Detecção e classificação de alvéolos em favos de apis mellifera,utilizando algoritmos de deep learning

Abstract

The use of systems with machine learning in tasks performed by humans has become an important tool, making slow and error-prone activities faster and more assertive. One of these activities is beekeeping, a very important job for the preservation of bees and food production, even with its importance, this is still a very manual process with slow and error-prone tasks. With the development of topics such as machine learning and deep learning, new tools have emerged for the beekeeper. The work proposes to develop a deep learning model that helps in the process of counting and classifying alveoli in combs, an important task in the management of hives. For this development, the DeepBee project was used as a base and with it the drone class was separated and included in the model, contemplating 8 classes (capped, dontcare, drone, eggs,honey,larve,nectar and pollen). In addition to providing an algorithm that facilitates the monitoring of hives, through files. Two model architectures were used (MobileNet and ResNeXt) and the best one presented an f1-score of 95%.