A pipeline signal recognition and direction calculation method

Through deep learning and signal processing technology, combined with energy density window method and three-dimensional matrix rotation, signal strengthening and identifying ground-penetrating radar data of water supply and drainage pipelines is solved, and the problem of low recognition accuracy and automation in the existing technology is achieved. Accurate identification and directional calculation of pipes made of various materials are suitable for complex underground environments.

CN118465757BActive Publication Date: 2024-11-15ZHEJIANG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410912638.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-09
Publication Date
2024-11-15
Estimated Expiration
2044-07-09

AI Technical Summary

Technical Problem

The existing technology lacks identification methods for different materials in the identification and direction calculation of ground-penetrating radar data in water supply and drainage pipelines. It relies on manual experience, has low processing efficiency and unstable processing, and has a lot of noise interference in underground environments, resulting in low recognition accuracy and automation. .

Method used

Using deep learning combined with signal processing, the three-dimensional radar data is acquired for array processing and data enhancement, and the pipeline identification model R3D for three-dimensional radar data is used for signal enhancement and identification, combining energy density window method and three-dimensional matrix rotation to achieve the Accurate identification and direction calculation of pipes of various materials.

Benefits of technology

It improves the accuracy and automation of pipeline signal recognition, reduces noise interference, and realizes accurate identification and directional calculation of pipelines of different materials. It is suitable for complex underground environments and improves the efficiency of urban infrastructure construction and management.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118465757B_ABST
    Figure CN118465757B_ABST
Patent Text Reader

Abstract

The present invention belongs to the technical field of water supply and drainage pipeline detection, and provides a pipeline signal recognition and direction calculation method, including: obtaining three-dimensional radar data of buried pipelines; determining the size and movement overlap rate of the three-dimensional radar data array block, arraying the three-dimensional radar data to form an array block data set, and manually marking the array block data set; performing signal enhancement and data enhancement on the array block data set; inputting the array block data set after signal enhancement and data enhancement into a pipeline recognition model trained for three-dimensional radar data; according to the pipeline recognition result of the pipeline recognition model, dyeing the array block data set and using the dyed array block to calculate the pipeline direction. It can realize accurate recognition of pipelines of various materials, reduce the interference of underground environmental noise characteristics, automatically and accurately calculate the direction of hidden pipelines, and facilitate construction and excavation; and the processing efficiency is high, providing reliable technical support for urban infrastructure construction and management.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of water supply and drainage pipeline detection, and in particular relates to a pipeline signal recognition and direction calculation method. Background Art

[0002] The accurate positioning of underground pipelines, represented by water supply and drainage pipes, plays a vital role in urban construction, infrastructure maintenance and avoiding engineering accidents. Although traditional underground pipeline detection methods, such as consulting construction materials, on-site excavation and audio-listening methods, have certain reliability, they often require a lot of time and manpower costs, and are less efficient in complex terrain or areas with dense underground pipelines. In recent years, radar technology has been widely used in the field of underground pipeline detection. As a non-destructive underground exploration technology, ground penetrating radar can quickly obtain the location, depth and morphology information of underground pipelines by emitting high-frequency electromagnetic waves and receiving their reflected signals. It has the advantages of high resolution and rapid data acquisition. The basic principle of ground penetrating radar technology is to use the characteristics of electromagnetic waves propagating and reflecting in underground media, and to infer the location and nature of underground pipelines by analyzing parameters such as the intensity, time delay and phase of the reflected signal.

[0003] However, the identification and direction calculation of ground penetrating radar data signals of water supply and drainage pipelines remains a challenging problem. First, pipelines are usually presented in the form of hyperbolic signals in ground penetrating radar data. Water supply and drainage pipelines made of different materials (such as metal pipes and non-metallic pipes) may show different signal characteristics, such as the real semi-axis length, vertex position, and focus position of the hyperbola, which requires targeted design of data sets or signal recognition algorithms; secondly, the complexity of the underground environment and the existence of noise interference make the data processing process more complicated, requiring effective data preprocessing and noise suppression methods. In addition, existing data signal identification and direction calculation methods often rely on manual experience or simple rules, with low intelligence and automation, resulting in low processing efficiency and unstable results.

