Rehabilitation Training System Based On Motor Imagery Brain-Computer Interface

Brain-computer interface is a technology that can establish direct connections between the human brain and machines.As a new type of human-computer interaction technology,it has broad application prospects in fields such as healthcare,nursing,entertainment,and industry.However,the difficulty of analyzing EEG signals is high,and the recognition accuracy is not high,making it difficult to perform multi classification recognition.And existing research mostly stays in the offline recognition stage of EEG signals,making algorithms difficult to meet the online parsing requirements of EEG signals in practical applications.This article aims to establish an efficient brain-computer interface real-time control system that can control the up,down,left,and right movements of rehabilitation robots.We propose a multi domain fusion EEG signal analysis algorithm to achieve multi classification recognition of motion imagination signals.And a real-time online brain-computer interface control system was constructed to detect and analyze user motor imagery signals in real-time,identify their imagination schemes,and control the rehabilitation robot to move up,down,left,and right.The main work of the paper is as follows:Firstly,a signal acquisition system and preprocessing scheme were designed based on the characteristics of EEG signals.Secondly,in order to solve the problems of low recognition accuracy,difficulty in multi-classification and low real-time performance of current signal analysis algorithms,a multi-domain fusion EEG signal analysis algorithm is proposed to achieve real-time analysis of user motion imagery signals.The scheme is based on a linear discriminant analysis algorithm combining spatial and frequency domains(WFCL).The feature extraction method is a combination of Feature Weighting and Regularization of Common Spatial Patterns,the Waveformlength Optimal Spatial Filter(WOSF)and the Power Spectral Density(PSD).Then,the linear Discriminant Analysis(LDA)algorithm is used to classify the feature.Finally,the user’s motor imagery signal is discriminated by voting.WFCL algorithm is fast in operation and has a strong discriminant ability to EEG signals.However,in a few cases,WFCL will get a maximum of 2 votes,at which point the user’s imaginary category cannot be identified.In this case,signal resolution is assisted by feature extraction based on Pearson correlation coefficient and signal resolution algorithm based on K-value nearest neighbor classification(PCC-KNN).In this way,we can make full use of the fast and accurate advantage of WFCL method,and make up for the defect that WFCL can not be effectively identified when the maximum voting number is 2,and finally realize the real-time analysis of EEG signal.This paper validates the data from several well-known public databases,and obtains the highest 94.4% discriminant accuracy and 80.3%average accuracy in the four classifications validation,which is better than the latest EEG processing algorithms of major research teams.In the actual online experiments of multiple subjects,the highest accuracy rate was 80.6%,with an average of 69.1%.Finally,this paper builds a rehabilitation system based on MI-BCI,which implements the real-time collection,preprocessing and analysis of EEG signals,and inputs the parsed control signals to the rehabilitation robot in real time to control the rehabilitation robot to move up,down,left,and right.The experimental results show that the system can extract and analyze the user’s EEG signal in real time and accurately,and control the movement of the rehabilitation robot accurately according to the user’s imagination.