The Processing of Electromyography Signal Based on Wavelet Neural Network

Abstract—This paper introduces the wristwork pattern recognition with the method of Auto-regressive (AR) Model and Wavelet Neural Network (WNN). The correlation between Surface Electromyography Signal (SEMG) and wristwork has been researched based on the analysis of these signals. A method is used to analyze SEMG signal due to its unsteady characteristic based on wavelet transform. The different motion pattern is recognized by extracting Four-order AR coefficient. Then we construct the coefficients as eigenvector and input it into WNN. The experiment shows that pattern recognition rate of four movements (flexor carpi, extensor carpi, intorsion and extorsion) are all more than 80%. This paper finds the WNN has many advantages such as self-learning, Self-adaptive, robust, fault-tolerance and generalization ability; so WNN should be better than BP neural network