[0004] At present, some research has been devoted to solving these problems, such as signal recognition algorithms based on machine learning, improved feature extraction methods, and intelligent data processing processes. However, existing methods are often limited to specific application scenarios or underground pipeline types, lacking versatility and robustness.

[0005] In summary, the problems of the existing technology are: lack of identification methods for signal characteristics of water supply and drainage pipelines of different materials; lack of data preprocessing and noise suppression methods for complex underground environments; existing pipeline direction calculation methods rely on manual experience or simple rules; the cost of collecting a large amount of data on pipelines of different materials is high; there are many interference targets and noise in the underground environment; and it is difficult to design pipeline direction calculation rules. Therefore, it is necessary to further explore new data processing methods and algorithm models to improve the accuracy and automation of ground penetrating radar data signal recognition and direction calculation of water supply and drainage pipelines to meet the needs of urban infrastructure construction and pipeline maintenance. Summary of the invention

[0006] The present invention aims to address the difficulties and challenges of ground penetrating radar data signal recognition and direction calculation for water supply and drainage pipelines, and proposes a new method by combining deep learning, signal processing and underground exploration technology to improve the existing data processing process, improve recognition accuracy and processing efficiency, and provide reliable technical support for urban infrastructure construction and management.

[0007] In order to achieve the above object, the present invention provides the following technical solutions:

[0008] The present invention provides a pipeline signal recognition and direction calculation method, comprising:

[0009] (1) Obtaining 3D radar data of buried pipelines;

[0010] (2) determining the size and movement overlap ratio of the 3D radar data array block, performing array processing (cutting) on ​​the 3D radar data according to the array block size to form an array block data set, and manually labeling the array block data set;

[0011] (3) performing signal enhancement on the array block data set;

[0012] (4) Perform data enhancement on the array block dataset;

[0013] (5) inputting the array block data set after signal enhancement and data augmentation into a pipeline recognition model R3D trained on 3D radar data;

[0014] (6) According to the pipeline recognition result of the pipeline recognition model R3D, the array block data set is dyed and the pipeline direction is calculated using the dyed array blocks.

[0015] Further, the size of the array block includes an actual size and an array size;

[0016] The actual size of the array block is determined according to the characteristics of the pipeline hyperbolic signal, which is expressed as:

[0017]

[0018] In the formula, x is the horizontal coordinate of the hyperbola, t is the vertical coordinate of the hyperbola, and x 0 is the horizontal coordinate of the vertex of the hyperbola, t 0 is the ordinate at the vertex of the hyperbola, v is the radar wave speed in the soil medium, R 0 is the distance from the center of the pipeline to the pipe wall when the radar survey line is not perpendicular to the pipeline, R is the radius of the pipeline, and c is the focus of the hyperbola;

[0019] The actual size of the array block in the X-axis and Y-axis directions in the horizontal plane includes at least half of the pipeline hyperbolic signal. Therefore, the actual size of the array block in the X-axis and Y-axis directions is x AB and AB for:

[0020]

[0021] Where, h is the buried depth of pipeline;

[0022] The actual size z of the array block in the depth Z axis direction AB for:

[0023]

[0024] Where D is the pipe diameter;

[0025] The array size T×N×S of the array block in the three directions of XYZ is:

[0026]

[0027] In the formula, Δx is the radar moving step length; Δy is the radar line spacing; Δt is the radar sampling point time interval. According to the principle that the size of the array block in the X and Y axis directions is equal to half of the pipeline hyperbolic signal, the moving overlap rate of the array block is set to 50%.

[0028] Furthermore, the manually labeled categories include background, metal pipes, non-metal pipes and cables.

[0029] Furthermore, the array block data set is enhanced by the energy density window method, and the energy distribution of the data points on the section at a certain time is statistically analyzed to describe the strength distribution trend of the radar signal. Specifically, the B-Scan diagram (two-dimensional profile diagram) in the XZ domain collected by the radar is composed of several waveform data recording the change of radar signal amplitude over time. The original section diagram at each moment is the distribution diagram of the radar wave amplitude value at that moment. The energy density formula is applied to the horizontal section C-Scan diagram (three-dimensional diagram) in the XY domain of the array block:

[0030]

[0031] In the formula, and are the amplitude value and energy density value of the ith sampling point of the jth radar A-Scan signal in the kth radar B-Scan image in the 3D radar data, respectively; C is the window width of the energy density window, and is the number of C-Scan layers included in the calculation range;

[0032] Apply the energy density formula to each horizontal slice on the XY domain of the array block to obtain the energy density C e section:

[0033]

[0034] Where T is the number of radar channels and N is the number of radar survey lines.

