Research On The Classification Of Surface EMG Based On Neural Network

Surface EMG (electromyography) is a bioelectric signal collected on the surfaceof skin when human muscle contraction. It is the electrical pulses produced bystimulated human motor neuron, reflecting the functional status of the human nervesand muscles to a certain extent. Surface EMG has been widely used in the fields ofclinical medicine, rehabilitation medicine and intelligent robots. Due to theadvantages of surface EMG, Using it to complete control system of intelligent bionicprosthesis has become one of the hot research within the field of rehabilitationmedicine. This paper is supported by the important development project of Jilinprovince “Research and development of the bionic arm with temperature, touch andslip telepresent”(2009030) and graduate innovation fund of Jilin University”Research on the classification of surface EMG based on integrated neural network”(20121107).The classification research of surface EMG in this paper is conducive tothe development and industrialization of myoelectric controlled bionic prosthesis,therefore, the study in this paper have important practical and social significance.The main work of this paper includes these following aspects:(1)We introduced the background and significance of this research, analyzed andcompared the research status at home and abroad. Then we introduced the generationmechanism and characteristics of surface EMG in detail. We selected EMG electrodes,EMG electrodes placement and five different hand movements combined with theknowledge of the regional anatomy. We also completed the experiment of surfaceEMG acquisition during five different hand movements.(2)We conducted feature extraction of two-channel surface EMG with timedomain, frequency domain and time-frequency methods. We compared featureelement’s variation and selected wavelet packet decomposition to extract featureelements of two-channel surface EMG.(3)We designed neural network classifier to carry out pattern recognition test ofextracted feature vectors of surface EMG. In order to contrast the pros and cons of thetest results, we totally designed BP neural network, Elman neural network andintegrated BP neural network. After the completion of the design of classifier, weconducted pattern recognition experiments with extracted feature vectors and carried out a statistical analysis of the correct classification rates of the three different neuralnetworks. The results show that integrated BP neural network has the highest correctclassification rates indicating that integrated BP neural network has the bestclassification performance.(4)We calculated the training time and training epochs of the three neuralnetworks. The result is that the training time and epochs of integrated BP neuralnetwork is bigger than the other two. These indicates that integrated BP neuralnetwork can improve the correct classification rates of surface EMG, but requiresmore training time and training epochs.(5)We used Matlab GUI method to complete the development of surface EMGoffline pattern recognition system, combining the aspects of surface EMG read,feature extraction, pattern recognition and recognition results display together,increasing the intuitive and visibility of surface EMG pattern recognition links.As the training stage and testing stage of neural network can be carried outseparately, the training stage can be done offline in advance, so the study in our paperstill has theoretical value and practical significance.