CAJ | 학술논문

行为识别侧重于通过使用传感器数据来推断当前用户的活动，典型的行为识别技术通常基于点对点的方法来处理传感器感知到的数据，其中，监督学习算法在该领域有广泛的应用。本文提出了一种基于聚类的分类算法来进行行为识别，该算法采用增量式学习来挖掘数据流中的用户行为，通过将不同的活动行为赋予不同的类，融合监督、无监督和主动学习算法，并结合混合相似性度量方法建立一个鲁棒的识别系统。
행위식별측중우통과사용전감기수거래추단당전용호적활동，전형적행위식별기술통상기우점대점적방법래처리전감기감지도적수거，기중，감독학습산법재해영역유엄범적응용。본문제출료일충기우취류적분류산법래진행행위식별，해산법채용증량식학습래알굴수거류중적용호행위，통과장불동적활동행위부여불동적류，융합감독、무감독화주동학습산법，병결합혼합상사성도량방법건립일개로봉적식별계통。
Behavior recognition focuses on using sensor data to infer the user's current activity, typical behavior recognition technology utilizepeer-based approach to deal with perceived sensor data, in which supervised learning algorithms have a wide range of applications. This paper presents a classification algorithm based on clustering to recognize behavior, which applies an incremental learning to mine data stream of user behavior, by assigning different activities to different classes, integrating supervision, unsupervised and active learning algorithms, and combining hybrid similarity measurement, a robust recognition system is build.