复杂恶劣的作业环境易诱发矿工不安全情绪，进而导致不安全行为产生，是煤矿安全事故的险兆之一。近红外光谱脑功能成像（Functional Near-Infrared Spectroscopy，fNIRS）和机器学习近年来逐渐被应用于情绪识别研究。为有效识别矿工不安全情绪，避免“险兆矿工”引发煤矿安全事故，基于便携式fNIRS设备进行了矿工不安全情绪识别模型研究。

       首先，在现有研究基础上明确了矿工不安全情绪的概念、分类标准及其视角下煤矿事故致因路径，在fNIRS和机器学习理论基础上分析并提出了可将氧合血红蛋白（Oxyhemoglobin，HbO₂）浓度中提取的特征指标结合相应情绪标签输入机器学习分类器构建矿工不安全情绪识别模型。然后，设计了基于fNIRS的矿工不安全情绪识别研究实验方案，其中包含300s静息态和7个情绪诱发任务，招募了48名受试者进行实验，收集了问卷、量表数据和前额叶（Prefrontal Cortex，PFC）24条通道的fNIRS信号。结合SPSS软件处理了问卷及量表数据，分析了实验任务对安全情绪与不安全情绪的诱发效果以及具体诱发了那些情绪。采用MATLAB软件借助Homer2和NIRS_KIT工具箱预处理了fNIRS信号，通过对所得HbO₂浓度一般线性模型建模并进行单样本T检验，分析了矿工在两种情绪状态下各测量通道的血液动力学响应强度（β值）和脑激活显著性是否存在差异。最后，从10种HbO₂浓度特征指标中使用配对T检验与多重比较校正法分析与选取了153条矿工不安全情绪识别相关特征通道，采用z-score方法对所选特征指标进行了标准化，并通过ReliefF算法选择了121个关键特征指标。以40名被试的特征指标和情绪标签为训练集，其余为测试集，采用决策树、支持向量机、集成学习（Ensemble Learning，EL）和神经网络训练了矿工不安全情绪识别模型，并测试了4种分类器各自最优模型。

       研究结果及结论：（1）发现了低熟悉度的情绪诱发素材有利于诱发不安全情绪，而高熟悉度的情绪诱发素材有利于诱发安全情绪。3个安全情绪诱发任务诱发情绪主要有愉悦、轻松、愉快和高兴4种，而4个不安全情绪诱发任务主要诱发了愤怒、忧伤、敌对、惊恐、害怕、紧张、忧伤、担心和悲观9种情绪。（2）安全情绪的24条通道β均值小于0，而不安全情绪则大于0，且愉悦度越高被试PFC的β均值越低。从PFC激活水平来看，左侧PFC是不安全情绪识别敏感区域。此外，发现从均值、峰值、方差、近似熵、平均峰值、偏度、峰度和均方根这8种HbO₂浓度特征指标中选取的特征通道在两种情绪状态下也存在显著差异。（3）ReliefF特征选择能有效提升支持向量机、EL和神经网络所训练矿工不安全情绪识别模型的准确率，所有训练模型中，EL模型训练准确率为84.5%，测试准确率为80%，平均准确率可达83.7%，比其他相似研究的准确率更高，且需采集fNIRS信号通道数更少。本研究可用于矿工不安全情绪的实时监测和预警，为提升煤矿人因事故防控水平提供了新途径，有助于推动煤矿安全管理模式向事前预防转型。

       Complex and harsh operating environments are prone to induce miners' unsafe emotions, which in turn lead to unsafe behaviours, and are one of the danger signs of coal mine safety accidents. Functional Near-Infrared Spectroscopy (fNIRS) and machine learning have been gradually applied to emotion recognition research in recent years. In order to effectively identify miners' unsafe emotions and avoid coal mine safety accidents caused by "dangerous omen miners", this paper carries out the research on miners' unsafe emotion recognition model based on fNIRS.

