摘要
为准确有效地对情绪状态进行识别,对4种情绪状态(Joy、Anger、Sadness、Pleasure)下的多生理信号(心电、肌电、呼吸、皮电)进行预处理和特征提取,利用ReliefF算法进行特征选择,利用J48决策树分类器最终实现对4种情绪状态的识别。J48决策树分类器对4种情绪状态的平均识别率为96.74%,对结果和数据进行分析发现,RSP信号对情绪状态识别十分重要;不同生理信号组合对情绪状态的识别效果不同;Sadness和Pleasure的相互误识率相对较高;使用J48决策树进行分类时采用的特征数量与样本数量正相关。
To recognize emotional states accurately and effectively,preprocessing and feature extraction for multi-physiological signals(ECG,EMG,RSP,SC)in terms of four emotions(Joy,Anger,Sadness,Pleasure)were undertaken,ReliefF algorithm was used for feature selection,and J48 decision tree classifier was used to achieve recognition of the four emotional states.Average recognition rate based on J48 decision tree classifier of the four emotional states is 96.74%,and analyzed results show that RSP signal is very important for emotional states recognition.Combinations with different physiological signals for emotional states recognition have different effects.The error recognition rate between Sadness and Pleasure is relatively high.The number of features J48 decision tree used for classification is positively correlated with the number of samples.
出处
《计算机工程与设计》
北大核心
2017年第3期761-767,共7页
Computer Engineering and Design
基金
广西高校重点实验室科学基金项目(GXSCIIP201411)
广西自然科学基金重点项目(2014GXNSFDA118037)
四川省科技计划基金项目(2015HH0036)