Based On Motor Imagery Brain-Computer Interface Algorithm

The research of Brain-computer Interface reached an unprecedented stage, owing to the influence of the four International Brain-computer Interface Competitions. In recent years, the Brain-computer Interface systems continuously appeared in reality which has shown a frontier for human life. In BCI technology, the signal processing of EEG is the core problem. While the core of the entire BCI system is supported by fast, accurate, and efficient signal processing.To carry on the experimental study, this thesis starts with EEG signal-processing, which employs the data set, IVa of the BCI Competition III, as the experimental data. IVa data set is brain-computer date set of small-sample learning of imagery movement. It begins with the setting of the experiment scheme, then gives the optimum experiment plan of classification and identification of motor electroencephalogram(EEG), as well as focuses on the study of preprocessing, feature selection, feature extraction and pattern recognition of EEG signal. In the process of signal prepocessing, four filters, which are Butterworth, Chebyshev I, Chebyshev II and Elliptic, were selected for comparing to confirm which one is the optimum filter. In the process of feature selection, feature selection algorithm of correlative coefficient is proposed based on feature selection algorithm of mutual information. Then, comparative experiment is carried on. The needed required parameters in the experimental process are chosen, also the advantages and disadvantages of these two feature selection algorithm performances are compared. In the process of feature extraction, feature selection algorithm of auto-regressive(AR) and Common Spatial Pattern(CSP) are compared. As a result, classification based on LDA classification algorithm of Fisher criterion and SVM are being used. At last, after analyzing the experimental results of all kinds of algorithm combinations, the best for classification is selected, namely, Buterworth filter, feature selection of correlative coefficient, CSP feature extraction and SVM classifier. Simulation was made for this combination. The final result is compared with the submitted result of the IVa of BCI Competition III, which confirmed the experimental result of this thesis is only after that of the first in the competiton. Therefore, the algorithm combination chosen in this th esis is precise and validity. It is also suitable for the study of motor imagery electroencephalogram signal.