Research On Characteristics And Classification Of Hybrid Brain-computer Interface Signal Based On Lower Limb MI And SSSEP

At present,the single-mode brain-computer interface system cannot meet the current control requirements.Various useful features can be fused together to form a hybrid brain-computer interface system to achieve accurate classification and feature research.In this paper,a hybrid brain-machine interface based on lower limb MI and SSSEP was studied.During the execution of movement imagination task,somatosensory potential was induced by electric stimulation,and SSSEP features and MI features were obtained at the same time.MI features and SSSEP features were combined to decode the system movement intention.Use the FBCSP algorithm to fuse the two types of features together,and send them to the SVM and KNN classifiers to realize the classification and recognition of left foot movement imagination and right foot movement imagination.Three experimental conditions,Hybrid paradigm,MI paradigm and SA paradigm,were designed in this paper.The Hybrid paradigm added somatosensory electrical stimulation that induced the characteristics of SSSEP on the basis of the motion imagination paradigm,and imagined the movement of left foot or right foot with or without stimulation.The selective attention(SA)paradigm requires no imaginative activity,and only the subjects are required to selectively pay attention to the somatosensory electrical stimulation sensation of a certain foot.Observing the characteristic distribution of the EEG signals collected in the three paradigms in the time-frequency domain and the spatial domain confirmed that reasonable steady-state somatosensory evoked potential characteristics can be induced when stimulating the posterior tibial nerve of the ankle.At the same time,the differences in the characteristics of the left foot somatosensory stimulation and the right foot somatosensory stimulation were compared.The SSSEP frequency band set in the experiment did not coincide with the frequency band of MI features,so FBCSP was selected to realize feature fusion,and SVM and KNN algorithms were used to classify EEG signals during motor imagination.The results show that when SVM is used,the average classification accuracy of lower limb motor imaging after feature fusion can reach 78.6%,which is about 15% higher than the result of classification based on MI features alone;when using KNN,the accuracy rate is 84.62%,which was 18.75% higher than that without feature fusion.Comparing the accuracy,ROC curve and AUC value of the two classifiers when classifying the same kind of signal,the results show that the KNN classifier is better than the SVM classifier in processing motor image recognition and classification.However,the effect of classification for SSSEP characteristics in the selective attention paradigm is slightly worse than that of SVM.In addition,real-time classification and recognition of lower extremity motor imagination is realized in this paper,and real-time transmission,processing and feedback of EEG classification results are completed in the real-time hybrid brain-computer interface system based on lower extremity MI and SSSEP.The experimental results showed that the average classification accuracy of the seven subjects was 71.07%.