Exact Fit Attention in Node-Holistic Graph Convolutional Network for Improved EEG-Based Driver Fatigue Detection

TL;DR

The paper tackles the challenge of EEG-based driver fatigue detection by addressing the loss of inter-channel spatial information in traditional CNNs. It introduces NHGNet, a node-holistic graph convolutional network that jointly learns per-channel temporal features through a three-branch temporal extractor with EF-attention and captures global channel interactions via a trainable global similarity adjacency matrix (GDSAM) within a graph convolutional framework. The model demonstrates state-of-the-art performance on two public fatigue-driving EEG datasets, with notable intra-subject and inter-subject improvements, and provides interpretable insights into brain-region involvement (e.g., central parietal Pz/Cz for fatigue; frontal/temporal regions for vigilance). These results highlight the practical potential of dynamic graph learning for robust, interpretable EEG-based fatigue monitoring and advance understanding of neurophysiological patterns associated with fatigue and vigilance.

Abstract

EEG-based fatigue monitoring can effectively reduce the incidence of related traffic accidents. In the past decade, with the advancement of deep learning, convolutional neural networks (CNN) have been increasingly used for EEG signal processing. However, due to the data's non-Euclidean characteristics, existing CNNs may lose important spatial information from EEG, specifically channel correlation. Thus, we propose the node-holistic graph convolutional network (NHGNet), a model that uses graphic convolution to dynamically learn each channel's features. With exact fit attention optimization, the network captures inter-channel correlations through a trainable adjacency matrix. The interpretability is enhanced by revealing critical areas of brain activity and their interrelations in various mental states. In validations on two public datasets, NHGNet outperforms the SOTAs. Specifically, in the intra-subject, NHGNet improved detection accuracy by at least 2.34% and 3.42%, and in the inter-subjects, it improved by at least 2.09% and 15.06%. Visualization research on the model revealed that the central parietal area plays an important role in detecting fatigue levels, whereas the frontal and temporal lobes are essential for maintaining vigilance.

Figures (7)

• Figure 1: Network structure of NHGNet. fs denotes the ultimate sampling rate. $\text{T}_1$ represents the new data dimension produced by dimensionality reduction. BN stands for batch normalization.
• Figure 2: The subtitle provides an elucidation of the data sources and the experimental approach. The curves of different colors in each figure represent various settings of attention mechanisms or activation functions.
• Figure 3: Inter-subject evaluation of adjacency matrix variants on two datasets. Comparative performance of dynamic similarity, dynamic random, and fixed similarity adjacency matrices in terms of accuracy, sensitivity, and f1-score.
• Figure 4: Visualization of EEG channel correlations across different datasets and subjects. Each subplot is marked with the data source, predicted probabilities for vigilance and fatigue stages from NHGNet, and the true labels of the samples.
• Figure 5: Visualization of the relationship between EEG channels in vigilance samples from two datasets. Each subplot, marked at the top, shows the data source, prediction probabilities for vigilance and fatigue states from NHGNet, and the true labels.