ECSA-EEGNet:基于高效通道空间注意力与多尺度融合的轻量化脑电解码模型

doi:10.3969/j.issn.0258-8021.2026.02.005

Abstract
Figure/Table
References
Related Citation (15)

Download: PDF (4932 KB) HTML (1 KB)
Export: BibTeX | EndNote (RIS)

Abstract To address challenges of low spatial-channel feature fusion efficiency, high computational complexity, and weak cross-subject generalization in existing motor imagery (MI) electroencephalogram (EEG) decoding models, this study aimed to construct a novel decoding architecture that balances high accuracy with lightweight design for resource-constrained embedded applications. We proposed the ECSA-EEGNet, a lightweight model incorporating three key designs. First, a dynamic channel-space attention module was designed, utilizing parallel dual-pooling and adaptive weighting mechanisms to replace traditional fully connected dimensionality reduction, thereby enhancing the perception of key brain region features without information loss. Second, the attention sub-modules were reconstructed by adopting a fully convolutional structure to reduce redundant parameters (approximately 40% reduction compared to traditional mechanisms) and introducing a convolution-batch normalization-mish (CBM) unit to strengthen non-linear modeling capabilities. Finally, a multi-scale channel-space fusion framework was constructed to improve robustness against individual differences through cross-dimensional feature interaction. The model was validated on the BCI Competition IV-2a (4-class) and IV-2b (2-class) datasets involving nine healthy subjects. Results indicated that ECSA-EEGNet achieved an average accuracy of 80.29% on the IV-2a dataset (a 7.89% improvement over the baseline EEGNet) with a Kappa coefficient of 0.72. On the IV-2b dataset, it reached an accuracy of 85.22% (a 9.50% improvement) with a Kappa coefficient of 0.70. The model comprises only 5,022 parameters, approximately one-third of those in TCNet-Fusion. In fine-grained evaluation, Subject 3 in the IV-2a dataset achieved recognition rates exceeding 90% for all four tasks (left hand, right hand, foot, and tongue), with a peak accuracy of 95.83%, confirming the model's stability in complex tasks. These findings demonstrated that ECSA-EEGNet achieved a significant leap in decoding performance with minimal parameter overhead, effectively balancing accuracy and efficiency, thus providing a viable lightweight solution for the clinical deployment of real-time brain-computer interface systems.

Key words： brain-computer interface electroencephalogram signal decoding efficient channel spatial attention multi-scale feature fusion lightweight

Received: 19 March 2025

PACS:

R318

Corresponding Authors: ^*E-mail: xlbao@nxu.edu.cn;pengzhang123@nxu.edu.cn

About author:: ^#Member,Chinese Society of Biomedical Engineering

	Service

	E-mail this article
	Add to my bookshelf
	Add to citation manager
	E-mail Alert
	RSS
	Articles by authors
	Yang Yabing
	Dong Zheng
	Bao Xueliang
	Zhang Peng

Cite this article:

Yang Yabing,Dong Zheng,Bao Xueliang, et al. ECSA-EEGNet: A Lightweight EEG Decoding Model with Efficient Channel-SpatialAttention and Multi-Scale Fusion[J]. Chinese Journal of Biomedical Engineering, 2026, 45(2): 178-187.

URL:

http://cjbme.csbme.org/EN/10.3969/j.issn.0258-8021.2026.02.005 OR http://cjbme.csbme.org/EN/Y2026/V45/I2/178

