Feature-knowledge-guided and direction-adaptive recognition of cross-sea bridges from high-spatial-resolution remote sensing imagery

Cross-sea bridges are critical components of coastal transportation networks, and their accurate recognition is essential for infrastructure mapping, geographic information updating, and emergency assessment. However, reliable recognition from high-spatial-resolution remote sensing imagery remains challenging because of complex coastal backgrounds and the large variation in bridge orientations. This study aims to improve the robustness and directional adaptability of cross-sea bridge recognition while retaining the interpretability and low sample dependence of knowledge-driven approaches. An integrated framework is developed by combining image enhancement, land–sea spatial constraints, bridge-specific feature knowledge, and direction-adaptive mathematical morphology. Water bodies are first extracted to constrain the recognition space, after which local water-connectivity directions are estimated to guide adaptive morphological reconstruction. Candidate bridge regions are then identified and screened using spatial and geometric knowledge. Experiments using GF-2 imagery over the Zhoushan Archipelago show that all four cross-sea bridges with different orientations were successfully identified, with good positional consistency and a maximum bridge-length deviation of 1.54%. The results demonstrate that integrating environmental context, feature knowledge, and directional adaptation provides an interpretable and sample-efficient solution for cross-sea bridge recognition, with potential applications in coastal infrastructure mapping, geographic information updating, and rapid bridge identification.