EEG-BASED MOTOR IMAGERY CLASSIFICATION USING DIGRAPH FOURIER TRANSFORMS AND EXTREME LEARNING MACHINES

Abstract

Brain-computer interface (BCI) systems have been widely proposed for rehabilitation and neural control of external devices. This thesis proposes a classification method for BCI EEG signals associated with motor imagery patterns. The proposed method uses a graph Fourier transform based on a symmetric graph Laplacian for directed and undirected graph models of multi-channel EEG signals. This method shows superior performance compared to other methods. Experiments were conducted using extreme learning machines (ELM) on the dataset Ia of BCI Competition 2003. The directed and undirected graph models resulted in accuracies of 96.58% and 95.9%, respectively. This work can be extended to larger BCI multi-channel EEG classification problems. For these problems, additional vertex-domain graph features and graph transform features can be considered to reveal hidden network patterns.