PhenoBot: an automated system for leaf area analysis using deep learning.

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Phenotyping

Background:

• Crop yield is significantly influenced by intercepted radiation, making Leaf Area Index (LAI) a crucial parameter for yield estimation.
• Accurate LAI determination is essential for crop growth models (CGMs) and understanding phenological processes.
• Existing leaf area estimation methods often lack dynamism and struggle with leaf variations.

Purpose of the Study:

• To develop a low-cost, dynamic image capturing and analysis pipeline for direct leaf area estimation.
• To improve the accuracy and throughput of leaf area calculations in commercial agricultural settings.
• To create a versatile system applicable to multiple crop types.

Main Methods:

• Developed a low-cost image acquisition setup and an automated data management pipeline.
• Utilized virtual machines running custom-trained deep learning models for leaf segmentation.
• Implemented color-based deep learning segmentation for precise leaf area calculation at whole-set and individual levels.
• Integrated open-source hardware, platforms, and algorithms for affordability and reproducibility.

Main Results:

• Achieved high accuracy in leaf area estimation for multiple crop types.
• The system dynamically segments leaves, calculates leaf area, and overlays information.
• Data is stored efficiently on-device and in the cloud.
• Generated compatible files for direct use in crop growth models.

Conclusions:

• A cost-effective, dynamic deep learning pipeline enables accurate leaf area estimation in commercial environments.
• The developed system enhances crop phenotyping and supports precise yield simulation.
• The use of open-source components ensures the system's affordability and reproducibility across different research settings.