Research On GPR Images Recognition For Subgrade Defects In Ballasted Railroad

At present the total mileage of railways are nearly 112,000 km in China, ballasted railway(Including bridge and tunnel)mileage will reach 96,000 km. GPR become the main technical means of continuous detection for ballasted subgrade. However, detection data is large, processing time is delayed.and recognition of subgrade defects has very strong subjectivity, different discriminant standard, low accuracy.It is urgent for research on recognition method of radar image of ballastless subgrade defects, in order to achieve rapid, accurate recognition of subgrade defects, provide technical support for the subgrade disease treatment.In this paper, with the support of National High-Tech R&D "863" Program (2009AA11Z212) and Hebei province funding research project (11963544D), based on the experimental data and theory study, Daqin railway is taken as the research target. The classification of railway subgrade defects is established. The occurrence mechanism, developmental characteristics and spatial distribution characteristics of typical subgrade diseases (defects) are analyzed. Parameters and identification methods of typical subgrade defects are studied. Rapid recognition methods of railway subgrade defects based on GPR are established.The main conclusions of this paper are summarized below:1. Based on fully understanding of the current classification methods, characterization, basic cause, shape characteristic of typical subgrade defects are analyzed. Combined with the characteristics of GPR technology and radar images, the the classification of GPR subgrade defects based on radar signal characteristics is established, to provide the initial basis for feature extraction and recognition technology of subgrade defects.2. Combined with the radar data of multiple operator lines, the occurrence mechanism and spatial distribution of typical subgrade defects are analyzed, the distribution of the corresponding defects in length and depth are obtained, in order to extract and provide a prior information for the characteristics of defects. According to the technical index of the subgrade detection and the parameters of ground penetrating radar, work parameters of ground penetrating radar and radar antenna configuration are designed, and system is installed in the track inspection train, with 100M and 400M frequency radar antenna, and different depth of the subgrade state is detected, the detection of part of Daqin lines is completed.3. In this paper, the feature extraction of typical radar images in subgrade defects is studied from the aspects of time domain, time-frequency domain and physical geometry. The optimal radar signal feature is established. The characteristics in time domain are determined by the principal component analysis, the frequency and the energy value of the sub block energy, the spatial characteristics.of the lower dimensional space are obtained; two-dimensional time frequency features of the distance and depth of radar image in typical subgrade defects are analyzed, and the low frequency information of the radar image and the energy spectrum of the wavelet multi-scale spatial are acquired;the feature vector of radar signal is extracted by sparse representation, and the vertical and horizontal projection line segments are obtained by projection transformation, which can be used as the basis for judging the types of subgrade defectss, and thus provides a theoretical basis for the image feature extraction.4. Based on the continuity and disorder of the phase axis of the subgrade, the time domain characteristics of the subgrade defects are established by principal component analysis. According to the extracted feature in time-domain, artificial neural network, generalized neural network clustering techniques and support vector machine technology artificial intelligence methods are compared with respectively. Support vector machine is used as the recognition algorithm of subgrade defects, the foundation for the recognition of railway subgrade and its typical defects.5. The physical features of the subgrade defects are analyzed, and the vertical and horizontal projection line segments are obtained by extracting the boundary curve of the effective radar image. And vertical projection line segment and the energy spectrum are used as effective characteristic values of subgrade defects, a novel approach to defects discrimination is proposed.The analysis show that energy spectrum is a quarter of the original data, collecting data is reduced,and recognition speed of subgrade defects is improved by nearly 4 times, HS-SVM method is identified, and subgrade defects recognition rate are all above 85%, the rapid recognition of subgrade defects is achieved.