Estimation of gait pattern and muscle fatigue based on partial swarm optimization process neural networks

The classification of lower limb movement is important to the design of patient training or auxiliary force control system of moving robot. A lower limb gait classification algorithm using surface electromyography (sEMG) based on the particle swarm optimization process neural network (PSO-PNN) is proposed. Electromyography sensors are adopted to acquire the sEMG of four muscles around the knee joint under different pattern movements of the lower limb and different muscle fatigue levels. In this paper, to avoid the information loss of feature extraction, process neural networks that can directly deal with time varying functions were used to construct the gait classifiers. To improve the learning speed and classification accuracy, partial swarm optimization (PSO) was utilized to tune the weight parameters of the neural network sEMG signals were acquired from 6 subjects for different scenarios to experimentally verify the effectiveness of the algorithm. The results show that higher classification accuracy and faster learning speed can be obtained compared with the BP neural network scheme. Consequently, the proposed method is more suitable for the classification of lower limb movement.