IDANet: Iterative D-LinkNets with Attention for Road Extraction from High-Resolution Satellite Imagery

Road information plays a fundamental role in many application fields, while satellite images are able to capture a large area of the ground with high resolution. Therefore, extracting roads has become a hot research topic in the field of remote sensing. In this paper, we propose a novel semantic segmentation model, named IDANet, which adopts iterative D-LinkNets with attention modules for road extraction from high-resolution satellite images. Our road extraction model is built on D-LinkNet, an effective network which adopts encoder-decoder structure, dilated convolution, and pretrained encoder for road extraction task. The attention mechanism can be used to achieve a better fusion of features from different levels. To this end, a modified D-LinkNet with attention is proposed for more effective feature extraction. With this network as the basic refinement module, we further adopt an iterative architecture to maximize the network performance, where the output of the previous network serves as the input of the next network to refine the road segmentation and obtain enhanced results. The evaluation demonstrates the superior performance of our proposed model. Specifically, the performance of our model exceeds the original D-LinkNet by 2.2% of the IoU on the testing dataset of DeepGlobe for road extraction.