Abstract
Decoding brain activity through electroencephalogram (EEG) signals is still challenging, due to the low signal-to-noise ratio and spatial resolution of EEG signals. To address these limitations, we propose an advanced end-to-end network, called the parallel multi-scale attention network (PMSA-Net). This novel architecture introduces the parallel structure and channel attention mechanism to enhance the accuracy and robustness of EEG signal classification. Specifically, PMSA-Net initially expands EEG signals into the depth dimension to improve spatial resolution, thereby better capturing and learning spatial information. Subsequently, the multi-scale technique is applied to enhance the feature extraction capability of the model. Finally, channel attention mechanisms are embedded in the parallel structure to recalibrate features across multiple dimensions. Experimental results based on the BCI Competition IV 2a dataset demonstrate that the model shows superior performance to other benchmark methods, achieving the highest average accuracy and kappa across all datasets. Moreover, a series of ablation experiments are conducted to further confirm the effectiveness of the structural improvements.
Funding Information
  • Jianbing Lingyan Foundation of Zhejiang Province, P.R.China (Grant No.2023C01022)
  • National Natural Science Foundation of P.R.China (NSFC:62073286)

This publication has 27 references indexed in Scilit: