Um estudo comparativo de modelos de aprendizado de máquina para classificação de flores apícolas: integrando extratores de texturas e classificadores

Abstract

The Beekeeping Flora comprises the group of plants from which bees extract floral resources, which may include pollen, nectar, or both. Since the beekeeping production potential of a region is determined by its floral coverage, understanding the foraging resources available to bees contributes to planned management and the optimization of apiaries. In this context, this study proposes the development of a computational model based on machine learning to classify flower images into three categories: pollen-producing, nectar-producing, and both nectar and pollen producing. The objective is to provide insights into the beekeeping production potential of a given region. The experiments were conducted using a dataset containing 1.145 flower images from the beekeeping flora of the Sertão Central region in the state of Ceará, Brazil. Feature extraction was performed using ten Convolutional Neural Networks (CNNs), while classification was carried out using SVM, KNN, MLP, and a proposed CNN. The study aimed to identify the best combination of feature extractor and classifier. The results demonstrated the efficiency of the proposed system, with the best combination being ResNet50 + SVM, achieving an MCC of 92.94%.