The Key Technology Of Brain-computer Codec Based On The Weak Event-related Electroencephalogram Features And Its Application

As a direct control and communication pathway between brain and output devices,brain-computer interface(BCI)can provide a new way for brain to interact with external information.The brain-computer interface system based on event-related potential(ERP)has the fastest interaction speed and the highest information transfer rate.However,due to the limitation of low amplitude and weak signal-to-noise ratio of event-related Electroencephalography(EEG)response,it is still difficult to realize natural and efficient coding and decoding of ERP-BCI,which also hinders the process of application of BCI technology.In view of the above problems,the research focuses on the development of new coding paradigm,the design of stable decoding algorithm and the build of interactive and friendly BCI system,in order to explore the key technologies and applications of brain-computer codec based on weak event-related EEG characteristics.Aiming at the problem that ERP-BCI mainly relies on strong visual stimulus and has limited coding features,the study carried out the characteristic induction experiment under 4 kinds of stimulation parameters for the purpose of evoking asymmetric visual evoked potential(a VEP),and clarified the multidimensional neural response characteristics of a VEP,meanwhile,designed the mixed coding paradigm of space-code division access(SCDMA),and then implemented the experiment of 7spatial coding paradigms in order to determine the optimal coding paradigm based on SCDMA.The optimal stimulation parameters which located away from the central visual field,had better classification effect and higher comfort,as well as the optimal coding paradigm based on SCDMA were determined.Aiming at the problem of low decoding efficiency of ERP with strong noise,wide scope and great differences,the study implemented many kinds of ERP induction experiments and proposed a universal decoding algorithm named discriminative canonical pattern matching(DCPM)combining the typical pattern spatial filter with discriminant pattern matching.This study compared DCPM with 7 typical classification methods for the single-trial classification of five ERP datasets(a VEP,P300,EPFL,RSVP,m VEP).The results showed that the DCPM outperformed other classifiers for all of the tested datasets,especially with small training sets.It suggests the DCPM has high robustness,strong generalization,and can effectively improve the intra-individual decoding efficiency of ERP-BCI.Aiming at the problem of low decoding efficiency of ERP with great changes in polarity and great individual differences,the study implemented a large sample of ERP induction experiments and proposed a within/cross-subject decoding algorithm named multi window discriminative canonical pattern matching(MWDEPM)using the thought of forward/backward stepwise regression and linear discriminant.The method was applied to the recognition of two kinds of ERP(P300,Err P),and established crossindividual general models of the two ERPs based on multi-person data.The results showed that the MWDEPM outperformed DCPM,and the accuracy of intra-and crossindividual recognition in P300 character recognition was improved by 8.18% and 3.65%on average,which proved that MWDEPM can further improve the intra-and crossindividual decoding efficiency of ERP.Aiming at the problem that ERP-BCI is difficult to identify and application weak EEG,the study implemented 3 kinds of ERP(a VEP,P300,SSVEP)induction experiments and offline/online experiments of a VEP-speller.The visual resources utilization was quantitatively compared using the retina-cortex mapping,the multi template decoding algorithm was designed,and the BCI-speller based on weak stimulation appeared in peripheral vision was established.The online information transfer rates achieved to 63.33 bits/min,and the recognition and application of extremely weak a VEP feature(<1?V)was realized.In summary,the reseach developed a new BCI encoding paradigm based on multidimensional EEG,and proposed a new BCI decoding method based on spatial filtering and feature enhancement.The reseach also implemented a wide variety of ERP experiments,as well as proved that the corfort of paradigm and the robustness of algorithm through the qualitative analysis and quantitative comparison.Finally,a set of new BCI codec method was formed.The results effectively improved the codec efficiency of ERP-BCI,which opens a new path and provides a new method for the development of BCI technology and its application transformation.