Study On Motor Imagery Based Online Brain Computer Interface Using Semi-supervised Algorithm

Brain-Computer Interface(BCI)is a new channel for information exchange between the human brain and the outside world.It not only provides a new information output channel for patients with shackles,but also provides services for normal people.It has broad application prospects in various fields such as medical,psychological,military,and entertainment.BCI research covers a wide range of disciplines,including biology,communications,and computer technology.In recent years,BCI has received widespread attention from scholars at home and abroad and has become a hot topic in current research.The subject of this study is BCI based on motor imagery.Its EEG signal is different from P300 and steady-state visual evoked potential,etc.Subjects need to be trained for a period of time(ranging from weeks to months)to get a more visible signal of motor imagery.In addition,the traditional supervised classification method also needs to obtain a large number of labeled motor imagery EEG samples to train better performance classifiers.Obtaining a large number of labeled EEG samples is a time consuming,labor intensive,and costly process.Therefore,constructing a motor imagery BCI system under a small number of labeled samples has great research significance and practical value.Aiming at the situation of small training samples,this paper studies the motor imagery BCI system based on semi-supervised algorithm.The main research contents are as follows:(1)This paper studies the feature extraction using Common Spatial Patterns(CSP)and the supervised classification model based on Support Vector Machine(SVM).Based on this,a self-training semi-supervised classification model based on CSP and SVM is given.In order to obtain better classification of motor imagery under small sample conditions,Particle Swarm Optimization(PSO)is used to optimize the relevant parameters of SVM,and a semi-supervised classification model based on PSO and SVM is constructed.In this paper,the proposed semi-supervised classification model based on PSO optimization is verified by experiments,and compared with the supervised classification model and the unoptimized classification model.The experimental results show that the proposed method can achieve higher classification accuracy.(2)Considering the importance of confidence discrimination in sample selection in semi-supervised algorithms,this paper proposes an improved self-training semi-supervised algorithm to help the main classifier SVM pick unmarked samples with high-confidence by adding a classifier kernel Fisher discriminant analysis,and we use these samples to update the training set and retrain the classification model.Using the BCI competition data,the improved self-training algorithm and the unimproved algorithm were analyzed and compared offline,and the effectiveness of the proposed algorithm was verified.The improved classification rate of motor imagery was significantly improved.(3)Based on the improved self-training algorithm,an online motor imagery BCI system based on semi-supervised learning is devised and achieved.By imagining the left and right hand movements,the BCI selection path can be manipulated to sequentially illuminate the small lights for online games.The overall design of the system is divided into five modules: real-time communication,preprocessing,signal storage,feature extraction and classification,and output display.It is worth noting that the system uses Neuroscan software to realize real-time acquisition and transmission of multi-channel EEG signals.The upper computer software was written by LabVIEW platform to realize signal preprocessing,CSP-based motor imagery feature extraction,and motor imagery classification based on improved self-training algorithm.In order to evaluate the designed online BCI system,an experimental study is carried out,which is divided into three parts: 1)subject training experiment,collecting offline data to train classification model and providing initial classifier for online experiment;2)PC software test experiment,through the BCI competition data to verify the functions of the online system PC software;3)online experiment to evaluate system performance.Two subjects operated the BCI system through motor imagery to complete the game operation and counted the experimental results.When more marked samples are used as the initial training set,the online BCI system based on semi-supervised learning can achieve better classification results.When the number of samples in the initial training set is reduced,the system uses the unlabeled samples for online learning,and the classification effect shows a clear upward trend.It is verified that the online BCI system constructed by the improved self-training algorithm is reliable and effective.