Research On Recognition Method Of Human Lower Extremity Movement Intention Based On SEMG Signal

The recognition of human action intention is of great significance to the human-robot interaction(HRI)of the rehabilitation exoskeleton.Because Surface Electromyography(sEMG)has the characteristics of being generated in advance of human motion and contains rich human motion information,the research of rehabilitation exoskeleton using sEMG signal as a signal source has received more and more attention.However,in the research process of applying sEMG signal to rehabilitation exoskeleton,due to the non-stationary and non-linear characteristics of sEMG signal,it is still difficult to extract the features and recognize the intention.Therefore,the research content of this article is mainly focused on the research of movement intention recognition based on sEMG signal of lower limbs:First,in order to obtain the raw sEMG signal with less redundant data and realize the acquisition of multi-channel signals,this paper proposes a multi-segment sEMG acquisition method based on Adaptive Energy Diffiential Product(AEDP).In the acquisition method,the first-level threshold for judging the starting point of the signal is set by calculating the energy differential product value of the signal,and the second-level threshold for judging the starting point of the signal is set by the sequential statistics method.Experiments have shown that the acquisition method can determine the starting point of the raw sEMG signal through dual thresholds.At the same time,the deviation between the starting point of the algorithm detection and the starting point of the real signal can be reduced,and the acquisition of multi-channel sEMG signals that are not synchronized at the starting time and duration can be realized.Subsequently,aiming at the problem of obtaining high-class separability features of the raw sEMG signal a feature extraction method of S-transformation’s Singular Value-Concentration Meature(ST-SCM)is proposed.This method improves the classification accuracy of human motion intention recognition by calculating the concentration characteristics of the original sEMG signal.And through performance verification experiments,it can be known that ST-SCM has better inter-class similarity in the evaluation of Spearman’s correlation coefficient;In the action classification of RF,its classification accuracy rate has superiority in multiple feature algorithms;In terms of computing time,although ST-SCM has a higher running time than the time domain algorithm,it satisfies the calculation requirements for short-term EMG signal data and belongs to a faster-running characteristic algorithm.Finally,in order to achieve high-accuracy recognition of motion intentions,based on the ST-SCM characteristics of sEMG signals,the classifiers suitable for each segmented signal were screened,and an intent recognition strategy based on the weighted threshold voting method of multi-segmented signals was developed.Through the classification experiment of the test data set,it can be seen that the multi-segment signal’s weighted threshold voting method is better than the single-segment signal in the classification accuracy.The classification of the six types movements for single joint can be used to train the wearer to perform preliminary exercises of the lower limbs and complete the muscle memory of simple movements.In addition,this paper explores the intent recognition of complex multi-joint leg motions,verifies the applicability of this intent recognition strategy to other types of actions,and has achieved good actual online test results.The main significance of the intent recognition of the complex multi-joint in the lower limbs is to help the wearer of the rehabilitation exoskeleton complete further basic movement training.In summary,this article has a certain application value for the research on human-computer interaction of rehabilitation exoskeleton based on sEMG signal.