[0035] Furthermore, the three-dimensional radar data is enhanced by using three-dimensional matrix rotation, specifically: the three-dimensional linear transformation matrix M is multiplied by the homogeneous coordinates of the three-dimensional radar data for transformation. The specific calculation formula is:

[0036]

[0037] Where [x, y, z, 1] is the homogeneous coordinate system representation of the original 3D radar data. It is the homogeneous coordinate system representation of the transformed 3D radar data. When the 3D radar data rotates positively around the x, y, and z axes by an angle β, the three components of M are:

[0038]

[0039] In order to make the rotation center of the three-dimensional matrix the center position of the three-dimensional radar data, the rotation center needs to be translated. The transformation matrix is:

[0040]

[0041] The final rotated three-dimensional radar data is expressed as:

[0042]

[0043] Furthermore, the pipeline recognition model R3D trained for three-dimensional radar data is established by modifying the number of convolutional layers and output categories of the C3D model and adding a residual module. After the modification, the number of convolutional layers of the C3D model is 5, the number of pooling layers is 5, the number of fully connected layers is 2, and the output categories are 4. The residual module is arranged on the 2nd to 4th convolutional layers.

[0044] Furthermore, the process of calculating the pipeline direction using the colored array blocks is:

[0045] Project the colored array block onto the plane, connect the midpoints of the array blocks at the start and end points, and the angle θ between the pipeline direction and the positive direction of the X-axis. When θ is not 90° or 0°:

[0046]

[0047] Where n y and n x Respectively represent the number of array blocks passed between the calculation starting point and the end point in the X and Y directions;

[0048] x AB and AB It is the actual size of the array block in the X-axis and Y-axis directions in the horizontal plane.

[0049] The present invention has the following beneficial effects:

[0050] (1) The present invention performs array preprocessing and category labeling on pipeline radar data, and uses the energy density window method to enhance pipeline signal characteristics and suppress interference signal characteristics; uses three-dimensional matrix rotation to perform data enhancement, and constructs an R3D model to input the data set for training and verification; the array data blocks are colored according to the recognition results, and the true direction of the pipeline is calculated, so that pipelines of various materials can be accurately identified, the interference of underground environmental noise characteristics can be reduced, and the direction of hidden pipelines can be automatically and accurately calculated, which is convenient for construction and excavation; and the processing efficiency is high, providing reliable technical support for urban infrastructure construction and management.

[0051] (2) The present invention can identify pipelines of different materials and accurately calculate their directions. The intelligent method has strong robustness and good generalization ability and is suitable for engineering applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a flow chart of the calculation method of the present invention.

[0053] Figure 2 It is a visualization flow chart of the calculation method of the present invention.

[0054] Figure 3 It is a schematic diagram of the result after the array block data set is enhanced in the embodiment.

[0055] Figure 4 is a network structure diagram of the R3D model in the embodiment.

[0056] Figure 5 These are typical sample diagrams of pipelines of various materials in the embodiments.

[0057] Figure 6 are the training and validation curves of the R3D model in the embodiment.

[0058] Figure 7 It is a pipeline direction calculation result diagram given based on the R3D model recognition result in the embodiment. DETAILED DESCRIPTION

[0059] The specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings. It should be pointed out that the embodiments are only specific explanations of the invention and should not be regarded as limitations of the invention. The purpose of the embodiments is to enable those skilled in the art to better understand and reproduce the technical solutions of the present invention. The protection scope of the present invention shall still be based on the scope defined by the claims.

