EMG Pattern Recognition based on Particle Swarm Optimization and Recurrent Neural Network

摘要

Surface electromyography signal (sEMG) plays an important role in gesture recognition and prosthetic control. Aiming at the problems of complex combination of RNN parameters, setting difficulty, and structure dependence of model quality, an EMG pattern recognition method based on particle swarm optimization recurrent neural network (PSO-RNN) is proposed. This method uses the characteristics of particle swarm optimization (PSO), such as high global search efficiency, fast convergence speed, and wide optimization range, and automatically finds the optimal structure of RNN through continuous iterative updating. On the Ninapro EMG database, the classification of 12 types of EMG actions by the PSO-RNN algorithm is tested, and the results are compared with four algorithms applied in the same data set. The results show that the proposed PSO-RNN algorithm model achieves a high accuracy of 94.1667%, and it has certain effectiveness and practicability.