Related Experiment Video
Updated: May 14, 2026
05:36
STFEEG-Tool: A Spatial-Temporal-Frequency EEG Analysis Tool for Motor Imagery Brain-Computer Interfaces
Published on: March 10, 2026
SE-Driven Dynamic Convolution for Adaptive EEG-Based Driver Fatigue Detection Across Spectral, Spatial, and Temporal
Tianle Zhou1, Jin Cheng2, Jinbiao Zhang2
1Mengxi Honors College, Jiangsu University, Zhenjiang 212013, China.
Sensors (Basel, Switzerland)
|May 13, 2026
Summary
This study introduces DCAMNet, a novel deep learning model for electroencephalogram (EEG)-based driver fatigue detection. DCAMNet achieves high accuracy in detecting driver fatigue, even for unseen drivers, paving the way for safer transportation.
Area of Science:
- Neuroscience
- Machine Learning
- Transportation Safety
Background:
- Driver fatigue is a major cause of road accidents.
- Existing EEG-based fatigue detection methods struggle with inter-subject variability and complex data dynamics.
- Current approaches often process spectral, spatial, and temporal features independently, limiting performance.
Purpose of the Study:
- To develop a unified, lightweight deep learning model for robust EEG-based driver fatigue detection.
- To address signal-level challenges including spectral variability and coupled dynamics.
- To improve the accuracy and efficiency of fatigue detection for real-world deployment.
Main Methods:
- Introduced DCAMNet, a compact Convolutional Neural Network (CNN) with 12.3K parameters.
- Utilized an SE-driven dynamic convolution block for adaptive spectral sensitivity.
- Employed a spatial convolution block for cortical pattern encoding and a temporal attention block for fatigue dynamics.
- Evaluated on SEED-VIG and MESD datasets using subject-mixed and leave-one-subject-out (LOSO) protocols.
Main Results:
- Under LOSO cross-validation, DCAMNet achieved 85.43% accuracy on SEED-VIG and 79±5% on MESD, outperforming baselines.
- Subject-mixed protocols yielded upper-bound accuracies of 97.47% (SEED-VIG) and 96.52% (MESD).
- Demonstrated a low inference latency of 1.35 ms on a standard GPU.
Conclusions:
- DCAMNet effectively addresses key challenges in EEG-based driver fatigue detection.
- The model's lightweight architecture and high accuracy suggest potential for real-time embedded systems in vehicles.
- Further on-device validation is recommended for automotive hardware integration.
