Research On Robot Semg Control Interface Based On Neural Network

In recent years,with the increasing demand for intelligent rehabilitation equipment,the birth of mechanical and electrical integration has emerged.In this paper,the humancomputer interaction control interface is established based on surface electromyography(sEMG).The sEMG acquisition circuit is designed and constructed,and optimized method of sEMG feature selection and pattern recognition in time domain features,then the control strategies of sEMG control interface with the Nao robot as hardware experimental platform.The main content of this article is as follows:First of all,a sEMG acquisition device based on Arduino UNO is designed and established,the characteristics of acquisition circuit are theoretically analyzed and simulated by Simulink,according to the analysis result,the improve acquisition circuit is reconstructed by modified the electrical component parameters,the frequency domain information of output signals are compared with improve circuit and original circuit with the actually test results.Secondly,the features of sEMG in time domain,frequency domain and time-frequency domain are extracted by continuous time windows,and each features' application scenarios are analyzed,an evaluation indicator of feature selection in time domain based on time series model is proposed,and tested performance of the evaluation indicator in time domain features of sEMG.Then,the neurons number of hidden layer in feedback neural network in sEMG gestures classification performance is analyzed,and a neural network gesture classification base on sEMG is designed and constructed,and an advanced neural network based on reverse sample labeling method is proposed,compared the recognition results of advanced networks and raw networks in no-prior sample,the result indicated the effectiveness of the method,and improve the generalization ability of networks when recognition the no-prior sample.Finally,a virtual interactive game based on sEMG is programed and established by Processing,and the walking stability of Nao robot is analyzed based on ZMP theory.The sEMG control interface is designed and established on two experiment platforms: virtual game and Nao robot,the experimental results verify the feasibility and practicability of the EMG control interface.The influence of signal acquisition delay on real-time gesture recognition accuracy is analyzed,and the identification accuracy of the active and antagonistic muscles under the same gesture is compared.The research content of this paper has important reference significance and reference value for the research of low cost and convenience of sEMG acquisition equipment,optimization of sEMG time domain feature selection method,improvement of sEMG pattern recognition algorithm and practical application of sEMG control interface.