Confidence-Aware Multimodal Network for Driver Drowsiness Detection: Leveraging EEG-EOG Signals Enhanced by Knowledge Distillation

Driver fatigue is a leading cause of traffic accidents, underscoring the urgent need for advanced fatigue detection systems. The integration of physiological signals, especially electroencephalography (EEG) and electrooculography (EOG), shows significant promise in this field. However, existing methods frequently encounter challenges with the fusion accuracy of multiple modalities and the computational burden of complex models. In this article, we introduce a novel multimodal fatigue detection framework, named the EEG-EOG confidence-aware network (EECNet), specifically designed to address these issues. EECNet adeptly extracts local and global EEG features using the deep channel attention transformer (DCAT) model, while EOG features are extracted via the EOGNet. A confidence-aware module is included to assess the reliability of each modality, thereby guiding the fusion process to improve both detection accuracy and robustness. Additionally, knowledge distillation (KD) is utilized to develop a lightweight model that effectively balances complexity and performance. Experimental results on a public dataset demonstrate that EECNet surpasses state-of-the-art methods, achieving an average accuracy of 95.68%, thus showcasing its superior performance in detecting driver drowsiness.