Rapeseed Yield Estimation Using UAV-LiDAR and an Improved 3D Reconstruction Method

Na Li, Zhiwei Hou, Haiyong Jiang, Chongchong Chen, Chao Yang, Yanan Sun, Lei Yang, Tian-Yu Zhou, Jingyu Chu, Qinqin Fan, Lijie ZhangAgriculture 2025 · DOI

TL;DR

This paper introduces a novel 3D reconstruction method (HybridMC-Poisson) for rapeseed plants using UAV-LiDAR data, enabling accurate canopy volume estimation. This volume is then used to build a robust yield prediction model, achieving a field-scale relative error of ~12%, which is crucial for precision agriculture.

What Problem Does This Paper Address?

Accurately estimating crop yield is vital for efficient agricultural management and sustainable development. Traditional manual methods are time-consuming, labor-intensive, and destructive, making them impractical for large-scale applications. While remote sensing, particularly UAV-based approaches, offers a promising alternative, challenges remain. Optical sensors suffer from spectral saturation under dense canopies, and 2D imagery lacks the crucial 3D structural information needed to accurately characterize complex plant architectures like rapeseed. This paper aims to bridge this gap by developing a reliable 3D reconstruction and yield estimation framework using UAV-LiDAR, overcoming the limitations of existing methods.

Methods

The core of this research is the development of an innovative HybridMC-Poisson algorithm for 3D reconstruction of individual rapeseed plants. This algorithm combines the strengths of two established methods: Marching Cubes (MC) for preserving local boundary details and Screened Poisson Reconstruction for achieving smooth surfaces. Key enhancements include:

  • Voxelization and Marching Cubes: The point cloud is converted into a binary voxel grid, and MC is used to extract an initial mesh, capturing fine boundary details.
  • Poisson Reconstruction and Adaptive Density Trimming: A smooth mesh is generated using Poisson reconstruction, followed by adaptive trimming based on vertex density to remove spurious patches in sparse regions, improving robustness against varying LiDAR data quality.
  • Rigid Registration and Fusion: The MC and Poisson meshes are rigidly registered and fused to retain global smoothness while preserving boundary fidelity.
  • Dynamic Error Feedback: A mechanism is implemented to adjust parameters and re-run reconstruction if the relative error between reconstructed and measured volume exceeds a threshold, ensuring accuracy.

For yield modeling, the study employed a linear regression approach. Initially, a through-origin model was developed to relate single-plant volume to single-plant yield. This was then scaled to the field level by introducing an integrated coefficient that accounts for planting density and other agronomic factors.

Datasets

The study utilized UAV-LiDAR data collected from a winter rapeseed field in Baoding City, China. Field data for yield measurement were collected from 60 quadrats, with plants categorized into low, medium, and high-density levels. For algorithm comparison, 20 representative rapeseed plants were selected and reconstructed using the proposed HybridMC-Poisson method and three baseline algorithms (Ball-Pivoting, Poisson, Alpha-Shape). Ground truth volumes were obtained using the water displacement method.

Key Insights and Ideas

  • 3D Structure is Key: The research strongly emphasizes that accurate 3D structural information, specifically canopy volume, is a more reliable predictor of rapeseed yield than traditional 2D spectral indices, which are prone to saturation.
  • Hybrid Approach Wins: Combining Marching Cubes and Poisson reconstruction in the HybridMC-Poisson algorithm effectively balances boundary detail preservation with surface smoothness, leading to superior reconstruction accuracy for complex plant structures.
  • Density Matters, But So Does Plant Number: While plant density significantly influences quadrat-level yield, individual plant volume showed less variation across density classes. This suggests that yield is primarily driven by the number of plants rather than their individual size.
  • Through-Origin Model is Biologically Sound: For yield modeling, a regression model passing through the origin (zero volume, zero yield) proved more biologically rational and provided better predictive power than models with an intercept.

Key Results

  • The HybridMC-Poisson algorithm demonstrated superior performance in single-plant reconstruction, achieving the lowest average relative volume error (2.96%) and better morphological accuracy (lower Hausdorff and Chamfer distances) compared to Poisson, Alpha-Shape, and Ball-Pivoting.
  • The algorithm achieved a favorable accuracy-efficiency trade-off, with an average runtime of approximately 3 seconds per plant.
  • A linear regression model relating single-plant volume to yield was established, with the field-scale model predicting a total yield with a relative error of approximately 12.4% compared to the measured yield.
  • The study found a significant negative correlation between actual plant volume and reconstruction error, indicating that larger plants are modeled more accurately.

Limitations and Future Work

The authors acknowledge that the study was conducted on a single experimental site and crop type, which may limit the generalizability of the model. Future work should focus on:

  • Extending validation to multiple crop varieties, ecological regions, and LiDAR configurations to enhance model adaptability and robustness.
  • Improving cross-regional adaptability and point cloud stitching efficiency.
  • Leveraging deep learning and multi-modal data fusion to further enhance robustness and scalability.
  • Investigating future improvements in enhancing spatial structure representation, optimizing quadrat design, and refining point cloud processing algorithms.

Summary updated

Abstract

Quantitative estimation of rapeseed yield is important for precision crop management and sustainable agricultural development. Traditional manual measurements are inefficient and destructive, making them unsuitable for large-scale applications. This study proposes a canopy-volume estimation and yield-modeling framework based on unmanned aerial vehicle light detection and ranging (UAV-LiDAR) data combined with a HybridMC-Poisson reconstruction algorithm. At the early yellow ripening stage, 20 rapeseed plants were reconstructed in 3D, and field data from 60 quadrats were used to establish a regression relationship between plant volume and yield. The results indicate that the proposed method achieves stable volume reconstruction under complex canopy conditions and yields a volume–yield regression model. When applied at the field scale, the model produced predictions with a relative error of approximately 12% compared with observed yields, within an acceptable range for remote sensing–based yield estimation. These findings support the feasibility of UAV-LiDAR–based volumetric modeling for rapeseed yield estimation and help bridge the scale from individual plants to entire fields. The proposed method provides a reference for large-scale phenotypic data acquisition and field-level yield management.
BibTeX
@inproceedings{li2025rapeseed,
  author = {Na Li and Zhiwei Hou and Haiyong Jiang and Chongchong Chen and Chao Yang and Yanan Sun and Lei Yang and Tian-Yu Zhou and Jingyu Chu and Qinqin Fan and Lijie Zhang},
  title = {{Rapeseed Yield Estimation Using UAV-LiDAR and an Improved 3D Reconstruction Method}},
  booktitle = {Agriculture},
  year = {2025},
  doi = {10.3390/agriculture15212265},
}
Figure