Research On Key Technologies For Multi-channel Surface Electromyography Feature Extraction

As a kind of non-invasive,safe and easily available biological signal,surface Electromyography(sEMG)is the electrophysiological source of muscle activity,and is widely used in the analysis and research of fields related to biomechanics of sports.Researches have shown that there was heterogeneous activation within skeletal muscles to accommodate different mechanical tasks,traditional sEMG acquisition tools using single or several pairs of electrodes can't capture the muscle activation heterogeneity information,thus failed to accurate muscle activation analysis and muscle force estimation.Multi-channel sEMG acquisition technique like high density sEMG(HD-sEMG)can capture high-resolution spatio-temporal sEMG and is the mainstream data acquisition tool for force estimation and movement analysis.Though the multichannel sEMG can obtain abundant information,it can also bring more redundant and intrusive information,so it should be processed in depth for practical application.Focusing on information fusion requirements for sEMG-muscle-force estimation and instability and incompleteness problems of muscle synergy extraction in crawling movement,this paper studied the multi-channel sEMG feature extraction in depth based on blind source separation algorithms(BSS)like principal component analysis(PCA),independent component analysis(ICA),nonnegative matrix factorization(NMF),factor analysis(FA),canonical correlation analysis(CCA).The main research contents and achievements of this paper can be summarized as follows:(1)The feature extraction technologies for array high density sEMG(HD-sEMG)-muscle force estimation.This paper divided the input feature(input signal)extraction of the muscle force estimation model into three procedures:signal preprocessing,data fusion and filter design.In the signal preprocessing part,noisy principal components were identified based on Signal Noise Ratio(SNR)or CCA algorithm,the non-noise principal components were used to reconstruct the denoised HD-sEMG.In the data fusion part,the heterogeneity information was analyzed based on BSS algorithms,a series of BSS-based HD-sEMG fusion frameworks were proposed and achieved,these frameworks contained ICA_Clustering framework based on ICA and weight combination clustering,PCA_Selecting framework based on heterogeneity maximization,PCA_FixedPoint framework based on fast multiple principal components fusion,and novel CCA-based source signals fusion framework(PCA-CCA?ICA-CCA?CCA-CCA and FA-CCA).In the filter design part,aiming at the filter parameter selection difficulties,a filter parameters optimization scheme using the EMG modulation model was proposed and an effective filter parameters optimization algorithm based on kurtosis was designed.Taking a static isometric elbow flexion task with linear growth-sustaining force pattern as research target,10 subjects were recruited in this study,subjects performed force tasks included 20%maximum voluntary force(MVC),40%MVC and 60%MVC and the data was recorded.The data analysis results showed:the proposed noise source identification method was effective,and the signal quality of those low-SNR channels was significantly improved by principal component reconstruction(p<0.05);data fusion frameworks based on heterogeneity can capture muscle activation information well,pass the fused signal through the designed kurtosis guided low-pass filter,the high quality input signal can be obtained(the Pearson correlation coefficient between the input signal and the measured force can reach up to more than 0.95);at last,import the input signal into the polynomial force model and high-precision force estimation result was obtained(root mean square difference(RMSD)less than 10%).(2)Study on the stable and effective crawling movement muscle synergy feature extraction.In view of the big difference in muscle activation in crawling movement,two types of difficulties in muscle synergy extraction by traditional NMF were summarized as the instability and separation incompleteness issues,two ways to improve NMF are explored.Firstly,to solve the problem of instability of extracted muscle synergy,the method of setting initial value of muscle synergy based on PCA was adopted.Secondly,aiming at the problem of incomplete muscle synergy separation,HNMF based on energy stratification and UNMF based on synergy uncorrelated constraint were proposed.In this study,6 subjects were recruited to perform crawling movements,the data analysis results showed:the method of PCA-based specifying the initial value of synergy significantly improved the stability of muscle synergy extraction based on NMF algorithm,in most cases,the synergy can converge to the same solution while the algorithm was separately run several times;The high-energy layer signals in HNMF mainly represented the information of a few strongly activated muscles,which cannot represent the characteristics of crawling movement,the low-energy layer signals represent the information of most weakly activated muscles,which can fully represent the characteristics of crawling movement;The high and low energy layer division of HNMF reduced the influence on synergy extraction caused by the big difference of muscle activation,and the synergy separation level was high;Through an unrelated constraint,UNMF adds regular terms to the iteration criterion of NMF,which significantly improves the degree of synergy separation and obtains a relatively sparse synergy structure,which had potential neural control significance.In conclusion,this paper systematically studies the multi-channel sEMG feature extraction based on blind source separation algorithm,the main innovations are as follows:(1)Considering the heterogeneity of muscle activation,a series of framework based on blind source separation was proposed and applied to signal preprocessing and data fusion;(2)a novel filter parameters optimization method based on EMG modulating model was proposed,high quality input feature of muscle force model for HD-sEMG-force estimation was obtained;(3)considering the muscle synergy extraction problems in the crawling movement caused by big muscle activation difference,try to improve NMF algorithm from different angles,achieved stable and effective muscle synergy extraction in crawling movement.The research results can be used in prosthesis control,human-computer interface design,rehabilitation training,motion evaluation and other related fields.