Coffee Leaf Diseases Identification and Severity Classification using Deep Learning

In this paper, we propose a method for automatic identification and classification of leaf diseases and pests in the Brazilian Arabica Coffee leaves. We developed a Machine Learning model, trained with the BRACOL public image dataset, to evaluate if a given image of a leaf has a disease or pest — Miner, Phoma, Cercospora and Rust — or if it is healthy. We then compared our model with other famous and well-known classification models, and we were able to achieve an accuracy of 98,04%, which greatly exceeds the accuracy of the other methods implemented. In addition, we developed an assessment to perform a classification related to the percentage of each leaf that is affected by the disease, achieving an accuracy of approximately 90%.