汽车工程 ›› 2022, Vol. 44 ›› Issue (5): 771-777.doi: 10.19562/j.chinasae.qcgc.2022.05.015
所属专题: 车身设计&轻量化&安全专题2022年
收稿日期:
2021-10-09
修回日期:
2021-12-17
出版日期:
2022-05-25
发布日期:
2022-05-27
通讯作者:
黄晶
E-mail:huangjing926@hnu.edu.cn
基金资助:
Jing Huang(),Yang Peng,Ye Huang,Xiaoyan Peng
Received:
2021-10-09
Revised:
2021-12-17
Online:
2022-05-25
Published:
2022-05-27
Contact:
Jing Huang
E-mail:huangjing926@hnu.edu.cn
摘要:
现有驾驶员精神负荷评价研究多以驾驶场景中有无次任务来给定驾驶员的精神负荷分类标签,但驾驶员在正常驾驶情景下也可能由于陷入自我思维而导致精神负荷的增加;此外,由于个体差异,同一驾驶次任务对不同驾驶员精神负荷的影响也不尽相同。因此,由传统方法所制作的数据集可能存在噪声标签,从而影响精神负荷评价模型的训练效果。针对此类问题,本文中采用置信学习的方法对驾驶员的精神负荷分类标签进行检测和滤除,使用处理过的标签,以脑电、心电和皮电信号特征作为模型输入,基于支持向量机、随机森林、K近邻、决策树、逻辑回归和多层感知机等多种算法构建驾驶员精神负荷模型,对比分析噪声标签处理对提高各类模型性能的效果。结果表明:使用置信学习进行噪声标签处理后,所构建的多种驾驶员精神负荷模型的性能均得到了明显的改善,其中,支持向量机模型的性能提升的效果最佳。
黄晶,彭扬,黄烨,彭晓燕. 考虑噪声标签影响的驾驶员精神负荷状态评价[J]. 汽车工程, 2022, 44(5): 771-777.
Jing Huang,Yang Peng,Ye Huang,Xiaoyan Peng. Evaluation of Driver's Mental Load State Considering the Influence of Noisy Labels[J]. Automotive Engineering, 2022, 44(5): 771-777.
表3
初始提取的生理信号特征表"
信号类型 | 特征类型 | 特征名称与解释 |
---|---|---|
ECG | 时域 | AVNN心跳间隔平均值;SDNN心跳间隔标准差;SDSD相邻RR间期差值序列的标准差; RMSSD相邻RR间期差值序列的均方根;CVSD连续差异的变异系数;CVNII变异系数; PNNI_50时间间隔大于50 ms的RR间期占所有间期的比重;Range_NNI最大和最小心跳间隔序列之间的差异; Mean_HR平均心跳;Max_HR最高心跳;Min_HR最低心跳;Std_HR心跳的标准差 |
频域 | LF低频频段能量;HF高频频段能量;LF_HF关于HF与LF的比值;TP频段内所含的总能量 | |
非线性 | SD1关于Poincaré 散点图短轴参数;SD2关于Poincaré 散点图长轴参数; SD1_SD2关于Poincaré 散点图短轴与长轴参数之比;CSI心脏交感指数;CVI迷走神经指数 | |
EEG | 时域 | EEG_mean脑电信号均值;EEG_std脑电信号标准差;EEG_ptp脑电信号峰峰值;EEG_max脑电信号最大值; EEG_min脑电信号最小值;EEG_cv脑电信号的功率;EEG_diff_mad脑电信号1阶差分绝对平均值; EEG_diff_nor_mad脑电信号归一化的1阶差分平均值;EEG_diff_2_mad脑电信号2阶差分绝对平均值; EEG_diff_2_nor_mad脑电信号归一化的2阶差分绝对平均值;EEG_power脑电信号的能量 |
频域 | EEG_θ:θ波段(4~8 Hz)功率谱能量;EEG_α:α波段(8~12 Hz)功率谱能量; EEG_δ:δ波段(1~3.5 Hz)功率谱能量;EEG_β:β波段(12~30 Hz)功率谱能量; EEG_γ:γ波段(30 ~ 50 Hz)功率谱能量 | |
非线性 | EEG_DE脑电信号微分熵;EEG_ApEn脑电信号近似熵;EEG_SampEn脑电信号样本熵; EEG_HFD脑电信号的Higuchi分形维数;EEG_LZCn脑电信号的L-Z复杂度 | |
EDA | 时域 | EDA_mean皮电信号均值;EDA_std皮电信号标准差;EDA_ptp皮电信号峰峰值;EDA_max皮电信号最大值; EDA_min皮电信号最小值;EDA_cv皮电信号的功率;EDA_diff_mad皮电信号1阶差分绝对平均值; EDA_diff_nor_mad皮电信号归一化的1阶差分平均值;EDA_diff_2_mad皮电信号2阶差分绝对平均值; EDA_diff_2_nor_mad皮电信号归一化的2阶差分绝对平均值;EDA_power皮电信号的能量 |
频域 | EDA_FD1皮电信号(0~0.1 Hz)的功率谱能量;EDA_FD2皮电信号(0.1~0.2Hz)的功率谱能量 | |
非线性 | EDA_DE皮电信号微分熵;EDA_ApEn皮电信号近似熵;EDA_SampEn皮电信号样本熵; EDA_HFD皮电信号的Higuchi分形维数;EDA_LZCn皮电信号的L-Z复杂度 |
表4
心电特征的卡方检验结果"
序号 | 特征名称 | 卡方检验p值 | 序号 | 特征名称 | 卡方检验p值 |
