24x faster animal weighing with a computer vision-based device
About the client
Barkom Ltd. is a Ukrainian agricultural company with over 20 years of experience in growing pigs and cattle, as well as selling self-produced food products.
The challenge
On pig farms, weighing animals is a daily, laborious task that also causes unnecessary stress for pigs. Since existing livestock weighing solutions are stationary, workers must herd the animals to the scales and keep them still. To simplify the weighing process, a new solution must address several challenges.
Delivered value
Solution
The portable computer vision-based solution we developed allows farmers to weigh animals even in motion. Due to its special camera, the device works in low-light conditions. The agricultural company can easily integrate it with the existing infrastructure as well as with third-party livestock weighing systems.
Using this device, farmers can find out a pig’s weight from above, at arm’s height. To make sure it delivers accurate results, our team tested the solution with almost 17TB of real data. The captured picture is processed on the device by a trained neural network, which evaluates the results. Due to machine learning, farmers can teach the device to weigh different breeds of pigs.
To make this device work, Lermberg Solution’s team developed an advanced weight recognition algorithm that uses neural networks with a custom image recognition algorithm. Such an approach gives 98% weighing accuracy.
Overall, the solution went through several stages of development. The early prototypes struggled with different light conditions, making weighing results inconsistent. To fix this, we upgraded the camera to a newer model that helps the neural network accurately identify a pig’s body shape and weight even in low light.
After rigorous testing on a farm, our team has completed the device — now, farmers can easily and accurately measure the weight of their animals.
