A power pipeline identification model construction method and a power pipeline identification method

Abstract

The present application relates to the technical field of radar detection, in particular to a power pipeline identification model construction method and a power pipeline identification method. The method can ensure that clutter in signals of different depths can be effectively removed through layered filtering and gain compensation of echo data, and improve the accuracy of subsequent model identification. At the same time, the gain compensation also processes the echo signals of each layer respectively, so that the energy of deep layer signals and the energy of shallow layer signals are in the same order of magnitude in numerical value, and the feature drift caused by depth difference is suppressed. The prior depth and dielectric constant are added to the features extracted from the echo signals, so that the sample has physical interpretability, the depth error caused by medium uncertainty is significantly reduced, and the pipeline features are more embodied. Moreover, the spatial local density is introduced as a penalty weight for the sample, which can strengthen the learning of sparse pipelines through the weighting strategy. Thus, the identification accuracy and applicability of the model can be significantly improved.

Description

Technical Field

[0001] This invention relates to the field of radar detection technology, specifically to a method for constructing a power line identification model and a method for identifying power lines. Background Technology

[0002] In recent years, the scale of urban underground engineering has been continuously expanding, and the laying of underground pipelines has entered a stage of rapid development. Soil disturbance caused by urban construction has frequently led to power cable breakage accidents, threatening the safety of power supply. The rapid development of geophysical exploration technology has provided an important means for detecting underground power pipelines.

[0003] Ground penetrating radar (GPR) technology is a technique that uses electromagnetic wave reflection to determine the distribution of underground media. It is widely used in underground pipeline detection due to its non-destructive, rapid, and shallow high-resolution advantages. Currently, GPR technology for underground power line detection mostly requires manual analysis of radar echoes line by line, which is time-consuming and labor-intensive, making it difficult to meet the needs of rapid urban construction and emergency rescue scenarios. Summary of the Invention

[0004] In view of this, the present invention provides a method for constructing a power pipeline identification model and a method for identifying power pipelines, so as to solve the problems of low accuracy and low efficiency of existing detection methods.

[0005] In a first aspect, the present invention provides a method for constructing a power pipeline identification model. The method includes: acquiring echo data from ground-penetrating radar detecting known underground pipelines, the dielectric constant of the underground soil, and the pipeline category; layering the echo data according to the wave velocity calculated from the dielectric constant and the transmission time, and performing layered filtering and gain compensation on the layered echo data to obtain preprocessed echo data; constructing a cutoff window based on the prior burial depth of the pipeline to extract features from the preprocessed echo data, and constructing a feature vector by combining the prior burial depth of the pipeline and the dielectric constant, wherein the feature vector serves as a sample and the pipeline category serves as a label to constitute a training sample set; constructing a penalty weight for the sample based on the spatial local density of the sample to obtain a weighted sample set; and training a support vector machine model using the weighted sample set to obtain a power pipeline identification model.

[0006] The power pipeline identification model construction method provided in this invention, through layered filtering and gain compensation of echo data, ensures that clutter and other noise in signals at different depths can be effectively removed, improving the accuracy of subsequent model identification. Simultaneously, gain compensation processes the echo signal of each layer separately, ensuring that the energy of deep and shallow signals are on the same order of magnitude, suppressing feature drift caused by depth differences. Prior burial depth and dielectric constant are added to the features extracted from the echo signal, making the sample physically interpretable, significantly reducing depth errors caused by medium uncertainty, and better reflecting pipeline characteristics. Furthermore, spatial local density is introduced as a penalty weight for the sample, thereby strengthening the learning of sparse pipelines through this weighted strategy. The pipeline identification model trained using the weighted training samples constructed in the above manner can significantly improve the model's identification accuracy and applicability.

[0007] In one optional implementation, the echo data is layered based on the wave velocity calculated from the dielectric constant and the transmission time, and the layered echo data is subjected to layered filtering and gain compensation to obtain preprocessed echo data. This includes: determining a time zero point based on the direct peak value in the echo data; determining the depth corresponding to the echo data based on the time zero point, the wave velocity determined by the dielectric constant, and the corresponding time of the echo data; layering the echo data based on the depth; performing layered filtering on the echo data at different depths according to different frequency ranges, wherein the upper and lower frequency limits in the frequency range decrease as the depth increases; and performing gain compensation on the layered filtered echo data using an exponential gain function, where the exponential gain function is a function of gain changing with time, and the coefficient of the exponential gain function is the skin attenuation coefficient.

[0008] In this invention, by determining the zero point of time, the time base drift caused by ground undulations can be eliminated; by setting the upper and lower frequency limits in the frequency range during filtering to decrease as the depth increases, effective removal of clutter in each layer can be achieved.

[0009] In one optional implementation, a cutoff window is constructed based on the prior burial depth of the pipeline to extract features from the preprocessed echo data. A feature vector is then constructed by combining the prior burial depth of the pipeline and the dielectric constant. The feature vector serves as a sample, and the pipeline category serves as a label to form a training sample set. This includes: constructing a cutoff window based on the ratio of the prior burial depth of the pipeline to the wave velocity; performing signal cutoff on the preprocessed echo data based on the cutoff window, and performing principal component analysis on the cutoff signal to obtain principal component features; constructing a feature vector based on the principal component features, the prior burial depth of the pipeline, and the dielectric constant; and constructing a training sample set using the feature vector as a sample and the pipeline category as a label.

