Upper Limb Motion Recognition Using Gated Convolution Neural Network via Multi-Channel sEMG

High accuracy in pattern recognition based on surface electromyography (sEMG) contributes to the effectiveness of human motion intention identification. However, due to the complexity and variability of sEMG signal, the improvement of recognition performance still faces many challenges. This study aims to improve accuracy and optimize the deep learning based on the gated convolution neural network (G-CNN) algorithm for classifying six upper limb motion from multi-channel sEMG signal. This algorithm introduces a "gate" structure to integrate multiple convolution layers for upper limb motion recognition. Based on multi-scale feature, a gate employs several filters to extract useful information and block noises by executing convolution operation simultaneously, thus a gate-based feature layer is more effective and efficient as compared to the convolution one. In addition, batch normalization (BN) layer is used to normalize the features after convolution layer. The experimental results show that the G-CNN model proposes in this paper has better recognition accuracy than the traditional machine learning and convolution neural network model.