| 引用本文: | 罗映雪,贾博,裘旭益,邓平煜,吴奇.层次狄利克雷过程隐半马尔科夫模型识别飞行员脑疲劳状态[J].控制理论与应用,2020,37(6):1196~1206.[点击复制] |
| LUO Ying-xue,JIA Bo,QIU Xu-yi,DENG Ping-yu,Wu Qi.Using hidden semi-Markov model with hierarchical Dirichlet process to infer pilots’ fatigue states[J].Control Theory and Technology,2020,37(6):1196~1206.[点击复制] |
|
| 层次狄利克雷过程隐半马尔科夫模型识别飞行员脑疲劳状态 |
| Using hidden semi-Markov model with hierarchical Dirichlet process to infer pilots’ fatigue states |
| 摘要点击 2477 全文点击 799 投稿时间：2019-05-02 修订日期：2019-10-25 |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| DOI编号 10.7641/CTA.2019.90311 |
| 2020,37(6):1196-1206 |
| 中文关键词 脑电信号 飞行员疲劳 基于剩余寿命的隐半马尔科夫模型 光滑伪仿射维格纳-维勒分布 |
| 英文关键词 electroencephalogram signals pilots’ fatigue hidden semi-Markov model smooth pseudo-affine Wigner- Ville distribution |
| 基金项目 国家自然科学基金项目(61671293), 国家自然科学民航联合基金项目(U1933125)资助. |
|
| 中文摘要 |
| 在民航及军用航空领域, 长期恶劣飞行环境的影响将导致飞行员精神不集中, 产生疲劳现象, 严重影响飞行员的判断能力. 因此, 研究飞行员的脑疲劳状态对飞行安全来说非常重要. 脑疲劳推理主要面临二个基本问题: 一是如何提取脑疲劳认知的特征, 二是如何识别带驻留时间的脑疲劳认知潜在状态. 针对第一个问题, 本文提出一种基于Kaiser窗函数的光滑伪仿射维格纳-维勒分布的方法, 提取主要认知脑区的脑电节律的瞬时频谱特征. 并与其他时频分布的特征进行对比, 本文的特征具有明显的局部显著性. 针对第二个问题, 本文建立一种基于剩余寿命的隐半马尔科夫模型 (HSMM) 的飞行员脑疲劳认知动力学推理模型, 解决了传统隐马尔科夫模型潜状态快速切换及无法对潜状态驻留时间进行建模的问题. 飞行员脑疲劳认知行为是由多通道的脑节律融合的整体表现行为, 建立多通道共享狄利克雷过程先验参数的层次学习网络, 形成共享疲劳认知主题的子任务学习机制. 实验结果显示本文的模型具有较高的推理飞行员脑疲劳状态的能力, 具有广泛的应用价值. |
| 英文摘要 |
| In the civil aviation and military aviation field, the long flight and harsh conditions will make pilots feel
fatigued. The pilots’ fatigue will seriously affect their judgment, so the inference of fatigue state is important to ensure
flight safety. Brain fatigue cognition faces two main problems: one is how to extract feature of fatigue cognition, and
the other is how to identify the latent state of brain cognition with duration. For the first problem, a method based on
smooth pseudo-affine Wigner-Ville distribution (SPAWVD) with kaiser window function was proposed to calculate the
instantaneous spectral features of Electroencephalogram (EEG) rhythms. Features extracted by this method had better
local significance. For the second problem, a residual life hidden semi-Markov model (HSMM) was established to learn the
dynamic mechanism of the brain. It modeled the duration of brain fatigue cognitive states and avoids fast switching between
states caused by hidden semi-Markov model (HMM). Then a multi-layer learning network based on hierarchical Dirichlet
process (HDP) was built to provide subtasks that share the subject of fatigue awareness. The result of the experiment was
satisfactory and proved that the model had a high ability to identify pilots’ latent brain cognitive state. |
|
|
|
|
|