Files
Abstract
Roots play a central role in how plants acquire water and nutrients and are therefore critical for plant growth and stress tolerance. Because of their importance, root traits are increasingly targeted in breeding programs. However, their measurement remains difficult due to their complex, three-dimensional branching structures. Capturing root traits therefore requires accurate image data and robust computational methods. Although imaging methods such as microscopy and Structure-from-Motion (SfM) have improved over the past decades, the resulting measurements often contain structural ambiguities that complicate trait extraction. This can be seen in toot hairs that overlap in 2D images, and in 3D reconstructions, where roots may cross and form intersections that are not part of the actual root architecture. These artifacts affect how roots are represented digitally and distort measurements of traits such as root hair length, lateral root density, or branching hierarchy.In this dissertation I address this problem by developing algorithms that resolve intersections in 2D and 3D root structures. I present a method that identifies and separates intersecting root hairs in microscopy images, allowing the measurement of individual hairs across different species. Further, I optimize camera calibration in a 3D scanner to improve the quality of root
reconstructions based on Structure from Motion, resulting in higher quality point clouds that can ultimately lead to better skeletonization and more reliable trait measurements. Finally, I develop an algorithm to resolve intersecting roots in 3D skeletons of root systems. Many root systems contain cycles that arise when branches cross in the point cloud. These cycles are removed when individual roots are detected, and the skeleton is restructured into a tree, allowing traits such as lateral root length and topology to be measured more accurately.
Together, these projects improve the accuracy of root phenotyping across different spatial scales. The algorithms I developed support measuring actual traits rather than indirect proxies or aggregated traits. Therefore, they contribute to a better understanding of root system architecture and enable the use of detailed root traits in studies of plant function and in breeding for more resilient crops.