A method for constructing a ballastless track apparent damage sample database

Abstract

This invention discloses a method for constructing a database of apparent damage samples for ballastless tracks, including: acquiring apparent images of ballastless tracks; constructing a coding standard for apparent damage of ballastless tracks; annotating the apparent images of ballastless tracks, and simultaneously encoding the damage in the apparent images of ballastless tracks based on the coding standard, thereby automatically generating legends and attribute files for apparent damage samples of ballastless tracks; and constructing a database of apparent damage samples of ballastless tracks based on the legends and attribute files. This method unifies and standardizes the description of damage to ballastless tracks, efficiently annotates and centrally manages ballastless track damage samples, and facilitates industry application.

Description

Technical Field

[0001] This invention relates to the field of railway track inspection technology, specifically to a method for constructing a database of apparent damage samples for ballastless tracks. Background Technology

[0002] The equipment for ballastless track inspection and monitoring is becoming increasingly sophisticated. Comprehensive inspection trains, comprehensive inspection vehicles, specialized inspection vehicles, and onboard monitoring equipment are becoming increasingly sophisticated. The scale of inspection and monitoring data is enormous, diverse, and characterized by multiple sources and massive amounts.

[0003] As the operating mileage and operating time increase, structural damage to ballastless track lines gradually becomes apparent, and the impact of structural damage inspection and maintenance on operations becomes increasingly prominent.

[0004] There are no clear definitions for the names and descriptions of structural damage during operation, and the results of inspections and data entry vary widely, which brings great difficulties to subsequent information processing, centralized and unified management of data and statistical analysis.

[0005] As the foundation of damage identification, the preliminary data annotation work is also indispensable. In railway traffic scenarios, it is often necessary to annotate relatively small linear damage images in large-size, low-quality damage images. Moreover, these linear damages often do not have the clear and consistent structure and appearance characteristics of natural object images, which also requires a certain level of professional knowledge from the annotators.

[0006] While existing image annotation tools such as Labelme and CVAT can meet the basic requirements for pixel-level annotation, the purely manual annotation method is inefficient for massive amounts of ballastless track appearance image data. Furthermore, common damages and cracks in ballastless tracks typically exhibit linear structures with blurred boundaries, rather than being closed regions composed of a single boundary line. However, the aforementioned annotation tools all employ closed-loop annotation methods without relaxation blur, which are not entirely suitable for annotation work in ballastless track scenarios. As the above analysis shows, pixel-level annotation in such professional scenarios is a very time-consuming and labor-intensive task, and researching efficient professional annotation methods and processes has extremely high practical application value.

[0007] Deep convolutional neural networks based on artificial intelligence technology are increasingly demonstrating their advantages in the field of image processing. For the rapid detection and intelligent semantic recognition of damage conditions such as apparent cracks, gaps, and defects in ballastless tracks, the industry has carried out a number of research projects and accumulated a relatively rich amount of image data. However, there is no formally established high-quality and large-scale damage sample database for ballastless tracks for unified and centralized management to provide support for the field. Summary of the Invention

[0008] To address the aforementioned problems, the purpose of this invention is to provide a method for constructing a database of apparent damage samples for ballastless tracks. This method unifies and standardizes the description of damage to ballastless tracks, efficiently labels and centrally manages damage samples of ballastless tracks, and facilitates industry application.

[0009] The technical solution adopted in this invention is: a method for constructing a database of apparent damage samples for ballastless tracks, comprising the following steps:

[0010] S100: Acquire visual images of the ballastless track;

[0011] S200: Constructing a coding standard for apparent damage to ballastless tracks;

[0012] S300: The apparent image of the ballastless track is annotated, and the damage in the apparent image of the ballastless track is encoded based on the apparent damage coding specification of the ballastless track, thereby automatically generating a sample legend of apparent damage of the ballastless track and its attribute file.

[0013] S400: Construct a database of apparent damage samples for ballastless tracks based on the legends and attribute files of the apparent damage samples.

[0014] Preferably, step S100 further includes image segmentation preprocessing of the acquired apparent image of the ballastless track.

[0015] Preferably, the apparent damage coding specification for ballastless track in step S200 includes four groups: dictionary code, feature code, information code, and image code; the four groups include eight levels: component name, damage type, location information, attribute features, damage level, order information, detection information, and image information; the eight levels include several fields.

