MAPPING OF WEEDS IN CROPS OF SUGARCANE BY IMAGES OBTAINED FROM AN UNMANNED AERIAL VEHICLE (UAV)

Brazil is a global leader in sugarcane production, a crop that provides numerous jobs, generates income, and drives economic development, making it one of the most vital sectors of the Brazilian economy. Weeds in sugarcane fields negatively impact both the productivity of the crop and the quality of the harvested product. Therefore, effective weed control is crucial, with post-emergent herbicides currently being the primary method. However, the extensive use of herbicides can lead to various problems, including high costs and negative effects on the soil’s physical, chemical, and biological properties. Consequently, mapping weeds can serve as a valuable tool for control, offering three key benefits: significant economic savings, reduced environmental impact, and a lower risk of herbicide-resistant weed emergence. To address this, numerous studies have explored weed mapping using aircraft and satellite imagery. While these remote sensing methods can cover large areas, their spatial and temporal resolution is often insufficient for accurate pattern recognition. A promising alternative is the use of unmanned aerial vehicles (UAVs), which capture images with higher spatial and temporal resolution. This study proposes a weed mapping system based on machine learning techniques, using RGB images captured by a UAV. The system was initially tested with three classifier models, and the Artificial Neural Network (ANN) yielded the best performance, achieving an overall accuracy of 76% and a Kappa coefficient of 0.72.