[1] Nishiura R, Tsuru K. Development of BCI application for the physically challenged using EEG measured with multi-pin electrodes[C]//2024 IEEE International Technical Conference on Circuits/Systems, Computers, and Communications (ITC-CSCC). Hong Kong: IEEE, 2024: 1-6.
[2] Kutlu İ Ç, Tashan W, Shayea I, et al. An introductory guide on creating a Pandas-based EEG analysis and action prediction tool for BCI systems[C]//2024 IEEE 13th International Conference on Communication Systems and Network Technologies (CSNT). Dubai: IEEE, 2024: 1372-1378.
[3] Akuthota S, Rajkumar K, Janapati R. Svelte EEG Net: Cross-session motor imagery classification for BCI-controlled prosthetic arms using novel compact CNN[C]//2024 IEEE 3rd World Conference on Applied Intelligence and Computing (AIC). Raipur: IEEE, 2024: 1285-1289.
[4] Naser MYM, Bhattacharya S. Towards practical BCI-driven wheelchairs: A systematic review study[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2023, 31: 1030-1044.
[5] Wolpaw JR, McFarland DJ, Neat GW, et al. An EEG-based brain-computer interface for cursor control[J]. Electroencephalography and Clinical Neurophysiology, 1991, 78(3): 252-259.
[6] Carlson T, Millán JDR. Brain-controlled wheelchairs: A robotic architecture[J]. IEEE Robotics & Automation Magazine, 2013, 20(1): 65-73.
[7] Koles ZJ. The quantitative extraction and topographic mapping of the abnormal components in the clinical EEG[J]. Electroencephalography and Clinical Neurophysiology, 1991, 79(6): 440-447.
[8] Müller-Gerking J, Pfurtscheller G, Flyvbjerg H. Designing optimal spatial filters for single-trial EEG classification in a movement task[J]. Clinical Neurophysiology, 1999, 110(5): 787-798.
[9] Ang Kaikeng, Chin Zhengyang, Zhang Haihong, et al. Filter bank common spatial pattern (FBCSP) in brain-computer interface[C]//2008 IEEE International Joint Conference on Neural Networks (IEEE World Congress on Computational Intelligence). Hong Kong: IEEE, 2008: 2390-2397.
[10] Schirrmeister RT, Springenberg JT, Fiederer LDJ, et al. Deep learning with convolutional neural networks for EEG decoding and visualization[J]. Human Brain Mapping, 2017, 38(11): 5391-5420.
[11] Lawhern VJ, Solon AJ, Waytowich NR, et al. EEGNet: A compact convolutional neural network for EEG-based brain-computer interfaces[J]. Journal of Neural Engineering, 2018, 15(5): 056013.
[12] 李红利,刘浩雨,陈虹宇,等。基于时间序列数据增强的运动想象脑电多尺度特征提取分类 [J]. 生物医学工程学杂志,2023, 40(3): 418-425.
[13] Ingolfsson TM, Hersche M, Wang X, et al. EEG-TCNet: An accurate temporal convolutional network for embedded motor-imagery brain-machine interfaces[C]//2020 IEEE International Conference on Systems, Man, and Cybernetics (SMC). Toronto: IEEE, 2020: 2958-2965.
[14] Altaheri H, Muhammad G, Alsulaiman M. Physics-informed attention temporal convolutional network for EEG-based motor imagery classification[J]. IEEE Transactions on Industrial Informatics, 2022, 19(2): 2249-2258.
[15] Zhao Ruoqi, Wang Yuwen, Cheng Xiangxin, et al. A multi-scale spatial-temporal convolutional neural network with contrastive learning for motor imagery EEG classification[J]. Medicine in Novel Technology and Devices, 2023, 17: 100215.
[16] Woo S, Park J, Lee JY, et al. CBAM: convolutional block attention module[C]//Proceedings of the European Conference on Computer Vision (ECCV). Cham: Springer, 2018: 3-19.
[17] Wang Yihan, Wang Jiaxing, Wang Weiqun, et al. TFTL: a task-free transfer learning strategy for EEG-based cross-subject and cross-dataset motor imagery BCI[J]. IEEE Transactions on Biomedical Engineering, 2024, 72(2): 810-821.
[18] Liu Chang, You Jia, Wang Kun, et al. Decoding the EEG patterns induced by sequential finger movement for brain-computer interfaces[J]. Frontiers in Neuroscience, 2023, 17: 1180471.
[19] Xiong Shiming, Wang Li, Xia Guoxian, et al. MBCNN-EATCFNet: a multi-branch neural network with efficient attention mechanism for decoding EEG-based motor imagery[J]. Robotics and Autonomous Systems, 2024: 104899.
[20] Zhu Lei, Wang Yunsheng, Huang Aiai, et al. A multi-branch, multi-scale, and multi-view CNN with lightweight temporal attention mechanism for EEG-based motor imagery decoding[J]. Computer Methods in Biomechanics and Biomedical Engineering, 2024,27: 1-15.
[21] Li Linglong, Cao Guangzhong, Zhang Yuepeng, et al. MACNet: A multidimensional attention-based convolutional neural network for lower-limb motor imagery classification[J]. Sensors, 2024, 24(23): 7611.
[22] Deng Haodong, Li Mengfan, Li Jundi, et al. A robust multi-branch multi-attention-mechanism EEGNet for motor imagery BCI decoding[J]. Journal of Neuroscience Methods, 2024, 405: 110108.
[23] Wang Qilong, Wu Banggu, Zhu Pengfei, et al. Efficient channel attention for deep convolutional neural networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2020: 11534-11542.
[24] Bashivan P, Rish I, Yeasin M, et al. Learning representations from EEG with deep recurrent-convolutional neural networks[C]//2016 International Conference on Learning Representations (ICLR). San Juan: ICLR, 2016.
[25] Musallam YK, AlFassam NI, Muhammad G, et al. Electroencephalography-based motor imagery classification using temporal convolutional network fusion[J]. Biomedical Signal Processing and Control, 2021, 69: 102826.
[26] Huang Qiangfeng, Yang Yuanpo, Li Jun, et al. A multi-level attention-based multi-scale fusion temporal convolutional network for decoding MI-EEG signals[J]. Biomedical Signal Processing and Control, 2025, 108: 107916.