The Research Of Motion Pattern Recognition And Joint Moment Analysis Of Human Lower Limb Based On SEMG

Surface electromyography signal (sEMG) is an electric wave generated by muscle contraction, which is not only related to physiological characteristics of muscle tissue, but also related to the neural control system, reflecting the activity and functional state of nerve and muscle. sEMG signal is highly correlated in varying degrees to the activity and functional status of muscle. Since different body movements are produced by different muscle contraction pattern and the difference exists in sEMG signal features, the different action mode could be distinguished by analyzing the characteristics of sEMG signal. Furthermore, sEMG signal has been widely used in many fields of clinical medicine, sports medicine, biomedical engineering and so on. The advantages of application in intelligent prostheses has made sEMG signal the ideal signal control of functional electrical stimulation.This paper aimed to identify the intention of sports person and study the relationship between sEMG signals and joint torques through the analysis of the collected sEMG signal processing, which was as the basis for intelligent prosthetic control.The main work includes the following four aspects:(1) This paper designed the experiment to collect sEMG signal of nine channels corresponding to rectus femoris, vastus medialis, vastus lateralis, biceps femoris, semitendinosus, tibialis anterior, medial gastrocnemius, lateral gastrocnemius, soleus muscles respectively, with four lower limb motions including up the stairs, down the stairs, up the slope and down the slope. In order to improve the signal to noise ratio and enhance the recognition performance, the paper chose the wavelet threshold denoising method. After wavelet decomposition, the coefficients of each layer were more than and less than a certain threshold value, and then inversed wavelet coefficients, reconstructing the signal.(2) Extract three feature values, including maximum features of db4 wavelet transform in a certain scale, decomposition coefficient features based on dmey wavelet and singular value features of bior3.1 wavelet transform, from the original sEMG signals. It was found that the effect of different muscles in different feature extraction methods varies. Through the analysis of the eigenvalues of single feature value, fusion features from single feature could make complementary advantages of the single feature and enhance the characterization ability which could improve the accuracy of pattern effectively.(3) BP neural network and Elman neural network were used to identify the characteristics of the human body movement patterns. BP neural network is a kind of multilayer feedforward network based on error inverse propagation training algorithm, and Elman neural network is a dynamic feedback network. Single feature value and fusion feature value were input to BP neural network and Elman neural network respectively. Better results of pattern recognition were obtained through the analysis of recognition results.(4) Input the sEMG signal to calculate muscle force of knee joint to obtain the joint torque, which contributes to foundation for future prosthetic control. This article did research from two aspects of positive and inverse dynamics. Inverse dynamics used kinematics data, force platform and equilibrium equation to obtain joint resultant force moment. Dynamics were first to calculate muscle activation from sEMG signal, calculating muscle length, systolic velocity and force arm using personalized human skeletal muscle model, and then put muscle activation, muscle length and systolic velocity to muscle contraction model to obtain muscle force. Finally, compare and analyze the joint torques obtained by the inverse dynamics and the positive dynamics.