CEEMDAN fuzzy entropy based fatigue driving detection using single-channel EEG

Fatigue-induced driving remains a prominent contributing factor to frequent traffic accidents. Extensive research has demonstrated the efficacy of utilizing Electroencephalogram (EEG) for accurate fatigue detection. However, the laborious and cost-intensive process of EEG labeling, compounded by the issue of label reliability, poses a substantial challenge. Most of the current studies are based on multi-channel EEG signals, which are not conducive to the application of intelligent vehicle systems because they require abundant complex wiring. Based on the EEG signals of the left forehead, this paper proposes a self-training semi-supervised method to transform unlabeled data into pseudo-labeled data and combines the fuzzy entropy feature after scale-transformation of the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) to establish a driving fatigue recognition model. In this work, three kinds of features of EEG in the time domain, frequency domain, and entropy were extracted to train the semi-supervised model of self-training. Unlabeled data is predicted using this model, and high-confidence pseudo-labeled data is amalgamated with the labeled data. The multi-scale fuzzy entropy algorithm based on CEEMDAN was used to establish the classification model. The results show that the CEEMDAN fuzzy entropy method can improve the recognition accuracy by about 8 %, and the use of pseudo-labeled data with high confidence obtained by single self-training for model training can improve the recognition accuracy. Compared with other recognition methods based on single-channel forehead EEG, this method has higher accuracy. Meanwhile, our methodology's effectiveness has been further validated through testing on publicly accessible datasets, underscoring its robustness and applicability.