详细信息
文献类型:期刊文献
中文题名:基于域自适应的水下图像质量提升
英文题名:Domain Adaptation for Underwater Image Enhancement
作者:邴雪雯[1];任温琦[1];唐漾[1]
机构:[1]华东理工大学能源化工过程智能制造教育部重点实验室,上海200237
年份:2025
卷号:32
期号:9
起止页码:1619
中文期刊名:控制工程
外文期刊名:Control Engineering of China
收录:;北大核心:【北大核心2023】;
基金:国家科技部重点研发计划项目(2021YFB1714300);国家自然科学基金重大项目(62293502)和重点项目(62233005);中央高校基本科研业务费专项资金资助项目(222202317006);高等学校学科创新引智计划(B17017);上海人工智能实验室资助。
语种:中文
中文关键词:水下图像质量提升;域自适应;深度学习;图像处理
外文关键词:Underwater image enhancement;domain adaptation;deep learning;image processing
摘要:水下图像质量提升在海洋探测中有着重要的地位,但真实数据难以获取,现有方法多采用合成数据训练的方式。然而,合成数据无法还原复杂的水下环境,导致模型的实际应用效果较差。为了解决上述问题,提出了一个从水上到水下的域自适应框架来进行水下图像的质量提升,框架分为颜色校正和域自适应两步。为了更好地提取全局信息,在域自适应步骤中引入了Transformer模块作为编码器来提取输入图像的特征。同时,还设计了特征增强模块来保留随空间变化的纹理细节和边缘信息。实验结果表明,该网络框架不仅在水下图像质量度量(underwater image quality measure, UIQM)性能指标中表现良好,同时还可达到更优越的视觉感知效果。
Underwater image enhancement is crucial for marine exploration,but real-world data is difficult to obtain,so most current methods rely on synthetic data for training.However,synthetic datasets often fail to describe the natural appearance and show poor capabilities of the generalization.An in-air to underwater image enhancement framework that overcomes the limitations of underwater synthetic datasets is proposed.Specifically,the presented framework consists of a color correction step and a domain adaptation step.To better extract global information,we introduce Transformer module as an encoder to extract features of the input image in the domain adaptation step.We also propose a feature enhancement module to accommodate spatially varying textures and edges.Experimental results indicate that the framework not only obtains remarkable UIQM scores than the previous methods do,but also achieves superior results in visual perception.
参考文献:
正在载入数据...