[0010] In one optional implementation, a weighted sample set is obtained by constructing a penalty weight for the sample based on the spatial local density of the sample, including: determining the penalty weight of each sample based on the number of samples within a preset radius centered on the sample and the penalty factor; and constructing a weighted sample set based on the penalty weight, the sample, and the label.

[0011] In one optional implementation, the power pipeline identification model is trained using the weighted sample set to obtain a power pipeline identification model, including: performing feature normalization processing on the samples in the weighted sample set to obtain a processed sample set; determining the initial values ​​of the kernel function parameters in the support vector machine model based on the attenuation model of electromagnetic waves in underground media; and training the support vector machine model using the processed sample set to obtain a power pipeline identification model.

[0012] In this invention, initial parameter values ​​are derived from the attenuation model of electromagnetic waves in underground media, providing a reasonable search starting point for subsequent optimization.

[0013] In an optional implementation, the method further includes: fixing the initial values ​​of the parameters, changing the penalty factor to perform cross-validation, and determining a better penalty factor; fixing the better penalty factor, changing the initial values ​​of the parameters to perform cross-validation, and determining the parameters of a better kernel function; and updating the power line identification model based on the better penalty factor and the parameters of the better kernel function to determine the hyperplane parameters of the model.

[0014] In this invention, the power pipeline identification model is updated by further adjusting the initial values ​​of parameters and the penalty factor, thereby achieving high-precision classification and identification.

[0015] Secondly, the present invention provides a method for identifying power pipelines, the method comprising: acquiring echo data of unknown pipelines detected by ground penetrating radar; performing layering, layer filtering, gain compensation, and feature extraction on the echo data to obtain echo features; inputting the echo features into a power pipeline identification model constructed by the first aspect and any one of the power pipeline identification model construction methods of the first aspect to obtain the pipeline category corresponding to the echo data; determining the location and size of the pipeline based on the peak value and half-width at half-maximum time and wave velocity of the echo data, and correcting the size based on the pipeline category.

[0016] Thirdly, the present invention provides a power pipeline identification model construction device, the device comprising: a first data acquisition module, used to acquire echo data, the dielectric constant of the underground soil, and the pipeline category when a ground penetrating radar detects a known underground pipeline; a preprocessing module, used to layer the echo data according to the wave velocity calculated by the dielectric constant and the transmission time, and to perform layered filtering and gain compensation on the layered echo data to obtain preprocessed echo data; a sample construction module, used to construct a cutoff time window based on the prior burial depth of the pipeline to extract features from the preprocessed echo data, and to construct a feature vector by combining the prior burial depth of the pipeline and the dielectric constant, wherein the feature vector serves as a sample and the pipeline category serves as a label to constitute a training sample set; a sample weighting module, used to construct a penalty weight for the sample based on the spatial local density of the sample to obtain a weighted sample set; and a training module, used to train a support vector machine model using the weighted sample set to obtain a power pipeline identification model.

[0017] In one optional implementation, the preprocessing module is specifically used for: determining a time zero point based on the direct peak value in the echo data; determining the depth corresponding to the echo data based on the time zero point, the wave velocity determined by the dielectric constant, and the time corresponding to the echo data, and layering the echo data based on the depth; performing layered filtering on the echo data at different depths according to different frequency ranges, wherein the upper and lower frequency limits in the frequency range decrease as the depth increases; and performing gain compensation on the layered filtered echo data using an exponential gain function, wherein the exponential gain function is a function of gain changing with time, and the coefficient of the exponential gain function is the skin attenuation coefficient.

[0018] In one optional implementation, the sample construction module is specifically used to: construct an interception window based on the ratio of the pipeline's prior burial depth to the wave velocity; intercept the preprocessed echo data based on the interception window, and perform principal component analysis on the intercepted signal to obtain principal component features; construct a feature vector based on the principal component features, the pipeline's prior burial depth, and the dielectric constant; and construct a training sample set using the feature vector as samples and the pipeline category as labels.

[0019] In one optional implementation, the sample weighting module is specifically used to: determine the penalty weight of each sample based on the number of samples within a preset radius centered on the sample and the penalty factor; and construct a weighted sample set based on the penalty weight, the sample, and the label.

[0020] In one optional implementation, the training module is specifically used to: perform feature normalization processing on the samples in the weighted sample set to obtain a processed sample set; determine the initial parameter values ​​of the kernel function in the support vector machine model based on the attenuation model of electromagnetic waves in underground media; and train the support vector machine model using the processed sample set to obtain a power pipeline identification model.

[0021] In an optional implementation, the device further includes: an optimization module, configured to fix initial parameter values, change the penalty factor for cross-validation, and determine a better penalty factor; fix the better penalty factor, change the initial parameter values ​​for cross-validation, and determine the parameters of a better kernel function; and update the power line identification model based on the better penalty factor and the parameters of the better kernel function to determine the hyperplane parameters of the model.

[0022] Fourthly, the present invention provides a power pipeline identification device, the device comprising: a second data acquisition module for acquiring echo data of an unknown pipeline detected by ground penetrating radar; a feature extraction module for performing layering, layered filtering, gain compensation, and feature extraction on the echo data to obtain echo features; an identification module for inputting the echo features into a power pipeline identification model constructed by the power pipeline identification model construction method of the first aspect of the present invention to obtain the pipeline category corresponding to the echo data; and a quantization module for determining the position and size of the pipeline based on the peak value and half-width at half-maximum time and wave velocity of the echo data, and correcting the size based on the pipeline category.

[0023] Fifthly, the present invention provides a computer device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the power pipeline identification model construction method of the first aspect or any corresponding embodiment thereof and the power pipeline identification method of the second aspect.

[0024] In a sixth aspect, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the power line identification model construction method of the first aspect or any corresponding embodiment thereof and the power line identification method of the second aspect.

[0025] In a seventh aspect, the present invention provides a computer program product, including computer instructions, which are used to cause a computer to execute the power pipeline identification model construction method of the first aspect or any corresponding embodiment thereof and the power pipeline identification method of the second aspect. Attached Figure Description

[0026] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0027] Figure 1 This is a flowchart illustrating the method for constructing a power pipeline identification model according to an embodiment of the present invention;

[0028] Figure 2 This is a flowchart illustrating the power line identification method according to an embodiment of the present invention;

[0029] Figure 3 This is a distribution map of buried power lines and non-power lines according to an embodiment of the present invention;

[0030] Figure 4 This is a diagram showing the intelligent identification results of power lines and non-power lines according to an embodiment of the present invention;

[0031] Figure 5 This is a structural block diagram of a power pipeline identification model construction device according to an embodiment of the present invention;

[0032] Figure 6 This is a structural block diagram of a power line identification device according to an embodiment of the present invention;

[0033] Figure 7 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] According to an embodiment of the present invention, an embodiment of a method for constructing a power line identification model is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0036] This embodiment provides a method for constructing a power line identification model, which can be used in electronic devices such as computers, mobile phones, and tablets. Figure 1 This is a flowchart of a power pipeline identification model construction method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps:

[0037] Step S101: Acquire echo data, the dielectric constant of the underground soil, and the pipeline type when the ground-penetrating radar detects known underground pipelines. In this embodiment, the echo data is acquired by the ground-penetrating radar in point measurement mode. That is, the transmitting antenna of the ground-penetrating radar is controlled to emit electromagnetic waves at a specific location, and the data is generated based on the echo signal of the target received by the receiving antenna. The target is the known pipeline mentioned in this embodiment. Specifically, the known pipeline includes power lines and non-power lines. Power lines are made of metal, i.e., the pipeline type is metal; non-power lines are made of non-metallic materials, i.e., the pipeline type is non-metallic.

[0038] In addition, to facilitate the construction of multiple subsequent samples, the echo data acquired here includes echo data from multiple known pipelines. Here, "known pipelines" refers to pipelines whose material type and burial depth are known. Simultaneously, the dielectric constant of the underground soil obtained can be the dielectric constant of the soil at the location where the underground pipeline is buried. This can be determined on-site using time-domain reflectometry (TDRS). Specific determination methods are detailed in relevant technical documents and will not be elaborated upon here.

[0039] Step S102: The echo data is layered according to the wave velocity and transmission time calculated based on the dielectric constant, and the layered echo data is subjected to layered filtering and gain compensation to obtain preprocessed echo data.

[0040] Specifically, the propagation speed (i.e., wave speed) of electromagnetic waves emitted by ground-penetrating radar varies in different media. Therefore, when echo data is detected in different media, the wave speed needs to be determined based on the dielectric constant of that medium. Furthermore, the acquired echo data includes both time and amplitude data; the time can be understood as the signal transmission time. By multiplying this time by the wave speed, the detection depth corresponding to the echo signal can be obtained. The echo signals can then be layered based on the detection depths corresponding to different echo signals.

[0041] Specifically, after the echo data is layered, different frequency ranges are used to filter the echo data in each layer to ensure that clutter and other noise in the signal at different depths can be effectively removed, thereby improving the accuracy of subsequent model recognition. At the same time, gain compensation is also applied to the echo signals of each layer to ensure that the energy of deep and shallow signals is on the same order of magnitude, suppressing feature drift caused by depth differences.

[0042] Step S103 involves constructing a truncation window based on the prior burial depth of the pipeline to extract features from the preprocessed echo data. A feature vector is then constructed by combining the prior burial depth and dielectric constant of the pipeline. This feature vector serves as a sample, and the pipeline category serves as a label, forming a training sample set. Specifically, the prior burial depth can be obtained from relevant data of known pipelines. When multiple echo data points correspond to different burial depths, a truncation window for the corresponding echo signal can be constructed based on the burial depth. This truncation window allows the extraction of a segment of the echo signal containing a peak value, thus achieving the acquisition of the core echo information.

[0043] For signals captured through a time window, further feature extraction is performed. The extracted features, combined with the pipeline's prior burial depth and dielectric constant, constitute a sample. This makes the sample physically interpretable, significantly reduces depth errors caused by medium uncertainties, better reflects pipeline characteristics, and avoids aliasing. Furthermore, this embodiment uses the acquired pipeline category as the label for each sample and combines these samples to construct a training sample set. This enables the subsequently trained model to identify pipeline categories.

[0044] Step S104: Construct penalty weights for the samples based on the spatial local density of the samples to obtain a weighted sample set; specifically, in this embodiment, spatial local density is introduced as a penalty weight for the samples, thereby enhancing the learning of sparse pipelines through this weighting strategy.

