Automated pipeline for leaf spot severity scoring in peanuts using segmentation neural networks.

Area of Science:

• Plant Pathology
• Agricultural Science
• Computer Vision

Background:

• Peanut leaf spot diseases cause significant global yield losses.
• Current phenotyping relies on subjective human scoring, limiting accuracy and scalability.
• Developing objective disease assessment tools is crucial for breeding resistant peanut varieties.

Purpose of the Study:

• To develop an objective, end-to-end image analysis pipeline for scoring peanut leaf spot severity.
• To replace subjective human rating scales with automated, quantitative metrics.
• To facilitate efficient phenotyping in peanut breeding programs.

Main Methods:

• Utilized image capture protocols and segmentation neural networks for lesion area extraction.
• Developed algorithms to convert image data into quantitative quality metrics.
• Trained and evaluated the pipeline on a large dataset of field images with varying disease severity.

Main Results:

• The pipeline accurately determined infected leaf surface area and identified dead leaves from cellphone imagery.
• Automated scoring achieved a root mean square error of 0.996 (single image) and 0.800 (three images) compared to expert visual scores.
• Demonstrated the pipeline's capability to provide objective, quantitative disease severity ratings.

Conclusions:

• The developed image processing pipeline serves as a viable alternative to subjective human scoring.
• Automated scoring reduces subjectivity, enabling non-experts to collect data and potentially facilitating drone-based assessments.
• This technology can accelerate the identification of new peanut lines and genes for improved leaf spot resistance.