U-Net Based Method for ARPES Spectra Processing

Angle-resolved photoemission spectroscopy (ARPES) is a powerful tool for investigating the electronic structure of materials. However, resolving the electronic dispersion from ARPES spectra can be challenging due to the broadening effects, presence of different types of noise, resolution limitations, etc. This paper proposes a new approach for determining dispersion from ARPES spectra based on the U-Net neural network. The energy band extraction problem is regarded as the semantic segmentation task. We will show that the U-Net trained only with generated data can determine band structure from the experimentally obtained spectra, without prior denoising.