[0016] Preferably, the annotation of the apparent image of the ballastless track in step S300 includes:

[0017] The precise location of the damage is marked on the apparent image of the ballastless track, thus obtaining the marked apparent image of the ballastless track.

[0018] Preferably, a linear damage annotation function with adjustable pen width is added to the annotation tool labelme to achieve precise annotation of the damage location.

[0019] Preferably, before annotating the apparent image of the ballastless track in S300, the method further includes automatically pre-identifying damage in the apparent image of the ballastless track to obtain the relative position of the damage in the apparent image of the ballastless track.

[0020] Preferably, the automatic pre-identification of damage in the apparent image of ballastless track includes the following sub-steps:

[0021] S311: Construct a damage detection model based on image patches;

[0022] S312: Filter out background blocks in the apparent image of ballastless track based on the damage detection model;

[0023] S313: Multi-scale fusion segmentation is performed on the apparent image of ballastless track based on image patches containing damage at different scales, thereby obtaining the relative position of damage in the apparent image of ballastless track.

[0024] Preferably, the image patch-based damage detection model is constructed in the following manner:

[0025] The undamaged image is segmented into small image patches and input into a deep convolutional neural network model, enabling the model to learn the feature distribution of the undamaged background image and train to form a damage detection model based on image patches.

[0026] Preferably, the attribute file in step S300 includes a unique coded identifier for each individual damage generated according to the apparent damage coding specification for ballastless track, and the coded identifier is associated with the name of the apparent damage sample legend for ballastless track.

[0027] Preferably, step S400 further includes: periodically refreshing the file at a specified location to synchronize the newly generated apparent damage sample illustrations and their attribute files into the apparent damage sample database of ballastless track, and displaying them on a web page.

[0028] The beneficial effects of the above technical solution are as follows:

[0029] (1) The method for constructing a database of apparent damage samples of ballastless track disclosed in this invention realizes the unified and standardized description of ballastless track damage, efficiently labels ballastless track damage samples and centrally manages them, which is convenient for industry application and has important significance.

[0030] (2) By compiling the apparent damage coding standard for ballastless tracks, this invention can standardize and normalize the damage description, so that after the data is labeled, the relationship between the damage legend and the corresponding code can be generated, which facilitates the damage retrieval and visualization in the database, and forms a unified industry standard.

[0031] (3) The efficient annotation method developed in this invention can automatically generate damage sample legends and their attribute files that meet the above coding specifications, effectively solving the problem of annotation difficulties caused by the large number of damage images and improving annotation efficiency; moreover, this annotation method is more in line with the linear damage annotation requirements of railway scenarios than existing annotation tools, and has better professionalism and security.

[0032] (4) This invention enables multiple users to access database content and perform operations such as retrieval and statistics simultaneously and remotely through the collaborative work of Web server, database server, GPU server, image processing workstation and massive storage device. Users do not need to consider the limitations of the device when accessing through the web page, which is friendly to common field application environments in the industry.

[0033] (5) Based on the dictionary of ballastless track of high-speed railway, this invention proposes a method for constructing a database of apparent damage to ballastless track. The constructed database of apparent damage to ballastless track can not only centrally manage a large number of samples, but also lay the foundation for detailed evaluation of the condition of ballastless track of high-speed railway.

[0034] (6) The deep convolutional neural network based on artificial intelligence technology is increasingly showing its advantages in the field of target pixel-level segmentation. It can build a large-scale and high-quality database of apparent damage to ballastless track, which helps to realize intelligent identification of damage and accurate fusion analysis. It is an important support for building an artificial intelligence-based ballastless track maintenance system. Attached Figure Description

[0035] Figure 1 A flowchart illustrating a method for constructing a database of apparent damage samples for ballastless tracks, as provided in an embodiment of the present invention;

[0036] Figure 2 An example diagram of the apparent damage coding specification for ballastless track provided in one embodiment of the present invention;

[0037] Figure 3 This is a schematic diagram of the original image (the acquired appearance image of the ballastless track) provided for an embodiment of the present invention;

[0038] Figure 4 This is a schematic diagram illustrating an example of apparent damage to a ballastless track according to an embodiment of the present invention.

