KR102519104B1 - A method of collecting GPR image data based on artificial intelligence learning and a method of diagnosing underground structures using it - Google Patents

Abstract

The present invention relates to an underground structure diagnosis method using a GPR image data collection method based on artificial intelligence learning, which improves exploration reliability, comprising: a tenth step of constructing information on various underground structures and a GPR image as a database; a twentieth step of the GPR image; a thirtieth step of selecting information on an underground structure for the selected GPR image; and a fortieth step of extracting and providing the information on the underground structure.

Description

A method of collecting GPR image data based on artificial intelligence learning and a method of diagnosing underground structures using it}

The present invention relates to artificial intelligence learning-based GPR image data collection and underground structure diagnosis. It relates to a method for diagnosing underground structures using a GPR image data collection method.

This section provides background information related to the subject matter of this application and is not necessarily prior art.

In general, structures such as underground facilities, buildings, and bridges undergo deformation over time, so maintenance is required for safety and life extension, and structure inspection, inspection, measurement, and safety judgment are required. .

In addition, there are various ways to inspect the structure, for example, it can be inspected by the GPR exploration technique.

The GPR probe (Ground Penetrating Radar) injects electromagnetic waves from the ground surface or the surface of a structure inward with a TX (Transmitter) antenna, then receives and stores the waves reflected from the medium interface through the RX (Receiver) antenna. Information such as the medium, location, size, and depth of the exploration target is determined through various signal processing processes.

Registered Patent No. 10-2205080 processes GPR raw data collected by a GPR exploration device that generates a signal on the surface of the earth, receives a signal reflected by a material under the surface of the earth, and explores a pipeline located under the surface of the earth. A GPR data processing method driven by a system for processing GPR data provided to a user by: (a) receiving GPR exploration raw data; (b) removing noise and correcting the received GPR exploration raw data; (c) imaging the noise-removed GPR exploration raw data through the following steps, (c-1) generating a plan view by collecting traces of the nth depth of all channels, (c-2) total depth of all channels Generates a cross-sectional view by collecting the m-th tray of (c-3) generates a longitudinal cross-section by collecting all traces of the entire depth of the l-th channel, and (c-4) 3D GPR using the generated plan, cross-section, and longitudinal cross-section generate data; (d) generating a virtual conduit on the 3D GPR data through the following steps, (d-1) generating a first location determined as a conduit in one of the plan view, cross section view, and longitudinal section view, (d-2 ) generating a second position (a position different from the first position) determined as a pipeline in one of the plan view, cross section view, and longitudinal section view, and (d-3) the first position and the second position on the 3D GPR data It is a method for processing GPR data with improved visibility, including generating a virtual conduit connecting the , and (e) outputting the GPR data in which the virtual conduit is generated.

Patent Registration No. 10-2178908 is an analysis system capable of learning a deep learning network to find a pipeline from GPR 3D-data: the location of the first point, the starting point of observation, and the location of the second point, the end point of observation a storage unit in which information and GPR 3D-data collected using a surface penetrating radar for a conduit for which information on a conduit diameter is known are stored; a correct answer data generator for receiving the GPR 3D-data collected from the storage unit and generating correct answer data; And a learning unit for providing the correct answer data to the deep learning network for learning, wherein the correct answer data generator uses information about the location of the conduit of the first and second points in the collected GPR 3D-data After creating a straight line, creating a first plane perpendicular to the generated straight line at the first point and a second plane perpendicular to the generated straight line at the second point, the space between the first and second planes and the generated A pixel of the GPR 3D-data located at the intersection of a cylindrical space created with information about a straight line and a diameter is labeled as a correct answer state, and a pipeline in the GPR 3D-data collected when the correct answer data generator generates a straight line The first and second points of are determined as the first uppermost position and the second uppermost position of the pipe, respectively, and a position lowered by the radius of the pipe from the first uppermost position and the second uppermost position downward in the height direction to the first It is determined by the center and the second center, and the straight line generated is an analysis system for finding a pipeline from GPR 3D-data generated by the equation of a straight line passing through the first center and the second center.

According to the prior art, raw data obtained through GPR exploration is processed in various ways to secure the visibility of the pipeline, and there is a problem of going through a complicated process. can't get it

Registered Patent No. 10-2205080 Registered Patent No. 10-2178908

The present invention is to solve the above-described problems, and provides basic data through artificial intelligence-based learning that collects and matches information on existing underground structures and GPR signals, thereby providing exploration results by comparing basic data and GPR signals. The purpose is to provide a method for diagnosing underground structures using an artificial intelligence learning-based GPR image data collection method that can be identified.

A method for diagnosing underground structures using an artificial intelligence learning-based GPR image data collection method according to the present invention includes the steps of constructing a database of information on various underground structures and GPR images using an artificial intelligence learning-based GPR image data collection method; a twentieth step of acquiring a GPR image while moving the GPR search device; A 30th step of comparing the GPR image obtained in step 20 with the GPR image of the database constructed in step 1, selecting the same GPR image from the database, and selecting information on the underground structure for the selected GPR image; ; A 40th step of extracting and providing information on the underground structure selected through the 3rd step, and the 10th step is the 1st step of collecting pipe diameter, pipe length, pipe direction, and pipe depth as information on various underground structures. , The second step of collecting GPR images by GPR exploration of the underground structure, information on various underground structures and GPR images by matching the information on the underground structure collected in the first step with the GPR image collected in the second step A third step of building a database, and the 40th step is characterized in that the pipe diameter, pipe length, pipe direction, and pipe depth are provided as information on the underground structure selected from the database.

delete

According to the underground structure diagnosis method using the artificial intelligence learning-based GPR image data collection method according to the present invention, the correlation between the GPR image and the underground structure information is derived as a result of learning, and the learning result is used in the GPR exploration of the underground structure to promptly GPR exploration work is possible and has the effect of improving the reliability of exploration.

