A camera above the flock paired with artificial intelligence can track individual turkeys and predict their body weight up to three weeks ahead with about 93% accuracy, according to researchers at Penn State. Body weight is one of the most important measurements in poultry production — it drives growth monitoring, flock uniformity, feed conversion and disease detection and sets the equipment at the processing plant — but frequent, accurate weighing of individual birds has remained a labour-intensive job that also stresses the animals.

The study at the Penn State Poultry Education and Research Center followed 30 male turkeys housed together from day 37 to day 133 of age, nearly 14 weeks. A camera positioned above the birds captured both standard colour images and depth images giving each bird's three-dimensional shape and size. Because several birds appear in each frame, the system had to learn which pixels belonged to which turkey — instance segmentation — and the researchers weighed the birds by hand five times a week to give the model its reference points.

A ResNet deep-learning network, widely used for image analysis, learned the relationship between a bird's appearance, size and shape and its actual weight. Its forecasts of future weight were nearly as accurate as its estimates of current weight — notable, the authors say, because earlier computer-vision systems could not forecast over time.

On a commercial farm with thousands of birds, cameras that estimate weight reliably would spare farmers from catching and weighing large numbers of turkeys, said senior author Enrico Casella, assistant professor of data science for animal systems, with uses in growth monitoring, spotting birds that are not growing and planning processing.

Casella cautioned that a study of 30 turkeys under controlled conditions does not mean the technology is ready for commercial use; larger trials are needed across farms, breeds, lighting and environments. First author Mireia Molins, who completed her master's degree this year, now works as a field engineer with NovaTech Engineering in Minnesota; the work was funded by Penn State's Institute for Computational and Data Sciences and USDA's National Institute of Food and Agriculture.

Source: The Poultry Site