       Firstly, on the basis of existing research, the concept of miners' unsafe emotions, the classification criteria and the causal path of coal mine accidents from their perspectives are clarified, and the feature indicators extracted from the concentration of oxyhemoglobin (HbO₂) combined with the corresponding emotion labels are inputted into a machine learning classifier to construct a miners' unsafe emotions recognition model based on the theories of fNIRS and machine learning. recognition model. Then, an experimental protocol for the fNIRS-based recognition of miners' unsafe emotions was designed, which included 300s resting state and 7 emotion-evoking tasks, and 48 subjects were recruited to conduct the experiment, and questionnaires, scale data, and fNIRS signals from 24 channels of the Prefrontal Cortex (PFC) were collected. The questionnaire and scale data were processed with the help of SPSS software to analyse the effect of the experimental task on the evocation of safe and unsafe emotions as well as those emotions that were specifically evoked. The fNIRS signals were preprocessed using MATLAB software with the aid of Homer2 and the NIRS_KIT toolkit, and whether there were differences in the haemodynamic response strengths (β-values) and brain activation significance of the measured channels in the miners' two emotional states was analysed by modelling the resulting general linear model of HbO₂ concentration and performing a one-sample t-test. Finally, 153 feature channels related to the identification of miners' unsafe emotions were analysed and selected from 10 HbO₂ concentration feature indicators using paired t-tests with multiple comparisons correction, the selected feature indicators were standardised using the z-score method, and 121 key feature indicators were selected by the ReliefF algorithm. With the feature indicators and emotion labels of 40 subjects as the training set and the rest as the test set, the miners' unsafe emotion recognition model was trained using decision trees, support vector machines, Ensemble Learning (EL) and neural networks, and the optimal model of each of the four classifiers was tested.

       Findings and conclusions: (1) It was found that low-familiarity emotion-evoking material favoured the induction of insecure emotions, whereas high-familiarity emotion-evoking material favoured the induction of secure emotions. The three secure emotion-evoking tasks induced four main types of emotions: pleasure, relaxation, cheerfulness, and gladness, whereas the four insecure emotion-evoking tasks induced four main types of emotions: anger, sadness, hostility, alarm, fear, nervousness, sadness, worry, and pessimism 9 kinds of emotions. (2) The mean β value of the 24 channels of safe emotions was less than 0, while that of insecure emotions was greater than 0, and the higher the pleasantness the lower the mean β value of the subjects' PFC. In terms of PFC activation level, the left PFC is a sensitive area for insecure emotion recognition. In addition, it was found that the feature channels selected from the eight HbO₂ concentration feature indicators, namely mean, peak, variance, approximate entropy, mean peak, skewness, kurtosis, and root mean square, also differed significantly between the two emotional states. (3) ReliefF feature selection can effectively improve the accuracy of miners' unsafe emotion recognition models trained by support vector machine, EL and neural network, and among all the training models, the training accuracy of EL model is 84.5%, the testing accuracy is 80%, and the average accuracy can be up to 83.7%, which is higher than that of other similar studies, and the number of channels of fNIRS signals need to be captured is less. This study can be used for real-time monitoring and early warning of miners' unsafe emotions, which provides a new way to improve the level of prevention and control of human-caused accidents in coal mines, and helps to promote the transition of the coal mine safety management mode to ex ante prevention.

[1]景国勋, 刘孟霞. 2015—2019年我国煤矿瓦斯事故统计与规律分析 [J]. 安全与环境学报, 2022, 22(03): 1680-1686.

[2]王云刚, 崔春阳, 张飞燕, 等. 2011—2020年我国较大及以上煤矿事故统计分析及研究 [J]. 安全与环境学报, 2023, (09): 3269-3276.

[3]LALITHA K, RAMYA G, SHUNMUGATHAMMAL M. AI-Based Safety Helmet for Mining Workers Using IoT Technology and ARM Cortex-M [J]. IEEE SENSORS JOURNAL, 2023, 23(18): 21355-21362.

[4]许鹏飞. 2000-2021年我国煤矿事故特征及发生规律研究 [J]. 煤炭工程, 2022, 54(07): 129-133.

[5]景国勋, 穆璐璐. 煤矿瓦斯爆炸事故统计分析及应急管理研究 [J]. 安全与环境学报, 2023, (10): 3657-3665.

[6]付恩三, 白润才, 刘光伟, 等. “十三五”期间我国煤矿事故特征及演变趋势分析 [J]. 中国安全科学学报, 2022, 32(12): 88-94.

