基于生理信号的压力情感数据库的建立及分析

Research on the Performance Comparing and Building of Affective Computing Database Based on Physiological Parameters

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,其结果直接影响后续的情感识别及分析。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。本文针对这一问题,分析与比较了国际上两个公开、经典的认知情感计算数据库的性能,即美国麻省理工学院(MIT)的认知情感计算数据库与德国Augsburg大学情感识别数据库,分别就数据库中数据的结构与数据类型进行了比较,并对基于该数据实现情感识别的效果进行了分析研究,结果表明,基于生理参数的分析,能有效地进行情感识别,是一种实现情感评估的可行方法。针对国内基于生理参数压力情感评估的数据缺乏这一问题,构建了一个面向高校中高压力人群的压力情感评估数据库。该数据库以应届硕士研究生作为受试,通过一定的认知任务刺激,采集其生理参数,并基于此数据库进行了压力分析,结果表明,该数据库的建立对于压力评估具有一定参考价值,希望通过此研究,为压力情感评估和分析提供一个参考与支持。

The validity and reasonableness of emotional data are the key issues in the cognitive affective computing research. Effects of the emotion recognition are decided by the quality of selected data directly. Therefore, it is an important part of affective computing research to build affective computing database with good performance, so that it is the hot spot of research in this field. In this paper, the performance of two classical cognitive affective computing databases, the Massachusetts Institute of Technology （MIT） cognitive affective computing database and Germany Augsburg University emotion recognition database were compared, their data structure and data types were compared respectively, and emotional recognition effect based on the data were studied comparatively. The results indicated that the analysis based on the physical parameters could get the effective emotional recognition, and would be a feasible method of pressure emotional evaluation. Because of the lack of stress emotional evaluation data based on the physiological parameters domestically, there is not a public stress emotional database. We hereby built a dataset for the stress evaluation towards the high stress group in colleges, candidates of postgraduates of Ph. D and master as the subjects. We then acquired their physiological parameters, and performed the pressure analysis based on this database. The results indicated that this dataset had a certain reference value for the stress evaluation, and we hope this research can provide a reference and support for emotion evaluation and analysis.