A method of contactless assessment of cow weight using ANN

The article presents a methodology for obtaining experimental data on remote measurement of the weight of black-and-white cows on a dairy farm. A variant of the arrangement of the cameras of the optical-electronic system for obtaining the geometric dimensions of the animal and the use of artificial neural networks for segmentation of images of a pre-trained DeepLab model when highlighting the contours of the animal are provided. A sequence of actions for determining the geometric dimensions of an animal from an image using significant lines and calculating the pixel size with subsequent conversion to the SI system to determine the animal's mass using zootechnical tables, calculations, or a regression equation. According to the results of studies on individual weight determination, the absolute measurement error was up to 38 kg, and the relative error was up to 6%. For a group of animals, the absolute error was 2 kg, and the relative error was less than 0.04%.