基于领域相似性迁移学习的行为识别

Human action recognition based on a transfer learning algorithm using domain similarity

提出了1种基于领域相似性的迁移学习算法,利用其他领域中的相关数据帮助完成当前领域的行为识别任务。首先通过典型相关性分析,得到领域间相似性的约束并与目标分类模型相联系,以充分利用相关领域中的有效信息;然后学习1种具有重建性、判别性、域适应性的跨域字典对,将不同领域的数据特征映射到同一空间;最后根据映射特征和分类模型对行为进行识别。利用网络中的大量图像,在UCF Sports Action数据集上的实验结果表明了算法的有效性。

Anew transfer learningalgorithmbasedondomain similarity isproposed inthispaper,whichutilizesrelevantdata from other domains to enhance the original learning system, First,by using canonical correlation analysis to add additional con-straints on the target classifiers,we can leverage the useful knowledge from the related domains. Then,we learn a reconstruc-tive, discriminative and domain-adaptive cross-domain dictionary pair to map data from different domains into a same abstractspace. Finally,the human behavior is classified according the mapping features and classification images from Web pages,and evaluate the proposed method for human a：tion recognition on the UCFchieving effective results.