AGL-Net: An Efficient Neural Network for EEG-Based Driver Fatigue Detection.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Road Safety Engineering

Background:

• Driver fatigue is a significant road safety concern globally.
• Electroencephalography (EEG) is the gold standard for fatigue detection due to its precision.
• Existing deep learning models for EEG-based fatigue detection lack computational efficiency for mobile deployment.

Purpose of the Study:

• To propose an efficient and accurate deep learning model for EEG-based driver fatigue detection.
• To address the limitations of existing models in terms of computational efficiency and parameter count.
• To develop a fatigue detection system suitable for mobile device implementation.

Main Methods:

• An attention-based Ghost-LSTM neural network (AGL-Net) was developed.
• The model employs an attention mechanism for feature relevance and Ghost bottlenecks for spatial feature extraction.
• Temporal features were extracted using a Long Short-Term Memory (LSTM) network, with both regression and classification models implemented.

Main Results:

• AGL-Net demonstrates superior computational efficiency with low FLOPs (2.67 M) and Params (103,530).
• The model achieved an average accuracy of approximately 87.3% and an RMSE of 0.0864 on the SEED-VIG dataset.
• Ablation experiments confirmed the effectiveness of individual AGL-Net modules, particularly the Ghost bottleneck for efficiency.

Conclusions:

• The proposed AGL-Net is a feasible and high-performing method for EEG-based fatigue detection.
• The Ghost bottleneck module significantly enhances computational efficiency.
• AGL-Net offers a practical solution with higher accuracy and efficiency compared to previous methods.