Railway roadbed disease intelligent identification method based on ground penetrating radar
Technical Field
The invention belongs to the technical field of railway roadbed detection, and relates to a railway roadbed disease intelligent identification method based on a ground penetrating radar.
Background
The detection and evaluation of the railway roadbed state is a key link of railway maintenance and repair. In the existing detection means, Ground Penetrating Radar (GPR) is the most ideal detection method. However, at present, GPR data processing and interpretation at home and abroad mainly depend on manual interpretation, and the efficiency is low and the timeliness is poor. Therefore, a rapid and accurate railway roadbed GPR data processing method is established, and is an urgent problem to be solved in railway roadbed detection in China.
In 2008, research of a small amount of railway roadbed diseases intelligent identification method begins to appear. At present, the research mostly takes railway professional workers as main bodies, the technical backgrounds of machine vision, mode recognition and the like are weak, the considered disease types are few, and main structures such as turnouts, bridges and the like are not considered; taking one-dimensional single-channel radar data as an identification unit, segmenting along the depth direction, and extracting a one-dimensional radar signal characteristic value; finally, the accuracy of pattern recognition is not high, and the method is difficult to be put into practical engineering application.
Disclosure of Invention
The invention aims to provide a railway roadbed disease intelligent identification method based on a ground penetrating radar, and the railway roadbed disease identification is fast, accurate and lossless.
The technical scheme adopted by the invention for solving the technical problems is as follows: a railway roadbed disease intelligent identification method based on a ground penetrating radar specifically comprises the following steps:
(1) intelligent recognition software system for railway roadbed diseases
a) Detecting normal railway subgrades, railway subgrades containing different types of subgrade diseases, railway bridges and turnouts by using a ground penetrating radar, and storing detection data;
b) pretreatment: carrying out zero line correction on the detection data, and converting the detection data into a gray image;
c) two-dimensional dispersion: dividing the obtained ground penetrating radar image into a plurality of identification units along the mileage direction, wherein each identification unit comprises 50-150 channels of data, dividing each identification unit into a plurality of identification subunits along the depth direction, and enabling two adjacent identification subunits to have 50% overlapping areas;
d) feature extraction: extracting each characteristic value by taking the identification subunits as a unit, wherein the characteristic values of all the identification subunits of each identification unit form a characteristic vector of the identification unit, and the initial dimension M = identification subunit number multiplied by characteristic value number of the characteristic vector;
e) and (3) feature dimensionality reduction: determining a dimensionality reduction N, and performing dimensionality reduction on the feature vector by utilizing principal component analysis to construct a low-dimensional feature vector;
f) constructing a recognition model: establishing a support vector machine classifier, inputting a low-dimensional feature vector into the classifier, training the classifier, and constructing a railway roadbed disease intelligent identification model based on a ground penetrating radar;
(2) railway roadbed defect identification
Detecting the railway subgrade to be identified by using a ground penetrating radar, and storing detection data; carrying out preprocessing, two-dimensional dispersion and feature extraction by established intelligent recognition software, and carrying out feature dimension reduction by using a dimension reduction dimension N; and identifying the section of the railway roadbed by using the identification model to obtain the roadbed type of each identification unit of the section of the railway roadbed.
The roadbed diseases comprise slurry pumping, ballast bed dirt, sinking, water containing and cavities.
The characteristic value comprises signal characteristics of the radar image, such as energy and variance; histogram statistical features such as mean, standard deviation, smoothness, third moment, consistency, entropy.
The specific process for determining the dimensionality reduction N is as follows:
(1) setting a series of dimension values within the range of 8-M, respectively utilizing principal component analysis to reduce the dimension of the feature vector, and acquiring a group of data sets after dimension reduction under each dimension value;
(2) reconstructing an original data set by using each group of data sets after dimensionality reduction respectively, and calculating the root mean square error of the reconstructed data set and the original data set;
(3) and selecting the smallest dimension value as the dimension reduction N from the dimension values with the root mean square error smaller than 0.5%.
