On The SEMG-Movement Relationship Modeling And Movement Pattern Recognition

The surface electromyography (sEMG) is a biological signal which is recorded from thesurface skeletal muscle through electrode when human body moves independently. It canreflect the different gestures in a certain extent, and has advantage in the field of movementpattern recognition. Therefore, EMG is widely used in biomedical engineering, such as therehabilitation medicine, sports science and pattern recognition.Intelligent bionic arm hasbeen widely used with the rapid development of rehabilitation medicine. Although there aremany places of the practical performance of intelligent prosthetic arm are not yet reachedthe satisfaction of the people, intelligent prosthetic research is still an attractive topic. ThesEMG pattern recogntion build up a good basis for intelligent bionic arm research.This paper focuses on the multi-channel sEMG pattern recognition. The main content asfollows:1) Collecting different movement motions of sEMG signals. A signal processingalgorithm is proposed. Analyzing human forearm muscle diagram and experiments, thelocation of electrodes are determined. After that, six kinds of sEMG movements have beengot. Equipment noise, ambient noise and man-made noise exist in the sEMG acquisitionprocess. In view of the situation above, Butterworth filter, the empirical modedecomposition (EMD) and independent component analysis (ICA) are proposed in signalpretreatment, to remove the noise. Then the effective activity sections have been got.2) sEMG modeling method is proposed. Different movement motions have differentcharacteristics. Therefore, signal feature extraction is the key to the movement patternrecognition. Considering the algorithm computational complexity, the feature extractionmethods based on the time domain, frequency domain and time-frequency domain areproposed. sEMG model is established in order to obtain more effective features to improvethe recognition rate and analyze the generation mechanism of the EMG. Through theanalysis and comparison, the best model coefficients, which has the most representative, arechosen as the input features for movement pattern recognition.3) Two sEMG movement pattern recognition methods are proposed. In view of thedeficiencies of traditional artificial neural network, an improved probabilistic neuralnetwork (PNN) is proposed. The probabilistic neural network is easily affected by theimpact of input data. Due to this problem, an improved ART2neural network which is a stable and not affected by any interference of the input data, is proposed in this paper.Finally, comparing those two methods mentioned above with identification results. The onewho has the short training time and high recognition rate has been selected.4) A method, which used to verify the effect of EMG movement pattern recognitionresults, is proposed. Adams software is used to design a virtual bionic hand model. Thefuzzy PID control system is build by Matlab. Then a bionic hand movement system hasbeen established by those two. Take movement recognition results as the input of the bionicartificial arm motion system. The bionic hand model can be completed based on therecognition results with the corresponding gestures to verify the accuracy of the gesturerecognition system effectively.