CLASSIFICANDO ESTRATOS VEGETAIS DE UMA AREA DO BIOMA CAATINGA COM IMAGENS DE VANTS

Resumo

This article presents a low-cost UAV that was assembled and used in the classification of vegetation strata in a Caatinga biome area, aiming to enable the capture and classification of images in a more accessible manner. The high cost of these aircraft has been a barrier to research and agricultural development in disadvantaged regions of Brazil. In this study, the capture and processing of images played an important role in binary classification, subjecting them to the MobileNetV2 Neural Network. The results achieved an accuracy of 93% for the Herbaceous stratum, 94% for the Shrub stratum, and 83% for the Tree stratum, and 91% in the multiclass classification of the three strata, highlighting the potential of the proposed approach.