Exploring the Diversity of Root System Architecture in Soybean using Plant Phenomics

Root system architecture (RSA) studies are tedious, susceptible to introduced variation and the extracted features may not translate to a meaningful outcome. With the advent of high-throughput phenotyping, computer vision and machine learning there is a renewed interest in uncovering “the hidden half”. Our study included 300 diverse soybean accessions from a wide geographical distribution (17 countries) of which genotypic information is available. We deployed a 2-D (in controlled conditions) and stereo imaging platforms (field tests), image processing algorithms and data analytic tools to deep phenotype for RSA traits using in-house software. The 2-D platform developed is non-destructive, adding observations throughout seedling growth and development. The stereo imaging platform of multiple cameras at multiple angles allows creation of a 3-D point cloud of a mature root.

Tens of thousands of images were collected from thousands of plants using the imaging platforms developed in this study. Moving forward, we are adapting machine learning techniques via convolutional neural networks will allow for the extraction and prediction of novel root architectural information. Utilizing phenotyping techniques has allowed us to capture tremendous RSA variability of these 300 diverse genotypes throughout various stages of development that will drive gene discovery and breeding methods forward.