Abstract
Clast‐fabric analysis is a widely used method for investigating depositional and deformation processes in glacial sediments. However, traditional field‐based approaches lack standardization, are time consuming and introduce sampling bias. This study aimed to develop a novel approach to automate clast‐fabric analysis using machine learning‐based image segmentation applied on X‐ray computed tomography scanned drill‐cores. By retraining a deep neural network and integrating it into state‐of‐the‐art image‐segmentation software, we establish a scalable and adaptable workflow for the analysis of sedimentary samples. This included the following: (i) clast segmentation, (ii) object‐based analysis, and (iii) fabric analysis. We demonstrate this on drill‐core samples of glacial diamicts (tills), achieving performance comparable to leading segmentation models used in geological sciences. We further use this automated workflow to identify grain size‐dependent small‐scale fabric variations, demonstrating the advantages of deep learning over conventional methods. This workflow provides a foundation for future applications, such as long, continuous drilled sections and field samples, including fluvial and colluvial sediments.
Funding Information
  • Eidgenössisches Nuklearsicherheitsinspektorat
  • Nationale Genossenschaft für die Lagerung radioaktiver Abfälle
  • Leibniz-Institut für Angewandte Geophysik
  • Deutsche Forschungsgemeinschaft (450654078)

This publication has 35 references indexed in Scilit: