Refinement and global connection method applied to remote sensing image road network extraction

The invention discloses a refinement and global connection method applied to remote sensing image road network extraction, which comprises the following steps: obtaining a road direction feature map from a road binary map extracted from a remote sensing image, and decomposing the road binary map into a plurality of road binary map layers according to direction features; carrying out data point fitting on each road binary image layer to obtain a refined road curve, extracting end points and end point tangential direction vectors, and then carrying out connection according to spatial Euclidean distances between the end points and included angle cosine distances between the end point connection directions and the tangential directions; and superposing all the refined and connected binary image layers to obtain a finally extracted road binary image. According to the method, road binary image refinement is carried out according to direction feature grouping, a road curve is smooth and freeof burrs, it can be guara