7 - Characterization of Feeder Cattle Behavior Using an Integrated Machine Vision Learning System

Abstract: Animal behavior can be a valuable indicator of the health and welfare status of an animal. Current assessments of cattle behavior in commercial settings rely upon human observers who are only capable of observing a relatively small proportion of animals for a small proportion of time. Moreover, the mere presence of human observers may alter normal behavioral patterns, making data difficult to interpret. To overcome these challenges, we developed a machine vision learning system to monitor cattle behavior on a commercial feedlot. Using continuous data collected from solar powered cameras, we set out to characterize the standing and lying patterns of feedlot cattle. A series of solar powered cameras were installed on a single pen of calves at a commercial feedlot (n=280; mean weight = 632lbs +/- 45). To characterize standing and lying behavior, a neural network model (YOLOv5) was trained and applied to pen images captured every 5 minutes during the day for weeks 2-20 of the feeding period (n=19,152). Algorithm precision and recall were 96% and 92%, respectively. Standing and lying behaviors showed a pattern of temporal cyclicity with the greatest proportion of cows standing at 0800 (92%) and 1700 (96%). The mean proportion of animals observed lying down increased during the feeding period (wk 2-8=36%; wk 9-14=39% and wk 15-20=45%). Machine vision learning systems can be an accurate and efficient means of quantifying behavioral patterns in commercial environments. Future work will examine how environmental and management factors (e.g. weather, pen moves) and morbidity alter behavioral patterns throughout the feeding period.