Automated discrimination of gait patterns based on sEMG recognition using neural networks

A set of schemes for automated discrimination of gait patterns based on recognition of surface electromyogram (sEMG) of human lower limbs is proposed to classify 3 different terrains and 6 different movement patterns. To compare the recognition performance of different classifiers, Back Propagation Neural Networks (BPNNs) and Process Neural Networks (PNNs) are deployed to discriminate gait patterns under different conditions. To obtain the discrete inputs to BPNNs, time-frequency parameters, wavelet variance and matrix singularity values are separately considered as the feature vector. Since PNNs can deal with time-varying functions without signal discretion or feature extraction, sEMG signal after filtering is directly fed to the neural networks to discriminate different gaits. To improve the learning efficiency and accuracy, partial swarm optimization (PSO) is used to obtain the weight parameters of PNNs. Simulations were conducted to validate the efficiencies and recognition accuracies of different neural classifiers. PNNs show good adaptability and robustness and have great potential in the application of bio-electrical signal processing.