Underground pipeline positioning method and device based on knowledge graph and ground penetrating radar

Abstract

The invention relates to an underground pipeline positioning method and device based on a knowledge graph and a ground penetrating radar. The method comprises the following steps: acquiring a radar image acquired by a ground penetrating radar for a target pipeline, and carrying out feature extraction on the radar image to respectively obtain a total image feature and a local feature of the radar image; determining a target pipeline feature vector based on the total image feature and the local feature of the radar image; and inputting the target pipeline feature vector into a knowledge graph to obtain a prediction result of the target pipeline, nodes of the knowledge graph indicating different entities, an edge of the knowledge graph indicating a relationship between entities respectively corresponding to two nodes connected with the edge, and the prediction result of the target pipeline indicating a pipeline burying depth and a pipeline attribute of the target pipeline. According to the embodiment of the invention, effective reasoning can still be carried out by utilizing the existing pipeline knowledge in the knowledge graph in a small-sample and high-noise application scene, so that the burial depth and the attribute of the pipeline can be accurately predicted under the condition of not depending on large-scale labeled data.

Description

Technical Field

[0001] This disclosure relates to the field of pipeline location technology, and in particular to a method and apparatus for locating underground pipelines based on knowledge graphs and ground penetrating radar. Background Technology

[0002] Underground pipelines are a crucial component of urban infrastructure, widely used in water supply, power supply, gas supply, and telecommunications. Accurately determining the location and distribution of underground pipelines is essential for urban planning, construction safety, and emergency repairs. Ground-penetrating radar (GPR), a non-contact, non-destructive underground detection technology, boasts advantages such as high detection speed and wide applicability, and has been widely applied to the detection and location of underground pipelines. However, in practical applications, the effectiveness of GPR-based underground pipeline location largely depends on the performance of the radar instrument and the comprehensive capabilities of the backend data processing and analysis algorithms.

[0003] Existing methods for locating underground pipelines using ground-penetrating radar (GPR) can be mainly divided into two categories: clustering-based feature extraction methods and neural network-based feature extraction methods. Clustering-based feature extraction methods are significantly affected by the detection instrument, the detection environment, and the type of pipeline. Changing the instrument, using different environments, or measuring different pipelines can all lead to significant changes in the features of the radar image, making it difficult for the algorithm to adapt. Neural network-based feature extraction methods depend on the size and quality of the dataset, resulting in low accuracy and insufficient interpretability in typical small-sample, high-noise application scenarios. Therefore, a novel underground pipeline location method is urgently needed to address the problems of insufficient algorithm adaptability, low accuracy in typical small-sample, high-noise application scenarios, and insufficient interpretability in existing technologies. Summary of the Invention

[0004] In view of this, this disclosure proposes a method and device for locating underground pipelines based on knowledge graphs and ground penetrating radar.

[0005] According to one aspect of this disclosure, a method for locating underground pipelines based on knowledge graphs and ground-penetrating radar is provided. The method includes:

[0006] The radar images collected by ground-penetrating radar targeting the pipeline are acquired, and feature extraction is performed on the radar images to obtain the full-image features and local features of the radar images.

[0007] Based on the full-image and local features of radar images, the feature vector of the target pipeline is determined;

[0008] The feature vector of the target pipeline is input into the knowledge graph to obtain the prediction result of the target pipeline. The nodes of the knowledge graph indicate different entities, the edges of the knowledge graph indicate the relationship between the entities corresponding to the two nodes of the connecting edge, and the prediction result of the target pipeline indicates the pipeline burial depth and pipeline attributes.

[0009] In one possible implementation, feature extraction is performed on the radar image to obtain both full-image features and local features, including:

[0010] Preprocessing is performed on radar images to extract full-image features and obtain preprocessed radar images. Full-image features include one or more of the following: scanning signal, reflected signal intensity histogram, DC component, and ringing intensity.

[0011] The target detection neural network is used to process the preprocessed radar image to identify target areas containing pipeline reflection features in the preprocessed radar image.

[0012] The target area is converted into local features of the radar image.

[0013] In one possible implementation, preprocessing is performed on the radar image to extract full-image features and obtain a preprocessed radar image, including:

[0014] The scanning signal is obtained by extracting signals from radar images;

[0015] The radar image is converted into a two-dimensional grayscale image, and the histogram of reflected signal intensity is determined based on the two-dimensional grayscale image;

[0016] The DC component is extracted and removed from the two-dimensional grayscale image to obtain the image after removing the DC component;

[0017] The ringing intensity is extracted from the image after removing the DC component, and the ringing intensity is removed to obtain the image after removing the ringing.

[0018] Gain processing is performed on the image after ringing removal to obtain the preprocessed radar image.

[0019] In one possible implementation, the method further includes:

[0020] Obtain environmental information about the target pipeline, including at least one of the following: the detection area related to the target pipeline, road surface properties, soil moisture, and surrounding buildings;

[0021] Environmental features are obtained by embedding representations based on environmental information.

[0022] Based on the full-image and local features of radar images, the feature vector of the target pipeline is determined, including:

[0023] Based on full-map features, local features, and environmental features, the feature vector of the target pipeline is determined.

[0024] In one possible implementation, the entities in the knowledge graph include full-image features of radar images, local features of radar images, environmental features, pipeline attributes, and pipeline burial depth. The pipeline attributes include at least one of pipeline material, pipeline outer diameter, pipeline wall thickness, pipeline function, and dielectric constant of the medium surrounding the pipeline.

[0025] In one possible implementation, the target pipeline feature vector is input into a knowledge graph to obtain the prediction results of the target pipeline, including:

[0026] Based on the feature vector of the target pipeline and the knowledge graph, the entities in the triplet to be predicted are predicted, and the candidate entities corresponding to the target pipeline are determined. The triplet to be predicted includes entities, relations and attribute values.

[0027] Based on the feature vectors of each candidate entity and the target pipeline, the attribute values ​​in the triplet to be predicted are predicted to determine the prediction result of the target pipeline.

[0028] In one possible implementation, based on the feature vectors of each candidate entity and the target pipeline, the attribute values ​​in the triples to be predicted are predicted to determine the prediction result of the target pipeline, including:

[0029] Based on the feature vectors of each candidate entity and the target pipeline, the set of nodes associated with the target pipeline is determined in the knowledge graph;

[0030] The prediction results for the target pipeline are obtained by predicting the attribute values ​​in the triplet to be predicted based on the node set.

[0031] According to another aspect of this disclosure, an underground pipeline location device based on knowledge graphs and ground-penetrating radar is provided. The device includes:

[0032] The first determining module is used to acquire radar images collected by ground penetrating radar against target pipelines, extract features from the radar images, and obtain full-image features and local features of the radar images respectively.

