A photometric stereo-based 3D imaging system using computer vision and deep learning for tracking plant growth

Area of Science:

• Plant Science
• Computer Vision
• Genetics

Background:

• Accurate plant growth tracking is vital for predicting climate change impacts.
• Automated phenotyping significantly increases data acquisition efficiency over manual methods.
• Computer vision and machine learning enhance automated plant phenotyping accuracy.

Purpose of the Study:

• Introduce PS-Plant, a novel, low-cost, portable 3D plant phenotyping platform.
• Utilize photometric stereo (PS) imaging for advanced plant growth analysis.
• Develop and validate automated algorithms for plant trait extraction.

Main Methods:

• Calibrated PS-Plant to monitor Arabidopsis thaliana growth over diel cycles.
• Developed custom computer vision algorithms and evaluated deep neural networks for image analysis.
• Created the first PS training dataset with 221 annotated Arabidopsis rosettes (1,768 images).

Main Results:

• Demonstrated PS-Plant's ability to track 3D plant growth and diel leaf movements.
• Successfully automated segmentation of rosettes and leaves for trait extraction.
• Illustrated the significant impact of environmental conditions on plant phenotype.

Conclusions:

• PS-Plant offers a powerful tool for high-resolution plant phenotyping.
• The system supports both small- and large-scale research initiatives.
• PS-Plant accelerates the bridging of the phenotype-to-genotype gap in plant research.