BGR-Beiträge zu der BGE-Veranstaltung
„Tage der Standortauswahl“ vom 08.–10.06.2022
Themengebiet: „Endlagerkonzepte und Geoprozesse“
Automated Segmentation of Pores in Opalinus Clay by Using Artificially Enhanced SEM Images
M. Brysch1, B. Laurich1, C. Schettler1, and M. Sester2 (1 Federal Institute for Geosciences and Natural Resources, Hannover; 2 Gottfried Wilhelm Leibniz University, Hannover)
Kurzbeschreibung:
Accurate knowledge of the microscopic pores of a host rock is crucial for the safe underground storage of radioactive waste. Their amount, morphology, size frequencies and their spatial distribution strongly control the rock’s physical properties, such as permeability and strength. A specialized method of scanning electron microscopy (SEM) has already been established to analyse rock pores[1–3]. Here, the pores are characterized and evaluated according to size, orientation and frequency, using segmentation masks. However, the creation of these masks involves some difficulties caused e.g. by the nano-scale resolution limit of SEM, that leaves a majority of small micro-pores undetected.
In this work, we present a method that first artificially improves the resolution of SEM images by one order of magnitude, generating super-resolution images. Our approach uses a modified super-resolution generative adversarial network [4], which we specifically trained to upsample our SEM images.
In a further step, nine machine-learning classifiers were combined into a voting classifier method that computes a pore probability field instead of binary segmentation masks.
This allows the derivation of distinct confidence levels that reduce false pore segmentation and capture pore edges more uniformly and consistently. Finally, we plan to combine the voting classifier and super-resolution to the so-called super-segmentation. In a test case with opalinus clay, we segmented many pores that remained undetected or insufficiently segmented in conventional SEM images. We were able to decrease the lower truncation limit of pores [5] by one order of magnitude, so that the extrapolation to even smaller pores becomes more accurate.
Referenzen:
[1] M. E. Houben, G. Desbois and J. L. Urai, “Pore morphology and distribution in the Shaly facies of Opalinus Clay (Mont Terri, Switzerland): Insights from representative 2D BIB–SEM investigations on mm to nm scale,” Applied Clay Science, Vol. 71, 2013, pp. 82–97.
[2] L. M. Keller, “3D pore microstructures and computer simulation: Effective permeabilities and capillary pressure during drainage in Opalinus Clay,” Oil Gas Sci. Technol. – Rev. IFP Energies nouvelles, Vol. 76, 2021, p. 44.
[3] J. Klaver, G. Desbois, J. L. Urai and R. Littke, “BIB-SEM study of the pore space morphology in early mature Posidonia Shale from the Hils area, Germany,” International Journal of Coal Geology, Vol. 103, 2012, pp. 12–25.
[4] X. Wang, K. Yu, S. Wu, J. Gu, Y. Liu, C. Dong et al., “ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks,” Proceedings of the European conference on computer vision (ECCV) workshops, 2018. https://arxiv.org/pdf/1809.00219.
[5] E. Bonnet, O. Bour, N. E. Odling, P. Davy, I. Main, P. Cowie et al., “Scaling of fracture systems in geological media,” Rev. Geophys., Vol. 39, No. 3, 2001, pp. 347–383.
