Research on EEG classification method of motor imagery based on EEG-transformer model

The EEG signal has the characteristics of strong randomness, low amplitude and non-stationary. The EEG signal itself is a signal sequence composed of a set of long and continuous eigenvalues. The traditional Transformer model cannot capture the dependence between longer sequences. Setting a fixed length sequence will cause fragmentation problems. Therefore, it needs to be further adjusted and optimized. This paper proposes an EEG-transformer network architecture for EEG motor imagery classification tasks. The architecture extracts EEG-specific features through a compact convolutional neural network EEGNet, fuses the output features with the features extracted by the Transformer encoder module, and directly inputs them into the classifier for classification, thereby significantly improving the classification performance of the model. In the subject dependence assessment, the architecture achieves a classification accuracy of 79.50% on the BCI competition IV-2a dataset. Compared with the existing methods, EEG-transformer shows excellent performance, which verifies the effectiveness of the method and provides a new solution for the field of EEG decoding.