Reactive frequency band based movement imagery classification

Abstract

This paper proposes a novel movement imagery feature extraction algorithm based on reactive frequency band (RFB) of user using electroencephalogram (EEG) signal for brain–computer interface (BCI). Based on the asymmetric nature of the human brain, the authors propose the algorithm for the feature extraction from EEG data related to the imagination of hand and leg movement. In this proposed algorithm, the RFB has been estimated by applying time–frequency series (TFS) based method over the training dataset. Further, based on this RFB, the TFS based feature extraction method has been applied to the testing dataset. An asymmetry coefficient based on Hjorth parameter as a feature extracted for each frequency bin obtained through the TFS, further subjected to support vector machine (SVM) classifier for their classification. The effectiveness of the proposed RFB estimation method of EEG signal is validated by self-generated artificial sine wave signal. The movement imagery classification method has been validated through the three open source BCI competition datasets, such as single subject, five subjects and nine subjects movement imagery dataset. The proposed method of EEG signal processing outperforms the different well established conventional methods of EEG signal processing like discriminative frequency band common spatial pattern (DFBCSP), least square support vector machine (LS-SVM), regularized common spatial pattern (R-CSP) etc. over the BCI competition II, III and IV dataset.