Reconhecimento de Plantas Alimentícias Não Convencionais (PANCs) em Ambientes Naturais Utilizando o Modelo YOLO: Uma Abordagem Baseada em Visão Computacional

Abstract

Non-Conventional Food Plants (PANCs) have gained prominence due to their nutritional value and sustainable cultivation, but they still face recognition chal lenges given Brazil’s vast biodiversity and the limited educational resources availa ble. This work aims to identify and localize PANCs in images using the YOLOv11 mo del, one of the most recent versions of the You Only Look Once (YOLO) architecture. For training and testing the model, seed 0 was used, with a total of 15,968 images and 19,037 instances, applying data augmentation techniques and distributing the sam ples across 14 classes. The images were randomly and stratifiedly divided into 70% for training, 20% for validation, and 10% for testing. All samples were obtained from the Pl@ntNet website. The effectiveness of the proposed method was evaluated using metrics such as precision, recall, confusion matrix, loss functions, mean Average Pre cision (mAP), and heatmaps generated with the Eigen-CAM algorithm. For the latter, the most activated regions were analyzed to identify the model’s areas of attention, improving its explainability. The model achieved a precision of 0.877, a recall of 0.778, an mAP50 of 0.878, and an mAP50-95 of 0.801, facing greater difficulties with classes that are similar or have a wide variety of characteristics.