A ground penetrating radar based underground pipeline target identification method and device

Abstract

The application relates to the technical field of underground pipeline detection, and discloses an underground pipeline target identification method and device based on a ground penetrating radar, which collects the multi-polarization scattering characteristics of underground pipelines based on a multi-polarization radar or a single-stage radar, obtains original image data of the underground pipelines, fuses the multi-polarization scattering characteristics by adopting a Laplacian pyramid algorithm to obtain fused image data, solves a plurality of polarization attributes of the fused image data, and identifies target pipelines in the underground pipelines based on a particle swarm AdaBoost algorithm and the polarization attributes. The application adopts a multi-polarization technology, combines a deep learning method to realize the identification and classification of underground pipeline targets, solves the problem that single-polarization ground penetrating radar information is not comprehensive, which leads to low target classification accuracy, improves target classification and identification speed, and provides technical support for further improving urban infrastructure.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of underground pipeline detection, in particular to an underground pipeline target identification method and device based on ground penetrating radar. BACKGROUND

[0002] With the continuous acceleration of economic development and modernization process, the city scale is expanding, at the same time, as the city infrastructure, the underground pipe network system is also more and more complex. However, due to the long time, part of the underground pipe network laying drawing cannot be traced, and in the process of urbanization, the city facilities are constantly updated, which brings the problems of underground pipe network system maintenance difficulty, safety accidents and so on. For example, when the underground pipe network is excavated and repaired, the original drawing may be lost or mismatched, which may cause gas explosion, cable leakage and other accidents, and even cause personnel casualties. In order to reduce the incidence of such accidents, it is of great significance to detect and classify the underground pipeline system.

[0003] As a non-destructive detection method, ground penetrating radar has the advantages of high detection accuracy, high efficiency, high image resolution and convenient operation. Ground penetrating radar obtains underground information by transmitting and receiving electromagnetic wave signals. When electromagnetic wave meets target body with different electrical properties, the echo signal will show different characteristics. For underground pipeline system, the materials, diameters and laying methods of water supply and drainage pipes, gas pipes, communication cables and power cables are different, based on these characteristics, ground penetrating radar can be used to identify and classify underground pipeline targets, and provide technical support for further improving city infrastructure.

[0004] For simple structure target body, traditional ground penetrating radar usually uses single polarization form, and identifies target body by analyzing the amplitude, frequency and phase of echo signal. However, with the continuous improvement of urbanization infrastructure, the underground pipe network system is very complex, and single polarization ground penetrating radar cannot meet the detection requirements. Multi-polarization technology can change the polarization characteristics of electromagnetic wave by changing the placement of transmitting and receiving antenna, enrich the echo signal information, and realize the identification and classification of distributed and complex structure target body by analyzing the polarization properties of target body. At the same time, in the classification process, the target body information obtained by ground penetrating radar needs to be extracted, and the feature extraction method relying on manual extraction and shallow learning module has low accuracy, which cannot meet the needs of large data underground pipeline system. SUMMARY

[0005] The present application aims to solve at least the technical problem in the prior art that the feature extraction method relying on manual extraction and shallow learning module has low accuracy in the classification process, which cannot meet the needs of large data underground pipeline system.

[0006] To this end, one object of the present application is to provide a ground pipeline target recognition method based on ground penetrating radar, comprising:

[0007] Collecting the multi-polarization scattering characteristics of the underground pipeline based on the multi-polarization radar or single-stage radar to obtain original image data of the underground pipeline;

[0008] Fusing the multi-polarization scattering characteristics by using a Laplacian pyramid algorithm to obtain fused image data;

[0009] Solving a plurality of polarization attributes of the fused image data, and identifying a target pipeline in the underground pipeline based on a particle swarm AdaBoost algorithm and the polarization attributes, wherein the polarization attributes include polarization basic attributes, Ferriman attributes, and polarization similarity attributes.

[0010] Further, the multi-polarization scattering characteristics are fused by using a Laplacian pyramid algorithm to obtain fused image data, comprising:

[0011] Constructing a Gaussian pyramid with the original image data as a bottom layer;

[0012] Constructing a Laplacian pyramid using the Gaussian pyramid, wherein the Laplacian pyramid includes a plurality of polarization direction Laplacian pyramids;

[0013] Reversing the Laplacian pyramid to obtain the fused image data.

[0014] Further, in the Gaussian pyramid, the image data of the Lth layer of the Gaussian pyramid is calculated according to the image data of the (L-1)th layer of the Gaussian pyramid and the convolution of the low-pass filter window U, and the image data of the Lth layer of the Gaussian pyramid with a total number of layers Q is:

[0015]

[0016] Wherein, i1 and j1 represent the number of rows and columns of data in the layer; m and n are layer numbers; S L is the image data of the Lth layer of the Gaussian pyramid of a single polarization direction; S L-1 is the image data of the (L-1)th layer of the Gaussian pyramid of a single polarization direction.

