基于对比学习的水面含雾图像复原方法-《集美大学学报（自然科学版）》

文章信息/Info

Title:: Restoration Method of Foggy Water Surface Images Based on Contrastive Learning

Author(s):: CHEN Yihan1; YU Qing2; (1.College of Artificial Intelligence,Xiamen Institute of Technology,Xiamen 361021,China；2.Navigation College of Jimei University,Xiamen 361021,China)

Keywords:: foggy water surface image; image restoration; contrastive learning; feature extraction

摘要:: 针对雾天水面交通场景下图像清晰度和可见度受到严重影响的问题，提出一种图像去雾模型。该模型基于对比学习网络框架，设计全局特征及局部特征对比模块、特征融合模块，引入混合损失函数引导模型训练，充分利用正负样本信息提高模型去雾性能，利用对比学习策略从含雾图像中恢复出清晰、无雾的图像。合成及真实水面含雾图像复原实验结果表明：该方法在主观视觉和客观评价指标上都表现出较好的性能，能见度增强后的图像可以显著改善水上图像的视觉效果，提高雾天水面交通场景下船舶目标检测的准确性，为驾驶员提供更加清晰的视野，提高航行安全性。

Abstract:: Targeting at the serious impact of image clarity and visibility in foggy water surface traffic scenes,an image defogging model is proposed.This model is based on a contrastive learning network framework to design global and local feature comparison modules,feature fusion modules,and introduces a mixed loss function to guide model training.It fully utilizes positive and negative sample information to improve the model’s defogging performance.Through a contrastive learning strategy,clear and fog free images are restored from foggy images.Through the synthesis and real water surface fog image restoration experiments,the results show that the method shows better performance in both subjective visual and objective evaluation indexes.The image with enhanced visibility can effectively improve the accuracy of ship target detection in foggy water surface traffic scenes.This can not only promote the progress of shipping technology,but also improve the safety and efficiency of navigation,and provide important support for intelligent decision-making and management.