Abstract:
In view of recent increase of brain computer interface (BCI) based applications, the importance of efficient classification of various mental tasks has increased prodi-giously now-a-days. For effective classification purpose, efficient feature extraction scheme is necessary. Most of the reported algorithms are performed on the EEG signals or the processed EEG signals taken from various channels while the inter-channel relationship has not been utilized. Depending on the nature of the mental tasks, different spatial locations of brain become more actuated compare to other locations. It is expected that the correlation obtained from different combination of channels will be different for different mental tasks which can be exploited to extract distinctive feature. To corroborate this idea, in the proposed method, a feature extraction scheme based on cross-correlation among data (or decomposed data) obtained from various channels is proposed. Instead of directly utilizing EEG signal, various decomposition techniques, such as empirical mode decomposition (EMD), spectral band division, wavelet decomposition (WD) and wavelet packet decomposition (WPD) are employed on a test EEG signal obtained from a channel. Different well defined narrow frequency bands, corresponding to state of vigilance, are also investigated for feature extraction. Since EEG is a non-stationary signal, EMD, WD and WPD have the potential to perform better than the conventional time-frequency analysis method. Correlation coefficients are extracted from inter-channel pre-processed EEG signal. At the same time, different statistical features of decomposed EEG signals are also obtained. Finally, the feature matrix is formed utilizing inter-channel features and intra-channel features (statistical features) of the decomposed EEG signals. Different kernels of support vector machine (SVM) clas-sifier are used to carry out classification result. For the purpose of demonstrating classification performance, ten different combinations of five different mental tasks, namely geometrical figure rotation, mathematical multiplication, mental letter com-posing, visual counting, base-line resting obtained from a publicly available dataset are utilized. It is found that the proposed scheme can classify mental tasks with a very high level of accuracy compared to some existing methods.
