Wide-Area Land Cover Mapping with Sentinel-1 Imagery using Deep Learning Semantic Segmentation Models

Land cover mapping is essential to monitoring the environment and understanding the effects of human activities on it. The automatic approaches to land cover mapping (i.e., image segmentation) mostly used traditional machine learning that requires heuristic feature design. On natural images, deep learning has outperformed traditional machine learning approaches for image segmentation. On remote sensing images, recent studies demonstrate successful applications of specific deep learning models to small-scale land cover mapping tasks (e.g., to classify wetland complexes). However, it is not readily clear which of the existing models are the best candidates for which remote sensing task. In this study, we answer that question for mapping the fundamental land cover classes using satellite radar data. We took Sentinel-1 C-band SAR images available at no cost to users as representative data. CORINE land cover map was used as a reference, and the models were trained to distinguish between the 5 major CORINE classes. We selected seven among the state-of-the-art semantic segmentation models so that they cover a diverse set of approaches: U-Net, DeepLabV3+, PSPNet, BiSeNet, SegNet, FC-DenseNet, and FRRN-B. The models were pre-trained on the ImageNet dataset and further fine-tuned in this study. All the models demonstrated solid performance with overall accuracy between 87.9% and 93.1%, and with good to a very good agreement (kappa statistic between 0.75 and 0.86). The two best models were FC-DenseNet and SegNet, with the latter having a much smaller inference time. Overall, our results indicate that the semantic segmentation models are suitable for efficient wide-area mapping using satellite SAR imagery and also provide baseline accuracy against which the newly proposed models should be evaluated.