Driving Cognitive Alertness Detecting Using Evoked Multimodal Physiological Signals Based on Uncertain...

Area of Science:

• Neuroscience
• Machine Learning
• Human-Computer Interaction

Background:

• Multimodal physiological signals are crucial for assessing driver stress and alertness.
• Challenges in cognitive alertness detection include data scarcity and diverse signal modalities.

Purpose of the Study:

• To develop a self-supervised learning model for accurate detection of driving cognitive alertness.
• To address limitations in existing methods for physiological signal analysis in driving contexts.

Main Methods:

• A novel cascaded attention mechanism processes multimodal physiological signal patches.
• Multiscale entropy (MSE) is used to evoke an attention mechanism for data feature extraction.
• A multimodal uncertainty-aware module with uncertain resampling enhances generalization.

Main Results:

• The model significantly outperforms existing baselines in both subject-dependent and independent settings.
• Achieved average improvements of 6.26% (Accuracy), 6.64% (Recall), and 7.75% (F1 Score) in linear probing.
• Demonstrated superior performance with 7.96% (Acc), 9.13% (Rec), and 9.2% (F1) gains in fine-tuning evaluations.

Conclusions:

• The proposed self-supervised model effectively utilizes multimodal physiological signals for cognitive alertness detection.
• The model shows robust performance and generalization capabilities, offering a promising solution for driver monitoring systems.