Abstract: Domain adaptation, as a critical technique in transfer learning, effectively addresses the issue of differing distributions between training and testing datasets. However, traditional domain adaptation methods were typically limited to scenarios where the number and types of categories in the target and source domain datasets were identical. In practical applications, this condition was often difficult to meet. Open - set Domain Adaptation (OSDA) emerged to address this challenge. In order to fill the gap in this field and provide a reference for related research, a summary and analysis of the open - set domain adaptation methods that had emerged in recent years were conducted. The related concepts and basic structures were introduced. The typical methods were combed and analyzed from three stages: data augmentation, feature extraction, and classifier. Future development directions were also looked forward to.

Key words: transfer learning, open set domain adaptation, data augmentation, feature extraction, classifier

摘要： 域适应作为迁移学习的关键技术，能很好地解决训练和测试数据集不同分布的问题。然而，域适应方法通常只适用于目标域和源域数据集所含类别数量和种类相同的情况，但在实际场景中该条件往往很难满足，开集域适应正是为了解决此问题应运而生。为了填补该领域的空白，为相关研究提供借鉴参考，对近年来出现的开集域适应方法进行归纳分析，介绍了相关概念与基本结构，分别从针对数据增强、针对特征提取以及针对分类器3个阶段梳理分析了相关典型方法，并对未来发展方向进行展望。

关键词: 迁移学习, 开集域适应, 数据增强, 特征提取, 分类器