Direction-aware structural and semantic alignment for SAR-optical image matching

Cross-modal image matching between synthetic aperture radar (SAR) and optical images is an essential task for applications such as change detection and scene understanding. However, it remains challenging due to significant structural distortions and semantic inconsistencies arising from different imaging mechanisms and sensor characteristics. To address this problem, we propose the Direction-Aware Interaction Learning Network (DAIL), which combines orientation-sensitive structural cues with cross-modal semantic interaction. Specifically, the proposed network comprises two modules: (1) the Orientation-Aware Dynamic Channel Fusion (OADCF) module, which adaptively enhances directional gradient features to provide robust structural representations; and (2) the Cross-Modal Interaction Attention Module (CMIAM) that performs bidirectional feature interaction to improve semantic consistency. Experimental results on the SEN1-2 and OSdataset as well as the newer OSDataset2.0 benchmark demonstrate that DAIL achieves competitive matching performance under various evaluation thresholds, demonstrating its effectiveness for robust SAR-optical image matching.