sEMG-Based Ankle Joint Torque Prediction by Combining CNN-LSTM and SENet Attention

Authors:
Publisher website
Google Scholar
Abstract
Joint torque prediction can be applied to the rehabilitation assessment of patients and the control of rehabilitation robots. The surface electromyography(sEMG) signal is an important indicator for judging muscle activity and human locomotor ability, and it is widely used in joint torque prediction. We propose a method to predict the ankle joint torque by fusing CNN and LSTM and inserting SENet attention mechanism to construct a CNN-LSTM-SE hybrid neural network. CNN is used to extract the higher order features of the sEMG signals and reduce the dimensionality of the data, LSTM is used to retain or forget the data, different calf muscles play different importance during ankle movement, so SENet is used to learn the importance of calf muscles. We had four subjects perform stationary bicycle exercise and collected their calf sEMG signals and ankle joint angular acceleration signals in toe flexion and dorsiflexion, as the input parameters of the neural network. Coefficient of determination ($\mathrm{R}^{2}$) and root mean square error (RMSE) were used as the criteria to judge the prediction accuracy.
Funding Information
  • National Natural Science Foundation of China

This publication has 13 references indexed in Scilit: