Center-aligned domain adaptation network for image classification

作者：

Highlights：

• The common class center on the source and target domains achieve semantic alignment.

• The compactness and discrimination of domain data are beneficial for semantic alignment.

• Update the common class center with source data and pseudo-labeled target data.

• Paired source and target domain data feed into network can avoid model perturbations.

摘要

•The common class center on the source and target domains achieve semantic alignment.•The compactness and discrimination of domain data are beneficial for semantic alignment.•Update the common class center with source data and pseudo-labeled target data.•Paired source and target domain data feed into network can avoid model perturbations.

论文关键词：Transfer learning,Domain adaptation,Center-aligned,Semantic alignment,Image classification

论文评审过程：Received 8 August 2020, Revised 1 November 2020, Accepted 24 November 2020, Available online 27 November 2020, Version of Record 5 December 2020.

论文官网地址：https://doi.org/10.1016/j.eswa.2020.114381