[0039] Figure 5 A schematic diagram of a property file provided in one embodiment of the present invention;

[0040] Figure 6 This is a schematic diagram of the structure of a management platform for a database of apparent damage samples of ballastless tracks, provided in one embodiment of the present invention. Detailed Implementation

[0041] The embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. The following detailed description of the embodiments and the accompanying drawings are used to illustrate the principles of the present invention by way of example, but should not be used to limit the scope of the present invention. That is, the present invention is not limited to the described preferred embodiments, and the scope of the present invention is defined by the claims.

[0042] In the description of this invention, it should be noted that, unless otherwise stated, "a plurality of" means two or more; the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance; those skilled in the art can understand the specific meaning of the above terms in this invention as appropriate.

[0043] Example 1

[0044] This invention focuses on apparent damage to ballastless tracks, and studies the construction process of an apparent damage database for ballastless tracks based on massive image data; such as... Figure 1 As shown in the figure, this embodiment discloses a method for constructing a database of apparent damage samples for ballastless tracks, including the following steps:

[0045] S100: Acquire visual images of the ballastless track;

[0046] The methods for acquiring surface images of ballastless track for high-speed railways include, but are not limited to, acquiring surface images of ballastless track through one or more of the following methods: integrated inspection vehicle, track flaw detection vehicle, hand-push inspection vehicle, electric-driven inspection vehicle, and manual photography; among which, the acquired surface images of ballastless track can achieve a pixel accuracy of up to 0.2 mm.

[0047] The above-mentioned data collection methods can collect damages such as cracks, gaps, and defects in different track bed types and various components, laying a massive image foundation for formulating damage coding standards and constructing a database of apparent damage samples for ballastless tracks.

[0048] Furthermore, in one embodiment, the acquired apparent image of the ballastless track is subjected to image segmentation preprocessing;

[0049] Because the acquired field images (appearance images of ballastless track) are too large, the requirements for graphics card memory are too high, resulting in low processing efficiency. Therefore, this invention performs image segmentation preprocessing on the acquired appearance images of ballastless track. Based on considerations for the subsequent recognition model, this invention sets the basic size of image segmentation, for example, 1024×1024, 2048×2048, 4096×4096, etc. Such sizes can better utilize the global information of the image while ensuring high operating efficiency.

[0050] For example, image acquisition is carried out on-site using integrated inspection vehicles, track flaw detection vehicles, and hand-push and electric-drive inspection vehicles. The acquired images of the ballastless track are stored on hard drives and brought back, and the information of the acquired images of the ballastless track is recorded. The acquired images of the ballastless track are preprocessed, such as segmentation, in an image processing workstation (Windows or Linux system).

[0051] S200: Constructing a coding standard for apparent damage to ballastless tracks;

[0052] To standardize and normalize the description of apparent damage to ballastless track, this invention describes the various attributes of a certain damage using a unique code. The proposed coding standard for apparent damage to ballastless track includes three parts: track bed, fasteners, and rails. This coding standard may include, for example, four groups, eight levels, and several fields; that is, four groups include eight levels, and eight levels include several fields. Figure 2 As shown, the four groups are, for example, dictionary code, feature code, information code, and image code. Dictionary code includes, for example, component name and damage type; feature code includes, for example, location information, attribute features, damage level, and priority information; information code includes, for example, detection information; and image code includes image information. Thus, the eight levels are component name, damage type, location information, attribute features, damage level, priority information, detection information, and image information. Fields in the component name include, but are not limited to, level 1 component, level 2 component, level 3 component, and level 4 component; fields in the damage type include, but are not limited to, damage category and damage sub-category; fields in the location information include, but are not limited to, bureau name, section name, interval, line name, line type, row type, line mileage, relative mileage, sleeper information, lateral area, track type, trackside, distance from the line centerline, and distance from the track centerline; fields in the attribute features include, but are not limited to, average width. The fields in the damage level include, but are not limited to, damage level, level of attention, whether it is newly added, whether it has developed, whether it has been rectified, and whether it is normal. The fields in the ranking information include, but are not limited to, plate number, plate number, channel number, damage ranking number, sample ranking number, and damage code. The fields in the detection information include, but are not limited to, detection equipment, detection time, and detection personnel. The fields in the image information include, but are not limited to, pixel length, pixel width, pixel depth, pixel scale, image type, image format, image name, XML file name, top left x coordinate, top left y coordinate, damage length, damage width, and damage pixel coordinates.

