脑电数据分析方法及其在压力情感状态评估中的应用

发布时间：2018-03-23 13:57

本文选题：脑电　切入点：压力评估　出处：《燕山大学》2014年硕士论文

【摘要】：长期的压力状态，容易导致各种疾病，情感应激状态评估是合理和有效的压力干预基础。本文以压力情感评估为研究目标，充分利用脑电数据中蕴含丰富情感信息的特点，结合复杂度、熵理论以及多重分形方法，实现了压力情感状态的分析评估，，设计并完成了基于脑电的压力情感评价系统。针对压力情感脑电数据特征提取的问题，本文提出了一种基于复杂度与熵相结合的方法。KC复杂度可以描述脑电数据的随机程度，小波熵和近似熵分别能够在时频两域中对脑电数据的复杂程度和能量分布状况进行量化；采用遗传算法改进的支持向量机融合以上3种特征参数，压力情感状态评估得以实现。本文中采集了14名被试总共92组压力脑电数据。基于上述三种特征融合算法对被试的压力情感状态进行分析评估，最高准确率为94.12%，平均准确率为82.06%。研究同时表明，不同脑区对压力敏感程度不同，左半球相对右半球来说，压力感受敏感。对情感脑电数据分析的结果显示，本文提出的近似熵、小波熵特征融合方法优于传统的统计学特征，比Sander Koelstra的频率能量特征方法唤醒度的分类准确率高出7.49%。脑电的多重分形谱为脑电的非线性分析提供了一个有效方法。根据信号的奇异性大小分析比较发现，压力脑电信号的奇异谱宽度大于无压力脑电信号的奇异谱宽度，这说明人们在不同压力状态下，脑电的多重分形特性不同，压力越小，脑电的复杂度趋于减弱。同样，人类不同情感状态下，脑电的多重分形特性也不同，基于多重分形谱特征对情感脑电进行分类，分类结果较现有研究结果最多高出28.88%，说明我们对数据的有效筛选和特征提取，在情感识别中是可行的。在Visual Studio2008编译器环境下利用C#和MATALB混合编程开发压力情感状态评估系统。对相关脑电数据进行预处理、特征提取、分类识别，最后得出被测者的情绪压力状态。
[Abstract]:The assessment of emotional stress state is a reasonable and effective basis for stress intervention. The purpose of this study is to make full use of the characteristics of EEG data containing rich emotional information. Combined with complexity, entropy theory and multifractal method, the analysis and evaluation of stress emotion state are realized, and a stress emotion evaluation system based on EEG is designed and completed. Aiming at the problem of feature extraction of pressure-emotional EEG data, a method based on complexity and entropy is proposed in this paper. KC complexity can describe the random degree of EEG data. Wavelet entropy and approximate entropy can quantify the complexity and energy distribution of EEG data in time-frequency domain respectively. In this paper, we collected a total of 92 groups of stress-EEG data from 14 subjects. Based on the above three feature fusion algorithms, we analyzed and evaluated the pressure-emotional state of the subjects. The highest accuracy rate is 94.12 and the average accuracy is 82.06.The study also shows that different brain regions are sensitive to stress, and the left hemisphere is more sensitive to stress than the right hemisphere. The wavelet entropy feature fusion method is superior to the traditional statistical feature, and the classification accuracy is 7.49% higher than that of Sander Koelstra's frequency energy feature method. Multifractal spectrum of EEG provides an effective method for nonlinear analysis of EEG. According to the analysis of singularity of signals, it is found that the singular spectrum width of pressure-EEG signals is larger than that of pressure-free EEG signals. This shows that the multifractal characteristics of EEG are different in different stress states. The lower the pressure is, the more complexity of EEG tends to weaken. Similarly, the multifractal characteristics of EEG are different in different emotional states of human beings. The classification of affective EEG based on multifractal spectrum features shows that the result of classification is 28.88% higher than that of existing research, which shows that our effective data selection and feature extraction are feasible in emotion recognition. In the environment of Visual Studio2008 compiler, C # and MATALB are used to develop the stress emotional state evaluation system. The related EEG data are preprocessed, feature extracted, classified and identified. Finally, the emotional stress state of the subjects is obtained.
【学位授予单位】：燕山大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.7

【参考文献】