[0017] Further, the Laplacian pyramid is constructed using the Gaussian pyramid, comprising:

[0018] An operator A is introduced to interpolate and enlarge the Gaussian pyramid to obtain S' L , wherein the size of S' L is the same as that of the image data of the (L-1)th layer of the Gaussian pyramid, and the operator A is represented as:

[0019]

[0020] wherein, i1 and j1 represent the number of rows and columns of data in the layer; m and n are layer numbers; S L is the image data of the Lth layer of Gaussian pyramid for a single polarization direction; U represents the convolution of a low-pass filter window; S' L is the intermediate value, which has no real meaning;

[0021] Let the Lth layer of Laplacian pyramid data for a single polarization direction be P L , and its formula is:

[0022] P L = S L - A(S L+1 )

[0023] wherein, S L+1 is the image data of the L+1th layer of Gaussian pyramid for a single polarization direction; the Lth layer of Laplacian pyramid data is equal to the image data of the Lth layer of Gaussian pyramid.

[0024] Further, based on the particle swarm AdaBoost algorithm and the polarization attribute, the target pipeline in the underground pipeline is identified:

[0025] A data set D is constructed based on a plurality of the polarization attributes, and a particle center PC of the data set D is calculated by iteration; a new data set Z is established according to the particle center PC, wherein Z = |D-PC|;

[0026] The data set D is projected to a new domain, and the weight of each sample point in the data set Z is determined, wherein η k = (ω k1 , ω k2 ,..., ω kl ,..., ω kN ), wherein k represents the iteration time; N represents the number of samples; ω kl is the weight of the lth sample point at the iteration time k;

[0027] According to the weight and error curve E k of each sample point, the error of the weak classifier is calculated, as follows:

[0028]

[0029] wherein, e k is the error of the weak classifier; y l is the label of the lth sample point; Z l is the lth sample point in the new data set Z; h k is the weak classifier, which is determined according to the error curve E k ;

[0030]

[0031] wherein r is the radius of the weak classifier;

[0032] updating the weight of the iterative sample point according to the error of the weak classifier, classifying the underground pipeline, and identifying the target pipeline.

[0033] Further, the updating of the weight of the iterative sample point according to the error of the weak classifier comprises:

[0034] If the error of the weak classifier is less than a preset error threshold, it is a correct classification, otherwise it is an incorrect classification, and in the next iteration, the weight of the sample point of the correct classification is reduced and the weight of the sample point of the incorrect classification is increased.

[0035] Further, the method further comprises:

[0036] calculating the velocity and the coordinate of each particle in the data set D in the iteration, wherein the velocity of the qth particle in the ith iteration is and the coordinate is respectively:

[0037]

[0038] wherein the subscript number of q represents the serial number of the polarization attribute, 1 represents the polarization entropy, 2 represents the average scattering angle, and 3-5 represent the three-dimensional parameter values of the Freeman attribute, and 6-8 represent the three-dimensional parameter values for representing the polarization similarity respectively;

[0039] constructing a loss function by using the distance between the particle center and all particles, wherein the loss function is:

[0040]

[0041] wherein Loss is the loss function; N is the total number of particles; b is the particle serial number; c represents the serial number of the polarization attribute, and the value range is 1-8, 1 represents the polarization entropy, 2 represents the average scattering angle, 3-5 represent the three-dimensional parameter values of the Freeman attribute, and 6-8 represent the three-dimensional parameter values for representing the polarization similarity respectively; D bc is the particle center when the serial number of the polarization attribute in the data set D is c;

[0042] calculating the minimum fitting value of the loss function of each particle from the first iteration to the ith iteration, and determining the minimum fitting value in all particles as the particle center PC.

[0043] Further, the method further comprises:

[0044] updating the speed and coordinate of the particle swarm with the minimum fitting value in all particles;

[0045] recomputing the minimum fitting value with the updated speed and coordinate of each particle until the difference between the minimum fitting value and the minimum fitting value in the previous iteration is less than a preset fitting value threshold.

[0046] Further, the multi-polarization scattering characteristics of the underground pipeline are collected based on a multi-polarization radar or a single-stage radar, including:

[0047] The multi-polarization radar is adopted to collect the multi-polarization scattering characteristics of the underground pipeline by using a chaotic signal as a detection signal.

[0048] The single-polarization radar is adopted to collect the multi-polarization scattering characteristics of the underground pipeline by rotating the single-polarization radar at the same position to change the polarization direction of the antenna.

[0049] The application provides an underground pipeline target identification device based on a ground penetrating radar, including: a collection module configured to collect multi-polarization scattering characteristics of an underground pipeline based on a multi-polarization radar or a single-stage radar to obtain original image data of the underground pipeline.

