Identificação automatizada do escore de condição corporal bovino

Abstract

This work proposes the automation of the identification of ECC in cattle using advanced PID and AM techniques. Given the variability and subjectivity inherent in traditional methods of ECC evaluation, the development of an automated system aims to provide a more objective and accurate approach. The methodology included the capture and processing of digital images of cattle, the selection and training of convolutional neural network models, and the validation of the models through a three-way holdout approach. The models explored included VGG16, ResNet50, and Inception-V3, with VGG16 demonstrating superiority in initial tests. The Hyperband optimization technique was used for hyperparameter tuning and the Optuna framework to refine the models. The results showed a reduction in EQM over the epochs, with the best model achieving an EQM value of 0.126 for the test set, with a 95% confidence interval of [0.089, 0.164]. These results indicate that the developed model can classify EQM with precision. This study contributes to herd management by providing a tool that, when applied, can assist producers in accurately monitoring the nutritional status of cattle.