Abstract: With the wide spread application of high-resolution satellite image remote sensing technology, it also plays a huge driving role in transportation. There are technical difficulties in extracting rural road networks, such as data acquisition difficulties, inconsistent road features, natural environment interference, and frequent road changes. This study mainly uses high-resolution satellites as the main data source, and based on deep learning algorithms, learns the spectral, texture, and separable feature sets of different types of rural roads to identify rural roads with low grades and poor road conditions, which provides accurate data guarantee for the identification and discrimination of rural road networks, timely obtaining objective, accurate, and comprehensive basic data of rural roads solves the problem of insufficient road network information in complex and difficult areas. This study can serve scientific decision-making basis for the planning and construction of road networks in complex and difficult areas, and facilitates travel and traffic guidance.

Key words: remote sensing data, road network extracting, deep learning