Feature Analysis And Extraction Of The Four-class Motor Imagery EEG

The brain-computer interface (BCI) is a non-muscular communication channel forexchanging messages and commands between the human brain and computer. To theparalyzed patients, especially those have normal thinking but lost of the basic bodymovement, it provides a new way to communicate with the outside world. For the analysisand extraction of the synchronization and desynchronization phenomenon in motor imagerysignals, the methods of the preprocessing, feature extraction and classification of the EEGdata have been researched on2008BCI competition statistics. Finally, a real-time BCIsystem based on four-class motor imagery is designed. The main research work is asfollows:(1) A filter based on independent component analysis (ICA) has been designed toremove the artifact signals, such as the interference of EOG and ECG. The results show thatthe artifacts are filtered successfully, the signal-noise ratio is improved and thecharacteristic in frequency domain is enhanced.(2) The following three methods: frequency-domain analysis, time-frequency analysis(Short-time Fourier Transform and Wavelet Transform) and Common Spatial Pattern (CSP)have been analyzed and compared. The results show that CSP is better than the other twomethods for the feature extraction of multichannel and multi-class EEG. Finally, a methodnamed Wavelet-CSP that is the combination of wavelet transform and CSP is proposed. Thecomparative results denominate that the Wavelet-CSP has the better classification resultsthan the three methods.(3) Combined the advantages and disadvantages of the traditional classificationstructure, such as OVO-SVM, OVR-SVM, DAG-SVM, a new four-classification structurewhich has less number of classifier and lower probability of misclassification was designed.Compared with traditional three methods, the proposed classifier has the best classificationresults.(4) A real-time BCI experimental system is devolved on the LabVIEW platform. TheEEG acquisition system is from the Thought Technology Company in Canada. Tenchannels of EEG can be collected and stored. The features can be extracted and classifiedonline by this system.