SPATIO-TEMPORAL CONVOLUTIONAL NETWORKS FOR MYOELECTRIC CONTROL

Abstract

Utilising both within-channel temporal and between-channel spatial dependencies of the surface Electromyographic (sEMG) signals improves the accuracy of machine learning-based models of myoelectric control. Here, we introduce the Spatio-Temporal Convolutional Network (STCN) to decode five hand gestures from eight EMG signals recorded from the forearm of eight able-bodied subjects. We compared our proposed STCN model with a combination of a conventional convolutional neural network (CNN) and a Long Short-Term Memory (LSTM) deep learning model, as well as the Linear Discriminant Analysis (LDA). The results show that STCN model can outperform both CNN-LSTM and LDA methods at a much lower computational complexity.