An improved CapsNet based on data augmentation for driver vigilance estimation with forehead single-channel EEG

Area of Science:

• Neuroscience
• Artificial Intelligence
• Traffic Safety

Background:

• Estimating driver vigilance is crucial for reducing traffic accidents.
• Current electroencephalography (EEG)-based methods often require multi-channel signals, limiting practical application due to cost and complexity.

Purpose of the Study:

• To develop a cross-subject vigilance estimation algorithm using single-channel EEG signals.
• To improve the utility and accuracy of EEG-based vigilance monitoring for practical applications.

Main Methods:

• A capsule network (CapsNet) algorithm was developed for single-channel EEG vigilance estimation.
• Novel input feature map construction and a self-attention mechanism were integrated.
• Conditional generative adversarial networks (cGANs) were used to augment single-channel data.

Main Results:

• The proposed algorithm demonstrated improved computing speed.
• Root-mean-square error (RMSE) was reduced by 3%, and Pearson Correlation Coefficient (PCC) improved by 12% compared to mainstream methods.
• Feasibility of using single-channel forehead EEG for cross-subject vigilance estimation was confirmed.

Conclusions:

• The developed CapsNet-based algorithm offers a feasible and efficient approach for cross-subject vigilance estimation using single-channel EEG.
• This research paves the way for lightweight and practical EEG vigilance monitoring devices.