Fourier kernel weighted, the similarity measurement method based on the logarithmic ratio and the weight of twodimensional gaussian function. Finally, the performance of the proposed method is tested in SAR image change detection experiments, and the result has proven the well performance of the proposed method.