Research On Natural Interactive Motion Intention Recognition Method Based On Bioelectrical Signal

With the development of smart devices,the natural human machine interaction(nHMI)method,based on bioelectric signals,has become a hot issue in the field of human-computer interaction research.Bioelectrical signal interaction,a new type of human-computer interaction,has developed rapidly in the past two decades.Based on electroencephalogram(EEG)and surface electromyogram(sEMG),it has the advantages of highly coupling and mutual causation.Therefore,it has a bright future in the field of human-computer interaction in the future.In this paper,the brain intent recognition and upper limb gesture gesture recognition are researched,which are the key technical issues of the natural interactive control problem using EEG and sEMG signals as the medium.The specific work is as follows:(1)EEG signal acquisition scheme design and system implementation.By analyzing the EEG signal and the distribution of cerebral cortex function,the electrode placement scheme and motion activation paradigm for four types of motor imaging are designed.In order to provide effective data acquisition method for the research of natural interactive control system based on EEG signal,the EEG signal of motor imaging of three experimental subjects is collected.(2)Adaptive removal of electrooculogram artifacts in EEG signals.The ICA algorithm cannot adaptively select the components of the electrooculogram artifacts and the CCICA algorithm requires additional electrooculogram channels.Then,based on EMD-CCICA that automatically removes electrooculogram artifacts,a noise reduction algorithm is proposed.Using the data set of the "Fourth BCI Contest",the performance of the EMD-CCICA algorithm is verified.The results show that the noise reduction algorithm proposed in this paper can adaptively remove the electro-oculus artifacts in the EEG signal.(3)Multi-domain feature extraction and pattern classification of EEG signals.Considering the weak energy and strong randomness of EEG signal,multi-domain features such as time domain,frequency domain,time-frequency domain and space domain,and 6 machine learning classification algorithms are analyzed.Using the "3rd BCI Contest" data set for simulation verification,the four types of sports intent have 86.5% recognition accuracy.When applying the parameters to the EEG data collected independently,the highest intent recognition rate among the objects is 81.3%,and the average intent recognition result is 68.3%.The results show that the multi-domain eigenvalues combined with the SVM algorithm have the best classification performance.It can recognize four types of motion intentions.(4)16 kinds of gesture recognition of sEMG signal of upper arm.By analyzing the signal characteristics of the sEMG signal,a feature extraction method based on sparse representation and time-domain fusion is used.It successfully recognizes 16 kinds of gestures.Among the 16 gestures of 3 experimental subjects,the highest recognition rate is 98.4%,and the average recognition rate is 97.1%.