Impact of viewing angles in high-resolution imagery of tomato canopy on yield and chlorophyll estimation

• Grant number: 24/23051-4
• Support Opportunities: Scholarships abroad - Research Internship - Doctorate
• Start date: June 02, 2025
• End date: December 01, 2025
• Field of knowledge: Physical Sciences and Mathematics - Geosciences
• Principal Investigator: Rouverson Pereira da Silva
• Grantee: Vinicius dos Santos Carreira
• Supervisor: Wouter Hendrik Maes
• Host Institution: Faculdade de Ciências Agrárias e Veterinárias (FCAV). Universidade Estadual Paulista (UNESP). Campus de Jaboticabal. Jaboticabal , SP, Brazil
• Institution abroad: Ghent University (UGent), Belgium
• Associated to the scholarship: 22/16084-8 - A framework for high-resolution remote sensing in tomato crop upon minicomputer and cloud computing, BP.DR

Abstract

Current digital and precision agriculture research largely relies on UAV workflows that generate orthomosaics from nadir-view imagery. This approach excludes valuable data from alternative viewing angles and introduces blending steps during image processing. This issue may be particularly critical for specialty crops, such as staking tomatoes, which are highly anisotropic, with vertical canopy and dense foliage that often weakens the correlation between spectral data and agronomic traits. Therefore, the objective of this internship project is to investigate the impact of multispectral sensor viewing angles on tomato canopy reflectance and their effects on the performance of machine learning models for estimating yield and chlorophyll. Advanced high-resolution imagery techniques, including extraction of viewing angles and multi-angular imagery, will be studied and developed at the UAV Research Center at Ghent University. Nadir (roll = 0º and pitch = 0º) and oblique (roll = 0º and pitch = 20º) UAV flights will be conducted to leverage multi-angular, high-resolution imagery on tomato canopy. An optimized workflow will be developed to extract VZA (View Zenith Angle) and VAA (View Azimuth Angle) for each pixel in multispectral images. The data will then be divided into four groups: nadir-view, multi-angular (all views), multi-angular (grouped views by VZA/VAA), and the standard orthomosaic. Each dataset will be used independently as input for machine learning algorithms and evaluated using state-of-the-art metrics to determine which approach provides the best fit for tomato yield and chlorophyll estimation. The results will potentially provide new insights into UAV imagery standards and machine learning estimations for tomato crops, contributing to challenges adressed in the ongoing FAPESP projects no. 2022/16084-8 and no. 2021/06029-7.