GPR Signal Automatic Identification Method And Engineering Application For Multi-type Defects In Underground Structures

The ground penetrating radar has been widely used in underground engineering detection,because of the advantages such as rapidity,convenience,high resolution and so on.At present,the interpretation of GPR data is mainly based on manual interpretation,which will lead to problems such as low efficiency and unstable precision in the interpretation process.This article focuses on the automatic identification method for multi-type defects in the GPR data,by the methods of theoretical analysis,FDTD forward modeling and on-site measurement,the automatic identification of single target,disease area and structural thickness is achieved through the combination of the wavelet transform,HHT transform,genetic algorithm and SVM algorithm,after the noises in GPR data is filtered by the wavelet domain KL transform.The main work and results are as follows:(1)Based on the theoretical analysis and fusion of the wavelet transform method and the KL transform method,a wavelet domain KL transform method which can be used in GPR data is proposed in this paper.The performance of this method is analyzed by comparison with the wavelet transform method and the KL transform method using diffierent denoising parameters,dealing with a simulated GPR data with known noise content acquired by FDTD simulation method.The test result shows that the wavelet domain KL transform method performs better in both denoising performance and denoising stability that the other two methods,and the noises in GPR data can be effectively screened,which will lead to a higher resolution in the following automatic identification process.(2)The automatic identification method of the single target is studied basing on the hyperbolic characteristics of single target echo signals.Through FK transform,image grayscale gradient search,imgae lining processes,a series of sub-regions containing single target echo signals are extracted from the whole GPR image,as the searching areas of the single target.This process can effectively reduces the size of the search data,meanwhile simplifies the search process.Assuming that the electromagnetic wave propagates uniformly in the medium,and ignoring the influence of the target size,a hyperbolic searching model can be established based on the simplified time-distance formula of the single target reflection.Then based on this hyperbolic model,the high precision automatic identification of single target can be achieved by searching and matching the hyperbolas in sub-regions using the genetic algorithm.(3)Based on the FDTD simulation method,an ideal disease model is constructed to obtain the disease reflections under ideal conditions without interference.Six typical identification features of the disease signal are extracted by theoretical analysis of the disease signals in time domain,frequency domain and time-frequency domain.The high precision automatic identification of the disease area in GPR data is finally proposed by firstly automatically identifies the horizontal distribution of the disease area using SVM,and then automatically identifies the vertical distribution of the disease area by the extraction and analysis of the IMF1 component of the disease signal.The verification results based on the FDTD simulation data and the measured data of the tunnel backfill layer show that this method has a strong detection capability for the disease areas in GPR data,and for the same type of GPR data,the two classification model gained by SVM can be used universally,which will lead to a higher detection efficiency.(4)The automatic identification method of layers in GPR data is studied basing on the analysis of the wave and phase characterastics of the layer echo signals.An ideal layer model and a non-ideal layer model with metals in it are constructed respectively by FDTD method.Through theoretical analysis and verification by the simulated data,the waveform characteristics and instantaneous phase characteristics of the echo signal in the layer position are studied,as well as their change regulation under the disturbance of the rebar reflections.After that,an automatic identification method based on the first-order difference curve of the instantaneous phase is proposed.The performance of this automatic method in practical application is verified by the measured data of the first lining and second lining of a tunnel,and the results show that this method has a strong identify ability for the layers in GPR data.(5)A complete framework for automatic identification of multiple types of defects in underground structures is constructed.Based on the detect data gained in Ciershan No.2 tunnel under 13 survey lines,the comprehensive performance of the above research results in the actual proj ect is evaluated.The number of steel arches,rebars,the disease areas,the thickness of the first lining and the thickness of the second lining in Ciershan No.2 tunnel can be obtained by automatically interpret the detect data gained under different survey lines,and finally,the automatic identification of multi-types of defects in the supporting structure of Ciershan No.2 tunnel is achieved.