[0033] The second determining module is used to determine the target pipeline feature vector based on the full-image features and local features of the radar image;

[0034] The third determination module is used to input the feature vector of the target pipeline into the knowledge graph to obtain the prediction result of the target pipeline. The nodes of the knowledge graph indicate different entities, the edges of the knowledge graph indicate the relationship between the entities corresponding to the two nodes of the connecting edge, and the prediction result of the target pipeline indicates the pipeline burial depth and pipeline attributes of the target pipeline.

[0035] In one possible implementation, the first determining module is used for:

[0036] Preprocessing is performed on radar images to extract full-image features and obtain preprocessed radar images. Full-image features include one or more of the following: scanning signal, reflected signal intensity histogram, DC component, and ringing intensity.

[0037] The target detection neural network is used to process the preprocessed radar image to identify target areas containing pipeline reflection features in the preprocessed radar image.

[0038] The target area is converted into local features of the radar image.

[0039] In one possible implementation, preprocessing is performed on the radar image to extract full-image features and obtain a preprocessed radar image, including:

[0040] The scanning signal is obtained by extracting signals from radar images;

[0041] The radar image is converted into a two-dimensional grayscale image, and the histogram of reflected signal intensity is determined based on the two-dimensional grayscale image;

[0042] The DC component is extracted and removed from the two-dimensional grayscale image to obtain the image after removing the DC component;

[0043] The ringing intensity is extracted from the image after removing the DC component, and the ringing intensity is removed to obtain the image after removing the ringing.

[0044] Gain processing is performed on the image after ringing removal to obtain the preprocessed radar image.

[0045] In one possible implementation, the device further includes:

[0046] The acquisition module is used to acquire environmental information of the target pipeline. The environmental information includes at least one of the following: the detection area related to the target pipeline, road surface properties, soil moisture, and surrounding buildings.

[0047] The embedding representation module is used to perform embedding representation based on environmental information to obtain environmental features;

[0048] The second determining module is used for:

[0049] Based on full-map features, local features, and environmental features, the feature vector of the target pipeline is determined.

[0050] In one possible implementation, the entities in the knowledge graph include full-image features of radar images, local features of radar images, environmental features, pipeline attributes, and pipeline burial depth. The pipeline attributes include at least one of pipeline material, pipeline outer diameter, pipeline wall thickness, pipeline function, and dielectric constant of the medium surrounding the pipeline.

[0051] In one possible implementation, the third determining module is used for:

[0052] Based on the feature vector of the target pipeline and the knowledge graph, the entities in the triplet to be predicted are predicted, and the candidate entities corresponding to the target pipeline are determined. The triplet to be predicted includes entities, relations and attribute values.

[0053] Based on the feature vectors of each candidate entity and the target pipeline, the attribute values ​​in the triplet to be predicted are predicted to determine the prediction result of the target pipeline.

[0054] In one possible implementation, based on the feature vectors of each candidate entity and the target pipeline, the attribute values ​​in the triples to be predicted are predicted to determine the prediction result of the target pipeline, including:

[0055] Based on the feature vectors of each candidate entity and the target pipeline, the set of nodes associated with the target pipeline is determined in the knowledge graph;

[0056] The prediction results for the target pipeline are obtained by predicting the attribute values ​​in the triplet to be predicted based on the node set.

[0057] According to another aspect of this disclosure, an underground pipeline location device based on knowledge graphs and ground penetrating radar is provided, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0058] According to another aspect of this disclosure, a non-volatile computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the above-described method.

[0059] According to embodiments of this disclosure, radar images of a target pipeline are acquired by ground-penetrating radar (GPR). Feature extraction is performed on the radar images to obtain full-image features and local features. Based on the full-image features and local features of the radar images, a feature vector of the target pipeline is determined. By inputting the target pipeline feature vector into a knowledge graph, GPR image features can be fused into the knowledge graph, and reasoning can be performed in combination with existing pipeline knowledge. This avoids relying solely on data-driven models for end-to-end learning. Even in typical small-sample, high-noise application scenarios, effective reasoning can still be performed using existing pipeline knowledge in the knowledge graph. Thus, the burial depth and pipeline attributes of the target pipeline can be accurately predicted without relying on large-scale labeled data. The knowledge graph can explicitly introduce structured knowledge in the pipeline domain, imposing prior constraints on the prediction results. This improves the adaptability of the algorithm under different detection instruments and complex environmental conditions. At the same time, the prediction results can be back-explained through corresponding entities and relationships between entities, enhancing the interpretability of the prediction results.

[0060] Other features and aspects of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0061] The accompanying drawings, which are included in and form part of this specification, illustrate exemplary embodiments, features, and aspects of this disclosure together with the specification and serve to explain the principles of this disclosure.

[0062] Figure 1 A flowchart illustrating an underground pipeline location method based on knowledge graphs and ground-penetrating radar according to an embodiment of the present disclosure is shown.

[0063] Figure 2 A schematic diagram of a pattern layer of a knowledge graph according to an embodiment of the present disclosure is shown.

[0064] Figure 3 A structural diagram of an underground pipeline location device based on a knowledge graph and ground penetrating radar according to an embodiment of the present disclosure is shown.

[0065] Figure 4 This is a block diagram illustrating an apparatus 1900 for locating underground pipelines according to an exemplary embodiment. Detailed Implementation

[0066] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0067] As used herein, the terms “comprising,” “including,” “having,” or variations thereof are open-ended and include one or more of the stated features, integrals, elements, steps, components, or functions, but do not exclude the presence or addition of one or more other features, integrals, elements, steps, components, functions, or groups thereof.

[0068] When an element is referred to as “connected,” “coupled,” “responding,” or a variation thereof relative to another element, it may be directly connected, coupled, or responding to another element, or there may be an intermediate element present.

[0069] Although the terms first, second, third, etc., may be used herein to describe various elements / operations, these elements / operations should not be limited by these terms. These terms are only used to distinguish one element / operation from another. Therefore, without departing from the teachings of the inventive concept, a first element / operation in some embodiments may be referred to as a second element / operation in other embodiments.

[0070] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0071] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0072] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, data stored, data displayed, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant regions.

[0073] Underground pipelines are a crucial component of urban infrastructure, widely used in water supply, power supply, gas supply, and telecommunications. Accurately determining the location and distribution of underground pipelines is essential for urban planning, construction safety, and emergency repairs. Ground-penetrating radar (GPR), a non-contact, non-destructive underground detection technology, boasts advantages such as high detection speed and wide applicability, and has been widely applied to the detection and location of underground pipelines. However, in practical applications, the effectiveness of GPR-based underground pipeline location largely depends on the performance of the radar instrument and the comprehensive capabilities of the backend data processing and analysis algorithms.

[0074] Existing methods for locating underground pipelines using ground-penetrating radar (GPR) can be mainly divided into two categories: clustering-based feature extraction methods and neural network-based feature extraction methods. Clustering-based feature extraction methods are significantly affected by the detection instrument, the detection environment, and the type of pipeline. Changing the instrument, using different environments, or measuring different pipelines can all lead to significant changes in the features of the radar image, making it difficult for the algorithm to adapt. Neural network-based feature extraction methods depend on the size and quality of the dataset, resulting in low accuracy and insufficient interpretability in typical small-sample, high-noise application scenarios. Therefore, a novel underground pipeline location method is urgently needed to address the problems of insufficient algorithm adaptability, low accuracy in typical small-sample, high-noise application scenarios, and insufficient interpretability in existing technologies.

