Classification Of Four Kinds Of Motor Imagination EEG Signals

BCI technology combines the knowledge of signal detection,signal processing,pattern recognition and nerve direction,so it can be used as a real-time communication system to connect the brain with external tools.Among them,the motor imagery EEG interface can directly convert all the information generated in the human brain into behaviors that conform to external activities,and the converted behaviors can supersede human limbs or language tissues to realize interchange with the exterior world and command the outer circumstances.At present,the application prospect of brain-computer interface technology is very broad,which can be applied to medical care,internet of things,military and so on.This thesis mainly researches the classification precision rate of EEG signals build on left hand,right hand,feet and tongue,primarily from two ways: characteristic extraction and f characteristic classification.The concrete study substances are as follows:1.Considering the non-stationary and nonlinear features of EEG,the time-frequency characteristics and spatial characteristics are conjunct in feature extraction,and two specific feature extraction methods are proposed:(1)The method of combining wavelet packet transform with common space pattern;(2)The method of combining discrete wavelet transform with common space mode.In this thesis,the features extracted by this method and the common space pattern feature extraction method are sent to the single-core SVM classifier for comparison.The experiment shows that the method combining wavelet packet transform with the common space pattern is superior.2.In order to study the classification methods of motor imagery EEG signals,two classification algorithms are applied respectively: linear discriminant method and logistic regression.In view of the problem that logistic regression needs to set regularization coefficients manually in actual operation,which is easy to lead to over-fitting,an optimization method of cross-validation for logistic regression algorithm is proposed.The improved algorithm is adapted to the feature classification,and compared with the classification results of the other two models.The experimental results show that the improved logistic regression is the best in the four classifications.3.In order to obtain more accurate classification results,a classification method based on ensemble learning is proposed.bagging method is used to integrate the three algorithm models of linear discriminant,logistic regression and improved logistic regression,and the results are predicted and compared by soft voting and hard voting respectively.The final classification accuracy can reach 94.1%,which is better than other existing paper methods.