Graph Signal Processing Based Cross-Subject Mental Task Classification Using Multi-Channel EEG Signals

Classification of mental tasks from electroencephalogram (EEG) signals play a crucial role in designing various brain-computer interface (BCI) applications. Most the current techniques consider each channel as independent, neglecting functional connectivity brain during activity and are primarily subject specific. This paper proposes graph signal representation to classify pair using multi-channel EEG (MTMC-EEG) with cross classification within database. Here, corresponds nodes task based whose time series resides on respective nodes. Functional between these is obtained smoothness constraint Graph Signal Processing (GSP) technique. spectral features namely, two-norm total variation eigen vector (TNTV) corresponding weighted adjacency matrix, Laplacian energy (GLE) eigenvalues matrix convex sum TNTV GLE form joint (JTVE) proposed this paper. The performance methodology evaluated publicly available two different databases MTMC benchmark classifiers compared state art. Further, superiority metric smoothened GSP technique validated by comparing it Pearson correlation Gaussian radial basis function (RBF) terms accuracy, F-Score, information transfer rate (ITR). robustness method adding white noise (AWGN) SNRs.