College of Physics and Electronic Engineering, Hanjiang Normal University, Shiyan 442000, China.
Hubei Key Laboratory of Advanced Technology for Automotive Components, Wuhan University of Technology, Wuhan 430070, China.
Sensors (Basel). 2024 Aug 7;24(16):5086. doi: 10.3390/s24165086.
The present study utilizes the significant differences in θ, α, and β band power spectra observed in electroencephalograms (EEGs) during distracted versus focused driving. Three subtasks, visual distraction, auditory distraction, and cognitive distraction, were designed to appear randomly during driving simulations. The θ, α, and β band power spectra of the EEG signals of the four driving attention states were extracted, and SVM, EEGNet, and GRU-EEGNet models were employed for the detection of the driving attention states, respectively. Online experiments were conducted. The extraction of the θ, α, and β band power spectrum features of the EEG signals was found to be a more effective method than the extraction of the power spectrum features of the whole EEG signals for the detection of driving attention states. The driving attention state detection accuracy of the proposed GRU-EEGNet model is improved by 6.3% and 12.8% over the EEGNet model and PSD_SVM method, respectively. The EEG decoding method combining EEG features and an improved deep learning algorithm, which effectively improves the driving attention state detection accuracy, was manually and preliminarily selected based on the results of existing studies.
本研究利用在注意力分散和集中驾驶时脑电图（EEG）中观察到的θ、α和β频段功率谱的显著差异。设计了三个子任务，即视觉分心、听觉分心和认知分心，在驾驶模拟中随机出现。提取了四种驾驶注意状态的 EEG 信号的θ、α和β频段功率谱，并分别使用 SVM、EEGNet 和 GRU-EEGNet 模型对驾驶注意状态进行检测。进行了在线实验。与提取整个 EEG 信号的功率谱特征相比，提取 EEG 信号的θ、α和β频段功率谱特征对于检测驾驶注意状态更为有效。与 EEGNet 模型和 PSD_SVM 方法相比，所提出的 GRU-EEGNet 模型的驾驶注意状态检测精度分别提高了 6.3%和 12.8%。基于现有研究结果，手动初步选择了一种结合 EEG 特征和改进的深度学习算法的 EEG 解码方法，该方法可有效提高驾驶注意状态检测精度。
