国际中文开源期刊平台

基于多人脑电一致性特征的情绪识别探索

Exploring Emotion Recognition Using EEG-based Inter- subject Correlation Features

中国心理学前沿 / 2019,1(3)：187-201 / 2019-05-31 6207 9345

查看全文 下载文档

• 作者： 胡鑫¹      喻纯²      张丹¹     
• 单位：

  1. 清华大学社会科学学院心理学系，北京，100084；
  2. 清华大学计算机科学与技术系，北京，100084

• 关键词： 脑电；多人一致性；情绪；分类
• EEG; Inter-subject correlation; Emotion; Classification
• 摘要： 寻找不同情绪状态的脑神经表征，对于情绪的认知神经机制研究与情绪识别等现实应用具有重要意义。现有基于脑电的情绪识别研究方法大多采用经典的事件相关分析方法，难以应用于包含连续视觉—听觉信息的自然情绪刺激情境。本研究探索了多人脑电一致性特征用于情绪识别的效力。多人脑电一致性特征关注多人在同一复杂视听刺激环境下的脑电响应时间过程的相似性，从有别于事件相关方法的多人视角刻画了大脑神经响应特性。研究发现，32名被试观看40段情绪诱发视频时，被试的左侧顶叶及前额叶脑电响应的多人一致性特征分别与主观情绪评分的唤起度和效价有显著相关（唤起度r=0.41，p=0.008；效价r=0.37，p=0.017）。进一步研究发现，基于多人脑电一致性特征对情绪唤起度和效价主观评分高/低的二分类所得正确率分别为77.5%和70.0%，而基于相应的单人脑电特征所得分类正确率仅为随机水平（44.1±13.9%和48.6±12.7%）。本研究结果首次展示了多人脑电一致性特征在情绪识别领域的效力与应用潜力。
• Human emotion plays an important role in our daily life. EEG-based emotion recognition is expected to facilitate our understanding of the neural mechanisms as well as applications of emotion recognition. Most EEG-based emotion recognition studies have employed the classical event-related approach, which may not be the most suitable tool for naturalistic situations with continuous audio-visual emotional information streams.In the present study, we explored emotion recognition using EEG-based inter-subject correlation (ISC) features. ISC measures the consistency of neural responses across a group of participants exposed to identical complex and continuous stimuli, characterizing the neural responses from a multi-person perspective. Using a publicly available EEG-based emotion database named DEAP, in which 32 participants watched 40 video clips with different emotional properties, ISCs over left parietal cortex and frontal region are found to be significantly correlated with arousal ratings and valence ratings, respectively (arousal: r=0.41, p=0.008; valence: r=0.37, p=0.017) . In addition, using ISCs as features, binary classification accuracies for arousal and valence reached 77.5% and 70%, which are superior to the traditional individual spectral power based method (44.1±13.9% and 48.6±12.7%). Our results suggest that the inter-subject correlation approach as an effective and promising candidate for investigating human emotion experiences.
• DOI： https://doi.org/10.35534/pc/0103013
• 引用： 胡鑫，喻纯，张丹． 基于多人脑电一致性特征的情绪识别探索［J］.中国心理学前沿，2019，1 （3）：187-201.
  https://doi.org/10.35534/pc/0103013