Composite attention multiscale spatial-frequency network for underwater image super-resolution

Jinpeng Zhang; Hao Wu; Maichen Xiong; Chengdong Wu; Xiujing Gao; Yaoming Zhuang

doi:10.1109/EECR69522.2026.11549262

Composite attention multiscale spatial-frequency network for underwater image super-resolution

Jinpeng Zhang
, Hao Wu
, Maichen Xiong
, Chengdong Wu
, Xiujing Gao
, Yaoming Zhuang

School of Computer, Data & Mathematical Sciences

Northeastern University China
Nanyang Technological University
Fujian University of Technology

Research output: Chapter in Book / Conference Paper › Conference Paper › peer-review

Abstract

Underwater images suffer from severe degradations such as color distortion and detail loss, restricting visual perception tasks. This paper proposes a Multiscale SpatialFrequency Information Network (MSFIN) for underwater image super-resolution. Integrating the strengths of CNNs and Transformers, MSFIN employs a Residual Compound Attention Block (RCAB) to fuse local and global features, and an Overlapping Cross-Attention Block (OCAB) to overcome the limited receptive field of standard windows. Furthermore, a Spatial-Frequency Enhancement Block (SFEB) is introduced to explicitly recover high-frequency textures. The experimental results demonstrate that this proposed method achieved excellent performance on both the UFO-120 and USR-248 datasets.

Original language	English
Title of host publication	Proceedings of the 12th International Conference on Electrical Engineering, Control and Robotics (EECR 2026), April 6-8, 2026, Suzhou, China
Place of Publication	U.S.
Publisher	IEEE
Pages	266-272
Number of pages	7
ISBN (Electronic)	9798319529329
DOIs	https://doi.org/10.1109/EECR69522.2026.11549262
Publication status	Published - 2026
Event	International Conference on Electrical Engineering, Control and Robotics - Suzhou, China Duration: 6 Apr 2026 → 8 Apr 2026 Conference number: 12th

Conference

Conference	International Conference on Electrical Engineering, Control and Robotics
Abbreviated title	EECR
Country/Territory	China
City	Suzhou
Period	6/04/26 → 8/04/26

Keywords

attention mechanism
superresolution
underwater image
wavelet transform

Access to Document

10.1109/EECR69522.2026.11549262

Cite this

Zhang, J., Wu, H., Xiong, M., Wu, C., Gao, X., & Zhuang, Y. (2026). Composite attention multiscale spatial-frequency network for underwater image super-resolution. In Proceedings of the 12th International Conference on Electrical Engineering, Control and Robotics (EECR 2026), April 6-8, 2026, Suzhou, China (pp. 266-272). IEEE. https://doi.org/10.1109/EECR69522.2026.11549262

@inproceedings{57c1f1c3f9e94cfabad0195d3b602d45,

title = "Composite attention multiscale spatial-frequency network for underwater image super-resolution",

abstract = "Underwater images suffer from severe degradations such as color distortion and detail loss, restricting visual perception tasks. This paper proposes a Multiscale SpatialFrequency Information Network (MSFIN) for underwater image super-resolution. Integrating the strengths of CNNs and Transformers, MSFIN employs a Residual Compound Attention Block (RCAB) to fuse local and global features, and an Overlapping Cross-Attention Block (OCAB) to overcome the limited receptive field of standard windows. Furthermore, a Spatial-Frequency Enhancement Block (SFEB) is introduced to explicitly recover high-frequency textures. The experimental results demonstrate that this proposed method achieved excellent performance on both the UFO-120 and USR-248 datasets.",

keywords = "attention mechanism, superresolution, underwater image, wavelet transform",

author = "Jinpeng Zhang and Hao Wu and Maichen Xiong and Chengdong Wu and Xiujing Gao and Yaoming Zhuang",

year = "2026",

doi = "10.1109/EECR69522.2026.11549262",

language = "English",

pages = "266--272",

booktitle = "Proceedings of the 12th International Conference on Electrical Engineering, Control and Robotics (EECR 2026), April 6-8, 2026, Suzhou, China",

publisher = "IEEE",

note = "International Conference on Electrical Engineering, Control and Robotics, EECR ; Conference date: 06-04-2026 Through 08-04-2026",

