An improved feature extraction algorithm for EEG-based driving fatigue recognition

Geng, Xiaozhong; Hu, Weixin; Liang, Qipeng; Chen, Cheng; Zhang, Xintong; Yu, Ping; Bao, Jie · 2025 · Crossref

DOI: 10.1038/s41598-025-18554-1

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 challenge of accurately detecting driving fatigue using electroencephalogram (EEG) signals, a major cause of traffic accidents. While EEG signals provide high temporal resolution for monitoring brain activity, their utility is often compromised by noise, particularly electrooculography (EOG) artifacts from eye movements, and the limitations of single-dimensional feature extraction methods. The authors propose an improved algorithm that integrates advanced signal preprocessing with a novel multi-feature fusion strategy to enhance detection accuracy and reliability. The methodology consists of two primary stages: artifact removal and feature extraction. To eliminate EOG artifacts, the study employs a hybrid approach combining Ensemble Empirical Mode Decomposition (EEMD) and Fast Independent Component Analysis (FastICA). FastICA decomposes the raw EEG signals into independent components, which are then analyzed using kurtosis to identify those associated with ocular activity. These specific components undergo EEMD, a noise-assisted decomposition technique that mitigates mode mixing, allowing for the precise identification and removal of artifact-related Intrinsic Mode Functions (IMFs) based on low-frequency characteristics, high kurtosis, and energy proportion. The clean EEG signals are then reconstructed via inverse ICA. For feature extraction, the authors introduce a fusion of Wavelet Packet Transform (WPT) and Sample Entropy (SampEn). WPT is used to decompose the purified signals into time-frequency sub-bands, capturing detailed spectral information. SampEn is subsequently applied to these sub-bands to quantify the nonlinear complexity of the signals. These time-frequency and nonlinear features are integrated into a comprehensive vector and classified using a Support Vector Machine (SVM). Experimental results demonstrate that the proposed EEMD-FastICA denoising method effectively filters out EOG artifacts while preserving the integrity of the underlying EEG signals. Furthermore, the multi-feature fusion approach, combining WPT and SampEn, significantly outperforms traditional single-method feature extraction techniques. By capturing both the time-frequency dynamics and the nonlinear complexity of fatigue-related EEG patterns, the integrated model achieves higher recognition accuracy in distinguishing between fatigued and non-fatigued states. The study validates the superiority of this combined strategy through comparative experiments, confirming its effectiveness in handling the nonstationary and nonlinear nature of EEG data. The significance of this work lies in its contribution to more robust and accurate driver fatigue detection systems. By addressing the dual challenges of signal noise and limited feature representation, the proposed algorithm offers a reliable solution for real-time monitoring applications. The integration of EEMD-FastICA for preprocessing and WPT-SampEn for feature extraction provides a comprehensive framework that enhances the sensitivity of EEG-based fatigue recognition, potentially reducing traffic accidents caused by driver drowsiness. This approach sets a new standard for processing complex physiological signals in safety-critical environments.

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

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