Research On Feature Extraction And Recognition Of Four-class Motor Imagination EEG Signals

There are people whose brains are intact and other body parts are damaged due to various reasons.Due to diseases,accidents and other reasons,the body loses part or all of the brain control,which affects the lives of these people.Motor imagery(MI)can stimulate the activity of the patient’s cerebral cortex.Through the feature extraction and processing of electroencephalography(EEG)signals of motor imagery,it can be used to control external equipment or feed back to disease patients to help them recover.Treatment to achieve the purpose of treating patients.The research of the Brain Computer Interface(BCI)system for motor imaging is of great significance.The BCI system based on motor imagination,that is,the subject performs motor imagination,collects the EEG signals generated by it,processes the collected motor imagination EEG signals,extracts and classifies features,and then converts them into control signals for transmission Into the machine equipment,so as to realize the connection between the human brain and the machine.Aiming at the preprocessing,feature extraction,and signal classification of four-category motor imaging EEG signals,this article mainly does the following work:(1)Collection and preprocessing of EEG signals for motor imaging.According to the characteristics of the motor imagery EEG signal,design a paradigm based on the motor imagery EEG signal brain-computer interface experiment,and use EEG acquisition equipment to perform data collection,and preprocess the collected EEG signal,including filtering,Independent component analysis,etc.(2)Use the Common Spatial Pattern(CSP)method to extract features of the motor imagery EEG signal.Through the simulation results,compared with the power spectrum,it is found that the characteristics of the ARMA(Auto-Regressive and Moving Average Model,ARMA)model are more obvious and the effect is better.For the dimensional problem of ARMA feature extraction,the CSP feature extraction method solves the problem of fewer signal leads in the ARMA model,and at the same time,improves the effectiveness of the features.(3)Use Support Vector Machine(SVM)algorithm for feature classification.In view of the different classification effects of different kernel functions of SVM,this paper proposes the use of genetic algorithm(GA)to optimize the parameters of SVM,and find out the best parameters of SVM for EEG classification.Comparing the results of linear classifiers(LDA),using GA’s optimized SVM classification method,on the collected data set,an average accuracy rate of 88.4% was obtained,which proved the validity of the data and the performance of the feature extraction algorithm and the classification algorithm.feasibility.