AI-Powered Detection of Muscle Fatigue in Long-Duration Activities Using Surface EMG

Abstract

Muscle fatigue is of great importance during any kind of physical activity, as it reduces the performance of the body part and, in extreme cases, can even cause harm to the individual. Surface electromyography (sEMG) has emerged as one of the most prevalent methods for assessing muscle state and fatigue-related changes. This paper suggests a deep learning-based approach to detect the onset of muscle fatigue caused by long-duration, low-intensity HCI tasks, namely, typing, gaming, and using a mouse for an extended period of time, which is not much talked about compared to high-intensity or athletic fatigue. The proposed method employs a CNN-LSTM model and physiologically meaningful time-frequency characteristics (RMS Amplitude, Zero Crossing Rate, and Median Frequency) to predict the occurrence of fatigue. A multistage signal-preprocessing chain, including a band-pass filter, full-wave rectification, RMS smoothing, and normalisation, is applied prior to classification. All experiments are conducted on the NinaPro DB2 public dataset to ensure reproducibility. Results from signal-level analysis revealed clear, nonstationary amplitude variations throughout the signal, in the form of sporadic high-amplitude events, which are attributed to motor unit recruitment during continuous muscle activation. This demonstrates the feasibility of the method and provides a basis for developing a portable muscle fatigue detection sensor.