Research On Detecting Underground Cavities Using Ground Penetrating Radar

Ground penetrating radar(GPR) is capable of fast, efficient, non-destructive underground target detection, and becomes preferred technology for urban roads diseases census. However, the underground environment is so complex that target detection becomes environmentally sensitive. Underground cavities exhibit large morphological differences, which add to the difficult of the underground cavity target detection and identification, so that the method relies on human interpretation often leads to false alarm or missing alarm. Also, the field of experimental conditions are often difficult to meet, obtaining subsurface cavities’ echo through a large database of experimental method is not feasible. To solve the above problems, this paper first obtained a variety of cavities’ GPR data by forward simulation and analyzed the reflection characteristics of those simulated cavities. Furthermore, the paper studied the ground-penetrating radar data clutter suppression, subsurface cavity detection and recognition problems, accumulating data samples for target detection, providing subsurface cavity detection and identification methods, presenting a strong basis of correct interpretation of cavity target from a large number of collected GPR data. The main contents are summarized as follows:(1) For impulse ground penetrating radar system, forward simulation of subsurface cavities was studied based on FDTD method. A variety of two-dimensional GPR images of subsurface cavities of different shapes were obtained by GPRMax2 D, a simulation software. The influences of shape, size and burial depth of cavities on reflection characteristics were also analyzed. All these work accumulates experimental data and experience for the subsequent target detection and recognition.(2) For air-ground direct reflection and background clutter, three clutter suppression methods were studies, namely mean subtraction, principal component analysis(PCA) and robust principal component analysis(RPCA). Noticing that the mean subtraction method is only applicable to uniform clutter and principal component analysis shows poor robustness of strong clutter, robust principal component analysis method is proposed for clutter suppression. Experimental comparison proved that robust principal component analysis giv es the best SCR improvement of ground penetrating radar B-scan.(3) GPR detection algorithm is proposed based on least squares estimation. Current A-scan is estimated using target-free reference signal previously observed, and the best estimate is obtained in the sense of minimum sum of squared estimation error. The sum of squared estimation error is used as a test statistic to compare with a fixed threshold and whether the presence of a target if finally determined. Experimental analysis shows the reliability of the detection algorithm.(4) Feature extraction problem of GPR signal is studied in the time domain, frequency domain and wavelet domain, respectively. Through qualitative and quantitative analysis, we summed up the differences between signals incl uding a cavity echo and that without. Finally, we extracted A-scan in time-domain, energy density spectrum in frequency domain and wavelet packet energy spectrum as features.(5) A GPR subsurface cavity recognition method is proposed based on support vector machine. Recognition experiments were carried out using the extracted three features as mentioned above. Simulation and real data experiments have proved the effectiveness of the subsurface cavity identification method. Cavit ies can be successfully recognized using any of the three features. And we found that feature in frequency domain and wavelet domain is of better stability than that of the time-domain feature.