Study on exercise muscle fatigue based on sEMG and ECG data fusion and temporal convolutional network

Area of Science:

• Biomedical Engineering
• Sports Science
• Machine Learning

Background:

• Muscle fatigue is a critical factor in monitoring exercise effectiveness and preventing injuries.
• Accurate assessment of muscle fatigue is essential for optimizing training regimens and ensuring athlete safety.

Purpose of the Study:

• To develop a novel data fusion method for muscle fatigue assessment using surface electromyography (sEMG) and electrocardiography (ECG) signals.
• To leverage morphological information from time-domain sEMG and ECG signals for enhanced fatigue detection.

Main Methods:

• Processed and normalized sEMG and ECG time series data.
• Employed a deep learning network with sequential convolution for feature extraction.
• Established a muscle fatigue evaluation model using the Dempster-Shafer (D-S) evidence theory for signal fusion.

Main Results:

• The proposed time-domain convolutional network (TCN) model outperformed traditional KNN and SVM algorithms.
• The D-S fusion model achieved high recognition accuracies: 89% for relaxed, 86% for excessive, and 88.5% for fatigue states.
• Achieved overall accuracy of 0.9055, recall rates of 0.9303-0.9570, and F-scores of 0.8764-0.8911 for the different muscle states.

Conclusions:

• The fusion of sEMG and ECG signals using a time-series convolutional network offers a robust method for recognizing motor muscle states.
• This approach demonstrates significant practical value for muscle evaluation, clinical diagnostics, and the development of wearable devices.