Study On Processing Software Of GPR Data From Railroad Trackbed Inspection

The existence of sleepers, random noise and the propagation attenuation of electromagnetic wave affect the authenticity and reliability of GPR image and disturb our explanation for it, when GPR is used to inspect for railroad trackbed. And the following heavy data processing needs much time and effort, which is not conductive to GPR spreading and applying in railroad trackbed inspection. Facing the lack of function of GPR supporting software at present, an automatic interpretation algorithm adapted for GPR image from railroad trackbed inspection is designed to meet mass data analysis demands.First, the reason for GPR data distortion and the results of GPR’s application in some fields were analyzed. The data processing method adapted for GPR detection of railroad trackbed was designed to reduce the effects. Parabolic Radon transform was used for eliminating the effect from railway sleepers. KL transform in wavelet domain was applied for removing random noise from railroad facilities and communications. High-passing filtering and instantaneous phase were chosen to improve the resolution of deep and thin layers. Then an interface-test algorithm was designed according to the characteristic of reflection echo form interfaces, the information of interface points from all traces data was extracted, an interface classification principle was proposed to correspond with the phenomenon of rupture, skip, fork and intersection interface, and the interfaces were intelligently traced. Next segmented energy, variance and interface for eigenvalue from GPR data were extracted. The eigenvalue could not only distinguish different kinds of defects, but also value extent of development of defects. Finally an expert system was established. All kinds of trackbed defects in different localities were stored in defect library and studied by learning vector quantization network model to get classification rules by which the trackbed defect types were judged. The algorithms were programmed by a hybrid code of VC and MATLAB, and the software was named RAILGPR1.0.Mass GPR data for Shanghai-Nanjing railroad trackbed was processed by using this software. The results indicate that the algorithm is good for the characteristics of high-speed computation, general-purpose, and less-parameters. After the processing, the GPR image shows strong signals in depth, clear interfaces, and defects positions. The software can automatically distinguish the defect types and define distances of them. The model’s recognition rate of trackbed defects is up to 90%.