3D morphological trait analysis of rice plants using light field imaging

Overview

This research proposes a novel methodology for estimating morphological traits of rice plants, including leaf count and axial leaf angles, using 3D models reconstructed from light field images captured with a Raytrix 42 plenoptic camera. The approach enables automated, non-destructive plant phenotyping, replacing labor-intensive manual measurements with a digital imaging pipeline. By combining light field imaging with machine learning algorithms, the technology aims to deliver accurate, scalable trait estimation for rice cultivation research and broader agricultural applications.

Technical specifications

Key features:

• Light field imaging captures multiple perspectives in a single exposure, enabling depth reconstruction without complex multi-camera setups
• Raytrix 42 plenoptic camera provides high-resolution 3D data acquisition suitable for detailed plant architecture analysis
• 3D model reconstruction from light field images allows measurement of leaf angles and leaf count across growth stages
• Machine learning algorithms trained on ground-truth measurements predict leaf angles and lengths with quantitative accuracy metrics including counting error and root mean squared error (RMSE)
• Calibrated acquisition platform designed specifically for repeatable imaging of rice plants at different developmental stages

Technology readiness level

This research is at an early stage, with the hypothesis not yet validated. The team plans to build and calibrate the acquisition platform, collect training data across growth stages, train machine learning models, and evaluate predictions using established error metrics. Previous work by the team demonstrated 4D rice plant modeling from 2D images with 3% growth prediction error, providing a strong foundation for extending these capabilities to light field-based 3D reconstruction.