Non-contact method for cattle weight estimation based on deep-learned convolutional neural network

Autores

  • Oleh Iliunin Kharkiv National University of Radio Electronics
  • Oleg Rudenko Kharkiv National University of Radio Electronics
  • Oleg Kobylin Kharkiv National University of Radio Electronics
  • Nataliia Serdiuk Kharkiv National University of Radio Electronics
  • Andrii Cherkasov Kharkiv National University of Radio Electronics

Palavras-chave:

augmentation, data set, deep learned convolutional neural network

Resumo

Improving the efficiency of meat and dairy farming and introducing innova-tive digital technologies into the industry are urgent tasks today. Rapid assessment of cattle weight is the main indicator for adjusting the diet and monitoring animal health. The proposed method uses the geometric dimen-sions of cows from camera images processed by computer technologies and is based on a deep learning convolutional neural network based on the Effi-cientNetB3 and Resnet18 models for segmentation and formation of key points on animal images for further identification of their characteristics. The developed system allows automating the process of measuring cattle weight, which significantly reduces labor costs and increases the economic efficiency of farms. The use of an intelligent information system (IIS) allows obtain-ing more accurate estimates of animal weight compared to traditional meth-ods, which showed an adequacy of RMSE = 0.93% (root mean squared er-ror) on an extended data set. After adaptation, the application can be used to estimate the weight of other farm animals.

Downloads

Não há dados estatísticos.

Publicado

2025-12-05

Como Citar

Iliunin, O. ., Rudenko, O. ., Kobylin, O. ., Serdiuk , N. ., & Cherkasov, A. . (2025). Non-contact method for cattle weight estimation based on deep-learned convolutional neural network. Anais Do Congresso Internacional De Conhecimento E Inovação – Ciki, 1(1). Recuperado de https://proceeding.ciki.ufsc.br/index.php/ciki/article/view/1742