改进的支持向量机在情感识别中的应用

Application of improved support vector machine in emotion recognition

针对传统支持向量机核函数参数σ、惩罚系数γ以及不敏感损失常数ε需要优化的问题,提出模拟退火免疫粒子群算法(SA-IPSO)优化支持向量机(SVM)关键参数σ、γ、ε的方法。并使用BIOPAC MP150对于630名被试者进行了情感激发状态下的心电生理信号采集,构建可靠的情感生理信号数据库,用该算法对其分类,与模拟退火支持向量机(SA-SVM)以及默认参数支持向量机相比,识别率更高,误报率更低,说明该算法在情感识别领域识别效果优于传统支持向量机。

This paper discussed the optimization method about parameters σ, ε, and γ for Support Vector Machine （SVM） and raised a new SVM algorithm which was improved by the Simulated Annealing-Immune Particle Swarm Optimization （SA- IPSO）. BIOPAC MP150 was used to test 630 subjects and get their biological signals to build a reliable database. The database was classified by the new algorithm and comparisons were made with SVM and Simulated Annealing-SVM. Results indicate that using this new algorithm, the True Positive Rate （TPR） is higher and the False Positive Rate （FPR） is lower, which means that it has better identifying effect in emotion recogniton.