Abstract: Focused on scientific goals of automatic extraction of low-grade road information from unmanned aerials vehicle (UAV) images, the paper proposes a method based on generative adversarial networks (GANs). First, GANs are built based on TensorFlow platform; second, the UAV images are preprocessed and the road information is tagged in such images; third, the road feature of UAV images is trained by GANs built in the research. Finally, the low-grade road information from UAV images can be extracted automatically. Comparison and verification show that the proposed method provide a reference for extraction of road information.