[7]程磊, 许艳之, 景国勋. 2019—2021年煤矿顶板事故统计分析 [J]. 煤炭技术, 2023, 42(02): 135-138.

[8]李乃文, 张文文, 牛莉霞. 领导非权变惩罚对矿工不安全行为的影响研究 [J]. 中国安全科学学报, 2019, 29(12): 1-6.

[9]田水承, 景国勋. 安全管理学 第2版 [M]. 北京: 机械工业出版社, 2016.

[10]田水承, 冯帅, 高晓颖, 等. 一线矿工不安全状态的熵权TOPSIS法评价 [J]. 安全与环境学报, 2023, 23(11): 4049-4056.

[11]李红霞, 田水承. 人的应急管理与生命安全教育研究—来自大学校园的观察与实践 [M]. 西安: 西北工业大学出版社, 2015.

[12]WANG X, ZHANG C, DENG J, et al. Analysis of Factors Influencing Miners' Unsafe Behaviors in Intelligent Mines using a Novel Hybrid MCDM Model [J]. Int J Environ Res Public Health, 2022, 19(12): 7368.

[13]田水承, 孙璐瑶, 唐艺璇, 等. 矿工不安全状态评价指标体系的构建与分析 [J]. 西安科技大学学报, 2021, 41(03): 402-409.

[14]李国为, 田水承. 煤矿岗前人员不安全状态智能检测系统研究——以红柳林煤矿为例 [J]. 中国安全生产科学技术, 2023, 19(02): 106-113.

[15]YANG L, WANG X, ZHU J, et al. Comprehensive Evaluation of Deep Coal Miners’ Unsafe Behavior Based on HFACS-CM-SEM-SD [J]. International Journal of Environmental Research and Public Health, 2022, 19(17): 10762.

[16]李广利, 田水承, 严一知, 等. 矿工关键不安全情绪识别研究 [J]. 西安科技大学学报, 2021, 41(05): 793-799.

[17]PETITTA L, PROBST T M, GHEZZI V, et al. Cognitive failures in response to emotional contagion: Their effects on workplace accidents [J]. Accident Analysis & Prevention, 2019, 125: 165-173.

[18]ZHANG G, LU D, LIU H. Strategies to Utilize the Positive Emotional Contagion Optimally in Crowd Evacuation [J]. Ieee Transactions on Affective Computing, 2020, 11(04): 708-721.

[19张舒, 王双, 史秀志. 不安全行为决策中情绪作用机制的脑电研究 [J]. 中国安全科学学报, 2023, 33(01): 32-40.

[20]YANG X, TIAN Y, FENG K, et al. Signal Game Analysis on the Effectiveness of Coal Mine Safety Supervision Based on the Affective Events Theory [J]. Complexity, 2020, 2020: 1-9.

[21]杨捷鸿, 焦学军, 曹勇, 等. 多生理信号信息融合技术的情绪识别发展 [J]. 生物医学工程研究, 2021, 40(04): 420-427.

[22]李梦琪, 龚安民, 南文雅, 等. 基于功能近红外光谱成像的神经反馈技术及应用 [J]. 生物医学工程学杂志, 2022, 39(05): 1041-1049.

[23]PAN T, WANG H B, SI H Q, et al. Research on the Identification of Pilots' Fatigue Status Based on Functional Near-Infrared Spectroscopy [J]. Aerospace, 2022, 9(03): 173.

[24]TIAN F Y, LI H X, TIAN S C, et al. Is There a Difference in Brain Functional Connectivity between Chinese Coal Mine Workers Who Have Engaged in Unsafe Behavior and Those Who Have Not? [J]. International Journal of Environmental Research and Public Health, 2022, 19(01): 509.

[25]LIN W Q, LI C. Review of Studies on Emotion Recognition and Judgment Based on Physiological Signals [J]. Applied Sciences, 2023, 13(04): 2573.

[26]BALCONI M, GRIPPA E, VANUTELLI M E. What hemodynamic (fNIRS), electrophysiological (EEG) and autonomic integrated measures can tell us about emotional processing [J]. Brain Cogn, 2015, 95: 67-76.

[27]张迪, 万柏坤, 明东. 基于生理信号的情绪识别研究进展 [J]. 生物医学工程学杂志, 2015, 32(01): 229-234.

