Gender-specific Gesture Recognition And Dexterous Hand Simulation Based On SEMG

Surface electromyography(sEMG)can reflect the most primitive motion features of human limbs,and can identify the motion intention of human body by analyzing and processing sEMG.However,as the random and non-stationary characteristics of sEMG signals,using the artificial intelligence prosthesis can not achieve the good control effect.Therefore,how to improve the recognition rate of sEMG to better control artificial intelligence(AI)prosthesis have become a hot research topic in rehabilitation medicine.The effect of hand gesture recognition for sEMG depends largely on the processing of original signals,the extraction of signal features and the construction of classification models.In order to solve the above problems,this paper studies the effects of different sampling locations on sEMG signals;the relationship between sex difference and surface EMG signal;and feature extraction and recognition algorithm;finally,the joint simulation of Matlab and Adams is established to realize the visual operation of gesture.The main research contents are as follows:(1)Firstly,the research status of sEMG at home and abroad is introduced,including feature extraction,recognition and classification methods,and the generation mechanism and characteristics of sEMG are introduced.(2)Secondly,according to the traditional way of collecting electrode,a new way of collecting electrode is put forward,that is,the position of collecting electrode is determined according to the stretching and contraction of corresponding muscle group when a single finger moves.In this paper,we designed 10 kinds of hand gestures,according to the difference between different sexes,we use the moving average energy to detect the starting point of the s EMG signal,and increase the characteristic identification degree of the motion segment by the way of energy compensation,the experimental results show that the recognition rate is improved about 6% after compensation.(3)In this paper,four feature extraction methods are used,including time domain,frequency domain,wavelet packet decomposition,Mel's Cepstrum Coefficient MFCC,and the high dimension MFCC feature coefficient is reduced by principal component PCA.The results of several feature extraction algorithms are sent to the classifier for recognition,and the results show that MFCC is the best feature extraction method.(4)The experiment designs four kinds of classifiers,including the traditional BP neural network,the support vector machine,particle swarm optimization support vector machine(PSO-SVM),and the improved particle swarm optimization support vector machine(CCPSO-SVM),the result shows that the improved CCPSO-SVM has the highest recognition rate,the same individual recognition rate is 98.5%,the recognition rate of different individuals is 93.5%.(5)In this paper,a bionic dexterous hand model is established according to real human hands,and 10 kinds of gesture movements are simulated by Adams controls module and Matlab,the control effectiveness of s EMG signal to the dexterous hand model is verified.