---|---|---|---|---|---|
1 | AVNN | 0.413 0 | 12 | Std_HR | 0.037 5 |
2 | SDNN | 0.021 8 | 13 | LF | 0.028 6 |
3 | RMSSD | 0.083 6 | 14 | HF | 0.043 6 |
4 | SDSD | 0.086 8 | 15 | LF_HF | 0.562 7 |
5 | CVSD | 0.049 4 | 16 | TP | 0.032 0 |
6 | CVNII | 0.023 6 | 17 | SD1 | 0.088 3 |
7 | PNNI_50 | 0.044 5 | 18 | SD2 | 0.011 4 |
8 | Range_NNI | 0.035 4 | 19 | SD1_SD2 | 0.250 7 |
9 | Mean_HR | 0.538 9 | 20 | CSI | 0.250 7 |
10 | Max_HR | 0.959 9 | 21 | CVI | 0.049 5 |
11 | Min_HR | 0.037 7 |
表5
不同类型特征组合下噪声标签处理前后驾驶员精神负荷分类模型的准确率和F1分数对比 (%)"
噪声标签处理 | 算法模型 | ECG | EEG | EDA | ECG+EEG | ECG+EDA | EEG+EDA | all | |||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
准确率 | F1 分数 | 准确率 | F1 分数 | 准确率 | F1 分数 | 准确率 | F1 分数 | 准确率 | F1 分数 | 准确率 | F1 分数 | 准确率 | F1 分数 | ||
处理前 | SVM | 59.77 | 59.70 | 81.77 | 81.76 | 66.13 | 66.11 | 84.38 | 84.37 | 72.85 | 72.84 | 86.10 | 86.10 | 87.05 | 87.05 |
RF | 60.58 | 60.53 | 83.32 | 83.32 | 67.33 | 67.32 | 85.82 | 85.81 | 73.88 | 73.87 | 87.83 | 87.83 | 88.61 | 88.61 | |
DT | 59.76 | 59.76 | 82.79 | 82.74 | 66.15 | 66.12 | 85.30 | 85.30 | 72.88 | 72.86 | 87.32 | 87.34 | 88.15 | 88.15 | |
MLP | 59.91 | 59.87 | 81.90 | 81.85 | 65.60 | 65.55 | 84.91 | 84.91 | 72.35 | 72.32 | 86.84 | 86.85 | 87.90 | 87.90 | |
LR | 59.24 | 59.19 | 80.43 | 80.38 | 64.87 | 64.83 | 83.07 | 83.07 | 70.94 | 70.92 | 84.90 | 84.91 | 85.97 | 85.97 | |
KNN | 59.53 | 59.49 | 80.69 | 80.65 | 64.71 | 64.67 | 83.50 | 83.50 | 71.04 | 71.02 | 85.34 | 85.35 | 86.54 | 86.54 | |
处理后 | SVM | 61.74 | 61.53 | 87.13 | 87.13 | 69.50 | 69.43 | 89.85 | 89.84 | 76.43 | 76.38 | 91.78 | 91.78 | 93.13 | 93.13 |
RF | 62.52 | 62.38 | 88.06 | 88.06 | 70.56 | 70.51 | 90.70 | 90.69 | 77.29 | 77.25 | 92.58 | 92.58 | 93.80 | 93.80 | |
DT | 61.79 | 61.68 | 87.33 | 87.35 | 69.26 | 69.22 | 90.04 | 90.06 | 75.91 | 75.87 | 91.94 | 91.97 | 93.10 | 93.10 | |
MLP | 61.68 | 61.54 | 86.78 | 86.79 | 68.93 | 68.88 | 89.95 | 89.96 | 75.65 | 75.61 | 91.93 | 91.96 | 93.21 | 93.21 | |
LR | 61.01 | 60.89 | 85.61 | 85.62 | 68.30 | 68.26 | 88.58 | 88.60 | 74.41 | 74.37 | 90.54 | 90.56 | 91.91 | 91.92 | |
KNN | 61.26 | 61.14 | 85.75 | 85.76 | 68.15 | 68.11 | 88.89 | 88.90 | 74.57 | 74.53 | 90.84 | 90.86 | 92.28 | 92.29 |
1 | HWANG S L, YAU Y J, LIN Y T, et al. A mental workload predicator model for the design of pre alarm systems[C]. International Conference on Engineering Psychology and Cognitive Ergonomics(EPCE 2007), 2007. |
