摘要
针对脉搏信号的情感识别问题,提出了一种相关性分析和最大最小蚁群算法相结合的方法,找出了对情感识别模型构建具有较好性能的稳定特征子集。首先将原始特征用序列后向选择(SBS)方法排序,然后利用线性相关系数分析法计算特征间的相关度,并根据排序结果去除部分相关度较大的特征,最后针对筛选后的特征子集用最大最小蚁群算法进行特征选择,并结合Fisher分类器对高兴、惊奇、厌恶、悲伤、愤怒和恐惧6种情感进行分类。实验结果表明,该方法能在原始特征集合中找出更稳定有效的特征子集,从而建立起有效的情感识别模型。
For the affective recognition from pulse signal,a new approach was presented,which combined correlation analysis and max-min ant colony algorithm.The effective feature subset which can identify the affective recognition modelwith better performance was found.Firstly,sequential backward selection(SBS) was used for sorting the original features.Secondly,the linear correlation coefficient was presented for calculating the correlation between features,and some features were removed which had greater correlation according to the result of sorting.Finally,max-min ant colony algorithm realized feature selection which searched for an optimal subset based on the compact feature subset,and combined with Fisher classifier to finish classification of six emotions which include happiness,surprise,disgust,grief,anger and fear.The experiments show that the proposed approach can find the more stable and effective feature subset from the original feature set,and establish effective affective recognition model.
出处
《计算机科学》
CSCD
北大核心
2012年第4期250-253,274,共5页
Computer Science
基金
国家自然科学基金(60873143)
国家重点学科基础心理学科研基金(NKSF07003)
中央高校基本科研业务费专项资金(XDJK2009B008)资助
关键词
情感识别
脉搏信号
相关性分析
最大最小蚁群算法
Affective recognition
Pulse signal
Correlation analysis
Max-min ant colony algorithm