Avaliação da conformação de cavalos da raça Mangalarga Marchador por fotogrametria

• Ano de defesa: 2025
• Autor(a) principal: Andrade, Millena Oliveira
• Orientador(a): Barcelos, Kate Moura da Costa
• Banca de defesa: Barcelos, Kate Moura da Costa, Procópio, Alessandro Moreira, Silva, Sergio Francisco da
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso embargado
• Idioma: por
• Instituição de defesa: Universidade Federal de Goiás
• Programa de Pós-Graduação: Programa de Pós-graduação em Zootecnia (EVZ)
• Departamento: Escola de Veterinária e Zootecnia - EVZ (RMG)
• País: Brasil
• Palavras-chave em Português: Equinos; Inteligência artificial; Redes neurais
• Palavras-chave em Inglês: Equines; Artificial intelligence; Neural networks
• Área do conhecimento CNPq: CIENCIAS AGRARIAS::ZOOTECNIA
• Link de acesso: http://repositorio.bc.ufg.br/tede/handle/tede/14279

The Brazilian Mangalarga Marchador breed stands out on the national scene for being the largest in number of specimens, approximately 747 thousand horses. These horses are registered by the Mangalarga Marchador Horse Breeders Association, considering zootechnical controls, mainly of morphometric measurements. Traditional methods for performing morphometric measurements, such as the use of a hippometer, can be used for this control, but their use can lead to measurement errors due to the movement of the animal, and risk to the professional and the animal due to contact during the measurement. Therefore, the investigation of other measurement methodologies is essential, such as the use of mathematical models that predict morphometric segments, using images taken by smartphone. The objective of this study was to use two-dimensional images to predict morphometric measurements and to evaluate the automation of the prediction of morphometric measurements through convolutional neural networks (CNN). For the first stage, the models were developed using multiple linear regression (MLR), Support vector regression (SVR), and random forest (RF) methodologies. The factors sex, weight, stud farm, and segment of interest were considered for the development of the models. As a result, only sex did not obtain a positive result regarding the influence on the results, since there was not an insufficient number of animals to conclude the influence, despite the literature suggesting that it is an important factor. The methodologies addressed had good results regarding weight prediction, with similar results among the three, thus the most indicated is the MLR due to its simplicity. The second stage consisted of analyzing the images through CNN, an automatic evaluation methodology. CNN obtained good results, reaching a MAPE value of less than 10%. Thus, it can be stated that both manual and automatic prediction are capable of reliably predicting equine morphometric measurements.