[0075] In view of this, this disclosure provides a method and apparatus for locating underground pipelines based on knowledge graphs and ground-penetrating radar. The method of this disclosure acquires radar images of the target pipeline collected by ground-penetrating radar, extracts features from the radar images to obtain full-image features and local features, and determines the target pipeline feature vector based on these features. By inputting the target pipeline feature vector into a knowledge graph, ground-penetrating radar image features can be fused into the knowledge graph, and reasoning can be performed using existing pipeline knowledge. This avoids relying solely on data-driven models for end-to-end learning. Even in typical small-sample, high-noise application scenarios, effective reasoning can still be performed using existing pipeline knowledge in the knowledge graph. Therefore, it can accurately predict the burial depth and pipeline attributes of the target pipeline without relying on large-scale labeled data. The knowledge graph can explicitly introduce structured knowledge in the pipeline domain, imposing prior constraints on the prediction results, which improves the algorithm's adaptability to different detection instruments and complex environmental conditions. Furthermore, the prediction results can be explained through the corresponding entities and relationships between entities, enhancing the interpretability of the prediction results.

[0076] The underground pipeline location system based on knowledge graphs and ground-penetrating radar according to embodiments of this disclosure may include a user terminal and a software terminal. The user terminal includes an input interface and an output interface, and the software terminal may include an updatable knowledge graph and radar image feature extraction program.

[0077] In the application scenarios of this disclosure embodiment, for pipelines of unknown nature, the user can input the radar image of the pipeline through the user terminal's input interface. The user terminal's input interface also supports other optional inputs (such as detection area, road surface properties, soil moisture, surrounding buildings, etc.). After receiving the user input, the radar image feature extraction program on the software side can extract the full-image features and local features of the radar image, input them into a knowledge graph, and use the method of this disclosure embodiment to output the prediction results of the pipeline (e.g., output the predicted probability distribution of the pipeline's burial depth, material, outer diameter, function, and dielectric constant of the medium, etc.), and send them to the user terminal's output interface for display.

[0078] The underground pipeline location system based on knowledge graphs and ground-penetrating radar of this disclosure can be deployed on terminal devices or servers. Terminal devices can be any one or more of the following: mobile phones, foldable electronic devices, tablets, desktop computers, laptop computers, handheld computers, ultra-mobile personal computers (UMPCs), netbooks, cellular phones, personal digital assistants (PDAs), and in-vehicle devices, possessing wired or wireless communication capabilities. This disclosure does not impose any special limitations on the specific type of terminal device.

[0079] Servers can be located locally or in the cloud, and can be physical devices or virtual devices such as virtual machines and containers. They have wireless communication capabilities, which can be configured in the server's chip (system) or other components. Wireless communication capabilities can be implemented through mobile communication technologies such as 2G / 3G / 4G / 5G, as well as Wi-Fi, Bluetooth, frequency modulation (FM), data radio, and satellite communication; they can also communicate via wired connections to interact with other devices.

[0080] Figure 1 A flowchart illustrating an underground pipeline location method based on knowledge graphs and ground-penetrating radar according to an embodiment of this disclosure is shown. This method can be used in the aforementioned underground pipeline location system based on knowledge graphs and ground-penetrating radar, such as... Figure 1 As shown, the method may include:

[0081] Step S101: Obtain radar images of the target pipeline collected by ground penetrating radar, extract features from the radar images, and obtain the full-image features and local features of the radar images respectively.

[0082] Ground penetrating radar is a non-contact electromagnetic detection device used to detect underground pipelines. The target pipeline can be represented as the underground pipeline facility to be detected, and the radar image can be a binary file in 16-bit or 32-bit format.

[0083] The full-image features of a radar image can represent the overall information of the entire image and can be extracted during subsequent preprocessing. Local features of a radar image can characterize target areas containing pipeline reflection information and can be obtained through target detection or local region analysis of the preprocessed radar image.

[0084] In one possible implementation, step S101 includes:

[0085] Preprocessing is performed on the radar image to extract full-image features and obtain a preprocessed radar image. The preprocessed radar image is then processed using a target detection neural network to identify target regions containing pipeline reflection features. Finally, the target regions are converted into local features of the radar image.

[0086] The full-image features may include one or more of the following: scan signal (such as A-scan), histogram of reflected signal intensity, DC component, and ringing intensity.

[0087] In one possible implementation, preprocessing is performed on the radar image to extract full-image features and obtain a preprocessed radar image, including:

[0088] Signal extraction is performed on the radar image to obtain the scanning signal; the radar image is converted into a two-dimensional grayscale image, and the histogram of reflected signal intensity is determined based on the two-dimensional grayscale image; the DC component is extracted and removed from the two-dimensional grayscale image to obtain the image after removing the DC component; the ringing intensity is extracted and removed from the image after removing the DC component to obtain the image after removing the ringing; gain processing is performed on the image after removing the ringing to obtain the preprocessed radar image.

[0089] The original radar image data (such as the binary file mentioned above) can be unpacked to obtain a two-dimensional array (called B-scan). The horizontal axis represents the displacement along the radar survey line, and the vertical axis represents the time axis. The value of each pixel in the B-scan reflects the electromagnetic wave intensity at the corresponding location and time. Based on this, each column of the B-scan signal can be used as a vector to extract the scanning signal (A-scan), which is used to characterize the change of electromagnetic waves received by the radar antenna at that survey line location over time.

[0090] B-scan can also be converted into a two-dimensional grayscale image for subsequent image processing and feature extraction. Based on the two-dimensional grayscale image, statistical methods can be used to calculate the distribution of pixel grayscale values, resulting in a histogram of reflected signal intensity, which characterizes the overall intensity features of electromagnetic wave reflection in the image. The histogram of reflected signal intensity forms a vector within the grayscale value range of 0 to 255, used to describe the global reflection features of the image.

[0091] Next, multiple filters can be used to process the two-dimensional grayscale image to remove interference items such as ringing, noise, and signal attenuation caused by factors such as antenna-to-ground distance, antenna sidelobes, external electromagnetic interference, and electromagnetic wave propagation attenuation. This will fully expose the pipeline reflection characteristics and obtain the preprocessed radar image.

[0092] This process involves performing initial time correction on the 2D grayscale image to align the starting points of each signal column and ensure consistency along the time axis. Next, the average value of each signal column can be calculated as the DC component, with each average value corresponding to an element within the DC component. Furthermore, this DC component is removed from each signal column to eliminate superimposed white noise, resulting in an image with the DC component removed.

