基于低秩交叉重构的领域自适应算法

为了解决现有领域在自适应方法中忽略了整个数据域内部结构的信息和源域与目标域之间的差异问题,提出了一种新的基于低秩交叉重构的领域自适应方法.通过对源域和目标域的交叉重构来构造新的源域与目标域,使得同类数据相互交织,缩短了同类数据之间的距离; 通过对重构矩阵施加低秩约束,将2个域的同类数据对齐,以此来充分挖掘源域和目标域同类数据之间的内在结构信息,并利用该结构信息来学习分类器,从而取得更好的跨域识别效果.在5个公开数据集上的实验结果表明:该方法具有较高的跨域识别准确率.

The Low-Rank Constraint-Based Cross Reconstruction for Domain Adaptation

In order to solve these problems,a new low-rank constraint-based cross reconstruction(LRCR)for domain adaptation is proposed.Through the cross reconstruction of the source domain and the target domain,the new source domain and the target domain are constructed so that the same label data are interweaved with each other.By applying the low-rank constraint,the same label data in the two domains are aligned to fully mine the internal structure information between the same label data in the source domain and the target domain.The structure information is used to learn the classifier to achieve better cross-domain recognition accuracy.The experimental results on five public datasets show that LRCR has high cross-domain recognition accuracy.