Ground Penetrating Radar Clutter Removal Based On Low-rank And Sparse Matrix Decomposition

Ground Penetrating Radar(GPR)is a geophysical method for underground target detection using electromagnetic wave propagation.It has the advantages of non-destructive testing,high detection accuracy,high detection efficiency,and convenient on-site implementation.GPR has been widely used in engineering quality inspection,geological exploration,archaeology,military detection,and other fields.Due to the serious attenuation of the detection signal in lossy media and the strong clutter caused by the ground reflection,the direct coupling between transmitting and receiving antennas,and the background clutter,the signal-to-noise ratio of the GPR echo signal is low,and the target reflection is often submerged in the clutter,which is not conductive to subsequent data interpretation,quantitative analysis,and target recognition.Therefore,clutter removal is a significant technology in GPR signal processing and a crucial link to improving target detection and recognition.Many methods have been proposed and applied to GPR clutter removal,among which the Low-Rank and Sparse matrix Decomposition(LRSD)method has been proved to be highly effective.However,these methods still have some disadvantages,such as high computation complexity,parameter sensitivity,poor performance in a complex environment,and the inability to process 3D data.To solve these problems,this thesis deeply studies clutter removal methods based on the LRSD.The main innovations and research results of this thesis are as follows:(1)To solve the problems of high computation complexity and parameter sensitivity of existing methods,an improved clutter removal method based on factor group sparse regularization is proposed.It uses non-convex matrix factorization as a surrogate for the nuclear norm in the existing methods to reduce the computational complexity.It also performs parameter-independent clutter removal by updating the over regularization parameter and soft thresholding.Simulation and experimental results show that the proposed method has higher performance and shorter running time than the state-of-the-art methods,and it is not sensitive to the parameters.(2)To account for the nonparallel case between antennas and the ground surface,a novel clutter removal method based on weighted nuclear norm minimization is proposed.It uses a non-convex weighted nuclear norm to replace the nuclear norm in the existing methods and assigns different weights to different singular values of the echo data.Simulation and experimental results demonstrate its superiority to the state-of-the-art methods in nonparallel cases.(3)Due to the limitations of current clutter removal methods for GPR data with missing entries,a novel clutter removal method based on deep autoencoder is proposed.Firstly,the missing GPR data is recovered by a deep autoencoder,and then another deep autoencoder is used to decompose the recovered GPR data into its low-rank and sparse components.The obtained sparse matrix is the clutter-free target image.Simulation and experimental results show that when the missing data rate is up to 50%,the proposed method can still complete the missing data successfully and suppress the strong clutter effectively.(4)Due to the lack of clutter removal methods for GPR three-dimensional(3D)data,a novel clutter removal method based on tensor robust principal component analysis is proposed.It utilizes the third-order tensor nuclear norm to approximate the rank of the 3D low-rank clutter tensor and uses the tensor l-1 norm to limit the sparsity of the target matrix.In addition,randomized singular value decomposition is applied to reduce the computational complexity of the algorithm.For the first time,it realizes the decomposition of a low-rank clutter tensor and a sparse target tensor from 3D data.Simulation and experimental results demonstrate that the proposed method has good clutter removal performance for GPR 3D data.