MFF-STC: a lightweight deep learning-based method for EEG signal classification

As Brain-Computer Interface (BCI) technology continues to advance, processing motor imagery EEG signals with deep learning methods to detect human brain intentions can effectively assist patients in their rehabilitation training activities. However, the inherently low signal-to-noise ratio (SNR) of EEG signals severely impacts the effectiveness of feature extraction. This paper proposes a method based on multi-scale frequency fusion and spatio-temporal convolution to address the problem of motor imagery signal classification. The method extracts features through adaptive band splitting and a cascaded structure that combines multi-dimensional convolutions. Furthermore, it utilizes a Temporal Convolutional Network (TCN) to model long-term cross-channel dependencies and performs the final classification using a Softmax classifier. On the BCI Competition IV-2a dataset, our method achieved an average accuracy of 83.6% in within-subject tests, demonstrating ideal classification performance for the MI-EEG classification task.