Objective phenotyping root system architecture using image augmentation and machine learning in alfalfa (Medicago sativa L.)

Authors: Xu, Zhanyou; YORK, LARRY - Oak Ridge National Laboratory; SEETHEPALLI, ANAND - Noble Research Institute; BUCCIARELLI, BRUNA - University Of Minnesota; CHENG, HAO - University Of California, Davis; Samac, Deborah - Debby

Submitted to: Plant Phenomics
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 3/7/2022
Publication Date: 4/7/2022
Citation: Xu, Z., York, L.M., Seethepalli, A., Bucciarelli, B., Cheng, H., Samac, D.A. 2022. Objective phenotyping root system architecture using image augmentation and machine learning in alfalfa (Medicago sativa L.). Plant Phenomics. 2022. Article 9879610. https://doi.org/10.34133/2022/9879610.
DOI: https://doi.org/10.34133/2022/9879610

Interpretive Summary: Roots are important for securing plants in soil and for nutrient and water uptake. They also serve important roles for perennial plants in winter survival for storing carbohydrates. In alfalfa, a perennial legume, selecting plants for specific root architecture traits has increased herbage yields; however, selecting plants with a desired root phenotype is challenging since scoring has relied on subjective classification and is prone to human error and bias. Roots from alfalfa plants selected for increased or decreased branching were excavated from field plots, digitally photographed, and the images processed to identify 38 specific root traits. Four different machine learning models and two deep learning models were tested to determine the model that most accurately predicted the root type. A prediction accuracy of 97% was achieved with image augmentation and random forest machine learning. This automated phenotyping method will give plant breeders greater confidence for advancing the best lines in their breeding programs to improve climate-resilient alfalfa plants with improved environmental services while retaining biomass yields.

Technical Abstract: Several protocols have been developed for high-throughput phenotyping of root system architecture (RSA) for annual crops. However, previous methods for phenotyping RSA of perennial plants used visual scoring as ordinal data or subjectively classified plants into different root types. This research aims to develop and compare objective RSA phenotyping methods using machine and deep learning algorithms with 617 root images from mature alfalfa plants collected from the field. Our results show that Random Forest and TensorFlow-based neural networks can classify the root types into branch-type (B), taproot-type (T), and an intermediate Taproot-Branch (TB) type with 86% accuracy. With image augmentation technology, the prediction accuracy was improved to 97%. Coupling the predicted root type with its prediction probability will give breeders a greater confidence level for decisions to advance the best and exclude the worst lines from their breeding program. This machine and deep learning approach enables accurate prediction of RSA phenotypes for genomic breeding of climate-resilient alfalfa.