EEG-Based Monitoring of Pilot Training: Transferring Full-Cap Representations to Headphone-Style Electrode Positions

Monitoring pilots’ mental states during training is important for ensuring safety and optimizing performance, yet existing full-cap EEG systems are impractical for operational use. In this work, we evaluate a headphone-style 9-channel electrode montage rather than a complete headphone EEG device, and propose a transfer learning framework to preserve performance under sparse coverage. Our two-stage approach first applies self-supervised pretraining on 64-channel full-cap EEG, and then adapts the learned representations to the headphone-style montage through a position correction module that accounts for electrode misplacement. We validate the framework in a flight simulator with 12 participants on two tasks: motor intention (left vs. right) and cognitive workload (low vs. high). Despite limited coverage, the headphone-style montage achieves within-subject accuracies of 85% for motor imagery and 89% for workload, compared to full-cap performance of 88% and 91%. Cross-subject accuracies reached 66% and 73%, demonstrating generalizability across users. Ablation analyses show that both self-supervised pretraining and position correction independently improve performance, and together provide a 12% boost in cross-subject decoding accuracy. These findings highlight the feasibility of headphone-style EEG montages for practical pilot monitoring, while clarifying methodological limitations and the need for future real-time and user-centered evaluation.