Classificação de Doenças em Folhas de Macieira utilizando Redes Neurais Convolucionais

Abstract

Since 2021, Brazil has joined the group of the world's largest producers of apple fruits. However, the phytosanitary diagnostic process relies on visual inspection, requiring specialized training that increases operational costs. So, the following work addresses the application of Convolutional Neural Networks (CNNs) as an effective strategy to identify diseases in apple leaves, aiming to provide a solution that can replace or enhance the conventional process of visual inspection widely adopted in agricultural f ields. The images used by the model were taken from the 'New Plant Diseases Dataset' available on Kaggle, resized, and subjected to a series of normalization enhancements. The goal was to identify whether an apple leaf was healthy or not, and in case of a negative diagnosis, to determine which of the three provided disease types the sample belonged to. The model achieved an accuracy of 83%, precision of 85%, recall of 80%, and f1-score of 82%.