[0045] Step S105: The weighted sample set is used to train the Support Vector Machine (SVM) model to obtain the power pipeline identification model. The Support Vector Machine (SVM) is a supervised learning-based classification algorithm whose goal is to find a hyperplane that can separate samples of different classes in the training data. Specifically, in this embodiment, the core objective of the SVM model is to find the optimal classification hyperplane in the feature space, accurately separating power pipelines of different materials and maximizing the margin between the two classes of samples and the hyperplane. Mathematically, this objective is a constrained optimization problem. To solve this type of constrained optimization problem, it needs to be transformed into an unconstrained optimization problem using the "Lagrange multiplier method," with the introduced auxiliary parameters represented as Lagrange multipliers. (Each training sample corresponds to one) Finally, by solving the transformed optimization problem, the result is obtained. The value is the contribution weight of the sample to the optimal classification hyperplane.

[0046] Therefore, when training a support vector machine model using a weighted sample set, the SVM solver can be invoked, the samples can be input, and the Lagrange multipliers returned by the solver can be used. This constitutes the initial model M0 (i.e., the power line identification model).

[0047] The power pipeline identification model construction method provided in this invention, through layered filtering and gain compensation of echo data, ensures that clutter and other noise in signals at different depths can be effectively removed, improving the accuracy of subsequent model identification. Simultaneously, gain compensation processes the echo signal of each layer separately, ensuring that the energy of deep and shallow signals are on the same order of magnitude, suppressing feature drift caused by depth differences. Prior burial depth and dielectric constant are added to the features extracted from the echo signal, making the sample physically interpretable, significantly reducing depth errors caused by medium uncertainty, and better reflecting pipeline characteristics. Furthermore, spatial local density is introduced as a penalty weight for the sample, thereby strengthening the learning of sparse pipelines through this weighted strategy. The pipeline identification model trained using the weighted training samples constructed in the above manner can significantly improve the model's identification accuracy and applicability.

[0048] This embodiment provides a method for constructing a power pipeline identification model, which includes the following steps:

[0049] Step S201: Obtain echo data, the dielectric constant of the underground soil, and the pipeline type when using ground-penetrating radar to detect known underground pipelines; for details, please refer to [link to relevant documentation]. Figure 1 Step S101 of the illustrated embodiment will not be described again here.

[0050] Step S202: The echo data is layered according to the wave velocity and transmission time calculated based on the dielectric constant, and the layered echo data is subjected to layered filtering and gain compensation to obtain preprocessed echo data.

[0051] Specifically, the above steps include:

[0052] Step S2021: Determine the zero point of time based on the peak value of the direct wave in the echo data; specifically, the echo of the ground penetrating radar will have a stronger signal near the starting time, which is called the direct wave. The peak value of the direct wave is used as the 0ns time, so that the detection depth can be more accurate.

[0053] Step S2022: Based on the zero time point, determine the depth corresponding to the echo data according to the wave velocity determined by the dielectric constant and the time corresponding to the echo data, and layer the echo data based on the depth; specifically, after determining the zero time point, the time in the echo data is based on this zero time point. Meanwhile, the wave velocity is calculated using the following formula:

[0054]

[0055] In the formula, c represents the speed of electromagnetic waves in a vacuum, i.e., the speed of light. This represents the dielectric constant.

[0056] After determining the time and wave velocity, the time corresponding to the wave peak in the echo data can be obtained. Multiplying this time by the wave velocity yields the depth. Then, the echo data is layered according to depth ranges such as 0m to 2m, 2m to 5m, or greater than 5m. In other words, the echo data is assigned to the corresponding layer based on its depth. For example, if an echo data point corresponds to a depth of 1.5 meters, it would be assigned to the 0m to 2m layer. It should be noted that this layering method is only an example. In practical applications, it can be set based on the depths corresponding to multiple echo data points.

[0057] Step S2023 involves performing layered filtering on the echo data at different depths according to different frequency ranges. The upper and lower limits of the frequency range decrease as the depth increases. Specifically, after layering the received echo data, different bandpass filtering ranges are used for each layer. Furthermore, other noise reduction methods can be employed to remove noise from different layers. When determining the bandpass filtering range, the center frequency and upper limit of the bandpass filtering range gradually decrease with increasing detection depth; that is, high-frequency bandpass is used for shallow layers, and low-frequency bandpass is used for deeper layers. For shallow echo data, which is mainly affected by ground interference signals (multiple high-frequency signals), high-frequency filtering is used. For deep echo data, which is mainly affected by reflections from deep soil interfaces and geological structure interference (most of which are low-frequency), low-frequency filtering is used.

[0058] For example, the bandpass filter range for echo data from 0m to 2m is 100 MHz–600 MHz; the bandpass filter range for echo data from 2m to 5m is 50 MHz–400 MHz, and wavelet soft thresholding can also be used for noise reduction; the bandpass filter range for echo data >5m is 0 MHz–300 MHz, and adaptive background removal filtering is used.

[0059] Step S2024: Gain compensation is performed on the layered filtered echo data using an exponential gain function. The exponential gain function is a function of gain changing with time, and its coefficients are skin attenuation coefficients. Specifically, the exponential gain function used in this embodiment is expressed as follows:

[0060]

[0061] In the formula, t represents time. , Indicates the skin attenuation coefficient. f represents the center frequency of the ground-penetrating radar antenna, such as 200MHz or 300MHz. Indicates the magnetic permeability of the soil medium. This indicates the electrical conductivity of the soil medium.