[0053] The order and number of fields in the above-mentioned ballastless track apparent damage coding specification can be adjusted, and the field names can be replaced with similar expressions. This invention constructs a ballastless track apparent damage coding specification, thereby generating a relationship between damage legends and corresponding codes after the apparent images of ballastless tracks are labeled. This not only facilitates damage retrieval and visualization in the ballastless track apparent damage sample database, but also unifies industry standards and standardizes practical applications.

[0054] S300: Annotate the apparent image of ballastless track, and encode the damage in the apparent image of ballastless track based on the apparent damage coding standard of ballastless track, thereby automatically generating a legend of apparent damage samples of ballastless track and its attribute file;

[0055] (1) The annotator manually annotates the surface image of the ballastless track to mark the precise location of the damage, thereby obtaining the annotated surface image of the ballastless track;

[0056] Unlike traditional segmentation, linear damage, such as apparent cracks in ballastless tracks, typically exhibits a linear structure rather than a closed region defined by a boundary line. Furthermore, general evaluation metrics for linear damage segmentation tasks consider the minuteness of the cracks themselves, incorporating the concept of fuzzy pixels—allowing for a prediction deviation of k pixels. Due to this unique property, traditional pixel-level annotation tools like Labelme, which employ closed-boundary, non-relaxed fuzzy annotation methods, are not entirely suitable for precisely locating linear damage such as cracks in ballastless track scenarios. Therefore, this invention optimizes the data annotation method based on the characteristics of linear damage and adds an adjustable pen width linear damage annotation function to the traditional annotation tool Labelme to achieve precise damage location annotation.

[0057] Compared to traditional annotation methods, the annotation method involved in this invention can adjust the width of the annotation pen by modifying the width parameter W to adapt to the annotation of the precise location of damage of different thicknesses and shapes, which is especially helpful for small cracks and damages; in addition, during the annotation process, a fuzzy pixel selection is provided to make the annotation closer to the actual damage situation; furthermore, the annotation method involved in this invention supports the use of different annotation colors to distinguish the same and different types of damage, so as to facilitate subsequent traceability and modification.

[0058] Furthermore, the precise location of the damage in this invention can also be marked using existing annotation tools such as labelme and CVAT.

[0059] Furthermore, in one embodiment, before manually annotating the apparent image of the ballastless track, unsupervised automatic pre-identification is performed on the damage in the image to be annotated (the acquired apparent image of the ballastless track). This allows the relative position of the damage in the image to be annotated to be obtained before manual annotation, thus assisting in manual annotation. By adopting unsupervised automatic pre-identification, the time spent on manual annotation can be greatly shortened and the difficulty of manual annotation can be reduced.

[0060] The automatic pre-identification of damages such as cracks, gaps, and defects in the surface images of ballastless track based on the damage detection model, in order to initially obtain the relative location of the damages in the images, includes the following steps:

[0061] S311: Construct a damage detection model based on image patches;

[0062] This invention uses a normal background image (i.e., an undamaged image) to train an image patch-based damage detection model based on a deep convolutional neural network. Specifically, the undamaged image is divided into small image patches and input into the deep convolutional neural network model, so that the model can learn the feature distribution of the undamaged background image and train to form an image patch-based damage detection model.

[0063] S312: Filter out background blocks in the apparent damage image of ballastless track based on the damage detection model;

[0064] The presence of damage defects is determined by the distance from the normal feature distribution, thereby filtering out most of the background blocks. In the first stage of damage segmentation, the damage detector filters out most of the background blocks in the damage image, which not only suppresses noise in the image, but also overcomes the influence of different lighting conditions on damage segmentation by filtering out background image blocks.

[0065] S313: Multi-scale fusion segmentation is performed on the apparent image of ballastless track based on image patches containing damage at different scales, so as to initially obtain the relative position of damage in the image to be labeled (apparent image of ballastless track).

[0066] Since feature fusion at different scales has a positive impact on semantic segmentation results, this invention proposes a maximum entropy thresholding method for multi-scale fusion. By setting image blocks of different scales during the segmentation process for multi-scale fusion segmentation, the problem of missing details in large-scale segmentation and noise interference in small-scale segmentation can be alleviated, thereby obtaining and annotating the relative position information of the damage in the image, providing a reference for later annotators.