[0050] A fusion module is configured to fuse the multi-polarization scattering characteristics by using a Laplacian pyramid algorithm to obtain fused image data.

[0051] An identification module is configured to solve a plurality of polarization attributes of the fused image data, identify a target pipeline in the underground pipeline based on a particle swarm AdaBoost algorithm and the polarization attributes, wherein the polarization attributes include polarization basic attributes, a Freeman attribute and a polarization similarity attribute.

[0052] The polarization basic attributes include polarization entropy and average scattering angle.

[0053] The application provides an underground pipeline target identification method and device based on a ground penetrating radar, which has the following beneficial effects:

[0054] (1) The multi-polarization information is obtained by changing the antenna position: due to the continuous modification in the urbanization construction process, the underground pipeline network system is very complex, and the single-polarization information cannot meet the accurate identification of the underground pipeline network target. Therefore, the multi-polarization ground penetrating radar is adopted to enrich the target information and improve the identification and classification accuracy. However, due to the limitation of the actual scene, there may be some special scenes where the multi-polarization radar is not convenient to use, and only the single-polarization radar can be used. To solve this problem, the application proposes that the direction of the single-polarization ground penetrating radar antenna can be changed continuously to obtain the full-polarization target information in four relative position relationships.

[0055] (2) Using Laplace pyramid algorithm to realize multi-polarization data fusion: after obtaining multi-polarization target information based on ground penetrating radar, the image data needs to be multi-polarization fused. When using basic Gaussian pyramid, in the operation process, the image is subjected to convolution and down-sampling operation, and part of the high-frequency information is lost, and the high-frequency information is used to describe the target detail information. When facing underground pipeline target, since the diameter and material difference between part of the pipeline or cable are subtle, the Gaussian pyramid cannot meet the requirement of detail identification, therefore, the present application uses Laplace pyramid, and uses interpolation method to improve the sensitivity to detail information, and avoids the classification error problem caused by information loss.

[0056] (3) Using particle swarm AdaBoost method to realize classification: since the position relationship of underground pipeline target is complex and the distinction degree is small, in the classification process, multiple attribute values need to be used as classification basis. Therefore, particle swarm AdaBoost method is used to improve the classification accuracy. When facing multiple parameter problems, the method can automatically select appropriate parameters through weak classifier according to different targets, and improve the classification efficiency.

[0057] The present application uses multi-polarization technology, and combines deep learning method to realize identification and classification of underground pipeline target, solves the problem that single-polarization ground penetrating radar information is not comprehensive, and leads to low target classification accuracy, improves the target classification and identification speed, and provides technical support for further improving urban infrastructure. BRIEF DESCRIPTION OF DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the embodiment or prior art description will be briefly introduced below, and obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0059] Figure 1 is a hybrid FDTD algorithm flow chart of the underground pipeline target identification method based on ground penetrating radar in the embodiment of the present application;

[0060] Figure 2 is a schematic diagram of four position relationships of transmitting antenna and receiving antenna in ground penetrating radar of the underground pipeline target identification method based on ground penetrating radar in the embodiment of the present application. DETAILED DESCRIPTION

[0061] Exemplary embodiments of the present disclosure are described herein with reference to the accompanying drawings, which are meant to be exemplary in nature, and include various details intended to facilitate understanding of the present disclosure. Accordingly, one skilled in the art should realize that various changes and modifications can be made to the embodiments described herein, without departing from the scope and spirit of the present disclosure. Likewise, for clarity and conciseness, descriptions of well-known functions and constructions are omitted from the following description.

[0062] As shown in the figure, according to the underground pipeline target identification method based on ground penetrating radar according to the embodiments of the present disclosure, the following steps are included: Figure 1

[0063] Step S1: Based on the multi-polarization radar or single-polarization radar, the multi-polarization scattering characteristics of the underground pipeline are collected to obtain the original image data of the underground pipeline.

[0064] Step S2: The multi-polarization scattering characteristics are fused by using the Laplacian pyramid algorithm to obtain the fused image data.

[0065] Step S3: A plurality of polarization attributes of the fused image data are solved, and the target pipeline in the underground pipeline is identified based on the particle swarm AdaBoost algorithm and the polarization attributes, wherein the polarization attributes include polarization basic attributes, Ferriman attributes and polarization similar attributes, and the polarization basic attributes include polarization entropy and average scattering angle.

[0066] Specifically, in the step S1, when the single-polarization radar can only be used due to environmental restrictions and the like, the multi-polarization scattering characteristics are obtained by measuring multiple times from multiple directions.

