Parametric Modeling of Band Powers for Electroencephalographic Signals

Electroencephalography (EEG) is the recording of the brain electrical activity. Classification of EEG signals into different mental tasks could be a hard problem, due to the relative low signal to noise ratios associated with EEG and the presence of artifacts originating from non brain activity. In this paper, we are concerned with the binary discrimination of EEG signals for brain-computer interface systems (BCI). In particular, the classification of EEG of left and right hand motor imagery is considered here. Feature extraction is an important issue for classification of EEG signals in BCI applications, autoregressive (AR) modeling of band powers using the Yule-Walker method and the Burg method is used to derive reliable parameters for accurate classification using linear discriminant analysis (LDA) classifier and support vector machines (SVM). Our results carried out on BCI competition datasets, showed the effectiveness of classification of EEG signals using an AR modeling of the power spectrum and common linear classifiers.