| The application of reverse time migration(RTM)imaging algorithm in ground penetrating radar plays an important role in obtaining accurate underground structures.As imaging scenarios become increasingly complex,RTM imaging faces three major challenges: large computation,low efficiency,and low-frequency noise.To address these issues,combining the finite-difference time-domain(FDTD)algorithm in the time domain with the discontinuous Galerkin time-domain(DGTD)algorithm,this paper proposed an electromagnetic wave RTM imaging algorithm based on a hybrid algorithm of FDTD-DGTD.The algorithm is optimized based on the wave field extension calculation process,reducing computation and improving performance.An RTM imaging denoising neural network(RDnet)applicable to electromagnetic wave reverse time migration imaging is proposed in this paper.This network can effectively improve the resolution of reverse time migration(RTM)images.The main research work and results are as follows:An electromagnetic wave reverse time migration imaging algorithm based on FDTD-DGTD hybrid algorithm is proposed,to improve the computational performance issues of reverse time migration imaging algorithm.Firstly,the electromagnetic field iteration equations of the FDTD-DGTD hybrid algorithm were derived and applied to the wavefield extrapolation process in electromagnetic wave inverse time migration imaging.Two-dimensional numerical examples were used to compare the imaging effects of different size metal cylinders using the FDTD,DGTD,and FDTD-DGTD hybrid algorithm.The results show that the proposed imaging algorithm effectively improves computational efficiency while maintaining the same imaging accuracy.It provides a smoother characterization of the curved edges of the metal cylinder.A reverse time migration imaging denoising neural network(RDnet),which is suitable to electromagnetic wave RTM images data,is proposed in this paper in order to solve the problem of low frequency noise.First,inverse time migration imaging was performed on different targets in different scenarios and positions through numerical simulation,and sample labels were obtained using wavefield decomposition,and an RTM image data set is produced.Then the RDnet network model is trained on the this RTM image data set.Finally,the noise reduction performance of the trained noise reduction network is compared with that of the traditional noise reduction methods.The results show that the RDnet network model produces good denoising performance,with a structural similarity(SSIM)of 0.963 and a peak signal-to-noise ratio(PSNR)of29.432 d B. |