2 | DE WAARD D, STUDIECENTRUM V. The measurement of drivers’mental workload[J]. Dissertationsubrugnl,1996, 198 (39). |
3 | YERKES R M, DODSON J D. The relation of strength of stimulus to rapidity of habit-formation[J]. Journal of Comparative Neurology and Psychology, 1908, 18(5):459-482. |
4 | REID G B. The subjective workload assessment technique: a scaling procedure for measuring mental workload[M]. Elsevier Science Publishers, 1988. |
5 | HART S G, STAVELAND L E. Development of NASA-TLX (task load index): results of empirical and theoretical research[J]. Advances in Psychology, 1988, 52(6):139-183. |
6 | GRIER R A. How high is high? a meta-analysis of NASA-TLX global workload scores[C]. Human Factors & Ergonomics Society, 2015:1727-1731. |
7 | BRAARUD P Ø. An efficient screening technique for acceptable mental workload based on the NASA Task Load Index—development and application to control room validation[J]. International Journal of Industrial Ergonomics, 2020, 76. |
8 | RAHMAN N I A, DAWAL S Z M, YUSOFF N. Subjective responses of mental workload during real time driving: a pilot field study[J]. IOP Conference Series: Materials Science and Engineering, 2017, 210. |
9 | BENEDETTO S, PEDROTTI M, MININ L, et al. Driver workload and eye blink duration[J]. Transportation Research Part F: Traffic Psychology and Behaviour, 2011, 14( 3): 199-208. |
10 | ALMAHASNEH H, CHOOI W T, KAMEL N, et al. Deep in thought while driving: an EEG study on drivers' cognitive distraction[J]. Transportation Research Part F :Traffic Psychology & Behaviour, 2014, 26(pt.A):218-226. |
11 | FLUMERI G D, BORGHINI G, PIETRO A, et al. EEG-based mental workload neurometric to evaluate the impact of different traffic and road conditions in real driving settings[J]. Frontiers in Human Neuroscience, 2018, 12. |
12 | ABDRAHMAN N I,DAWAL S Z M, YUSOFF N. Driving mental workload and performance of ageing drivers[J]. Transportation Research Part F: Traffic Psychology and Behaviour, 2020, 69:265-285. |
13 | HUANG Jing, LUO Xiong, PENG Xiaoyan. A novel classification method for a driver’s cognitive stress level by transferring interbeat intervals of the ECG signal to pictures[J]. Sensors, 2020, 20, 1340. doi:10.3390/s20051340. |
14 | TJOLLENG A, JUNG K, HONG W, et al. Classification of a driver's cognitive workload levels using artificial neural network on ECG signals[J]. Applied Ergonomics, 2017. |
