| Muscle strength estimation can reflect the body's movement intention,limb move-ment parameters,muscle health status,and fatigue level.It has wide application value and important theoretical significance in many fields such as exercise guidance,muscle disease diagnosis,rehabilitation status evaluation,and human-computer interaction.In order to solve the problem of muscle strength estimation of knee joint motion,in this dissertation,the quadriceps femoris of knee joint is taken as the research object,and a method of human muscle strength estimation based on multi-channel Mechanomyo-gram(MMG)signal is proposed,which mainly focuses on four aspects:MMG signal acquisition,signal processing,feature extraction and regression model construction.The results show it can further improve the accuracy of muscle strength estimation and promote the application of muscle strength estimation in engineering practice.The main work and innovations of the dissertation are as follows:(1)The principle of MMG signal generation and its correlation with muscle strength were discussed,and the main factors determining muscle strength were analyzed.A multi-channel MMG signal acquisition device was designed,which improved the in-stallation method of the MMG sensor Compared with electromyography and elec-troencephalogram signals,this device could acquire the original MMG signal through clothing,which has better robustness to skin impedance and makes the acquisition of MMG signal more flexible and convenient.(2)A signal processing method based on intrinsic mode function was proposed by analyzing the noise and artifact of MMG signals.This method uses multivariate empirical mode decomposition algorithm to decompose multi-channel MMG signals to a series of intrinsic mode functions;then,white noise and motion artifacts are re-moved respectively according to the autocorrelation function and energy distribution characteristics of each intrinsic mode function,and pure MMG information is obtained.This method realizes a lower level of distortion introduced in the MMG signal after the suppression of white noise and motion artifacts and has better adaptive capabilities for different groups of people.(3)The characteristics of time domain,frequency domain,entropy,correlation and instantaneous frequency of MMG signals were studied.A feature extraction method based on adaptive length of the sliding window was proposed to improve the accuracy of feature representation.The method first obtains the instantaneous frequency and in-stantaneous energy information of the MMG signal using the Hilbert-Huang transform;then,the sliding window length adjusts by the change of instantaneous energy due to the correlation between the instantaneous energy and the muscle strength.The results show the adaptive sliding window can better capture the change of muscle strength and further improve the accuracy of the constructed model.(4)A muscle force estimation regression model based on multi-channel MMG signals was constructed,which achieves a continuous estimation of the contraction force of the quadriceps muscle.The input of the regression model is composed of 12 features,which is mean absolute value(MAV),mean power frequency(MPF),sample entropy(SampEn)and the correlation coefficient between the three MMG signal channels;then the relevance vector machine(RVM)algorithm of the sparse Bayesian model is used to establish the MMG-force regression model.Finally,the accuracy of the muscle strength estimation regression model was verified by the isometric contraction experiment of the quadriceps.The results show that the proposed method achieves a continuous estimation of muscle strength.The root mean square error(RMSE)of the estimated force value and the actual force value is 8.7%MVC,and the coefficient of determination(R2)is 0.817,which is superior to most muscle strength estimation based on sEMG and MMG sig-nals.This method has advantages in acquiring signal hardware requirements,installa-tion methods,and robustness to skin impedance. |
PDF Full Text Request