Extracted SEED-VIG dataset for cross-dataset driver drowsiness recognition

The SEED-VIG dataset contains EEG data from subjects conducting a monotonous driving task in a virtual simulator. The eye closure (PERCLOS) information was obtained from Senso-Motoric-Instrument (SMI) eye-tracking glasses, which was used to label the data. The dataset is officially available from:<br>https://bcmi.sjtu.edu.cn/~seed/seed-vig.html<br>We have extracted a small portion of the dataset for research on cross-dataset driver drowsiness recognition. Specifically, we down-sampled the EEG signals to 128Hz and processed the data with a low-pass filter of 1Hz. EEG samples with a 3-second length were extracted prior to the PERCLOS evaluation event. We followed the procedure adopted in [1] by labeling the samples as ‘alert’ when PERCLOS is lower than 0.35 and samples as ‘drowsy’ when PERCLOS is higher than 0.7, while the samples in the middle range were discarded. We further discarded sessions with less than 50 samples of either class and balanced the class for each session by selecting the most alert or drowsiest ones. In this way, we have 4566 samples in total from 12 subjects.<br>The data file contains 3 variables and they are EEGsample, substate and subindex.<br>"EEGsample" contains 4566 EEG samples of size 17x384 from 12 subjects. Each sample is a 3s EEG data with 128Hz from 17 EEG channels."subindex" is an array of 4566x1. It contains the subject indexes from 1-11 corresponding to each EEG sample."substate" is an array of 4566x1. It contains the labels of the samples. 0 corresponds to the alert state and 1 correspond to the drowsy state.<br>Details about our work of cross-dataset driver drowsiness recognition based on this dataset can be found from [2] and the codes are available from:<br>https://github.com/cuijiancorbin/Benchmarking-EEG-based-cross-dataset-driver-drowsiness-recognition-with-deep-transfer-learning<br><br><br>[1] W.-L. Zheng, and B.-L. Lu, "A multimodal approach to estimating vigilance using EEG and forehead EOG," Journal of neural engineering, 2017. 14(2): p. 026017.[2] Cui, Jian, et al. "Benchmarking EEG-based cross-dataset driver drowsiness recognition with deep transfer learning." 2023 45th Annual International Conference of the IEEE Engineering in Medicine &amp; Biology Society (EMBC). IEEE, 2023. DOI: 10.1109/EMBC40787.2023.10340982SEED-VIG数据集包含了受试者在虚拟驾驶模拟器中完成单调驾驶任务时采集的脑电图（Electroencephalogram, EEG）数据。眼闭合度（PERCLOS）信息来自感官运动仪器（Senso-Motoric-Instrument, SMI）眼动追踪眼镜，该信息被用于该数据集的标注。 该数据集的官方获取地址为：<br>https://bcmi.sjtu.edu.cn/~seed/seed-vig.html<br> 我们提取了该数据集的一小部分样本，用于开展跨数据集驾驶员疲劳识别研究。具体而言，我们将脑电图信号下采样至128Hz，并使用1Hz低通滤波器对数据进行预处理。我们在PERCLOS评估事件前截取了时长为3秒的脑电图样本。我们遵循文献[1]中的标注规则：当PERCLOS低于0.35时，将样本标记为“清醒”；当PERCLOS高于0.7时，标记为“嗜睡”；处于两个阈值中间区间的样本则予以丢弃。我们进一步移除了任一类别样本数少于50的会话，并通过选取最清醒或最嗜睡的样本，实现了每个会话的类别平衡。最终，我们共得到来自12名受试者的4566条样本。 该数据文件包含3个变量，分别为EEGsample、substate与subindex。 "EEGsample"包含了来自12名受试者的4566条脑电图样本，每条样本的维度为17×384。每条样本为来自17个脑电图通道、采样率128Hz的3秒脑电图数据。"subindex"为一个4566×1的数组，存储了每条脑电图样本对应的受试者索引（取值范围为1至11）。"substate"为一个4566×1的数组，存储了样本的标签：0对应清醒状态，1对应嗜睡状态。 基于该数据集开展的跨数据集驾驶员疲劳识别研究的详细内容可参见文献[2]，相关代码可从以下地址获取：<br>https://github.com/cuijiancorbin/Benchmarking-EEG-based-cross-dataset-driver-drowsiness-recognition-with-deep-transfer-learning<br><br><br> [1] W.-L. Zheng与B.-L. Lu，“基于脑电图与前额眼电图的警觉度估计多模态方法”，《Journal of Neural Engineering》，2017年，第14卷第2期，论文编号026017。 [2] Cui, Jian等人，“基于深度学习迁移学习的脑电图跨数据集驾驶员疲劳识别基准研究”，2023年第45届IEEE工程医学与生物医学学会年会（EMBC），IEEE，2023年，DOI: 10.1109/EMBC40787.2023.10340982