The On-line Identification Of BCI Based On Motor Imagery

BCI（Brain Computer Interface）is a device with which the human brain cancontrol the external environment directly. By gathering and interpreting the humanbrain mental task signals, BCI can translate them into a control commands. And then,the external devices will be controlled by using the commands. BCI has importantapplications in medical rehabilitation, device control in special environment. BCI hasbecome crossover hotspot in the area of science, neurophysiology, signal analysis,control science and computer science.At present BCI research has made some progresses, but still has a wide gapforthe practical application. The practical application of BCI is determined by twopreconditions. First, BCI needs a person to produce the mental task signals that canbe distinguished. Second, a system identifying the mental task signals online isneeded. One of the key technical problems need to be resolved is how to identify themental task signals online, and complete the control of external devices. Thedifficulty in it is how to make online BCI system reduce the time of the mentalsignals identification with high identification accuracy rate.In this paper, in order to achieve the online identification of four kinds ofconscious commands in motor imagery BCI system, taking the online EEG featureextraction and classification as a core, studied the key technology of onlineidentification speed and accuracy of the EEG online feature extraction andclassification. The main findings include:The feature extraction and classification algorithms in BCI online and offlinesystem were compared, when they were used to analysis the EEG signal that wouldconsist of four categories of motor imagery tasks consciousness signal. For theproblem of BCI online system real-time and continuous identifying the single EEGsignal, the wavelet and AR algorithm was used. The features that consist of theidentifiable information in specific frequency band of EEG were extracted onlinewith the algorithm. Then, the SVM was used to rea-time identify the EEG. To improve the real-time and the accuracy of identification systems, theoptimization algorithm of the feature and SVM model parameter was further studied.In order to reduce the features dimension that the classification algorithm needed, theGA algorithm was used to select the optimal feature set the fitness function. Toimprove the classification accuracy, the GA algorithm was also used to train the SVMmodel parameter. The best online system model parameters were determined by jointoptimization features and model parameters. So that, reducing system onlineidentification system EEG signals recognition algorithms time consuming and errorwould be achieved.In the testing model of BCI system, the GA was used to get the best systemmodel by the joint optimization. Based on the model which contained the best featureset and SVM model parameters for the specific subject, the individual differences wassolved. Furthermore, the universality of online system for different subjects wasenhanced and the overall recognition accuracy of the system was improved.BCI online system was designed to carry out the online experiments. Based onthe experiments, the proposed feature extraction and classification algorithm wasverified online. When carrying out the online experiment, the experimental paradigmbased on motor imagery was written, the C3, C4and the other electrodes around themacquisition channels were selected. For the special subject, the system modelparameters which were trained in the testing model were called to carry out the onlineexperiment. The online results showed that: the proposed wavelet and AR featureextraction algorithms and SVM classification algorithm have a higher identificationspeed and accuracy and meet the requirements of BCI online system for the real-timealgorithms and identification accuracy.