Research On The Representation And Prediction Method Of Muscle Fatigue Based On SEMG Signal Of Lower Limbs

As one of the important products of human-computer interaction technology,the exoskeleton has been widely used in military and civilian fields.To improve security collaboration between the exoskeleton and the human,human-computer interaction techniques based on surface electromyography(surface electromyography,s EMG)signals emerged.Muscle fatigue will greatly interfere with the s EMG signal,thereby reducing the stability of human-computer interaction.Therefore,it is necessary to use muscle fatigue as a characteristic parameter to compensate the control system.However,how to properly characterize muscle fatigue is still a major difficulty in related fields.Besides,to improve the flexibility and stability of the exoskeleton,it is necessary to predict the changes in muscle fatigue.Therefore,this article focuses on muscle fatigue characterization and muscle fatigue prediction.Aiming at the problem of poor performance of muscle fatigue characterization due to changes in muscle shape and action differences in isotonic contraction,a detection algorithm based on EMG power and integrated EMG(IEMG)combined with dynamic threshold is proposed.This algorithm uses myoelectric power instead of Zero Crossing Rate(ZCR),which solves the problem of false detection caused by ZCR oscillating up and down in the active segment in the combined dynamic threshold algorithm of IEMG and ZCR;besides,before the threshold is set,the algorithm uses the intermediate value of the maximum and minimum values to intercept the data,solves the problem of unreasonable dynamic threshold setting caused by abnormal values,and reduces the missed detection rate caused by abnormal values.Aiming at the problem of lower limb muscle fatigue characterization,the lower limb muscle contraction is divided into isometric contraction and isotonic contraction.In isometric contraction,it is proposed to use Muscle Fiber Propagation Velocity(MFPV)for fatigue characterization;in isotonic contraction,it is proposed PCA is used to extract the fatigue feature of s EMG time-frequency domain combination features,and perform fatigue characterization according to the contraction area.Compared with the traditional time-frequency domain features,the muscle fatigue features proposed in this article have better sensitivity,individual applicability,and stability.Aiming at the problem of muscle fatigue prediction,a Random Forest(RF)-Gated Recurrent Unit(GRU)muscle fatigue prediction method is proposed.To reduce the interference of the original input characteristic noise on the GRU model,RF is used to predict the changing trend of muscle fatigue characteristics for the first time,and the data after RF correction and fitting is input into GRU,which solves the problem of prediction lag in GRU to a certain extent.Compared with other neural network models,the model proposed in this paper can realize long-term prediction of muscle fatigue characteristics and has the advantages of high accuracy and good generalization.In summary,the contents of this paper are the follow-up in human-computer interaction,the muscle fatigue as a myoelectric control compensation signal parameters to achieve compliance control exoskeleton is of great significance.