Deciphering Motor Imagery EEG Signals of Unilateral Upper Limb Movement using EEGNet

Brain Computer Interfaces (BCI) face challenges in achieving sufficient control dimensions from decoding movements of left and right limb. To address this limitation, the motor imagery (MI) of fine movements from the same arm or leg can provide natural control of external equipment and increase the available control dimensions in a BCI system. However, conventional feature extraction and machine learning techniques have shown limited potential in detecting variations in EEG signals during the imagination of movements involving unilateral limb joints. In this study, we analyse the classification of movements specific to a single limb by utilizing EEGNet. We investigate the performance of EEGNet in classifying three different states: right-hand MI, right-elbow MI, and the rest state EEG signal. Our findings demonstrate that EEGNet achieves mean classification accuracy of 71.24% for the three-class classification task. The lowest accuracy observed was 58.89%, while the highest classification accuracy reached 84.44%. The results indicate that EEGNet has the potential to effectively differentiate MI signals of joints located on the same limb, offering promising avenues for intuitive control of external equipment in BCI applications. By surpassing the limitations of conventional techniques, EEGNet opens up new possibilities for improving control dimensions and enhancing the functionality of BCI systems.