多分支残差注意力机制融合的图像超分辨率重建  

Image Super-Resolution Reconstruction Based on Multi-Branch andResidual Attention Combination

作　　者：时维国[1] 王佳依 SHI Weiguo;WANG Jiayi(School of Automation and Electrical Engineering,Dalian Jiaotong University,Dalian 116028 China)

机构地区：[1]大连交通大学自动化与电气工程学院,辽宁大连116028

出　　处：《大连交通大学学报》2023年第3期103-108,112,共7页Journal of Dalian Jiaotong University

基　　金：辽宁省教育厅科学研究计划资助项目(JDL2019011);人工智能四川省重点实验室开放基金资助项目(2020RYJ04)。

摘　　要：针对图像超分辨率重建时存在特征映射不完全,以及映射特征易丢失而导致的重建效果不佳的问题,提出了一种多分支残差注意力机制融合的图像超分辨率重建算法。该算法在深层特征映射部分引入多分支架构,增加了网络宽度;并在每个分支中引入包括融合注意力机制的残差结构,对不同的特征信息进行侧重学习,提高了模型提取特征的能力;然后在非线性映射输出层对多个分支映射的信息进行叠加,进一步使特征信息融合,促进参数传递。试验结果表明,该算法在客观评价指标上和主观图像效果上均优于其他算法,有效提高了重建图像的细节清晰度。Aiming at the problems of incomplete feature mapping and easy loss of mapped features in image super-resolution reconstruction,an image super-resolution reconstruction algorithm is proposed based on multi branch residual attention mechanism fusion.The algorithm introduces multi branch architecture in the deep feature mapping part to increase the network width.The residual structure including attention fusion mechanism is introduced into each branch to focus on learning different feature information,which improves the ability of feature extraction.Then,the information of multiple branch mappings is superimposed in the nonlinear mapping output layer to fuse the feature information further and promote the transmission of parameters.The experiment results show that the algorithm is superior to other ones in objective evaluation index and subjective image effect with effective improvements of the detail definition of the reconstructed image.

关 键 词：图像超分辨率重建 残差网络 注意力机制 多分支架构 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]