[0060] like Figure 1-Figure 2 As shown, the present invention provides a pipeline signal recognition and direction calculation method, comprising:

[0061] S1, using ground penetrating radar and simulation software to obtain three-dimensional radar data of buried pipelines;

[0062] Specifically, in the pipeline leakage test, ground penetrating radar is used to collect data on buried pipelines, where the materials of buried pipelines include metal pipes, non-metallic pipes and cables, and the buried depth, length and spacing of the pipelines are uniform. In the simulation software, a three-dimensional soil environment is randomly generated, and small cavities, water-containing cavities and stones are added to make the pipeline imaging close to the real working conditions. From top to bottom, the first layer is air, the second layer, the third layer, and the fourth layer are three-dimensional soil media with water, air and soil with different water contents randomly distributed. As the depth increases, the amount of air in the soil decreases and the water content increases. The fourth layer is composed of metal pipes, non-metallic pipes, cables, cavities, water-containing cavities and stone subsoil backgrounds.

[0063] S2, determining the size and movement overlap rate of the three-dimensional radar data array block, performing array processing (cutting) on ​​the three-dimensional radar data according to the array block size to form an array block data set, and manually labeling the array block data set;

[0064] The size of the array block includes an actual size and an array size;

[0065] The actual size of the array block is determined according to the characteristics of the pipeline hyperbolic signal, which is expressed as:

[0066]

[0067] In the formula, x is the horizontal coordinate of the hyperbola, t is the vertical coordinate of the hyperbola, and x 0 is the horizontal coordinate of the vertex of the hyperbola, t 0 is the ordinate at the vertex of the hyperbola, v is the radar wave speed in the soil medium, R 0 is the distance from the center of the pipeline to the pipe wall when the radar survey line is not perpendicular to the pipeline, R is the radius of the pipeline, and c is the focus of the hyperbola;

[0068] The actual size of the array block in the X-axis and Y-axis directions in the horizontal plane includes at least half of the pipeline hyperbolic signal. Therefore, the actual size of the array block in the X-axis and Y-axis directions is x AB and AB for:

[0069]

[0070] Where, h is the buried depth of pipeline;

[0071] The actual size z of the array block in the depth Z axis direction AB for:

[0072]

[0073] Where D is the pipe diameter;

[0074] The array size T×N×S of the array block in the three directions of XYZ is:

[0075]

[0076] In the formula, Δx is the radar moving step length; Δy is the radar line spacing; Δt is the radar sampling point time interval. According to the principle that the size of the array block in the X and Y axis directions is equal to half of the pipeline hyperbolic signal, the moving overlap rate of the array block is set to 50%.

[0077] Preferably, the manually labeled categories include background, metal pipes, non-metal pipes and cables.

[0078] S3, signal enhancement of array block dataset;

[0079] The array block data set is enhanced by the energy density window method, and the energy distribution of the data points on the slice at a certain time is statistically analyzed to describe the strength distribution trend of the radar signal. Specifically, the B-Scan image on the XZ domain collected by the radar consists of several waveform data that record the change of the radar signal amplitude over time. The original slice image at each moment is the distribution map of the radar wave amplitude value at that moment. The energy density formula is applied to the horizontal slice C-Scan image on the XY domain of the array block:

[0080]

[0081] In the formula, and are the amplitude value and energy density value of the ith sampling point of the jth radar A-Scan signal in the kth radar B-Scan image in the 3D radar data, respectively; C is the window width of the energy density window, and is the number of C-Scan layers included in the calculation range;

[0082] Apply the energy density formula to each horizontal slice on the XY domain of the array block to obtain the energy density C e section:

[0083]

[0084] Where T is the number of radar channels and N is the number of radar survey lines.

[0085] S4, data enhancement is performed on the array block data set; data enhancement is performed on the three-dimensional radar data using three-dimensional matrix rotation, specifically: transformation is performed by multiplying the three-dimensional linear transformation matrix M with the homogeneous coordinates of the three-dimensional radar data. The specific calculation formula is:

[0086]

[0087] Where [x, y, z, 1] is the homogeneous coordinate system representation of the original 3D radar data. It is the homogeneous coordinate system representation of the transformed 3D radar data. When the 3D radar data rotates positively around the x, y, and z axes by an angle β, the three components of M are:

[0088]

[0089] In order to make the rotation center of the three-dimensional matrix the center position of the three-dimensional radar data, the rotation center needs to be translated. The transformation matrix is:

[0090]

[0091] The final rotated three-dimensional radar data is expressed as:

