Identificação de ferrugem, cercospora e bicho-mineiro em folhas de cafeeiro por meio de técnicas de inteligência artificial

Abstract

Phytosanitary events affecting coffee leaves are among the main challenges in coffee production, especially coffee leaf rust, cercospora leaf spot, and the coffee leaf miner, which cause premature defoliation, reduction of the photosynthetic area, and consequently decreased productivity. Traditional diagnostic methods based on visual inspection present limitations related to subjectivity, analysis time, and the need for specialists, compromising the efficiency of crop monitoring. In this context, this study proposes the development and evaluation of models for the automatic classification of coffee leaf diseases using computer vision and machine learning techniques. Established convolutional neural network architectures, such as ResNet, DenseNet, and EfficientNet, are investigated in order to identify those capable of more accurately distinguishing healthy leaves from those affected by rust, cercospora, or leaf miner. Additionally, Principal Component Analysis (PCA) and the Minimally Random Convolutional Kernel Transform (MiniRocket) are applied for dimensionality reduction and processing optimization, enabling the use of traditional classifiers such as Support Vector Machines (SVM) and Random Forest, and promoting computationally lighter solutions. Model evaluation is conducted using metrics such as accuracy, precision, recall, and F1-score, along with an analysis of carbon impact and processing time associated with training and inference, aiming to promote more parsimonious and sustainable artificial intelligence approaches within the context of digital agriculture. The results indicate good model performance (accuracy ranging from 0.7311 to 0.9750, precision from 0.7453 to 0.9751, recall from 0.7410 to 0.9751, and F1-score from 0.7323 to 0.9784 between training and final validation, respectively) under controlled conditions, although significant challenges remain when applied to real field scenarios