DOI: 10.11992/tis.202001032
多模态情绪识别研究综述
潘家辉
1
,何志鹏
1
,李自娜
2
,梁艳
1
,邱丽娜
1
(1. 华南师范大学 软件学院,广东 佛山 528225; 2. 华南师范大学 计算机学院,广东 广州 510641)
摘 要:本文针对多模态情绪识别这一新兴领域进行综述。首先从情绪描述模型及情绪诱发方式两个方面对
情绪识别的研究基础进行了综述。接着针对多模态情绪识别中的信息融合这一重难点问题,从数据级融合、特
征级融合、决策级融合、模型级融合4种融合层次下的主流高效信息融合策略进行了介绍。然后从多种行为表
现模态混合、多神经生理模态混合、神经生理与行为表现模态混合这3个角度分别列举具有代表性的多模态混
合实例,全面合理地论证了多模态相较于单模态更具情绪区分能力和情绪表征能力,同时对多模态情绪识别方
法转为工程技术应用提出了一些思考。最后立足于情绪识别研究现状的分析和把握,对改善和提升情绪识别
模型性能的方式和策略进行了深入的探讨与展望。
关键词:情绪识别;情绪描述模型;情绪诱发方式;信息融合;融合策略;情绪表征;模态混合
中图分类号:TP391.4 文献标志码:A 文章编号:1673−4785(2020)04−0633−13
中文引用格式:潘家辉, 何志鹏, 李自娜, 等. 多模态情绪识别研究综述[J]. 智能系统学报, 2020, 15(4): 633–645.
英文引用格式:PAN Jiahui, HE Zhipeng, LI Zina, et al. A review of multimodal emotion recognition[J]. CAAI transactions on in-
telligent systems, 2020, 15(4): 633–645.
A review of multimodal emotion recognition
PAN Jiahui
1
,HE Zhipeng
1
,LI Zina
2
,LIANG Yan
1
,QIU Lina
1
(1. School of Software, South China Normal University, Foshan 528225, China; 2. School of Computer, South China Normal Uni-
versity, Guangzhou 510641, China)
Abstract: This paper reviews the emerging field of multimodal emotion recognition. Firstly, the research foundation of
emotion recognition is summarized from two aspects: emotion description model and emotion-inducing mode. Then,
aiming at the key and difficult problem of information fusion in multi-modal emotion recognition, some mainstream and
high-efficiency information fusion strategies are introduced from four fusion levels: data-level fusion, feature-level fu-
sion, decision-level fusion, and model-level fusion. By exemplifying representative multi-modal mixing examples from
three perspectives: the mixing of multiple external presentation modalities, the mixing of multiple neurophysiological
modalities, and the mixing of neurophysiology and external presentation modalities, it fully demonstrates that multi-
modality is more capable of emotional discrimination and emotional representation than single-modality. At the same
time, some thoughts on the conversion of multi-modal recognition methods to engineering technology applications are
put forward. Finally, based on the analysis and grasp of the current situation of emotion recognition research, the ways
and strategies for improving and enhancing the performance of the emotion recognition models are discussed and pro-
spected.
Keywords: emotion recognition; emotion description model; emotion inducing mode; information fusion; fusion
strategy; emotion representation; modality blend
1 相关研究
1.1 背景与研究意义
情绪,是一系列主观认知经验的高度概括,由
收稿日期:2020−01−30.
基金项目:国家自然科学基金面上项目(61876067);广东省自
然科学基金面上项目(2019A1515011375);广州市科
技计划项目重点领域研发计划项目(202007030005).
通信作者:潘家辉,E-mail:panjh82@qq.com.
第 15 卷第 4 期
智 能 系 统 学 报
Vol.15 No.4
2020 年 7 月
CAAI Transactions on Intelligent Systems
Jul. 2020
评论0