A passive brain-computer interface for operator mental fatigue estimation in monotonous surveillance operations:...

Area of Science:

• Human-Computer Interaction
• Cognitive Science
• Biomedical Engineering

Background:

• Visual search in search and rescue missions relies on human operators, making them susceptible to mental fatigue (MF).
• Accurate detection of MF is critical for enhancing operational safety and efficiency.
• Existing methods for MF detection, particularly time-on-task (TOT), have limitations in real-world applicability and direct comparison with performance metrics.

Purpose of the Study:

• To evaluate MF detection using a realistic visual search task.
• To compare the effectiveness of time-on-task (TOT) versus behavioral performance metrics for MF estimation.
• To investigate the influence of labeling type on the accuracy of intra-participant fatigue detection.

Main Methods:

• Participants performed a 1-hour monotonous visual search task divided into four 15-minute blocks.
• Physiological data, including cardiac activity, electroencephalography (EEG), and eye movements, were recorded.
• Machine learning algorithms were employed to detect MF using both TOT and behavioral performance data.

Main Results:

• Machine learning model performance for MF detection was highly dependent on the definition of MF used.
• High classification accuracies (e.g., 99.3%) were achieved when using time-on-task (TOT).
• When MF was estimated based on behavioral performance metrics, classification accuracies dropped to chance level (52.2%).

Conclusions:

• While TOT-based MF detection yields high accuracy, it overlooks practical usability and the cognitive construct of MF.
• Behavioral performance metrics, though operationally more valuable, result in significantly lower detection accuracy.
• Future systems require a more nuanced approach to MF detection that balances accuracy with real-world relevance and cognitive validity.