Deep Learning Models for Classification of Motor Imagery EEG Signals

The Brain-Computer Interface (BCI) based on motor imagery electroencephalogram (MI-EEG) signals is a system that aims to recognize the movement intention of people who suffer from various severe motor impairment, in order to improve the quality of daily life. This technology is in rapid development whose main challenge is the accuracy of the classification of EEG signals. The objective of this work is to achieve the best classification method based upon deep learning methods by comparing both models; CNN (convolutional neural network) and RNN (recurrent neural network). The assessment was performed using BCI Competition IV 2a dataset which includes 4 motor imagery classes from 9 subjects. Our findings demonstrated the RNN-LSTM model's superior performance relative to the proposed CNN model. It also outperforms other state-of-the-art methods with an accuracy rate equal to 98% and using the Adam optimizer.