Researchers from the Chelyabinsk State University in Russia have designed a neural network to analyse video surveillance for the non-contact weighing of cattle.
The scientists explained that weighing cows is necessary to assess their production performance and timely identify health problems. On the other hand, traditional livestock weighing is stressful for animals. Today, the weight of most cows in Russia is approximately estimated by farm workers, but the assessment ‘by eye’ is extremely inaccurate.
The neural network uses computer vision to not only calculate the current weight of animals but also to evaluable their morphological characteristics and even to predict their future production performance, the scientists claimed.
The new system could help farmers adjust operational costs, said Aleksey Ruchai, head of the Department of Computer Security at the Chelyabinsk State University and one of the authors of the study. Among other things, it could flag the right time to slaughter a cow.
Currently, scientists are working on establishing a comprehensive database to teach algorithms to better fulfil their duties. The size of the database is crucial for machine learning.
In its current version, the system has a margin of error ranging between 5% and 10%, which is comparable with conventional weighing. As the size of the database grows, the system is expected to improve accuracy.
Currently, scientists are working on teaching the neural network to rely on footage from a single camera and calculate the weight of a moving animal in a matter of just a few seconds.
The new development is not the first attempt in Russia to use machine vision to improve operations in the dairy industry. In 2021, another group of scientists from the Novosibirsk State University rolled out the VectorMoo system designed to analyse footage from video surveillance to timely identify animal diseases and formulate recommendations helping farmers to shape up operations.
The scientists said that their system could be integrated into a modern farm within 12 hours. The investment was expected to pay off in a year. However, no further information on this system is available.