[0093] Alternatively, a moving average filter, bandpass filter, or destriping algorithm can be used to extract and remove ringing intensity from the image after removing the DC component, in order to eliminate strip noise caused by ground reflection and antenna sidelobes, and obtain the image after removing the ringing.

[0094] Time gain processing can be applied to the image after ringing removal to amplify signals from deeper layers and make the reflection characteristics of underground pipelines more apparent. An example of time gain processing can be found in the following formula:

[0095]

[0096] in, This represents the preprocessed radar image. This represents the image after ringing has been removed. The time point at which the signal was received. For the time resolution of ground-penetrating radar, The start time of the signal. , , These are preset parameters used to control the gain amplitude and gain variation curve.

[0097] In the above preprocessing process, all radar images can use the same processing flow and parameter settings to ensure that the preprocessing steps do not introduce additional signal features, thereby ensuring the accuracy and consistency of subsequent feature extraction.

[0098] Target detection neural networks can be used to locate target areas that may contain pipeline reflection features from preprocessed radar images. Pipeline reflection features refer to the areas with specific geometric shapes and intensity distribution characteristics formed in radar images after underground pipelines reflect electromagnetic waves from ground-penetrating radar. They are usually represented by hyperbolic reflection trajectories. The position of the apex, curvature, and intensity distribution of the hyperbola are related to the burial depth, outer diameter, and surrounding medium conditions of the pipeline.

[0099] Object detection neural networks can be derived by modifying existing neural network architectures. For example, the YOLOv8 open-source object detection network architecture can be simplified and adjusted. The original YOLOv8 detection head is typically used to simultaneously perform object localization and object category classification tasks, and its structure is relatively complex. However, in this embodiment, the object detection neural network only needs to detect whether there are regions containing pipeline reflection features in the radar image, without needing to distinguish between multiple target categories. Therefore, the YOLOv8 detection head can be simplified and replaced with a detection head structure consisting of at least two convolutional layers, so that the feature tensor output by the network backbone can be directly mapped to a tensor form used to calculate the loss function. Where N represents the batch size, B represents the number of anchor frames, H and W represent the height and width of the feature map, respectively. The first four parameters are used to specify the size information of the detected region containing pipeline reflection features. The fifth parameter is used to characterize the confidence of the region predicted by the target detection neural network. When the confidence of the target region output by the target detection neural network is greater than a preset threshold, the region is regarded as the target region.

[0100] The target detection neural network can be obtained by training on the above framework using a sample dataset. The sample dataset may include: (1) radar image samples obtained by ground penetrating radar in urban environments to measure underground pipelines; (2) radar image samples simulated under certain boundary conditions, which can fit the common application scenarios of urban underground pipelines as closely as possible; (3) radar image samples generated by generative networks in existing technologies, so as to expand the training data scale and enhance the generalization ability of the model when the sample data of the first two items is insufficient.

[0101] The identified target region can be processed by feature representation. For example, the two-dimensional feature map corresponding to the target region can be converted into a one-dimensional vector through a flattening operation, which can be used as a local feature of the radar image. The flattening operation can be implemented by convolutional neural networks or fully connected layers in existing technologies to ensure the effective expression of local features in the vector space.

[0102] In one possible implementation, the method further includes:

[0103] Obtain environmental information of the target pipeline; perform embedding representation based on the environmental information to obtain environmental features.

[0104] The environmental information can be unstructured knowledge information that can be optionally input by the user, used to characterize the external environmental conditions when the ground-penetrating radar acquires radar images of the target pipeline. The environmental information may include at least one of the following: the detection area related to the target pipeline, road surface properties, soil moisture, and surrounding structures.

[0105] The detection area is used to indicate the functional area type where the target pipeline is located, such as residential area, main road, industrial area, etc.; the road surface properties are used to indicate the material type of the road surface above the target pipeline, such as cement, asphalt, soil, etc.; soil moisture is used to indicate the water content of the soil around the target pipeline; surrounding buildings and structures are used to indicate whether there are buildings or municipal facilities near the target pipeline, such as residential buildings, industrial buildings, manhole covers, water pumping stations, etc.

[0106] Embedded representation can be implemented based on existing feature encoding or representation learning methods. Through embedded representation, environmental information can be converted into a unified vector representation, so that different types of environmental information can be mapped to the same feature space, thereby obtaining environmental features.

[0107] This can further improve the accuracy of subsequent prediction results and increase the accuracy of pipeline identification.

[0108] Step S102: Determine the target pipeline feature vector based on the full-image features and local features of the radar image.

[0109] The target pipeline feature vector can be used as input for subsequent knowledge graphs. For example, the full-image features and local features obtained above can be combined (e.g., by splicing) as the target pipeline feature vector.

[0110] When the user inputs environmental information as described above, step S102 includes:

[0111] Based on full-map features, local features, and environmental features, the feature vector of the target pipeline is determined.

[0112] At this point, the features of the whole image, local features, and environmental features can be combined as the feature vector of the target pipeline.

[0113] Step S103: Input the target pipeline feature vector into the knowledge graph to obtain the prediction result of the target pipeline.

[0114] The knowledge graph can be a pre-constructed graph model used to represent structured knowledge related to underground pipelines. Nodes in the knowledge graph indicate different entities. Entities can include full-image features of radar images, local features of radar images, environmental features, pipeline attributes, and pipeline burial depth. Pipeline attributes include at least one of pipeline material, outer diameter, wall thickness, function, and dielectric constant of the surrounding medium. Each entity can include different attributes and attribute values, and edges in the knowledge graph can indicate the relationship between the entities corresponding to the two nodes connecting the edges.

[0115] The prediction results of the target pipeline can indicate the pipeline burial depth and pipeline properties.

[0116] According to embodiments of this disclosure, radar images of a target pipeline are acquired by ground-penetrating radar (GPR). Feature extraction is performed on the radar images to obtain full-image features and local features. Based on the full-image features and local features of the radar images, a feature vector of the target pipeline is determined. By inputting the target pipeline feature vector into a knowledge graph, GPR image features can be fused into the knowledge graph, and reasoning can be performed in combination with existing pipeline knowledge. This avoids relying solely on data-driven models for end-to-end learning. Even in typical small-sample, high-noise application scenarios, effective reasoning can still be performed using existing pipeline knowledge in the knowledge graph. Thus, the burial depth and pipeline attributes of the target pipeline can be accurately predicted without relying on large-scale labeled data. The knowledge graph can explicitly introduce structured knowledge in the pipeline domain, imposing prior constraints on the prediction results. This improves the adaptability of the algorithm under different detection instruments and complex environmental conditions. At the same time, the prediction results can be back-explained through corresponding entities and relationships between entities, enhancing the interpretability of the prediction results.

[0117] The following section introduces the methods for constructing knowledge graphs.

[0118] Data sources used to construct the knowledge graph can include current domestic standards and specifications, research papers related to ground-penetrating radar (GPR) for locating underground pipelines, and optional municipal pipeline building information modeling (BIM) data. These data sources can be uniformly organized into knowledge text to provide prior knowledge support for underground pipeline location, thereby improving the accuracy of location and identification.

