基于迁移子空间的半监督领域自适应方法

Semi-supervised domain adaptation via transfer subspace

传统子空间学习方法在对齐领域总体分布时往往忽略样本类别信息,若原始样本判别力不足,将难以保证投影后子空间中样本的判别性。针对该问题,提出迁移子空间的半监督领域自适应方法。通过充分利用样本类别标签先验信息,在得到具有判别性子空间的同时充分挖掘重构矩阵中蕴含的鉴别信息,增强子空间跨领域特征表达的鉴别力和鲁棒性,提高模型的分类性能。在领域自适应问题常用的基准图像数据集上进行实验,其结果表明,该算法有较好的分类效果。

Traditional subspace learning methods tend to ignore category information when aligning the overall distribution of domains.If discrimination ability of original data is insufficient,it will be difficult to ensure the discrimination of data in subspace after projection.To alleviate this problem,a semi-supervised domain adaptation via transfer subspace(SSDTS)was proposed.The class label information of sample was used to collect the discrimination information contained in the reconstruction matrix while obtaining the discriminative subspace.Results of experiments on benchmark image datasets which were commonly used in domain adaptive problems show that SSDTS has good classification accuracy.