量子粒子群和相关性分析在心电特征选择中的应用  

Application of QPSO Algorithm and Correlation Analysis in Feature Selection from ECG Signal

作　　者：曹军[1] 刘光远[1] 赖祥伟[2] 

机构地区：[1]西南大学电子信息工程学院,重庆400715 [2]西南大学计算机与信息科学学院,重庆400715

出　　处：《计算机科学》2012年第3期209-211,221,共4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(NKFS07003);中央高校基本科研业务费专项资金(XDJK2009B008)资助

摘　　要：针对心电(ECG)信号情感识别中特征选择的问题,首先运用相关性分析方法,去除原始特征集中的高相关度特征,实现原始特征集的降维;其次,为了在降维后的特征空间中进行有效的特征选择,提出了一种改进的二进制量子粒子群算法(SBQPSO)。实验结果表明,基于本算法结合Fisher分类器建立的ECG信号情感识别系统能够对高兴、惊奇、厌恶、悲伤、愤怒和恐惧6种情感达到良好的识别效果。This paper discussed the feature selection from ECG signal in affective recognition. At first, the original features with high correlation were deleted to reduce dimensionality of original feature set by correlation analysis. And then, an improved quantum-behaved particle swarm optimization with binary encoding algorithm was proposed to achieve effective feature selection in the feature space with reduced dimension. The experimental results shows that the affective recognition system based on this algorithm and fisher classifier recognize the anger, disgust, fear, grief, joy and surprise successfully.

关 键 词：特征选择 相关性分析 二进制量子粒子群算法 情感识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]