MAPEAMENTO DIGITAL DE PLANTAS DANINHAS EM ÁREAS DE PRODUÇÃO DE SOJA UTILIZANDO AERONAVES REMOTAMENTE PILOTADAS

Abstract

Weeds are one of the most important pests in agroecosystems, causing considerable economic losses in global production. Thus, the first study aimed the identification of weeds through images as well the quantification of the weed seed bank in commercial areas in southwest Goiás-GO. In this way, the sampling was carried out through images and soil collections to carry out the seed bank present in 12 plots distributed in 12 municipalities. In the second study, the objective was to identify, quantify and map the weed species present in 384 plots distributed in 16 municipalities in commercial agricultural areas in southwest Goiás via aerial imaging. When evaluating the classic versus digital monitoring methods combined with the soil seed bank, it was possible to affirm that the RGB images generated by drones can be used for further training (machine learning), obtaining an 88% of accuracy. Thus, making it feasible to carry out a survey of weedy flora to identify, quantify and map the species in commercial agricultural areas in southwest Goiás. The high number of identified species of the same family, as well as the higher frequency of species observed in the ten-day periods before the starting of soybean crop, indicated the dominance of some species. The number of species identified, the size of the area, and the sampling region were not correlated, because sites with low species diversity were more grouped than sites with high diversity.