Research On High Efficient And Accurate Electromagnetic Imaging Methods For Composite Material

The composite materials have been widely used in aerospace,automobile,medical equipment and many other fields because they have advantages of high strength,high stiffness,low weight,and corrosion resistance,etc.As an artificial material,it has the characteristics of multi-material,multi-layer,periodicity,anisotropy,and so on.However,due to the instability in its manufacturing process and the fatigue accumulation during the use,many defects could appear such as displacement,cracks,bubbles,and pores.These defects can greatly affect the properties of composite material.Therefore,how to realize the rapid and accurate nondestructive testing of these materials is the key.The electromagnetic imaging system has many advantages such as strong penetration,low attenuation,sensitive polarization and high resolution,which makes it very suitable for nondestructive testing of composite materials.However,these defects are mostly subwavelength compared with the operating wavelength of the electromagnetic imaging system,which greatly increases the difficulty of detection.Therefore,the research of super-resolution electromagnetic imaging algorithm based on full-wave scattering effect is carried out in this dissertation.Several high-efficiency,highprecision and high-stability imaging algorithms are proposed for electromagnetic imaging.The main work of this dissertation can be summarized as follows:Firstly,the research of electromagnetic imaging algorithms is summarized.The advantages and disadvantages of different imaging algorithms as well as their great potential are analyzed.Meanwhile,the deep learning technology,which has received widespread attention in recent years,is briefly introduced.The advantages and feasibility of deep learning technology within electromagnetic imaging algorithm are discussed.Secondly,to overcome the low efficiency of the Born iterative method,a fast forward solver based on the Born series is proposed to replace the large matrix inverse operation of the traditional operator.By this method,each horizontal iteration is integrated into the vertical iteration,which improves the computational efficiency.Numerical and experimental results show that the new algorithm based on fast forward operator can greatly reduce the imaging time.Thirdly,two new iterative algorithms based on the differential-equation model and the full-wave scattering effect are proposed.For those algorithms based on integralequation model,the Green’s function is an indispensable part.However,the solution of customized numerical Green’s function under complex background is very complicated and time-consuming.In order to solve this problem,two new iterative algorithms based on differential-equation model are proposed,which do not need Green’s function.At the same time,compared with other similar algorithms,the new algorithm is not constrained by additional conditions in imaging accuracy.In addition,for the variable parameter in iterations,an adaptive regularization factor is proposed to choose the appropriate value of regularization factor in iterations,which realized the purpose of automatic selection in the iterative process.The experimental and simulation results show that the two new algorithms are flexible when dealing with complex background.Meanwhile,they can obtain better imaging accuracy and higher imaging quality.Fourthly,the above algorithm still needs iteration to complete convergence.To improve the imaging efficiency,an effective approximation method based on signal subspace theory is proposed.It can provide a more accurate result in each iteration step so that the purpose of improving the convergence speed can be achieved.Simulated numerical results show that the new algorithm based on signal subspace approximation can complete convergence with fewer iterative steps.Finally,a novel complex neural network is proposed for electromagnetic imaging.In recent years,deep learning has been widely applied in the field of inverse scattering problem(ISP)imaging.However,the input data of traditional neural network is mostly real data,which makes the complete information of scattered data partially discarded,resulting in low accuracy of imaging results.To solve this problem,in this dissertation,a complex neural network is proposed for electromagnetic imaging,which can directly use the scattered data as the input and realize the purpose of utilizing the complete imaging information.Simulation results show that the algorithm based on complex neural network and deep learning can provide good results almost in real time.And in the aspect of imaging quality and efficiency,it greatly outperforms the traditional iterative algorithms.