Advantageous effects
Compared with the background art, the invention has the beneficial effects that:
(1) by utilizing machine vision and mode recognition technology, the rapid, accurate and lossless intelligent recognition of various railway roadbed diseases can be realized;
(2) performing two-dimensional dispersion on a GPR image, extracting various two-dimensional characteristic values such as signal characteristics, histogram statistical characteristics and the like, optimizing the characteristic representation of roadbed diseases, and obviously improving the recognition rate;
(3) the dimensionality reduction is determined by the root mean square error of the reconstructed data set and the original data set, so that the representation performance of the low-dimensional characteristic vector subjected to dimensionality reduction on the original data set can be ensured, the data redundancy is reduced, and the identification efficiency is improved.
Drawings
FIG. 1 is a flow chart of the software system construction of the present invention.
FIG. 2 is a schematic diagram of a two-dimensional discretized process of the present invention.
Detailed Description
The invention is further described with reference to the following figures and embodiments.
A railway roadbed disease intelligent identification method based on a ground penetrating radar comprises two parts of establishing a railway roadbed disease intelligent identification software system and carrying out railway roadbed disease identification.
Firstly, establishing a railway roadbed defect intelligent identification software system.
(1) Detecting normal railway subgrades, railway subgrades containing different types of subgrade diseases, railway bridges and turnouts by using a ground penetrating radar, and storing detection data; the roadbed diseases comprise slurry pumping, ballast bed dirt, sinking, water containing and cavities;
(2) pretreatment: carrying out zero line correction on the detection data, and converting the detection data into a gray image;
(3) two-dimensional dispersion: as shown in fig. 2, the obtained ground penetrating radar image a is equally divided into a plurality of identification units { a ] along the mileage direction1,A2,…,AkDividing each identification unit into a plurality of identification subunits equally along the depth direction, wherein each identification unit comprises 50-150 channels of data, and every two adjacent identification subunits have a 50% overlapping areaDomains, as shown in FIG. 2, identify units A2Is divided into { A21,A22,A23,A24,…};
(4) Feature extraction: extracting characteristic values including signal characteristics of radar images, such as energy and variance, by taking the identification subunit as a unit; histogram statistical features such as mean, standard deviation, smoothness, third moment, consistency, entropy; the characteristic values of all the identification subunits of each identification unit form a characteristic vector of the identification unit, and the initial dimension M = of the characteristic vector is the number of the identification subunits multiplied by the number of the characteristic values;
(5) and (3) feature dimensionality reduction: setting different dimension values { D ] within the range of 8-M1,D2,…,DkPerforming dimensionality reduction on the feature vectors by using principal component analysis respectively, and acquiring a group of dimensionality-reduced data sets under each dimensionality value; reconstructing the original data set by using each group of data sets after dimension reduction, and calculating the root mean square error { E) of the reconstructed data set and the original data set1,E2,…,Ek}; selecting the minimum dimension value as a dimension reduction dimension N from the dimension values with the root mean square error smaller than 0.5%;
(6) constructing a recognition model: and establishing a support vector machine classifier, inputting the low-dimensional feature vector into the classifier, training the classifier, establishing a railway roadbed disease intelligent identification model based on the ground penetrating radar, and completing the establishment of a railway roadbed disease intelligent identification software system.
And secondly, identifying the railway roadbed diseases.
(1) Detecting the railway subgrade to be identified by using a ground penetrating radar, and storing detection data;
(2) carrying out preprocessing, two-dimensional dispersion and feature extraction by established intelligent identification software;
(3) performing feature dimension reduction by using the dimension reduction dimension N to construct a low-dimensional feature vector;
(4) and inputting the low-dimensional feature vector into the identification model, identifying the section of the railway roadbed by using the identification model, and acquiring the roadbed type of each identification unit of the section of the railway roadbed.
The above-mentioned embodiments are provided only for the purpose of the present invention, not for limiting the present invention, and those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, therefore, all equivalent technical solutions should also fall into the scope of the present invention, and should be defined by the claims.