An end-to-end multi-task motor imagery EEG classification neural network based on dynamic fusion of spectral-temporal features

Average classification accuracy reached 85.1% ± 6.19% on a 4-class EEG BCI task.

• An end-to-end deep neural network was developed to automatically extract and combine features from EEG signals for motor imagery-based brain-computer interfaces.
• Spectral features were learned through compact convolutional neural network layers, while temporal patterns were learned using gated recurrent unit layers.
• An attention mechanism was applied to dynamically combine extracted features across EEG channels, aiming to reduce redundancy.
• The method showed comparable accuracy to recent advancements in the field with low variability among participants.
• The average classification accuracy on a 6-class dataset was 64.4% ± 8.35%, indicating variability in performance across different tasks.

AI simplified