Improving pear fruit quality without yield loss through 3D point cloud-based estimation of reasonable fruit load

Area of Science:

• Horticulture
• Agricultural Engineering
• Computer Science

Background:

• Fruit quality and yield in pear production depend on appropriate fruit load (FL).
• Accurate leaf number estimation is crucial for FL regulation but remains challenging.
• Current methods limit practical precision FL management.

Purpose of the Study:

• To develop a data-driven framework for estimating leaf number and reasonable FL in pear trees.
• To integrate 3D point cloud-derived canopy structure with machine learning for trait extraction.
• To enable rapid and field-deployable FL regulation for improved orchard management.

Main Methods:

• Developed a pipeline using the FTPCT software for 3D architectural trait extraction from point clouds.
• Identified key traits associated with leaf number using correlation analysis and VIF screening.
• Employed and optimized five machine learning models, including random forest regression, using Bayesian optimization.

Main Results:

• The optimized random forest regression model achieved high performance (R²=0.85, RMSE=239.74, MAE=149.26) for leaf number estimation.
• SHAP analysis revealed tree crown volume as the dominant predictor for leaf number.
• Field validation showed improved fruit weight and size with the proposed FL regulation strategy.

Conclusions:

• The proposed framework offers an efficient and accurate method for leaf number and FL estimation in pear trees.
• This approach reduces data requirements and computational costs, making it suitable for large-scale precision agriculture.
• The data-driven method significantly enhances fruit quality and size, providing economic benefits for pear growers.