Thin-Pavement Thickness Estimation Using GPR With Compressive Sensing And Sparse Representation

In the field of civil engineering, Ground-Penetrating radar (GPR) Is commonly used to sounding the top layer of carriageways, i.e., the pavement layer. The resolution of GPR radar is bandwidth dependent. The thickness of layers can be deduced from the time delays of backscattered echoes and the dielectric constant of the medium. There is already a series of subspace based methods which have been developed to estimate the time delays, such as rotational invariance techniques (ESPRIT), multiple-signal classification (MUSIC) algorithm, Min-Norm. But these methods need long calculation time which is the main disadvantage. The comprcssivc sensing (CS) has been attracted significant attention in the recent years, we try to introduce it to solve the time-delay estimation (TDE) model. There are two main CS algorithms, Orthogonal Matching Pursuit (OMP), model-based iterative hard thresholding (IHT), have been introduced in this paper. Unfounately, the proposed TDE model does not satisfy the Restricted Isometry Property (RIP) strictly, the performance of CS method is not good as we expected. A model-based CS has been introduced here to overcome these barriers by making some necessary adaptations. Meanwhile, this algorithm sacrifices the efficiency compared to CS. An alternative strategy to avoid this dissatisfaction of RIP is sparse reconstruction. The use of11norms to achieve sparsity has been known for a decade, but the time consumption is very expensive. The re-weighted12norms have been proposed recently, which can enforce sparsity and efficiency at the same time. FOcal Underdetermined System Solver (FOCUSS) is introduced in this thesis. The regularization can be used to deal with the noise, which yields a regularized FOCUSS. Because the FOCUSS is an iterative algorithm, so the regularization parameter need to be set independent in each iteration. We find the traditional methods to estimate the regularized parameter are expensive and invalid hi sparse representation. So we proposed a heuristic method here to estimate the parameter.And the performance of these proposed methods will be made a comparison with the subspace methods. The advantages and disadvantages will be presented in the thesis.