Quantitative Assessment of Muscle Fatigue Based on Improved Gramian Angular Difference Field

Accurately quantifying muscle fatigue and identifying relative risk is crucial for manual laborers and patients with muscle diseases. Existing studies have insufficiently quantified muscle fatigue and have not fully utilized subjective fatigue information. This study proposes a dynamic muscle fatigue recognition method based on the Gaussian normalized Gramian Angular Difference Field (G-GADF) image encoding method, utilizing the multi-channel surface electromyography (sEMG) signal fusion method. Thirty-one healthy participants performed a focused bicep curl movement and the sEMG signals were categorized into ten fatigue states based on the rating of muscle fatigue (RMF) values. The multi-channel sEMG signal data are preprocessed and fused into a single-channel signal using the local maxima of different channels at each time. Then, the signals are encoded into 2D images using the G-GADF method, and these images are fed into the EfficientNet_B0 model. Finally, the EfficientNet_B0 model is compared with other models. The results indicate that the image dataset, generated through preprocessing, multi-channel fusion, and G-GADF methods, achieves a test accuracy of 97.058% for ten categorized muscle fatigue classifications using the EfficientNet_B0 model. These methods improve the accuracy of identifying muscle fatigue for individuals engaged in high-intensity physical labor and muscle rehabilitation, aiding in exercise intensity management and muscle injury prevention due to overexertion.