Research On Ground Penetrating Radar Imaging Techniques

GPR (Ground Penetrating Radar) uses electromagnetic waves to probe theunderground region nondestructively. GPR can be used to reconstruct the information ofunderground scenes and targets by the reflection and scattering phenomena stimulatedby the discontinuity of the underground medium, as well as underground targetdetection and recognition. The information of underground targets can be displayeddirectly by GPR imaging technique, which is also convenient for the interpretation ofunderground targets. Because of the advantages such as nondestructive, high resolution,fast probing and high safety, GPR has been widely used both in military and civilianapplications, and it possesses great developing potential. Unlike traditional radarimaging, the observing scene of GPR is relatively special because GPR works in nearfield conditions and the characteristics of both the imaging scene and targets are rathercomplicated. In order to solve these problems, an in-depth and detailed research hasbeen carried out in this thesis in the aspects of time domain imaging technique,frequency domain imaging technique, super resolution imaging technique andcompressive sensing imaging technique.First of all, the BP (Back Projection) algorithm in time domain is investigated. Thelayered mediums in the scene of GPR imaging cause the refraction phenomenon ofelectromagnetic waves on the interface between different mediums. In the procedure ofimaging the calculation of the coordinates of the refraction point must be performed,which has a huge workload and high complexity. An optimized method for solving theposition of refraction point is proposed by improving the approximation method anddecresing the redundancy. The accuracy and effectiveness of the imaging algorithm areboth improved. It turns out that standard BP algorithm suffers from low calcualtionspeed and severe interference in imaging results. In order to accelerate the speed of BPalgorithm, a novel FBP (Fast Back Projection) algorithm is proposed by subaperturedivision and processing in polar coordinate. The FBP algorithm is fast than the originalalgorithm, and it is convenient to perform parallel or real time calculation. In order tosuppress the interference in imaging results, a novel CBP (Cross-correlation BackProjection) algorithm is proposed by using the cross-correlation information of thereceiving data. The CBP algorithm can receive better performance of artifactssuppression.Then the RM (Range Migration) algorithm in frequency domain is investigated.The imaging results of classic RM algorithm will not focus correctly because in layeredmediums the the discontinuity in space results in the discontinuity of the wavefield.Base on the exploding source model for layered mediums, a novel LRM (LayeredRange Migration) algorithm is proposed. The LRM algorithm divides the discontinuous propagation of the wavefield into several local continuous propagation. Not only can theLRM algorithm focus effectively, but also it maintains the computational efficiency ofclassic RM algorithm.After the studies on traditional imaging algorithms in time and frequency domain,the super resolution imaging algorithms based on the theory of spectral estimation areinvestigated. Subjected to the constraint on limit bandwidth, there are lots of artifactslike sidelobes and clutters in the imaging results. Considering that in the field of modernspectral estimation the theory of both RCB (Robust Capon Beamforming) and APES(Amplitude and Phase Estimation of a Sinusoid) have the advantage of high resolution,good artifacts suppression and high robustness, the RCB and APES algorithms areproposed and better imaging results with high resolution and artifacts suppression areachieved. However, the proposed algorithms are computationally ineffective because ofthe calculation of high dimensional matrixes. More efficient algorithms called DR-RCB(Dimensionality Reduced RCB) and DR-APES (Dimensionality Reduced APES) areproposed by pre-beamforming in the sub-image domain and constructing lowdimensional beamforming vectors, that is equivalent to reducing the dimensionality ofcovariance matrix. The DR-RCB and DR-APES algorithms accelerate the RCB andAPES algorithms dramatically, meanwhile, they preserve the high imaging quality.In the end, the CS (Compressive Sensing) algorithm is investigated. The CS theory,which can reconstruct the original sparse signal with very few measurements, breakthrough the traditional framework of Nyquist sampling theory. The relational work suchas data acquisition, coding, decoding and data transmission could be simplified.Combining with the fact that in practical GPR imaging scene the targets are sparsecomparing to the background, the CS imaging algorithm is proposed. The effects ofparameters like the dimension of measurement matrix, SNR (Signal to Noise Ratio),data lacking ratio, compactness of targets on imaging results are systematic analysed. Ithas been proved that the CS algorithm has remarkable ability on imaging performance,artifacts suppression and robustness.