New AI framework FatigueFormer uses Transformer models to accurately detect muscle fatigue from electrical signals across varying intensity levels.

FatigueFormer combines saliency-guided feature separation with deep temporal modeling to interpret muscle fatigue from surface electromyography (sEMG) signals

Uses parallel Transformer-based sequence encoders to separately capture static and temporal features, improving robustness across different Maximum Voluntary Contraction (MVC) levels from 20-80%

Tested on 30 participants and achieves state-of-the-art accuracy with strong generalization performance under mild-fatigue conditions

Provides attention-based visualization of fatigue dynamics for interpretability, addressing previous challenges with signal variability and low signal-to-noise ratio (SNR)