| In the process of urban underground pipeline construction,due to historical development and other reasons,a unified underground pipeline management system has not been formed,and its layout and planning are relatively complex,so it is necessary to know the underground structure in advance.Ground Penetrating Radar(GPR)is widely used in the construction and maintenance of urban underground pipelines with its characteristics of simple operation,fast speed and no damage.In the actual detection process,due to the interference of various electromagnetic noises from the equipment and the external environment,the collected images contain a large number of random noises and clutter,which seriously affects the accurate judgment of the real target.Therefore,for GPR pipeline detection images,based on bayesian theory combined with multi-filtering and non-negative matrix decomposition technology,a combined bayesian network filtering denoising method and a bayesian non-negative matrix decomposition clutter suppression method are proposed in this paper.The specific research contents of this paper are as follows:(1)A combined filtering denoising method based on bayesian network is proposed.In bayesian network construction,filter coefficients are taken as network nodes,and network parameters are set by maximum likelihood estimation.Then,the Junction Tree algorithm is selected to accurately infer the posterior probability value of noise and effective signal for each pixel,and the maximum probability value is selected to judge whether is noise,thus denoising is carried out.The method combines the advantages of multiple filters to make the processing result not only remove the noise but also retain the effective signal.In this paper,real detection images are used for experiments,and the results show that the model has good stability and effectiveness for GPR denoising.(2)A clutter suppression method based on bayesian non-negative matrix factorization is proposed.The horizontal gradient of the captured images is taken as the input matrix to be decomposed,and the Probability Nonnegative Matrix Factorization(PNMF)is carried out by using the variational bayes in the probability model.Based on the assumption that the matrix and the decomposed basis matrix and coefficient matrix obey a certain nonnegative probability distribution,the low-rank matrix representation of clutter components is obtained by using variational bayes approximation,and then clutter is separated from the image.In this paper,the simulation and real data are compared to evaluate the robustness of the proposed method to GPR clutter suppression. |
