Classification of Coffee Leaves Using Smartphone Images and Convolutional Neural Networks

The coffee is part of Brazilians’ life, an essential product in agribusiness in Brazil and the world. Brazilian coffee plantations present problems, such as the leaf miner pest; this disease and the pest, if not identified in a timely manner, can wipe out coffee plantations. This article aims to classify coffee leaves with the leaf miner pest, rust disease, and healthy leaves using Convolutional Neural Networks (CNN). The model that best classified the two classes was the MobileNet, measuring F 0.95414 on the test data. This MobileNet model was integrated into an application for the Android system, developed in Flutter, where it is possible to send an image from the gallery to perform the prediction. This proposed methodology can bring consistent results so the farmer can make quick and accurate decisions.