Native social bees have shown a high agronomic potential in the pollination of plants in Brazil,
with bees of the genus Scaptotrigona standing out for improving the productivity of açaí palm
(Euterpe oleracea) by up to 70% and coffee plants (Coffea arabica) by up to 30%, with the
help of these pollinators. However, a drastic reduction in bee populations has been observed,
attributed to several causes, such as the destruction of natural habitats, the increase in
agricultural practices, deforestation leading to the loss of plant diversity, climate change, and
the use of pesticides. These threats not only directly affect the bees but also compromise
pollination, which is essential for maintaining ecosystems and food production worldwide.
Considering this scenario, this work aims to monitor the behavior of Scaptotrigona bees, also
known as "canudo" bees, using Artificial Intelligence (AI) tools to obtain information that will
support further research for the technological development of beekeeping and the dissemination
of knowledge about bees and their importance. For this, a new image acquisition methodology
was developed, using 3D printing, to create an unprecedented database with 7,806 images of
canudo bees, containing 19,954 annotations, which supported the construction of a neural
network model using the YOLOv8 network to classify the scapto, scapto_garbage and
scapto_polen classes, with 96% accuracy in this task. Additionally, the model demonstrated
great potential for estimating hive populations, selecting the most hygienic hives, and analyzing
the bees' preference for certain blooms, while also indirectly assisting in botanical studies to
better understand the flowering period.
|
