MADnet: a Multiple Attention Decoder Network for Segmentation of Remote Sensing Images

Segmentation of high-resolution remote sensing images is a challenging task. Some recent studies have used complex convolutional neural networks to address this problem, and they can only extract multi-scale features and improve object boundary discrimination. However, these neural networks are difficult to train due to high complexity. In this paper, we propose two new attention modules, Local Channel Spatial Attention and 16-piece Local Channel Spatial Attention, which can focus on improving the feature representation of each local area. In order to effectively utilize multi-scale features, we also propose a Twice Decoder Module, which is integrated into our model to improve the discrimination ability of target boundary. Our proposed method is verified through extensive experiments and experimental results show that our model can better improve the overall accuracy compared with other state-of-the-art methods.