Road Detection in High-resolution SAR Images with Improved Multiple Feature Fusion

In this paper, we propose a novel method for road region detection in high-resolution SAR images based on the fusion of multiple features. Compared with traditional SAR road detection methods with feature fusion, we exploit more useful features such as the standard deviation of directional radiance for distinguishing between roads and buildings or flatland. Then, the features are binarized with dynamic thresholds related to the cumulative possibility distribution of features. Finally, we define a membership parameter to fuse the binarized features and select the road candidate regions according to their geometric features, thereby ensuring better detection rate and lower false alarm rate. Experimental results of GF-3 SAR images show the effectiveness of the proposed method in the detection of both urban and suburban road regions.