Camera AI Predicts Turkey Weight Three Weeks Ahead

Poultry farmers face a daily challenge that is both labor-intensive and stressful for the animals: weighing birds. Traditionally, this requires catching individual animals, placing them on scales, and recording data, a process that disrupts the flock and consumes significant human resources. Researchers at Penn State University have introduced a camera-based alternative that claims to solve this by using artificial intelligence to track growth without physical contact.

The system, detailed in a study published in Frontiers of Animal Science, does more than just weigh a bird in the moment. It forecasts body weight up to three weeks into the future with approximately 93% accuracy. For commercial operations, this predictive capability offers a way to plan processing schedules and feed strategies based on projected growth rather than reactive measurements, potentially reducing the need to handle large numbers of birds.

Visual analysis replaces manual weighing

The technology relies on a camera positioned above the poultry housing that captures both standard color images and depth data. This depth information helps the software understand the three-dimensional shape and size of each turkey, which correlates with its mass. Because multiple birds are often in the same frame, the AI must distinguish individual animals from the background and from one another, a process known as instance segmentation.

To train the system, researchers used a deep learning model called ResNet. The AI learned to associate specific visual features and spatial dimensions with actual body weight. During the testing phase, which spanned nearly 14 weeks, the team manually weighed the turkeys five times a week to provide the reference data needed to calibrate the camera's predictions.

Predicting future growth patterns

The most significant finding is the system's ability to forecast weight weeks in advance. Previous computer vision tools in agriculture could estimate current weight but lacked the capacity to project future growth trajectories. This temporal forecasting allows farmers to identify birds that are not growing at the expected rate, enabling early intervention for health or feed issues before they become costly problems.

Trade-offs in automated monitoring

While the 93% accuracy rate is promising, it is not perfect. In a commercial setting with thousands of birds, even a small margin of error can complicate equipment settings for processing lines, which rely on average flock weight. Furthermore, the system requires consistent camera placement and quality image data to function correctly, which may present technical challenges in varied farm environments.

As reported by GN technics/ai (en-US), this development marks a step toward reducing animal stress and labor costs. However, the transition from research center to widespread commercial use will depend on whether the technology can maintain its accuracy across diverse flock densities and lighting conditions without constant human oversight.