[28]姜劲, 焦学军, 潘津津, 等. 基于功能性近红外光谱技术识别情绪状态 [J]. 光学学报, 2016, 36(03): 164-174.

[29]靳来鹏. 前概念评测的近红外脑影像表征 [D]. 南京: 东南大学, 2019.

[30]王思锐. 基于近红外光谱技术的情绪识别算法的研究 [D]. 兰州: 兰州大学, 2019.

[31]马亚静. 关于矿工安全心理问题的思考 [J]. 煤矿安全, 2007, (07): 87-90.

[32]刘玉姣, 朱宝岩. 中国煤矿工人心理健康状况的元分析 [J]. 中国安全生产科学技术, 2016, 12(06): 187-192.

[33]刘波. 安全情感文化对矿工不安全情绪的影响研究 [D]. 西安: 西安科技大学, 2018.

[34]马亚静. 矿工情绪与安全问题研究 [J]. 中国煤炭, 2007, (01): 60-62.

[35]郑莹. 煤矿员工不安全行为的心理因素分析及对策研究 [D]. 唐山: 河北理工大学, 2008.

[36]褚建平, 蔡云生, 常向东, 等. 煤矿工人工伤事故与人格、心理健康因素的研究 [J]. 中国健康心理学杂志, 2011, 19(06): 664-665.

[37]代丽芳. 矿工情绪管理研究 [D]. 阜新: 辽宁工程技术大学, 2012.

[38]李焕. 情绪与矿工不安全行为关系实验研究 [D]. 西安: 西安科技大学, 2015.

[39]惠璐. 不安全情绪对矿工风险决策影响的实验研究 [D]. 西安: 西安科技大学, 2018.

[40]房俨然, 魏薇, 罗萍, 等. 员工负性情绪对情绪劳动策略的影响 [J]. 心理学报, 2019, 51(03): 353-365.

[41]田水承, 杨鹏飞, 李磊, 等. 矿工不良情绪影响因素及干预对策研究 [J]. 矿业安全与环保, 2016, 43(06): 99-102.

[42]沈剑, 李红霞. 矿工作业疲劳对煤矿险兆事件的影响机理——基于情感耗竭中介变量的分析 [J]. 安全与环境学报, 2019, 19(02): 527-534.

[43]许莎, 庞淑燕, 何晓丽, 等. 不同民族文化对负性情绪调节的影响：来自ERPs的证据 [J]. 心理科学, 2021, 44(05): 1201-1207.

[44]CHATURVEDI V, KAUR A B, VARSHNEY V, et al. Music mood and human emotion recognition based on physiological signals: a systematic review [J]. Multimedia Systems, 2021, 28(01): 21-44.

[45]李广利. 矿工不安全情绪形成路径研究 [J]. 安全与环境学报, 2023, 23(11): 4056-4064.

[46]郭晓栋, 郑泓, 阮盾, 等. 认知和情感共情与负性情绪：情绪调节的作用机制 [J]. 心理学报, 2023, 55(06): 892-904.

[47]FU W, LUO Z M, WANG J G, SHU C M. Structural equation model of the influence of noise annoyance on miners' unsafe behavior in coal mines [J]. Journal of Loss Prevention in the Process Industries, 2023, 82: 104978.

[48]屈慧君. 突发事件网络舆情中的负性情绪表达与疏导 [J]. 郑州大学学报(哲学社会科学版), 2022, 55(01): 121-125.

[49]李广利, 彭斌. 基于SIRS模型的矿工不良情绪传播研究 [J]. 西安科技大学学报, 2022, 42(06): 1096-1103.

[50]李乃文, 冀永红, 唐水清, 等. 高危岗位矿工情绪稳定性影响机制SD仿真 [J]. 中国安全生产科学技术, 2018, 14(05): 167-173.

[51]张思慧, 卞诚, 买晓琴, 等. 睡眠对违反道德的负性情绪记忆的调节 [J]. 科学通报, 2021, 66(13): 1642-1652.

[52]李王杰, 芦金峰. 初中生参加学校体育锻炼对负性情绪的影响——基于CEPS数据的实证分析 [J]. 沈阳体育学院学报, 2023, 42(01): 29-35.