[0067] More specifically, in the step S1, the multi-polarization radar is used to collect the multi-polarization scattering characteristics by using chaotic signals as detection signals; when the multi-polarization radar cannot be used in actual situations, the single-polarization radar is used, the single-polarization radar is rotated at the same position to change the polarization direction of the antenna, and the multi-polarization scattering characteristics of the underground pipeline target are collected. Figure 2 As shown in the figure, four kinds of position relationships between the transmitting antenna and the receiving antenna in the ground penetrating radar are shown, wherein TX represents the transmitting antenna, RX represents the receiving antenna, and the four kinds of relationships are horizontal transmitting and horizontal receiving (HH), vertical transmitting and vertical receiving (VV), horizontal transmitting and vertical receiving (HV), and vertical transmitting and horizontal receiving (VH).

[0068] The present disclosure provides an underground pipeline target identification method and device based on ground penetrating radar:

[0069] ​(1) Obtain multi-polarization information by changing the antenna position: due to the continuous modification in the process of urbanization construction, the underground pipe network system is very complex, and single polarization information cannot meet the accurate identification of underground pipe network targets. Therefore, multi-polarization ground penetrating radar is used to enrich target information and improve the classification accuracy. However, due to the limitation of the actual scene, there may be some special scenes where it is not convenient to use multi-polarization radar, and only single-polarization radar can be used. To solve this problem, the application proposes to change the direction of the single-polarization ground penetrating radar antenna continuously, and obtain full-polarization target information in four relative position relationships.

[0070] (2) Multi-polarization data fusion is realized by using Laplace pyramid algorithm: after obtaining multi-polarization target information based on ground penetrating radar, image data needs to be fused. When using basic Gaussian pyramid, in the operation process, part of the high frequency information will be lost after convolution and downsampling operation, and the high frequency information is used to describe the target detail information. When facing underground pipeline targets, due to the slight difference between the diameters and materials of part of the pipelines or cables, the Gaussian pyramid cannot meet the requirement of detail identification, therefore, the application uses Laplace pyramid, uses interpolation method to improve the sensitivity to detail information, and avoids the classification error problem caused by information loss.

[0071] (3) Classification is realized by using particle swarm AdaBoost method: due to the complex position relationship of underground pipeline targets and small differentiation, multiple attribute values need to be used as classification basis in the classification process. Therefore, particle swarm AdaBoost method is used to improve the classification accuracy. When facing multi-parameter problems, this method can automatically select appropriate parameters through weak classifiers according to different targets, and improve the classification efficiency.

[0072] In one embodiment, after obtaining the multi-polarization scattering characteristics of the underground pipeline target, in order to improve the accuracy, the underground pipeline target identification method based on ground penetrating radar further comprises the following steps in step S2:

[0073] This step includes three processes of decomposition, superposition and reconstruction, and the specific process is as follows:

[0074] Step S21: construct a Gaussian pyramid with the original image data as the bottom layer, and obtain the image data of the Lth layer of the Gaussian pyramid with a total number of Q by using the convolution of the image data of the L-1th layer of the Gaussian pyramid and the low-pass filter window:

[0075]

[0076] Where, i1 and j1 represent the number of rows and columns of data in the layer; m and n are layer numbers, U(m, n) is the low-pass filter window of the layer; S L is the image data of the Lth layer of the Gaussian pyramid for a certain polarization direction; SL-1 image data of the Lth layer of the Gaussian pyramid of a certain polarization direction;

[0077] Step S22: obtaining Laplacian pyramid data from the image data of the Gaussian pyramid, wherein the Laplacian pyramid comprises a plurality of Laplacian pyramids of polarization directions;

[0078] Specifically, an operator A is introduced to interpolate and enlarge the Gaussian pyramid to obtain S' L , wherein the size of S' L is the same as the image data of the L-1th layer of the Gaussian pyramid, and the operator A is expressed as:

[0079]

[0080] wherein i1 and j1 represent the number of rows and columns of data in the layer; m and n are layer numbers; S L is image data of the Lth layer of the Gaussian pyramid of a single polarization direction; U represents convolution of a low-pass filter window; S' L is an intermediate value without real meaning;

[0081] Let the Lth layer of the Laplacian pyramid data of a single polarization direction be P L , and its formula is:

[0082] P L = S L -A(S L+1 )

[0083] wherein S L+1 is image data of the L+1th layer of the Gaussian pyramid of a single polarization direction; the Lth layer of the Laplacian pyramid data is equal to the image data of the Lth layer of the Gaussian pyramid.

[0084] Step S23: obtaining the fusion image data by inversely changing the Laplacian pyramid, including horizontal transmission horizontal reception Laplacian pyramid image data horizontal transmission vertical reception Laplacian pyramid data and vertical transmission horizontal reception Laplacian pyramid data

[0085] S L-1 = P L-1 -A(S L )

[0086] In the image data of the Laplacian pyramid, a certain threshold value is taken to divide the high-frequency region and the low-frequency region, the high-frequency describes the detailed information, and the low-frequency describes the outline information. Therefore, for the high-frequency part, the maximum value is taken; for the low-frequency part, the average value is taken.

