Cross-subject and cross-motion muscle fatigue detection

Develop and validate muscle fatigue detection methods that simultaneously generalize across different subjects and across different motions, achieving robust fatigue state recognition beyond a single movement type and addressing practical confounds such as sensor offset and skin sweating.

Background

The paper proposes IADAN, a deep learning framework that combines an Inception-attention feature extractor with a domain adversarial classifier and supervised contrastive learning to improve cross-subject generalization in sEMG-based fatigue detection. While the method attains high accuracy for three-class fatigue classification in a single movement (calf raises), the authors note that broader generalization across different motion patterns has not been addressed.

They explicitly identify the simultaneous generalization across subjects and motions (e.g., walking, jumping) as an unresolved challenge due to practical constraints such as sensor offsets and skin sweating, pointing to the need for future research to tackle this problem.