The Application of Support Vector Machine in Pattern Recognition

Abstract

To nonstationary characteristics of surface electromyography (sEMG) signals, a novel sEMG pattern recognition method, which is based on wavelet packet transformation and support vector machine (SVM), is proposed. Raw four channels sEMG signals from four corresponding muscles are first analyzed with wavelet packet transformation. And then the energy of different frequency bands in the wavelet packet decomposition coefficients is extracted as the signal character to construct eigenvector. A new multi-class SVM classifier is designed with "one versus one" classification strategy and binary tree. Experiment results show that eight upper-limb movement patterns can be well identified after training by the SVM and average identification ratio is 99.375%, and that the SVM can sort out sEMG eight movement patterns more accurately than traditional BP neural network, Elman neural network and RBF neural network. And the SVM recognition result is robust. It offers a new method for sEMG pattern recognition, which can be directly applied to the other nonstationary bioelectric signals pattern recognition study.