Underground pipeline detection method and system based on multi-source data fusion and intelligent analysis

Abstract

The invention discloses an underground pipeline detection method and system based on multi-source data fusion and intelligent analysis, and belongs to the technical field of pipeline detection. The method comprises the steps of collecting multi-source data of a target area; wherein the multi-source data comprises electromagnetic signal data, electromagnetic wave reflection data and sound wave propagation data; performing filtering and normalization preprocessing on the multi-source data to obtain target data; constructing a multi-source data fusion model to perform feature level fusion on the target data to generate a fusion feature vector; an underground pipeline recognition model based on deep learning is constructed to intelligently analyze and recognize the fusion feature vector, and a recognition result of the underground pipeline is obtained; and marking and displaying the identification result on an electronic map in a visual mode. According to the invention, by fusing data collected by various detection technologies and applying an intelligent analysis algorithm, the accuracy, reliability and efficiency of underground pipeline detection are improved, and the limitation problem of the existing detection technology is solved.

Description

Technical Field

[0001] This invention belongs to the field of pipeline detection technology, specifically relating to an underground pipeline detection method and system based on multi-source data fusion and intelligent analysis. Background Technology

[0002] With the rapid development of urban construction, the continuous addition of underground structures has led to increasingly complex underground pipeline networks. Accurately detecting the location, direction, and materials of underground pipelines is crucial for urban planning, construction, and daily maintenance. Currently, commonly used underground pipeline detection technologies include electromagnetic induction, ground-penetrating radar (GPR), and acoustic detection. However, these technologies have certain limitations. For example, electromagnetic induction is significantly affected by the surrounding electromagnetic environment, resulting in lower detection accuracy in complex electromagnetic environments; GPR is ineffective for detecting non-metallic pipelines; and acoustic detection is significantly affected by soil properties, leading to substantial differences in results under different geological conditions. Moreover, most existing detection technologies are single-technology applications, failing to fully leverage the advantages of various technologies and hindering comprehensive and accurate detection of underground pipelines. Summary of the Invention

[0003] In view of this, the purpose of this invention is to provide an underground pipeline detection method and system based on multi-source data fusion and intelligent analysis. By integrating data collected from multiple detection technologies and applying intelligent analysis algorithms, the accuracy, reliability and efficiency of underground pipeline detection are improved, and the limitations of existing detection technologies are solved.

[0004] To achieve the above objectives, the present invention provides the following technical solution: The method for detecting underground pipelines based on multi-source data fusion and intelligent analysis includes: Collect multi-source data of the target area; the multi-source data includes electromagnetic signal data, electromagnetic wave reflection data and sound wave propagation data; filter and normalize the multi-source data to obtain the target data; A multi-source data fusion model is constructed to perform feature-level fusion of target data and generate a fused feature vector. A deep learning-based underground pipeline identification model is constructed to intelligently analyze and identify fused feature vectors, thereby obtaining the identification results of underground pipelines. The recognition results are marked and displayed on an electronic map in a visual way.

[0005] Furthermore, multi-source data of the target area is collected, including: Electromagnetic induction detectors are used to detect the target area and collect electromagnetic signal data of underground metal pipelines to obtain rough location and direction information of the underground metal pipelines. Ground-penetrating radar is used to scan the target area and collect electromagnetic wave reflection data of underground pipelines to obtain information on the burial depth and diameter of underground pipelines. Acoustic wave detectors are used to conduct acoustic wave detection in the target area and collect acoustic wave propagation data of the soil medium around underground pipelines. This data is used to obtain information on the differences between underground pipelines and the surrounding soil medium, providing auxiliary information for the location of non-metallic underground pipelines.

[0006] Furthermore, the multi-source data is preprocessed by filtering and normalization to obtain the target data, including: The collected electromagnetic signal data, electromagnetic wave reflection data, and sound wave propagation data are filtered to remove noise interference. The filtered data is normalized to convert data of different types and dimensions into a unified numerical range to obtain the target data. For electromagnetic signal data, an adaptive filtering algorithm is used to automatically adjust the filtering parameters according to the characteristics of the signal, effectively removing noise generated by the surrounding electromagnetic environment. For electromagnetic wave reflection data, wavelet filtering algorithm is used to analyze and process signals of different frequency components to remove high-frequency noise and low-frequency interference; For sound wave propagation data, a bandpass filtering algorithm is used to retain only the signal within the effective frequency range and remove noise at irrelevant frequencies.

