基于交感神经指数和副交感神经指数的情绪识别
文献类型：学位论文
| 作者 | 向 鋆 |
| 答辩日期 | 2022-06 |
| 文献子类 | 硕士 |
| 授予单位 | 中国科学院大学 |
| 授予地点 | 中国科学院心理研究所 |
| 其他责任者 | 刘正奎 |
| 关键词 | 情绪识别 心电信号 交感神经指数 副交感神经指数 |
| 学位名称 | 理学硕士 |
| 学位专业 | 应用心理 |
| 其他题名 | Emotion recognition based on sympathetic index and parasympathetic index |
| 中文摘要 | Emotion recognition refers to assessing people's emotional states based on their behaviors and reactions. In the field of emotion recognition, emotion recognition based on physiological signals is considered to be more reliable and objective. Among them, heart rate variability (HRV) is a commonly used emotion recognition indicator, but it cannot recognize real-time emotion and is not sensitive to emotional changes in a short period of time. The newly proposed sympathetic activity index (SAI) and parasympathetic activity index (PAI) can respectively assess the influence of activity produced by the two autonomic nervous systems. Importantly, SAI and PAI can simulate the fluctuations of sympathetic and parasympathetic nerves in real time, which indicates that SAI and PAI can be used for emotion recognition and real-time monitoring of emotions. However, whether SAI and PAI can identify emotions has not been tested and remains to be explored and studied. This study is divided into two parts to explore the ability of the emotional valence recognition model constructed based on SAI and PAI to identify different emotional valences in laboratory scenarios and life scenarios. The first research is divided into two parts: pre-experiment and formal experiment. The specific contents of the research are as follows: Study 1 Pre-experiment: Provide emotion-inducing materials for formal experiments. Through online platforms such as Tencent, use keywords such as "funny" and "sad" to search for video materials. After the video was edited, 4 interns were invited to preliminarily screen the edited video according to the degree of understanding and differentiation. Then 55 subjects were recruited to evaluate the videos, and 2 positive videos and 2 negative videos were finally selected. Study 1 Formal experiment: mainly to explore the identification of positive and negative emotional valence in the laboratory based on the emotional valence recognition model based on SAI and PAI. Emotional videos were used to evoke emotions in 95 subjects and their emotions were subjectively scored. At the same time, a wearable wrist sensor is used to collect ECG signals. According to the calculation formula of SAI and PAI, the values of SAI and PAI were calculated from the ECG signal, and 14 features such as standard deviation and autoregressive coefficient were extracted. After filtering the features, SAI and PAI-based emotional valence recognition models were constructed using support vector machines. The final emotional valence recognition accuracy was 83%. The results show that the emotional valence recognition model based on SAI and PAI can accurately identify different emotional valences in the laboratory environment. It shows that SAI and PAI have the ability to identify different valence emotions. Study 2: Mainly to verify the ability of the emotional valence recognition model constructed in Study 1 to identify different emotional valences in life scenarios. The same sensor was used to collect the ECG signals of 600 subjects in a living scene, and the collection time was one month. At the same time, subjects were asked to perform subjective emotional scores on their smartphones every day. Using the model of Study 1 to identify the emotional valence of the subjects, the accuracy rate was 56.81%. This study explored the ability of SAI and PAI to recognize emotion from the perspective of emotional valence. The results show that the emotional valence recognition model based on SAI and PAI has a good effect of recognizing emotional valence in the laboratory environment, reaching 83%; the accuracy of emotional valence recognition in life scenes is 56.81%. This study shows that the ecological validity of the emotional valence recognition model based on SAI and PAI needs to be improved. |
| 英文摘要 | 情绪识别是指根据人们的行为和反应来评估人们的情绪状态。在情绪识别领域当中，基于生理信号的情绪识别被认为更可靠、客观。其中，心率变异性（heart rate variability, HRV）是一种常用的情绪识别指标，但是其不能识别出实时的情绪，短时间内对情绪变化的敏感度不高。 新近提出的交感神经指数（sympathetic activity index, SAI）和副交感神经指数（parasympathetic activity index，PAI）能分别评估两种自主神经系统各自产生的活动影响。重要的是 SAI 和 PAI 能够实时模拟交感神经和副交感神经的波动情况，这表明 SAI 和 PAI 可用于情绪识别，实时监测情绪。然而，SAI 和 PAI 是否能够识别情绪还未得到检验，有待于探索和研究。 本研究分为两个部分，探讨基于 SAI 和 PAI 构建的情绪效价识别模型在实验室场景和生活场景中识别不同情绪效价的能力。其中研究一又分为预实验和正式实验两个部分。研究的具体内容如下： 研究一预实验：为正式实验提供情绪诱发材料。通过腾讯等网络平台，使用“搞笑”、“悲伤”等关键词搜索视频材料。视频剪辑后，邀请 4 名所内实习生依据理解度和分化度对剪辑过的视频初步筛选。然后招募 55 名被试评定视频，最终选出正性视频和负性视频各 2 个。 研究一正式实验：主要探究基于 SAI 和 PAI 的情绪效价识别模型在实验室中对正、负性情绪效价的识别。使用情绪视频诱发 95 名被试情绪并对其情绪进行主观评分。同时使用可穿戴式的腕部传感器采集心电信号。依据 SAI 和 PAI 的计算公式从心电信号中计算出 SAI 和 PAI 的值，提取标准差、自回归系数等 14个特征。筛选特征后，使用支持向量机构建基于 SAI 和 PAI 的情绪效价识别模型。最终情绪效价识别准确率为 83%。结果表明基于 SAI 和 PAI 的情绪效价识别模型在实验室环境中能够准确识别出不同的情绪效价。说明 SAI 和 PAI 具备识别不同效价情绪的能力。 研究二：主要验证研究一构建的情绪效价识别模型在生活场景中识别不同情绪效价的能力。使用同样的传感器采集 600 名被试在生活场景中的心电信号，采集时间为一个月。同时要求被试每天在智能手机上进行主观情绪评分。使用研究一的模型来识别被试的情绪效价，准确率为 56.81%。 本研究从情绪效价角度探讨 SAI 和 PAI 识别情绪的能力。结果表明基于 SAI和 PAI 的情绪效价识别模型在实验室环境下，识别情绪效价的效果较好，达到83%；生活场景中识别情绪效价的准确率为 56.81%。本研究表明基于 SAI 和 PAI的情绪效价识别模型的生态效度有待改进。 |
| 语种 | 中文 |
| 源URL | [http://ir.psych.ac.cn/handle/311026/43201]  |
| 专题 | 心理研究所_应用研究版块 |
| 推荐引用方式 GB/T 7714 | 向 鋆. 基于交感神经指数和副交感神经指数的情绪识别[D]. 中国科学院心理研究所. 中国科学院大学. 2022. |
入库方式： OAI收割
来源：心理研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。