Strip Attention Networks for Road Extraction

In recent years, deep learning methods have been widely used for road extraction in remote sensing images. However, the existing semantic segmentation networks generally show poor continuity due to high-class similarity between roads and buildings surrounding images, existence of shadows occlusion. To deal with this problem, paper proposes strip attention (SANet) extracting Firstly, a module (SAM) is designed extract contextual information spatial position roads. Secondly, channel fusion (CAF) fuse low-level features high-level features. The network trained tested using CITY-OSM dataset, DeepGlobe CHN6-CUG dataset. test results indicate that SANet exhibits excellent performance can better solve problem compared other networks.