Research On Hand Motion Information Identification Technology Based On Surface Electromyography Signal

The sEMG(Surface Electromyography)signal is the superposition of the action potential generated by the motor unit in the muscle fiber on space and time when the muscle is excited.And it contains a lot of human motion information.By analyzing the signals generated by the upper arm muscles during the movements of hand,gestures can be recognized and hand force can be estimated.This article mainly does research on hand motion information identification technology based on s EMG.The main contents are:(1)The basic theories and methods of hand information acquisition based on s EMG include the acquisition of s EMG and force signal,feature extraction,active segment detection,training and testing for classification or fitting model.After introducing these,experimental systems for recognizing gestures and estimating fingertip clamping force based on s EMG are designed and constructed.(2)6 types of target gestures is recognized by using BPNN(Back Propagation Neural Network)and SVM(Support Vector Machine).Based on this,an optimal feature combination selection method based on ANOVA(Analysis of Variance)is proposed.After obtaining the optimal feature combination through this method,a classification model based on it is constructed to realize online gesture recognition.And a bionic multi-degree of freedom upper limb is controlled according to the results of online gesture recognition.The results of the experiment show that the feature combination has a significant effect on the rate of gesture recognition.The optimal feature combination is the zero crossing and integral s EMG whose highest gesture recognition rate is 95.4%.On the whole,BPNN is better than SVM.(3)The clamping force between the tip of thumb and forefinger when the arm is in different postures is estimated by using BPNN and GRNN(Generalized Regression Neural Network).The factors which influence the force estimation are analyzed,including the position of the arm,the length of window for feature extraction,combinations of different positions where s EMG signals acquire,combinations of different features extracted from s EMG and fitting algorithms.The results of experiment show that the generally reasonable window length for feature extraction is 600-800 points(0.6s-0.8s)when the arm is in different postures.In this range,GRNN is better than BPNN.When the window length is 600 points and the fitting algorithm is GRNN,the force estimation is the best when the arm posture is horizontal.And the RMS(Root Mean Square)between the estimated force value and the actual value is 0.33 N,the MAE(Mean of Absolute Error)is 0.18 N,and the correlation coefficient is 99.8%.