[0119] To make the knowledge graph more suitable for underground pipeline detection and location tasks, the schema layer of the knowledge graph can be predefined in this embodiment.

[0120] Figure 2 This diagram illustrates a schema layer of a knowledge graph according to an embodiment of the present disclosure. The schema layer is used to define the entity types appearing in the knowledge graph and the relationships between entity types. Figure 2 As shown, the ontology (i.e., entity type) of the knowledge graph in this embodiment may include full-image features of the radar image (as shown in "Full-image Features", where full-image features are represented by A-scan, signal strength histogram, DC component, ringing intensity, etc.), local features of the radar image (as shown in "Local Features", where local features are represented by the output of the neural network), environmental features (as shown in "Surrounding Environment"), pipeline attributes, and pipeline burial depth (as shown in "Buried Depth"). Pipeline attributes may include pipeline material, pipeline outer diameter, pipeline wall thickness, pipeline function, and other ontology elements.

[0121] The "full-image features" ontology contains entities that correspond to full-image features under different radar images, and their attribute values ​​can be represented as the feature vector values ​​of the corresponding full-image features. For example, full-image feature A and full-image feature B are two different entities under the "full-image features" ontology, where the attribute value of full-image feature A is feature vector 1, and the attribute value of full-image feature B is feature vector 2.

[0122] Similarly, "local features" as an ontology contain entities that correspond to different local features output by the neural network, and their attribute values ​​can be represented as the feature vector values ​​of the corresponding local features. "Environmental features" as an ontology contain entities that correspond to different environmental features, and their attribute values ​​are the feature vector values ​​of the corresponding environmental features.

[0123] The "Pipeline Attributes" entity serves as the core, encompassing various sub-entities such as pipeline material, outer diameter, wall thickness, and function. The "Pipeline Material" entity can contain entities that are either steel pipes or non-steel pipes; this entity indicates the pipeline material, and its attribute values ​​are identical to those of the entity itself. Similarly, the "Pipeline Function" entity can contain entities that are water supply pipes, drainage pipes, etc.; this entity indicates the pipeline function, and its attribute values ​​are identical to those of the entity itself.

[0124] The "pipeline outer diameter" is the entity itself, and the entities it contains can correspond to different pipeline outer diameters. Its attribute values ​​can be represented as the corresponding nominal specification values. Similarly, the "pipeline wall thickness" is the entity itself, and the entities it contains can correspond to different pipeline wall thicknesses. Its attribute values ​​can be represented as the corresponding nominal specification values.

[0125] The "Pipeline Burial Depth" entity is a body containing entities that correspond to different pipeline burial depths, with their attribute values ​​being specific burial depth values. For example, Pipeline Burial Depth A and Pipeline Burial Depth B are two different entities under the "Pipeline Burial Depth" entity, where Pipeline Burial Depth A has an attribute value of 3 and Pipeline Burial Depth B has an attribute value of 6.

[0126] In this model layer, semantic relationships are predefined between different entity types. For example, the surrounding environment of a pipeline is related to other attributes of the pipeline (such as burial depth, outer diameter, function, etc.). For instance, pipelines near industrial areas are more likely to carry out water or fuel transportation functions; pipelines under main roads and residential areas usually have significant differences in burial depth, outer diameter, and function; road surface properties and soil moisture affect the propagation characteristics of ground-penetrating radar signals, thus affecting the overall and local features of radar images; the outer diameter and wall thickness of pipelines are usually nominal values ​​specified by standards or specifications and are related to the pipeline function; the pipeline material is related to the function it performs, and the range of materials that can be selected for pipelines with different functions is usually limited.

[0127] When constructing a knowledge graph, one can draw upon the ontology library (i.e., named entity set) specified by the aforementioned schema layer. Starting with different entity types, the knowledge text is preprocessed and entities and relations are extracted to construct a knowledge graph.

[0128] The preprocessing steps may include converting the aforementioned standards, specifications, and municipal pipeline data related to pipeline attributes, installation, and location into a readable text format; splitting the text into multiple sentences; and further dividing each sentence into several word units, wherein each sentence is mapped to a word vector sequence through word segmentation. .

[0129] Based on this, entity and relation extraction is performed on the aforementioned word vectors. Entity and relation extraction can employ rule-based methods, using pre-designed ontologies (entity types), relationships between ontologies, and ontology attributes in the schema layer to automatically extract corresponding entities, relationships between entities, and attribute values ​​corresponding to entities from the text.

[0130] Because underground pipeline detection is a relatively specialized field, there are many prior mapping relationships in the domain knowledge. For example, the ontology (entity type) corresponding to "cast iron" is "pipeline material," and the entity corresponding to the ontology "pipeline material" can only be one of a finite set (such as PE, PVC, steel, etc.). Therefore, an underground pipeline detection domain dictionary can be constructed by adopting the experience of experts and engineers to define the entity text corresponding to each ontology. In the process of entity and relation extraction, the entities in the text can be initially identified based on this underground pipeline detection domain dictionary to determine the ontology to which the entity belongs. On this basis, classification methods such as Support Vector Machine (SVM) can be used to further classify the entity and relation identification results to improve the extraction accuracy. The output of entity and relation extraction can be represented as: for each sentence Generate the corresponding tag set In this context, when there are relationships between ontologies defined in the schema layer, the corresponding entities can form knowledge graph triples. For example, if a sentence contains both "main road" and "water supply pipeline," and the schema layer defines the relationship between "surrounding environment" and "pipeline function" as "installation," where the surrounding environment is the ontology corresponding to the main road and the pipeline function is the ontology corresponding to the water supply pipeline, then a triple of "main road – installation – water supply pipeline" can be formed. The "main road" can be represented by the feature value corresponding to the environmental features. Furthermore, triples can also represent the relationship between a specific full-graph feature and a specific pipeline function.

[0131] In addition to unstructured knowledge extracted from text, knowledge graphs can also store a large number of radar image features, such as full-image features and local features obtained in the above manner, and connect them with environmental information, pipeline attributes and burial depth corresponding to the radar image acquisition.

[0132] In knowledge graphs, triples can take the form of "entity-relationship-entity" or "entity-attribute-attribute value". For example, for the entity "steel pipe", its attribute "wall thickness" can correspond to a set of nominal wall thickness values ​​of steel pipes; the feature vector values ​​corresponding to the full graph features and local features can also be stored as attribute values.

[0133] By collecting measured data or simulating real-world scenarios, ground-penetrating radar image features (including the aforementioned full-image features and local features) under different pipeline specifications and burial conditions can be obtained. For example, for a steel pipe with an outer diameter of 25 cm and a burial depth of 0.5 m, the feature vector values ​​of its corresponding full-image features and local features can be stored in the knowledge graph as the attribute values ​​of the entity under the corresponding attribute conditions.

