Open AccessImage Compressed Sensing and Reconstruction of Multi-Scale Residual Network Combined with Channel Attention Mechanism
Author(s) -
Chunyan Zeng,
Zhenghui Wang,
Zhifeng Wang,
Yan Kang,
Yingjie Yu
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/2010/1/012134
Subject(s) - computer science , discriminative model , feature (linguistics) , residual , convolutional neural network , artificial intelligence , block (permutation group theory) , channel (broadcasting) , pattern recognition (psychology) , feature learning , process (computing) , representation (politics) , image (mathematics) , scale (ratio) , compressed sensing , computer vision , algorithm , mathematics , computer network , philosophy , linguistics , physics , geometry , quantum mechanics , politics , political science , law , operating system
Most of the existing compressed sensing reconstruction algorithms based on deep learning only use simple stacked convolutional layers to extract image feature information, which cannot extract sufficient image information; and each feature maps are treated equally in the reconstruction process, which lacks the ability of discriminative learning across feature channels and can’t make full use of the representation capabilities of convolutional neural network. This paper proposes a new multi-scale image compressed sensing reconstruction network based on residual network and channel attention mechanism. We use the residual network to increase the network depth, which reduces the information loss in the convolutional process through the skip connections in the residual block. It can obtain richer image information and protect the integrity of data information. We also add channel attention mechanism to adaptively scale each feature maps to enhance effective feature information and suppress invalid feature information. The experimental results show that the performance of image reconstruction is greatly ameliorated compared with the existing the state-of-the-art methods.