A novel edge-weight based fuzzy clustering method for change detection in SAR images.

Abstract Change detection has found wide application in several fields, and in this paper we put forward a novel change-detection approach in synthetic aperture radar (SAR) images. The approach is implemented to the difference image (DI) through the modification of conventional fuzzy c-means (FCM) clustering method. In order to reduce the impact of speckle noise, the objective function is modified by introducing piecewise prior, which serves as the use of local spatial information. The approach mainly includes an edge pre-estimation step and an objective function optimization step. In the first step, the areas containing the edges in the DI is detected by the level set method, in which an energy functional is established to find out the final level set function. Then a weight which serves as a smooth parameter in the second step is output according to the computed level set function. In the second step, the objective function is optimized by the modified accelerated proximal gradient (APG) algorithm, in which the Lagrange multiplier method is applied to determine some other unknown variables. Our contribution lies in two aspects. Firstly, the approach is capable of reducing the impact of speckle noise in the homogeneous region and preserving blurred edges due to the edge pre-estimation step along with its output weight. Secondly, the approach converges in a fast speed because of the use of the APG algorithm that super-linearly converges. Theoretical analysis and experimental results on simulated and real SAR datasets show that the proposed approach is able to detect the real changes by reaching a trade-off between noise reduction and edge preservation. The results also demonstrate its fast convergence speed. [ABSTRACT FROM AUTHOR]