Analysis Of Motor Imagery EEG

Users can directly communicate with the outside world or control the outside system by the brain using the Brain Computer Interface technology. The BCI based on the motor imagery is one of the important research directions. Feature extraction and classification of motor imagery EEG signal is the core of the system. And it is also the core of this thesis. Motor imagery refers to the imagination of movement and not the actual movements. In the process of imaging these actions, the energy of the EEG in specific brain regions will increase or decrease. Through the feature extraction and classification of EEG signal, it can judge the intention of the imagine movement. The results can be used for BCI device controlling, realize its value of application.This thesis uses the subjects K3b's data in the data sets IIIa of BCI Competition in 2005. The whole experiment process is 8 seconds×90 time×4 classes: 8 seconds is the whole time of single imagination trial from preparation action to the end of imagination; 90 represents the total number of a certain movement imagination; 4 class represents four kinds of motion imagination(left hand, right hand, foot, tongue). The whole process consists of 360 experimental data including 90 times of left hand movement imagination, 90 times of right hand movement imagination, 90 times of foot movement imagination, 90 times of tongue movement imagination.The thesis uses a feature extraction algorithm which is combined with CSP, Hilbert transform algorithm. And this thesis uses the Grid Search algorithm and PSO algorithm to improve the SVM performance. It also compares the optimized support vector machine(SVM) and learning vector quantization neural network classifier performance in the movement imagination EEG. The results show that with the proper parameters, SVM and LVQ's classification accuracy is as high as 88.33% and 90.213%.Finally, the EEG of different regions is used to be compared. The result shows that the data from full brain regions show the best performance. And the data from C3 and C4?Cz?CP6 channels are good at the classification of body movements. These results construct a foundation of the online application.