Ground Penetrating Radar Reverse Time Migration And Full Waveform Inversion Based On GPU Parallel Acceleration

The underground structure is an important indicator that is inseparable from human production and life.It plays an important role in oil and gas exploration,archaeological research,road disease detection,building quality detection,and cave detection.The underground structure includes both the interface between different media and the internal parameters of different media.At present,many studies have proved that reverse time migration is an effective method to obtain the spatial shape of underground structures,and full waveform inversion is considered to be an effective method to obtain the internal parameter information of underground structures.On the basis of previous research,this paper improves the ground penetrating radar reverse time migration and full waveform inversion algorithm,reduces the storage amount,improves the calculation efficiency and convergence speed,and enhances the practicability of these two methods.The preliminary conclusions are as follows:(1)By introducing the CUDA parallel architecture and calling multi-core parallel computing electric field and magnetic field component nodes,the computing power of the computer is fully released,and the fast ground penetrating radar forward algorithm based on the GPU parallel architecture is programmed and implemented,and the calculation is performed with the traditional CPU algorithm.After the efficiency comparison,it is found that with the increase of the number of grids,the speedup ratio gradually becomes larger,and finally can reach more than 10.Through the model trial calculation of the abnormal body and the complex,it is concluded that for the abnormal body model with a grid size of 200×200,the speedup ratio is 4.20;for the complex medium model with a grid size of 600×400,the speedup ratio reaches 11.04.(2)Aiming at the problems of large memory occupation and low computational efficiency of reverse time migration,the strategy writing algorithms of wave field reconstruction and GPU parallel acceleration based on CUDA platform are adopted respectively.By storing the wave fields at different times on the boundary grid points,the wave field of all grid points is avoided to be stored,and the wave field is reconstructed at the same time of back propagation.By comparing the normal wavefield and reconstructed wavefield of the uniform model and the complex model,it is found that the absolute errors are all in the order of 10e-5,which proves that the accuracy of the wavefield reconstruction is high enough,and the wavefield heavy The structural technology is reliable;the results of the classical anomalous body and the cave model trial calculation show that the reverse time migration enables the interface and the anomaly body to be presented with high precision through the homing of the reflected wave and the convergence of the diffracted wave.(3)The L-BFGS method is used to calculate the descending direction,and the Hessian matrix is further approximated on the basis of BFGS,and the complete approximate Hessian matrix is no longer stored,which reduces the storage capacity and improves the speed of dual-parameter inversion.When writing algorithms,the CUDA parallel architecture is introduced to improve computational efficiency.Three models are constructed to test the algorithm in this paper.The results show that: the overall convergence speed is faster and the convergence accuracy is higher;The effect is poor;combined with reverse time migration,high-precision imaging of the structural shape and internal parameters of the cave model is achieved,which proves that the method in this paper can obtain accurate underground structures.