Use of deep learning for structural analysis of computer tomography images of soil samples

Abstract

Soil samples from several European countries were scanned using medical computer tomography (CT) device and are now available as CT images. The analysis of these samples was carried out using deep learning methods. For this purpose, a VGG16 network was trained with the CT images (X). For the annotation (y) a new method for automated annotation, ‘surrogate’ learning, was introduced. The generated neural networks (NNs) were subjected to a detailed analysis. Among other things, transfer learning was used to check whether the NN can also be trained to other y-values. Visually, the NN was verified using a gradient-based class activation mapping (grad-CAM) algorithm. These analyses showed that the NN was able to generalize, i.e. to capture the spatial structure of the soil sample. Possible applications of the models are discussed.

Grassland management intensity determines root development, soil structure, and their interrelationship: Results of a regional study of Leptosols in the Swabian Alps

“…These relationships make µCT an objective and standardizable method to quantify soil quality and soil health. In addition, AI can play an important role in future work in the analysis of µCT data, for which basic principles have already been developed (Wieland et al, 2021).…”

Numerical Study of Gas Flow in Super Nanoporous Materials Using the Direct Simulation Monte-Carlo Method

“…An open-source code, PoreSpy [ 37 ], was used to replicate the geometry of the porous material for numerical simulations. PoreSpy includes a variety of predefined functions to extract data from images of porous material (for example, those obtained using X-ray tomography) and to generate artificial geometries of porous materials [ 37 , 38 , 39 ]. The Blobs function from the PoreSpy code generates an image with random noise and then applies a Gaussian blur to the image, creating a correlated field with a Gaussian distribution.…”

Application of X-ray computed tomography in soil and plant -a review

“…The transfer learning approach was used to train the convolutional deep network to select a feature space, which was then used to train a support vector machine (SVM) for the segmentation task. Wieland et al (2021) used a deep learning classification technique that can deal with low phase contrast X-ray CT data due to the neural network's robustness to solve the root segmentation challenge. Despite being trained with only synthetic soil columns, the deep network was able to successfully segment roots from the soil.…”