Classificação de galinhas para o manejo em avicultura utilizando redes neurais convolucionais

Abstract

The livestock sector, with poultry farming as one of its main exponents, has been one of the main driving forces in the global economy for generating income, jobs and moving the agribusiness. The development of poultry farming in Brazil was due to new production systems related to biosecurity, health, quality of animals and meat. On the other hand, the sector faces difficulties related to the handling, monitoring and welfare of animals, together with problems related to the lack of qualified labor. Artificial intelligence systems appear more and more in the agricultural sector and become a solution to this type of problem, as they simulate the human brain, managing to have their own understanding and thus, enabling themselves to replace the traditional workforce. Convolutional neural networks (CNN) are a subtype of machine learning that uses large amounts of data, seeking to detect patterns in images and learn different levels of abstraction. This work aims to develop a method capable of detecting chickens in images, through artificial intelligence, to help professionals in poultry management. For the development of this work, 468 photographic images with the same dimension pattern acquired in the field were used. The photos went through a labeling process and after that they were separated into 80% for training, these subdivided into 80% training and 20% for validation, and 20% for testing. The photographs labeled with their respective originals were passed to a YOLO network that carried out the training, generating a model at the end. The model presented a 50% accuracy rate in classification. It is concluded that this work presented promising results in the classification with the acquired images. For the future objective, it is intended to increase the amount of images for each base and make changes in some parameters of the neural network.