[0062] When using an exponential gain function to compensate for the gain of layered echo data, the echo data can be multiplied by the exponential gain function. Since the exponential gain function changes over time, multiplying the echo data by this function can provide different levels of gain for data at different depths, making the intensity of data at each depth more uniform. For example, this can make the energy of deep signals and shallow signals of the same order of magnitude, suppressing feature drift caused by depth differences.

[0063] Step S203: Based on the prior burial depth of the pipeline, a cutoff time window is constructed to extract features from the preprocessed echo data, and a feature vector is constructed by combining the prior burial depth of the pipeline and the dielectric constant. The feature vector is used as a sample and the pipeline category is used as a label to form a training sample set.

[0064] Specifically, step S203 includes:

[0065] Step S2031: Construct an extraction window based on the ratio of the prior burial depth of the pipeline to the wave velocity. Specifically, as mentioned above, this extraction window allows the extraction of a segment of the echo signal containing peak values, thus achieving the acquisition of the core echo information. Based on this, the extraction window constructed in this embodiment is represented as [d / v–10 ns, d / v+10 ns], where d represents the prior burial depth of the pipeline and v represents the wave velocity. This extraction window enables the extraction of signals containing peak values ​​from the echo data. It should be noted that the prior burial depth of the pipeline and the wave velocity may differ for different echo data, and can be determined based on the actual situation when acquiring the echo data in step S101.

[0066] Step S2032: Based on the truncation window, the preprocessed echo data is truncated, and principal component analysis is performed on the truncated signal to obtain principal component features. Specifically, for the echo data truncated using the corresponding truncation window, principal component analysis algorithm is used for processing to achieve feature extraction. In this embodiment, when using the principal component analysis algorithm, the top k principal components with a cumulative contribution rate ≥ 95% are retained, and k ≤ 10 is restricted. The specific processing procedure of principal component analysis can be implemented with reference to relevant technologies, and will not be elaborated here.

[0067] Step S2033: Construct a feature vector based on the principal component features, the prior burial depth of the pipeline, and the dielectric constant; specifically, the constructed feature vector is represented as: F=[PC1,…,PC1…]. k , d, ] T This allows for the explicit introduction of physical constraints on depth and dielectric constant. Where PC1,…,PC… k , represents the extracted principal component features, and d represents the prior burial depth of the pipeline. This represents the dielectric constant.

[0068] Step S2034: Construct a training sample set using the feature vector as samples and the pipeline category as labels. This training sample set is represented as S = {( , )}, i = 1…N, label ∈ {+1, –1} (+1: power lines, –1: non-power lines). Let F represent the eigenvector.

[0069] Step S204: Construct penalty weights for the samples based on the spatial local density of the samples to obtain a weighted sample set.

[0070] Specifically, step S204 includes:

[0071] Step S2041: Determine the penalty weight for each sample based on the number of samples within a preset radius centered on the sample and the penalty factor; specifically, the spatial local density is represented by ρ( ) indicates that in a three-dimensional coordinate system, with The ratio of the number of samples N to the volume of a sphere centered at a radius of 0.5m is given. The radius of the sphere can be adjusted based on actual conditions. When determining ρ( After that, the penalty weight is expressed as:

[0072]

[0073] In the formula, Based on the normalized feature scale determination This represents the maximum value among all spatial local densities.

[0074] Step S2042: Construct a weighted sample set based on the penalty weights, samples, and labels. Specifically, the weighted sample set is represented as {( , , )}.

[0075] Step S205: The support vector machine model is trained using the weighted sample set to obtain the pipeline identification model.

[0076] Specifically, step S205 includes:

[0077] Step S2051 involves performing feature normalization on the samples in the weighted sample set to obtain a processed sample set. Specifically, before using the obtained samples to train the model, they need to be normalized to prevent the kernel function in the support vector machine model from becoming ineffective due to differences in units. Before normalization, the mean and variance of the samples in the sample set are obtained. For example, the mean and variance can be packaged into a persistent object (e.g., a .pkl or .model file) for subsequent normalization processing of each sample. This normalization process is expressed by the following formula:

[0078]

[0079] In the formula, x represents the sample, and m represents the mean. Indicates variance.

[0080] Step S2052: Determine the initial values ​​of the kernel function parameters in the support vector machine model based on the attenuation model of electromagnetic waves in underground media; specifically, in the support vector machine model, this embodiment uses the radial basis function as the kernel function, which is expressed by the following formula:

[0081]

[0082] In the formula, for the parameter Its initial value is determined by the attenuation model of electromagnetic waves in underground media. This provides a reasonable search starting point for subsequent optimization. Specifically, the parameters... initial value The following formula is used to determine it:

[0083]

[0084] In the formula, Skin depth represents the depth to which an electromagnetic wave penetrates when its amplitude decays to 1 / e (≈37%) of the surface value, reflecting the effective range of the signal in the underground medium. Angular frequency, This is the center frequency of the ground-penetrating radar. Indicates the magnetic permeability of the soil medium. This indicates the electrical conductivity of the soil medium.

[0085] Step S2053: The support vector machine model is trained using the processed sample set to obtain the power pipeline identification model. Specifically, the processed sample set is represented as {( , , During training, parameters can be fixed. The remaining parameters can be set to default values. Then, the Lagrange multipliers returned by the solver... This constitutes the power pipeline identification model M0.

