情绪的生理特征及特征变异性研究

发布时间：2018-04-14 23:00

本文选题：情绪 + 生理特征　；参考：《山东大学》2017年硕士论文

【摘要】：心理疾病发病率逐年升高,临床上心理疾病的预防与诊断主要根据心理医生的经验判断,缺乏更加科学客观的评估依据。情绪识别能够科学、客观地诊断与预防心理疾病,为后期的疾病治疗提供可靠依据。基于生理信号科学客观地评估每个个体的情绪状态,对心理疾病的预防和诊断具有重要的社会意义和临床价值。但现有的情绪识别研究更多地关注分类率的提高,通过使用不同的机器学习算法进行特征选择和情绪分类以提高分类正确率,缺乏生理信号的固有特性和情绪状态之间的特异性研究。针对目前生理特征与情绪状态的特异性关系尚不明确的问题,本文研究了不同情绪状态的生理特征及特征变异性。论文以心率变异性和脉搏传播时间序列变异性为研究对象,通过提取RR间期和脉搏传播时间序列的时域、频域和非线性特征,探讨分析不同情绪状态下生理信号变异性特征的差异性,最后构造基于生理信号变异性特征、网格搜索参数寻优的支持向量机分类器进行情绪分类。本文的主要研究工作如下:(1)计算心率变异性和脉搏传播时间序列变异性特征。使用symlet8小波阈值去噪的方法对心电、脉搏信号进行去噪,采用自适应差分阈值法检测心电信号的R波,并使用Wapb方法检测脉搏波的起始点,进而构造RR间期和PTT时间序列,并计算了心率变异性的9个特征:rrMEAN、rrSDNN、rrRMSSD、rrPNN50、rrLFn、rrHFn、rrLF/HF、rrSampEn 和 rrFuzzyMEn,以及脉搏传播时间序列变异性的 8 个特征:pttMEAN、pttSDNN、pttRMSSD、pttLFn、pttHFn、pttLF/HF、pttSampEn 和 pttFuzzyMEn。(2)对比研究两种截然相反情绪状态下的心率变异性指标和脉搏传播时间序列变异性指标。设计严谨的情绪生理信号采集实验,同步采集高兴和悲伤情绪的心电、脉搏信号;计算两种情绪下的HRV、PTTV指标,并对得到的17个变异性指标进行差异性和相关性分析。研究结果表明,6个HRV指标rrMEAN、rrSDNN、rrLFn、rrHFn、rrLF/HF、rrFuzzyMEn,以及 5 个 PPTV 指标 pttRMSSD、pttLFn、pttHFn、pttLF/HF、pttFuzzyMEn在高兴悲伤两种相对情绪之间具有显著性差异,且两种情绪下的HRV、PTTV的频域、非线性指标具有一致的变化趋势。(3)探究多种情绪对生理信号变异性指标的影响。设计更为完备的中性、高兴、悲伤、恐惧、愤怒和厌恶六种情绪下信号采集实验;然后计算六种情绪状态下的HRV、PTTV指标,以中性情绪为基准,比较多种情绪与中性情绪下序列变异性指标的统计差异性。结果发现,HRV、PTTV的时域指标在某些情绪之间具有显著性差异,但不存在某一种时域指标在所有情绪之间均有显著性差异;HRV、PTTV的频域指标具有显著性差异的是高兴与中性情绪,而中性与负性情绪之间没有明显差异;对于非线性分析,HRV指标中除了中性与高兴情绪、中性与恐惧情绪之间的rrFuzzyMEn没有显著性差异外,其他均有显著性差异,PTTV非线性指标中仅有高兴情绪的pttFuzzyMEn指标与中性情绪具有显著性差异,而负性情绪与中性情绪之间没有明显差异。(4)基于生理信号变异性特征、网格搜索算法和SVM构建了情绪识别模型。本文分别以提取的所有变异性特征和具有显著性差异的变异性特征作为特征向量,通过GSA-SVM训练并建立了不同情绪状态的识别决策器,对比两组特征作为输入特征向量的识别效果。结果表明,单纯使用所有特征,和单纯使用有统计差异的特征,两种方式的识别效果差别不大,使用生理信号变异性特征进行情绪识别是行之有效的。
[Abstract]:Mental disease incidence increased year by year, the clinical prevention and diagnosis of mental illness according to the doctor's psychological experience, the lack of assessment more scientific and objective basis. Emotion recognition can be scientific, objective diagnosis and prevention of mental illness, and provide a reliable basis for the treatment of late stage disease. Physiological signal scientific and objective assessment of each individual's emotion. Based on the social significance and important clinical value in prevention and diagnosis of mental illness. But the study of emotion recognition of existing pay more attention to improve the classification rate, by using different machine learning algorithms for feature selection and classification of emotions in order to improve the classification accuracy, the lack of specific physiological signal characteristic and emotion state. Aiming at the specific characteristics and the relationship between the physiological and emotional state is not clear the problem, this paper studies the different emotional state The physiological characteristics and variability characteristics. Based on the heart rate variability and pulse propagation time sequence variability as the research object, through the time domain RR interval and pulse propagation time sequence, frequency domain and nonlinear characteristics, analysis of differences in physiological signal variability characteristics in different emotional states, the final structure of physiological variability based on the grid search support vector machine classifier parameter optimization for sentiment classification. The main contents are as follows: (1) the calculation of heart rate variability and pulse propagation time sequence variability. Methods using symlet8 wavelet threshold denoising of ECG, pulse signal denoising, R wave using adaptive differential detection of ECG signal sub threshold method, and use the Wapb method to detect the starting point of the pulse wave, and then construct the RR interval and PTT time series, the 9 characteristics of heart rate variability and calculation: rrMEAN, RR SDNN, rrRMSSD, rrPNN50, rrLFn, rrHFn, rrLF/HF, rrSampEn and rrFuzzyMEn, as well as 8 variability of pulse propagation time sequence: pttMEAN, pttSDNN, pttRMSSD, pttLFn, pttHFn, pttLF/HF, pttSampEn and pttFuzzyMEn. (2) of the two opposite emotion condition index of heart rate variability and pulse propagation time sequence the variability of the index contrast. Emotional physiological signal acquisition and rigorous experimental design, synchronous acquisition of grief and joy, ECG, pulse signal; calculating the two emotions under the HRV, PTTV index, and analyzed the difference and correlation between the 17 variability index. The results showed that 6 HRV rrMEAN index, rrSDNN rrLFn, rrHFn, rrLF/HF, rrFuzzyMEn, and PPTV, 5 pttLFn, pttHFn, pttRMSSD, pttLF/HF, pttFuzzyMEn in two kinds of emotions between relatively happy sadness has significant differences, and two emotions under HRV PTTV, the frequency domain, the nonlinear index had the same trend. (3) to explore the impact of emotion on various physiological signal variability. The design is more complete neutral, happy, sad, fear, anger and disgust six emotion signal acquisition experiment; and then calculate the six emotional state of the HRV, PTTV index and to the neutral emotion as the benchmark, statistical differences of emotion and neutral emotion sequence variability index of comparison. The results showed that HRV, the time domain indexes of PTTV have significant differences in some mood, but there is no one time domain index showed significant differences in all emotions between HRV, the frequency domain index PTTV; the significant difference was happy and neutral emotion, and there is no significant difference between the neutral and negative emotions; for the nonlinear analysis, the HRV index in addition to neutral and happy mood, between neutral and emotional rrFuzzyMEn fear No significant differences, other significant differences, PTTV nonlinear index only happy mood pttFuzzyMEn index had significant differences with neutral emotion, and there is no significant difference between negative emotion and neutral emotion. (4) physiological variability based on grid search algorithm and SVM model. This paper constructs emotion recognition to extract all variability and significant differences in the variability as the feature vector by the GSA-SVM training and established the recognition decision maker different emotional state, characteristics between the two groups as input vectors of the recognition results. The results show that the simple use of all the features, characteristics and use simple statistical differences. Little difference between two kinds of recognition results, using physiological variability of emotion recognition is effective.

【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R33;B842.6

【相似文献】