Analysis Of Surface Electromyography Signal And Research On Sports Mode Classification

Surface electromyography signal is the superposition of bio-electricity signal on skin surface that is produced when nervous system of human body controls muscle contraction. Characterized by non-invasive, safe and easy-obtained, it is widely used in the field of intelligent mechanical arm as a control signal source. At present, the problems of research on taking surface electromyography signal as a control source are as followings:1) The de-noising effect of traditional analog filter is poor because the noise has much influence on the process of gathering the Surface electromyography signal.2) The real-time of controlling the intelligent mechanical arm can’t be satisfied with the delay of signal during the process of gathering the Surface electromyography signal.3) The accuracy of intelligent mechanical arm could be affected by the low recognition rate of the multi-mode actions of upper arm by general feature classifiers.In this paper, we put forward an approach which is more suitable for the optimization of signal control source. In this method, the surface electromyography signal has been gathered, de-noised, forecasted and analyzed, and then do the recognition of multi-mode actions of upper arm.1) Based on the wavelet de-noising method, we present a method of de-noising with integrates independent component analysis. It has a better effect comparing with the traditional filter and wavelet de-noising method, not only wiping off the noise signal but also reserving the useful ones.2) It is relatively better that the forecast result of building mathematics model for surface electromyography signal through the combination of Gaussian process model and wavelet regression.3) After classifying separately by neural network classifier and support vector machines classifier, the result turns out that the latter is superior to neural the former. However, in neural network, they are better that momentum BP algorithm and LM algorithm than traditional BP network algorithm.This paper builds data gathering system for surface electromyography signal through the real-time workspace of MATLAB, collects surface electromyography signal by eight different actions of upper arm, and then makes filtration through blending independent component analysis and wavelet de-noising method. According to the production mechanism and feathers of surface electromyography signal, we build mode by Gaussian process regression and extracts characteristic values from forecast result according to the analysis of time domain, frequency domain and time-frequency domain. Finally, we make sports mode classification for the feature vector extracted before by using support vector machines classifier. The design scheme in this article not only improves the signal in terms of signal-to-noise ratio and instantaneity, but also promotes the recognition rate of sports mode.