Rural road extraction from remote sensing images based on multi-view contextual information and multi-stage features

Accurate extraction of rural roads from high-resolution optical remote sensing images is of great significance to the development of rural areas, road navigation, rural land resource planning and other applications. Different from urban roads, rural ones with complex terrain backgrounds are often slender and winding, thereby making them more susceptible to vegetation cover. In order to improve the reliability and accuracy of rural road extraction, a Complex Rural Road Extraction Network (CRRENet) is proposed in this work, which consists of five parts: feature encoder, Multi-view Contextual Information Extraction Module (MCIEM), Multi-stage Feature Fusion Module (MFFM), Channel Coordinate Attention Mechanism (CCAM) and feature decoder. The MCIEM extracts the multi-view contextual information by the parallel dilated convolution with different dilation rates. To avoid the loss of image details, the MFFM integrates different feature maps from the downsampling stages. By adjusting the weights of feature maps, the CCAM enables the network to self-adaptively suppress the background noise and focus on the road foreground. Ablation and comparison validate CRRENet's superiority.