[0087] The polarization basic attribute, the Freiman attribute, and the polarization similarity attribute include polarization entropy H, average scattering angle a, three-dimensional parameter value P S d v and three-dimensional parameter value rs s v d for representing polarization similarity;

[0088] The above attributes are obtained by the fusion image data based on theoretical formula, wherein the subscripts s, v, and d respectively represent scattering, double-backscattering scattering, and volume scattering;

[0089] In an embodiment, the step S3 of the ground penetrating radar-based underground pipeline target identification method further includes the following steps of identifying the target pipeline in the underground pipeline based on the particle swarm AdaBoost algorithm and the polarization attributes:

[0090] Step S31: Construct a data set D based on a plurality of the polarization attributes, and calculate the particle center PC of the data set D by iteration; according to the particle center PC, a new data set Z is established, wherein Z = |D-PC|;

[0091] Step S32: Project the data set D to a new domain, and determine the weight of each sample point in the data set Z, wherein η k = (ω k1 , ω k2 ,..., ω kl ,..., ω kN ), wherein k represents the iteration time; N represents the sample quantity; ω kl is the weight of the lth sample point at the iteration time k;

[0092] Step S33: According to the weight and error curve E k of each sample point, the error of the weak classifier is calculated;

[0093] Specifically, as follows:

[0094]

[0095] wherein e k is the error of the weak classifier; y l is the label of the lth sample point; Z l is the lth sample point in the new data set Z; h k is the weak classifier, and the weak classifier is determined according to the error curve E k ;

[0096] ​​​​

[0097] wherein r is the radius of the weak classifier.

[0098] Step S34: updating the weight of the iterative sample point according to the error of the weak classifier, classifying the underground pipeline, and identifying the target pipeline. If the error of the weak classifier is less than a preset error threshold, it is a correct classification, otherwise it is an incorrect classification, and in the next iteration, the weight of the sample point of the correct classification is reduced and the weight of the sample point of the incorrect classification is increased.

[0099] In the step S31, the particle center PC of the data set D includes the following steps:

[0100] Step S311: calculating the velocity and coordinates of each particle in the data set D in the iteration, wherein the velocity of the qth particle in the i3th iteration is and the coordinates are respectively:

[0101]

[0102] wherein the subscript number of q represents the serial number of the polarization attribute, 1 represents the polarization entropy, 2 represents the average scattering angle, 3-5 represent the three-dimensional parameter values of the Freeman attribute, and 6-8 represent the three-dimensional parameter values for representing the polarization similarity; the specific values are as follows:

[0103] is the velocity value of the qth particle of the polarization entropy H in the i3th iteration; is the velocity value of the qth particle of the average scattering angle a in the i3th iteration; is the velocity value of the qth particle of the scattering parameter value P S representing the Freeman attribute in the i3th iteration; is the velocity value of the qth particle of the double reflection scattering parameter value P d representing the Freeman attribute in the i3th iteration; is the velocity value of the qth particle of the volume scattering parameter value P v representing the Freeman attribute in the i3th iteration; is the velocity value of the qth particle of the scattering parameter value rs s representing the polarization similarity in the i3th iteration; is the velocity value of the qth particle of the double reflection scattering parameter value rs v representing the polarization similarity in the i3th iteration; is the velocity value of the qth particle of the volume scattering parameter value rs d representing the polarization similarity in the i3th iteration;

[0104] is the coordinate value of the qth particle of the average scattering angle a in the ith iteration; is the coordinate value of the qth particle of the average scattering angle a in the ith iteration; is the scattering parameter value P of the qth particle of the Freeman attributes; S is the coordinate value of the qth particle of the Freeman attributes in the ith iteration; is the scattering parameter value P of the qth particle of the double reflection Freeman attributes; d is the coordinate value of the qth particle of the double reflection Freeman attributes in the ith iteration; is the scattering parameter value P of the qth particle of the volume Freeman attributes; v is the coordinate value of the qth particle of the volume Freeman attributes in the ith iteration; is the scattering parameter value rs of the qth particle of the polarization similarity; s is the coordinate value of the qth particle of the polarization similarity in the ith iteration; is the scattering parameter value rs of the qth particle of the double reflection polarization similarity; v is the coordinate value of the qth particle of the double reflection polarization similarity in the ith iteration; is the scattering parameter value rs of the qth particle of the volume polarization similarity; d is the coordinate value of the qth particle of the volume polarization similarity in the ith iteration;

[0105] Step S312: using the particle center PC and the distance between all particles, the loss function is constructed as follows:

[0106]

[0107] wherein, Loss is the loss function; N is the total number of particles; b is the particle serial number; c represents the serial number of the polarization attribute, and the value range is 1-8, 1 represents the polarization entropy, 2 represents the average scattering angle, 3-5 represents the three-dimensional parameter value of the Freeman attribute, and 6-8 respectively represents the three-dimensional parameter value for representing the polarization similarity; D bc is the particle center when the serial number of the polarization attribute in the data set D is c;

[0108] Step S313: calculating the minimum fitting value of the loss function of each particle from the first iteration to the ith iteration, and determining the minimum fitting value in all particles as the particle center PC;

[0109] updating the speed and coordinates of the particle group by using the minimum fitting value in all particles;

[0110] recomputing the minimum fitting value by using the updated speed and coordinates of each particle until the difference between the minimum fitting value and the minimum fitting value in the previous iteration is less than the preset fitting value threshold.

