Logistic Regression for Enhancing FOPEN SAR Change Detection Performance

The SAR change detection processing sequence includes SAR image stack formation (reference and surveillance), detection, and classification. The currently used approaches for classification in FOPEN SAR change detection are based on morphological operations such as erosion and dilation. In this paper, logistic regression, a supervised machine learning algorithm, is proposed for classification in FOPEN SAR change detection. The proposal is tested on the experimental data collected by the CARABAS system in 2002 over a dense forest. The test results indicate that logistic regression enhances the FOPEN SAR change detection performance. Specifically, the detection probability is 95% and the false alarm rate is 0.51 per square kilometer.