A piece of rock can reveal a great deal: what minerals it consists of, how it was formed, and what traces more recent processes have left behind. This becomes visible in wafer-thin sections under the microscope. Funded by the German Research Foundation (DFG), the ‘AI Rocks!’ project at the TU Bergakademie Freiberg (TUBAF) is transferring this knowledge from the collection cabinets into a global database.
Over the next two years, around 11,000 so-called thin sections are to be digitally recorded and linked to scientific descriptions. The rock sections, which are approximately 0.03 millimetres thick, are mounted on glass slides. Under special lighting, minerals appear in different colours and levels of brightness. An automated microscope scanner at Freiberg University Library captures this information in high-resolution images. The teaching collection is of particular importance here. Only universities have 20 identical thin sections of the same rock – for teaching purposes. The aim is to create a dataset containing many very similar images. This variation is very important, as the information contained in each thin section is identical. The resulting data volume will amount to approximately 300 TB.
What a photograph of a rock can reveal
The digital recordings and their descriptions are to be made freely available to various international specialist communities; primarily researchers and lecturers in the field of AI, but also those in the Geosciences, materials research and engineering should be able to work with them. Initial datasets will be published via the TUBAFmedia university platform as the project progresses; a scientific data archive is intended to ensure their long-term accessibility.
The respective data packages will be published alongside specific use cases and appropriate benchmark tests. Such standardised comparative tests make it possible to verify how effectively a method removes images, for example. To this end, the project provides processed training and test data, evaluation criteria, as well as sample programmes and reference models. Research teams from all over the world can thus work directly with the data and compare their results under the same conditions.
As part of the project, a Data Steward and a Scientific Data Engineer will be employed for two years, whilst student assistants will be tasked with scanning the thin sections.
Many specialist areas, a shared database
Other experts from the TU Bergakademie Freiberg are also indispensable to the project: Prof. Guido Meinhold and Dr Michael Magnus from the Department of Sedimentology and Volcanology, Dr Christin Kehrer, curator of the Petrology and Mineral Deposits Collections, and Prof. Jan-Michael Lange, curator of the Petrographic Collection at the Senckenberg Natural History Collections in Dresden, are contributing their expertise in the Geosciences and in collection management. Prof. Holger Lieberwirth and Patricia Vivanco are contributing the thin-section collection of the Institute for Processing Machines and Recycling Systems Technology. Dr Stefanie Nagel and Oliver Löwe from the University Library, as well as Dr Friederike Röver and Dr Sebastian Borrmann from the University Computing Centre, are supporting the preparation, publication, storage and processing of the data. Student assistants are helping with digitisation and documentation.
The principal applicants are Prof. Gerhard Heide from the Institute of Mineralogy, together with Prof. Bernhard Eidel from the Institute of Mechanics and Fluid Dynamics (IMFD) (both at TU Bergakademie Freiberg), Dr Julia Meyer (Freiberg University Library) and Dr Andreas Kluge (Freiberg University Computer Centre). Heide combines his expertise in mineralogy with the Geosciences collections, whilst Eidel contributes his experience in artificial intelligence (AI) and the modelling of material properties. With Meyer as Director of the University Library and Kluge as Director of the University Computing Centre, data management, mass data storage and high-performance computing for the benchmark tests also form the core of the project.
