A real-time phenotyping framework using machine learning for plant stress severity rating in soybean

Area of Science:

• Plant science
• Agricultural technology
• Computational biology

Background:

• Phenotyping is crucial for plant research and crop improvement.
• Automated, high-throughput phenotyping is needed for large populations across diverse environments.
• Imaging technology, data analytics, and machine learning offer solutions for efficient phenotyping.

Purpose of the Study:

• To develop and illustrate an end-to-end phenotyping workflow for plant stress severity.
• To rapidly and automatically assess iron deficiency chlorosis (IDC) severity in soybean across thousands of field plots.
• To deploy a machine learning model as a smartphone app for real-time field assessment.

Main Methods:

• An integrated workflow involving image capture, data curation, trait extraction, and machine learning classification was designed.
• Approximately 4500 soybean canopies with varying IDC levels were analyzed.
• Multiple classification models were trained, and the best performing model was selected for app deployment.

Main Results:

• A hierarchical classifier achieved a mean per-class accuracy of approximately 96% for IDC severity prediction.
• A 'population canopy graph' was constructed, linking extracted canopy traits to stress severity.
• The workflow was integrated into a smartphone app for automated, real-time IDC scoring via digital canopy images.

Conclusions:

• The high-throughput framework enhances genetic gain by providing a robust system for other abiotic and biotic stresses.
• The workflow can be integrated with ground vehicles and unmanned aerial systems for automated stress detection and quantification.
• This technology supports plant research, breeding programs, and stress scouting applications.