Banca de QUALIFICAÇÃO: NAISILA CARVALHO PRESTES

Uma banca de QUALIFICAÇÃO de MESTRADO foi cadastrada pelo programa.
DISCENTE : NAISILA CARVALHO PRESTES
DATA : 22/06/2026
HORA: 09:00
LOCAL: Meet
TÍTULO:

HYPERSPECTRAL DATA ANALYSIS AND SPECTRAL DIFFERENTIATION OF IPÊ SPECIES (HANDROANTHUS SPP. AND TABEBUIA SPP.)

PALAVRAS-CHAVES:

remote sensing; spectroradiometry; machine learning; spectral signatures; forest species.

PÁGINAS: 60
GRANDE ÁREA: Ciências Agrárias
ÁREA: Agronomia
SUBÁREA: Agrometeorologia
RESUMO:

Hyperspectral sensing associated with machine learning algorithms presents high potential for the spectral differentiation of native forest species, especially in tropical environments with high vegetation diversity. This study aimed to evaluate the efficiency of hyperspectral spectroradiometry in discriminating four ipê species (Tabebuia rosea, Tabebuia roseoalba, Handroanthus chrysotrichus, and Handroanthus impetiginosus) located in the municipality of Sinop, Mato Grosso, Brazil. Leaf samples were collected in an urban environment and subjected to spectral curve acquisition using the Ocean Optics STS-VIS-L-50-400-SMA sensor, within the spectral range from 450 to 824 nm. Spectral data were organized into specific bands and reflectance inflection difference (RID) indices and subsequently analyzed using multivariate statistical techniques and machine learning algorithms, including Random Forest, Support Vector Machine, Artificial Neural Networks, REPTree and J48 decision trees, as well as Logistic Regression. The spectral curves revealed differences in reflectance behavior among the species, mainly in the visible and near-infrared regions, associated with the biochemical and structural characteristics of the leaves. Principal component analysis enabled the identification of distinct groupings among the species, demonstrating a high contribution of spectral bands and RID indices to spectral discrimination. Among the evaluated algorithms, Random Forest and Support Vector Machine showed the best performance for the metrics of correct classification, Kappa coefficient, and F-score. The results demonstrated that the integration of hyperspectral sensing and machine learning constitutes an efficient and promising approach for the automated identification of tree species, contributing to applications in forest inventories,
environmental monitoring, urban afforestation, and biodiversity conservation.

MEMBROS DA BANCA:
Presidente - 131916001 - RIVANILDO DALLACORT
Interno - 265126001 - CARLOS ANTONIO DA SILVA JUNIOR
Externo à Instituição - PAULO EDUARDO TEODORO - UFMS