Research On The Prediction Method Of Lower Extremity Knee Joint Angle Based On SEMG

With the development of rehabilitation robots in the field of medical rehabilitation,human motion classification based on surface electromyography(sEMG)has been widely used in human-computer interaction,and its motion information estimation results are often used as control instructions for lower limb rehabilitation equipment.Since the accurate prediction of the continuous joint angles of the lower limbs can significantly improve the performance of human-machine coordination,related scientific research work has always been a research hotspot in this field.However,considering the complex interactions between muscle groups during lower extremity movement,in this paper,based on the collected sEMG data,a novel filtering process and feature extraction algorithm were studied,and the prediction of knee joint angle for continuous lower extremity movement was explored.The method finally improves the accuracy and stability of the joint angle prediction model.The main research contents are as follows:(1)Acquisition and preprocessing of sEMG signals.By analyzing the relationship between lower extremity muscles and knee joint motion,a synchronous acquisition system for lower extremity motion information was built to collect sEMG signals of four muscles in three motion modes,and to analyze various noise signal sources in the process of sEMG signal acquisition.Finally,an improved wavelet threshold-empirical mode decomposition sEMG denoising experiment is designed.The results show that after denoising by this method,a large amount of noise in the sEMG signal is filtered out,the original characteristics of sEMG are preserved,the quality of sEMG signal is improved,and data preprocessing is prepared for sEMG feature extraction.(2)sEMG signal feature extraction.The generation principle and characteristics of sEMG signals were explored,and three time-domain characteristics with high computational efficiency,such as root mean square value,variance and wavelength,were taken as the research objects.Based on the above feature selection,a novel fusion feature based on time dimension is proposed.The method uses the sliding window method to extract the time advance feature and time delay feature of the sEMG signal.The two features contain the lower limb motion information at different times,and finally the fusion feature can be used as the input of the subsequent knee joint angle prediction model.(3)Construction of knee joint angle prediction model.This paper proposes a knee joint angle prediction model(WOA-Attention-LSTM)that optimizes the Attention-LSTM network using the Whale Optimization Algorithm(WOA).The weight of sEMG features is increased by increasing the Attention mechanism,and the optimal LSTM network parameters are obtained by using WOA.In the experiment,the root mean square error is used as the evaluation index,and the WOA-AttentionLSTM is compared with the LSTM and Attention-LSTM models respectively.The results show that the RMSE value of WOA-Attention-LSTM(average RMSE=1.49448)is lower than that of the control group.The model not only saves the cost of selecting parameters by manual experience,but also effectively improves the accuracy of predicting the knee joint angle of the lower limbs.(4)Model validation and performance analysis.The influence of different features on the prediction accuracy of the model and the stability performance of each model under different motion modes are analyzed and compared.One-way analysis of variance(ANOVA)was used to analyze the significant difference in the root mean square error value of the prediction results of knee joint angle.The results show that the fusion feature proposed in this paper has better advantages in the prediction of lower limb knee joint angle(the root mean square error of the prediction result is extremely significantly smaller than that of TDF-LSTM,p<0.01);The mean absolute error(MAE)was used as the evaluation index to analyze the stability of each model in different motion modes.Experiments show that using WOA-Attention-LSTM has higher stability(average MAE=2.7970)in joint angle prediction under different motion modes.