Research On Real-time Classification And Action Switching Based On SEMG Signals Of Lower Limbs

Online and real-time action pattern recognition is of great significance to the new type of human-machine interaction(HMI).Compared with other sensor signals,surface electromyography(sEMG)signals have the characteristic of generating actions ahead of the human body and can greatly improve the real-time performance of action recognition.Therefore,they are widely concerned in the field of rehabilitation medicine and human-computer interaction.However,the EMG signal itself has the properties of non-stationary,non-periodic and chaotic.This also leads to the problem of low accuracy and compatibility of real-time and accuracy in the pattern recognition technology that uses it as the research object under unstable and time-varying online application scenarios.At the same time,in practical applications,the signal capture delay of the action switching part will directly affect the recognition accuracy of the action.Therefore,the research content of this article mainly focuses on the real-time classification and state switching of lower limb movements based on sEMG:This paper proposes an improved energy kernel feature extraction method based on sEMG signal,which is suitable for real-time classification of sEMG signal.This paper evaluates the operational efficiency of the improved energy kernel method through two different experiments,and proves that it has a higher operational efficiency than the traditional energy kernel method.;Compared with the commonly used time and frequency domain feature extraction methods,it has better class separability;in terms of stability,the improved energy core method The feature extraction method is better than the commonly used temporal feature extraction method.Subsequently,a new system framework suitable for sEMG signals was proposed,and a complete acquisition and test system was constructed based on this framework.In order to accurately classify and respond in time,the system variables were studied.The data collected when the subjects performed six different lower limb movements were used to construct the model.An online test was conducted on the subjects,and relatively satisfactory results were obtained.In the online situation,the classification effect is lower than the offline situation,which is analyzed and explored,and hypotheses and solutions are proposed.Finally,a prediction-classification research idea based on multi-channel sEMG signals is proposed,and preliminary exploration experiments are carried out mainly for the study of action switching in continuous motion.Combining the predictive model with the diagnostic model and using the simulated online method to test the data collected by the subject,the average handover delay obtained is within an acceptable range.Experiments have confirmed the feasibility of this research idea and laid the foundation for future research on multiple continuous action switching.The experimental results show that the research done in this paper has certain reference value and significance for online application research based on sEMG signals.