Research On Feature Extraction And Pattern Classification Of Motor Imagery Electroencephalogram

Brain-computer interface technology makes brain communicate with the machine directly.Its main feature is that it doesn't require peripheral nerve and muscle tissue involvement.People just collect EEG signal to process,analyze and identify during thinking,and then convert the ideas into external device(such as a computer,a robot and wheelchairs,etc.)to communicate with the outside world.The technology has been used in medical health,military defense,automotive driving,game entertainment and other fields.So it not only has important theoretical research value,but also has good prospects,which has become a hot topic of the current individual countries.Considering Motor Imagery(MI)EEG that contains abundant motor information can avoid risk caused by electrodes implantation and can be obtained with low cost,it is selected to be as object of study in this paper.The subject study is supported by the National Natural Science Foundation.Firstly,the background and importance of EEG study is introduced.Secondly,research status of brain-computer interface(BCI)is reviewed and the characteristics of EEG and composition of BCI is introduced.And then the method of preprocessing,feature extraction and pattern classification about MI EEG is analyzed and summarized.Lastly,some unsolved problem is proposed and then relevant study is started.The thesis involves following content:(1)In the stage of EEG preprocessing:Nonlinear cell-average multiscale signal representations de-noising method is used to filter a set of signals that is not concerned with MI EEG and can make the Signal-to-Noise-Ratio(SNR)enhance,so as to provide precondition to extract feature and classify.(2)In the stage of EEG feature extraction: Brain functional network is introduced to feature extraction of electroencephalography(EEG),and a novel method is proposed based on Lasso-Granger causality between regions of interest(ROI)in the brain,in order to overcome the inherent deficiencies of research methods based on isolated brain region.Firstly,the maximum principal component of ROIs is extracted by principal component analysis,and then causality values between ROIs are calculated by Lasso-Granger.Finally,the values are used as the input vector.This method provides a new idea for the study of extracting EEG features.(3)In the stage of EEG pattern classification: to make classification more accurate and fast,and improve adaptive capacity of classifier,as well as enhance thepracticality BCI system,a method based on incremental twin SVM(INC_TVM)is proposed in this paper.The idea of the method is given as follows: Firstly,considering the issue of BCI system training samples with long time,a new extension of support vector machine-TSVM is used as the basic model for subsequent incremental expansion algorithm.Secondly,incremental learning is introduced into the classification model building to make individuals with good self-learning and correcting performance to enhance the adaptive capacity of BCI system to achieve accurate classification.In addition,Particle Swarm Optimization(PSO)is used in parameter optimization of twin SVM to get the best classification results.In this paper,all the algorithms are based on motor imagery BCI signal processing methods and are carried out on the MATLAB implementation.Simulation results show the effectiveness of the algorithm.