A Deep Learning-Based Phenotypic Analysis of Rice Root Distribution from Field Images

Area of Science:

• Agricultural Science
• Plant Biology
• Computational Biology

Background:

• Root distribution is critical for plant water and nutrient uptake, directly impacting crop yield.
• Traditional methods for assessing root systems, like the trench profile method, are labor-intensive for quantification.
• Automated root segmentation is needed to efficiently analyze root architecture.

Purpose of the Study:

• To develop and validate a convolutional neural network (CNN) model for automated root segmentation in trench profile images.
• To establish quantitative parameters (Depth50, Width50) for root distribution analysis.
• To assess the phenotypic diversity of root distributions in rice accessions.

Main Methods:

• Utilized the trench profile method to capture underground root images.
• Developed and trained a convolutional neural network for segmenting root areas in the images.
• Defined and calculated root distribution parameters (Depth50, Width50) using the CNN model and manual tracing for validation.

Main Results:

• The CNN model achieved high correlation with manual tracing for root distribution parameters (Depth50, Width50) in rice.
• The developed method enables rapid and accurate quantification of root traits from trench profile images.
• Phenotypic diversity in root distribution was revealed across 60 rice accessions.

Conclusions:

• Convolutional neural networks offer a reliable and efficient tool for root phenotyping using the trench profile method.
• This approach significantly facilitates the study of crop root systems in field conditions.
• The findings support the use of AI in accelerating crop breeding and understanding plant adaptation.