[0007] Furthermore, a multi-source data fusion model is constructed to perform feature-level fusion of the target data, generating a fused feature vector, including: A multi-source data fusion model is established to fuse preprocessed electromagnetic signal data, electromagnetic wave reflection data, and sound wave propagation data; During the fusion process, a feature-level fusion method is adopted to extract feature vector information of various types of data, and the extracted feature vector information is combined to form a fusion feature vector containing multiple aspects of underground pipeline information, thereby achieving a comprehensive description of underground pipeline information.

[0008] Furthermore, fusion methods that utilize multi-source data fusion models to perform feature-level fusion of target data include: Obtain the preprocessed target data; Three pre-constructed independent feature extraction sub-networks are used to extract features from electromagnetic signal data, electromagnetic wave reflection data, and sound wave propagation data in the target data, respectively, to obtain feature vector information of the three types of data; the three independent feature extraction sub-networks include electromagnetic signal feature extraction sub-network, electromagnetic wave reflection feature extraction sub-network, and sound wave feature extraction sub-network. The attention weights among the three types of feature vectors are calculated based on the modal attention mechanism. The attention weights are then used to enhance each feature vector. The enhanced feature vectors are then concatenated and compressed through a fully connected layer to obtain a fused feature vector.

[0009] Furthermore, a deep learning-based underground pipeline identification model is constructed to intelligently analyze and identify fused feature vectors, yielding the identification results of underground pipelines, including: We obtained several labeled underground pipeline data as training samples, and used a convolutional neural network as the basic architecture to build an underground pipeline identification model based on deep learning. The fused feature vector is input into the underground pipeline identification model, and the identification results of the underground pipeline features are output, so as to accurately identify the type, location, direction, burial depth and diameter of the underground pipeline.

[0010] Furthermore, underground pipeline detection methods based on multi-source data fusion and intelligent analysis also include: The collection of underground pipeline change events based on crowd-sensing involves collecting multi-source sensing data from corresponding-level data sources within the target area through data source hierarchical rules; the multi-source sensing data includes text information and image information. Multi-source sensing data is subjected to hierarchical cleaning and standardization to obtain preprocessed data; hierarchical cleaning involves filtering highly relevant data based on the data source level, and standardization involves unifying the geographic description format and image coordinate system. Natural language processing and computer vision processing technologies are used to extract underground pipeline-related information from the preprocessed data to obtain target data, which is then used to generate a predicted map of underground pipeline information distribution. By using the underground pipeline information distribution prediction map, the identification results of underground pipelines are optimized to obtain complete distribution information of underground pipelines in the target area; The complete distribution information is marked and displayed on an electronic map in a visual way.

[0011] Furthermore, the data source classification rules categorize data sources into Level 1, Level 2, and Level 3 data sources based on their credibility. Primary data sources include official pipeline reporting systems and shared data from authorized construction units; Secondary data sources include industry communities and government new media interaction data; Level 3 data sources include public content from social media platforms and historical archive OCR data; Targeted data collection includes: Real-time data synchronization between primary and secondary data sources is achieved through API integration or compliant web crawling. Data is collected through a dual-filtering approach using keyword search and geofencing to access three levels of data sources. Hierarchical cleaning of multi-source sensing data includes: The data collected from the primary data source is filtered for format errors using a format anomaly detection method. By using location and keyword filtering rules, content without location information or pipeline keywords in secondary and tertiary data sources is filtered out. Then, the TextCNN text classification model and ResNet image classification model are used to filter highly relevant data with a confidence level ≥ 0.7. The target data to be extracted includes: Natural language processing technology is used to extract pipeline entity information and spatial relationships from text information, and computer vision processing technology is used to identify pipeline features and scene auxiliary information from image information, and the target data is obtained by integrating them.