[53]GUO Y Y, LI J Z. The effect of emotional intelligence on unsafe behavior of miners: the role of emotional labor strategies and perceived organizational support [J]. Int J Occup Saf Ergon, 2023, 29(02): 515-527.

[54]LI W, ZHANG Z, SONG A G. Physiological-signal-based emotion recognition: An odyssey from methodology to philosophy [J]. Measurement, 2021, 172: 108747.

[55]阮凯. 情绪识别技术的问题、风险与规治 [J]. 自然辩证法通讯, 2023, 45(02): 82-90.

[56]YAN W, LIU X J, SHAN B A, et al. Research on the Emotions Based on Brain-Computer Technology: A Bibliometric Analysis and Research Agenda [J]. Front Psychol, 2021, 12: 771591.

[57]PINTI P, TACHTSIDIS I, HAMILTON A, et al. The present and future use of functional near-infrared spectroscopy (fNIRS) for cognitive neuroscience [J]. Annals of the New York Academy of Sciences, 2020, 1464(01): 5-29.

[58]HOSHI Y, HUANG J, KOHRI S, et al. Recognition of human emotions from cerebral blood flow changes in the frontal region: a study with event-related near-infrared spectroscopy [J]. J Neuroimaging, 2011, 21(02): 94-101.

[59]BALCONI M, GRIPPA E, VANUTELLI M E. Resting lateralized activity predicts the cortical response and appraisal of emotions: an fNIRS study [J]. Social Cognitive and Affective Neuroscience, 2015, 10(12): 1607-1614.

[60]BALCONI M, VANUTELLI M E. Emotions and BIS/BAS components affect brain activity (ERPs and fNIRS) in observing intra-species and inter-species interactions [J]. Brain Imaging and Behavior, 2016, 10(03): 750-760.

[61]YU J, ANG K K, HO H S, et al. Prefrontal cortical activation while viewing urban and garden scenes: A pilot fNIRS study [J]. Annual International Conference of the IEEE Engineering in Medicine and Biology Society IEEE Engineering in Medicine and Biology Society Annual International Conference, 2017, 2017: 2546-2549.

[62]王星. 不同情绪下过度网络使用者前额叶认知功能的fNIRS研究 [D]. 昆明: 昆明理工大学, 2020.

[63]BALTERS S, LI R, ESPIL F M, et al. Functional near-infrared spectroscopy brain imaging predicts symptom severity in youth exposed to traumatic stress [J]. J Psychiatr Res, 2021, 144: 494-502.

[64]王会会. 大学生早期应激的特点及对情绪工作记忆的影响 [D]. 郑州: 河南大学, 2021.

[65]SKLIVANIOTI GREENFIELD M, WANG Y, MSGHINA M. Similarities and differences in the induction and regulation of the negative emotions fear and disgust: A functional near infrared spectroscopy study [J]. Scand J Psychol, 2022, 63(06): 581-593.

[66]YAO L Y, ZHOU L, QIAN Z Y, et al. Exploring the impact of 3D movie watching on the brain source activities and energy consumption by ESI and fNIRS [J]. Biomedical Signal Processing and Control, 2022, 71: 103194.

[67]王玲莉. 不同情绪对大学生工作记忆的影响：结合功能性近红外脑成像（fNIRS）的研究 [D]. 贵阳: 贵州医科大学, 2022.

[68]NASEER N, HONG K S. fNIRS-based brain-computer interfaces: a review [J]. Frontiers in Human Neuroscience, 2015, 9: 3.

[69]BANDARA D, VELIPASALAR S, BRATT S, HIRSHFIELD L. Building predictive models of emotion with functional near-infrared spectroscopy [J]. International Journal of Human-Computer Studies, 2018, 110: 75-85.

[70]HU X, ZHUANG C, WANG F, et al. fNIRS Evidence for Recognizably Different Positive Emotions [J]. Frontiers in Human Neuroscience, 2019, 13: 120.

[71]DUAN L, FENG Q, XU P F. Using Functional Near-Infrared Spectroscopy to Assess Brain Activation Evoked by Guilt and Shame [J]. Frontiers in Human Neuroscience, 2020, 14: 197.

[72]MA T F, LYU H, LIU J J, et al. Distinguishing Bipolar Depression from Major Depressive Disorder Using Fnirs and Deep Neural Network [J]. Progress In Electromagnetics Research, 2020, 169: 73-86.