[0092]

[0093] Since the sizes of the three-dimensional data array blocks in the three directions may not be equal, data gaps may occur after rotation. When gaps appear in the rotated data matrix, bilinear interpolation is used to fill the gaps, that is, the gaps in the rotated data matrix f(Q 1 ) can be used with the help of the data of the four points near the position of the original matrix f(Q 11 ), f(Q 21 ), f(Q 12 ) and f(Q 22 )get:

[0094]

[0095] in,

[0096]

[0097] S5, inputting the array block data set after signal enhancement and data enhancement into a pipeline recognition model R3D trained for three-dimensional radar data; training the R3D model on the three-dimensional data array block data set with labeled categories to obtain a deep learning model for pipeline signals of different materials,

[0098] The pipeline recognition model R3D trained for three-dimensional radar data is established by modifying the number of convolutional layers and output categories of the C3D model and adding residual modules. After the modification, the number of convolutional layers of the C3D model is 5, the number of pooling layers is 5, the number of fully connected layers is 2, and the output categories are 4. The residual modules are arranged on the 2nd to 4th convolutional layers.

[0099] S6, dyeing the array block data set according to the pipeline recognition result of the pipeline recognition model R3D and calculating the pipeline direction using the dyed array blocks;

[0100] The process of calculating the pipeline direction using the colored array blocks is:

[0101] Project the colored array block onto the plane, connect the midpoints of the array blocks at the start and end points, and the angle θ between the pipeline direction and the positive direction of the X-axis. When θ is not 90° or 0°:

[0102]

[0103] Where n y and n x Respectively represent the number of array blocks passed between the calculation starting point and the end point in the X and Y directions;

[0104] x AB and AB It is the actual size of the array block in the X-axis and Y-axis directions in the horizontal plane.

[0105] Example

[0106] S1, using ground penetrating radar and simulation software to obtain three-dimensional radar data of buried pipelines;

[0107] Specifically, in the pipeline leakage test, ground penetrating radar is used to collect data on buried pipelines, where the materials of buried pipelines include metal pipes, non-metallic pipes and cables, and the buried depth, length and spacing of the pipelines are uniform. In the simulation software, a three-dimensional soil environment is randomly generated, and small cavities, water-containing cavities and stones are added to make the pipeline imaging close to the real working conditions. From top to bottom, the first layer is air, the second layer, the third layer, and the fourth layer are three-dimensional soil media with water, air and soil with different water contents randomly distributed. As the depth increases, the amount of air in the soil decreases and the water content increases. The fourth layer is composed of metal pipes, non-metallic pipes, cables, cavities, water-containing cavities and stone subsoil backgrounds.

[0108] There are 1,400 sets of data collected at the test site and 792 sets of data obtained by simulation. The above two types of data are used as training sets and validation sets, and an additional 280 sets of real data are collected at the engineering site as test sets.

[0109] S2, determining the size and movement overlap rate of the three-dimensional radar data array block, array-cutting the three-dimensional radar data according to the array block size to form an array block data set, and manually annotating the array block data set; the manually annotated categories include background, metal pipes, non-metal pipes and cables.

[0110] Among them, the pipeline burial depth is 100cm, the pipeline diameter is 70cm, the radar moving step length Δx is 1cm, the radar line spacing Δy is 10cm, the radar wave velocity v in the soil medium is 10cm / ns, and the radar sampling point time interval Δt is 1 / (7035*10^6)s.

[0111] The size of the array block includes an actual size and an array size;

[0112] The actual size of the array block in the X-axis and Y-axis directions in the horizontal plane includes at least half of the pipeline hyperbolic signal. Therefore, the actual size of the array block in the X-axis and Y-axis directions is x AB and AB for:

[0113]

[0114] The actual size z of the array block in the depth Z axis direction AB for:

[0115]

[0116] Where D is the pipe diameter;

[0117] The array size T×N×S of the array block in the three directions of XYZ is:

[0118]

[0119] The mobile overlap is 50%.

