摘要： 为使个人运动管理系统更好地感知用户行为，给予科学的运动指导，提出一种基于通用模型迁移的自适应行为识别方法。该方法无需对个体数据进行标定，通过将群体行为的共性知识迁移到个体行为，使通用识别模型可以随着个体行为样本的增多，自适应地调整共性知识，从而形成针对特定个体的个性化行为识别模型。实验结果表明，个性化模型的平均识别精度可以从67.31%提高到83.54%。

关键词: 运动管理, 普适计算, 行为识别, 迁移学习, 机器学习, 决策树

Abstract: To help personal sports management system accurately detect users’ activity and give them scientific sports instruction, a common model transferred adaptive activity recognition method is presented. With this method, the common knowledge of many persons can be transferred to a new one and with the new one’s data increasing, the model can be automatically adapted to fit the new user. The method proposed is evaluated on real-world dataset to demonstrate its performance.

Key words: exercise management, pervasive computing, activity recognition, transfer learning, machine learning, Decision Tree(DT)

中图分类号: 

• TP393