Coffee Leaf Rust Detection Using Convolutional Neural Network

Abstract

Rust is a severe disease affecting many productive coffee regions. It is caused by a pathogenic fungi that attacks the underside of coffee leaves and it is characterized by the presence of yellow-orange and powdery points. If not treated, rust can cause a drop in coffee production of up to 45%. In this sense, this paper presents a contribution to the problem of rust identification that doesn't use “handcrafted” features, i.e., features extracted according to rules established by human programmers. Instead, we propose to train a Convolutional Neural Network (CNN) to learn to identify rust infection. We evaluated our CNN in a set of images provided by an expert and comparison results show that our approach is able to to detect the infection with a high precision, as corroborated by the high Dice coefficient obtained.