The broad emergence of neural networks has influenced nearly every aspect of human activity, and research in physics is no exception. Nowadays, we are at a unique moment to re-evaluate many research procedures and consider whether neural networks can accelerate and enhance these activities. In this presentation, I will show you how to use two state-of-the-art neural networks, YOLOv8 and MONAI, to detect and analyze objects in different types of images, such as dark-field frames, diffraction patterns, and electron maps.
Mr Zelenka will focus on three intertwined projects that he is working on at FZU. The first project aims at developing an automated procedure to search for suitable flakes of van der Waals materials for further device processing. He will explain how he uses a motorized microscope and a software that he developed to automatically screen samples with flakes of layered materials, such as TaRhTe4, and create a catalogue of their parameters and coordinates.
In the second project, he optimizes and improves diffraction pattern analysis. Heuses YOLOv8, a powerful object detection model, to identify and select reflections in complicated diffraction frames. He also developed a small tool to convert Rigaku RASX files to standard image files.
In the last project, he aims to automate electron map analysis. He will show you how he uses MONAI, a medical imaging framework, to detect and classify types of molecules in electron density maps, and how this can help to understand the structure and properties of complex materials.
In this presentation, he will demonstrate how machine learning can be a useful and versatile tool for physics research. He will present a detailed guide on how he employs two projects for training neural networks, YOLOv8 and MONAI, to perform object detection and analysis.