How's it growing? Estimating growth from depth images.

Neural networks can be pretty good at estimating the weight of a plant from a single top-down image. "Can be", because it only works well with small plants that have a simple, open structure. Complex plants, or plants in a dense canopy, are too occluded for an accurate estimate from a single image. Luckily, there might be a simple solution to this problem: more images. Visually tracking plants in 3D as they grow might provide a well-designed neural network with the additional information it needs to properly estimate a crop's growth even in highly occluded situations. Or it might prove that simple, classical methods are good enough. This project aims to create a new benchmark dataset and a matching growth estimation model, and test it against several baselines. Tasks involve data collection, processing, and training a weight estimation model. Do you want to help create a new state-of-the-art in automated crop growth estimation? Join the smartfarming project.

Used skills

• Data analysis
• (Basic) Python
• (Optional) Designing neural networks

Interested in doing a BSc or MSc thesis at HPP? Please contact Katharina Hanika or Kim Vanderwolk via the HPP office (office.hpp@wur.nl).