摘要
图像是众多任务的基础使用数据,对于本身的质量有较高的要求。但是图像的质量受到很多因素的影响,例如空气中的雾。因此图像去雾的研究就十分有必要。兴起的深度学习在各种计算机视觉任务中发挥了重要作用,在单幅图像去雾中也是如此。论文在卷积神经网络的基础上,提出设计了一种结合小波变换和注意力机制的U-NET图像去雾模型,小波变换替代原始U-NET中的上下采样,保留更多的细节信息,同时将像素注意力机制与通道注意力机制结合成注意力模块,与U-NET模块并行,作为特征补充存在。在视觉效果和定量分析中,证明该模型有较好的去雾效果。
Images are the basic usage data for many tasks,and have high requirements for their own quality.But the quality of the image is affected by many factors,such as fog in the air.Therefore,the study of image dehazing is very necessary.The emerging deep learning has played an important role in various computer vision tasks,as well as in single image dehazing.Based on the convo-lutional neural network,this paper proposes and designs a U-NET image dehazing model that combines wavelet transform and atten-tion mechanism.Wavelet transform replaces the up and down sampling in the original U-NET,and retains more detailed informa-tion.At the same time,the pixel attention mechanism and the channel attention mechanism are combined into an attention module,which is parallel to the U-NET module and exists as a feature supplement.In the visual effect and quantitative analysis,it is proved that the model has better dehazing effect.
作者
邱雨珉
郭剑辉
楼根铨
张文俊
QIU Yumin;GUO Jianhui;LOU Genquan;ZHANG Wenjun(School of Computer Science and Engineering,Nanjing University of Science and Technology,Nanjing 210094;Jiangnan Shipyard(Group)Co.,Ltd.,Shanghai 201913)
出处
《计算机与数字工程》
2024年第6期1859-1863,共5页
Computer & Digital Engineering
关键词
图像去雾
卷积神经网络
U-NET
小波变换
注意力机制
image dehazing
convolutional neural network
U-NET
wavelet transform
attention mechanism
