A Muscle Force Estimation Framework Based On Factorization Algorithms And Fast Orthogonal Search Method

As a branch of sports biomechanics,muscle force estimation is of great significance in many research fields,such as artificial prosthesis design,clinical rehabilitation medicine and professional sports training.Surface electromyogram(sEMG)generated by muscular activities is regarded as an important kind of bioelectric signals,which can reflect the muscle contraction state and activation condition.Due to its safe and noninvasive data collecting process,sEMG has become the research focus of muscle force estimation.In recent years,many related researches have applied high-density sEMG grids instead of the conventional single or bipolar separated electrodes to capture heterogeneous muscle activities more comprehensively and to estimate muscle force more precisely.This study proposed a sEMG-force estimation framework based on fast orthogonal search(FOS)method combined with factorization algorithms.To evaluate the feasibility and superiority of the novel framework,twelve healthy male subjects were recruited to perform static variable force elbow flexion movement,during which high-density sEMG from upper limb skeletal muscles and the wrist force signal were collected synchronously.The main research work and research results can be summarized as follows:Firstly,since many skeletal muscles could be subdivided into muscle-tendon units(MTUs),and there were substantial heterogeneities in muscle activation during voluntary contractions,this research collected sEMG signals from upper limb muscles with the aid of high-density electrode grids and explored different activation levels of MTUs within skeletal muscles by means of principal component analysis(PCA),independent component analysis(ICA),and nonnegative matrix factorization(NMF).The experimental results demonstrated that,all three methods successfully extracted different activation patterns and corresponding activation curves of different MTUs during the isometric contraction.Furthermore,compared to the conventional method using the mean value of all sEMG channels' envelopes(Mean-ENVLP),factorization algorithms could extract the activation information which was more similar to the force signal.Secondly,considering the complex relations between sEMG signals and muscle force,a variety of functional forms like linear function,polynomial function,exponential function and sigmoid function were used to map their relationship.The results demonstrated that FOS method could select suitable functions from these candidates to build force estimation model and provided more precise force estimation.Also,this research analyzed the performances of muscle force estimation under conditions of different related muscles,different activate signal extraction method and different force following patterns respectively.It was found that the more muscles that were taken into account,the more precise force estimation was;compared to Mean-ENVLP method,factorization algorithms especially NMF could offer better performance;the force models trained by force patterns with high percent of maximum voluntary contraction had stronger generalizations.Unlike many previous researches which-took a piece of muscle as a whole,this study proposed a novel sEMG-force estimation framework based on fast orthogonal search(FOS)method combined with factorization algorithms,which could deeply explore the activation patterns and activation levels of MTUs within target muscles and significantly improve the muscle force estimation performance.The research results had the potential applied value in the fields of daily exercise guidance,sports assessment and medical rehabilitation.