HopBox: An image analysis pipeline to characterize hop cone morphology

Authors: Altendorf, Kayla; Heineck, Garett; WAKHOLI, COLLINS - Oak Ridge Institute For Science And Education (ORISE); Tawril, Anna; RAJA, PRANAV - University Of California, Davis; Rippner, Devin

Submitted to: The Plant Phenome Journal
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 7/17/2023
Publication Date: 8/22/2023
Citation: Altendorf, K.R., Heineck, G.C., Wakholi, C., Tawril, A., Raja, P., Rippner, D.A. 2023. HopBox: An image analysis pipeline to characterize hop cone morphology. The Plant Phenome Journal. 6(1). Article e20080. https://doi.org/10.1002/ppj2.20080.
DOI: https://doi.org/10.1002/ppj2.20080

Interpretive Summary: Hop cones are the female flowers from hop plants that serve as the primary bittering agent for brewing beer. In hop breeding and research programs, it is customary to evaluate hop cone morphological traits, such as shape, size, density, and color, because these characteristics can influence picking and drying ability, and can be indicative of quality as various biotic and abiotic stresses can cause discoloration and stunting. Collecting this data is tedious, however, as hop plants produce thousands of cones and the variation within a plant can be extensive. To automate this process and increase consistency across research studies, we developed an easily constructed light box and camera system along with a publicly available image analysis pipeline which we call the HopBox, that can quantify hop cone length, area, width, perimeter, openness, weight, color, and density. To test the HopBox, we imaged 500 cones each from fifteen experimental lines. We were able to detect significant differences between hop lines for selection purposes. Using subsampling from the full dataset, we determined that only 5-10 cones were needed from each experimental line to represent the variation within the entire sample of 500 cones. The HopBox has broad utility for hop breeding and hop research programs.

Technical Abstract: Hop cone morphology can influence picking and drying ability, and color can impact consumer preference and may be indicative of quality. However, these characteristics are not generally evaluated in hop breeding programs due to the tedious nature of trait quantification and the extensive variation among cones within a genotype. We developed the HopBox, which is a simply constructed light box with a camera mount, and a publicly available image processing pipeline that identifies hop cones within color-corrected images, reads a QR code within the image, and outputs data on hop cone length, width, area, perimeter, openness, weight, color, and density. The trained model was applied to images of 500 cones each from fifteen replicated advanced hop genotypes from the USDA-ARS breeding program in Prosser, WA. Analysis of variance revealed significant (P < 0.001) differences between genotypes for all traits measured, enabling breeders to discriminate between genotypes for selection purposes. Broad sense heritability for all traits ranged from 0.23-0.59. A random sampling of hop cones from the complete dataset revealed that imaging only 5-10 cones adequately captured genotypic variation and provided acceptable rank correlations (rs > 0.75), however increasing the sample size to 30 provided optimal precision. Instructions for constructing a HopBox and the code for the analysis pipeline are publicly available online and have wide applicability for hop breeding and research.