[0120] S3, the signal enhancement is performed on the array block data set. The result of signal enhancement is shown in Figure 3 As shown,

[0121] The array block data set is enhanced by the energy density window method, and the energy distribution of the data points on the slice at a certain time is statistically analyzed to describe the strength distribution trend of the radar signal. Specifically, the B-Scan image on the XZ domain collected by the radar consists of several waveform data that record the change of the radar signal amplitude over time. The original slice image at each moment is the distribution map of the radar wave amplitude value at that moment. The energy density formula is applied to the horizontal slice C-Scan image on the XY domain of the array block:

[0122]

[0123] In the formula, and are the amplitude value and energy density value of the ith sampling point of the jth radar A-Scan signal in the kth radar B-Scan image in the 3D radar data, respectively; C is the window width of the energy density window, which is the number of C-Scan layers included in the calculation range, and is taken as 10;

[0124] Apply the energy density formula to each horizontal slice on the XY domain of the array block to obtain the energy density C e section:

[0125]

[0126] Where T is the number of radar channels and N is the number of radar survey lines.

[0127] S4, data enhancement is performed on the array block data set; data enhancement is performed on the three-dimensional radar data using three-dimensional matrix rotation, specifically: transformation is performed by multiplying the three-dimensional linear transformation matrix M with the homogeneous coordinates of the three-dimensional radar data. The specific calculation formula is:

[0128]

[0129] Where [x, y, z, 1] is the homogeneous coordinate system representation of the original 3D radar data. It is the homogeneous coordinate system representation of the transformed 3D radar data. When the 3D radar data rotates positively around the x, y, and z axes by an angle β, the three components of M are:

[0130]

[0131] In order to make the rotation center of the three-dimensional matrix the center position of the three-dimensional radar data, the rotation center needs to be translated. The transformation matrix is:

[0132]

[0133] The final rotated three-dimensional radar data is expressed as:

[0134]

[0135] Since the sizes of the three-dimensional data array blocks in the three directions may not be equal, data gaps may occur after rotation. When gaps appear in the rotated data matrix, bilinear interpolation is used to fill the gaps, that is, the gaps in the rotated data matrix f(Q 1 ) can be used with the help of the data of the four points near the position of the original matrix f(Q 11 ), f(Q 21 ), f(Q 12 ) and f(Q 22 )get:

[0136]

[0137] in,

[0138]

[0139] S5, input the array block data set after signal enhancement and data enhancement into a pipeline recognition model R3D trained for 3D radar data; train the R3D model on the 3D data array block data set with labeled categories to obtain a deep learning model for pipeline signals of different materials. Typical samples of pipelines of various materials are as follows Figure 5 shown.

[0140] like Figure 4 As shown, the pipeline recognition model R3D is established by modifying the number of convolutional layers and output categories of the C3D model and adding a residual module. After the modification, the number of convolutional layers of the C3D model is 5, the number of pooling layers is 5, the number of fully connected layers is 2, and the output categories are 4. The residual module is arranged on the 2nd to 4th convolutional layers.

[0141] like Figure 6 As shown in the figure, after training, the training and validation curves of the R3D model, including the loss value change curve and the accuracy change curve. The R3D model converged quickly in 100 iterations, and the final accuracy was close to 1.0, performing well on the training set and validation set.

[0142] S6, according to the pipeline recognition result of the pipeline recognition model R3D, the array block data set is dyed and the pipeline direction is calculated using the dyed array blocks, such as Figure 7 As shown;

[0143] Specifically, the R3D model was tested on the 280 sets of real field data. The array blocks were colored according to the pipeline identification results, the colored array blocks were projected in the plane, and the midpoints of the array blocks at the starting point and the end point were connected. y and n x Taking the values ​​of 8 and 1 respectively, the angle θ between the pipeline direction and the positive direction of the X-axis is:

[0144]

[0145] Through actual comparison based on the pipeline direction, the absolute error with the actual angle is 5°, and the relative error is 6.41%, which is within the allowable error range. Therefore, the present invention can accurately identify pipelines of various materials and provide reliable technical support for urban infrastructure construction and management.

[0146] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0147] It should be noted that the technical features not described in detail in the present invention can be implemented by any existing technology.