15 | 黄晶,杨梦婷.考虑初始情绪的个性化驾驶负荷状态评价[J].中国公路学报,2021,34(1):167-176. |
HUANG Jing, YANG Mengting. Personalized driving load state evaluation considering initial emotions[J]. China Journal of Highway and Transport, 2021, 34(1): 167-176. | |
16 | CHIHARA T, KOBAYASHI F, SAKAMOTO J. Evaluation of mental workload during automobile driving using one-class support vector machine with eye movement data[J]. Applied Ergonomics, 2020,89. |
17 | EGOVNIK T, STOJMENOVA K, JAKUS G, et al. An analysis of the suitability of a low-cost eye tracker for assessing the cognitive load of drivers[J]. Applied Ergonomics, 2018, 68:1-11. |
18 | KRAMER A F. Physiological metrics of mental workload: a review of recent progress [M]. Navy Personnel Research and Development Center, 1990. |
19 | BARUA S, AHMED M U, BEGUM S. Towards intelligent data analytics: a case study in driver cognitive load classification[J]. Brain Sciences, 2020, 10(8):526. |
20 | HE D, DONMEZ B, LIU C C, et al. High cognitive load assessment in drivers through wireless electroencephalography and the validation of a modified N-back task[J].IEEE Transactions on Human-Machine Systems, 2019,49(4): 362-371.doi: 10.1109/THMS.2019.2917194. |
21 | FOY H J, PETER C. Mental workload is reflected in driver behaviour, physiology, eye movements and prefrontal cortex activation[J]. Applied Ergonomics, 2018, 73:90-99. |
22 | YAN Shengyuan, TRAN Cong Chi, WEI Yingying, et al. Driver's mental workload prediction model based on physiological indices[J]. International Journal of Occupational Safety and Ergonomics, 2017. |
23 | SUK K H, DAESUB Y, SOON S H, et al. Predicting the EEG level of a driver based on driving information[J]. IEEE Transactions on Intelligent Transportation Systems, 2018, PP:1-11. |
24 | CHEN Z, WU C, ZHONG M, et al. Identification of common features of vehicle motion under drowsy/distracted driving: a case study in Wuhan, China[J]. Accident Analysis and Prevention, 2015, 81:251-259. |
25 | CHAKRABORTY B, NAKANO K. Automatic detection of driver's awareness with cognitive task from driving behavior[C]. 2016 IEEE International Conference on Systems, Man, and Cybernetics (SMC). IEEE, 2016. |
26 | MIYAKE S. Multivariate workload evaluation combining physiological and subjective measures[J]. International Journal of Psychophysiology, 2001, 40(3):233-238. |
