Uma abordagem de aprendizado profundo para a identificação automática das principais doenças de cultivares de laranja

Abstract

The orange is a fruit produced on a large scale worldwide, with Brazil and China standing out among the main producers. Diseases that affect oranges can cause significant losses, making early detection essential to prevent their spread. This work aims to implement and analyze Convolutional Neural Networks (CNNs) for the detection of diseases in oranges, using artificial intelligence techniques. The work presents the main diseases that affect orange-producing trees, such as Citrus Canker, Melanose, Citrus Greening, and Black Spot, and also presents the tools that were used to achieve the objective of the study, such as TensorFlow and Keras.In addition, two studies in the same area of plant disease detection through images using neural networks are presented, in order to compare the results of this work with others in the same field. A database of images of orange tree leaves is presented, where data augmentation and data preproces- sing techniques were applied, later used in training and validating the neural network. The work focuses on the process of building convolutional layers, defining activation functions, and using optimization techniques such as gradient descent and backpropagation. After applying the concepts presented during the methods, it was possible to obtain results when tested on images not shown during training. Accuracy and PR-AUC metrics of 72.34% and 90.75% stood out. Despite being a promising result, its performance was lower than that of the related works presented, highlighting the possibility of future improvements