[0086] Step S206: Fix the initial values ​​of the parameters, change the penalty factor and perform cross-validation to determine a better penalty factor; fix the better penalty factor, change the initial values ​​of the parameters and perform cross-validation to determine the parameters of a better kernel function; based on the better penalty factor and the parameters of the better kernel function, update the power line identification model and determine the hyperplane parameters of the model.

[0087] Specifically, for the parameters in the model and punishment factors After obtaining the pipeline identification model, it can be further optimized to update the power pipeline identification model.

[0088] The optimization process includes (the optimization process for the penalty factor) (Using C as a denoting)

[0089] (1) Fixed Five-fold cross-validation is performed with a step size of 0.5 within the range log2C∈[–5,5], and the option that maximizes the F1-score is selected. ;

[0090] (2) Fix C= In log2 ∈[log2(0.5 ), log2(2 Within the range, perform 5-fold cross-validation again with a step size of 0.2, and select... ;like Exceeding [0.5] ,2 ], then Retreat to the nearest boundary to ensure It will not deviate from the range of physical meaning;

[0091] (3), using ( , M0 was retrained to obtain a high-precision intelligent identification model for power pipelines, M1.

[0092] This embodiment provides a method for identifying power pipelines, such as... Figure 2 As shown, the method includes the following steps:

[0093] Step S301: Obtain echo data of unknown pipelines detected by ground penetrating radar; specifically, the unknown pipelines detected in this embodiment refer to pipelines that are known to exist underground, but whose relevant parameters, such as burial depth, pipeline type and other parameters, are unknown and need to be determined by echo data.

[0094] Step S302 involves layering, layered filtering, gain compensation, and feature extraction of the echo data to obtain echo features. Specifically, upon obtaining the echo data detected by ground-penetrating radar, the electromagnetic wave propagation velocity is determined based on the pre-acquired dielectric constant of the pipeline location. Simultaneously, zero-point calibration is performed on the acquired echo data to determine the time within the echo data, such as the time corresponding to the wave peak, based on the zero point (t=0 ns). This time is then multiplied by the wave velocity to obtain the depth. Subsequently, the echo data is layered based on the depth, and layered filtering, gain compensation, and feature extraction are performed on the layered data.

[0095] For layered filtering, the echo data is filtered according to the range to which it belongs after being layered. For example, if the layered data belongs to the 2m-5m range, a bandpass filter with a range of 50 MHz–400 MHz and wavelet soft thresholding denoising are used. After filtering, the echo data is multiplied by an exponential gain function to achieve gain compensation. Then, a truncation window is constructed based on the calculated depth (wave velocity multiplied by time) and wave velocity. After truncation of the echo data using the truncation window, principal component analysis is performed. The obtained principal component features, along with depth and dielectric constant, constitute the echo features.

[0096] Step S303: Input the echo features into the power pipeline identification model constructed by the power pipeline identification model construction method of the above embodiment to obtain the pipeline category corresponding to the echo data; specifically, the pipeline category obtained here is either a power pipeline or a non-power pipeline.

[0097] Step S304: Determine the location and size of the pipeline based on the peak value and half-width at half-maximum (HWHM) of the echo data, as well as the wave velocity, and correct the size based on the pipeline category. Specifically, this embodiment mainly focuses on the relevant parameters of power pipelines. Therefore, when the pipeline category is determined to be a power pipeline through step S303, other parameters of the pipeline are determined based on the echo data. The pipeline location is determined by the calculated depth, while the pipeline size, such as the diameter, can be determined using relevant techniques. For example, it can be determined using the time corresponding to the HWHM of the echo data, and the wave velocity, etc. Furthermore, after determining the size, the pipeline size is corrected based on the pipeline category. The correction factor is 1.1 for non-power pipelines and 0.8 for power pipelines. Multiplying the calculated pipeline size by this correction factor achieves the size inversion, reducing the size error from ±30mm to ±5mm.

[0098] As a specific application embodiment of the present invention, the power pipeline identification method uses a GR-IV type ground penetrating radar host and ground penetrating radar antennas of different frequencies. The acquisition parameters are set as follows: (1) the center frequency of the antenna is 400MHz; (2) the sampling time window is 100ns; (3) the sampling rate is 1024 points. The pipeline identification method is used to detect pipelines. During detection, a 400MHz antenna is used to detect along the vertical pipeline direction using point measurement mode. The obtained echo data is subjected to zero-point calibration, filtering and gain processing in sequence. The preprocessed data is imported into the pipeline identification model to intelligently identify the pipeline information, and the intelligent identification result is obtained. In order to verify the pipeline identification method, such as Figure 3 As shown in the figure, in this embodiment, 10 pipelines were buried in the experimental pit, and the above-described method was used for detection. The detection results are as follows. Figure 4 As shown.

[0099] This embodiment also provides a power line identification model construction device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0100] This embodiment provides a power pipeline identification model construction device, such as... Figure 5 As shown, it includes:

[0101] The first data acquisition module 41 is used to acquire echo data, the dielectric constant of the underground soil, and the pipeline type when the ground penetrating radar detects known underground pipelines.

[0102] The preprocessing module 42 is used to layer the echo data according to the wave velocity and transmission time calculated based on the dielectric constant, and to perform layered filtering and gain compensation on the layered echo data to obtain preprocessed echo data.