In the following description of the present invention, if it is determined that a detailed description of a related known function or configuration may unnecessarily obscure the subject matter of the present invention, the detailed description will be omitted. In addition, terms to be described later are terms defined in consideration of functions in the present invention, which may vary according to the intention or custom of a user or operator. Therefore, the definition should be made based on the contents throughout this specification.

The artificial intelligence learning-based GPR image data collection method according to the present invention includes the steps of collecting information on underground structures - collecting GPR images of underground structures - building GPR images of underground structures into a database.

1. Information collection of underground structures.

Data (pipe diameter, pipe length, pipe direction, pipe depth, etc.) are collected as information on underground structures, and this information can be obtained from competent authorities such as competent local governments. In order to build a database, it is desirable to collect various data.

2. Acquisition of GPR images (signals) of underground structures.

The GPR search is performed while moving the GPR search device, and this method is the same as the known method, and the GPR signal is extracted through the search. GPR signals are extracted as hyperbola-type events. GPR images are mainly in binary format or ASCII format, and GPR image data format conversion can be performed to convert them into data formats (PNG, JPG, etc.) that are easy to process and easy to learn and infer.

The specific method of collecting GPR images is as follows.

Since the GPR signal obtained by the GPR search device moving along the road is enormous and very long, it requires segmentation (tiling) into certain lengths.

Subsequently, an event (hyperbola type) for diagnosing an underground structure is extracted from the GPR signal.

After extracting the event, a database is constructed by matching with the information of the underground structure.

Meanwhile, data in the form of continuous multiple image segments is created through the tiling process of the GPR signal. At this time, an event can be shared between two consecutive image segments (formed in the slice area of the image segments), In this case, since events may be distorted due to tiling, it is necessary to perform a more delicate tiling operation.

For example, a GPR image is firstly tiled so that a plurality of image segments having a certain length are continuous, and secondly tiled to generate image segments including tiling portions of the image segments that are consecutive by the first tiling. That is, the size of the second-order tiling area includes the slice boundary of two adjacent first-order tiling areas. In this case, the primary tiling area and the secondary tiling area may have both the same size and different sizes.

Therefore, GPR images are collected targeting both areas sliced by the first tiling and areas sliced by the second tiling.

Meanwhile, although events can be collected without distortion as a result of the first and second tiling, distortion of the GPR signal may occur during the tiling process, and therefore, the reliability of data collection results may be increased by filtering out distorted events. . For example, if the linearity of the collected GPR signal does not satisfy the linearity of the reference curve by comparing the collected GPR signal with the reference curve by selecting the reference curve of the event in advance (need to select various samples) and comparing the collected GPR signal with the reference curve. In this case, the collected GPR signals are deleted and filtered.

3. Build a database of GPR images of underground structures.

A database is built by learning by matching the GPR signals acquired in the previous step, 2. GPR signal collection of underground structures, with the information collected in 1. Information collection of underground structures.

That is, in the database, various GPR signals and information on underground structures for these various GPR signals are matched and stored.

The underground structure diagnosis method using the artificial intelligence learning-based GPR image data collection method according to the present invention is as follows.

1. Build a database.

Through the above-described method, GPR images of underground structures are built into a database.

2. GPR exploration.

A GPR image is acquired while moving the GPR exploration device, and this process is the same as a known exploration process.

3. Comparison of GPR images and selection of underground structure information.

The controller receives the GPR image as a result of the exploration of the GPR exploration device, compares the GPR image of the exploration result with the GPR image built in the database, and selects a matching GPR image from the database.

4. Extraction of underground structure information.

In the database, together with the GPR image, information on the underground structure for this GPR image is managed, and the information on the underground structure is extracted.

The information of the underground structure is processed in various ways such as text and image and displayed on the screen.

Claims (5)

A tenth step of constructing a database of information and GPR images on various underground structures using an artificial intelligence learning-based GPR image data collection method;
a twentieth step of acquiring a GPR image while moving the GPR search device;
A 30th step of comparing the GPR image obtained in step 20 with the GPR image of the database constructed in step 10, selecting the same GPR image from the database, and selecting information on the underground structure for the selected GPR image; ;
A 40th step of extracting and providing information on the underground structure selected through the 30th step,
The tenth step is the first step of collecting the pipe diameter, pipe length, pipe direction, and pipe depth as information on various underground structures, the second step of GPR exploration of the underground structure and the collection of GPR images, and the collection in the first step. A third step of matching the information of one underground structure with the GPR image collected in the second step to build a database of information and GPR images on various underground structures,
The 2nd step is the 2-1st step of tiling the GPR image so that a plurality of image segments having a certain length are continuous. Step 2-2 of extracting a bola-type event, step 2-3 of building a database with the information of the underground structure on the hyperbola-type event extracted through step 2-2,
Step 2-1 performs primary tiling to divide the GPR image into a predetermined length such that a plurality of image segments having a predetermined length are continuous, and includes slice boundaries of two adjacent primary tiling regions divided by the primary tiling. Secondary tiling to create image segments with a size that
The step 40 is an underground structure diagnosis method using an artificial intelligence learning-based GPR image data collection method, characterized in that for providing the pipe diameter, pipe length, pipe direction, and pipe depth as information of the underground structure selected from the database. delete delete The method according to claim 1, wherein step 2-2 compares an event with a reference curve and deletes the event if the event does not satisfy the linearity of the reference curve. Methods for diagnosing structures.
delete