Robust Myoelectric during Long‐Time Use by Transfer Learning

Muscle fatigue significantly impacts the accuracy of force estimation, posing a challenge for the long‐term use of myoelectric interfaces. In this study, we propose a novel transfer learning‐based method to enhance force estimation accuracy during muscle fatigue. The approach involves pretraining a force estimator using electromyography (EMG) data from the non‐fatigue state and fine‐tuning the model with data from the fatigue state. A deep end‐to‐end convolutional neural network was employed as the force estimator, eliminating the need for hand‐crafted features and enabling direct learning from raw EMG signals. The results demonstrate a 75% reduction in the root mean square error (RMSE) of force estimation during muscle fatigue using the proposed transfer learning algorithm. Furthermore, the influence of target training dataset size on transfer learning performance was systematically analyzed, providing valuable insights into optimizing model adaptation. This study highlights the effectiveness of transfer learning in mitigating the effects of muscle fatigue on EMG‐based force estimation, enabling the reliable, long‐term use of myoelectric interfaces. The findings have significant implications for improving the functionality and convenience of prosthetic devices, ultimately enhancing user satisfaction and quality of life.