文献类型：期刊文献

中文题名：基于多源域迁移学习的脑电情感识别

英文题名：EEG emotion recognition based on multi-source domain transfer learning

作者：娄晓光[1];陈兰岚[1];宋振振[1]

机构：[1]华东理工大学信息科学与工程学院化工过程先进控制和优化技术教育部重点实验室,上海200237

年份：2020

卷号：41

期号：7

起止页码：2011

中文期刊名：计算机工程与设计

外文期刊名：Computer Engineering and Design

收录：CSTPCD;;北大核心:【北大核心2017】；

基金：国家自然科学基金项目(61976091);中央高校基本科研业务费专项基金项目(222201917006)。

语种：中文

中文关键词：情感识别;迁移学习;多源域选择;迁移特征选择;集成学习

外文关键词：emotion recognition;transfer learning;multi-source domain selection;transfer feature selection;ensemble learning

摘要：针对普通机器学习算法与单源域迁移学习在应用方面的局限性,利用多源域迁移学习算法解决跨被试情感识别中正确率低的问题。为提高迁移学习的计算效率并避免负迁移现象的产生,分别从样本和特征两个方面对迁移数据进行优化。用多源域选择算法筛选出最优源域集合,用迁移特征选择算法得到最优特征集合,训练出多个迁移学习模型并对之集成。在数据集SEED上对该算法进行验证,验证结果表明,该模型相比其它情感识别模型具有更优的跨被试情感识别能力。
Considering the limited applications of the traditional machine learning algorithms and the single source domain transfer learning algorithm,multi-source domain transfer learning algorithm was adopted to improve the detection accuracy in the cross-subject emotion recognition.To improve the computational efficiency of transfer learning and to avoid the occurrence of negative transfer,the transfer data were optimized from both the sample and the feature aspects.The multi-source domain selection(MDS)algorithm was used to select the optimal source domain set.The transfer feature selection(TFS)algorithm was used to get the optimal feature subset.Multiple transfer learning models were trained and integrated.The model is validated on the SEED data set,which shows better cross-subject emotion recognition ability than other emotion recognition models.