A Compact Depthwise Convolutional Neural Network for Motor Imagery EEG Classification

The identification of motor imagery electroencephalogram (EEG) signals has been a significant research direction in the field of non-invasive brain-computer interfaces. In recent years, the successful application of convolutional neural network in EEG-based brain-computer interfaces has enhanced the classification and recognition technology of motor imagery. However, faced with challenges such as the non-linear nature of EEG signals, low signal-to-noise ratio, individual differences and time variability, the accuracy of motor imagery classification remains suboptimal. Addressing these challenges, We proposes a novel, compact depthwise-based convolutional neural network for the classification of EEG signals. We have conducted experiments on BCI Competition IV Dataset 2a and 2b to validate the performance and effectiveness of the model. We test the parameter sensitivity of pooling kernel size, and compared the model against the state-of-art models. Experimental results demonstrate that the proposed network architecture significantly improves the classification performance on the BCI Competition IV Dataset 2a and 2b.