}

Zhang, J, Wu, H, Xiong, M, Wu, C, Gao, X & Zhuang, Y 2026, Composite attention multiscale spatial-frequency network for underwater image super-resolution. in Proceedings of the 12th International Conference on Electrical Engineering, Control and Robotics (EECR 2026), April 6-8, 2026, Suzhou, China. IEEE, U.S., pp. 266-272, International Conference on Electrical Engineering, Control and Robotics, Suzhou, China, 6/04/26. https://doi.org/10.1109/EECR69522.2026.11549262

Composite attention multiscale spatial-frequency network for underwater image super-resolution. / Zhang, Jinpeng; Wu, Hao; Xiong, Maichen et al.
Proceedings of the 12th International Conference on Electrical Engineering, Control and Robotics (EECR 2026), April 6-8, 2026, Suzhou, China. U.S.: IEEE, 2026. p. 266-272.

Research output: Chapter in Book / Conference Paper › Conference Paper › peer-review

TY - GEN

T1 - Composite attention multiscale spatial-frequency network for underwater image super-resolution

AU - Zhang, Jinpeng

AU - Wu, Hao

AU - Xiong, Maichen

AU - Wu, Chengdong

AU - Gao, Xiujing

AU - Zhuang, Yaoming

N1 - Conference code: 12th

PY - 2026

Y1 - 2026

N2 - Underwater images suffer from severe degradations such as color distortion and detail loss, restricting visual perception tasks. This paper proposes a Multiscale SpatialFrequency Information Network (MSFIN) for underwater image super-resolution. Integrating the strengths of CNNs and Transformers, MSFIN employs a Residual Compound Attention Block (RCAB) to fuse local and global features, and an Overlapping Cross-Attention Block (OCAB) to overcome the limited receptive field of standard windows. Furthermore, a Spatial-Frequency Enhancement Block (SFEB) is introduced to explicitly recover high-frequency textures. The experimental results demonstrate that this proposed method achieved excellent performance on both the UFO-120 and USR-248 datasets.

AB - Underwater images suffer from severe degradations such as color distortion and detail loss, restricting visual perception tasks. This paper proposes a Multiscale SpatialFrequency Information Network (MSFIN) for underwater image super-resolution. Integrating the strengths of CNNs and Transformers, MSFIN employs a Residual Compound Attention Block (RCAB) to fuse local and global features, and an Overlapping Cross-Attention Block (OCAB) to overcome the limited receptive field of standard windows. Furthermore, a Spatial-Frequency Enhancement Block (SFEB) is introduced to explicitly recover high-frequency textures. The experimental results demonstrate that this proposed method achieved excellent performance on both the UFO-120 and USR-248 datasets.

KW - attention mechanism

KW - superresolution

KW - underwater image

KW - wavelet transform

UR - https://www.scopus.com/pages/publications/105042711950

UR - https://go.openathens.net/redirector/westernsydney.edu.au?url=https://doi.org/10.1109/EECR69522.2026.11549262

U2 - 10.1109/EECR69522.2026.11549262

DO - 10.1109/EECR69522.2026.11549262

M3 - Conference Paper

AN - SCOPUS:105042711950

SP - 266

EP - 272

BT - Proceedings of the 12th International Conference on Electrical Engineering, Control and Robotics (EECR 2026), April 6-8, 2026, Suzhou, China

PB - IEEE

CY - U.S.

T2 - International Conference on Electrical Engineering, Control and Robotics

Y2 - 6 April 2026 through 8 April 2026

ER -

Composite attention multiscale spatial-frequency network for underwater image super-resolution

Abstract

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this