EEG-DBNet: A Dual-Branch Network for Temporal-Spectral Decoding in Motor-Imagery Brain-Computer Interfaces

TL;DR

This work tackles the challenge of decoding motor imagery EEG signals under low SNR and limited spatial resolution by introducing EEG-DBNet, a dual-branch network that simultaneously decodes temporal and spectral sequences. Each branch uses a local convolutional block to reshape and extract local features, followed by a global convolutional block that employs sequence splitting, SE-inspired feature reconstruction, and dilated causal CNNs to capture global patterns; the two branches are fused for final classification. Across two public benchmarks (BCI IV-2a and IV-2b), EEG-DBNet achieves state-of-the-art accuracies (approximately $85.9\%$ and $91.6\%$ Pa, with corresponding Kappa above $0.81$), outperforming multiple baselines and displaying strong generalization. The results highlight the value of parallel temporal-spectral decoding and the efficacy of sequence-level restructuring in deep MI-EEG decoding, with practical implications for robust, end-to-end BCI systems.

Abstract

Motor imagery electroencephalogram (EEG)-based brain-computer interfaces (BCIs) offer significant advantages for individuals with restricted limb mobility. However, challenges such as low signal-to-noise ratio and limited spatial resolution impede accurate feature extraction from EEG signals, thereby affecting the classification accuracy of different actions. To address these challenges, this study proposes an end-to-end dual-branch network (EEG-DBNet) that decodes the temporal and spectral sequences of EEG signals in parallel through two distinct network branches. Each branch comprises a local convolutional block and a global convolutional block. The local convolutional block transforms the source signal from the temporal-spatial domain to the temporal-spectral domain. By varying the number of filters and convolution kernel sizes, the local convolutional blocks in different branches adjust the length of their respective dimension sequences. Different types of pooling layers are then employed to emphasize the features of various dimension sequences, setting the stage for subsequent global feature extraction. The global convolution block splits and reconstructs the feature of the signal sequence processed by the local convolution block in the same branch and further extracts features through the dilated causal convolutional neural networks. Finally, the outputs from the two branches are concatenated, and signal classification is completed via a fully connected layer. Our proposed method achieves classification accuracies of 85.84% and 91.60% on the BCI Competition 4-2a and BCI Competition 4-2b datasets, respectively, surpassing existing state-of-the-art models. The source code is available at https://github.com/xicheng105/EEG-DBNet.

Figures (6)

• Figure 1: Structure of the proposed EEG-DBNet model.
• Figure 2: Structure of the local convolutional blocks. \ref{['fig_2a']} local convolutional block in the temporal branch. \ref{['fig_2b']} local convolutional block in the spectral branch.
• Figure 3: The structure of the global convolutional blocks with six sliding windows, four dilated causal convolution layers, and the kernel size is set to 4. \ref{['fig_3a']} global convolutional block in the temporal branch. \ref{['fig_3b']} global convolutional block in the spectral branch.
• Figure 4: Confusion matrices of EEG-DBNet. "L" stands for left hand, "R" stands for right hand, "F" stands for feet, and "T" stands for tongue. \ref{['fig_4a']} BCI Competition IV-2a. \ref{['fig_4b']} BCI Competition IV-2b.
• Figure 5: Classification performance ($P_a (\%)$ and $K$) on BCI Competition IV-2a for different values of $s$.