Reverse-Time Migration Applied To Ground Penetrating Radar And Intelligent Recognition Of Subsurface Targets

Ground penetrating radar(GPR),as a subsurface geophysical method,has the advantages of non-destructive,high resolution and rapidity.It has been widely used in geological exploration,civil engineering detection and lunar exploration.The processing of GPR data plays a key role in the later interpretation of data,and is a key step in the analysis of the geometric and electrical parameters of subsurface structure.In this article,related researches are carried out from two aspects:reverse-time migration(RTM)applied to GPR and intelligent recognition of subsurface targets.RTM can accurately reconstruct the geometry of subsurface structures and targets,and two-dimensional RTM has a unique advantage in calculation cost.However,in practical applications,data measurement is performed in three-dimensional space.Therefore,this article proposes to use 3D-2D data converter in the frequency domain to preprocess the radar reflected signals,and then perform 2D RTM imaging.First,the accuracy of this data converter was measured by a numerical simulation model.After that,the RTM imaging results of a concrete pipe simulation model illustrate the necessity of this data converter.The results of laboratory measured data from the Lunar Structure Electromagnetic Detector carried by the Chang-E 5 detector have once again verified the effectiveness of this data converter and the high resolution characteristics of RTM algorithm.This article then carried out a comparative study of time domain RTM based on finite difference time domain and frequency domain RTM based on layered medium Green's function,and analyzed the cost-effectiveness of frequency domain RTM with equal accuracy.The advantages show that the layered structure can be quickly imaged by the frequency domain RTM.Intelligent detection based on image processing is used to detect the GPR image.Combined with the parameter fitting,the true position and other parameters of the subsurface target can be calculated.It has great advantage in computational efficiency compared to migration imaging,and it can achieve real-time interpretation of GPR data.In this article,the convolutional neural network(CNN)based on Single Shot Multibox Detector(SSD)is used to detect the hyperbolic characteristics of GPR image.This method can achieve real-time and high-precision detection compared with the existing target detection algorithms of sliding window feature extraction combined with classifier or region proposal network combined with CNN in the field of GPR.This article uses the dataset HDXMU to train and test the neural network.In addition,transfer learning method is used to improve the detection accuracy and accelerate the convergence.A data augmentation strategy is also applied to improve the robustness of this deep neural network.The detection results show that this method can realize the real-time intelligent recognition of hyperbolic targets in GPR image,and GPR data can be interpreted in time.