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