Drowsiness detection of EEG signals using image-based convolutional neural network

Hong, Yunzhe; Zheng, Mingyuan; Li, Peiheng; Lyu, Xinjie · 2023 · Crossref

DOI: 10.54254/2755-2721/27/20230305

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the critical safety issue of drowsy driving, particularly in the context of autonomous vehicles where reduced driver engagement may lead to increased fatigue and accidents. Motivated by the high fatality rates and societal costs associated with drowsy driving, the authors propose a method to detect driver attention levels using electroencephalogram (EEG) signals. The research specifically investigates whether transforming EEG data into image-based representations improves classification accuracy when processed by Convolutional Neural Networks (CNNs), compared to traditional machine learning methods applied to raw signal data. The experimental design involved five participants who controlled a computer-simulated train for 35 to 55 minutes while wearing a 12-channel EEG acquisition device. The dataset comprised 25 hours of recordings sampled at 128Hz, capturing three distinct mental states: focused, unfocused, and drowsed. Data from seven specific channels (F3, F4, Fz, C3, C4, Cz, and Pz) were utilized. The preprocessing pipeline involved applying Fast Fourier Transform (FFT) and Hanning Windowing to convert time-domain data into the frequency domain. The signal amplitudes were then divided into theta (4-8Hz), alpha (8-12Hz), and beta (12-40Hz) bands. These frequency values were mapped onto a 2D head map using Azimuthal Equidistant Projection (AEP) and converted into RGB color images, where pixel colors reflected data magnitude. These images served as the input for a CNN model, which included convolutional, excitation, and pooling layers to extract features and classify the mental states. The results demonstrated that the CNN model trained on image-based EEG signals achieved a classification accuracy of 83.49% after 200 iterations. This performance was compared against a Support Vector Machine (SVM) model trained on raw EEG signals. While previous studies indicated SVM could reach 96.38% accuracy with large datasets (3,664 samples), its accuracy dropped to 86.59% with smaller datasets (786 samples). In contrast, the proposed CNN method maintained robust performance with limited data, achieving 75.92% accuracy at 100 iterations and improving to 83.49% at 200 iterations. The authors note that the CNN approach is more suitable for embedded systems due to its efficiency with smaller sample sizes. The study concludes that transforming EEG signals into images for CNN classification is an effective method for detecting drowsiness, offering a viable solution for enhancing driving safety in autonomous vehicle systems. The authors acknowledge a limitation regarding the use of a train simulator rather than a car, which may affect signal frequency and amplitude characteristics. They suggest that future offline experiments with actual driving conditions could further refine the model's applicability.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-nvidia on 2026-08-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• drowsiness detection algorithms
• drowsiness

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Empirical Findings: physiological data