中文摘要：

随着智能手机用户数量的逐年增长,很多情景感知相关的研究也逐步开展。基于智能手机的人体行为识别已成为用户自适应感知服务中的重要研究课题。尽管有很多研究者已经尝试使用移动设备进行用户行为识别,但依旧难于从不确定的、不完整的以及不充足的移动设备传感器数据中推测出用户的行为。文中提出一种基于自动标签机制的人体行为识别模型迁移方法,利用集成学习分治思想以及深度学习网络（MLP）构建自动标签系统对新用户数据进行打标签,将打完标签的数据划归到通用模型的训练集中进行重新训练,以此完成模型迁移。实验结果表明,迁移学习后的行为识别模型能有效提高行为识别准确率。

英文摘要：

With the number of smart phone users increasing, a lot of context aware research is gradually carried out. Human behavior recognition based on smart phone has become an important research topic in user adaptive sensing service. Although there are a lot of researchers have tried to use mobile devices for user behavior recognition, But it is still difficult to recognize the user＇s behavior from the uncertainty, incomplete and inadequate sensor data of the mobile device. In this paper, a method of human behavior recognition model migration based on automatic tagging mechanism is proposed, Using the ensemble learning partition thought and deep learning network （MLP） construction of automatic labeling system on the new user data playing tag, The finished tag data transferred to the general model of training for retraining, in order to complete the model migration. The experimental results show that the model can effectively improve the accuracy of behavior recognition.