Fast Forward Simulation Of Ground Penetrating Radar And Buried Target Detection Based On Machine Learning Method

Ground Penetrating Radar(GPR)is an instrument that uses high-frequency radar pulse waves to detect and image conditions below the surface.This is a non-destructive testing method that uses electromagnetic waves in the microwave band on the radio spectrum,and receives radar reflection waves caused by various object structures under the surface.GPR forward modeling is an important branch of GPR research.It plays an important role in real radar data interpretation and full waveform inversion.Therefore,this paper proposes three nearreal-time GPR forward simulation methods based on machine learning.First,the detection of steel bars in concrete is used as the GPR application scenario,and the water content of concrete,steel bar radius,and buried depth are used as model parameters,and the scattered echo signal is generated by using the Finite-Difference Time-Domain(FDTD)numerical simulation software gpr Max,as the data set;use principal component analysis(Principal Component Analysis,PCA)to reduce the dimensionality of the echo data to obtain the corresponding principal component weight coefficient,and use it as the output of the machine learning network.Secondly,a multi-layer recurrent network architecture and learning strategy based on random forest is designed,which not only fully exploits the internal causal relationship between the learning model parameters and the weight coefficients of the principal components,but also shares the interrelationships between the principal components,and has the ability to predict each The function of principal component improvement and correction can realize the fast forward simulation of ground penetrating radar based on machine learning.Compared with the neural network-based multi-layer recurrent network architecture and learning strategy and traditional random forest machine learning,it effectively improves the accuracy and stability of forward modeling.In addition,the current technology of ground penetrating radar for detecting and locating steel bars in concrete is relatively mature,but it is a challenging problem to quantify the diameter of steel bars.Therefore,this study designs a novel Machine learning framework:Combining two deep neural networks with random forests,completely taking the principal component coefficient of the synthesized echo data as input,and the water content of concrete,steel bar radius and buried depth as output for training,and established a machine learningbased The scene parameter prediction model realizes near-real-time detection of buried targets.The maximum error of the predicted concrete moisture content is 2%,and the maximum error of the buried depth of steel bars is 6.7%.The studied steel bars can be estimated within the resolution range of the antenna used.diameter of.