Banca de DEFESA: THAÍS MACHADO DE SOUZA

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : THAÍS MACHADO DE SOUZA
DATE: 28/07/2025
TIME: 09:00
LOCAL: Ambiente Virtual
TITLE:

APPLICATION OF MACHINE LEARNING TECHNIQUES FOR WEED IDENTIFICATION IN MAIZE: A COMPARISON BETWEEN MAXIMUM LIKELIHOOD, RANDOM FOREST, AND SVM

KEY WORDS:

remote sensing; supervised classification; vegetative stages; spectral analysis.

PAGES: 69
BIG AREA: Ciências Agrárias
AREA: Engenharia Agrícola
SUBÁREA: Máquinas e Implementos Agrícolas
SUMMARY:

Maize (Zea mays L.) is an of the most important crops in Brazilian and global agribusiness, serving as an essential source of food for both humans and animals, as well as a raw material for the pharmaceutical, chemical, energy, and food industries. However, its productivity can be severely affected by the presence of weeds, which compete with the crop for water, nutrients, and sunlight. The visual similarity between species, particularly at early growth stages, hinders their identification. In this context, this study applied Remote Sensing (RS) techniques for the spectral discrimination between maize and the weed Cyperus rotundus, using multispectral imagery acquired via Remotely Piloted Aircraft Systems (RPAS). Three supervised classifiers – Maximum Likelihood (ML), Random Forest (RF), and Support Vector Machine (SVM) – were evaluated and applied across different phenological stages and in two crop cycle (second-crop and crop). Overall Accuracy (OA), along with precision, recall, and F1-score per class, were used to assess the algorithms’ performance in classifying four targets: Zea mays L., Cyperus rotundus, soil, and shadow. An improvement in classification performance was observed as the crop advanced phenologically. In the first year (second-crop), the V8 stage proved to be the most suitable for plant differentiation, with ML achieving the highest performance (OA = 89%). For the class Zea mays L., precision was 0.88, recall 0.81, and F1-score 0.85; for Cyperus rotundus, precision was 0.68, recall 0.89, and F1-score 0.77. In the second year (crop), the best differentiation occurred at the V6 stage. Once again, ML showed the best performance (OA = 88%), with values of 1.00 for precision, 0.78 for recall, and 0.88 for F1-score in the Zea mays L. class, and 0.52 for precision, 1.00 for recall, and 0.68 for F1-score in the Cyperus rotundus class.

COMMITTEE MEMBERS:
Presidente - 1050615 - MURILO MACHADO DE BARROS
Interno - 2161955 - ANDERSON GOMIDE COSTA
Externo à Instituição - FLÁVIO CASTRO DA SILVA - UFF