DP-MP: a novel cross-subject fatigue detection framework with DANN-based prototypical representation and mix-up...

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Electroencephalography (EEG) is a key tool for fatigue detection.
• Practical EEG fatigue detection faces challenges with individual differences and noisy data.
• Existing methods struggle with cross-subject generalization.

Purpose of the Study:

• To develop an effective framework for cross-subject fatigue detection using EEG.
• To address challenges of bio-individual differences and noisy labels in fatigue detection.
• To introduce a novel pairwise learning approach for fatigue detection.

Main Methods:

• Proposed a novel DP-MP framework combining domain-adversarial neural networks and Mix-up pairwise learning.
• Utilized prototypical representation to encode fatigue-related semantic structures in EEG.
• Conceptualized fatigue detection as a pairwise learning task to reduce label interference.

Main Results:

• Achieved state-of-the-art performance on SEED-VIG and FTEF datasets with accuracies of 88.14% and 97.41%, respectively.
• Demonstrated the model's effectiveness and generalization capability in cross-subject fatigue detection.
• The Mix-up pairwise learning approach enhanced sample relationships for better detection.

Conclusions:

• This work pioneers fatigue detection as a pairwise learning task, offering a new perspective.
• The DP-MP framework successfully mitigates bio-individual differences and noisy labels.
• The findings advance practical applications of brain-computer interfaces for fatigue monitoring.