Modern pig ranches are gargantuan complexes devised for a comprehensive technological cycle — from reproduction to slaughter and processing of pig meat. Up to numerous thousand creatures can be situated at such enterprises concurrently. One of the most imperative errands for pig farm staffers is to regularly observe the weight of each individual in accordance with benchmarks on conventional scales. Moreover, for precise measurement, the personnel of the complex necessitates placing and fastening each animal individually on them. However, this procedure occupies ample time and labor, and can also generate stress in pets and negatively impact their physiological parameters. Consequently, evolutions that augment the rapidity and accuracy of defining the weight of pigs are relevant today.

"We have created a neural network archetype that automatically gauges the weight of pigs according to geometric dimensions determined grounded on recordings from a conventional video camera. Since such a framework necessitates only a few seconds of recordings for each creature, this will significantly accelerate and facilitate the weighing procedure for both the personnel of the complex and the pigs themselves," denotes Daria Andreevna Valenkova, an engineer at the Department of Automation and Control Processes of St. Petersburg State Technical University.

Defining weight by investigating geometric dimensions is feasible because there is a relationship between the weight of a pig and the dimensions of its body, for example, such as chest and abdomen girth, trunk height, body length and width. Apprehending these patterns, scientists have determined how they interrelate. In the most straightforward case, we are referring to the length and width of the body. As data for training and examinations, video recordings with several dozen creatures were utilized, which were furnished by two immense pig ranches from the village of Sarai, Ryazan region and the village of Zborov, Gomel region. The error in defining the weight was only about 5 kg.

"In the future, our framework can be employed in any pig ranch, irrespective of its magnitude and number of livestock. Staffers will only necessitate installing a camera under the ceiling, for instance, at the pig drive location. All animals will be documented on video, according to which, applying our neural network archetype, weight information will be defined automatically and conveyed to the computer of the system operator. In addition, we are currently laboring on engendering an even more expedient mobile application to execute weighing directly from a smartphone," notes Daria Andreevna Valenkova.

LETI scientists persist to toil on refining the complex for defining the weight of domestic pigs applying computer vision. In the imminent future, together with the industrial partner of the project, the database of animal videos will be expanded. In addition, it is planned to polish the parameters of the mathematical archetype for gauging geometric dimensions so that it employs not only the length and width of the pig's body, but also several other paramount parameters. All this will significantly refine the accuracy of the framework.

The maturation was backed by a grant from the II experimental Blue Sky Research tournament, which is aimed at sustaining interdisciplinary scientific projects of young scholars at the junction of artificial intelligence and the agro-industrial complex, as well as the food commerce. This exertion is one of the projects of the Department in the field of artificial intelligence application in the real sector of the economy, in particular, in agriculture. Previously, scholars created a framework for observing the condition of cows applying computer vision.