Font Size: a A A

The Application Of Hybrid Brain Computer Interface In Rehabilitation

Posted on:2018-05-17Degree:MasterType:Thesis
Country:ChinaCandidate:J Q ZhuFull Text:PDF
GTID:2348330515466702Subject:Control Science and Engineering
Abstract/Summary:PDF Full Text Request
Brain-computer interface(BCI)builds a kind of new way of communication between the brain and the outside world through the use of a computer or other external devices,which doesn't depend on the human body normal nerve and muscle tissue.It has a broad application prospect in assistive rehabilitation,intelligent life,entertainment fields and so on.Using motor imagery(MI)and P300 EEG as the breakthrough point,the paper builds the hybrid BCI system from the control for the rehabilitation training robot for rehabilitation training.The paper mainly does the following works:(1)Classic common spatial pattern(CSP)is used for the binary-class feature extraction,it can be applied to the multi-class problem through the extension of CSP.In the paper,we studied the multi-class method of one versus rest CSP(OVR-CSP)first.Since the performance of OVR-CSP depends on the selected band,When performing classification on the feature which is filtered on an inappropriate frequency,the classification accuracy is generally poor.Further research was carried out on the method of filter bank common spatial pattern in which the division of frequency bands was fixed.Although this method could further improve the classification accuracy,it was still far below the binary-class problem.(2)Contraposing the problem of low accuracy of common multi-class CSP algorithm,a two-level feature extraction method for multi-class variable bands motor imagery electroencephalogram(EEG)was proposed by employing the stacked denoising autoencoders(SDA).Firstly,the OVR-CSP was adopted to convert EEG into low dimensional space in which the discrimination of signal variances was maximized.Secondly,SDA network was used to extract the higher level abstract features which could characterize the category attributes more effectively.Thirdly,using Relief F method for feature selection of the obtained feature,and then the best feature was selected which had the maximal weight.Finally,the motor imagery tasks were classified with Softmax classifier.In the classification experiment with four-class motor imagery tasks from data sets 2a of the BCI competition IV,this method achieved the average Kappa value of 0.70.The results show that the proposed method is effective and robust.(3)Based on the research of the existing P300 paradigm,we put forward a kind of variable probability stimulus paradigm(VPP),in which the characters are uneven distribution and their density decrease from middle to both sides in turn.The character recognition is divided into two steps,the lines are flashed randomly to confirm the target's line first,and then the characters of selected line are also flashed randomly to confirm the target's character.The processing results of the signal acquired in laboratory show that the information transmission rate of VPP is increased by 10% than the stimulus paradigm based on region,proving the feasibility of the proposed paradigm.(4)In order to realize the multidimensional control of the rehabilitation robot,a hybrid BCI control strategy was designed based on motor imagery and P300 signal in the paper.We used P300 as a "switch" of the two different signals,VPP with game icon was used for the control panel of game menu,and as a control signal of the robot,MI could realize the rehabilitation training of patient.The result of the simulation control which is based on the experiment of off-line data acquisition shows the feasibility of this system.
Keywords/Search Tags:brain-computer interface, feature extraction, motor imagery, P300, stimulus paradigm, common spatial pattern, stacked denoising autoencoders
PDF Full Text Request
Related items