关键词: CNN, 深度学习, 道路提取, 遥感解译

Abstract: Road extraction comprehensively consider the accuracy, computing ability and adaptability to the environment of Guiyang in the interpretation process,so several links in the deep learning neural network model are decomposed. Through multiple rounds of comparative experiments and analysis,a model for automatic extraction of remote sensing images of road elements in Guiyang is established in this paper,the data extracted in batches are analyzed and optimized to complete the filling of some road attributes, which largely realizes the automatic intelligent and efficient extraction of road entities.The practical problems and technical routes involved in the process can provide reference for the natural resources business carried out by the municipal and county level satellite remote sensing application technology departments.

Key words: CNN, deep learning, road feature extraction, remote sensing interpretation