SWCARE: Switchable learning and connectivity-aware refinement method for multi-city and diverse-scenario road mapping using remote sensing images

Lixian Zhang, Shuai Yuan, Runmin Dong, Juepeng Zheng, Bin Gan, Dengmao Fang, Yang Liu, Haohuan Fu

Abstract: Highlights•A connectivity-aware model is proposed to improve connectivity in road extraction.•Four types of auxiliary supervision are utilized to improve extraction performance.•We build a diverse-scenario dataset to improve the robustness in the real world.•The proposed method is proved effective on multi-city and complex road datasets.