Hybrid input method for solving electromagnetic inverse scattering problem based on deep learning

Abstract

The invention discloses a hybrid input method for solving an electromagnetic inverse scattering problem based on deep learning, and the method comprises the following steps: 1, obtaining the quantitative information of an unknown scatterer through employing a quantitative inversion method, and enabling the quantitative information to comprise a contrast ratio; 2, obtaining qualitative informationof an unknown scatterer by utilizing a qualitative inversion method, wherein the qualitative information comprises normalized numerical values, an index function is defined on the interested domain and used for judging whether the sampling points are located inside or outside the unknown scatterer, a set of normalized numerical values are obtained according to the index function, and the normalized numerical values indicate whether each sampling point is located at a point inside the unknown scatterer or not; 3, performing point multiplication on the normalized numerical value and the contrastvalue, and converting a point multiplication result into a combined dielectric constant value; 4, taking the combined dielectric constant value as the input of the neural network, taking the real dielectric constant value of the scatterer as the output of the neural network, and training the neural network.

Description

technical field [0001] The invention relates to the field of quantitative microwave imaging, in particular to a mixed input method for solving electromagnetic inverse scattering problems based on deep learning. Background technique [0002] The purpose of electromagnetic inverse scattering problems (ISPs) is to determine the properties of unknown scatterers such as their shape, position and electrical properties (permittivity and conductivity) from measured scattered field data. It has a wide range of applications in the fields of non-destructive testing, microwave remote sensing, biomedical imaging diagnosis, through-wall imaging and security inspection. Non-linearity and ill-conditionedness are two major difficulties faced by the electromagnetic inverse scattering problem. To solve this problem, a variety of electromagnetic inverse scattering imaging methods have emerged on the basis of the theoretical research of electromagnetic inverse scattering. These methods can be d...

AI Technical Summary

Problems solved by technology

[0006] As a qualitative inversion method, DSM can quickly reconstruct the shape and position of unknown scatterers with low computational cost, but it only involves the basic solution in a uniform background and the simple inner product It can indicate the shape and position of unknown scatterers, but cannot recover the dielectric constant of unknown scatterers. When detecting targets with the same shape but different dielectric constants, DSM cannot distinguish between these two scatterers

Images