A Multiobjective Fuzzy Clustering Method For Change Detection In SAR Images

In recent years, Synthetic Aperture Radar(SAR) technology has got a rapid development, the spaceborne SAR systems have observed the earth for years, and have acquired a plenty of multi- temporal ground observation data. Most of studies about remote sensing have tried to develop the techniques which can make good use of the information obtained by the SAR systems, including object extraction, object classification, border detection, interferometry, change detection, etc. In particular, among these researches, the research on change detection technology is the hottest one. Image change detection is based on the comparative analysis of two images obtained from the same region at two different times, in order to find the change region between them. It has been broadly applied on a lot of domains, such as remote sensing monitoring, medical diagnosis, and video monitoring.In this paper, based on the study of change detection approaches for SAR images, aiming at the existing problems that universally exist in the SAR image change detection methods, we propose a novel change detection approach. It mainly makes contributions in the following two aspects:1. On account of the existence of image noise, the trade-off between details preservation and noise removing is crucial for change detection work in Synthetic Aperture Radar images. It understands change detection problem from the perspective of the multiobjective optimization problem for the first time. And it constructs two objective functions for multiobjective optimization from the view of details preservation and noise removing respectively, and ultimately transforms the change detection work into a multiobjective optimization problem(MOP).2. Based on the transformation from the SAR image change detection work into a multiobjective optimization problem, an incorporation of MOEA/D and the update of membership degree matrix is put forward so as to improve effects further. And Lagrange multiplier method is selected to calculate each subproblem that is obtained by decomposing multiobjective optimization problem. Therefore, update formula of membership degree matrix is obtained by solving some necessary conditions forminimizing this cost function. 3. Apply the aforementioned improvements to some datasets of real SAR images, compare them with several other existing approaches for change detection, and analyze experimental results in detail. As a result, these improvements in this paper are confirmed practical and effective.