Utilizing Graph Signal-Processing-Based Spectrum to Classify Mental Tasks With Multichannel EEG Signals

The precise identification of mental tasks through the analysis of electroencephalogram (EEG) signals plays an important role in the field of brain–computer interfaces (BCIs). However, existing techniques often overlook the functional connectivity that occurs within the brain during mental activities by treating each EEG channel independently. To address these limitations, this research paper presents a novel approach based on graph signal processing (GSP). The proposed method utilizes graph signal representation to classify pairs of mental tasks based on EEG signals. In this approach, each channel of the EEG signal is associated with a node in a task-based graph, and the EEG time series corresponding to each node is represented within the graph structure. To capture the functional connectivity between these nodes, the method employs GSP techniques while incorporating a smoothness constraint. In this article, graph spectral features such as the graph Fourier transform (GFT), which leverages the eigenvalues of the Laplacian matrix, along with the graph wavelet transform (GWT), incorporating statistical measures such as mean, variance, standard deviation, kurtosis, and energy is used. Two openly accessible databases are employed for evaluation purposes: the Keirn and Aunon dataset (Dataset I) and the EEG mental arithmetic task database (EEGMAT Dataset II). Furthermore, the proposed model achieves an impressive accuracy of 99.5% for database I and 99.75% accuracy for the BT-SS task pair in EEGMAT database II. By leveraging the graph signal representation and incorporating GSP techniques, this article aims to achieve accurate classification of different mental tasks.