From Organelle Morphology to Whole-Plant Phenotyping: A Phenotypic Detection Method Based on Deep Learning

Area of Science:

• Plant Biology
• Computational Biology
• Genetics

Background:

• Plant phenotype analysis is crucial for crop breeding and understanding plant development.
• Accurate and efficient phenotyping methods are needed to accelerate breeding programs.

Purpose of the Study:

• To apply deep learning for classifying *Arabidopsis thaliana* at macro (whole plant) and micro (organelle) levels.
• To develop a multi-output model for identifying accession lines and predicting growth status.
• To assess the generalizability of the deep learning model across different time intervals and biological scales.

Main Methods:

• Utilized a deep learning multi-output model for image classification and regression tasks.
• Trained and validated the model on *Arabidopsis thaliana* images at various growth stages.
• Expanded the dataset by varying image time intervals and applied the model to organelle classification.

Main Results:

• Achieved high accuracy (99.92%) in identifying *Arabidopsis thaliana* accession lines.
• Demonstrated good performance in predicting plant growth status with a root mean square error (RMSE) of 1.536.
• Confirmed model robustness across different time intervals and its generalizability to organelle classification.

Conclusions:

• Deep learning offers a powerful approach to broaden plant phenotype detection methods.
• The developed model can significantly facilitate the design of high-throughput plant phenotyping platforms.
• This research paves the way for accelerated crop improvement through advanced computational phenotyping.