Physics-Assisted Learning Scheme for Quantitative Imaging via Near-Field Scanning Microwave Microscopy

Yulong Zhou,Ning Leng,Zhun Wei,Tiantian Yin,Ming Bai,Xudong Chen
DOI: https://doi.org/10.1109/tmtt.2023.3238208
IF: 4.3
2023-01-01
IEEE Transactions on Microwave Theory and Techniques
Abstract:Quantitative determination of the dielectric property is important to the analysis and evaluation of the investigated samples’ performance. This article proposes a physics-assisted learning scheme (PALS) for the quantitative imaging via the near-field scanning microwave microscopy (NFSMM) in a nondestructive way. To avoid using the machine learning in a black-box way, an efficient input generation is realized by utilizing the Rayleigh approximation method. Numerical and experimental examples demonstrate that the PALS recovers the dielectric parameters of the subsurface perturbation more accurately and shows a better generalization ability. The proposed learning scheme provides a reasonable way to combine the machine learning with underlying physics in the NFSMM, rather than utilizing machine learning in a black-box way. The same principle is also potential to be applied in other types of scanning microscopies, such as optical and thermal ones.
engineering, electrical & electronic
What problem does this paper attempt to address?