摘要
本文针对源域样本与目标域样本分布不同的特性而导致分类器失效的问题,提出了联合子空间对齐与极限学习机无监督领域自适应方法。首先通过利用极限学习机自编码器ELM-AE获取的域不变性特征权重替代ELM初始化随机权重;然后拆分ELM输出层权重使其更加的灵活;最后联合ELM拆分的输出层权重与低秩约束的子空间对齐方法,得到具有迁移能力的ELM输出权重,通过域不变性特征权重和迁移能力的输出权重构建的分类器可以很好的适应跨域分类任务。在基准数据集上进行的实验证明,本文的方法在跨域视觉识别方面优于其他最新方法。
Aiming at the problem that the distribution of the samples in the source domain and the target do-main causes the classifier to fail,a Joint Subspace Alignment and Extreme Learning Machine of unsupervised domain adaptation is proposed.First,replace the ELM initialization random weights by using the domain in-variant feature weights which obtained from the ELM-AE of the extreme learning machine;then split the ELM output layer weights to make it more flexible;and finally combine the ELM split output layer weights and low rank The constrained subspace alignment method obtains the ELM output weights with transfer ability.The classifier constructed by the domain invariant feature weights and the transfer weight output weights can be well adapted to cross-domain classification tasks.Experiments conducted on the benchmark data set prove that the method in this paper is superior to other latest methods in cross-domain visual recognition.
作者
陶洋
胡昊
鲍灵浪
TAO Yang;HU Hao;BAO Ling-lang(School of Communication and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)
出处
《新一代信息技术》
2021年第1期14-23,共10页
New Generation of Information Technology
基金
国家自然科学基金项目(项目编号:61801072)。
关键词
领域自适应
极限学习机
子空间对齐
图像分类
Domain adaptation
Extreme learning machine
Subspace learning
Image classification
