摘要
卷积神经网络是目前相对普遍且去噪性能较好的图像处理方法。传统的深度卷积神经网络(DnCNN)中同一层中的特征通道间的重要程度是平等的,不利于特征的提取。将DnCNN与通道注意力,以及空洞卷积神经网络构成的稀疏块相结合,提出了一种通道注意力引导的卷积神经网络CDNet用于图像去噪。不仅更有效地提取图像复杂背景下的更有用的信息,还降低网络训练的复杂性。对比试验结果表明该网络在不同公开数据集上的PSNR值以及SSIM值都优于其余去噪网络,去噪效果相对较好。
Convolutional neural network’s utilization is a relatively common image processing method with good denois-ing performance.In traditional deep convolutional neural network(DnCNN),the importance of feature channels in the same layer is equal,which is not conducive to feature extraction.Combining DnCNN with channel attention,and sparse blocks com-posed of null convolutional neural networks,a channel attention-guided convolutional neural network CDNet is proposed for image denoising.It can not only extracts more useful information from the complex background of the image more effectively,but also reduces the complexity of network training.The experimental results show that the PSNR value and SSIM value of the network on different public data sets are better than other denoising networks,and the denoising effect is relatively good.
作者
孙光灵
彭欣仪
SUN Guanging;PENG Xinyi(School of Electronic and Information Engineering,Anhui Jianzhu University,Hefei 230601,China;Anhui Laboratory of Intelligent Interconnection System,Hefei University of Technology,Hefei 230009,China)
出处
《安庆师范大学学报(自然科学版)》
2023年第4期60-65,共6页
Journal of Anqing Normal University(Natural Science Edition)
基金
国家自然科学基金资助项目(62001004)
安徽省高校协同创新项目(GXXT-2021-024)
合肥工业大学“智能互联系统安徽省实验室”开放基金(PA2021AKSK0107)。
关键词
图像处理
图像去噪
通道注意力
空洞卷积神经网络
稀疏机制
image processing
image denoising
channel attention
dilated convolutional neural network
sparse mecha-nism