Sub-Pixel Width Road Network Extraction Using Sentinel-2 Imagery

Nowadays, road maps play a key role in our society. Therefore, keeping those maps up-to-date is highly important. The extraction of road networks from satellite imagery is a complex problem, not only because of occlusions, shadows produced by non-road objects, but also due to the limited spatial resolution of the imagery used. The feasibility to detect a road depends on its width, which can reach sub-pixel size in some satellite products. In the last decade, many attempts have been carried out to automatize this labour. However, the vast majority of methods rely on aerial imagery, whose costs are not yet affordable for maintaining up-to-date maps. This work demonstrates that it is also possible to accurately detect roads using freely available Sentinel-2 imagery, regardless of their width. For that purpose, a new deep learning architecture which combines semantic segmentation and super-resolution techniques is proposed. As a result, fine-grained road network maps at 2.5 m are generated from 10 m imagery taken as input. To evaluate this proposal a data-set composed of 20 cities spread across the Spanish territory is used.