Academy of Mathematics and Systems Science, CAS
Colloquia & Seminars

Speaker:	闫亮，东南大学
Inviter:
Title:	A hybrid deep learning framework for the inverse scattering problem with the impedance boundary
Language:	Chinese
Time & Venue:	2023.05.23 09:00-10:00 腾讯会议:663664161
Abstract:	In this talk, we are concerned with the inverse scattering problem with the impedance boundary condition, given far-field data for one incident direction at a fixed frequency. It is noticed that, to solve such an inverse scattering problem, iterative numerical algorithms based on the PDE-constrained optimization are commonly need a reasonable initial guess and a forward solver at each iteration step. Moreover, traditional iterative algorithms need to solve the inverse problem again with given new far-field data. To overcome these issues, a hybrid deep learning framework based on the scattering informed neural network(SINN) and a least squares method is proposed to simultaneously reconstruct the shape and the impedance. In this framework, SINN is trained to learn both the shape of the obstacle and the density function of the single layer potential, and then, the learned shape and density function are applied in the least squares method to recover the impedance. Once SINN is trained, the inverse problem could be computed directly via forward propagation through deep learning. Several numerical examples are presented to show the promising feasibility of our method.