情感数据库建立及情感特征提取算法研究

发布时间：2018-05-28 20:53

本文选题：心电情感数据库 + 小波变换　；参考：《燕山大学》2014年硕士论文

【摘要】：情感数据获取的有效性与合理性是认知情感计算研究中的关键问题，因此，建立性能良好的情感计算数据库是情感计算研究的重要部分，也是该领域学者研究的热点。心电情感信号是非常重要的生理信号，大量研究已证明其与人体情感状态有极大相关性，并且，心电信号在临床应用广泛，采集分析技术相对成熟。因此，本文通过情感诱发实验，采集心电情感信号，建立心电信号情感数据库。研究发现影视片段比图片和音乐等素材更能成功诱发人的情感，本文从大量的影视素材中剪辑出能够诱发特定情感的电影片段进行情感诱发。情感诱发实验采集到身体健康、无心脏病史的25名学生，分别在高兴、愤怒、悲伤、恐惧种情感状态下的心电信号。从心电情感信号中截取出60秒有效数据作为样本，通过小波变换进行去除基线漂移等预处理，建立心电信号的情感数据库。情感特征提取直接影响情感分析与识别效果，心率变异性反映了人体内外环境对心血管系统的扰动以及心血管系统通过自主神经及体液调节对这种扰动的反应，对不同生理状态甚至情感状态的变化都比较敏感，蕴含了有关情感的大量信息，，因此，能够用来进行情感识别。本文提出一种基于小波变换与独立成分分析结合的算法来进行心率变异性时频分析，共获得21个心率变异性特征。同时结合提取的心电信号时域特征79个、心电信号小波特征36个，常规心率变异时频特征21个，共提取出157个心电情感特征。对情感特征集进行心电情感识别分析，并进一步采用Relief-F、最大相关最小冗余算法、主成分分析和独立成分结合算法进行特征选择及优化组合，并采用遗传算法优化后的支持向量机分别进行单一情感识别，平均识别正确率均达到90%以上。结果表明，本文建立的心电信号情感数据库具有较好性能，并且提取的情感特征具有较强的情感识别能力。与常规心率变异性特征、心电时域特征、心电小波特征进行比较，该算法提取的心率变异性特征对情感识别同样取得较显著的识别能力。
[Abstract]:The validity and rationality of emotional data acquisition is a key issue in cognitive affective computing. Therefore, the establishment of an affective computing database with good performance is an important part of affective computing, and also a hot topic for researchers in this field. Electrocardiogram (ECG) emotional signal is a very important physiological signal. A large number of studies have proved that it has a great correlation with the emotional state of human body. Furthermore, ECG signal is widely used in clinical practice, and the technology of collection and analysis is relatively mature. Therefore, this paper collects ECG emotional signals and establishes ECG emotional database by affective induction experiment. The study found that film and television clips can induce human emotion more successfully than pictures and music. The affective induction experiment collected ECG signals from 25 healthy students who had no history of heart disease in the emotional states of joy anger sadness and fear. The effective data of 60 seconds was extracted from the ECG emotion signal as a sample, and the baseline drift was removed by wavelet transform to establish the emotion database of ECG signal. Emotion feature extraction directly affects the effect of emotion analysis and recognition. Heart rate variability reflects the disturbance of cardiovascular system in human body and the response of cardiovascular system to this disturbance through autonomic nerve and body fluid regulation. It is sensitive to the changes of different physiological states and even emotional states, and contains a lot of information about emotion, so it can be used for emotion recognition. In this paper, an algorithm based on wavelet transform and independent component analysis is proposed to analyze heart rate variability (HRV). A total of 21 HRV features are obtained. At the same time, 79 ECG temporal features, 36 ECG wavelet features and 21 time-frequency features of conventional heart rate variability were extracted. A total of 157 ECG emotion features were extracted. The emotion feature set is analyzed by ECG emotion recognition, and further, Relief-F, maximum correlation and minimum redundancy algorithm, principal component analysis and independent component combination algorithm are used to select and optimize the feature. The support vector machine (SVM), which is optimized by genetic algorithm, is used for single emotion recognition, and the average recognition accuracy is over 90%. The results show that the ECG emotion database established in this paper has a good performance and the extracted emotion features have a strong ability of emotion recognition. Compared with conventional heart rate variability (HRV), ECG time domain (ECG) and ECG wavelet (ECG wavelet), the HRV feature extracted by this algorithm can also achieve significant recognition ability for emotion recognition.
【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】：TP311.13;TN911.7

【参考文献】