Research And Application Of Intention Recognition Of Human Lower Limb Movement Based On Surface Electromechanics

With the gradual maturity of exoskeleton design technology,exoskeleton is not only used for rehabilitation treatment of paralysis and other diseases,but also gradually used to improve workers’ work ability,increase human body strength and endurance.Therefore,the accuracy and robustness of exoskeleton control is a higher challenge.The purpose of this study is to propose action intention recognition and joint angle prediction algorithms based on surface EMG data collection and feature analysis,to achieve accurate prediction of bone and joint motion,and to provide basic data for exoskeleton motion control.First,based on the kinematic characteristics of the lower limb skeletal muscle system,a lower limb kinematics measurement experiment was established.Six single joint movements and three speeds of gait movement experimental tests were performed.The NDI motion capture system and surface electromyography system were used to collect motion tracking data and sEMG signal;then based on the sEMG data obtained by filtering,segmentation and other preprocessing and PCA feature parameter analysis,and this paper calculated the angle history data of three joints.According to the processing and analysis results,the sEMG signal characteristic values that are sensitive to joint motion characteristics are determined.The research results provide a selection basis for intelligent motion intention recognition.In this paper,the LDA,SVM and DNN neural network models are established,and the three models are used to identify the movement intention of the human lower limb single joint.The comparative analysis results show that the DNN network model has superiority in recognition accuracy and training time.Using the DNN network to perform intent recognition training and recognition reliability analysis on the 6 actions,the results show that the recognition accuracy rate reaches 95.2%,and the training time is only 2.17 s.Therefore,this paper proposes a more feasible intent recognition scheme,which can provide a reference for subsequent intent recognition research.At the same time,this paper uses LSTM neural network to learn the sEMG characteristic parameters and motion data to predict the joint angle in gait movement.Through the optimization of the loss function,learning rate,optimizer and dropout value,this paper determines the parameters of the LSTM network.The gait data of three speeds are used to train and predict the joint motion angle of gait in the LSTM network.The results show that the prediction result of each joint angle at each speed has a maximum RMSE value of 1.587 and a maximum error rate of 4.6%.The obtained LSTM model has an ideal prediction effect.This paper proves the feasibility of using LSTM model to predict the joint angle,and gives suggestions for optimal tuning.The motion intent recognition algorithm and the lower limb joint angle prediction algorithm in this paper will provide the exoskeleton control system with a high accuracy and recognition speed control strategy,and contribute to the human-machine coupling and flexibility of the exoskeleton system.