Jiang et al. (2025) Improving sugar beet canopy mapping through UAV image analysis

Identification

• Journal: Scientific Reports
• Year: 2025
• Date: 2025-11-17
• Authors: Jianjun Jiang, Donghui Li, Qinru Qiu, Xiao Ling
• DOI: 10.1038/s41598-025-23868-1

Research Groups

• School of Electrical Automation and Information Engineering, Tianjin University, China
• Zhejiang Polytechnic University of Mechanical and Electrical Engineering, China
• Zhejiang Xinneng Photovoltaic Technology Co., Ltd, China

Short Summary

This study evaluated 18 image segmentation methods for estimating sugar beet fractional vegetation cover (FVC) from Unmanned Aerial Vehicle (UAV) RGB imagery. It found that the Excess Green (ExG) index combined with Otsu or Ridler–Calvard (RC) thresholding provided the most accurate FVC estimations, significantly outperforming other combinations.

Objective

• To accurately estimate canopy cover fraction (CCF) in sugar beet using UAV imagery and to compare the performance of various segmentation algorithms based on hybrid thresholding and discrimination techniques across different growth stages.

Study Configuration

• Spatial Scale: Sugar beet fields in Heilongjiang Sheng Province, China. Images covered an area of 16 m × 9 m (144 m²) with a spatial resolution of 0.004 m per pixel. Validation was performed across 30 plots, each divided into 746 × 573 pixel sub-regions. Row spacing was approximately 0.39 m, with an average plant spacing of 0.18 m.
• Temporal Scale: 2022 growing season, covering four distinct growth stages (from early four-leaf to late six-leaf stage). UAV images were acquired on May 15, 2022.

Methodology and Data

• Models used:
  • Vegetation Indices (6): Excess Green (ExG), Excess Green minus excess Red (ExGR), Excess Green minus excess Blue (ExGB), Green Leaf Index (GLI), Visible Atmospherically Resistant Index (VARI), Red-Green-Blue Vegetation Index (RGBVI).
  • Thresholding Algorithms (3): Otsu, Ridler–Calvard (RC), Two-Peaks (2P).
  • Ground Truth Generation: Supervised classification using the Mahalanobis Distance Classification (MDC) algorithm in ENVI 5.6 software.
• Data sources:
  • UAV Imagery: Acquired using a DJI Matrice 100 UAV equipped with a Zenmuse X3 RGB camera at flight altitudes of 8 to 12 m.
  • Ground Truth Data: Collected from 30 plots across four growth stages through manual inspection and supervised classification.

Main Results

• The combination of the Excess Green (ExG) index with either Otsu or Ridler–Calvard (RC) thresholding achieved the highest overall accuracy for sugar beet vegetation cover estimation (Normalized Root Mean Square Error (NRMSE) = 5.1%, Coefficient of Determination (R²) = 0.96).
• Green Leaf Index (GLI) with Otsu and RC thresholding followed closely in performance (NRMSE ≈ 6.7%, R² = 0.94).
• The weakest performance was observed with ExGB combined with the Two-Peaks (2P) method (NRMSE = 42.3%, R² = 0.34).
• ExG-based methods consistently showed superior performance across all four growth stages, with R² values up to 0.98 and average RMSE of approximately 4.7%.
• GLI-based methods demonstrated moderate accuracy (R² ranging from 0.93 to 0.96, average NRMSE ≈ 22.8%).
• RGBVI-based methods exhibited the weakest performance (R² ranging from 0.68 to 0.74, average NRMSE ≈ 28.8%).
• Lighting conditions significantly influenced the performance of vegetation indices; GLI showed higher accuracy under lower light intensity, while ExG's performance decreased under fluctuating light conditions.

Contributions

• Systematically evaluated 18 hybrid segmentation methods for accurate fractional vegetation cover (FVC) estimation in sugar beet using UAV RGB imagery.
• Identified the optimal combination of vegetation index (ExG) and thresholding algorithms (Otsu or RC) for sugar beet canopy mapping.
• Demonstrated the critical influence of varying lighting conditions on the performance of different spectral indices and highlighted the robustness of the Otsu algorithm.
• Provided practical guidelines for agricultural practitioners on selecting appropriate UAV flight timing, vegetation indices, and thresholding algorithms for precision agriculture applications.

Funding

• "Pioneer” and “Leading Goose” R&D Program of Zhejiang (project number: 2022C01105)
• "Pioneer” and “Leading Goose” R&D Program of Zhejiang (project number: 2024C01064)
• Zhejiang Transport Department’s 2021 Sci-Tech Project (2021032)

Citation

@article{Jiang2025Improving,
  author = {Jiang, Jianjun and Li, Donghui and Qiu, Qinru and Ling, Xiao},
  title = {Improving sugar beet canopy mapping through UAV image analysis},
  journal = {Scientific Reports},
  year = {2025},
  doi = {10.1038/s41598-025-23868-1},
  url = {https://doi.org/10.1038/s41598-025-23868-1}
}