Abstract: This paper proposes a fatigue driving detection method based on synthetic minority over-sampling technique (SMOTE), gated recurrent unit (GRU) neural network, and Attention mechanism. Fatigue driving is a significant cause of traffic accidents, especially during long-distance driving or nighttime driving, when the driver's alertness decreases. This paper analyzes steering wheel angle data to extract driving behavior features and determine the driver's fatigue state. To address the data imbalance issue, SMOTE algorithm is used to oversample the minority class samples, alleviating the impact of class imbalance on model training. A total of 4 320 driving segment samples are collected, and the fatigue levels are classified into three categories: normal, fatigued, and severely fatigued. By combining the GRU neural network with the Attention mechanism (GRU can handle long sequence data, while the Attention mechanism helps the network focus on corner transformation at critical turning moments), enhancing the recognition ability of fatigue features. Experimental results show that the proposed method achieves an probability of detection (POD) of over 98% in fatigue driving detection, effectively identifying different levels of fatigue and holding significant importance for traffic safety

Key words: fatigue driving; SMOTE algorithm; GRU neural network; Attention mechanism

摘要： 文章提出了一种基于合成少数类过采样技术（SMOTE）、门控循环单元（GRU）神经 网络与注意力（Attention）机制的疲劳驾驶检测方法。疲劳驾驶是导致交通事故的重要因素， 尤其在长途驾驶或夜间行车时，驾驶员的警觉度会下降。文章通过分析方向盘转角数据，提 取驾驶员行为特征，判断其疲劳状态。为解决数据不平衡问题，文章采用 SMOTE 算法对少 数类样本进行过采样，缓解了类别不平衡对模型训练的影响，研究共采集了 4 320 个驾驶片 段样本，并将疲劳程度划分为三类：正常、疲劳、非常疲劳。结合 GRU 神经网络和 Attention 机制（GRU 能够处理长时序数据，而 Attention 机制能够帮助网络关注关键时刻的转角变化）， 提高疲劳特征的识别能力。实验结果表明，所提出的方法在疲劳驾驶检测中取得了 98%以上 的命中率（POD），能够有效识别不同疲劳程度，对交通安全具有重要意义。

关键词: 疲劳驾驶；SMOTE 算法；GRU 神经网络；Attention 机制