AVALIAÇÃO DE MODELOS YOLO PARA INDENTIFICAÇÃO DE ESPÉCIES DE ABELHAS GUAXUPÉ E IRAPUÃ

Abstract

Bees develop a fundamental role as pollinating agents, being indispensible for the preserva tion and conservation of the biodiversity and the balance of the ecosystems, especially on the biome cerrado. However, the identification of native species is hindered by the so-called “taxonomic impediment,” intensified by the lack of specialists and the high morphological similarity among species. This study used YOLO convolutional neural networks to classify native Cerrado bees, Trigona hyalinata (Guaxup´e) e Trigona spinipes (Irapu˜a) because these species represent a higher level of difficulty in identification due to their similarity and the resulting taxonomic impediment. The objective was to minimize this limitation and contribute to the monitoring and conservation of these pollinators using a dataset of manually labeled images. In this work, the versions YOLO v8, v9, v10, v11, v12, and 26 were tested, comparing their performance in terms of precision, recall, mAP50, and F1 score.