Pattern Classification of Surface Electromyography Based on AR Model and High-order Neural Network

In order to implement the bionic control of powered prostheses for the handicapped effectively, it is necessary to achieve pattern classification for myoelectric signals. In this paper, a method combining AR model with high-order neural network to process surface electromyography is presented. After pro-processing of the acquired electromyography signals, a quartuple eigenvector is constructed by using the coefficients of AR model corresponding to two channels as the inputs of a high-order neural network and the nerve cells in the output layer corresponds to four movement patterns of hand: opening, grasping, twist flexion and twist extension. Before being used, the high-order neural network is trained rigorously with typical sample. The experiments indicate that the method combining AR model with high-order neural network to implement SEMG pattern classification can reduce workload of calculation, and get a relatively good recognition of movements of hand