Your new experience awaits. Try the new design now and help us make it even better

ORIGINAL RESEARCH article

Front. Plant Sci.

Sec. Plant Breeding

Volume 16 - 2025 | doi: 10.3389/fpls.2025.1655350

Deep Learning Driven, Image-Based Phenotyping of Seed Processing Efficiency in Sainfoin (Onobrychis viciifolia)

Provisionally accepted
  • The Land Institute, Salina, United States

The final, formatted version of the article will be published soon.

Sainfoin (Onobrychis spp.) is a perennial legume traditionally cultivated as a forage crop and is now emerging as a promising candidate for development as a perennial grain legume. However, no research has focused on understanding and breeding varieties with superior grain processing properties. In this study, we conducted a multifactorial experiment to depod and dehull seeds from five commercially available sainfoin varieties using two different methods and five sample sizes. We then fine-tuned a pre-trained a Faster R-CNN (Region-based Convolutional Neural Network) object detection model to detect intact pods, whole seeds, and split seeds from images of the processed seed mixtures and used these predictions to calculate the processing efficiency (PE) for each variety. We performed a comprehensive power analysis to determine the minimum sample size of sainfoin fruit pods required to detect differences in PE between breeding lines with high statistical power. Our results revealed strong varietal differences in PE, as well as clear effects attributed to the processing method. Samples processed with a belt thresher contained more intact pods, whereas the impact dehuller produced a higher proportion of split seeds. Larger sample sizes led to more intact pods in the mixture across all varieties and methods, and notably decreased the seed proportion in belt-threshed samples. Combining statistical modeling with seed count data from the Faster R-CNN predictions, we found that a minimum of 2g of seed is needed to reliably detect an absolute proportional difference of 0.25 between two breeding lines' PE with 80% power. Our findings emphasize that while deep learning models provide a powerful and cost-effective tool for phenotyping in plant breeding, their outputs must also be integrated with robust statistical design and methodology to yield reliable and actionable insights.

Keywords: deep learning, perennial legume breeding, power analysis, Seed imaging, Small object detection

Received: 27 Jun 2025; Accepted: 05 Sep 2025.

Copyright: © 2025 Meyering, Barriball and Schlautman. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

* Correspondence:
Bo Meyering, The Land Institute, Salina, United States
Brandon Schlautman, The Land Institute, Salina, United States

Disclaimer: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.