Abstract: In order to identify and count the daily behavior of cattle, accurately judge the living habits and health status of cattle in the farm, and provide decision-making basis for farm management, the daily behavior video of cattle in the farm were collected, and labelimage was used tolabel four daily behaviors of cattle, such as standing, lying, feeding and drinking, to construct the daily behavior data set of cattle. The YOLOv5 s network model was used to train the daily behavior data set of cattle, and four daily behaviors of cattle including standing, lying, feeding and drinking were identified. The accuracy of YOLOv5 s network model was compared with that of characteristic part relationship method in cattle daily behavior recognition. The video frames of the above four kinds of cattle daily behavior in the farm were counted, and the time statistical algorithm of the daily behavior of cattle was established. The occurrence time of the daily behavior of the above four kinds of cattle was counted through the time statistical algorithm. The results showed that the YOLOv5 s network model had a high recognition accuracy for the four daily behaviors of cattle（standing, lying down, feeding and drinking） in the farm, and the statistical error of the four daily behavior time of the four cattle was low. The results indicated that the identification and time statistics of daily behaviors of cattle by computer vision could basically meet the needs of farms and provide technical services for precision breeding industry.