Research On Algorithm Of Motor Imagery Based Asynchronous Brain-Computer Interfaces

Brain-Computer Interfaces (BCI) can establish a connection of communication and control between brain and devices, without the assistantance of muscle, the electroencephalogram (EEG) detected from the scalp can be processed to parse out the mental states of the subject, it is a new way of interaction, which can be used to fields such as medical, military, ertertainment and so on, resulting in the vigorous development in the last decades. In this paper, EEG based on left-right hand motor imagery is processed, some issues of practicable BCI is discussed and some solutions are gaven in the point of view of algorithm.EEG is very weak, in which there are various of artifacts, unfavourable to the subsequent research. In this paper, spatial and temporal features of each component derived from the algorithm Independent Component Analysis (ICA) are calculated, then the algorithm Expectation Maximization (EM) gives out the classification threshold wether the component is of artifacts, distinguish and remove the artifacts automatically, via comparison of the Event Related Desynchronization (ERD) and Event Related Synchronization (ERS) of the motor imagery EEG before and after artifacts removal, It is shown that the algorithm of artifacts removal can highlight the features related to the research, and solve the problem of inconvenient usage resulting from that specialists are needed to pick out components which belongs to the artifacts.Pattern recognition of the EEG is the core parts of the algorithm in BCI, which consist of feature extraction and classification. In this paper, the algorithm of Common Spatial Patterns which can maximize the difference of two classes of EEG is used to extract features of two and three classes of motor imagery EEG, Naive Bayes Classifier, Support Vector Machine and Linear Discriminant Analysis are used to classified the features in the scene of synchronization, and the accuracies of classification is gaven; Aiming at the practical use, method of processing in the scene of a synchronization is gaven, and with the corresponding evaluation:sensitivity and false positive rate of motor imagery detection, accuracy of the classification of left-right motor imagery.Because EEG is unstable data sequences, in this paper, we propose to use the Semi-supervised Learning for reference, making use of the train data set in which the samples are of typical characters, and uniting EEG generated in real-time to update the parameters of classifier, which can be used to the EEG in the coming duration. Compared with the scene in which the parameters of the classifier is not updated, our method leads to a better results.Comprehension from the discussions above, the system EEG_Processing is designed, similar to the practical application scene, the system can parse the original EEG to mental states of the subject.