Banca de QUALIFICAÇÃO: JHIORRANNI FREITAS SOUZA

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
STUDENT : JHIORRANNI FREITAS SOUZA
DATE: 18/12/2025
TIME: 13:00
LOCAL: https://conferenciaweb.rnp.br/users/aroldo-fereira-lopes-machado
TITLE:

Use of an unmanned aerial vehicle (UAV) to survey weeds in forest plantations for the restoration of the Atlantic Forest

KEY WORDS:

Image processing, remote sensing, plant competition, forest management.

PAGES: 53
BIG AREA: Ciências Agrárias
AREA: Agronomia
SUBÁREA: Fitotecnia
SPECIALTY: Matologia
SUMMARY:

The presence of weeds poses a significant challenge in forest restoration, requiring
efficient management of plant competition. Although herbicides are widely used due to
their low cost and effectiveness, their uniform application without precise identification
of the invasive species can lead to waste, reduced productivity, and environmental
impacts. Therefore, accurate identification of these plants is essential for more sustainable
management. In this context, high-resolution spectral data emerge as a promising tool for
weed identification, supporting rapid and non-destructive control in restoration areas.
This study was conducted in a restoration area of the Atlantic Forest at the Guapiaçu
Ecological Reserve (REGUA), in Cachoeiras de Macacu, Rio de Janeiro, Brazil. To
develop digital models for weed identification using Unmanned Aerial Vehicles (UAVs),
multispectral images were captured in May 2024 with a Mavic 3M at an altitude of 40 m.
Supervised classification algorithms, namely Random Forest (RF), Support Vector
Machine (SVM), and k-Nearest Neighbors (KNN), were tested, and the effect of
including spectral indices on accuracy was also evaluated. For each algorithm, two
approaches were compared: one using only RGB and multispectral bands, and another
incorporating 10 spectral indices derived from the images. Supervised classification
models showed high accuracy for the Bare Soil class (F1 between 0.83 and 1.00). The
Mulch class presented F1 between 0.61 and 0.85, Decomposed Mulch between 0.80 and
1.00, Tree class between 0.80 and 0.89, Assa-Peixe between 0.90 and 1.00, Anil between
0.71 and 1.00, and Brachiaria between 0.73 and 0.85. Increasing the number of samples
and incorporating spectral indices resulted in punctual improvements in accuracy. These
results indicate that the use of high-resolution spectral data combined with digital
supervised classification models is an effective tool for supporting weed management and
vegetation cover monitoring in restoration areas.

 

COMMITTEE MEMBERS:
Interno - 1905333 - AROLDO FERREIRA LOPES MACHADO
Interna - 2466219 - CAMILA FERREIRA DE PINHO
Externo ao Programa - 2161955 - ANDERSON GOMIDE COSTA - UFRRJExterno ao Programa - 2145654 - BRUNO ARAUJO FURTADO DE MENDONCA - UFRRJExterno à Instituição - CARLOS MAGNO MOREIRA DE OLIVEIRA