[0111] Specifically as follows:

[0112] The minimum fitting value of the particle from the first iteration to the ith iteration is calculated, and the minimum fitting value of the qth particle is The minimum fitting value in all particles is PC i3 The calculation formula is as follows:

[0113]

[0114] The underground pipeline target identification device based on the ground penetrating radar provided by the embodiments of the present disclosure comprises a collection module configured to collect the multi-polarization scattering characteristics of an underground pipeline based on a multi-polarization radar or a single-stage radar to obtain original image data of the underground pipeline.

[0115] A fusion module is configured to fuse the multi-polarization scattering characteristics by using a Laplacian pyramid algorithm to obtain fused image data.

[0116] An identification module is configured to solve a plurality of polarization attributes of the fused image data, and identify a target pipeline in the underground pipeline based on a particle swarm AdaBoost algorithm and the polarization attributes, wherein the polarization attributes comprise polarization basic attributes, Ferriman attributes, and polarization similarity attributes.

[0117] The embodiments of the present disclosure also provide an electronic device comprising at least a memory and a processor, wherein the memory stores a computer program, and the processor implements the above-mentioned underground pipeline target identification method based on the ground penetrating radar when executing the computer program on the memory, and the method comprises the following steps:

[0118] Step S1: collecting the multi-polarization scattering characteristics of an underground pipeline based on a multi-polarization radar or a single-stage radar to obtain original image data of the underground pipeline.

[0119] Step S2: fusing the multi-polarization scattering characteristics by using a Laplacian pyramid algorithm to obtain fused image data.

[0120] Step S3: solving a plurality of polarization attributes of the fused image data, and identifying a target pipeline in the underground pipeline based on a particle swarm AdaBoost algorithm and the polarization attributes, wherein the polarization attributes comprise polarization basic attributes, Ferriman attributes, and polarization similarity attributes, and the polarization basic attributes comprise polarization entropy and average scattering angle.

[0121] In some embodiments, the processor executing the computer program can be a processing device including one or more general-purpose processing devices, such as a microprocessor, a central processing unit (CPU), a graphics processing unit (GPU), and the like. More particularly, the processor can be a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a processor running other instruction sets, or a processor running a combination of instruction sets. The processor can also be one or more special-purpose processing devices such as an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), a digital signal processor (DSP), a system on a chip (SoC), and the like.

[0122] The memory can be read-only memory (ROM), random access memory (RAM), phase change random access memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), electrically erasable programmable read-only memory (EEPROM), other types of random access memory (RAM), flash memory, or other forms of flash memory, cache, register, static memory, compact disc read-only memory (CD-ROM), digital versatile disc (DVD), or other optical storage, magnetic cassettes or other magnetic storage devices, or any other possible non-transitory medium that can be used to store information or instructions that can be accessed by a computing device, and the like.

[0123] The electronic device of the embodiments of the present disclosure can include, but is not limited to, fixed terminal devices such as MCU controllers, servers, desktop computers, digital TVs, and the like, and mobile terminal devices such as in-vehicle devices (for example, head-up display devices), handheld devices (for example, mobile phones, tablets, and the like), wearable devices (for example, smart watches, smart bands, and the like), and the like.

[0124] The embodiments of the present disclosure also provide a computer readable storage medium, the computer readable medium stores a computer program, the computer program is executed by a processor to implement the above-mentioned ground pipeline target identification method based on ground penetrating radar, and the method comprises the following steps:

[0125] Step S1: based on a multi-polarized radar or a single-polarized radar, a multi-polarized scattering characteristic of an underground pipeline is collected, and original image data of the underground pipeline is obtained;

[0126] Step S2: a Laplacian pyramid algorithm is used to fuse the multi-polarized scattering characteristic, and fused image data is obtained;

[0127] Step S3: solving a plurality of polarization attributes of the fusion image data, identifying a target pipeline in the underground pipeline based on a particle swarm AdaBoost algorithm and the polarization attributes, wherein the polarization attributes include polarization basic attributes, a Ferriman attribute and a polarization similarity attribute, and the polarization basic attributes include polarization entropy and an average scattering angle.

