Research Of EEG Processing Algorithms In Brain-Computer Interface System

Brain-computer interface (BCI) is a technology, which is not dependent on the conventional output channel (peripheral nerve and muscle tissue) of brain, to achieve the communication and control of the human brain to a computer or other electronic devices. It is a brand new approach for information exchange and communication. By acquiring the cerebral cortex electroencephalogram (EEG) and using signal processing techniques to interpret the brain’s awareness-raising activities, it makes the dream to be true that using human brain thinking activities (awareness) to communicate with the outside world. Due to the great value for research and broad prospects for application, brain-computer interface technology becomes a hot topic to research in recent years.With respect to the key issue of brain-computer interface system, EEG signal processing technology, this thesis in-depth studies EEG preprocessing techniques, feature extraction techniques, feature selection and classification techniques. Based on the research result of a variety of methods, an asynchronous BCI system based on master-slave features has been proposed and established. By verifying in a real system, it is proved that the method proposed in this thesis is effective in practice.EEG preprocessing is to remove the noise from EEG signal. This is the basis of the good performance of the brain-computer interface system. Regarding to the fact that EEG is often mixed with ocular artifact, a method which combines independent components analysis and EEG topographic map is proposed to remove the ocular artifact in the EEG. The method of independent components analysis is used to separate individual sources in the EEG, and the method of EEG topographic map is used to identify the ocular artifact accurately. By setting the source of ocular artifact to zero and compose the individual sources back, ocular artifact is successful eliminated in the EEG.Feature extraction is the most critical step in the whole BCI system. This thesis takes the EEG of movement imagery as object of study. Based on the event-related synchronization/desynchronization phenomena in the movement imagery EEG, the wavelet packet transform, wavelet entropy algorithm and AR model are used respectively to achieve the EEG feature extraction, and six types of feature sets are constructed. Based on the technology of wavelet packet decomposition, the feature extraction method of EEG rhythm energy ratio is proposed. The features extracted by this method fully reflect the characteristics of the EEG movement imagery. High recognition accuracy is obtained, which proves the efficiency of the method.Feature selection is to optimize the result of feature extraction. This thesis proposes a feature selection method based on the combination of SFFS and PLN. Sequential forward floating search (SFFS) algorithm is used to generate the feature subsets, and piecewise linear network (PLN) is applied to assess the feature subsets in order to determine the optimal feature subset, which is the result of feature selection. The experimental result shows that this method effectively reduces the redundant features and improves the recognition accuracy.Feature classification and identification is to determine the mental tasks of the EEG The result is used to generate the control instructions sent out by the brain-computer interface system. This thesis studies the K-nearest neighbor classification algorithm and BP neural network recognition algorithm. With regard to the problems of slow training speed and hard to convergence in BP neural network, this thesis proposes a weight optimized BP neural network training algorithms, which effectively solves the above problems and makes BP neural network algorithm has practical value in actual BCI system.Finally, putting together the research results of the algorithms in this thesis, an asynchronous BCI experimental system is discussed. Regarding to the difficulty of idle/working status detection in BCI system, this thesis proposes a method to establish the asynchronous BCI system based on master-slave features. Utilizing BP feature as the master feature can detect the idle/working status effectively in an asynchronous BCI system. Utilizing EEG rhythm energy ratio as the slave feature can identify the different working status. With respect to the EEG of imagination of the left/right hand movement, an asynchronous BCI experimental system based on master-slave features is established. According to the actual testing, this system has performed very well, which proves the effectiveness of programs and algorithms.