3D Point Cloud Phenotyping of Soybean Growth in Early Stage

Phenotyping is important in breeding science even if molecular breeding strategies are applied. For example, phenotypic data are required for training a prediction model in genomic selection. Point cloud data can help us to evaluate morphological data cost-effectively. In this study, we developed a workflow for measuring the canopy growth of three soybean cultivars to compare growth speed of canopy based on point cloud acquired from a Structure from Motion (SfM) and Multi-View Stereo (MVS) pipeline, which is a technique to reconstruct a 3D surface as point cloud from a series of 2D images taken from different angles. First, we took multi-view 2D images of soybean plants growing in a field with a digital camera (EOS 60D, Canon). The images of plants in a single plot were taken from ca. 40 different directions. To scale the images and to calibrate camera parameters, we used Calibrated Photogrammetric Scale Bars (Cultural Heritage Imaging). Second, we segmented the point cloud of plants and their leaves for each plot from the reconstructed point cloud. Finally, we fitted several models to the point cloud for estimating phenotypic values of plant organs constituting canopy architecture (e.g. leaf area, leaf shape, curvature). For example, we reconstructed 3D surfaces of leaves from the point cloud with the penalized B-spline surface fitting. Using the workflow, we successfully acquired the high-resolution point cloud data of canopies of soybean plants growing in the field. Even when leaves slightly overlapped each other, a single leaf was successfully distinguished from the point cloud.