Maize Plant Height Estimation Using UAV and PlanetScope Imagery: A Comparison of 3-D Point Cloud and Machine Learning Approaches

Plant height (PH) is an essential indicator of crop growth and development, playing a critical role in field management and yield estimation. While recent advancements in remote sensing technologies have improved PH measurement capabilities, existing methods often encounter challenges related to accuracy, spatial resolution, and environmental conditions. In this study, we utilized low-altitude uncrewed aerial vehicle (UAV) imagery and Planet satellite remote sensing data to explore PH measurement methods based on 3-D reconstructed point clouds and inversion from spectral and texture features. First, a 3-D point cloud method was used to generate a canopy height model (CHM) from a digital surface model (DSM) and a digital terrain model (DTM). Second, this study compared the accuracy of four machine learning methods: eXtreme gradient boosting (XGBoost), random forest (RF), partial least-squares regression (PLSR), and support vector regression (SVR), for PH inversion at the field scale. The results show that the direct 3-D point cloud method is the most accurate for PH estimation, achieving an$R^{2}$of 0.98. For the machine learning inversion, the fusion of texture and spectral features consistently yielded the best performance for both data sources. Using high-resolution UAV imagery, XGBoost and RF models achieved the highest performance with an$R^{2}$of 0.74. With 3-m PlanetScope imagery, the XGBoost model was the top performer, reaching a maximum$R^{2}$of 0.59. On the contrary, the SVR model was consistently outperformed across all scenarios, highlighting the importance of appropriate model selection. Ultimately, high-resolution UAV imagery provided more accurate and reliable data for PH inversion compared to PlanetScope imagery. This study confirms the superiority of direct 3-D measurement and provides a clear guide for feature engineering and model selection in spectral-based PH estimation, providing new directions for future crop monitoring and precision agriculture research.