Research On Ground Penetrating Radar Data Reconstruction Method Based On Time Series Analysis

Ground penetrating radar is one of the most commonly used technologies in geophysical prospecting technology.The ground penetrating radar can be used to obtain imaging profiles reflecting underground pipelines.However,noise interference often affects the judgment of the target signal in actual detection,resulting in pipeline identification errors.Due to the complexity of the underground structure and the limitations of the equipment itself,the result of pipeline detection may appear unexpected situations such as breaks and twists.Based on this,this thesis focuses on the problem of ground penetrating radar signal noise during the acquisition process and the problem of missing values between adjacent B-scan data during the detection process,the reconstruction method of ground penetrating radar data is studied.The specific research content of this thesis is as follows:(1)A similarity reconstruction method of B-scan data based on Curvelet transformation is proposed.This method is based on the Curvelet transform combined with the time series similarity search,uses the convex set projection method and the conjugate symmetry of the time signal spectrum to perform the interpolation reconstruction method of the ground penetrating radar data.Compared with the original detection data,the virtual profile constructed by the interpolation reconstruction method compensates for the missing values in the adjacent B-scan data,which is conducive to the understanding of the overall structure of the region.Compared with the other five interpolation methods introduced in this thesis,this method is more accurate in judging between adjacent B-scan data,and the reconstruction effect is better.(2)A signal optimization reconstruction method based on EMD decomposition and KSVD B-scan data is proposed.This method combines the K-SVD algorithm with the empirical mode decomposition method,and uses singular value decomposition to improve the overall analysis capability of the virtual profile between adjacent B-scan data.Simultaneously,combined with the empirical mode decomposition ability to decompose the non-stationary and non-linear characteristics of the ground penetrating radar data signal,the effect of clutter suppression is effectively improved.Finally,a local weighted regression method is added to smooth the data and retain a large number of effective signals of hyperbolic waves.Experimental results show that under noisy conditions,this method can not only optimize the noise data well,but also shows a good suppression effect on the clutter part of the original B-scan data.(3)A B-scan data denoising reconstruction method based on sparse Naive Bayes is proposed.This method constructs a sparse naive bayes discriminant model.It combines sparsity and information entropy to enhance the classification performance of the naive Bayes model,and extracts the horizontal and vertical time series data of the GPR sampling points as the input of the model.By calculating the posterior probability of each sampling point belonging to a noise point,it is judged whether the sampling point is a noise point.Combining the advantages of bilateral filtering and wiener filtering for noise removal,the effective signal of hyperbolic wave is retained while removing noise in a targeted manner.This experiment uses B-scan data with different concentrations of noise as the experimental data,which has a higher peak signal-to-noise ratio compared with a single filtering method and hyperbolic wave data is retained in the process of removing noise data.