Oak Ridge National Laboratory Collaboration (Spring 2025, Fall 2025, Spring 2026)
Students in the ETSU MS in Applied Data Science program have partnered with Oak Ridge National Laboratory since January 2025. This collaboration centered on two main projects using data collected through the Advanced Plant Phenotyping Laboratory (APPL) at Oak Ridge National Laboratory (https://www.ornl.gov/appl). This is a state-of-the-art imaging facility designed to automate and accelerate discoveries related to crop yield and domestic energy for the nation.
One collaborative project focused on semantic segmentation of unstructured 3D point clouds. This project is ongoing, but the goal is to automatically obtain geometric plant information from unstructured images of the plant. In this project, students created labeled data from the images where different portions on the plants were tagged as leaves, stem, stake or background. Using the labeled data, they then trained a neural network model to systematically identify the different portions of the plant from these annotated images. Students are now fusing together RGB imagery and binary segmentation masks with the 3D LiDAR point clouds in order to improve structural refinement and background removal. Once this is done satisfactorily, they plan to extract relevant information automatically from the images for analysis of plant growth.
A second project focused on tabular structured data which corresponded to the unstructured image data collected through APPL. This was rich data and thus a variety of research questions arose throughout this collaboration. The first semester, students focused on modeling and analyzing genotypic variation in plant growth dynamics. The second semester, students moved to predictive modeling. In one portion of the predictive modeling, students wanted to identify early spectral signals that could be used in predicting long-term plant performance. Another aspect of the predictive modeling focused on the ability to predict final biomass from early growth (the first two weeks of growth). Students are currently examining RGB data and if one can use RGB data early in the growth to determine whether the plants will ultimately be healthy. Additionally, analysis is ongoing about whether there are systematic days one might be able to use for imaging in order to predict the final area and height of the plant earlier in the growing cycle.
Final semester project presentations can be accessed below. .Furthermore, a GitHub repository was created for a portion of this research and can be found at the following link:
Barrera-Gomez, A. A., Jung, I., & Hussung, L. (2025). Plant organ segmentation from 3D LiDAR point clouds via geometric deep learning (unpublished manuscript). https://github.com/angomezu/geometric-deep-learning-plant-organ-segmentation.
