Name Surname
Generative AI Assisted super-resolution and artificial 3D tomography of crystal zoning patterns: Implications for timescale estimates and volcanic hazard assessment
Main Supervisor: Monika Sester (Gottfried Wilhelm Leibniz Universität Hannover)
Co-Supervisors: Lukas Schlatt (NU Instruments), Andras Lukacs (Eötvös Loránd Tudományegyetem)
Objectives
The project focuses on investigating volcanic eruptive products (e.g., phenocrysts or microlites) in 2D and 3D. To this end, Deep Learning methods, as well as image processing methods will be used to segment the crystal populations and their geometric properties. The analysis will be based on available data sources, e.g., 2D BSE and 2D EDX images, as well as 1D EPMA measurements. The data will be complemented by 2D trace element maps obtained by LA-ICP-MS. In addition, based on maps obtained for a series of parallel sections, Deep Learning will be used to reconstruct (pseudo) 3D structures, which will be validated by X-ray tomography for elements with high-density contrast. The underlying idea is to exploit the additional/redundant information from the neighbouring slices to increase the quality and reliability of 3D textural information. The crystals will be clustered based on their compositional properties. Subsequently, a system diagram will be created from those clusters. Case studies will include Pico (Azores) and Fogo (Cape Verde).
Epected Results
Automatic processes to analyze volcanic eruptive products quickly and thus support the domain specialist in their subsequent analysis; Analysis in 2D and pseudo 3D. Results include:
- The identification of individual crystals, their geometry and their composition;
- The identification of clusters to infer processes, coded in a system diagram;
- The evaluation of benefits of 3D against 2D approach.