Weighing the flock without touching a bird

Tyson Foods wanted to feed its flocks precisely, but there was no practical way to know what the birds actually weighed without handling them one by one. Feed was set without that number, so grain was over-supplied and cost and waste climbed with it.

Any solution had to work at the scale and pace of a working hen house, not a lab, and without disturbing the birds.

We built a computer vision system that estimates the weight of each chicken from images alone. A camera captures the flock twice a day, and models trained on a large dataset of chicken images and known weights turn those pictures into a per-bird weight estimate.

Those estimates chart each flock's growth curve across the 56-day cycle, and deviations from the expected curve drive feed adjustments, so grain distribution matches the birds' real nutritional need. The estimation and feed strategy sharpen cycle on cycle as more data comes in.

Feed could finally be matched to actual weight rather than a guess, cutting grain waste and the cost that comes with it while keeping growth on track. An initial trial in a 20,000-bird facility gave Tyson a blueprint for rolling the approach out across its operations nationwide.