Abstract: Aiming at the problems of complex feature extraction and low classification accuracy of traditional

methods in human behavior recognition（HAR），a behavior recognition model based on deep belief network-support vector machine（DBN-SVM）was proposed. Firstly，in order to better capture the periodic changes of user

behavior，a filter was introduced to denoise，filter and window segment processing the behavior signal. Secondly，a DBN-SVM model was proposed，in which the behavior time-frequency signals preprocessed were taken as the

visual input of the model. The characteristics of the behavior data were extracted automatically by DBN，and the

behavior classification and recognition were realized by combining with SVM. Finally，the model was applied to

several behavioral data sets，and compared with the traditional machine learning method，the results show that the

efficiency of the proposed method is improved by 4%～15% compared with the traditional machine learning

method，which can achieve more accurate activity classification and improve the performance of behavior recognition.

      Key words: behavior recognition；feature extraction；machine learning；deep belief network-support vector

machine（DBN-SVM）