[73]吴文一. 基于EEG-fNIRS特征融合的多强度负性情绪识别研究 [D]. 天津: 天津大学, 2021.

[74]许博俊, 李梦琪. 基于VR-fNIRS的平静、沮丧和恐惧情绪的分类研究 [J]. 现代电子技术, 2023, 46(22): 121-125.

[75]SI X P, HE H, YU J Y, et al. Cross-Subject Emotion Recognition Brain-Computer Interface Based on fNIRS and DBJNet [J]. Cyborg Bionic Syst, 2023, 4: 0045.

[76]李锦瑶, 杜肖兵, 朱志亮, 等. 脑电情绪识别的深度学习研究综述 [J]. 软件学报, 2023, 34(01): 255-276.

[77]权学良, 曾志刚, 蒋建华, 等. 基于生理信号的情感计算研究综述 [J]. 自动化学报, 2021, 47(08): 1769-1784.

[78]吕宝粮, 张亚倩, 郑伟龙. 情感脑机接口研究综述 [J]. 智能科学与技术学报, 2021, 3(01): 36-48.

[79]SLAMA M E. Emotions and life: Perspectives from psychology, biology, and evolution [J]. Psychology and Marketing, 2005, 22(01): 97-101.

[80]秦天鹏, 生慧, 岳路, 等. 脑电信号情绪识别研究综述 [J]. 计算机工程与应用, 2023, 59(15): 38-54.

[81]HWANG S, JEBELLI H, CHOI B, et al. Measuring Workers’ Emotional State during Construction Tasks Using Wearable EEG [J]. Journal of Construction Engineering and Management, 2018, 144(07): 04018050.

[82]朱朝喆, 杨玉芳. 认知神经科学书系 近红外光谱脑功能成像 [M]. 北京: 科学出版社, 2020.

[83]周志华. 机器学习 [M]. 北京: 清华大学出版社, 2016.

[84]周志华, 李楠. 集成学习：基础与算法 周志华教授专著森林书 机器学习之利器 [M]. 北京: 电子工业出版社, 2020.

[85]邢雪娇. 基于认知的建筑工人心理负荷响应及干预方案研究 [D]. 武汉: 华中科技大学, 2021.

[86]VEERANKI Y R, KUMAR H, GANAPATHY N, et al. A Systematic Review of Sensing and Differentiating Dichotomous Emotional States Using Audio-Visual Stimuli [J]. Ieee Access, 2021, 9: 124434-124451.

[87]SUN Y, AYAZ H, AKANSU A N. Multimodal Affective State Assessment Using fNIRS + EEG and Spontaneous Facial Expression [J]. Brain Sciences, 2020, 10(02): 85.

[88]IYILIKCI E A, BOGA M, YUVRUK E, et al. An extended emotion-eliciting film clips set (EGEFILM): assessment of emotion ratings for 104 film clips in a Turkish sample [J]. Behav Res Methods, 2024, 56(02): 529-562.

[89]AASTED C M, YUCEL M A, COOPER R J, et al. Anatomical guidance for functional near-infrared spectroscopy: AtlasViewer tutorial [J]. Neurophotonics, 2015, 2(02): 020801.

[90]HUPPERT T J, DIAMOND S G, FRANCESCHINI M A, et al. HomER: a review of time-series analysis methods for near-infrared spectroscopy of the brain [J]. Appl Opt, 2009, 48(10): D280-D298.

[91]HOU X, ZHANG Z, ZHAO C, et al. NIRS-KIT: a MATLAB toolbox for both resting-state and task fNIRS data analysis [J]. Neurophotonics, 2021, 8(01): 010802.

[92]FISHBURN F A, LUDLUM R S, VAIDYA C J, et al. Temporal Derivative Distribution Repair (TDDR): A motion correction method for fNIRS [J]. Neuroimage, 2019, 184: 171-179.

[93]M R-Š, I K. Theoretical and empirical analysis of ReliefF and RReliefF [J]. Machine learning, 2003, 53: 23-69.

[94]许博俊, 刘化东, 李梦琪. 基于fNIRS的恐惧情绪分级研究 [J]. 现代电子技术, 2023, 46(20): 142-146.