Research On Phaseless Electromagnetic Inversion Method Based On Deep Learning

Electromagnetic inverse scattering imaging technology is one of the research focuses in computational electromagnetics.It predicts the position,shape and constitutive parameters of objects in the area according to the information of the incident field and the scattered field in the target area obtained by the transmitting device and the receiving device(Such as dielectric constant distribution,conductivity,etc.).Although there have been many advances in its related research in recent years,there are still many problems in practical application.The content of this article aims to improve some of the shortcomings encountered in traditional electromagnetic inverse scattering imaging technology.First of all,this thesis proposes three electromagnetic inverse scattering imaging algorithms based on phase-free data to solve the problem of difficulty in obtaining scattered field data in actual scenes.The phaseless total field data required by the phaseless algorithm only needs to measure the amplitude of the total field data without measuring the phase information,which effectively reduces the difficulty of data measurement in actual situations.Secondly,based on the use of phaseless data,this paper also introduces the current popular deep learning(DL),combining DL with electromagnetic inverse scattering imaging technology,and uses the excellent learning ability and generalization ability of DL,more accurately predict the physical properties of the unknown object,and on the calculation speed has been greatly promoted,can realize real-time imaging.This paper also proposes a phase recovery scheme based on DL,which can recover phaseless data into phase data,and then perform scatterer reconstruction.This scheme also achieves the purpose of avoiding measuring phase.Finally,this paper proposes the application of autoencoders in electromagnetic inverse scattering filed.The autoencoders is divided into two parts: encoding and decoding.Encoding can compress large-dimensional data into small-dimensional data;decoding can restore compressed data to original data.This scheme uses deep learning to map the phaseless total field data to the encoder output,and then restores the original image after decoding.