[0067] The annotator manually corrects the relative positions of the damage obtained initially (manual annotation) to obtain an annotated appearance image of the ballastless track.

[0068] (2) Based on the apparent damage coding standard for ballastless track, each damage in the labeled apparent image of ballastless track is coded, and the apparent damage sample legend and its attribute file of ballastless track are automatically generated;

[0069] The program will automatically generate legends and attribute files for apparent damage samples of ballastless track based on the manually corrected annotation results (annotated appearance images of ballastless track). The attribute files contain a unique code identifier for each individual damage generated according to the coding standard for apparent damage of ballastless track. The code identifier is associated with the name of the legend of apparent damage samples of ballastless track.

[0070] For example, based on the pre-identification results, each damage is classified and labeled at the pixel level. Then, based on the ballastless track apparent damage coding standard, each damage in the labeled ballastless track apparent image is coded, automatically generating ballastless track apparent damage sample legends and their attribute files with sample sizes of 1024×1024, 2048×2048, 4096×4096, rectangles, irregular shapes, etc. Taking a crack as an example, its original image (the acquired ballastless track apparent image) is as follows: Figure 3 As shown, the apparent damage sample image of the ballastless track is as follows: Figure 4 As shown, the attribute file is simplified as follows: Figure 5 As shown.

[0071] S400: Construct a ballastless track apparent damage sample database based on the legend of the ballastless track apparent damage sample and its attribute file, that is, store the ballastless track apparent damage sample legend and its attribute file in a database such as MySQL, Oracle, DB2 or SQL Server to form a ballastless track apparent damage sample database;

[0072] The automatically generated apparent damage sample images and attribute files of ballastless tracks are stored in a designated location on a mass storage device, forming a ballastless track apparent damage sample database. The system periodically refreshes the files in the designated location to synchronize newly generated ballastless track apparent damage sample images and attribute files into the ballastless track apparent damage sample database, and displays them on a web page, allowing authorized users to remotely search, download, and perform statistics.

[0073] The storage of massive image data samples requires the use of an excellent database platform to ensure secure, complete, and rapid storage of the data, facilitating remote and multi-user queries. This invention proposes a database structure framework that includes information such as the apparent damage attributes of ballastless tracks, legend organization, and storage structure. The database framework is designed to support data processing, data statistics and analysis, and efficient remote retrieval and querying of data.

[0074] This invention preferably uses a MySQL database, which offers advantages such as multi-threading, small space footprint, large storage capacity, high security, and ease of use. The apparent damage sample information of ballastless track (illustrations of the apparent damage samples and their attribute files) is stored in the MySQL database. Then, J2EE is used to retrieve the information content through SQL queries and output it to a web server in HTML format for display. Alternatively, data entered by the user in a form can be stored in the MySQL database by executing SQL queries in a J2EE program. Furthermore, other user operations on the web, such as exporting and statistical analysis, can be accepted in the J2EE script, and the data stored in the database can be managed through SQL queries. This invention can also use other server scripting languages ​​besides J2EE, such as J2SE, J2ME, PHP, and Python.

[0075] Example 2

[0076] like Figure 6 As shown, the present invention provides a management platform for a database of apparent damage samples of ballastless tracks, including a massive storage device, an image processing workstation, a GPU server, and a database server;

[0077] Massive storage devices are used to store the collected images of the ballastless track surface, and also to store, quickly transfer, and restore the legends of the apparent damage samples of the ballastless track and their attribute files. Quick transfer and restoration means that the disks in the storage devices can be quickly transferred to other machines by plugging and unplugging. An index is built based on the mileage information to enable fast retrieval and querying.

[0078] The image processing workstation is used to annotate the surface image of the ballastless track to mark the precise location of the damage, thereby obtaining an annotated surface image of the ballastless track;

[0079] Furthermore, in one embodiment, the image processing workstation is also used to perform image segmentation preprocessing on the acquired ballastless track appearance images; that is, the image processing workstation accesses all ballastless track appearance image data through direct-connect storage and deploys ballastless track appearance image processing programs to realize all human-computer interaction processes such as image preprocessing and data annotation.