[0012] Furthermore, the identification results of underground pipelines are optimized using the underground pipeline information distribution prediction map to obtain complete distribution information of underground pipelines within the target area, including: The rule base of safety standards for the construction of urban underground pipelines is obtained as a constraint condition, and a data fusion model is generated using training data. This model is used to fuse and optimize the underground pipeline information distribution prediction map based on the identification results of underground pipelines, so as to obtain the complete distribution information of underground pipelines in the target area. The training steps for a data fusion model include: Acquire multiple sets of training sample data; each set of training sample data corresponds to the baseline data and detailed data within a target area. The baseline data is the identification result of underground pipelines within the target area, and the detailed data is the predicted distribution map data of underground pipeline information within the target area. A dual-branch feature extraction module consisting of a baseline branch and a detail branch is constructed. The baseline branch transforms the recognition results into low-dimensional vectors that preserve topological relationships through a fully connected layer. The detail branch includes text feature extraction and image feature extraction. Text feature extraction uses a pre-trained BERT model to semantically encode the text information in the detail data and outputs a text feature vector. Image feature extraction uses a CNN model to perform coordinate mapping on the image information in the detail data and extracts the spatial coordinate features of pipeline endpoints and inflection points, outputting a coordinate feature vector. The text feature vector and the coordinate feature vector are concatenated and input into a GNN model to learn the pipeline detail features, outputting the pipeline detail feature vector. Load the rule library of urban underground pipeline construction safety standards containing quantitative constraint indicators, perform safety compliance screening on the pipeline detail feature vectors output by the detail branches, and retain the pipeline detail feature vectors that comply with the urban underground pipeline construction safety standards, which are denoted as compliance feature vectors. Using the low-dimensional vector output by the baseline branch as the anchor point, calculate the spatial Euclidean distance between the pipeline coordinates in the compliance feature vector output by the detail branch and the pipeline coordinates in the baseline data. If the spatial Euclidean distance is less than the preset distance, it is determined to be a high-matching region, and the baseline data is supplemented with the detail data in the compliance feature vector; otherwise, it is determined to be a low-matching region, and only the baseline data is retained. By integrating the completion results from high-matching regions and the retained baseline data from low-matching regions, preliminary fusion results are obtained, forming a complete framework for the fusion model. Set a loss function and train the complete framework of the fusion model by combining multiple sets of training sample data until the preset training conditions are met, and then output the data fusion model.

[0013] An underground pipeline detection system based on multi-source data fusion and intelligent analysis includes: The data acquisition module includes an electromagnetic induction detector, a ground-penetrating radar, and a sound wave detector, used to collect electromagnetic signal data, electromagnetic wave reflection data, and sound wave propagation data of the target area; The data preprocessing module is used to filter and normalize the collected data to generate target data; The data fusion module is used to perform feature-level fusion on preprocessed data using a multi-source data fusion model to generate a fused feature vector. The intelligent analysis module is used to analyze and identify fused feature vectors using a deep learning-based underground pipeline identification model, and obtain the identification results to determine relevant information about underground pipelines. The results output module is used to visually label and display the recognition results on an electronic map.

[0014] The beneficial effects of this invention are as follows: 1. Improved Detection Accuracy: By integrating data from multiple technologies such as electromagnetic induction, ground-penetrating radar, and acoustic detection, the advantages of each technology are fully utilized, overcoming the limitations of a single technology. This allows for more comprehensive and accurate information on underground pipelines, effectively improving detection accuracy. For example, for metal pipelines, combining electromagnetic induction and ground-penetrating radar can more precisely determine their location, direction, and diameter; for non-metallic pipelines, the fusion of data from ground-penetrating radar and acoustic detection can improve the success rate and accuracy of detection.

[0015] 2. Enhanced Reliability: The fusion of multi-source data and the application of intelligent analysis algorithms reduce misjudgments and omissions caused by single data or algorithms, thus improving the reliability of detection results. Even under complex geological conditions and electromagnetic environments, accurate underground pipeline information can be stably obtained.

[0016] 3. Improved Detection Efficiency: Automated data collection, processing, and analysis significantly reduce manual intervention and data processing time, thereby improving detection efficiency. Simultaneously, the visualized output of results allows users to quickly understand the distribution of underground pipelines, providing timely and accurate decision-making support for urban planning and construction.

[0017] 4. Wide applicability: This method and system are applicable to different types of underground pipelines (metal pipelines, non-metal pipelines, etc.) as well as various complex geological conditions and urban environments, and have strong versatility and practicality.

