Dual-Domain Fusion Graph Convolutional Network for EEG-Based Driving Fatigue Detection

Area of Science:

• Neuroscience
• Machine Learning
• Road Safety Engineering

Background:

• Driving fatigue is a major cause of traffic accidents.
• Electroencephalography (EEG) offers objective and accurate fatigue detection.
• Existing EEG methods underutilize brain topology and inter-electrode information.

Purpose of the Study:

• To propose a novel Dual-Domain Fusion Graph Convolutional Network (DDFGCN) model for enhanced driving fatigue detection.
• To leverage both local and global brain connectivity for multi-level feature aggregation.
• To improve the accuracy and reliability of EEG-based fatigue detection systems.

Main Methods:

• Utilized multi-scale temporal convolution for extracting dynamic EEG features.
• Developed two brain map construction methods to capture local and global channel dependencies.
• Integrated features and employed classification modules for fatigue state prediction.

Main Results:

• Achieved high accuracy rates of 94.67% on the SADT dataset and 95.6% on the SEED-VIG dataset.
• Demonstrated superior classification performance compared to existing methods.
• Validated the model's ability to integrate local brain activity and remote dependencies.

Conclusions:

• The DDFGCN model effectively enhances relational modeling of the entire scalp for improved fatigue detection.
• This approach offers a promising new method for fatigue driving detection technology.
• The findings highlight the importance of considering brain topology in EEG-based fatigue analysis.