注意力机制下的多尺度图像超分辨率重建  

Multi-scale image super-resolution reconstruction using channel attention

作　　者：何启琛 何蕾[1] HE Qichen;HE Lei(School of Mathematics,Hefei University of Technology,Hefei 230601,China)

机构地区：[1]合肥工业大学数学学院,安徽合肥230601

出　　处：《合肥工业大学学报（自然科学版）》2024年第9期1255-1261,共7页Journal of Hefei University of Technology：Natural Science

摘　　要：文章结合目前较流行的多尺度卷积和通道注意力机制,提出一种新颖的卷积神经网络(convolutional neural network,CNN)结构,即注意力机制下的多尺度卷积神经网络。该网络结构中加入大量的残差结构,加深了网络的深度;多尺度卷积的使用使该网络能从图片中提取更加丰富的信息;注意力机制的引入使网络处理高频信息时有更大的权重。实验结果表明,多尺度注意力机制卷积神经网络在图像超分辨率(super-resolution,SR)重建上取得了良好的表现,图像细节恢复效果令人满意。This paper combines the popular multi-scale convolution and channel attention mechanism,and proposes a novel convolutional neural network(CNN)structure,namely the multi-scale CNN under attention mechanism.A large number of residual structures are added to the proposed network structure,which deepens the depth of the network.The utilization of multi-scale convolution enables the network to extract richer information from pictures.The introduction of the attention mechanism enables the network to have greater weight in processing high-frequency information.Experimental results show that the multi-scale CNN under attention mechanism has achieved good performance in image super-resolution(SR)reconstruction,and the effect of image detail restoration is satisfactory.

关 键 词：超分辨率(SR) 深度学习 卷积神经网络(CNN) 注意力机制 多尺度 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]