[0080] The GPU server is used to automatically pre-identify cracks, gaps, and defects in the surface images of ballastless track based on a damage detection model, in order to initially obtain the relative position of the damage in the image. Specifically, the GPU server shares and accesses the acquired surface images of ballastless track stored in a massive storage device via a network, and deploys intelligent recognition algorithms to achieve pre-identification of surface damage of ballastless track. At the same time, it encodes the damage in the surface images of ballastless track based on the aforementioned surface damage coding standard, thereby automatically generating sample legends of surface damage of ballastless track and their attribute files.

[0081] The database server is used to construct a database of apparent damage samples of ballastless track based on the legends and attribute files of the apparent damage samples. Specifically, the database server establishes data tables based on key parameters such as damage type (cracks, gaps, defects, etc.), damage level (A, B, C), and damage magnitude (length, width, area). The data tables are objects used to store specific data and are one of the most important components of the database of apparent damage samples of ballastless track, realizing the centralized storage and management of apparent damage sample information of ballastless track.

[0082] Furthermore, in one embodiment, a web server is also included. The web server is used to set up a B / S architecture to realize functions such as visualization, retrieval, batch export, and statistical analysis of damage samples. That is, it can use J2EE server scripts to retrieve information content from the ballastless track apparent damage sample database through SQL queries and output it to the web server for display in HTML format.

[0083] like Figure 6 As shown, the ballastless track apparent damage sample database management platform proposed in this invention can achieve secure and unified storage, querying, and statistics of ballastless track apparent damage samples by working collaboratively through a web server, database server, GPU server, image processing workstation, and massive storage device.

[0084] Although the invention has been described with reference to preferred embodiments, various modifications can be made and components can be replaced with equivalents without departing from the scope of the invention. In particular, the technical features mentioned in the various embodiments can be combined in any manner, provided there is no structural conflict. The invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims. Parts of the invention not described in detail are well-known to those skilled in the art.

Claims

1. A method for constructing a database of apparent damage samples for ballastless tracks, characterized in that, Includes the following steps: S100: Acquire visual images of the ballastless track; S200: Constructing a coding standard for apparent damage to ballastless tracks; S300: Automatically pre-identify damage in the apparent image of the ballastless track to obtain the relative position of the damage in the apparent image; annotate the apparent image of the ballastless track, and encode the damage in the apparent image of the ballastless track based on the apparent damage coding specification, thereby automatically generating a sample legend of apparent damage of the ballastless track and its attribute file; wherein, the automatic pre-identification of damage in the apparent image of the ballastless track includes the following sub-steps: S311: Construct a damage detection model based on image patches; the damage detection model based on image patches is constructed in the following way: the undamaged image is divided into small image patches and input into a deep convolutional neural network model, so that the model can learn the feature distribution of the undamaged background image and train to form a damage detection model based on image patches. S312: Filter out background blocks in the apparent image of ballastless track based on the damage detection model; S313: Multi-scale fusion segmentation is performed on the apparent image of ballastless track based on image patches containing damage at different scales, thereby obtaining the relative position of damage in the apparent image of ballastless track; S400: Construct a database of apparent damage samples for ballastless tracks based on the legends and attribute files of the apparent damage samples.

2. The construction method according to claim 1, characterized in that, Step S100 also includes image segmentation preprocessing of the acquired apparent image of the ballastless track.

3. The construction method according to claim 1, characterized in that, The apparent damage coding standard for ballastless track in step S200 includes four groups: dictionary code, feature code, information code, and image code; the four groups include eight levels: component name, damage type, location information, attribute features, damage level, sequence information, detection information, and image information; the eight levels include several fields.

4. The construction method according to claim 1, characterized in that, The annotation of the apparent image of the ballastless track in step S300 includes: The precise location of the damage is marked on the apparent image of the ballastless track, thus obtaining the marked apparent image of the ballastless track.

5. The construction method according to claim 4, characterized in that, By adding a linear damage annotation function with adjustable brush width to the annotation tool labelme, the precise location of damage can be annotated.

6. The construction method according to claim 1, characterized in that, The attribute file in step S300 includes a unique coded identifier for each individual damage generated according to the apparent damage coding specification for ballastless track, and the coded identifier is associated with the name of the apparent damage sample legend for ballastless track.

7. The construction method according to claim 1, characterized in that, Step S400 further includes: periodically refreshing the files at a specified location to synchronize the newly generated apparent damage sample illustrations and their attribute files into the apparent damage sample database of ballastless track, and displaying them on a web page.

Images