KCL-STN：一种基于脑电图信号的时空融合疲劳驾驶检测方法

KCL-STN: A Spatiotemporal Fusion Fatigue Driving Detection Method Based on EEG Signals

摘要: 疲劳驾驶是交通事故的重要诱因，其检测对于保障交通安全至关重要。脑电图(EEG)信号因能反映大脑活动状态而被广泛用于疲劳驾驶检测，但现有深度学习方法常面临对预处理的依赖或对EEG信号时空信息的处理不足等挑战。本文提出了一种基于脑电图信号的时空融合疲劳驾驶检测方法——KCL-STN (KAN-CNN-LSTM Spatio-Temporal information Network)。该方法巧妙地结合了卷积神经网络和长短期记忆网络，分别从原始脑电信号中提取空间和时间特征，并进行有效融合，实现了端到端的疲劳驾驶检测。针对脑电数据稀缺问题，本文还提出了一种脑电信号滑动窗口增强算法，以增加样本数量并提高模型训练的稳定性。在公开数据集上的实验结果表明，KCL-STN在分类准确度、召回率和精确率等指标上均优于多种现有方法，准确率达到86.05%。消融实验证实了关键组件KAN线性层和滑动窗口数据增强方法的有效性。跨被试实验也证明了模型良好的泛化性能和鲁棒性。研究结果表明，KCL-STN能够有效地从原始脑电信号中提取疲劳相关特征，是鲁棒且高性能的疲劳驾驶检测方法。

Abstract: Fatigue driving is a significant factor contributing to traffic accidents, making its detection crucial for traffic safety. Electroencephalogram (EEG) signals, which reflect brain activity, have been widely utilized in fatigue detection. However, existing deep learning approaches often suffer from excessive reliance on signal preprocessing or insufficient exploitation of the spatio-temporal characteristics inherent in EEG data. In this study, we propose a novel spatio-temporal fusion framework for EEG-based fatigue detection, named KCL-STN (KAN-CNN-LSTM Spatio-Temporal Information Network). The proposed model integrates Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks to extract spatial and temporal features from raw EEG signals, respectively, enabling end-to-end fatigue detection. To address the issue of limited EEG data, we introduce a sliding window-based EEG data augmentation algorithm, which increases sample diversity and enhances training stability. Experimental results on a public dataset demonstrate that KCL-STN outperforms several existing methods in terms of classification accuracy, recall, and precision, achieving an accuracy of 86.05%. Ablation studies validate the effectiveness of the KAN linear layer and the proposed data augmentation strategy. Furthermore, cross-subject experiments confirm the model’s strong generalization capability and robustness. Overall, the findings indicate that KCL-STN is a robust and high-performance method for fatigue detection based on raw EEG signals.