We develop machine learning methods and computer vision algorithms to extract information from phenotyping data collected by different sensor platforms. For field scale phenotyping, we analyze data collected using RGB, multispectral cameras and 3D-LiDAR. We also analyze 2D and 3D imaging data collected in the lab to understand plant shoot architecture.
Canopy cover Fast canopy expansion can help suppress weed and improve photosynthetic efficiency. Using drones with RGB and multispectral cameras, we collect field scale data for soybean and edamame and to generate growth curves of these plants. Genome wide association study will be performed to determine genetic variations that are associated with fast canopy expansion. We also use 3D-LiDAR to study plant canopy height and ground cover.
Shoot architecture We collect imaging data for edamame plants and we use computer vision algorithms to automatically detect pod locations on the plant. Understand the distribution of pod locations on a plant can help determine the factors that affect harvest efficiency. We are also interested in identify genetic control of plant shoot architecture traits. Using persistent homology, we have identified hidden correlation between plant shoot structure and pod locations.
Corn project Using object detection algorithms, we are improving the speed and accuracy of drone-based plant stand counting. The result of this project can help producers make better management decisions based on early season phenotyping data collected by cheap, consumer grade UAV.