SiamEEGNet: Siamese Neural Network-Based EEG Decoding for Drowsiness Detection

Abstract

Recent advancements in deep-learning have significantly enhanced EEG-based drowsiness detection. However, most existing methods overlook the importance of relative changes in EEG signals compared to a baseline, a fundamental aspect in conventional EEG analysis including event-related potential and time-frequency spectrograms. We herein introduce SiamEEGNet, a Siamese neural network architecture designed to capture relative changes between EEG data from the baseline and a time window of interest. Our results demonstrate that SiamEEGNet is capable of robustly learning from high-variability data across multiple sessions/subjects and outperforms existing model architectures in cross-subject scenarios. Furthermore, the model's interpretability associates with previous findings of drowsiness-related EEG correlates. The promising performance of SiamEEGNet highlights its potential for practical applications in EEG-based drowsiness detection. We have made the source codes available at http://github.com/CECNL/SiamEEGNet.