Abstract:

BACKGROUND: Brain-computer interface (BCI) has provided a direct communication pathway between brain and external devices. The BCI based on motor imagery research has developed from classification of the two types to classification of multi-class.
OBJECTIVE: To investigate accurate and effective method for extracting and classifying electroencephalogram (EEG) for multi-class imagery movement.
METHODS: First, the common average reference method was used to reduce the correlation among leads and improve the signal to noise ratio of EEG signal. The one versus the rest common spatial patterns (OVR-CSP) was used to extract the feature of EEG data, then use support vector machine of decision tree to classify the feature data. For the insufficient sample, combined with support vector machines and Bayesian classifier, expanded to the training set form the test set with a high probability of classification results, and finally re-classified by using support vector machines.
RESULTS AND CONCLUSION: The best accuracy was 92.78%. The OVR-CSP and the decision tree based on support vector machine could effectively identify multi-class EEG signal, and the expansion of the sample could improve classification accuracy.