Freiburg 2024 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

Q: Fachverband Quantenoptik und Photonik

Q 5: Magnetometry

Q 5.7: Talk

Monday, March 11, 2024, 12:45–13:00, HS 1221

Physics-informed neural networks for analyzing NV-diamond wide-field images of magnetic field distributions measured with a lock-in camera — •Mykhailo Flaks^1,2, Joseph S. Rebeirro¹, Muhib Omar¹, David A. Broadway², Patrick Maletinsky², Dmitry Budker^1,3,4, and Arne Wickenbrock³ — ¹Helmholtz Institut Mainz, 55099 Mainz, Germany — ²Department of Physics, University of Basel, Basel, CH-4056, Switzerland — ³JGU Mainz, 55128 Mainz, Germany — ⁴Department of Physics, University of California, Berkeley, California 94720-7300, USA

We use a novel approach with physics-informed neural networks (PINNs) for analyzing magnetic field distributions. We focus on widefield images acquired from nitrogen-vacancy center-ensembles in diamond using a lock-in camera. Our method allows to reconstruct source distributions such as currents or magnetization. The inverse reconstruction technique can be used for mapping current distributions in conductors, studying superconductor vortices, and exploring magnetization textures.

We apply these techniques to the images acquired with a microwave-free NV-based imaging device, that uses the ground state level anticrossing (GSLAC) feature. With the addition of lock-in acquisition of the magnetic field image and the PINN to the inverse problem analysis, we alleviate the effect of the ill-posed nature of the inverse problem and the presence of noise in data. We address the improved sensitivity of the underlying source distribution to advance the measurement method towards a biocompatible sensor for neurons.

Keywords: magnetic fields; signal processing; neural networks; photoluminescence; optically detected magnetic resonance