DGGV-E-Publikationen
Titel: Pore Detection and modelling with Deep Neural Networks
Autoren:
Christoph Schettler
Institutionen:
Federal Institute for Geosciences and Natural Resources, Germany
Veranstaltung: Abstract GeoUtrecht2020
Datum: 2020
DOI: 10.48380/dggv-jkgd-ya77
Zusammenfassung:
The description of drilling probes due to their porosity is an important research subject in terms of searching for nuclear waste repositories. Th current techniques use raster electron microscope pictures together with various edge detection and growing seed algorithms coming from the field of image processing. Those algorithms suffer from certain shortcomings. The BGR has started the project ITERATOR to overcome those shortcomings with the usage of artificial intelligence, mainly with deep neural networks. The first step is to produce enough REM images and use them within the second step to train a neural network. The first results of the used neural network are promising. Within the speech the starting material and the used network will be described. A closer look inside the network will be provided to better demonstrate the function of the neural network. To complete the presentation of the project ITERATOR the next steps within the project (3D-Models and detection of rock grains).
Ort: Germany