基于fNIRS的恐惧情绪分级研究

Research on fear emotion grading based on fNIRS

功能性近红外光谱技术(fNIRS)能够实现对恐惧情绪的分级。文中设计情绪诱发范式,并对恐惧程度进行分级,将恐惧水平分为三级(无恐惧、弱恐惧和强恐惧)。其次,采集20位受试者在三种情绪诱发视频下的fNIRS实验数据,采用支持向量机(SVM)、K近邻算法(KNN)和随机森林(RF)等三种算法作为分类器,提取fNIRS领域常用统计学特征和熵特征进行比较研究。结果表明,常用统计学特征最高准确率达到84%,而通过集合经验模态分解(EEMD)分解的模糊熵(FuEn)特征最终获得的准确率高达93.98%。研究结果表明,通过EEMD分解的FuEn是一种相较于常用统计学特征更加优秀的恐惧情绪分级特征,可为后续其他情绪的分级奠定基础。

Functional near⁃infrared spectroscopy(fNIRS)can realize the grading of fear emotions.The emotion induction paradigm is designed,and the fear degree is graded and divided into three levels(no fear,weak fear and strong fear).The fNIRS experimental data of 20 subjects under three kinds of emotion induction videos is collected,support vector machine(SVM),K⁃nearest neighbor(KNN)algorithm and random forest(RF)are used as classifiers,and the statistical features and entropy features commonly used in the field of fNIRS are extracted for comparative research.The results show that the highest accuracy rate of commonly used statistical features is 84%,while the highest accuracy rate of fuzzy entropy(FuEn)features decomposed by ensemble empirical mode decomposition(EEMD)is as high as 93.98%.The research results indicate that FuEn decomposed by EEMD is a superior fear emotion grading feature compared with commonly used statistical features,which can lay the foundation for subsequent grading of other emotions.