Classificação de Culturas de Batata Usando Deep Learning Avançado com Mixup

Uma Abordagem para Aumentar a Precisão na Identificação de Variedades de Batata em Imagens Agrícolas

Abstract

This work explores the application of advanced deep learning techniques for the classification of potato varieties, a widely consumed tuber. Using 1,200 potato images, the study implemented a Convolutional Neural Network (CNN) with the Mixup method, which combines training examples to improve model generalization. The model achieved an average accuracy of 99.75% on the test set, with 100% accuracy in one of the validations. The paper discusses the importance of data augmentation approaches and the effectiveness of Mixup in preventing sample memorization, as well as highlighting the relevance of deep learning in smart agriculture. The results demonstrate the feasibility of this methodology, which can be applied to the automatic classification of various potato varieties, contributing to more sustainable and profitable agricultural practices. The data and experiments are publicly available in a GitHub repository for validation and replication.