IDENTIFICAÇÃO DE ABELHAS NATIVAS NO CERRADO COM O USO DE INTELIGÊNCIA ARTIFICIAL

Abstract

This study proposed and evaluated the use of convolutional neural networks, through the YOLO architecture, for automatic identification of native Cerrado bees from images, contributing to biodiversity conservation and to the advancement of tools supporting pollinator monitoring. Images obtained from different sources and properly labeled were used, covering several native species. In the general experiment, the model achieved significant metrics, with strong precision and recall rates for most species, demonstrating its efficiency in detection and classification. Additionally, a second experiment was carried out focusing on the morphologically similar species Trigona hyalinata (Guaxupé) and Trigona spinipes (Irapuã), in which the model achieved an overall accuracy of 79.67%, a precision of 79.35% for Guaxupé, and 80.00% for Irapuã, confirming its discrimination capability even in scenarios of high visual similarity. The results demonstrate the feasibility and robustness of the approach, highlighting its potential for application in monitoring programs, sustainable management, and conservation of native bees in the Cerrado.