U-Net segmentation of 3D micro-CT images of rooted soils using label data from multi-physics simulators
- Project site
- Team Nabla's repository (private access for eScience members only)
The health of soil is equally important as the health of humans; soil has to be strong enough to sustain the built infrastructures while being functional to allow water and nutrient transport to plants. While image processing has been a great help to support medical doctors in diagnosis, the use of images is not so straightforward for soil characterization. Difficulties arise simply because soil is random in nature (it contains e.g., solid grains, water, and plant roots) and has living organisms (e.g., plants) constantly modifying its structure. The project UNSAT tries to improve the identification capability of 3D image processing by training machine learning classifiers with high-fidelity, physics-based simulation data. We are interested in how new machine-learning techniques can help us to observe water transport and how root and soil react to water cycles during the growth of young maize.
- References
- 3D U-NET Learning dense volumetric segmentation from sparse annotation link.
- Trainable Weka Segmentation: a machine learning tool for microscopy pixel classification link.
- Hydro-micromechanical modeling of wave propagation in saturated granular crystals link.
- Micro-scale investigation of unsaturated sand in mini-triaxal shearing using X-ray CT link.
- Investigating the effect of porosity on the soil water retention curve using the multiphase Lattice Boltzmann Method link.
- Root-soil interaction. Floriana Anselmucci's thesis: Root-soil interaction: effects on soil microstructure.
- Traditional image processing vs deep learning link
- Software
- Datasets
- Learning materials
Project logo: Soil icons created by Freepik - Flaticon