27 | HAN J, LUO P, WANG X. Deep self-learning from noisy labels[C]. 2019 IEEE/CVF International Conference on Computer Vision (ICCV) 2019. |
28 | LI J, SOCHER R, HOI S. DivideMix: learning with noisy labels as semi-supervised learning[J]. arxiv,2020. |
29 | 欧阳全会,陈定方.高速公路交通流状况自动检测模拟器的研究与试验[J].武汉理工大学学报(交通科学与工程版),2004(3):416-419. |
OUYANG Quanhui, CHEN Dingfang. Research and experiment on the automatic detection simulator for expressway traffic flow conditions [J]. Journal of Wuhan University of Technology (Transportation Science and Engineering Edition), 2004(3): 416-419. | |
30 | 徐良, 陈日清, 王祥,等. 基于零相位的心电信号基线漂移滤波方法的实时性改进[J]. 中国医药导刊, 2015, 17(S1):31-35. |
XU Liang, CHEN Riqing, WANG Xiang, et al. Real-time improvement of the baseline drift filtering method of ECG signal based on zero phase[J]. China Medical Guide, 2015, 17(S1): 31-35. | |
31 | 史玉君. 基于经验模态分解的眼电伪迹去除方法的研究[D].兰州:兰州大学,2015. |
SHI Yujun. Research on electrooculogram artifact removal method based on empirical mode decomposition[D]. Lanzhou :Lanzhou University, 2015. | |
32 | 孙凯. 基于混合生理信号处理的情绪分类技术研究[D].哈尔滨:哈尔滨工业大学,2020. |
33 | 翟晓雪,张皓.非线性动力学分析方法在神经康复领域中的应用进展[J].中国康复医学杂志,2019,34(4):483-486. |
34 | RUSSAKOVSKY O, DENG J, SU H, et al. ImageNet large scale visual recognition challenge[J]. International Journal of Computer Vision, 2015, 115(3):211-252. |
35 | KRIZHEVSKY A, HINTON G. Learning multiple layers of features from tiny images[J]. Handbook of Systemic Autoimmune Diseases, 2009, 1(4). |
36 | NORTHCUTT C G, JIANG L, CHUANG I L. Confident learning: estimating uncertainty in dataset labels[J]. arxiv,2019. |
[1] | 陈吉清,李子涵,兰凤崇,蒋心平,潘威,陈继开. 基于非线性降维IC特征的实车电池SOH估计[J]. 汽车工程, 2023, 45(2): 199-208. |
[2] | 崔格格,吕超,李景行,张哲雨,熊光明,龚建伟. 数据驱动的智能车个性化场景风险图构建[J]. 汽车工程, 2023, 45(2): 231-242. |
[3] | 姜俊昭,杨文豪,彭彬,郭婷,徐业凯,王国卓. 基于能耗加权策略的燃料电池汽车续驶里程预测[J]. 汽车工程, 2023, 45(12): 2357-2365. |
[4] | 何智成,谢泽军,刘侃,周恩临,唐谦,黄元毅. 纯电动汽车传动系统-电机结构参数协同设计优化[J]. 汽车工程, 2023, 45(11): 2113-2122. |
[5] | 廉玉波,凌和平,王钧斌,潘华,谢朝. 基于混合高斯-隐马尔可夫模型的动力电池实时热失控检测[J]. 汽车工程, 2023, 45(1): 139-146. |
[6] | 赵健,李雅欣,佟静,朱冰,武维祥,孙博华,韩嘉懿. 基于车辆动力学响应特征的越野地面分类方法[J]. 汽车工程, 2022, 44(6): 909-918. |
[7] | 胡杰,朱雪玲,何陈,杨光宇. 基于实车数据的电动汽车电池健康状态预测[J]. 汽车工程, 2021, 43(9): 1291-1299. |
[8] | 谢奕展,程夕明. 锂离子电池状态估计机器学习方法综述[J]. 汽车工程, 2021, 43(11): 1720-1729. |
[9] | 胡杰,高志文. 基于数据驱动的电动汽车动力电池SOC预测[J]. 汽车工程, 2021, 43(1): 1-9. |
[10] | 颜诗旋, 朱平, 刘钊. 基于改进LightGBM模型的汽车故障预测方法研究*[J]. 汽车工程, 2020, 42(6): 815-819. |
|