A Multi-class Motor Imagery BCI Study Based On FNIRS

Brain-computer Interface(BCI)is a cutting-edge human-computer interaction technology,which can realize the direct information exchange between the brain and external devices.In recent years,Functional Near-Infrared Spectroscopy(fNIRS)has become an important new method to develop BCI by virtue of its many advantages.BCI based on motor imagery(MI)has great application value in many fields such as military,rehabilitation medicine,entertainment and intelligent robot control.However,most of the studies on fNIRS-BCI based on motor imagery are still in the stage of 2 or 3 classes,and there are few studies with more than 3 classes.In order to increase the number of control commands for fNIRS-BCI,this paper conducts a multi-class pattern recognition study on the motor imagery tasks in fNIRS-BCI.We designed a new experimental paradigm with 4-class motor imagery tasks(left-and right-hand motor imagery,walking imagery and rest),and collected fNIRS signal data during the tasks.Then the classification effect was studied based on the three hemodynamic response signal indicators of Hb O,Hb R and Hb T.The results show that these tasks have good separability and are suitable for developing BCI application equipment with practical value.Moreover,the equipment developed based on this has more control commands,more powerful and perfect functions.In addition,the equipment will be more user-friendly through the control of simple and intuitive motor imagery tasks.Besides,this thesis also focuses on the study of hemodynamic response signal indicators in fNIRS-BCI.We proposed a new hemodynamic response signal indicator,the proportion of oxygenated hemoglobin concentration changes in total hemoglobin concentration changes P_{O T}.In order to study the classification effect of P_OT,the three classification algorithms of SVM,LDA and KNN and the eight features(mean,average rectified value,maximum,minimum,variance,kurtosis,root mean square and peak)of P_OT data were combined in pairs,and the combination with the best classification effect is SVM+kurtosis.Moreover,to verify the classification effect of P_OT fully,we made a comparative analysis of the experimental results and those of the above research,and additionally tested and compared the classification effect of the combined data of Hb O and Hb T.Experimental results show that the best average classification accuracy of the new indicator P_OT is 76.82%±3.36%,which exceeded that of the traditional indicators Hb O,Hb T and the combination of the two indicators.It shows that P_OT could be used as an effective signal indicator for classification,which verifies the feasibility of its application in the development of fNIRS-BCI,and also adds a new alternative for future research.