Comparative analysis of change detection methods in Polarimetric SAR images based on probability statistical models

Abstract

Change detection method in polarimetric SAR images considers polarization information and features. In order to make better use of polarization information and features for change detection, this paper compared and analyzed different change detection methods in polarimetric SAR image based on probability statistical method. Firstly, based on RADARSAT-2 polarimetric SAR data, four change detection methods were used to generate four change degree maps. Then, through four probability statistical model, we built the fitting models for the four kinds of change degree maps, and obtained 16 kinds of change detection results. The results showed that the 16 combination of change detection methods made full use of more ground information which contains in different polarization modes. Compared with other distribution fitting, the lognormal fitting distribution function can better fit the four change degree maps, and the change detection results are better. The overall accuracy (OA) can be improved by more than 1.53%, and the kappa coefficient can be improved by no less than 0.026. In addition, compared with 16 combined change detection methods, the Wishart-lognormal distribution change detection results are the best. The overall accuracy is 87.76% and kappa coefficient is 0.7469, which realizes the high-precision change detection of the study areas.