Li, Qing; Hao, Dalei; Zeng, Yelu
Abstract
Leaf inclination angle (LIA) distribution is a key canopy architecture trait that influences the light use efficiency and radiation transmission. Existing LIA measurement methods are accurate but difficult to implement at the large scale, creating a "phenotyping bottleneck" in both agriculture and ecology. In this study, we present a systematic framework for extracting field-level mean leaf angle (MLA) using Unmanned Aerial Vehicle (UAV) photogrammetry. The framework integrates three key steps: deep learning-based super-resolution to enhance orthomosaic resolution to 0.28 cm/pixel; U-Net-based semantic segmentation for precise individual leaf delineation (achieving an Accuracy of 0.93, with an F1 Score of 0.84, a Recall of 0.84, and a Precision of 0.85); and two distinct LIA extraction methods, LIA_PC (point cloud-based) and LIA_DSM (Digital Surface Model-based), both employing Random Sample Consensus plane fitting. The performance of the LIA_PC method, which uses direct 3D point cloud analysis, demonstrated high accuracy with a coefficient of determination (R² ) of 0.94 and a Root Mean Square Error (RMSE) of 1.74 degrees when validated against ground-truth measurements. In contrast, the LIA_DSM method, while more computationally efficient, had a lower accuracy (R² = 0.54, RMSE = 9.43 degrees). Furthermore, this study demonstrated the functional importance of the structural measurements by revealing that leaf orientation is a key driver of the leaf-scale apparent reflectance, impacting the Near-Infrared (NIR) and RedEdge reflectance. The results show that this framework can effectively generate field-scale MLA maps with high spatial resolution. By providing quantitative metrics of canopy architecture across large populations, this approach serves as a scalable method for applications in ecology and agriculture.
A UAV-based framework for field-scale extraction of leaf inclination angles in soybean canopies.pdf
