Style-Shift-Aware Category Contrastive Learning for Domain Generalization Semantic Segmentation of Optical Remote Sensing Images

Domain generalization semantic segmentation of optical remote sensing images aims to train a model on one or multiple source domains such that it can generalize to unseen target domains without accessing target-domain data during training. Existing methods are mainly divided into style randomization and feature normalization. These methods encourage the model to focus more on domain-invariant information by reducing the impact of style information on the model training. However, owing to the entanglement between style information and domain-invariant information, style randomization tends to partially change domain-invariant information, while feature normalization inevitably leads to the partial loss of domain-invariant information. To address these limitations, we propose style-shift-aware category contrastive learning, which can further explore the domain-invariant information embedded in the source domain. Then, the domain-invariant information is leveraged to compensate for the regions impacted by style randomization. Furthermore, to enrich the style diversity of the auxiliary domain, we introduce a dual-branch augmentation strategy. Strong augmentations first simulate variations caused by different image times, followed by a dual-branch design that simulates resolution-related differences caused by different image equipments: RandConv simulates resolution degradation, while an edge-texture augmentation module simulates resolution enhancement. Extensive experiments show that our method consistently outperforms state-of-the-art approaches. It achieves at least 6.22/5.03 mIoU gains on WHU dataset and 3.48/5.78 mIoU gains on OM dataset under ResNet-50/101, attaining the highest average performance across subsets. Our code could be available at https://github.com/yuan3ee/SSAC.