APPLICATION OF UAS-BASED RGB IMAGING FOR HIGH-THROUGHPUT MAIZE PLANT HEIGHT PHENOTYPING

UAS-derived RGB imagery offers a rapid, efficient, and non-invasive approach to high-throughput phenotyping, enabling cost-effective assessment of key agronomic traits in maize. During the 2025 growing season, a panel of 400 maize genotypes, comprising 200 hybrids and their respective parental components, was evaluated for plant height across two field locations in Serbia. RGB images were acquired weekly throughout the growing season, with acquisition frequency increased to twice weekly during pollination, using a DJI Mavic 3M UAS at a ground sampling distance (GSD) of 0.42–0.54 cm pixel⁻¹. RGB orthomosaics and digital elevation models (DEMs) were processed for plot-level data extraction, with plant height estimated using a surface differencing approach. A Digital Surface Model (DSM) derived from UAV imagery was differenced with a bare-soil DEM to generate Canopy Height Models (CHMs), from which the mean, median, 95th percentile (P95), maximum height, and standard deviation were calculated for each plot. Regression analysis revealed nonlinear temporal dynamics in maize height development, with the cubic model providing the best fit (R² = 0.885), compared with quadratic (R² = 0.869) and linear models (R² = 0.729). Across both environments, UAV-derived maximum canopy height showed the strongest correlation with manually measured plant height (r = 0.594 and 0.509; p < 0.001), whereas mean and median canopy height metrics showed weaker and less consistent associations. These findings indicate that maximum canopy height is the most informative UAV-derived metric for estimating maize plant height. Future work will focus on improving DSM normalization and ground-surface estimation, minimizing canopy occlusion and plot-edge effects, and enhancing the robustness and stability of UAV-derived height metrics across growth stages and varying canopy densities.