[0134] A complete knowledge graph can be constructed based on the entities and relationships between entities indicated by each triple, as well as the relationships between entities and attributes and attribute values. This knowledge graph can express the entities and relationships between entities indicated by each triple, as well as the relationships between entities and attributes and attribute values.

[0135] The completed knowledge graph can contain both textual knowledge and radar image feature data, and be stored in a graph database, such as GraphDB, Stardog, or other database systems that support reasoning functions.

[0136] This knowledge graph possesses certain reasoning capabilities. Upon receiving new input information (i.e., the feature vector of the target pipeline), it can make predictions and output the prediction result that best matches the input features. This can be achieved using an embedding method based on representation learning. The problem to be predicted is abstracted into a prediction triple (i.e., (h′, r′, t′)). By predicting h′ and t′, the prediction result of the target pipeline is obtained, thereby enabling the identification of the underground pipeline's burial depth and attributes. Here, h′ represents the entity, r′ represents the relation, and t′ represents the attribute value.

[0137] In one possible implementation, step S103 includes:

[0138] Based on the feature vector of the target pipeline and the knowledge graph, the entities in the triplet to be predicted are predicted to determine the corresponding candidate entities of the target pipeline; based on each candidate entity and the feature vector of the target pipeline, the attribute values ​​in the triplet to be predicted are predicted to determine the prediction result of the target pipeline.

[0139] The triple to be predicted, as described above (h′, r′, t′), includes an entity, a relation, and an attribute value. During the prediction of entity h′ in the triple, r′ and t′ are known values. The feature values ​​corresponding to the full-image features, local features, and environmental features in the target pipeline feature vector can be mapped to the known attribute values ​​t′ of the corresponding entities in the triple. For example, the feature values ​​of the full-image features in the target pipeline feature vector can be mapped to the attribute value corresponding to the entity "full-image features". The relation r′ in the triple is a known value, representing the sum of relations from entity h′ in the knowledge graph to attribute value t′. For example, in the knowledge graph, starting from an entity h′, passing through relation r1 to entity x2, and from x2 passing through relation r2 to attribute value t′, r′ represents r1 and r2. h′ is the target to be predicted.

[0140] At this stage, starting from all possible other entity nodes h′ in the knowledge graph, and based on the aforementioned r′, paths can be determined from all possible other entity nodes h′ to the attribute values ​​t′ corresponding to entity nodes with known full-graph features, local features, and surrounding environment types. Here, "possible other entity nodes" refers to entity nodes in the knowledge graph that can reach attribute value t′ through a path formed by a series of relationships. For each path, each triple contained in the path is mapped to a vector, and the vector sum is calculated. The vector sum is then input into a scoring function to obtain the score corresponding to the path. Finally, the entity node h′ contained in the path with the highest score is selected as the candidate entity. The candidate entity can include pipeline material type entities and pipeline function type entities.

[0141] For example, a knowledge graph contains entities such as: full graph feature A, full graph feature B, local feature A, local feature B, environmental feature A, steel pipe, PVC pipe, water supply pipeline, gas pipeline, pipeline burial depth, pipeline outer diameter, pipeline wall thickness, dielectric constant of the surrounding medium, etc., as well as the relationships between these entities, such as "corresponding feature", "related feature", "has attribute", "applies to", etc.

[0142] Assume that the eigenvalue of the full-map feature in the target pipeline feature vector is 0.85, the eigenvalue of the local feature is 0.72, and the eigenvalue of the environmental feature is 0.63.

[0143] These feature values ​​can be mapped to the knowledge graph: The knowledge graph finds the node whose attribute value is closest to the feature value of the full-image feature, local feature, or environment feature of the target pipeline feature vector, and maps that feature value to the attribute value of that closest node. For example, since the attribute value of full-image feature A in the knowledge graph is 0.85, which is closest to the feature value of full-image feature A in the target pipeline feature vector, the feature value 0.85 of full-image feature A is mapped to the attribute value corresponding to the entity "full-image feature A". Similarly, the feature value 0.72 of local feature A is mapped to the attribute value corresponding to the entity "local feature B", and the feature value 0.63 of environment feature A is mapped to the attribute value corresponding to the entity "environment feature A".

[0144] The knowledge graph can be searched for paths from various possible entities (such as steel pipes, PVC pipes, water supply pipelines, gas pipelines, etc.) to these known attribute values. For example, the search yields path 1 as "steel pipe → has attribute → pipeline outer diameter → associated feature → full graph feature A → has attribute → 0.85", and path 2 as "PVC pipe → corresponding feature → environmental feature A → has attribute → 0.63". For each path, the vector sum is calculated for the vectors mapped to each triple contained in the path. The vector sum is then input into a scoring function to obtain the score corresponding to that path. Finally, the entity included in the path with the highest score (e.g., path 1) is selected as the candidate entity. For example, the candidate entities could be "steel pipe" and "water supply pipeline". The specific form of the scoring function in each embodiment of this application is not limited.

[0145] Next, based on the feature vectors of each candidate entity and the target pipeline, the attribute values ​​in the triplet to be predicted are predicted to determine the prediction result of the target pipeline. Here, we need to predict attribute values ​​other than those determined by the feature values ​​corresponding to the full image features, local features, and environmental features mentioned above.

[0146] Since the values ​​for pipeline material and function are discrete and derived from a predefined set of finite entities in the knowledge graph (e.g., steel pipes, PVC pipes, water supply pipelines, gas pipelines, etc.), the pipeline material and function of the target pipeline can be directly determined through the candidate entities selected above. For example, if the candidate entities obtained above are "steel pipe" and "water supply pipeline", then the pipeline material of the target pipeline can be determined to be steel pipe, and the pipeline function of the target pipeline can be determined to be water supply.

[0147] For continuous attribute values ​​such as pipeline burial depth, pipeline outer diameter, pipeline wall thickness, and the dielectric constant of the surrounding medium, the attribute value of the corresponding entity type can be used as the predicted attribute value t′ in the predicted triple (h′, r′, t′). In this case, h′ and r′ in the predicted triple (h′, r′, t′) are known values. Here, h′ is the entity corresponding to the candidate entity predicted by the above process, as well as the feature values ​​of the full-image features, local features, and environmental features in the feature vector of the target pipeline; r′ is the sum of the relationships from h′ to the predicted attribute value t′.

[0148] In actual reasoning, attribute value prediction is not based solely on a single triple, but rather involves a comprehensive evaluation of multiple reasoning paths within the knowledge graph. Specifically, nodes corresponding to the entities in the target pipeline can be identified based on the feature vectors of each candidate entity and the target pipeline. This involves finding nodes corresponding to candidate entities within the knowledge graph, as well as nodes corresponding to the feature values ​​of all graph features, local features, and environmental features in the target pipeline's feature vector, and these nodes form a node set. Starting from these nodes, all possible reasoning paths from the nodes to the attribute value t′ to be predicted are constructed within the knowledge graph. For each path, each triple is mapped to a vector, and the vector sum of these vectors is calculated and input into a scoring function for evaluation. This results in a path quality score reflecting the degree of matching between nodes within the path, which is then used as the score for the candidate attribute value in that path. Candidate attribute values ​​can be sorted according to their scores, and the candidate attribute value with the highest score or that meets pre-defined confidence conditions is determined as the predicted attribute for the target pipeline.

