Spectral analysis of brain function network for the classification of motor imagery tasks

In order to deal with the classification for multi-class motor imagery(MI) tasks, a novel approach was presented in this paper. It is different from classical methods which classified the MI task with time-frequency analysis on EEG signals. It employs the brain function network(BFN) as a new characteristic to describe MI tasks. The BFN enlarges the features with respect to traditional time-frequency methods. Unlike analysis of statistical parameters of network such as average clustering coefficient (C) and the average pathlength (L), the proposed method employed spectral decomposition performing on BFNs, and considered the eigenvalue vector of threshold matrix as features for classification by SVM. Hence, it is speedy enough to meet the requirement of real-time in BCI-based application systems. The result of experiment demonstrates that proposed method can achieve satisfied accuracy of classification on multi-class MI tasks.