Coffee cultivation is vital to the global economy, but faces significant challenges with diseases such as rust, miner, phoma, and cercospora, which impact production and sustainable crop management. In this scenario, deep learning techniques have shown promise for the early identification of these diseases, enabling more efficient monitoring. This paper proposes an approach for detecting diseases and pests on coffee leaves using an efficient single-shot object-detection algorithm. The experiments were conducted using the YOLOv8, YOLOv9, YOLOv10 and YOLOv11 versions, including their variations. The BRACOL dataset, annotated by an expert, was used in the experiments to guarantee the quality of the annotations and the reliability of the trained models. The evaluation of the models included quantitative and qualitative analyses, considering the mAP, F1-Score, and recall metrics. In the analyses, YOLOv8s stands out as the most effective, with a mAP of 54.5%, an inference time of 11.4 ms and the best qualitative predictions, making it ideal for real-time applications.

Coffee cultivation is of extreme economic importance in many regions of the world, but productivity is hampered by the various diseases and pests that affect the leaves of the plants, damaging both the quality and yield of the harvest. In this context, deep learning presents itself as a promising solution for the automatic identification of plant diseases, reducing dependence on human inspection and increasing efficiency in crop management. In this sense, this study proposes a novel two-stage approach, detecting the diseased region of the coffee leaf and classifying the diseases into Miner, Rust, Cercospora and Phoma on coffee leaves. A new dataset, derived from the BRACOL and Diseases and Pests in Coffee Leaves datasets, was created and used to improve class balance and robustness. In the first stage, the YOLOv8 model is being used to detect the diseased regions. For the second stage, the InceptionResNetv2, DenseNet169, Resnet50 and ShuffleNet models are being trained and used to classify the detected region, and a modification to a low computational cost classification architecture called PavicNet-MCv2 is being proposed. The results obtained are compared and the performance analysis of the detection models shows that YOLOv8 obtained the best performance with a mAP (Mean Average Precision) of 85.1% and for classification the DenseNet169 model obtained the highest average accuracy with 97.93%. The PavicNet-MCv2 model presents itself as the best alternative with reduced complexity and an accuracy of 97.77%. The combination of promising performance and reduced computational cost suggests that PavicNet-MCv2 can be integrated into plantation monitoring systems, contributing to more effective management of diseases and pests affecting coffee production.