A Novel Hand Gesture Recognition Method Using sEMG-Based PSODE-BPNN

Abstract

Aim to improve the deficiencies of slow convergence speed and prone to trapped into local minimum for back propagation neural network (BPNN) in sEMG-based hand gesture, a novel hand gesture recognition method associating particle swarm optimization (PSO) with differential evolution (DE) is proposed to optimize BPNN, namely PSODE-BPNN. Firstly, initial parameters of weights and biases are adjusted and updated to yield optimal values, via PSO and DE incorporation with the gradient descent method of BPNN. Secondly, the optimal values are utilized as the initial parameters to train the BPNN. Then the proposed method is evaluated on sEMG-based hand gesture dataset Ninapro DB4. The performance for the PSODE-BPNN is compared with the PSO-BPNN, the BPNN, and other related competing methods in literatures. The experimental results show that the PSODE-BPNN improves the inadequacy of slow convergence rate and local minimum of BPNN, and achieves higher gesture recognition accuracy than other comparison listed methods. The effectiveness and superiority of the proposed method is significantly validated.