A SPATIOTEMPORAL DUAL-BRANCH NETWORK BASED ON CHANNEL ATTENTION AND RESIDUAL CONNECTION IN MOTOR IMAGERY EEG CLASSIFICATION
Keywords:
Motor imagery, Electroencephalogram, Channel attention, Residual connectionAbstract
Accurate decoding of motor imagery (MI) electroencephalogram (EEG) signals is a core bottleneck for the practical application of brain-computer interface (BCI) technology. Due to the characteristics of low signal-to-noise ratio, non-stationarity, and multi-dimensional feature coupling of these signals, a single feature extraction method is difficult to fully mine discriminative information. To address this problem, this study proposes a spatiotemporal dual-branch MI-EEG decoding method based on hierarchical channel attention and learnable residual connections. Taking a 3D EEG tensor as input, the method captures scalp electrode topological features and temporal dynamic dependencies through parallel spatial and temporal branches, respectively. It also introduces a channel attention mechanism with learnable residual connections to adaptively enhance the representation ability of key feature channels. Experimental results on the public BCI Competition IV 2A dataset show that the proposed method achieves an average classification accuracy of 79.63% and a Kappa coefficient of 0.7284 in four types of motor imagery tasks, significantly outperforming mainstream models such as EEGNet and FBCNet. Research indicates that the organic combination of the spatiotemporal dual-branch structure and the hierarchical channel attention mechanism can effectively improve the ability to extract spatiotemporal features and channel dependency relationships of EEG signals. The method exhibits excellent classification performance and cross-subject generalization in motor imagery EEG decoding tasks, providing strong support for feature construction and model design in related fields.References
[1] Wang X, Liesaputra V, Liu Z, et al. An in-depth survey on Deep Learning-based Motor Imagery Electroencephalogram (EEG) classification. Artificial Intelligence in Medicine, 2024, 147: 102738.
[2] Tibrewal N, Leeuwis N, Alimardani M. Classification of motor imagery EEG using deep learning increases performance in inefficient BCI users. PLOS ONE, 2022, 17(7): e0268880.
[3] Altaheri H, Muhammad G, Alsulaiman M, et al. Deep learning techniques for classification of electroencephalogram (EEG) motor imagery (MI) signals: a review. Neural Computing and Applications, 2023, 35(20): 14681–14722.
[4] Lawhern V J, Solon A J, Waytowich N R, et al. EEGNet: a compact convolutional neural network for EEG-based brain–computer interfaces. Journal of neural engineering, 2018, 15(5): 056013.
[5] Li X, Chu Y, Wu X. 3D convolutional neural network based on spatial-spectral feature pictures learning for decoding motor imagery EEG signal. Frontiers in Neurorobotics, 2024, 18: 1485640.
[6] Mo Y, Li Y, Zhang B X, et al. Motor imagery EEG classification based on weight fusion feature recalibration network. Journal of Electronic Measurement and Instrumentation, 2025, 39(1): 70-79.
[7] Brunner C, Leeb R, Müller-Putz G, et al. BCI Competition 2008–Graz data set A. Institute for knowledge discovery (laboratory of brain-computer interfaces), Graz University of Technology, 2008, 16(1-6): 1.
[8] Mane R, Chew E, Chua K, et al. FBCNet: A Multi-view Convolutional Neural Network for Brain-Computer Interface. 2021.
[9] Yu S, Wang Z, Wang F, et al. Multiclass classification of motor imagery tasks based on multi-branch convolutional neural network and temporal convolutional network model. Cerebral Cortex, 2024, 34(2): bhad511.
[10] Yu S, Wang Z, Wang F, et al. Multiclass motor imagery classification based on multi-branch CNN and TCN for BCI. Neurocomputing, 2023, 543: 126-138.