Skip to main content
Please refrain from downloading and saving posters in the FHB Forum Virtual Poster room. Not all authors have agreed to share their posters. If you would like a copy of a poster please contact the Corresponding Author.

Related Posters

Poster # 509
Poster Title: An RGB Based Deep Neural Network Approach for Field-Based High Throughput Phenotyping of Fusarium Head Blight in Wheat Using Mobile Images
Authors: Julian Cooper 1, Chuan Du 2, Zach Beaver2, Ming Zheng 2, Rae Page 1, Joseph R. Wodarek 3, Oadi Matny 1, Tamas Szinyei 1, Alejandra Quiñones 1, James A. Anderson 4, Ce Yang 5, Brian J. Steffenson 1, and Cory D. Hirsch 1
1. Department of Plant Pathology, University of Minnesota, Saint Paul, MN 55108
2. Mineral Earth Sciences LLC., Mountain View, CA 94043
3. Northwest Research and Outreach Center, University of Minnesota, Crookston, MN 56716
4. Department of Agronomy and Plant Genetics, University of Minnesota, Saint Paul, MN, 55108
5. Department of Bioproducts and Biosystems Engineering, University of Minnesota, Saint Paul, MN 55108
Corresponding Author: Julian Cooper, coop0409@umn.edu
Presenting Author:   Julian Cooper
Poster Video:



Fusarium head blight (FHB) is a major disease of wheat that can cause yield losses exceeding 50%. Breeding for resistance remains the most effective control method, however, traditional scoring methods are labor-intensive and subjective. This study presents a deep learning pipeline for wheat spike detection and FHB quantification, trained using images collected by the Google Moonshot Mineral X phenotyping rover. The pipeline detects and segments spikes and diseased tissue to quantifying disease severity at a spike and plot scale. To validate this high throughput FHB phenotyping pipeline, disease inferences  from field based images were compared to disease scoring in the field and manual image analysis of diseased wheat spikes by raters. Disease assessments from the imaging pipeline correlated strongly with rater disease scores at the spike and plot level, and outperformed traditional methods in precision and throughput. Furthermore, the pipeline was found to be generalizable across years, environments, and disease progressions, which allows for reliable assessments under diverse conditions. To improve access to this phenotyping tool, the image analysis pipeline has been adapted to analyze mobile images at single row plot resolution. This new tool is an objective and accessible phenotyping method for the wheat breeding and research community interested in low cost and scalable FHB field phenotyping.