Multilevel Assessment of Exercise Fatigue Utilizing Multiple Attention and Convolution Network (MACNet) based on...

Area of Science:

• Biomedical Engineering
• Sports Science
• Machine Learning

Background:

• Accurate assessment of exercise fatigue is vital for optimizing performance and preventing injuries.
• Surface electromyography (sEMG) offers quantitative insights into muscle fatigue.
• Existing sEMG fatigue assessment methods often lack precision in multilevel evaluations.

Purpose of the Study:

• To develop a novel deep learning model for a more robust, multilevel assessment of muscle fatigue using sEMG.
• To enhance the extraction of fatigue-related information from sEMG signals.

Main Methods:

• A multiple attention and convolution network (MACNet) was designed for three-level muscle fatigue assessment.
• sEMG signals were collected from 48 subjects under a 50% maximum voluntary contraction paradigm.
• MACNet integrated temporal attention, multiscale convolution, and channel-spatial attention for improved analysis.

Main Results:

• MACNet achieved high average classification F1-Scores and accuracies (e.g., 84.11% subject-wise accuracy).
• GradCAM visualization identified key muscles (flexor digitorum superficialis/profundus) and time-domain features influencing fatigue assessment.
• The model demonstrated superior performance in both subject-wise and cross-subject evaluations.

Conclusions:

• MACNet significantly outperforms state-of-the-art methods in exercise fatigue classification.
• The model effectively extracts nuanced insights from sEMG channels and time-domain features.
• This advancement offers improved tools for understanding and managing exercise-induced muscle fatigue.