Microwave Measurement And Imaging Based On Inverse Scattering Theory

Firstly,a remote measurement method for complex permittivity of samples with arbitrary cross section is proposed.In this thesis,based on the scattering integral equation commonly used in microwave imaging,combined with the prior information about the sample under test(SUT),a complex permittivity inversion algorithm for samples with arbitrary cross section is proposed,which can be used to realize a bistatic dielectric constant measurement system.In this paper,simulation data are used to verify the effectiveness of this algorithm for SUTs with arbitrary cross sections.With the same prior information of SUT,the inversion performance and efficiency of this algorithm is compared with inverse scattering methods such as contrast source inversion method,born iteration method and distorted born iteration method.The influence of different incident wave angle and different receiving angle on the inversion result of this algorithm is discussed.Secondly,extended from common electromagnetic inverse scattering imaging systems,a time-division multiantenna system is designed for dielectric constant measurement.the calibration equation is derived,and experimental measurements on various SUTs with complex cross sections are conducted.The results are compared with conventional measurement methods.The influence of different incident wave angles and different receiving angles on the inversion results in experimental measurement is explored.Experiments show that for various low-loss SUTs with complex cross sections,using the scattered field data detected by the measurement system in this thesis,and the complex permittivity inversion algorithm proposed in this article,the permittivity can be effectively extracted.At last,this thesis studies and improves the deep learning inverse scattering imaging algorithm based on the backpropagation scheme(BPS).The existing U-Net and the sum of square loss are used to implement the BPS,and the imaging results shows the problems of local blur and background noise.Then,based on the inherent sparsity of inverse scattering problem,a sparse attention U-Net based BPS is proposed.The U-Net-based BPS is improved from two aspects: a combined target loss is proposed to replace the sum of square loss,and the combined target loss combines absolute value loss and total variation loss to emphasize the sparse characteristics of the imaging results;a sparse attention U-Net network structure with a mechanism is proposed for feature selection.Experiments on cylinder data sets generated in this paper prove that the use of combined target loss can effectively reduce the relative error of the imaging results and suppress the local blur and background noise in the BPS imaging results.In addition,by using sparse attention U-Net to achieve inverse scattering imaging,the results significantly exceed the original U-Net in terms of boundary definition,value accuracy,and suppression of background noise.