Identifying Motor Tasks From EEG Signals of Planning and Execution Phases by CNNs in a Motor Learning Procedure

In this research we analyzed EEG signals to classify motor planning and execution phases during a set of motor learning task. We began by preprocessing the EEG data and then displaying topographical plots (topoplots) of different frequency bands of EEG data to classify the phases. Afterward, we tested convolutional neural network (CNN) models, ResNet18 and EfficientNet to further enhance the processing and classification of the EEG signals. The CNN models were examined based on performance measures like accuracy, precision, recall, and F1-score. Primary findings suggested that ResNet-18 and EfficientNet models successfully distinguished the motor planning and execution phases, particularly EfficientNet presented better results. This study highlights the utilization of deep learning models as a potential way to classify EEG signals. This will help us to understand motor learning in humans better. Ultimately, after classifying the phases, we proceeded to classifing different drawn patterns and identify them using these two models. For classifying the phases using ResNet18 and EfficientNet networks, we managed to achieve accuracies of $86.3 \%$ and $89.2 \%$, respectively. Additionally, for pattern classification, accuracies of $\mathbf{8 7. 9 \%}$ for ResNet18 and $\mathbf{9 6. 9 \%}$ for EfficientNet were obtained.