[0103] The sample construction module 43 is used to construct an interception window based on the prior burial depth of the pipeline to extract features from the preprocessed echo data, and to construct a feature vector by combining the prior burial depth of the pipeline and the dielectric constant. The feature vector serves as a sample and the pipeline category serves as a label to form a training sample set.

[0104] The sample weighting module 44 is used to construct the penalty weight of the sample based on the spatial local density of the sample to obtain a weighted sample set;

[0105] Training module 45 is used to train the support vector machine model using the weighted sample set to obtain the power pipeline identification model.

[0106] In one optional implementation, the preprocessing module is specifically used for: determining a time zero point based on the direct peak value in the echo data; determining the depth corresponding to the echo data based on the time zero point, the wave velocity determined by the dielectric constant, and the time corresponding to the echo data, and layering the echo data based on the depth; performing layered filtering on the echo data at different depths according to different frequency ranges, wherein the upper and lower frequency limits in the frequency range decrease as the depth increases; and performing gain compensation on the layered filtered echo data using an exponential gain function, wherein the exponential gain function is a function of gain changing with time, and the coefficient of the exponential gain function is the skin attenuation coefficient.

[0107] In one optional implementation, the sample construction module is specifically used to: construct an interception window based on the ratio of the pipeline's prior burial depth to the wave velocity; intercept the preprocessed echo data based on the interception window, and perform principal component analysis on the intercepted signal to obtain principal component features; construct a feature vector based on the principal component features, the pipeline's prior burial depth, and the dielectric constant; and construct a training sample set using the feature vector as samples and the pipeline category as labels.

[0108] In one optional implementation, the sample weighting module is specifically used to: determine the penalty weight of each sample based on the number of samples within a preset radius centered on the sample and the penalty factor; and construct a weighted sample set based on the penalty weight, the sample, and the label.

[0109] In one optional implementation, the training module is specifically used to: perform feature normalization processing on the samples in the weighted sample set to obtain a processed sample set; determine the initial parameter values ​​of the kernel function in the support vector machine model based on the attenuation model of electromagnetic waves in underground media; and train the support vector machine model using the processed sample set to obtain a power pipeline identification model.

[0110] In an optional implementation, the device further includes: an optimization module, configured to fix initial parameter values, change the penalty factor for cross-validation, and determine a better penalty factor; fix the better penalty factor, change the initial parameter values ​​for cross-validation, and determine the parameters of a better kernel function; and update the power line identification model based on the better penalty factor and the parameters of the better kernel function to determine the hyperplane parameters of the model.

[0111] This embodiment also provides a power line identification device, such as... Figure 6 As shown, the device includes:

[0112] The second data acquisition module 51 is used to acquire echo data of ground penetrating radar detecting unknown pipelines;

[0113] Feature extraction module 52 is used to perform layering, layered filtering, gain compensation and feature extraction on the echo data to obtain echo features;

[0114] The identification module 53 is used to input the echo features into the power pipeline identification model constructed by the power pipeline identification model construction method described in the above embodiment to obtain the pipeline category corresponding to the echo data;

[0115] The quantization module 54 is used to determine the position and size of the pipeline based on the peak value and half-width at half-maximum time and wave velocity of the echo data, and to correct the size based on the pipeline type.

[0116] Further functional descriptions of the above modules are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0117] This invention also provides a computer device having the above-described features. Figure 5 The power line identification model building device shown and Figure 6 The power line identification device shown.

[0118] Please see Figure 7 , Figure 7 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 7 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 7 Take a processor 10 as an example.

[0119] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0120] The memory 20 stores instructions executable by at least one processor 10 to cause at least one processor 10 to perform the method shown in the above embodiments.

[0121] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device as shown by a landing page for an app. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, which can be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0122] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0123] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0124] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0125] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0126] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for constructing a power line identification model, characterized by, The method comprises: acquiring echo data when ground penetrating radar detects underground known pipelines, dielectric constant of underground soil and pipeline category; layering the echo data according to wave velocity calculated based on the dielectric constant and transmission time, and performing layer filtering and gain compensation on the layered echo data to obtain preprocessed echo data; based on pipeline prior depth, constructing a time window for feature extraction of the preprocessed echo data, and combining the pipeline prior depth and the dielectric constant to construct a feature vector, the feature vector as a sample and the pipeline category as a label constitute a training sample set; based on the spatial local density of the sample, constructing a penalty weight of the sample to obtain a weighted sample set; training a support vector machine model using the weighted sample set to obtain a power pipeline identification model; layering the echo data according to wave velocity calculated based on the dielectric constant and transmission time, and performing layer filtering and gain compensation on the layered echo data to obtain preprocessed echo data, comprising: determining a time zero point according to the direct wave peak value in the echo data; based on the time zero point, determining the depth corresponding to the echo data according to the wave velocity determined by the dielectric constant and the time corresponding to the echo data, and layering the echo data based on the depth; performing layer filtering on the echo data at different depths according to different frequency ranges, the upper and lower limits of the frequency range decreasing as the depth increases; performing gain compensation on the layered filtered echo data using an exponential gain function, the exponential gain function being a function of gain changing with time, and the coefficient of the exponential gain function being a skin attenuation coefficient; based on pipeline prior depth, constructing a time window for feature extraction of the preprocessed echo data, and combining the pipeline prior depth and the dielectric constant to construct a feature vector, the feature vector as a sample and the pipeline category as a label constitute a training sample set, comprising: constructing a time window according to the ratio of pipeline prior depth and the wave velocity; based on the time window, performing signal interception on the preprocessed echo data, and performing principal component analysis on the intercepted signal to obtain principal component features; based on the principal component features, pipeline prior depth and the dielectric constant, constructing a feature vector; constructing a training sample set with the feature vector as a sample and the pipeline category as a label.

