Remote sensing image road extraction method based on multi-scale channel attention improvement

The invention discloses a remote sensing image road extraction method based on multi-scale channel attention improvement. The method comprises the following steps: preparing a data set; building a remote sensing image road extraction network; and training the remote sensing image road extraction network. According to the method, ResNeSt-50 is used as an encoder of a remote sensing image road extraction network, semantic information of images is extracted in a cross-channel mode, the diversity of feature maps is guaranteed, and therefore the road extraction network can better understand and capture road features in the remote sensing images. According to the method, the cavity convolution module is introduced into the central part of the road extraction network, so that the road extraction network has multi-scale sensing capability, the adaptability of the road extraction network to road pairs with different sizes is improved, and more accurate road and non-road region division is realized. According to the me