Fine-tuning improves deep learning-based digital phenotyping of lettuce genetic resources
- 1Department of Agricultural Engineering, National Institute of Agricultural Sciences, Rural Development Administration
- 2National Agrobiodiversity Center, National Institute of Agricultural Sciences, Rural Development Administration
- 3Agricultural AI and Data Division, Rural Development Administration
Abstract
Introduction: Morphological characterization of agricultural genetic resources relies largely on visual assessment by experts, and applying the same criteria repeatedly to large collections is labor-intensive. Image-based deep learning can support this process, but systematic evidence on training strategies suited to genebank trait data remains limited.
Methods: This study targeted six morphological traits of lettuce genetic resources characterized according to the test guidelines of the International Union for the Protection of New Varieties of Plants (UPOV). Using genebank images accumulated from 2012 to 2025, six convolutional neural networks were trained with three training strategies under identical conditions, and their performance was compared.
Results: Fine-tuning achieved the highest mean macro F1-score in 35 of the 36 backbone–trait combinations and significantly outperformed training from scratch and feature extraction, whereas no single backbone was superior across all traits. Misclassifications were concentrated in morphologically adjacent classes rather than randomly distributed, and for the ordinal traits, errors spanning two or more class levels were rare.
Discussion: These patterns suggest that the remaining classification difficulty was associated in part with the morphological continuity between neighboring classes. The attainable performance also differed among traits, distinguishing traits for which the training strategy alone achieved relatively high performance from those for which it did not. These results provide empirical evidence for selecting training strategies and for determining the role of model-assisted characterization on a trait-by-trait basis in genebanks.
Related
- Plant–Atmosphere Interactions in Urban Vineyards of Rome: A Digital Phenotyping Approach for the Functional Selection of Grapevine Cultivars
- Harnessing artificial intelligence in plant breeding: innovations in digital phenotyping and breeding methodologies
- Stress-Symptom Interaction in Inflammatory Bowel Disease: Longitudinal Digital Phenotyping Study.
- Aerial Phenotyping of Coffee Flowering Using Deep Learning for Early Yield Prediction
- Multi-temporal UAV multispectral phenotyping and machine learning for modeling cotton defoliation dynamics across diverse accessions
BibTeX
@article{lee2026finetuning,
title = {{Fine-tuning improves deep learning-based digital phenotyping of lettuce genetic resources}},
author = {Eunju Lee and Yuseok Jeong and Hyo Jun Bae and Yoon-Jung Lee and Seong-Hoon Kim and Young-Wang Na and Hwan-suk Cheong and Inchan Choi},
journal = {Frontiers in Plant Science},
year = {2026},
month = oct,
doi = {10.3389/fpls.2026.1970840},
url = {https://www.frontiersin.org/articles/10.3389/fpls.2026.1970840/full},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Fine-tuning improves deep learning-based digital phenotyping of lettuce genetic resources
AU - Eunju Lee
AU - Yuseok Jeong
AU - Hyo Jun Bae
AU - Yoon-Jung Lee
AU - Seong-Hoon Kim
AU - Young-Wang Na
AU - Hwan-suk Cheong
AU - Inchan Choi
PY - 2026
DA - 2026/10/09
JO - Frontiers in Plant Science
DO - 10.3389/fpls.2026.1970840
UR - https://www.frontiersin.org/articles/10.3389/fpls.2026.1970840/full
AB - Introduction: Morphological characterization of agricultural genetic resources relies largely on visual assessment by experts, and applying the same criteria repeatedly to large collections is labor-intensive. Image-based deep learning can support this process, but systematic evidence on training strategies suited to genebank trait data remains limited. Methods: This study targeted six morphological traits of lettuce genetic resources characterized according to the test guidelines of the International Union for the Protection of New Varieties of Plants (UPOV). Using genebank images accumulated from 2012 to 2025, six convolutional neural networks were trained with three training strategies under identical conditions, and their performance was compared. Results: Fine-tuning achieved the highest mean macro F1-score in 35 of the 36 backbone–trait combinations and significantly outperformed training from scratch and feature extraction, whereas no single backbone was superior across all traits. Misclassifications were concentrated in morphologically adjacent classes rather than randomly distributed, and for the ordinal traits, errors spanning two or more class levels were rare. Discussion: These patterns suggest that the remaining classification difficulty was associated in part with the morphological continuity between neighboring classes. The attainable performance also differed among traits, distinguishing traits for which the training strategy alone achieved relatively high performance from those for which it did not. These results provide empirical evidence for selecting training strategies and for determining the role of model-assisted characterization on a trait-by-trait basis in genebanks.
SN - 1664-462X
ER -