2. The method of claim 1, wherein, based on the spatial local density of the sample, constructing a penalty weight of the sample to obtain a weighted sample set, comprising: based on the number of samples within a preset radius centered on the sample and a penalty factor, determining the penalty weight of each sample; based on the penalty weight, sample and label, constructing a weighted sample set.

3. The method of claim 2, wherein, training a support vector machine model using the weighted sample set to obtain a power pipeline identification model, comprising: performing feature normalization processing on the samples in the weighted sample set to obtain a processed sample set; based on the attenuation model of electromagnetic waves in underground media, determining the initial value of the kernel function in the support vector machine model; training the support vector machine model using the processed sample set to obtain a power pipeline identification model.

4. The method of claim 3, wherein, The method further comprises: Fix the initial value of the parameter, change the penalty factor for cross-validation, and determine the optimal penalty factor; Fix the optimal penalty factor, change the initial value of the parameter for cross-validation, and determine the optimal parameter of the kernel function; Based on the optimal penalty factor and the optimal parameter of the kernel function, update the power pipeline identification model to determine the hyperplane parameter of the model.

5. A power line identification method characterized by, The method comprises: Obtain echo data of a ground penetrating radar detecting an unknown pipeline; Layer the echo data, filter the layered echo data, compensate the gain, and extract the features to obtain echo features; Input the echo features into the power pipeline identification model constructed by the power pipeline identification model construction method of any one of claims 1-4 to obtain a pipeline category corresponding to the echo data; Determine the position and size of the pipeline according to the time corresponding to the peak value and half-height width of the echo data and the wave velocity, and correct the size based on the pipeline category.

6. An electric power line identification model construction apparatus characterized by comprising: The device comprises: A first data acquisition module configured to acquire echo data of a ground penetrating radar detecting an underground known pipeline, a dielectric constant of underground soil, and a pipeline category; A preprocessing module configured to layer the echo data according to a wave velocity calculated based on the dielectric constant and a transmission time, and to filter the layered echo data and compensate the gain to obtain preprocessed echo data; A sample construction module configured to extract features from the preprocessed echo data based on a pipeline prior depth to construct a time window for signal interception, and to construct a feature vector based on the pipeline prior depth and the dielectric constant, the feature vector serving as a sample and the pipeline category serving as a label to form a training sample set; A sample weighting module configured to construct a penalty weight of the sample based on a spatial local density of the sample to obtain a weighted sample set; A training module configured to train a support vector machine model using the weighted sample set to obtain a power pipeline identification model; The preprocessing module is specifically configured to determine a time zero point according to a direct wave peak value in the echo data, determine a depth corresponding to the echo data based on a wave velocity determined based on the dielectric constant and a time corresponding to the echo data based on the time zero point, and layer the echo data based on the depth; filter the layered echo data at different depths according to different frequency ranges, the upper limit and lower limit of the frequency range decreasing as the depth increases; compensate the gain of the layered filtered echo data using an exponential gain function, the exponential gain function being a function of gain changing with time, and a coefficient of the exponential gain function being a skin attenuation coefficient; The sample construction module is specifically configured to construct a time window according to a ratio of the pipeline prior depth and the wave velocity, intercept signals from the preprocessed echo data based on the time window, and perform principal component analysis on the intercepted signals to obtain principal component features, construct a feature vector based on the principal component features, the pipeline prior depth, and the dielectric constant, and construct a training sample set by taking the feature vector as a sample and the pipeline category as a label.

7. The apparatus of claim 6, wherein, The sample weighting module is specifically configured to determine a penalty weight of each sample based on the number of samples within a preset radius centered on the sample and a penalty factor; and construct a weighted sample set based on the penalty weight, the samples, and the labels.

8. The apparatus of claim 7, wherein, The training module is specifically configured to perform feature normalization processing on the samples in the weighted sample set to obtain a processed sample set; and determine initial values of parameters of a kernel function in a support vector machine model based on an attenuation model of electromagnetic waves in underground media. The support vector machine model is trained using the processed sample set to obtain the power pipeline identification model.

9. The apparatus of claim 8, wherein, The device further includes an optimization module configured to fix the initial values of the parameters, change the penalty factor to perform cross validation, determine a relatively optimal penalty factor, fix the relatively optimal penalty factor, change the initial values of the parameters to perform cross validation, determine relatively optimal parameters of the kernel function, update the power pipeline identification model based on the relatively optimal penalty factor and the relatively optimal parameters of the kernel function, and determine hyperplane parameters of the model.

10. A power line identification device, characterized by The device includes: The second data acquisition module is configured to acquire echo data of an unknown pipeline detected by the ground penetrating radar; The feature extraction module is configured to perform layering, layering filtering, gain compensation, and feature extraction on the echo data to obtain echo features; The identification module is configured to input the echo features into the power pipeline identification model constructed by the power pipeline identification model construction method of any one of claims 1-4 to obtain a pipeline category corresponding to the echo data; The quantification module is configured to determine a position and a size of the pipeline according to a time corresponding to a peak value and a half-height width of the echo data and a wave velocity, and correct the size based on the pipeline category.

Images