Ship image dehazing method based on improved cyclic adversarial generative network

Foggy weather significantly degrades ship visibility and image quality, posing serious risks to navigation safety. Enhancing the dehazing performance of ship navigation images is therefore of great importance. To address the insufficient fog removal and poor detail restoration of existing dehazing methods in maritime scenarios, this study proposes an end-to-end ship image dehazing method that integrates an improved CycleGAN with attention mechanisms. A Squeeze-and-Excitation (SE) channel-attention module is introduced to aggregate feature maps, compress spatial information, and strengthen the network's ability to learn global representations. Multi-scale channel fusion is achieved through skip connections, which not only reduces computational complexity but also enables the model to better capture fog characteristics under complex atmospheric conditions and to process ship targets of different sizes. Furthermore, a Channel Attention module is incorporated to enhance feature selection and improve the restoration of ship contours and fine structural details. Quantitative evaluations and real fog-navigation experiments confirm the robustness of the proposed method, demonstrating consistent improvements over existing dehazing approaches across all tested metrics and navigation scenarios.