多频段时域分解的行为识别特征优选方法.pdfVIP

优先出版 计 算 机 应 用 研 究 第32 卷 多频段时域分解的行为识别特征优选方法* 王忠民， 王 斌 (西安邮电大学 计算机学院，西安 710121 ) 摘 要：为了降低外界环境对移动用户行为识别的影响，保留行为敏感特征，提高行为识别的准确率，提出了一种多频 段时域分解的人体行为识别特征优选方法，该方法首先对行为样本数据进行多频段分解，计算样本数据在不同频段信号 的特征，利用遗传算法以决策树作为分类器进行特征优选，在多组特征中搜索出近似最优的特征组合。实验结果表明， 该方法优选出的特征组合能有效提高行为识别的准确率。 关键词：多频段时域分解；行为识别；特征优选；遗传算法；决策树 中图法分类号 TP391 Feature selection method for mobile user behavior recognition based on multiband time domain decomposition Wang Zhong-min, Wang Bin (School of Computer Science Technology, Xi’an University of Posts Telecommunications, Xi’an 710121, China) Abstract: In order to reduce the influence of the external environment to mobile user behavior recognition, select the sensitive features, and improve the accuracy of behavior recognition, this paper proposed an acceleration signal feature selection method to mobile user behavior recognition based on multiband time domain decomposition. In the method, it firstly decomposed the sample data to different time domain components. Then it computed the features of the different time domain components. Fi- nally, it used the genetic algorithm and the decision tree as a classifier to choose the feature set which is sensitive to the human ’s behaviors . Experiment results show that the multiband time domain decomposition feature optimization method can improve the accuracy of human behavior recognition effectively. Key Words: multiband time domain decomposition; behavior recognition; feature optimization; genetic algorithm; decision tree 设备三轴加速度数据四分位差和小波能量,使用 kernel discri- 0 引