Multi-class MI-EEG Classification: Using FBCSP and Ensemble Learning Based on Majority Voting

How to decode multi-class motor imagery EEG and decipher an individual’s intents more accurately has always been a problem worthy of study in biomedical field. In the process of recognition and classification of 2-class motor imagery EEG, the Common Spatial Pattern (CSP) algorithm detects ERD /ERS by calculating the spatial filter and achieves great success. In this paper, we use self-frequency Segmented Filter Bank Common Spatial Patterns (FBCSP) with the one versus one (OVO) multi class extension on the classification of 4-class MI-EEG (BCI Competition IV Datasets 2a), and the Majority Voting strategy is adopted for the individual classifiers we select. We finally get a considerable classification result and the comparison of different methods shows that the performance of ensemble learning based on voting strategy is better than that of using individual classifier.