基于改进迁移学习的运动想象分类识别算法  

Classification and Recognition Algorithm of Motion Imagination Based on Improved Transfer Learning

作　　者：杜义浩[1] 常超群 杜正 张延夫 曹添福 范强 谢平[1] DU Yi-hao;CHANG Chao-qun;DU Zheng;ZHANG Yan-fu;CAO Tian-fu;FAN Qiang;XIE Ping(Institute of Electric Engineering,Yanshan University,Qinhuangdao,Hebei 066004,China)

机构地区：[1]燕山大学电气工程学院,河北秦皇岛066004

出　　处：《计量学报》2023年第11期1740-1748,共9页Acta Metrologica Sinica

基　　金：国家自然科学基金联合资助基金(U20A20192);国家自然科学基金(62076216);河北省自然科学基金(C2020203012)。

摘　　要：利用迁移学习算法提高分类识别的准确率是运动想象脑机接口应用的热点研究问题,其中样本迁移和特征迁移的传统模型算法在样本量较少或源域数据和目标域数据差异较大情况时,各自的迁移效果并不理想。基于欧式对齐(EA)和改进联合类质心匹配和局部流形自学习(CMMS)迁移学习的运动想象分类算法,将样本迁移和特征迁移的优势有机结合,在考虑样本本身的同时,进一步提高了分类准确率。首先,对样本进行源域和目标域的EA,减少源域和目标域的数据分布差异;其次,基于最小化最大均值差异(MMD)改进CMMS方法,筛选源域数据,再次减小源域样本与目标域的分布差异;最后,将该方法应用于BCI竞赛数据集进行离线测试和在线实验。实验结果表明:所研究的方法与SVM、JDA、BDA、EasyTL、GFK、CMMS相比较,迁移学习模型的识别准确率分别提高了14.38%,8.5%,5.8%,10.4%,11.8%,5.7%。The use of transfer learning algorithms to improve the accuracy of classification recognition is a hot research problem in the application of motion imagination brain-computer interface.Traditional model algorithms for sample transfer and feature transfer do not achieve ideal transfer results when the sample size is small or there is a significant difference between the source domain data and the target domain data.The motion imagery classification algorithm based on Euclidean alignment(EA)and improved jointly class centroid matching and local manifold self-learning(CMMS)migration learning organically combines the advantages of sample migration and feature migration to further improve the classification accuracy while considering the sample itself.Firstly,the samples are subjected to EA in the source and target domains to reduce the differences in data distribution between the source and target domains.Secondly,the CMMS method is improved based on minimizing the maximum mean difference(MMD)to filter the data in the source domain and again reduce the differences in distribution between the samples in the source and target domains.Finally,the method is applied to the BCI competition dataset for offline testing and online experiments.The experimental results show that compared with SVM,JDA,BDA,EasyTL,GFK,and CMMS,the recognition accuracy of the migration learning model is improved by 14.38%,8.5%,5.8%,10.4%,11.8%,and 5.7%,respectively.

关 键 词：计量学 脑机接口 运动想象 迁移学习 EA 改进CMMS 

分 类 号：TB973[一般工业技术—计量学]