Skip to main content
ARS Home » Southeast Area » Stuttgart, Arkansas » Dale Bumpers National Rice Research Center » Research » Publications at this Location » Publication #421947
Research Project: Broadening and Strengthening the Genetic Base of Rice for Adaptation to a Changing Climate, Crop Production Systems, and Markets

Location: Dale Bumpers National Rice Research Center

Title: Plant biomass estimations with minimal equipment using deep learning-based image segmentation

Author
item MITCHELL, JOHN - University Of Arkansas At Pine Bluff
item Sookaserm, Tiffany
item Rohila, Jai
item Huggins, Trevis
item PONNIAH, SATHISH - University Of Arkansas At Pine Bluff
item Edwards, Jeremy

Submitted to: Rice Technical Working Group Meeting Proceedings
Publication Type: Proceedings
Publication Acceptance Date: 12/13/2024
Publication Date: 1/5/2026
Citation: Mitchell, J., Sookaserm, T.B., Rohila, J.S., Huggins, T.D., Ponniah, S., Edwards, J. 2026. Plant biomass estimations with minimal equipment using deep learning-based image segmentation. Rice Technical Working Group Meeting Proceedings. New Orleans, Louisiana. February 17-20, 2025.

Interpretive Summary:

Technical Abstract: Accurate biomass measurement is essential for crop improvement and breeding efforts, yet traditional methods are destructive, labor-intensive, and unsuitable for continuous monitoring across plant growth stages. This study introduces an accessible, non-invasive approach for estimating above-ground biomass in rice plants, leveraging digital imaging and deep learning. Using convolutional neural networks (CNNs), side-view RGB images of rice plants were analyzed to estimate shoot biomass. The experiment involved a diverse set of rice genotypes cultivated in a greenhouse, with RGB images taken against a uniform background using a standard digital camera and a reference object for scale calibration, eliminating the need for specialized equipment. Image segmentation was performed with the R package “imageseg,” which calculated leaf area as a proxy for biomass. Results revealed strong consistency between repeated image-derived leaf area estimates and a robust correlation (R² = 0.763) between estimated leaf area and measured dry biomass. The approach proved reliable under ambient lighting and with variable plant orientations. This affordable CNN-based method offers a practical solution for high-throughput phenotyping, enabling efficient and accurate biomass estimation in rice and other crops under realistic conditions, while requiring only minimal resources and equipment.