Deep Learning-Based Classification Research of Multi-class Motor Imagery EEG Signals

This paper addresses the challenges of large parameter quantities and low operational efficiency faced by traditional deep learning models in EEG signal processing by proposing a model that integrates a One-dimensional Convolutional Neural Network (1DCNN) with a Temporal Convolutional Network (TCN). The model enhances signal class separability using One Versus Rest the Common Spatial Pattern (OVR-CSP) and strengthens key features through Squeeze-and-Excitation (SE), forming a dual-branch 1DCNN-TCN model for motor imagery EEG signal classification. Experimental validation on the BCI Competition IV 2a dataset demonstrates an average accuracy of $\mathbf{8 3. 1 3 \%}$. Compared to other advanced deep learning models, the training time is reduced by 8% to 23%, showcasing superior classification performance and efficient processing speed. This model offers a novel solution for EEG signal processing, making it particularly suitable for brain-computer interface systems that require rapid responses.