Cross-Domain Cooperative Fusion Network for Few-Shot Remote Sensing Imagery Segmentation

Abstract

Few-shot semantic segmentation (FSSS) in remote sensing faces challenges due to limited annotated samples, significant domain shifts, and complex interclass variations. In this article, a cross-domain cooperative fusion network (CDCFNet) is proposed for few-shot remote sensing imagery segmentation across heterogeneous domains. The framework incorporates cross-domain cooperative attention that performs joint channel-spatial attention to achieve dynamic feature alignment between source and target domains. The integration of positional and global–local cooperative attention strengthens domain invariance and enhances the robustness of target recognition. Furthermore, a gated adaptive feature-aligned pyramid transformation (GAF-PTM) is designed to decouple domain-specific and transferable components. This module further employs hierarchical feature weighting and prototype-guided sparse alignment to mitigate overfitting under limited-sample conditions. Comprehensive experiments conducted on multiple cross-domain remote sensing imagery segmentation tasks demonstrate that the proposed method consistently outperforms existing few-shot segmentation baselines in both one-shot and five-shot settings, achieving high accuracy and strong generalization across heterogeneous domains. The experimental results verify the effectiveness and adaptability of the proposed framework.