[0149] For example, in the knowledge graph example above, after determining the candidate entities as "steel pipe" and "water supply pipeline," it is necessary to further predict the continuous value attribute of the pipeline's outer diameter, denoted as t′. At this point, the node set includes the candidate entity nodes "steel pipe" and "water supply pipeline," as well as "full-image feature A," "local feature B," and "environmental feature A." Starting from these nodes, all possible paths pointing to the pipeline's outer diameter attribute value t′ can be searched in the knowledge graph. For example, path 1 might be "steel pipe → corresponding feature → full-image feature A → associated parameter → pipeline outer diameter 150mm," and path 2 might be "water supply pipeline → common specifications → pipeline outer diameter 100mm." It can be specified that if a path contains nodes corresponding to entities with full-image features, local features, or environmental features, then those nodes should correspond to the feature values ​​of the full-image features, local features, and environmental features in the target pipeline feature vector, such as the full-image feature A, local feature B, and environmental feature A determined above.

[0150] For path 1, calculate the vector sum of each triplet: (steel pipe, corresponding feature, full-image feature A), (full-image feature A, associated parameter, pipeline outer diameter), and (pipeline outer diameter, value, 150mm). Input the vector sum into the scoring function to obtain the score corresponding to this path. For example, the score for path 1 is 0.8; similarly, the score for path 2 is 0.6. Finally, the highest score, "150mm", is selected as the predicted pipeline outer diameter. Through a similar method, the pipeline wall thickness, burial depth, and dielectric constant of the surrounding medium are predicted, thus completing the prediction of all pipeline attributes of the target pipeline.

[0151] In the above process, if the user inputs knowledge other than radar images (i.e., the aforementioned environmental information), then when predicting the pipeline burial depth, pipeline outer diameter, pipeline wall thickness, and dielectric constant of the surrounding medium of the target pipeline, there will be more relevant triples available and the chains will be more complete, which will be more conducive to obtaining more accurate inference results.

[0152] Compared with existing technologies, this disclosure introduces ground-penetrating radar image features and pipeline domain knowledge into a knowledge graph, and utilizes the semantic association and reasoning capabilities of the knowledge graph to achieve joint prediction of underground pipeline burial depth and pipeline attributes. On the one hand, the interpretability of underground pipeline location results is improved by leveraging the modeling and reasoning capabilities of the knowledge graph for implicit semantic associations; on the other hand, the phased entity prediction and attribute prediction mechanism effectively shortens the identification time and reduces the dependence on large-scale labeled data. Furthermore, the knowledge graph constructed in this disclosure can be continuously updated and iterated based on new data added during actual pipeline construction operations, possessing self-evolution capabilities; it also integrates domain knowledge for different pipeline burial methods, environmental conditions, and construction specifications, which helps improve the adaptability and prediction accuracy of underground pipeline location methods under different detection equipment and complex application scenarios.

[0153] Figure 3 A structural diagram of an underground pipeline location device based on knowledge graphs and ground-penetrating radar according to an embodiment of this disclosure is shown. Figure 3 As shown, the device includes:

[0154] The first determining module 301 is used to acquire radar images collected by ground penetrating radar against target pipelines, extract features from the radar images, and obtain full-image features and local features of the radar images respectively.

[0155] The second determining module 302 is used to determine the target pipeline feature vector based on the full-image features and local features of the radar image;

[0156] The third determining module 303 is used to input the feature vector of the target pipeline into the knowledge graph to obtain the prediction result of the target pipeline. The nodes of the knowledge graph indicate different entities, the edges of the knowledge graph indicate the relationship between the entities corresponding to the two nodes of the connecting edge, and the prediction result of the target pipeline indicates the pipeline burial depth and pipeline attributes of the target pipeline.

[0157] In one possible implementation, the first determining module 301 is used for:

[0158] Preprocessing is performed on radar images to extract full-image features and obtain preprocessed radar images. Full-image features include one or more of the following: scanning signal, reflected signal intensity histogram, DC component, and ringing intensity.

[0159] The target detection neural network is used to process the preprocessed radar image to identify target areas containing pipeline reflection features in the preprocessed radar image.

[0160] The target area is converted into local features of the radar image.

[0161] In one possible implementation, preprocessing is performed on the radar image to extract full-image features and obtain a preprocessed radar image, including:

[0162] The scanning signal is obtained by extracting signals from radar images;

[0163] The radar image is converted into a two-dimensional grayscale image, and the histogram of reflected signal intensity is determined based on the two-dimensional grayscale image;

[0164] The DC component is extracted and removed from the two-dimensional grayscale image to obtain the image after removing the DC component;

[0165] The ringing intensity is extracted from the image after removing the DC component, and the ringing intensity is removed to obtain the image after removing the ringing.

[0166] Gain processing is performed on the image after ringing removal to obtain the preprocessed radar image.

[0167] In one possible implementation, the device further includes:

[0168] The acquisition module is used to acquire environmental information of the target pipeline. The environmental information includes at least one of the following: the detection area related to the target pipeline, road surface properties, soil moisture, and surrounding buildings.

[0169] The embedding representation module is used to perform embedding representation based on environmental information to obtain environmental features;

[0170] The second determining module 302 is used for:

[0171] Based on full-map features, local features, and environmental features, the feature vector of the target pipeline is determined.

[0172] In one possible implementation, the entities in the knowledge graph include full-image features of radar images, local features of radar images, environmental features, pipeline attributes, and pipeline burial depth. The pipeline attributes include at least one of pipeline material, pipeline outer diameter, pipeline wall thickness, pipeline function, and dielectric constant of the medium surrounding the pipeline.

[0173] In one possible implementation, the third determining module 303 is used for:

[0174] Based on the feature vector of the target pipeline and the knowledge graph, the entities in the triplet to be predicted are predicted, and the candidate entities corresponding to the target pipeline are determined. The triplet to be predicted includes entities, relations and attribute values.

[0175] Based on the feature vectors of each candidate entity and the target pipeline, the attribute values ​​in the triplet to be predicted are predicted to determine the prediction result of the target pipeline.

[0176] In one possible implementation, based on the feature vectors of each candidate entity and the target pipeline, the attribute values ​​in the triples to be predicted are predicted to determine the prediction result of the target pipeline, including:

[0177] Based on the feature vectors of each candidate entity and the target pipeline, the set of nodes associated with the target pipeline is determined in the knowledge graph;

[0178] The prediction results for the target pipeline are obtained by predicting the attribute values ​​in the triplet to be predicted based on the node set.

