Research On Electromagnetic Inverse Scattering Reconstruction Algorithm Based On Convolutional Neural Network

With the development of society and the progress of science and technology,electromagnetic reconstruction technology is more and more widely used in our daily life.Electromagnetic reconstruction belongs to electromagnetic inverse scattering problems.It refers to the technology of solving the dielectric constant distribution in the target imaging area when the scattering field is known.In contrast,the problem of electromagnetic forward scattering is to solve the radiation field when the dielectric constant distribution in the target imaging area is known.Electromagnetic inverse scattering problem is a nonlinear and ill posed problem,which can not be solved directly.At present,the mainstream methods for solving electromagnetic inverse scattering problems can be divided into linear methods and nonlinear methods.The linear method has high speed and is suitable for real-time imaging,but its accuracy is low;The nonlinear method has high accuracy,but the calculation process is complex.Therefore,the traditional electromagnetic inverse scattering method can not meet the requirements of solution speed and accuracy.Thanks to the development of deep learning technology,we try to use deep learning technology to realize an electromagnetic reconstruction method that can quickly obtain high-precision reconstruction results.Firstly,this paper studies the traditional electromagnetic inverse scattering algorithm,and combined with the current popular neural network technology,proposes an electromagnetic inverse scattering reconstruction algorithm based on convolutional neural network(CNN).CNN adds a convolution module to the network structure,which can well analyze the numerical change boundary in the input data.It has a wide application prospect in the field of image.In the electromagnetic inverse scattering problem,because all data are in matrix form,it is also possible to use CNN to solve the electromagnetic inverse scattering problem.Aiming at solving electromagnetic scattering problems with CNN,this paper puts forward two innovations: firstly,CNN is directly used to solve the forward and inverse problems in electromagnetic scattering problems,and the network structure of CNN is improved,and a more stable u-net network is introduced;Secondly,a hybrid electromagnetic reconstruction algorithm is proposed by combining neural network technology with traditional diffraction tomography(DT).In the innovative content of the first part,we directly train the scattering field in the electromagnetic scattering problem and the dielectric constant distribution matrix in the imaging area as the input and output of CNN respectively,and improve the network structure of standard CNN,add deconvolution and skip structure to the network structure,and introduce a better u-net network.The innovation of the second part is to analyze the traditional DT imaging algorithm,then combine the DT imaging algorithm with the neural network,use the solution of DT algorithm as the input training network of the neural network,and propose a hybrid electromagnetic reconstruction method.Finally,at the end of each chapter,the effectiveness of the new algorithm is proved by simulation results.