Vision-Driven RhizoNet: Foundations for Systematic Measurement of Plant Root Biomass
Authors:
Daniela Ushizima1,2,3*, Zineb Sordo1, Peter Andeer1, Marcus Noack1, James Sethian1,2, Trent Northen1, Susannah Tringe1
Institutions:
1Lawrence Berkeley National Laboratory; 2University of California–Berkeley; 3University of California–San Francisco
Goals
The goal of Center for Restoration of Soil Carbon by Precision Biological Strategies (RESTOR-C) is to harness plants and microbes to increase carbon flux into soil carbon storage pools to form persistent carbon that is stable for >100 years. This will address the Carbon Negative Shot goal to remove carbon dioxide (CO2) from the atmosphere and durably store it at meaningful scales for less than $100 per net metric ton of CO2-equivalent within a decade.
Abstract
Improving root traits is an important research area for soil carbon sequestration. To accelerate this research, we have developed the EcoBOT, an innovative robotic system designed for plant imaging including growth and monitoring of plants in specialized devices called EcoFABs that enable detailed root scans. The EcoBOT allied to EcoFABs can generate many hundreds of root scans each week and so automated computer vision tools based on machine learning are needed to rapidly process the data. To address this challenge, we are building RhizoNet, a deep learning–based workflow tailored for precise semantic segmentation of plant root imagery, including modules for analysis and measurements of root biomass growth. RhizoNet overcomes many challenges by employing a sophisticated Residual U-Net architecture that significantly improves prediction accuracy. This is complemented by a convex hull operation aimed at precisely delineating the primary root component over time, thus facilitating a more accurate assessment of root biomass and its growth. Its robust root detection model has demonstrated generalization capabilities across a wide range of experimental conditions, underscoring its utility in standardizing and objectifying the analysis of thousands of root images. By integrating RhizoNet into EcoBOT’s operational framework, the process of acquiring and analyzing root scans can be significantly streamlined, reducing the need for manual intervention, and thereby increasing throughput and accuracy in root growth studies. This automation is crucial for real-time monitoring and autonomous decision- making. Furthermore, the application of RhizoNet to plant root analysis highlights the broader implications of semantic segmentation technologies fields to optimize plant growth, enhance crop yields, and contribute to sustainable agricultural practices.
Funding Information
This research is supported by the U.S. DOE, Office of Science, BER program, and Advanced Scientific Computing Research program under contract number DE-AC02-05CH11231.