Classification of Motor Imagery EEG Signals Using Divergence Based CNN
Vibin Mammen Vinod, G. Murugesan · 2024
Motor imagery-based brain-computer interfaces (BCIs) have garnered significant attention in neuro engineering for their potential to enable communication and control in individuals with motor impairments. Electroencephalography (EEG) signals captured during motor imagery tasks offer valuable insights into the intention to execute specific movements. This study introduces a novel approach that utilizes a Divergence-based Convolutional Neural Network (DCNN) for classifying motor imagery EEG signals. The DCNN aims to augment the network’s discriminative power and enhance classification accuracy. The proposed approach incorporates three main components: Daubechies wavelet transform, Common Spatial Patterns (CSP), and CNN model. The Daubechies wavelet transform is employed to extract relevant features from the EEG signals. CSP is utilized to select discriminative features that capture the differences between motor imagery tasks, finally the CNN model is trained to classify the EEG signals into different motor imagery categories. Experimental results demonstrate that the proposed DCNN surpasses baseline CNN models and achieves higher accuracy in classifying motor imagery EEG signals. Furthermore, the DCNN exhibits robustness to variations in EEG signal characteristics with accuracy of $85 \%$ while the existing methods classifies with accuracy of 79%. The findings suggest that the integration of divergence-based regularization within the DCNN architecture holds promise for enhancing the classification performance of motor imagery-based BCIs. The proposed approach has the potential to contribute to the development of more accurate and reliable BCIs, thereby facilitating improved communication and control for individuals with motor disabilities.