Research On Continuous Motion Estimation Of Hand And Wrist Joints Based On Surface Electromyography And Acceleration Signals

Surface electromyography(s EMG)has attracted much attention in the field of humancomputer interaction.In the research of myoelectric prosthetic hand control,myoelectric control based on pattern recognition can only complete the switching of limited actions,while myoelectric control based on continuous motion estimation can decode s EMG signals through supervised regression algorithm to estimate the angle of multiple joints of hands and wrists,this method can better describe the flexibility and continuity of the multi degree of freedom movement of the human hand joints.However,the s EMG signals have some problems such as non-stationarity and many interference factors,which lead to abnormal values and jitters in the estimated joint angles,and pose challenges to the safety and coordination of myoelectric control.In order to propose continuous motion estimation methods with higher accuracy and better stability,the main works are as follows:In this paper,the continuous estimation of three degrees of freedom of wrist joints is studied.In order to achieve accurate wrist continuous motion estimation and improve the problem of joint estimation jitter caused by s EMG non-stationary,this paper adopts robust s EMG features and integrates acceleration information to improve the accuracy and stability of joint estimation.In this paper,three kinds of s EMG features and three kinds of s EMG features fused with acceleration information are selected to complete the model training and testing in support vector regression and XGBoost algorithm,the results show that using support vector regression algorithm combined with the fusion feature of average acceleration feature and time dependent power spectrum descriptors feature of s EMG can achieve the highest estimation accuracy of joint angles,the Pearson correlation coefficient is 0.95 and root mean square error is 7.6°,which achieves the highest joint angle estimation accuracy among various combination methods.Moreover,compared with the s EMG features,the fusion features significantly reduce the abnormal jitter of the estimated angle and improve the estimation accuracy,which provides a flexible and continuous myoelectric control scheme for the three degrees of freedom of the wrist.Moreover,this paper studies the continuous estimation method of hand joint angles.The deep learning algorithm based on recurrent neural network is often used to extract the internal timing features of s EMG,so as to improve the accuracy of joint estimation.Aiming at the problems of recurrent neural networks,such as large number of model parameters,difficulty in parallelization,and difficulty in modeling long-distance dependencies,this paper proposes a deep learning algorithm based on convolution and attention mechanism is proposed to mine s EMG timing information instead of recurrent neural networks.The algorithm has small parameters,can be parallelized and high estimation accuracy.Under six grasping movements of 15 subjects,the average Pearson correlation coefficient,root mean square error and determination coefficient are 0.88,8.496° and 0.762 respectively,which are higher than the other two comparison algorithms,and proves the effectiveness of the algorithm.In addition,in order to further improve the accuracy of hand multi joint continuous motion estimation,the s EMG features integrated with acceleration information are sent into the algorithm proposed in this paper for testing.By analyzing the regression performance index and observing the fitting curve,it can be seen that integrating the acceleration information into the s EMG feature can further improve the accuracy and stability of hand multi joint continuous motion estimation of supervised learning algorithm,it provides a more intuitive myoelectric continuous control scheme for myoelectric prosthetic hand.