[0128] The present application has the following beneficial effects:

[0129] (1) Obtain multi-polarization information by changing the antenna position: due to the continuous modification in the process of urbanization construction, the underground pipe network system is very complex, and single polarization information cannot meet the accurate identification of underground pipe network targets. Therefore, multi-polarization ground penetrating radar is used to enrich target information and improve the classification accuracy. However, due to the limitation of the actual scene, there may be some special scenes where it is not convenient to use multi-polarization radar, and only single-polarization radar can be used. To solve this problem, the present application proposes to change the direction of the single-polarization ground penetrating radar antenna constantly, and obtain full-polarization target information under four relative position relationships.

[0130] (2) Multi-polarization data fusion is realized by using Laplacian pyramid algorithm: after obtaining multi-polarization target information based on ground penetrating radar, the image data needs to be fused. When using a basic Gaussian pyramid, in the operation process, the image will lose part of the high-frequency information after convolution and downsampling operation, and the high-frequency information is used to describe the target detail information. When facing underground pipeline targets, due to the slight difference between the diameters and materials of part of the pipelines or cables, the Gaussian pyramid cannot meet the requirements of detail identification, therefore, the present application uses the Laplacian pyramid, uses interpolation method to improve the sensitivity to detail information, and avoids the classification error problem caused by information loss.

[0131] (3) Particle swarm AdaBoost method is used to realize classification: due to the complex position relationship of underground pipeline targets and small differentiation, a plurality of attribute values need to be used as classification basis in the classification process. Therefore, particle swarm AdaBoost method is used to improve the classification accuracy. When facing multi-parameter problems, this method can automatically select appropriate parameters through weak classifiers according to different targets, and improve the classification efficiency.

[0132] The present application uses multi-polarization technology, combines deep learning method to realize the identification and classification of underground pipeline targets, solves the problem that single-polarization ground penetrating radar information is not comprehensive, which leads to low target classification accuracy, improves the target classification and identification speed, and provides technical support for further improving urban infrastructure.

[0133] The computer readable storage medium of the embodiments of the present disclosure can adopt any combination of one or more computer readable media. The computer readable medium can be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium may, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination of the above. In the embodiments of the present disclosure, the computer readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or apparatus, for example, the memory described above.

[0134] The computer program of the embodiments of the present disclosure can be organized into one or more computer executable components or modules. Any number and combination of such components or modules can be used to implement aspects of the present disclosure. For example, aspects of the present disclosure are not limited to the specific computer executable instructions or specific components or modules illustrated in the figures and described herein. Other embodiments can include different computer executable instructions or components with more or less functionality than described herein.

[0135] The above description is merely the preferred embodiments of the present disclosure and the explanation of the principles of the applied technology. It should be understood by those skilled in the art that the disclosed scope of the present disclosure is not limited to the technical solutions formed by the specific combinations of the technical features described above, and should also cover other technical solutions formed by any combinations of the above technical features or their equivalent features without departing from the disclosed concept. For example, the technical solutions formed by replacing the above features with the technical features disclosed in the present disclosure (but not limited to) having similar functions.

Claims

1. A ground penetrating radar based underground pipeline target identification method, characterized by, The method comprises the steps of: Collecting the multi-polarization scattering characteristics of underground pipelines based on a multi-polarization radar or a single-stage radar to obtain original image data of the underground pipelines; Fusing the multi-polarization scattering characteristics by using a Laplacian pyramid algorithm to obtain fused image data; Solving a plurality of polarization attributes of the fused image data, and identifying a target pipeline in the underground pipelines based on a particle swarm AdaBoost algorithm and the polarization attributes, wherein the polarization attributes comprise polarization basic attributes, Freeman attributes, and polarization similarity attributes; The step of fusing the multi-polarization scattering characteristics by using the Laplacian pyramid algorithm to obtain the fused image data comprises the steps of: Constructing a Gaussian pyramid with the original image data as a bottom layer; Constructing a Laplacian pyramid with the Gaussian pyramid, wherein the Laplacian pyramid comprises a plurality of polarization direction Laplacian pyramids; Reversely changing the Laplacian pyramid to obtain the fused image data; Identifying the target pipeline in the underground pipelines based on the particle swarm AdaBoost algorithm and the polarization attributes: A data set D is constructed based on the plurality of polarization properties, and a particle center PC of the data set D is calculated by iteration; a new data set Z is established according to the particle center PC, wherein, ; projecting the dataset D into a new domain and determining a weight for each sample point in the dataset Z, wherein where k denotes the iteration time; N denotes the number of samples; is the weight of the l-th sample point at iteration time k. According to the weight and error curve of each sample point , the error of the weak classifier is calculated as follows: ， wherein, is the error of the weak classifier; is the label of the lth sample point; is the lth sample point in the new dataset Z; is a weak classifier, which is determined according to the error curve is determined: ， Wherein r is the radius of a weak classifier; According to the error of the weak classifier, updating the weight of an iterative sample point, classifying the underground pipelines, and identifying the target pipeline.

