CLASSIFICAÇÃO AUTOMÁTICA DE ESTÁGIOS DE MATURAÇÃO DO TOMATE USANDO REDES NEURAIS CONVOLUCIONAIS YOLO

Abstract

The tomato is one of the most widely consumed fruits globally, but its manual harvesting and color-based classification remain labor-intensive processes prone to human errors, such as subjectivity and inconsistency among evaluators. With advancements in agricultural automation, computer vision methods have been applied to enhance efficiency and standardization in fruit detection. This study utilized YOLO convolutional neural networks to classify cherry tomatoes into three ripening stages (green, intermediate, and ripe) based on a manually labeled image dataset. YOLO versions v8, v9, v10, and v11 were tested, comparing performance in terms of precision, recall, mAP50, and training time. YOLO v9 achieved the highest precision, while YOLO v11 stood out for its training speed and strong recall. Additionally, segmentation proved to be more effective than bounding boxes for image annotation. Considering all evaluated metrics, YOLO v11 was identified as the best option for automated tomato classification, offering a balance between efficiency and performance.