Classificação de imagens de sobras de dieta de bovinos confinados através de redes neurais convolucionais

Abstract

Animal nutrition is an important field in livestock production. An inappropriate formulation and bunk management can harm the animals health and the production. For years the feed bunk management was conducted subjectively and inconsistently, through manual and visual methods of classification. The continuous improvement of artificial neural networks and deep learning have contributed positively to problems in several fields with the use of computer vision techniques. The present work proposes a computational model capable of analyzing various images of leftover diets of confined bovines, in order to classify these images between scores established by the literature. Thus, a convolutional neural network was developed to classify bunk score images, achieving 89.58% of accuracy. In addition, the final model achieved computational performance metrics favorable for a low cost and a real time application for farmers.