Affiliations: [a]
College of Electronic Information & Control Engineering, Beijing University of Technology, Beijing, China
| [b] Beijing Key Laboratory of Computational Intelligence and Intelligent System, Beijing, China
Correspondence:
[*]
Corresponding author. Li Mingai, [email protected].
Abstract: The Common Spatial Pattern (CSP) algorithm is known to be effective in extracting discriminative features from Motor Imagery electroencephalograms (MI-EEG). However, its performance depends on the frequency bands that relate to brain activities associated with MI tasks. To achieve an accurate classification, several methods have been proposed to determine such a set of frequency bands. However, the existing methods cannot find the multiple subject-specific frequency bands adaptively. Based on the Orthogonal Empirical Mode Decomposition (OEMD), FIR filter and CSP algorithm, a novel feature extraction method called OEFCSP is proposed to effectively perform the autonomous extraction and selection of key individual spatial discriminative CSP features. A channel selection algorithm is applied to the band-pass filtered EEG signals to reduce the number of channels. Then, each remaining channel of the EEG signal is adaptively decomposed into multiple orthogonal Intrinsic Mode Functions (IMFs) by OEMD, and each IMF is further equally divided into multiple sub-band signals by the band-pass filters. Subsequently, the CSP features are extracted from each sub-band signal and a feature ranking algorithm is employed to reorder the CSP features. Finally, a feature selection and classification algorithm is optimized to classify the selected CSP features. Experiments are conducted on a publicly available dataset, and the experimental results show that OEFCSP yields relatively higher classification accuracies compared to the existing approaches.
Keywords: Motor imagery electroencephalogram, feature extraction, orthogonal empirical mode decomposition, common spatial pattern, adaptability
