[17p-S2_204-1~18] 23.1 Joint Session N "Informatics"

3:30 PM - 3:45 PM JST(6:30 AM - 6:45 AM UTC)Presentation by Applicant for JSAP Young Scientists Presentation Award

[17p-S2_204-8] Estimation of Vector Magnetic Field Distribution Using Magneto-Optical Imaging and Deep Learning

〇Haruna Takaki¹, Kinjo Ryota¹ (1.Osaka inst.)

Keywords:

Magneto-optical imaging,Deep learning,Inverse problem

We propose a deep learning method to estimate 2D vector magnetic field distributions from polarization rotation angle maps obtained via magneto-optical imaging. Training data were generated using a physical simulation integrating the oblique incidence Faraday effect and the polar Kerr effect to train a U-Net-based network (PF-Net). For this underdetermined problem of estimating three components from a single input, we achieved a relative error of approximately 1% and an angular error of approximately 0.5° under the tested conditions. We also report on the robustness against nonlinear responses and device inhomogeneity.