Abstract
Brain-computer interface (BCI) can interchange messages and orders between the user's brain and the computer. The motor imagery (MI) is presented by specific signal features that reflect the user's intention to be extracted and interpreted as commands. This paper focuses on the classification of two types of MI tasks (Right Hand and Foot). We deployed various feature extraction techniques for EEG data using wavelet transform and common spatial pattern. For the wavelet features, statistical values, energy, entropy and band power were used to form the desired feature vectors. Before extracting wavelet coefficients, we performed two scenarios, with and without surface laplacian filter around the channels C3, C4 and Cz. Three types of classifiers were employed for classification, linear discriminant analysis (LDA), support vector machines (SVM) and artificial neural network (ANN). The aim of this work is to compare between them and to recommend the suitable combination for synchronous two-class motor-imagery-based brain-computer interface experiments. The data were recorded from five subjects, provided by BCI-Competition III. The results show that SVM is more suitable with the features than those extracted from wavelet coefficients and combination of entropy-energy-band power, and LDA is more suitable with common spatial pattern. Overall, the results from CSP-LDA are better than those obtained from WT-SVM with the average classification accuracy of 84.79% and 82.64%, respectively.