Sub-Saharan Sheep Breeds classifcation and generation based on CNN and GANs

Abstract

This thesis examines the application of “deep learning” techniques for computerized sheep breed identifcation. Traditional methods for identifying sheep breeds are Time-consuming and prone to errors. To confront this problem, we develop a robust deep learning model using Convolutional Neural Networks (CNN1s), “data-augmentation” and Generative Adversarial Networks (GAN2s). Our dataset of sheep images is used for training and adjusting the models. We compare the performance of our “deep learning” Procedure to traditional methods, taking into account factors such as image quality. The results show the Capability of deep learning to precisely recognize sheep breeds, which has benefts for livestock and Farming techniques.