Research On The Application Of SEMG-based Action Recognition In Upper Limb Rehabilitation Training

At present,there are more than 100 million people with upper limb dysfunction due to diseases,accidents and aging population in China,so,using upper limb rehabilitation robots to assist upper limb rehabilitation training has attracted increasing attention.In order to increase the active participation of patients in rehabilitation phase,the motion recognition based on sEMG signals has become the emphasis research that rehabilitation robot assists upper limb rehabilitation training.However,the current research on upper limb motion recognition based on sEMG has the problems of simple recognition motion?single joint motion?,few types of motion recognition?4-8 types?,and no online verification.To make up for the shortage of existing research,in this paper,14 kinds of complex movements of upper limb multi-joint are identified and classified offline through the process of 8-channel sEMG data collection,pre-processing,feature extraction and fusion,recognition classification,etc.Finally,the offline action recognition algorithm proposed in this paper is used for online verification of M2 upper limb rehabilitation training platform.The main research work of this article is as follows:?1?Collection of sEMG.Based on the analysis of the main movements of the human upper limb joint and corresponding muscle groups,14 kinds of actions and 8 pieces of muscles are determined.The collection scheme of sEMG is designed to obtain high-quality data,and all sEMG data are stored into csv format by action category.?2?Preprocess of sEMG.In terms of denoising,a composite wavelet denoising method is proposed in this paper,which can effectively remove high-frequency noise and baseline drift in sEMG.By analyzing the shortcomings of wavelet denoising comprehensive evaluation index T used in sEMG,an improved comprehensive evaluation index P is proposed and verified by experiments.The P index is used to evaluate the denoising effect of wavelet transform threshold denoising,wavelet transform digital filtering threshold denoising and compound wavelet denoising.The P value after compound wavelet denoising is the smallest and the effect is optimal.In terms of extracting effective signal segment,the framed energy method is used to quickly and accurately extract the effective signal segment.The preprocessing algorithm proposed in this paper removes the redundant signals in sEMG and shortens the processing time.?3?Feature extraction and analysis of sEMG.The common features of sEMG including 5 time domain features,3 frequency domain features and 2 time frequency domain features are extracted in this article.On this basis,for the problem of large calculation amount when the nonlinear entropy features are extracted directly,12 sEMG fuzzy entropy and distributed entropy features based on single-window and multi-window analysis methods are studied and extracted.A comprehensive analysis of different types of features reveals that the time-domain features are most unstable,and the value of entropy feature is less than 1 and its stability is the strongest.?4?Feature fusion based on Fisher discrimination and recognition classification.The Fisher discriminant method is used to evaluate the separability of 22 features,and then 9feature fusion vectors based on the result of Fisher are obtained.9 feature fusion vectors are input into the Support Vector Machine?SVM?optimized by Particle Swarm Optimization?PSO?and Grid Search?GS?respectively.The test recognition rate is used as the evaluation index.The results show that the recognition rate of the best feature fusion vector₆F and classifier PSO-SVM is 93.66%.?5?The offline sEMG motion recognition algorithm is applied to the M2 upper limb rehabilitation training platform for the verification experiments of upper limb rehabilitation training.Taking 6 kinds of actions as examples,the final recognition rate of online action is79.79%,which confirms that the motion recognition based on sEMG can be applied to upper limb rehabilitation training.