Automatic Road Extraction from Very High Resolution Orthophoto Using DeepLab V3+

Road extraction, one of the processes in map-making, is widely used by various services such as intelligent transportation systems, disaster navigation and urban planning. So far, road extraction is done manually, which takes a long time, costs a lot, and needs to be carried out by a team of experts. Automated semantic segmentation can speed up the road extraction process. The author proposes the application of Deeplab V3+ model for road extraction from very high resolution orthophoto with the Indonesian study area. From the study, the model achieved mean Intersection Ratio Union value 88% and Mean Dice loss 6.8%.