Research On Encoding Paradigm And Decoding Algorithm Of EEG Information Induced By Sequential Finger Movement

Brain-computer interface(BCI)is an information communication channel which does not depend on peripheral nerve and muscle tissue.BCI system based on the decoding of brain motor-related mental information which can convert subjective motional intention into output instructions,not only establish a direct information communication path between motor intentions and limb movements,but also have important scientific significance and application value in motor rehabilitation,replacement and enhancement.However,the related BCI technology still faces two bottlenecks,one is the small instruction set,the other is the low decoding efficiency.Aiming at the above key problems,this paper explores a new paradigm for instruction set extension and a new method for feature extraction based on finger sequence keypresses,in order to promote the rapid development of BCI in intelligent control and rehabilitation based on motor mental information decoding.Firstly,the repetitive keypress paradigm of left and right index fingers was selected to explore the effect of number of keystrokes that is data length on classification accuracy.Ten healthy subjects participated in this experiment.In this paper,we used task-related components and discriminant spatial pattern spatial filtering algorithm to extract motor-related cortical potential(MRCP)feature,and common spatial pattern spatial filtering algorithm to extract event-related desynchronization(ERD)feature.The above features were screened and fused based on mutual information.Finally,support vector machine was used to classify and recognize.The results showed that with the increase of data length,the classification accuracy also increased significantly,and when the data length is only 1 second(single key),the average classification accuracy was more than 90%.The above results showed that increasing the number of keypresses in a single trial could improve the classification accuracy.Then,on the basis of repetitive paradigm,we designed the experiment of sequential keypress of left and right index fingers,including four motor task patterns:left and left,right and right,left and right,right and left.Ten healthy subjects participated in the experiment.The average classification accuracy of the four classifications was 77.15%.The above results showed that the left and right index finger sequential keypress paradigm effectively could expand instruction set in a shorter instruction time(only 2 s).Finally,two repetitive keypress modes(left and left,right and right)and two sequential keypress modes(left and right,right and left)were tested online.A total of12 participants participated in the online experiment,of which the average online results of left and left and right and right keypress action were 83.33%,and the average online results of left and right and right and left keypress action were 82.71%.These results further proved the feasibility and effectiveness of the left and right index finger sequence keypress paradigm and feature extraction and screening fusion method based on multi-class spatial filtering and mutual information analysis.In conclusion,the offline and online experiments of repetitive and sequential keypress of the left and right index finger keypress paradigm were conducted successfully in this paper.The feature extraction and selection method based on spatial filtering and mutual information is used to extract,fuse and classify the features of MRCP and ERD effectively,which provides theoretical basis and technical support for improving BCI performance of motor-related mental EEG information decoding.