Open RGB imaging workflow for morphological and morphometric analysis of fruits using deep learning: a case study on...

Area of Science:

• Plant Science
• Agricultural Technology
• Bioinformatics

Background:

• High-throughput phenotyping is crucial for overcoming bottlenecks in plant breeding.
• Imaging tools and Artificial Intelligence (AI) are revolutionizing phenotyping efficiency and data generation for genomic selection.
• Current methods require accessible, AI-enhanced imaging solutions for broader breeding program adoption.

Purpose of the Study:

• To develop an open Python workflow utilizing AI for analyzing plant organ morphology, color, and morphometric traits.
• To apply and validate this workflow in almond (Prunus dulcis) breeding, a species with a long breeding cycle.
• To assess the workflow's performance, accuracy, and scalability for real-world phenotyping applications.

Main Methods:

• Development of an open Python workflow integrating AI for image analysis.
• Application of the workflow to phenotyped almond kernels, nuts, and individuals (over 25,000 kernels, 20,000 nuts, 600 individuals).
• Evaluation of segmentation and reconstruction accuracy, and estimation of traits like kernel thickness using weight and area variables.

Main Results:

• The workflow achieved high accuracy with error rates below 1% for segmentation and reconstruction.
• Accurate estimation of kernel thickness with a Root Mean Squared Error (RMSE) of 0.47.
• Identification of 55 heritable morphological, morphometric, and color traits in almond, suitable for breeding programs.

Conclusions:

• The AI-driven workflow demonstrates robust performance across diverse datasets, even with limited training data.
• Compatibility with AI-based labeling tools reduces manual effort and accelerates dataset preparation.
• The workflow enhances scalability and practical applicability for breeding programs, streamlining segmentation model fine-tuning.