Monitored Versus Non-Monitored Stimuli in Brain-Computer-Interface Methods for Classifying Workload States During Piloting Tasks

Applying wireless sensors for measurement of pilot mental workload using electroencephalography (EEG) in passive brain-computer interfaces (pBCI) has proven difficult due to weak workload signatures in artifact-laden environments. While active monitoring of task-irrelevant stimuli is used with pBCI...

Full description

Saved in:
Bibliographic Details
Published in2023 IEEE Sensors Applications Symposium (SAS) pp. 1 - 6
Main Authors Van Benthem, Kathleen, Gard, Stefanie, Herdman, Chris
Format Conference Proceeding
LanguageEnglish
Published IEEE 18.07.2023
Subjects
aviation
brain-computer interface
Electroencephalography
mental workload measurement
Solid modeling
Task analysis
Time measurement
Virtual reality
Wireless communication
Wireless sensor networks
wireless sensors
Online AccessGet full text

Cover

More Information
Summary:Applying wireless sensors for measurement of pilot mental workload using electroencephalography (EEG) in passive brain-computer interfaces (pBCI) has proven difficult due to weak workload signatures in artifact-laden environments. While active monitoring of task-irrelevant stimuli is used with pBCI designs to improve the signal to noise ratio (SNR), this method introduces non-piloting tasks, and thus adds to pilot workload. A fully implicit non-monitoring approach, where stimuli are presented but ignored decreases superfluous artifacts associated with monitoring and responding to the stimuli, but also paradoxically weakens the workload signal of interest. We compared the misclassification rates for a variety of pBCI paradigms in monitoring versus non-monitoring conditions where auditory stimuli were used in the classification of moderate versus high workload conditions. The pBCI models incorporated spectral features of the EEG recorded wirelessly during a virtual reality flight simulation with non-pilots. Classification of workload states, using an open source BCI toolbox, found similar misclassification rates between the monitoring versus non-monitoring datasets (Bayes Factors between 2.1 and 2.9 showed there was no effect of group on classification errors). The results show that reasonable measurement of adjacent workload states (moderate versus high) can be undertaken using EEG data from wireless sensors using non-monitoring pBCI methods.
DOI:10.1109/SAS58821.2023.10254130