Research On Microwave Electromagnetic Inverse Scattering Imaging System And Method Based On Deep Learning

Microwave electromagnetic inverse scattering technology is a quantitative imaging method.As an accurate and non-destructive measurement modality for imaging,this technology is widely used in many fields such as science,engineering,military and medicine.It reconstructs the permittivity distribution inside the target through the scattered field,and it has the characteristic of non-contact.A large number of algorithms have been proposed and developed over the past few decades to solve electromagnetic inverse scattering problem.However,due to its inherent strong nonlinearity,ill-posedness and expensive computation cost,these algorithms still have many problems and challenges.In recent years,Deep Neural Network(DNN)has been widely used in the imaging field.The strategy based on DNN can greatly reduce imaging time and improve the imaging quality significantly.Therefore,combining the shortcomings of electromagnetic inverse scattering problem and the advantages of DNN,this thesis propose a deep learning algorithm for microwave imaging.The main work of this thesis is as follows:(1)A microwave imaging system is made,and the important components of the system are introduced in detail.First,the design of the balanced Vivaldi antenna is introduced.Secondly,the hardware and software of the data acquisition system are introduced in detail.(2)The problem of TM electromagnetic wave inverse scattering in two-dimensions is described in detail,and the methods for solving the forward problem by the Green’s function integral equation are studied in detail.(3)Traditional Contrast Source Inversion imaging algorithm(CSI),Multiplicative Regularized Contrast Source Inversion imaging algorithm(MR-CSI)and Multiplicative Regularized Contrast Source Inversion based on Multi-Frequency imaging algorithm(MF-MRCSI)are studied.(4)The feasibility of traditional CSI,MR-CSI and MF-MRCSI algorithms is verified by simulation data and measured data from the University of Manitoba.The three inversion imaging algorithms are compared from the aspects of different frequencies and different permittivity.The anti-interference and reconstruction effect under an arbitrary background of the three algorithms are discussed.The results show that the MR-CSI method is superior to the other two methods.At the same time,a microwave imaging application system has been developed,and it greatly simplifies the calculation process by integrating the forward and inverse calculation processes into a convenient and efficient user interface.(5)A deep learning algorithm called complex-valued pix2pix(CVP2P)is proposed.The principle and structure of the algorithm are introduced in detail,and compared its results with the MR-CSI method.The results show that for strong scattering objects,the result of CVP2 P method is superior to that of MR-CSI method.It is worth mentioning that the reconstruction time of the CVP2 P method is less than 1s,which is several orders of magnitude faster than traditional solutions.At the same time,the generalization ability of algorithm has been verified.