A Graph-Cut-Based Method for Road Labels Making With OSM Data

Aiming at producing the road labels for deep neural networks (DNNs), this letter proposes a graph-cut-based method to make road annotations on very high-resolution (VHR) remote-sensing images. With the aid of OpenStreetMap (OSM), a superpixel method and the graph cut method are employed for road segmentation. After that, the road areas are refined by the OSM. In this process, the road annotations are made automatically. In experiments, two traditional methods, two deep learning methods, and the proposed method are utilized to segment the roads on two types of satellite images in Tianjin port area. The results show that the proposed method creates more accurate and integrated road labels compared with other methods.