Claims

1. A pipeline signal recognition and direction calculation method, characterized in that: include: (1) Obtaining 3D radar data of buried pipelines; (2) determining the size and movement overlap ratio of the 3D radar data array block, performing array processing on the 3D radar data according to the array block size to form an array block data set, and manually labeling the array block data set; (3) performing signal enhancement on the array block data set; The array block data set is enhanced by the energy density window method, and the strength distribution trend of the radar signal is described by statistically analyzing the energy distribution of data points on the slice at a certain time. Specifically, the B-Scan graph on the XZ domain collected by the radar is composed of several waveform data recording the amplitude change of the radar signal over time. The original slice graph at each moment is the distribution graph of the radar wave amplitude value at that moment. The energy density formula is applied to the horizontal slice C-Scan graph on the XY domain of the array block: In the formula, and are the amplitude value and energy density value of the i-th sampling point of the j-th radar A-Scan signal in the k-th radar B-Scan image in the three-dimensional radar data, respectively; C is the window width of the energy density window, is the number of C-Scan layers included in the calculation range, T is the number of radar channels, and N is the number of radar survey lines; Apply the energy density formula to each horizontal slice on the XY domain of the array block to obtain the energy density C e section: (4) Perform data enhancement on the array block dataset; The array size of the array block in three directions of XYZ is O×P×Q; The three-dimensional radar data is enhanced by using three-dimensional matrix rotation. Specifically, the three-dimensional linear transformation matrix M is multiplied by the homogeneous coordinates of the three-dimensional radar data to perform the transformation. The specific calculation formula is: Where [x, y, z, 1] is the homogeneous coordinate system representation of the original 3D radar data. It is the homogeneous coordinate system representation of the transformed 3D radar data. When the 3D radar data rotates positively around the X, Y, and Z axes by an angle β, the three components of M are: In order to make the rotation center of the three-dimensional matrix the center position of the three-dimensional radar data, the rotation center needs to be translated. The transformation matrix is: The final rotated three-dimensional radar data is expressed as: (5) inputting the array block data set after signal enhancement and data enhancement into a pipeline recognition model R3D trained for three-dimensional radar data; the pipeline recognition model R3D is established by modifying the number of convolutional layers and output categories of the C3D model and adding a residual module. After modification, the number of convolutional layers of the C3D model is 5, the number of pooling layers is 5, the number of fully connected layers is 2, and the output categories are 4. The residual module is arranged on the 2nd to 4th convolutional layers; (6) According to the pipeline recognition result of the pipeline recognition model R3D, the array block data set is dyed and the pipeline direction is calculated using the dyed array blocks.

2. A pipeline signal recognition and direction calculation method according to claim 1, characterized in that: The size of the array block includes an actual size and an array size; The actual size of the array block is determined according to the characteristics of the pipeline hyperbolic signal, which is expressed as: Where x is the abscissa of the hyperbola, t is the ordinate of the hyperbola, x0 is the abscissa at the vertex of the hyperbola, t0 is the ordinate at the vertex of the hyperbola, v is the radar wave velocity in the soil medium, R0 is the distance from the center of the pipeline to the pipe wall when the radar survey line is not perpendicular to the pipeline, R is the radius of the pipeline, and f is the focus of the hyperbola; The actual size of the array block in the X and Y axis directions AB and AB for: Where, h is the buried depth of pipeline; The actual size z of the array block in the depth Z axis direction AB for: Where D is the pipe diameter; The array size O×P×Q of the array block in the three directions of XYZ is: Where Δx is the radar moving step length; Δy is the radar line spacing; Δt is the radar sampling point time interval; According to the principle that the size of the array block in the X and Y axis directions is equal to half of the pipeline hyperbolic signal, the movement overlap rate of the array block is set to 50%.

3. A pipeline signal recognition and direction calculation method according to claim 1, characterized in that: The manually labeled categories include background, metal pipes, non-metal pipes, and cables.

4. A pipeline signal recognition and direction calculation method according to claim 1, characterized in that: The process of calculating the pipeline direction using the colored array blocks is: Project the colored array block onto the plane, connect the midpoints of the array blocks at the start and end points, and the angle θ between the pipeline direction and the positive direction of the X-axis. When θ is not 90° or 0°: Where n y and n x Respectively represent the number of array blocks passed between the calculation starting point and the end point in the X and Y directions; x AB and AB It is the actual size of the array block in the X-axis and Y-axis directions in the horizontal plane.