Cotton field population phenotyping analysis based on 3D Gaussian reconstruction and dynamic spatial constraints

Area of Science:

• Agricultural Science
• Computer Vision
• Genetics

Background:

• High-throughput field phenotyping (HTFP) is crucial for understanding plant genetics.
• Challenges exist in acquiring 3D data and performing single-plant analysis in field settings.

Purpose of the Study:

• To develop an integrated framework for field crop reconstruction and phenotypic analysis.
• To enable accurate, efficient extraction of individual plant traits from 3D point cloud data.

Main Methods:

• Utilized 3D Gaussian splatting for field-scale cotton population modeling and 3D point cloud generation.
• Developed a geometry-aware dynamic constraint algorithm for instance segmentation.
• Proposed a crop localization domain for longitudinal phenotypic mapping.

Main Results:

• PhenotypeAI reconstructed nine cotton populations with high fidelity (PSNR > 30.0 dB).
• Achieved high accuracy in instance segmentation (91.32% F-score) and trait extraction (91.35%).
• Extracted traits (height, leaf area) showed strong correlation (R² ≈ 0.91) with manual measurements.

Conclusions:

• The proposed framework offers a low-cost solution for HTFP in cotton.
• This method significantly enhances the efficiency of cotton breeding programs.