基于心率变异性参数的压力状态描述研究

发布时间：2018-06-28 02:04

本文选题：心率变异性 + 心电信号　；参考：《天津理工大学》2017年硕士论文

【摘要】：生活在竞争激烈的社会,压力时刻伴随着我们,因此压力的自动、快速识别具有重要研究意义。基于生物信号的压力情绪识别方法不易伪装,较易检测出内在的、潜藏的情感,因而被广泛使用。其中,从心电信号(Electrocardiogram,ECG)中提取出的心率变异性(Heart rate variability,HRV)被证明可用于识别压力。HRV由一组参数组成,参数值作为特征被用于支撑识别系统,识别压力系统一般通过建立分类模型完成,而且以随机森林、逻辑回归、朴素贝叶斯等算法为多。但此类系统无法将不同压力状态和HRV特征之间的关系以量化的规则表达出来。为了解不同压力状态下每个HRV特征的近似量化表达,本文引入决策树方法,将特征以组合规则的方式表达出来,而规则中的每个特征也因此获得近似量化,研究利用CART算法生成二叉树,从而获得简单有效的规则,并且获得HRV参数范围。为克服规则冗余,采用基于错误率降低(Reduced Error Pruning,REP)的剪枝算法对规则优化,提取出易理解的规则。实验证明,经优化后规则识别压力的平均识别率达89.4%。经过对出错规则的分析,发现在参数范围值边界上发生错误的频率较高,即刚性边界值在进行压力识别时会缺乏包容性,进而导致规则的识别准确率不高,为加大参数值的包容性,本文提出一种将HRV参数值离散化为三部分的策略,该方法对每个HRV特征按照基于箱线图和正态分布两种方案分别将数据离散,并以此构建树、提取规则并使用REP算法优化,通过实验对比两种方案,基于箱线图策略要优于正态分布,平均识别率达到96.1%。
[Abstract]:Living in a competitive society, stress is always with us, so automatic and rapid recognition of stress is of great significance. The pressure emotion recognition method based on biological signal is not easy to disguise and easy to detect internal and latent emotion, so it is widely used. Among them, heart rate variability (HRV) extracted from electrocardiograms (ECG) has been proved to be used to identify pressure. HRV is composed of a set of parameters, which are used as features to support the recognition system. Moreover, random forest, logical regression, naive Bayes and other algorithms are more. However, such systems can not express the relationship between different pressure states and HRV characteristics in quantitative rules. In order to understand the approximate quantization of each HRV feature under different pressure states, the decision tree method is introduced in this paper. The feature is expressed as a combination rule, and each feature in the rule is approximately quantified. By using cart algorithm to generate binary tree, simple and effective rules are obtained, and HRV parameter range is obtained. To overcome rule redundancy, pruning algorithm based on reduced error pruning (rep) is used to optimize rules and extract rules that are easy to understand. The experimental results show that the average recognition rate of the rule recognition pressure is 89.4%. Through the analysis of the error rules, it is found that the frequency of errors on the boundary of parameter range is higher, that is, the rigid boundary value will not be inclusive in the process of pressure recognition, and the recognition accuracy of the rules will not be high. In order to increase the inclusiveness of parameter values, this paper proposes a strategy of discretization of HRV parameter values into three parts. The method discretizes the data for each HRV feature according to two schemes based on box diagram and normal distribution, and constructs a tree by this method. The rules are extracted and optimized by rep algorithm. By comparing the two schemes through experiments, the strategy based on box diagram is better than normal distribution, and the average recognition rate is up to 96.1%.
【学位授予单位】：天津理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R318

【相似文献】