Hybrid CNN-Transformer with PSD Features for EEG-Based Driver Drowsiness Detection
الكلمات المفتاحية:
EEG، Drowsiness Detection، CNN Transformer، Power Spectral Density، Cross-Subject Generalization، Leave-One-Subject-Outالملخص
Driver drowsiness is a major cause of fatal accidents worldwide. EEG signals can capture neural changes linked to fatigue before outward signs appear, but reliable detection is challenging due to high inter-subject variability and the complexity of multi-channel recordings, making robust modeling essential for effective monitoring. This paper proposes a hybrid deep model for EEG-based driver drowsiness detection: a 1-D CNN stem with sinusoidal encoding, Bahdanau-style transformer encoder attention pooling, and Power Spectral Density (PSD) feature pathway concatenated before classification, which addresses the high dimensionality and inter individual variation that make the use of multichannel a difficult task. Using the public EEG Driver Drowsiness Database 11 subjects, 30 channels with a leave-one subject-out and subject-calibration scheme, the study achieved 84.7% mean accuracy, 83.8% mean F1, and 0.912 mean AUC. A 3-limb ablation (CNN-Transformer, PSD, hybrid) shows PSD contributes most discriminative power, with the hybrid matching but not significantly exceeding PSD-only (p < 0.01) over no-calibration baseline, suggesting that PSD features are the main driver of cross-subject performance.
المراجع
[1] “A Review on Deep Learning Techniques for EEG-Based Driver Drowsiness detection systems”, IJCAI, vol. 48, no. 3, Sep. 2024, doi: 10.31449/inf.v48i3.5056.
[2] I. Latreche, S. Slatnia, O. Kazar, and S. Harous, “An optimized deep hybrid learning for multi-channel EEG-based driver drowsiness detection,” Biomed. Signal Process. Control, vol. 99, p. 106881, Jan. 2025. doi: 10.1016/j.bspc.2024.106881.
[3] I. Stancin, M. Cifrek, and A. Jovic, “A review of EEG signal features and their application in driver drowsiness detection systems,” Sensors, vol. 21, no. 11, p. 3786, 2021. doi: 10.3390/s21113786.
[4] R. T. Schirrmeister et al., “Deep learning with convolutional neural networks for EEG decoding and visualization,” Human Brain Mapping, vol. 38, no. 11, pp. 5391–5420, 2017.
[5] Y. Song, Q. Zheng, B. Liu, and X. Gao, “EEG conformer: Convolutional transformer for EEG decoding and visualization,” IEEE Trans. Neural Syst. Rehabil. Eng., vol. 31, pp. 710–719, 2023. doi: 10.1109/TNSRE.2022.3230250.
[6] Y. Zhang et al., “Real-time driver drowsiness detection using transformer architectures: A novel deep learning approach,” Scientific Reports, vol. 15, p. 17493, May 2025. doi: 10.1038/s41598-025- 02111-x.
[7] Z. Cheng, X. Bu, Q. Wang, T. Yang, and J. Tu, “EEG-based emotion recognition using multi-scale dynamic CNN and gated transformer,” Scientific Reports, vol. 14, p. 31319, 2024. doi: 10.1038/s41598-024- 82705-z.
[8] X. Feng, Z. Guo, and S. Kwong, “ID3RSNet: Cross-subject driver drowsiness detection from raw single-channel EEG with an interpretable residual shrinkage network,” Front. Neurosci., vol. 18, p. 1508747, Jan. 2025. doi: 10.3389/fnins.2024.1508747.
[9] C. Xu, Y. Song, Q. Zheng, Q. Wang, and P.-A. Heng, “Unsupervised multi-source domain adaptation via contrastive learning for EEG classification,” Expert Syst. Appl., vol. 261, p. 125452, 2025. doi: 10.1016/j.eswa.2024.125452.
[10] W. Lu, X. Zhang, L. Xia, H. Ma, and T.-P. Tan, “Domain adaptation spatial feature perception neural network for cross-subject EEG emotion recognition,” Front. Hum. Neurosci., vol. 18, p. 1471634, Dec. 2024. doi: 10.3389/fnhum.2024.1471634.
[11] Z. Cao, C.-H. Chuang, J.-K. King, and C.-T. Lin, “Multi-channel EEG recordings during a sustained-attention driving task,” Scientific Data, vol. 6, no. 1, p. 19, 2019. doi: 10.1038/s41597-019-0027-4.
[12] P. D. Welch, “The use of fast Fourier transform for the estimation of power spectra: A method based on time averaging over short, modified periodograms,” IEEE Trans. Audio Electroacoust., vol. 15, no. 2, pp. 70–73, 1967.
[13] A. Vaswani et al., “Attention is all you need,” in Advances in Neural Information Processing Systems (NeurIPS), vol. 30, 2017.
[14] I. Loshchilov and F. Hutter, “Decoupled weight decay regularization,” in Proc. Int. Conf. Learn. Represent. (ICLR), 2019.
[15] J. Cui, Z. Lan, Y. Liu, R. Li, F. Li, O. Sourina, and W. Müller-Wittig, “A compact and interpretable convolutional neural network for crosssubject driver drowsiness detection from single-channel EEG,” Methods, vol. 202, pp. 173–184, 2022, doi: 10.1016/j.ymeth.2021.04.017.
[16] J. Cui, Z. Lan, O. Sourina, and W. Müller-Wittig, “EEG-based crosssubject driver drowsiness recognition with an interpretable convolutional neural network,” IEEE Trans. Neural Netw. Learn. Syst., vol. 34, no. 10, pp. 7921–7933, 2023, doi: 10.1109/TNNLS.2022.3147208.
[17] J. Cui, Z. Lan, T. Zheng, Y. Liu, O. Sourina, L. Wang, and W. Müller-Wittig, “Subject-independent drowsiness recognition from single-channel EEG with an interpretable CNN-LSTM model,” in Proc. 2021 Int. Conf. Cyberworlds (CW), 2021, pp. 201–208, doi: 10.1109/CW52790.2021.00041.
[18] Y. Ding, Y. Li, H. Sun, R. Liu, C. Tong, C. Liu, X. Zhou, and C. Guan, “EEG-Deformer: A dense convolutional transformer for brain-computer interfaces,” IEEE J. Biomed. Health Inform., pp. 1–10, 2024. doi: 10.1109/JBHI.2024.3504604.
التنزيلات
منشور
إصدار
القسم
الرخصة
الحقوق الفكرية (c) 2026 Journal of Electrical and Electronic Engineering and Information Technology
هذا العمل مرخص بموجب Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