[0179] According to embodiments of this disclosure, radar images of a target pipeline are acquired by ground-penetrating radar (GPR). Feature extraction is performed on the radar images to obtain full-image features and local features. Based on the full-image features and local features of the radar images, a feature vector of the target pipeline is determined. By inputting the target pipeline feature vector into a knowledge graph, GPR image features can be fused into the knowledge graph, and reasoning can be performed in combination with existing pipeline knowledge. This avoids relying solely on data-driven models for end-to-end learning. Even in typical small-sample, high-noise application scenarios, effective reasoning can still be performed using existing pipeline knowledge in the knowledge graph. Thus, the burial depth and pipeline attributes of the target pipeline can be accurately predicted without relying on large-scale labeled data. The knowledge graph can explicitly introduce structured knowledge in the pipeline domain, imposing prior constraints on the prediction results. This improves the adaptability of the algorithm under different detection instruments and complex environmental conditions. At the same time, the prediction results can be back-explained through corresponding entities and relationships between entities, enhancing the interpretability of the prediction results.

[0180] In some embodiments, the functions or modules of the apparatus provided in this disclosure can be used to perform the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0181] This disclosure also provides an underground pipeline location device based on knowledge graphs and ground penetrating radar, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0182] This disclosure also provides a non-volatile computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described method.

[0183] Figure 4 This is a block diagram illustrating an apparatus 1900 for locating underground pipelines according to an exemplary embodiment. For example, apparatus 1900 may be provided as a server or terminal device. (Refer to...) Figure 4 The apparatus 1900 includes a processing component 1922, which further includes one or more processors, and memory resources represented by memory 1932 for storing instructions, such as application programs, that can be executed by the processing component 1922. The application programs stored in memory 1932 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 1922 is configured to execute instructions to perform the methods described above.

[0184] Device 1900 may also include a power supply component 1926 configured to perform power management of device 1900, a wired or wireless network interface 1950 configured to connect device 1900 to a network, and an input / output interface 1958 (I / O interface). Device 1900 can operate on an operating system, such as Windows Server, stored in memory 1932. TM macOS X TM Unix TM Linux TM FreeBSD TM Or similar.

[0185] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions that can be executed by a processing component 1922 of the device 1900 to perform the above-described method.

[0186] Computer-readable storage media can be tangible devices capable of holding and storing programs / instructions used by instruction execution devices. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0187] The computer program (or computer-readable program instructions) described herein can be downloaded from a computer-readable storage medium to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage medium in the respective computing / processing device.

[0188] The computer program (or computer program instructions) used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions to implement various aspects of this disclosure.

[0189] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0190] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0191] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0192] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0193] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for underground pipeline positioning based on a knowledge graph and ground penetrating radar, characterized in that, The method includes: The radar image collected by the ground penetrating radar targeting the pipeline is acquired, and the radar image is subjected to feature extraction to obtain the full-image features and local features of the radar image respectively. Based on the full-image features and local features of the radar image, the feature vector of the target pipeline is determined; The feature vector of the target pipeline is input into a knowledge graph to obtain the prediction result of the target pipeline. The nodes of the knowledge graph indicate different entities, the edges of the knowledge graph indicate the relationship between the entities corresponding to the two nodes connecting the edges, and the prediction result of the target pipeline indicates the pipeline burial depth and pipeline attributes of the target pipeline.

2. The method of claim 1, wherein, The step of extracting features from the radar image to obtain full-image features and local features of the radar image includes: Based on the radar image, preprocessing is performed to extract the full-image features of the radar image and obtain the preprocessed radar image. The full-image features include one or more of the following: scanning signal, reflected signal intensity histogram, DC component, and ringing intensity. The preprocessed radar image is processed using a target detection neural network to determine the target region in the preprocessed radar image that contains pipeline reflection features. The target area is converted into local features of the radar image.

3. The method of claim 2, wherein, The preprocessing based on the radar image, extracting full-image features from the radar image and obtaining the preprocessed radar image, includes: The scanning signal is obtained by extracting signals from the radar image. The radar image is converted into a two-dimensional grayscale image, and the reflected signal intensity histogram is determined based on the two-dimensional grayscale image; Based on the two-dimensional grayscale image, the DC component is extracted and removed to obtain an image after removing the DC component; Based on the image after removing the DC component, the ringing intensity is extracted and removed to obtain the image after removing the ringing. Gain processing is performed on the image after ringing removal to obtain the preprocessed radar image.

4. The method of claim 1, wherein, The method further includes: Obtain environmental information of the target pipeline, the environmental information including at least one of the following: detection area related to the target pipeline, road surface properties, soil moisture, and surrounding buildings; Environmental features are obtained by embedding representation based on the environmental information; Based on the full-image features and local features of the radar image, the target pipeline feature vector is determined, including: The target pipeline feature vector is determined based on the full-map features, the local features, and the environmental features.

5. The method of claim 4, wherein, The entities in the knowledge graph include full-image features of radar images, local features of radar images, environmental features, pipeline attributes, and pipeline burial depth. The pipeline attributes include at least one of pipeline material, pipeline outer diameter, pipeline wall thickness, pipeline function, and dielectric constant of the medium surrounding the pipeline.

6. The method of claim 5, wherein, The step of inputting the target pipeline feature vector into a knowledge graph to obtain the prediction result of the target pipeline includes: Based on the target pipeline feature vector and the knowledge graph, the entities in the triplet to be predicted are predicted to determine each candidate entity corresponding to the target pipeline. The triplet to be predicted includes an entity, a relation, and an attribute value. Based on the candidate entities and the feature vector of the target pipeline, the attribute values ​​in the triplet to be predicted are predicted to determine the prediction result of the target pipeline.

7. The method of claim 6, wherein, The step of predicting the attribute values ​​in the triplet to be predicted based on the candidate entities and the feature vector of the target pipeline, and determining the prediction result of the target pipeline, includes: Based on the candidate entities and the feature vector of the target pipeline, a set of nodes associated with the target pipeline is determined in the knowledge graph; Based on the node set, the attribute values ​​in the triplet to be predicted are predicted to obtain the prediction result of the target pipeline.

8. An underground pipeline positioning device based on a knowledge graph and ground penetrating radar, characterized by, The device includes: The first determining module is used to acquire radar images collected by ground penetrating radar against target pipelines, and to extract features from the radar images to obtain full-image features and local features of the radar images respectively. The second determining module is used to determine the target pipeline feature vector based on the full-image features and local features of the radar image; The third determining module is used to input the feature vector of the target pipeline into a knowledge graph to obtain the prediction result of the target pipeline. The nodes of the knowledge graph indicate different entities, the edges of the knowledge graph indicate the relationship between the entities corresponding to the two nodes connecting the edges, and the prediction result of the target pipeline indicates the pipeline burial depth and pipeline attributes of the target pipeline. 9.A device for underground pipeline positioning based on a knowledge graph and ground penetrating radar, comprising a memory, a processor and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

10. A non-transitory computer readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images