Identifying particulate organic matter in X-ray Computed Tomography images of soil aggregates with a deep learning approach.
EMSL Project ID
51847
Abstract
X-ray computed tomography (X-ray CT) allows for the visualization of the interior of objects that are opaque to visible light. Historically, X-ray CT has been widely used to identify pores and solids in porous media based on the degree of X-ray absorption by different materials. However, in heterogenous porous media like soil aggregates, containing particulate organic matter, only mineral solids strongly absorb x-rays. X-ray absorption by POM is size and composition dependent making it difficult to distinguish from both mineral solids and air space. We aim to circumvent these identification issues by training networks to identify POM in X-ray CT images of soil aggregates. Positional information for POM and soil minerals including quartz, phyllosilicates, and Fe/Al oxides in the aggregates will be determined and confirmed by combining X-ray CT analysis with Fourier transform infrared microscopy (FTIR microscopy) and scanning electron microscopy and energy dispersive X-ray spectroscopy (SEM-EDX) analysis . Maps of POM generated by FTIR microscopy, SEM-EDX, and corresponding X-ray CT images will be used to train a Fully Convolutional Network (FCN) to rapidly identify POM and different minerals in X-ray CT images of soil aggregates. Using the segmentation results from the FCN we can reconstruct aggregates in 3d and can quantify the size, surface chemistry, and tortuosity of the pores connecting POM to the aggregate surface. This will allow us to model oxygen and solute diffusion to the surface of POM in soil aggregates, processes that control POM mineralization. Additionally, after initial training and use on known aggregates, we will apply the neural network to images of soil aggregates it was not trained on to evaluate its predictive performance for generalized use with tomographic images of soil aggregates.
Project Details
Project type
Large-Scale EMSL Research
Start Date
2021-10-01
End Date
2023-12-31
Status
Closed
Released Data Link
Team
Principal Investigator
Co-Investigator(s)
Team Members