Estimating Radicle Length of Germinating Elm Seeds via Deep Learning

Area of Science:

• Plant biology and computational imaging.
• Development of AI-driven solutions for agricultural science.

Background:

• Accurate seedling trait measurement is vital for plant phenotyping, especially for species like elm (Ulmus spp.) with challenging curved morphologies.
• Existing manual and automated methods struggle with nonlinear seedling shapes, leading to inaccuracies and inefficiencies in growth and stress response studies.

Purpose of the Study:

• To introduce GLEN, a deep learning model for precise detection and length estimation of germinating elm seeds, particularly their curved radicles.
• To develop a robust and scalable framework for automated morphological quantification in plant science.

Main Methods:

• Development of GLEN, a deep learning model with a dual-path architecture integrating spatial and semantic features for curved radicle measurement.
• Creation of GermElmData, a novel annotated dataset, and a synthetic data generation pipeline to enhance model training and generalization.

Main Results:

• GLEN achieved millimeter-level estimation error, significantly outperforming existing automated measurement models for elm seedlings.
• The model demonstrated robust performance on curved radicle morphologies, a common challenge in plant phenotyping.

Conclusions:

• GLEN offers a highly accurate and efficient solution for measuring germinating elm seeds, advancing plant phenotyping capabilities.
• The model's architecture and data strategies provide a scalable framework applicable to broader biomedical and agricultural imaging challenges.