基于循环生成对抗网络的海事图像去雾方法  

Marine Image Dehazing Based on Cycle Generative Adversarial Network

作　　者：肖昌云 王梓创 刘虹余 韩冰[2] 陈信强 XIAO Changyun;WANG Zichuang;LIU Hongyu;HAN Bing;CHEN Xinqiang(COSCO SHIPPING Bulk Co.,Ltd.,Guangzhou 510335,China;Shanghai Ship and Shipping Research Institute Co.,Ltd.,Shanghai 200135,China;Institute of Logistics Science and Engineering,Shanghai Maritime University,Shanghai 201306,China)

机构地区：[1]中远海运散货运输有限公司,广州510335 [2]上海船舶运输科学研究所有限公司,上海200135 [3]上海海事大学物流科学与工程研究院,上海201306

出　　处：《上海船舶运输科学研究所学报》2024年第6期14-19,共6页Journal of Shanghai Ship and Shipping Research Institute

基　　金：上海市“科技创新行动计划”优秀学术/技术带头人项目(22XD1431000)。

摘　　要：为提升船舶在雾天等能见度不良条件下对周围环境的感知能力,提出一种基于循环生成对抗网络的海事图像去雾方法,并引入循环一致性和恒等映射损失,避免图像过度变形和重要特征信息丢失,提升去雾效果的稳定性。在模拟海雾图像和真实海雾图像上对该方法的有效性进行验证,并选取当前主流的海事图像去雾方法进行对比分析。研究结果表明,该方法在峰值信噪比、结构相似性和颜色差异等指标上的表现均优于其他方法,峰值信噪比平均值为21.92 dB,相比其他方法至少提升8.79 dB,能有效去除图像中的海雾,恢复图像的细节和纹理信息。In order to enhance the perception ability of ships to the surrounding environment under poor visibility conditions such as foggy weather,a maritime image dehazing method based on the cycle generative adversarial network is introduced.It enhances the stability of dehazing effects,avoiding excessive image deformation and the loss of important features by introducing cycle consistency loss and identity loss.Experimental validation is conducted on both simulated and real foggy maritime images,and then compared with mainstream dehazing methods.The results demonstrates that our method outperforms others in terms of PSNR(Peak Signal-to-Noise Ratio),structural similarity,and color difference.An average PSNR of 21.92 dB,surpassing traditional dehazing methods by at least 8.79 dB,is achieved.It is seen that the fog is effectively removed from images,and the details and texture information are restored.

关 键 词：船舶态势感知 雾天航行 图像去雾 对抗生成网络 

分 类 号：U675.5[交通运输工程—船舶及航道工程] TP391.41[交通运输工程—船舶与海洋工程]