Banca de DEFESA: BILTON GILBERTO NHANTUMBO

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : BILTON GILBERTO NHANTUMBO
DATE: 09/02/2026
TIME: 13:00
LOCAL: REMOTA
TITLE:

Digital phenotyping and artificial neural networks applied to estimate biometric variables of Moringa oleifera Lam seedlings

KEY WORDS:

Spectral reflectance; Computational learning; Vegetative monitoring; Morphometric attributes; Image analysis.

PAGES: 78
BIG AREA: Ciências Agrárias
AREA: Agronomia
SUBÁREA: Agrometeorologia
SUMMARY:

The growing demand for high-quality seedlings has driven the use of digital phenotyping techniques in the forestry sector, especially for fast-growing species such as Moringa oleifera Lam. With technological advances and the increasing need to improve efficiency in seedling production, digital phenotyping has emerged as a promising alternative for non-destructive monitoring, enabling fast, accurate, and low-cost evaluations. In this context, multispectral imaging techniques have been widely applied to characterize biometric attributes, support decision-making in forest nurseries, and optimize the management of forest species. The aim of this research was to assess the potential of multispectral images for estimating Leaf Area Index (LAI) and stem diameter of Moringa oleifera seedlings using multiple linear regression models and artificial neural networks (ANN). The experiment was carried out during the early growth of the seedlings, with weekly assessments involving the collection of traditional biometric variables (LAI and diameter) and the acquisition of images using a MAPIR Survey3 camera (red - R, green - G, and near infrared - NIR bands). For each assessment, vegetation indices NDVI, GNDVI, RVI, and CVI were calculated after radiometric correction and image segmentation. Pearson correlation showed strong associations between LAI and NDVI (r = 0.89), GNDVI (r = 0.98), and SAVI (r = 0.90), while stem diameter displayed high correlations with NDVI (r = 0.88) and GNDVI (r = 0.79). Multiple linear regression models demonstrated high performance in estimating LAI (R² = 0.935) and stem diameter (R² = 0.90), with GNDVI standing out as the most influential variable. The artificial neural network analysis produced superior results, achieving R² = 0.994 for LAI and R² = 0.992 for stem diameter, indicating excellent generalization capacity. The results demonstrate that the combination of digital phenotyping and statistical modeling or artificial intelligence algorithms constitutes a robust tool for estimating biometric attributes in seedlings, enabling significant advances in forest nurseries monitoring and supporting more efficient and sustainable management practices.

COMMITTEE MEMBERS:
Presidente - 1574025 - GUSTAVO BASTOS LYRA
Externo ao Programa - 1698763 - RICARDO VILAR NEVES - UFRRJExterno à Instituição - MARCEL CARVALHO ABREU