IEGC-FSDN: Foggy Ship Detection Network With Integrating Edge and Global Constraints
Abstract
With advances in aerial technology, remote-sensing object interpretation has found ubiquitous applications across diverse domains. Owing to atmospheric interference, haze severely degrades the quality of optical remote sensing images—an effect that is particularly pronounced over water, where evaporative moisture frequently produces dense fog in harbors and open-sea scenes. Consequently, ship detection under hazy conditions has become an extremely challenging task. To address this issue, we propose a Foggy Ship Detection Network with Integrating Edge and Global Constraints (IEGC-FSDN). The network introduces a Progressive Restoration Strategy that simultaneously leverages edge and global constraints to progressively recover haze-corrupted features within the detector, thereby enhancing the model’s perceptual capacity for ships in fog. In addition, a Multi-Scale Atmospheric Prior Module (MSAPM) is embedded to explicitly incorporate atmospheric physical priors, further strengthening robustness. The experimental results demonstrate that IEGC-FSDN achieves superior performance compared to state-of-the-art alternatives, with only a marginal increase in parameters over the baseline.