2. The underground pipeline target identification method based on the ground penetrating radar according to claim 1, wherein in the Gaussian pyramid, the image data of an Lth layer Gaussian pyramid is calculated according to the image data of an (L-1) th layer Gaussian pyramid and the convolution of a low-pass filter window U, and the image data of an Lth layer Gaussian pyramid of a Gaussian pyramid with a total number of layers of Q is: The step of constructing the Laplacian pyramid with the Gaussian pyramid comprises the steps of: ， wherein, and denote the number of rows and columns of data in the layer; m and n are layer indices; is image data of the Lth layer of Gaussian pyramid for a single polarization direction; is image data of the (L-1)th layer of Gaussian pyramid for a single polarization direction.

3. The ground penetrating radar based underground pipeline target identification method of claim 1, wherein, The step of updating the weight of the iterative sample point according to the error of the weak classifier comprises the steps of: An operator A is introduced to interpolate and enlarge the Gaussian pyramid to obtain wherein the The size of the image data of the Lth layer Gaussian pyramid is the same as that of the image data of the (L-1)th layer Gaussian pyramid, and the operator A is expressed as ， wherein, and denote the number of rows and columns of data in the layer; m and n are layer numbers; is the image data of the Lth layer of Gaussian pyramid for a single polarization direction; U denotes the convolution of a low-pass filter window; is the intermediate value, which has no real meaning; Let the Lth layer Laplacian pyramid data with a single polarization direction be The formula is: ， wherein, is the image data of the (L+1)th layer of the Gaussian pyramid with a single polarization direction; the Lth layer of Laplacian pyramid data is equal to the image data of the Lth layer of the Gaussian pyramid.

4. The ground penetrating radar based underground pipeline target identification method of claim 1, wherein, If the error of the weak classifier is less than a preset error threshold, it is correct classification, otherwise it is incorrect classification, and in the next iteration, the weight of the sample point of correct classification is reduced and the weight of the sample point of incorrect classification is increased. The step of iteratively calculating the particle center PC of the data set D comprises the steps of:

5. The ground penetrating radar based underground pipeline target identification method of claim 1, wherein, Wherein the subscript number of q represents the serial number of the polarization attribute, 1 represents polarization entropy, 2 represents average scattering angle, and 3-5 represent three-dimensional parameter values of the Freeman attribute, and 6-8 represent three-dimensional parameter values for representing polarization similarity respectively; The velocity and coordinates of each particle in the data set D at each iteration are calculated, where the velocity and coordinates of the qth particle at the ith iteration are: and respectively. ， The distance between the particle center and all particles is used to construct a loss function, wherein the loss function is: The minimum fitting value of the loss function of each particle from the first iteration to the i3th iteration is calculated, and the minimum fitting value in all particles is determined as the particle center PC. ， Wherein, Loss is a loss function; N is the total number of particles; b is the particle serial number; c represents the serial number of the polarization attribute, and the value range is 1-8, 1 represents the polarization entropy, 2 represents the average scattering angle, 3-5 represent the three-dimensional parameter values of the Freeman attribute, and 6-8 represent the three-dimensional parameter values for representing the polarization similarity respectively; is the particle center when the serial number of the polarization attribute in the data set D is c; 6. The underground pipeline target identification method based on the ground penetrating radar according to claim 5, further comprising: The speed and coordinates of the particle swarm are updated by using the minimum fitting value in all particles; The minimum fitting value is recalculated by using the updated speed and coordinates of each particle until the difference between the minimum fitting value and the minimum fitting value in the previous iteration is less than a preset fitting value threshold. The step of collecting the multi-polarization scattering characteristics of underground pipelines based on a multi-polarization radar or a single-stage radar comprises the steps of:

7. The ground penetrating radar based underground pipeline target identification method of claim 1, wherein, ​ The multi-polarization radar is used to collect the multi-polarization scattering characteristics of the underground pipeline by using a chaotic signal as a detection signal. The single-polarization radar is used to collect the multi-polarization scattering characteristics of the underground pipeline by rotating the single-polarization radar at the same position to change the polarization direction of the antenna.

8. A ground penetrating radar based underground pipeline target identification apparatus, characterized by, The method is implemented by using the underground pipeline target identification method based on ground penetrating radar according to any one of claims 1-7, comprising: A collection module configured to collect the multi-polarization scattering characteristics of the underground pipeline based on a multi-polarization radar or a single-polarization radar to obtain original image data of the underground pipeline. A fusion module configured to fuse the multi-polarization scattering characteristics by using a Laplacian pyramid algorithm to obtain fused image data. An identification module configured to solve a plurality of polarization attributes of the fused image data, identify a target pipeline in the underground pipeline based on a particle swarm AdaBoost algorithm and the polarization attributes, wherein the polarization attributes include a polarization basic attribute, a Ferriman attribute and a polarization similarity attribute.

Images