Research On Human Lower Limb Movement Intention Recognition Technology Based On Surface Electromyography

In today's society,with the increasing population of hemiplegic patients caused by aging and cerebrovascular diseases,exoskeleton robots that can provide wearers with functions such as exercise assistance and rehabilitation training have become hotspots,but the interaction between humans and exoskeleton robots need to actively understand human motion intentions,and traditional human-machine interaction based on program control cannot meet the needs.Surface electromyography(sEMG)signals,as electrical signals generated by the skin surface during muscle movement,contain a wealth of human motion information,but the technology for decoding human motion intentions from sEMG signals remains to be further studied.In this paper,the kinematic intention of the knee joint is characterized by the joint angle,and the sEMG signal and motion information acquisition platform for the lower limbs is established.Based on the pre-processing of the sEMG signal,two types of problems are proposed based on frequency estimation and knowledge base and feature matching Knee joint angle predictor to recognize the intention of lower limb movement of the human body.The specific work is as follows:(1)Establishment of software and hardware platform for lower limb sEMG and motion information collection.Hardware:Based on the study of the human knee motion mechanism,a three-channel lower limb sEMG signal acquisition platform was established using differential patch electrodes,an A/D conversion circuit board,and the Arduino Mega 2560 control board;a six-axis attitude gyro was used for lower limb motion information Instrument sensor for acquisition.On the software:The MATLAB Arduino support package is used to communicate with the Arduino control board,and the M file is written in MATLAB to complete the sEMG signal acquisition;the knee motion angle can be displayed and recorded in real time on the host computer software.Finally,the experimental scheme was designed and the above platform was used to complete the acquisition of sEMG signals and knee angle data.(2)sEMG pretreatment.First,for the situation where sEMG is generated ahead of the actual muscle movement,a TKE operator is used to detect the starting point of sEMG.Then,in order to solve the problem that the high-frequency part of the signal is lost in the existing noise reduction methods,a noise reduction method based on the combination of empirical mode decomposition and wavelet threshold noise reduction is proposed,and the threshold function in wavelet threshold noise reduction is improved.The hyperbolic threshold function proposed in this paper combines the advantages of traditional soft and hard threshold functions and requires fewer parameters to be adjusted than other threshold functions,and simulations verify the effectiveness of the proposed denoising algorithm.Finally,a muscle activation model is established to obtain the mapping relationship between sEMG and muscle activation.(3)Design of knee joint angle predictor based on frequency estimation.Aiming at the actual scenes of training movements that require different frequencies at different stages of hemiplegia rehabilitation,and the existing angle estimation model's prediction accuracy decreases under the frequency of lower limb motion,a knee joint angle predictor based on frequency estimation is proposed.Finally,eight groups of sEMG signals and joint angles of four subjects at two motion frequencies were collected simultaneously,and the proposed knee joint angle predictor was simulated and verified using MATLAB.(4)Design of knee joint angle predictor based on knowledge base and feature matching.First,the study found that the sEMG feature is relatively stable,and the feature vector constructed by it has instability,which will affect the performance of the angle predictor.Then,for this situation,the method of combining K-means and Gaussian mixture model(GMM)was used to cluster the feature space,and the features in each cluster were trained for their local angle predictors.Then,in the online prediction process,feature matching is performed first,and the local angle predictor of the matching cluster is obtained to perform knee joint angle prediction.Finally,simulation experiments were performed to verify the effectiveness of the proposed knee joint angle predictor based on knowledge base and feature matching.