Self-attention-based convolutional neural network and time-frequency common spatial pattern for enhanced motor imagery classification

Mean accuracies of 79.28% and 86.39% were achieved on two EEG datasets for motor imagery classification.

• A self-attention-based Convolutional Neural Network (CNN) was designed to enhance classification of motor imagery EEG signals.
• Data augmentation techniques were used to increase the size of the training datasets due to limited available data.
• The self-attention module calculates channel weights to identify and select the most active EEG channels.
• Multiscale time-frequency-space features were extracted using a time-frequency common spatial pattern (TFCSP) approach.
• The combination of self-attention-based CNN and TFCSP improved overall classification performance compared to existing methods.

AI simplified