Identificação de bicho-mineiro e ferrugem no cafeeiro utilizando processamento digital de imagens e redes neurais convolucionais

Abstract

The world’s demand for coffee increases every year, reaching 178.5 million 60 kg bags in the period 2022-2023, an increase of 1.7% compared to the previous period 2021-2022. The total production of Brazil’s coffee harvest in 2022 was calculated at 50.92 million bags of 60 kg of processed coffee 2, thus making it the world’s largest producer of the product. With this production volume, there is a growing need to improve product quality due to the demands of national and international markets. However, pests such as leaf miner and rust cause extensive damage to coffee plantations, resulting in crop losses annually. Various methods and techniques have been developed and applied to assess the level of infestation and control of these pests. Among these techniques are the use of computer vision and convolutional neural networks (CNNs). Thus, the objective of this work was to develop computational tools to correctly identify the presence of pests, reducing evaluation time, evaluator error, and labor costs. The accuracies of these methods developed were between 99.67% and 97.00%. In addition, a tool was developed to quantify the degree of infestation, achieving an accuracy of 86.67%.