Deep USRNet Reconstruction Method Based on Combined Attention Mechanism

Long Chen, Shuiping Zhang (), Haihui Wang, Pengjia Ma, Zhiwei Ma and Gonghao Duan
Additional contact information
Long Chen: Hubei Provincial Key Laboratory of Intelligent Robots, Wuhan Institute of Technology, Wuhan 430205, China
Shuiping Zhang: Hubei Provincial Key Laboratory of Intelligent Robots, Wuhan Institute of Technology, Wuhan 430205, China
Haihui Wang: Hubei Provincial Key Laboratory of Intelligent Robots, Wuhan Institute of Technology, Wuhan 430205, China
Pengjia Ma: Hubei Provincial Key Laboratory of Intelligent Robots, Wuhan Institute of Technology, Wuhan 430205, China
Zhiwei Ma: Hubei Provincial Key Laboratory of Intelligent Robots, Wuhan Institute of Technology, Wuhan 430205, China
Gonghao Duan: Hubei Provincial Key Laboratory of Intelligent Robots, Wuhan Institute of Technology, Wuhan 430205, China

Sustainability, 2022, vol. 14, issue 21, 1-19

Abstract: Single image super-resolution (SISR) based on deep learning is a key research problem in the field of computer vision. However, existing super-resolution reconstruction algorithms often improve the quality of image reconstruction through a single network depth, ignoring the problems of reconstructing image texture structure and easy overfitting of network training. Therefore, this paper proposes a deep unfolding super-resolution network (USRNet) reconstruction method under the integrating channel attention mechanism, which is expected to improve the image resolution and restore the high-frequency information of the image. Thus, the image appears sharper. First, by assigning different weights to features, focusing on more important features and suppressing unimportant features, the details such as image edges and textures are better recovered, and the generalization ability is improved to cope with more complex scenes. Then, the CA (Channel Attention) module is added to USRNet, and the network depth is increased to better express high-frequency features; multi-channel mapping is introduced to extract richer features and enhance the super-resolution reconstruction effect of the model. The experimental results show that the USRNet with integrating channel attention has a faster convergence rate, is not prone to overfitting, and can be converged after 10,000 iterations; the average peak signal-to-noise ratios on the Set5 and Set12 datasets after the side length enlarged by two times are, respectively, 32.23 dB and 29.72 dB, and are dramatically improved compared with SRCNN, SRMD, PAN, and RCAN. The algorithm can generate high-resolution images with clear outlines, and the super-resolution effect is better.

Keywords: super-resolution; USRNet network; attention mechanism; multi-channel mapping; loss function (search for similar items in EconPapers)
JEL-codes: O13 Q Q0 Q2 Q3 Q5 Q56 (search for similar items in EconPapers)
Date: 2022
References: View complete reference list from CitEc
Citations: View citations in EconPapers (1)

Downloads: (external link)
https://www.mdpi.com/2071-1050/14/21/14151/pdf (application/pdf)
https://www.mdpi.com/2071-1050/14/21/14151/ (text/html)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:gam:jsusta:v:14:y:2022:i:21:p:14151-:d:957869

Access Statistics for this article

Sustainability is currently edited by Ms. Alexandra Wu

More articles in Sustainability from MDPI
Bibliographic data for series maintained by MDPI Indexing Manager ().