ASABE Technical Library - Abstract

Abstract. In genomic research and breeding programs, a major bottleneck is the laborious process of measuring phenotypic traits, motivating research into high-throughput phenotyping systems. Previous studies have demonstrated that point-based sensors can measure plant height under field conditions, but the lack of spatial information limits the potential of extracting geometric traits such as leaf area index. The goal of this paper was to develop and evaluate an imaging-based high-throughput phenotyping system to measure cotton plant heights under field conditions. A high-clearance tractor equipped with a real time kinematic GPS was used as the system base. A Microsoft Kinect v2 camera was installed inside a partially-covered metal frame that was attached to the back end of the tractor. The camera was oriented towards nadir and controlled by a computer program that was developed using LabVIEW. The system collected depth images of 18 plots of cotton plants that were planted in 2015 in Watkinsville, GA on three dates (28 August, 9 October, and 16 November). The acquired depth images were segregated and stitched to form depth images for each plot. The stitched depth images were used to measure the maximum and average heights of each plot by using a peak-finder-based algorithm. The accuracy of the system was evaluated on the measurements acquired on 16 November. The results showed that the system‘s measurements were strongly correlated (R²=0.88-0.95) with manual measurements, with relative errors of less than 9%, indicating that the imaging-based HTP system could be an effective and accurate approach for field-based plant phenotyping.