High Throughput Phenotyping for Essential Crop Traits

Michael McGuire, Thomas Aref, Arif Syraj, Shubham Shukla, Akshay Kumar, Stephen Moose, Brian Diers

Keywords: Agricultural robotics, High-throughput field phenotyping, Yield prediction, Precision agriculture, Autonomy

TL;DR: We show positive results on predicting several key crop breeding phenotypes with AI applied to crop data collected with autonomous robots.

Abstract: We report promising results for high-throughput estimation of several key phenotypic traits through the use of small mobile robots and machine-learning based machine-vision algorithms. Our autonomous robotic data collection system, data association pipeline, and analytics algorithms can provide accurate estimation of Stem Width, Stand Count, Ear Height, and Plant Height traits for corn, as well as the Pod Count trait for soybeans. Collecting data for these phenotypes is manually extremely labor intensive, and has been difficult to automate. Our results are significant because they show that autonomous robots equipped with multiple relatively low-cost RGB-vision, RGB-Depth, LiDAR, inertial, and GPS sensors can significantly increase throughput in phenotyping tasks across corn and soybean at a variety of growth stages.