Research On High-Density Myoelectric Control Method Based On Pattern Recognition And Muscle Force Estimation

Myoelectric control is a technique that interprets the skeletal muscle contraction and joint state reflected by Electromyography(EMG)signal into physical movement intention,and controls external agent as machine command.It has been widely applied in prosthetic control,rehabilitation medicine,etc.Meanwhile,it has potential commercial value in the fields of human-computer interaction,electronic consumption,etc.Surface electromyography(sEMG)is often used as the driving signal of myoelectric control due to the advantages of non-invasiveness,safety,and portability.In particular,high-density surface electromyography(HD-sEMG)which can provide much information of muscle activation and fine motion control is widely used in myoelectric control.With modern information technology and electronic technology,myoelectric control has formed three major control strategies:threshold control,proportion control,and pattern recognition control.The natural human motion control strategy can be regarded as the parallel implementation of multivariant pattern recognition control and proportion control.The myoelectric control scheme can provide the recognition of intention and estimation of control force synchronously with good practical application value.However,myoelectric control is faced with many challenges in realistic scenarios.The research of myoelectric control based on synchronous myoelectric pattern recognition and muscle force estimation is still in its infancy,which is far from the natural control goal of actual limb function.Aiming to provide a series of feasible solutions for the realization of a robust and natural myoelectric control system,the HD-sEMG is taken as the signal acquisition method.Some key problems of the two core technologies,myoelectric pattern recognition and EMG-force estimation are deeply investigated in this paper.The main research work and innovation of this paper are as follows:(1)The research on dynamic autonomic contraction force estimation based on HD-sEMG and nonnegative matrix factorization(NMF)algorithm.To overcome the influence of the heterogeneity of skeletal muscle activation and unknown muscle contraction mode on the precision of muscle force estimation,a new scheme of muscle force estimation based on channel optimization is proposed.Based on the NMF algorithm,the HD-EMG is decomposed into the activation pattern matrix and time-varying coefficient matrix,the muscle activation region is located by the vector distribution of the activation pattern.The analysis of activation intensity is carried out with the time-varying coefficient curve.A new channel optimization and model excitation signal extraction method based on activation intensity ratio is proposed creatively.At the same time,a static contraction data acquisition and model training scheme,which includes diverse muscle activation modes and force levels,is designed to enhance the generalization ability of the force estimation model.In this study,the flexor of the upper arm is taken as the research object,10 participants are recruited to carry out the data acquisition and muscle force estimation experiments of the dynamic autonomic contraction task of forearm rotation.The results show that the framework can effectively locate the muscle activation region related to the dynamic autonomic contraction task and obtain accurate force estimation results.Meanwhile,the static sinusoidal mode force is proved to be a potential model calibration force mode.(2)The research on muscle force estimation based on the fusion of multi-muscle sEMG information.To solve the problem of the multi-muscle sEMG information fusion in muscle force estimation,a research idea is proposed to realize high precision comprehensive force estimation by fusing the sEMG information of multiple muscles with deep belief net(DBN).And the influence of different muscle groups on the total force is evaluated by mean impact value(MIV).In this study,some main muscles of the forearm and upper arm are taken as objects,13 participants are recruited to carry out the data acquisition and comprehensive contraction force estimation experiments of two representative multi-muscle contraction tasks,elbow flexion and palm pressing.The results show that the critical muscles with high activation level can better track the muscle force in the multi-muscle contraction task.It is more suitable for the high precision force estimation than muscle combination.Besides,according to the order of MIV,the priority of muscle group can be obtained,which can provide a reference for the selection of target muscle in muscle force estimation.(3)The research on electrode calibration in myoelectric pattern recognition.To overcome the influence of electrode shift on myoelectric pattern recognition accuracy,an adaptive electrode calibration method based on muscle core activation region is proposed.Based on the fast independent component analysis(FastICA)algorithm,the HD-sEMG signal is decomposed into the source signal matrix and mixed coefficient matrix.The mixed coefficient vector corresponding to the largest energy of the source signal is selected as the major pattern.The muscle core activation region is extracted by traversing the major pattern with a sliding window.The core activation region is aligned by unsupervised ways to realize the adaptive electrode calibration.In this study,9 types of gestures are taken as examples,the data acquisition and analysis of two experiments,the supervised electrode shift(S-ES)experiments and the unsupervised non-rotation electrode shift(UN-ES)experiments are carried out on 11 participants.The results show that the proposed calibration algorithm can overcome the influence of electrode shift on myoelectric pattern recognition accuracy and can reduce the user training burden of myoelectric control system.(4)The research on pattern recognition and muscle force estimation for myoelectric control.A myoelectric control scheme supporting pattern recognition and muscle force estimation is proposed to realize the myoelectric control mode of synchronous motion intention recognition and control force estimation.The synchronous prediction of gesture category and instantaneous force is realized by the multi-task learning(MTL)technique.Especially,a post-processing algorithm based on metric learning is proposed to overcome the influence of force variation on the accuracy of gesture recognition.In this study,11 participants are recruited to carry out the synchronous myoelectric pattern recognition and muscle force estimation experiments on 11 types of gestures in different life scenes.The results show that the proposed framework can effectively support the myoelectric control of synchronous pattern recognition and muscle force estimation and meet the real-time requirements of the myoelectric control system in the response delay.