| Texto completo | |
| Autor(es): Mostrar menos - |
Curti, Paula F.
;
Souza, Andre M.
;
Kobayashi, Livia L.
;
Castanheira, Aline M.
;
Pinto, Diogenes L.
;
Ruiz, Alexandre M.
;
Andrietta, Lucas T.
;
Balieiro, Julio C. C.
;
Renno, Francisco P.
;
Sousa, Elaine P. M.
;
Ventura, Ricardo V.
Número total de Autores: 11
|
| Tipo de documento: | Artigo Científico |
| Fonte: | COMPUTERS AND ELECTRONICS IN AGRICULTURE; v. 239, p. 9-pg., 2025-12-01. |
| Resumo | |
Respiratory Rate (RR) is a key physiological indicator in dairy cattle, providing insights into herd and individual health, as well as reactions to environmental factors like heat stress, thermo-resilience, and respiratory disease. Traditionally, RR is monitored by visually counting abdominal fluctuations, an approach that becomes impractical in large herds. This challenge creates an opportunity to implement Computer Vision (CV) methods, particularly deep learning networks designed for video assessment. In our study, we captured video recordings of Holstein cows during different milking sessions in the milking parlor. A total of 688 videos of the animals' dorsal view were manually annotated to count breaths over 30 s, identify suitable frames, and define the Region of Interest (ROI). The resulting dataset supported a comparative analysis between two architectures: R2Plus1D18 and MViTv2. We tested three ROI extraction techniques: Center Crop (CC), Full Abdomen Bounding Box (FABB), and One-Sided Portion of the Abdomen Bounding Box (OSPABB). R2Plus1D18 outperformed MViTv2, achieving MAPE of 13.21 %, 14.03 %, and 12.94 %, for CC, FABB, and OSPABB, respectively, and RMSE of 5.11, 5.48, and 4.66, respectively, against MViTv2 ' s RMSE of similar to 7.5 and MAPE of similar to 23.5 % on the three scenarios. To the best of our knowledge, this is the first study in the field to concurrently evaluate three ROI approaches using two state-of-the-art architectures on the same manually annotated dataset. Additionally, we contribute by releasing our source code to facilitate reproducibility. (AU) | |
| Processo FAPESP: | 16/19514-2 - Desenvolvimento de um banco de dados genômicos para a raça Nelore e criação de ferramentas computacionais objetivando a implementação de estudos em larga escala |
| Beneficiário: | Ricardo Vieira Ventura |
| Modalidade de apoio: | Auxílio à Pesquisa - Jovens Pesquisadores |
| Processo FAPESP: | 23/02851-0 - Uso de Visão Computacional Aplicado a Problemas Locomotores em Gado de Leite |
| Beneficiário: | Paula de Freitas Curti |
| Modalidade de apoio: | Bolsas no Brasil - Mestrado |
| Processo FAPESP: | 16/17078-0 - Mineração, indexação e visualização de Big Data no contexto de sistemas de apoio à decisão clínica (MIVisBD) |
| Beneficiário: | Agma Juci Machado Traina |
| Modalidade de apoio: | Auxílio à Pesquisa - Temático |
| Processo FAPESP: | 21/03101-9 - COWADAPT: detecção de variantes genéticas associadas com a adaptação de bovinos a ambientes desafiadores utilizando um multi-assembly graph como referência |
| Beneficiário: | Ricardo Vieira Ventura |
| Modalidade de apoio: | Auxílio à Pesquisa - Temático |