DDCAttNet: Road Segmentation Network for Remote Sensing Images

Semantic segmentation of remote sensing images based on deep convolutional neural networks has proven its effectiveness. However, due to the complexity of remote sensing images, deep convolutional neural networks have difficulties in segmenting objects with weak appearance coherences even though they can represent local features of object effectively. The road networks segmentation of remote sensing images faces two major problems: high inter-individual similarity and ubiquitous occlusion. In order to address these issues, this paper develops a novel method to extract roads from complex remote sensing images. We designed a Dual Dense Connected Attention network (DDCAttNet) that establishes long-range dependencies between road features. The architecture of the network is designed to incorporate both spatial attention and channel attention information into semantic segmentation for accurate road segmentation. Experimental results on the benchmark dataset demonstrate the superiority of our proposed approach both in quantitative and qualitative evaluation.