Vision Language Foundation Models based on CLIP architecture for remote sensing primarily rely on short text captions, which often result in incomplete semantic representations. Although longer captions convey richer information, existing models struggle to process them effectively because of limited textencoding capacity, and there remains a shortage of resources that align remote sensing images with both short text and long text captions. To address this gap, we introduce DGTRSD, a dual-granularity remote sensing image-text dataset, where each image is paired with both a short text caption and a long text description, providing a solid foundation for dual-granularity semantic modeling. Based on this, we further propose DGTRS-CLIP, a dual-granularity curriculum learning framework that combines short text and long text supervision to achieve dual-granularity semantic alignment. Extensive experiments on four typical zero-shot tasks: long text cross-modal retrieval, short text cross-modal retrieval, image classification, and semantic localization demonstrate that DGTRS-CLIP consistently outperforms existing methods across all tasks. The code has been open-sourced and is available at https://github.com/MitsuiChen14/DGTRS.
12 December 2025 in IEEE Transactions on Geoscience and Remote Sensing
In conventional Unsupervised domain adaptation (UDA), model knowledge is transferred to the target domain with access to annotated source data, which is an unsuitable strategy for cross-domain scenarios involving data privacy and confidentiality. In this paper, we focus on source-free domain adaptation (SFDA) for semantic segmentation tasks, which adapts source-trained models to unannotated target domains without relying on source data. Self-training paradigms dominate the existing approaches, but often suffer from significant performance degradation due to unreliable pseudolabels and knowledge forgetting. To address these challenges, we propose innovative self-supervised spatial–temporal consistency learning for source-free domain adaptive segmentation (S3T-SFDA). Specifically, to address pseudolabel ambiguity in various spatial contexts, a spatial multiview consistency (SMVC) mechanism is proposed to constrain the semantic consistency across different spatial views of the same image. To mitigate the knowledge forgetting problem caused by a lack of supervisory information, a temporal dynamic consistency (TDC) mechanism, which harnesses historical knowledge consistency to regularize the current model evolution direction, is proposed. Furthermore, to improve the category-discriminative representation capabilities under domain shifts, a spatial-temporal contrastive (STC) strategy is designed to promote the intrinsic semantic association of features belonging to different categories in the target domain. Extensive experiments conducted on two domain-adaptive remote sensing (RS) segmentation benchmarks. The results demonstrate the flexibility of the proposed method when integrated into various advanced segmentation architectures, as well as its excellent generalization performance across different cross-domain RS scenarios.
Semantic segmentation techniques for remote sensing images (RSIs) have been widely developed and applied. However, most segmentation methods depend on sufficiently annotated data for specific scenarios. When a large change occurs in the target scenes, model performance drops significantly. Therefore, unsupervised domain adaptation (UDA) for semantic segmentation is proposed to alleviate the reliance on expensive per-pixel densely labeled data. In this paper, two key issues of existing domain adaptive (DA) methods are considered: (1) the factors that cause data distribution shifts in RSIs may be complex and diverse, and existing DA approaches cannot adaptively optimize for different domain discrepancy scenarios; (2) domain-invariant feature alignment, based on adversarial training (AT), is prone to excessive feature perturbation, leading to over robust models. To address these issues, we propose an AdvCDA method that guides the model to adapt adversarial perturbation consistency. We combine consistency regularization to consider interdomain feature alignment as perturbation information in the feature space, and thus propose a joint AT and self-training (ST) DA method to further promote the generalization performance of the model. Additionally, we propose a confidence estimation mechanism that determines network stream training weights so that the model can adaptively adjust the optimization direction. Extensive experiments have been conducted on Potsdam, Vaihingen, and LoveDA remote sensing datasets, and the results demonstrate that the proposed method can significantly improve the UDA performance in various cross-domain scenarios.
23 October 2026 in ISPRS Journal of Photogrammetry and Remote Sensing