Arm Motion Recognition Based On EMG Signals And Virtual Simulation

Surface EMG signal,as an important physiological electrical signal,can be applied to intelligent system for prosthesis control.At present,in the research of intelligent prosthesis,it is crucial to action recognition and motion control by using EMG signal features.This paper focuses on the surface EMG signal acquisition and pretreatment,EMG feature extraction,recognition,virtual arm model design,virtual prosthetic motion simulation and other relevant research.The main contents are as follows:1)Surface EMG signal collect and preprocess.Design the eight action modes including wrist and palm,and collect EMG signals of finger extensor,flexor shallow muscle,wrist flexor and palm muscle.The method to wavelet transform combined with adaptive filtering was applied to preprocess the original surface electromyography signal and get a pure EMG signal.2)Surface EMG feature extract.First,extract the absolute mean value and the root mean square value from EMG signal as the time domain features.Next,extract the average power frequency and the median frequency from EMG signal spectrum as the frequency domain features.Then,use db3 wavelet for five layers wavelet decomposition.Calculate the root mean square and variance,by the fifth layer approximate coefficient as well as the third,fourth,fifth layer detail coefficients,which are as time-frequency feature of EMG Signals.Last,the differences between EMG features of the different action modes were analyzed by statistical method.3)Surface EMG feature recognize.Firstly,BP neural network was applied to classify action modes respectively by the time domain feature,frequency domain feature and time-frequency feature of surface EMG signals.The average accuracy of eight action recognitions is 89%,77%,91% corresponding to time domain,frequency domain and time-frequency features.Then,the stack self-encoding algorithm was designed to classify action modes under the time domain feature,frequency domain feature,and time-frequency feature of surface EMG signals.The average recognition accuracy is up to 95%,89%,96% respectively.It is shown that the timefrequency feature can better reflect the differences between different action modes,and the stack self-coding deep learning algorithm is superior to BP neural network method in applying surface EMG signal for action mode recognition.4)Virtual 3D prosthetic design and movement simulate.Use improved D-H method to build linkage arm model,then carry out kinematics analysis and trajectory planning.Utilize linkage arm model to complete the simulation of drinking water by the MATLAB platform.Apply SolidWorks software to design a virtual 3D prosthesis,and verify the virtual 3D prosthetic model is rational by computer simulation in a virtual reality environment.Finally,by using internet and Java programs,the action recognition results were transmitted to virtual prosthesis,which carried out various arm movement according to the trajectory planning.The simulation experimental results have shown that the pattern recognition method is effective as well as the action execution is integrity and accuracy.The research results of this paper can be applied to artificial intelligence human-computer interaction,virtual reality,bionic prosthesis and the other fields.And it has the dual value of science and application.