Image Processing for Analyzing Broiler Feeding and DrinkingBehaviors

摘要

Understanding broiler feeding and drinking behaviors can provide insights to farm managements and system designs. Currently, there is no automatic system for continuous behavioral monitoring of group-reared broilers. The objectives of this study wereto (I) develop algorithms to automatically detect bird number at feeder (BNF) and at drinkers (BND) for group-reared broilers, and (2) analyze these behaviors with the algorithms. Sixty Ross *Ross 708 broilers at 26-28 days of age were kept in a pen (2.91.40.7 m, LWH) with a tube feeder and five nipple drinkers. A camera was installed above the pen to record broiler behaviors in video files, which were converted to images. The images were firstly processed to extract broiler-representing pixels of concerned areas (i.e. feeder and drinker). The pixels around the feeder were used to develop a linear regression model for estimating BNF, and the pixels on the perimeters of the segmented drinkers were used to determine BND. The algorithms were trained andtested with 19,200 images. Broiler feeding and drinking behaviors (e.g. spatial and temporal preferences) were analyzed for three consecutive days with the algorithms. The results show that the accuracy for estimating BNF was 89%, and the mean square error was 0.4 bird, indicating small detection errors. The sensitivity, specificity, and accuracy for estimating BND were 87%, 97%, and 93%, respectively. Broilers showed spatial and temporal preferences for feeding and drinking. It is concluded that our algorithms had acceptable accuracies in determining bird number at feeder and drinker and thus are useful components for image-based automatic behavioral monitoring systems.