Banca de QUALIFICAÇÃO: MARCOS PAULO CAVALCANTE FONSECA

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
STUDENT : MARCOS PAULO CAVALCANTE FONSECA
DATE: 28/08/2026
TIME: 13:00
LOCAL: https://conferenciaweb.rnp.br/users/aroldo-fereira-lopes-machado
TITLE:

Use of Unmanned Aerial Vehicles (UAVs) and Artificial Neural Networks for Weed Monitoring in Atlantic Forest Restoration Areas

KEY WORDS:

Artificial Intelligence, Imaging, Georeferencing, Reforestation.

PAGES: 27
BIG AREA: Ciências Agrárias
AREA: Agronomia
SUBÁREA: Fitotecnia
SPECIALTY: Matologia
SUMMARY:

In Brazil, ecological restoration is expanding. One of the major challenges is the management of weed competition. Many areas where restoration projects are carried out are occupied by weeds, which hinder the establishment of native species and therefore require appropriate management. Monitoring and identifying these weeds is the first step toward the sustainable management of weed competition. Remote sensing can become an important tool for decision-making in weed management. Artificial intelligence and deep learning can be valuable tools for accurately identifying weeds and distinguishing them from species of interest. The objective of this project is to select algorithms capable of detecting critical periods for weed management interventions in forest restoration areas using UAV-acquired imagery and deep learning models. Field plots will be established in ecological restoration areas in the municipality of Cachoeiras de Macacu, Rio de Janeiro, Brazil. The study area covers 2,400 m² (40 × 60 m) and will be divided into four homogeneous blocks of 600 m², with individual plots of 100 m². Three sampling points will be established in each plot using a 1 × 1 m quadrat. These points will be georeferenced using the Global Positioning System (GPS). Plants within each quadrat will be identified and quantified, and their aboveground biomass will be collected for subsequent dry matter determination. Initially, UAV flights will be conducted over the experimental area at an altitude of approximately 60 m to acquire multispectral imagery, which will subsequently be compared with field data. Data from each flight and field survey will be used to validate the machine learning models under evaluation. Based on the field data, Relative Frequency (RF), Relative Density (RD), Relative Dominance (RDo), and the Importance Value Index (IVI) will be calculated. Five image segmentation tools and approaches will be evaluated: RStudio®, Detectron2®, QGIS®, Google Earth Engine®, and U-Net®. During the training and classification stages, different algorithms will be tested and compared in terms of their consistency and efficiency against ground-truth field data. The expected outcomes are to monitor and characterize weed populations in forest restoration areas using UAVs and remote sensing; generate phytosociological data to support more accurate weed management decisions; and select an AI-based approach capable of recognizing weed species in forest restoration areas. Its performance will be compared with conventional field-based identification methods to determine the most consistent and accurate approach.

COMMITTEE MEMBERS:
Interno - 1905333 - AROLDO FERREIRA LOPES MACHADO
Externo ao Programa - 2145654 - BRUNO ARAUJO FURTADO DE MENDONCA - UFRRJExterno ao Programa - 1050615 - MURILO MACHADO DE BARROS - UFRRJ