Continuous Motion Estimation Of Lower Extremity Based On SEMG

Surface electromyography(s EMG)is the physiological signal induced by muscle contraction/relaxation,which contains information related to human motion and can be utilized to recognize the movement intention of the lower limb joint.Because the s EMG signals are rich in information and easy to collect,they are used in human-computer interaction,medical rehabilitation,prosthetic control,and exoskeleton robots.In this dissertation,the continuous motion estimation of the lower extremity joints is performed via the s EMG.The main work is as follows:(1)For noise characteristics of the s EMG signals,a Butterworth filter is proposed to filter the noise,and the de-noised s EMG signals can increase the signal to noise ratio,and improve the signal quality.(2)Five kinds of time-domain features,i.e.,mean,variance,zero-crossing,logarithm,and waveform length,are selected as the objects to be extracted.Compare the differences in the effects of the single feature and the combined features,bring features into the regression fitting model to judge the index of the Pearson correlation coefficient and root mean square error,and select the optimal features.The results show that the combination of the absolute value mean and logarithm is the most suitable.(3)Due to the correlation between the various s EMG features,too many dimensions easily lead to complex model structure and increase calculation time.A method of the principal component analysis(PCA)is presented to reduce the s EMG feature dimension,simplify the model and shorten the training time.(4)According to the characteristics of different algorithms,the regularized extreme learning machine(RELM),BP neural network,and support vector machine(SVM)regression algorithm,respectively,are proposed to construct the models for the knee joint angle estimation.The Pearson correlation coefficient,root mean square error and model training time are as the indicators to compare the accuracy and speed in angle estimation.The results show that the root mean square error and the correlation coefficient of the RELM are similar to SVM,but slightly different to BP neural network.For three algorithms mentioned above,RELM model is the shortest training time,with two orders of magnitude higher than that of the BP neural network model.This dissertation systematically studies the s EMG-based method of continuous motion estimation of the knee joint angles,and completes the signal denoise,feature extraction and knee angle estimation of the s EMG.Comparing to three methods,it is found that the RELM algorithm has some advantages in real time performance,which has a valuable reference for real-time recognition and control of lower limb movement.