Quartz grains play a crucial role in studying the natural environment. Their surface properties carry traces of where they came from and how long a journey they’ve taken—whether they were transported by water, wind, or glaciers. Thanks to these characteristics, scientists can better understand landscape evolution and the processes that have shaped it. A student from the Faculty of Information Technology at the Czech Technical University in Prague (FIT CTU) has come up with a way to improve and simplify the analysis of these quartz grains. Bc. Denys Anopa proposed how artificial intelligence can speed up and enhance the accuracy of this research process. For his bachelor’s thesis, he received the Dean’s Award for the summer semester of 2024/2025.
Until now, scientists had to examine quartz grains and their features manually under an electron microscope—a process that was not only time-consuming but also heavily dependent on the subjective judgment of experts, often leading to inaccuracies.
To address this, Denys decided to make the process faster and more precise using artificial intelligence. He created a dataset of microscopic images and tested several deep learning models capable of automatically recognizing surface features. The best of these models was able to identify certain properties with relatively high accuracy—and much faster than a human expert, needing only a few seconds to analyze a single image.
Although the results are not yet perfect, they demonstrate great potential. In the future, quartz grain analysis could become faster, more accurate, and less dependent on human judgment.
“It turned out that even a model trained on a small amount of data can achieve fairly good results. In the future, it could be used not only for analysis but also for generating new data more efficiently for further research,” says Bc. Denys Anopa.
“This bachelor’s thesis stands significantly above the usual level. During its development, two new contributions to the scientific community were created—a dataset and an algorithm for automated processing,” says thesis supervisor Ing. Jakub Novák.
The work also produced a practical outcome that may help other researchers. During the project, Denys had to manually annotate more than two hundred microscopic images, as no publicly available datasets of quartz grains existed at the time. The resulting dataset will therefore be published so that other experts can build upon it and use it in their own experiments.
“I definitely want to continue working on this. My first goal is to publish the dataset in a way that allows it to be expanded and used to test new approaches,” Denys adds.
The project output also includes a demonstration Colab Notebook, which allows anyone to try out online how the model makes predictions on their own data.