[0018] Other advantages, objectives, and features of the invention will be set forth in the following description and will be apparent to those skilled in the art in some respects, or may be learned by practice of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0019] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the detection process of an underground pipeline detection method and system based on multi-source data fusion and intelligent analysis in an embodiment of the present invention. Figure 2 This is a schematic diagram of another detection process in an underground pipeline detection method and system based on multi-source data fusion and intelligent analysis, as described in an embodiment of the present invention. Figure 3 This is a flowchart illustrating the execution process of an underground pipeline detection method and system based on multi-source data fusion and intelligent analysis, as well as the swarm intelligence sensing module, in an embodiment of the present invention. Detailed Implementation

[0021] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0022] like Figure 1 As shown, this invention proposes an underground pipeline detection method based on multi-source data fusion and intelligent analysis, comprising: Collect multi-source data of the target area; the multi-source data includes electromagnetic signal data, electromagnetic wave reflection data and sound wave propagation data; filter and normalize the multi-source data to obtain the target data; A multi-source data fusion model is constructed to perform feature-level fusion of target data and generate a fused feature vector. A deep learning-based underground pipeline identification model is constructed to intelligently analyze and identify fused feature vectors, thereby obtaining the identification results of underground pipelines. The recognition results are marked and displayed on an electronic map in a visual manner; The working principle and beneficial effects of the above technical solution are as follows: By integrating data collected from multiple technologies such as electromagnetic induction, ground-penetrating radar, and acoustic detection, the advantages of each technology are fully utilized, and the limitations of a single technology are overcome. This allows for the acquisition of more comprehensive and accurate information on underground pipelines, effectively improving the accuracy of detection. For example, for metal pipelines, combining electromagnetic induction and ground-penetrating radar methods can more accurately determine their location, direction, and diameter; for non-metallic pipelines, the fusion of data from ground-penetrating radar and acoustic detection methods can improve the success rate and accuracy of detection. The fusion of multi-source data and the application of intelligent analysis algorithms reduce misjudgments and omissions caused by single data or single algorithms, improving the reliability of detection results. Even under complex geological conditions and electromagnetic environments, accurate underground pipeline information can be stably obtained. The automated data acquisition, processing, and analysis process greatly reduces manual intervention and data processing time, improving detection efficiency. Simultaneously, the visualized output of results allows users to quickly understand the distribution of underground pipelines, providing timely and accurate decision-making basis for urban planning and construction. This method and system are applicable to different types of underground pipelines (metal pipelines, non-metal pipelines, etc.) as well as various complex geological conditions and urban environments, and have strong versatility and practicality.

[0023] In one embodiment, collecting multi-source data of the target area includes: Electromagnetic induction detectors are used to detect the target area and collect electromagnetic signal data of underground metal pipelines to obtain rough location and direction information of the underground metal pipelines. Ground-penetrating radar is used to scan the target area and collect electromagnetic wave reflection data of underground pipelines to obtain information on the burial depth and diameter of underground pipelines. Acoustic wave detectors are used to detect acoustic waves in the target area and collect acoustic wave propagation data of the soil medium around underground pipelines. This data is used to obtain information on the differences between underground pipelines and the surrounding soil medium, providing auxiliary information for the location of non-metallic underground pipelines. The working principle and beneficial effects of the above technical solution are as follows: Electromagnetic induction detectors are used to detect the target area, collecting electromagnetic signal data of underground metal pipelines to obtain their approximate location and direction. Based on the principle of electromagnetic induction, when a metal pipeline is located in an alternating magnetic field, an induced current is generated in the pipeline. This current generates a secondary magnetic field, which the electromagnetic induction detector uses to detect the secondary magnetic field to determine the presence and location of the metal pipeline. Ground-penetrating radar (GPR) is used to scan the target area, collecting electromagnetic wave reflection data of the underground pipelines to obtain information such as burial depth and diameter. GPR emits high-frequency electromagnetic waves underground; these waves are reflected when they encounter interfaces between different media. By receiving and analyzing the characteristics of the reflected waves, relevant pipeline parameters can be determined. Acoustic wave detectors are used to detect sound waves in the target area, collecting sound wave propagation data of the soil medium surrounding the underground pipelines to obtain information on the differences between the pipeline and the surrounding soil medium, especially providing auxiliary information for locating non-metallic pipelines. Acoustic wave detectors emit sound waves; the propagation speed and attenuation of sound waves differ in different media. By analyzing the propagation characteristics of the sound waves, the presence and location of the pipeline can be determined.

[0024] In one embodiment, filtering and normalization preprocessing are performed on multi-source data to obtain target data, including: The collected electromagnetic signal data, electromagnetic wave reflection data, and sound wave propagation data are filtered to remove noise interference. The filtered data is normalized to convert data of different types and dimensions into a unified numerical range to obtain the target data. For electromagnetic signal data, an adaptive filtering algorithm is used to automatically adjust the filtering parameters according to the characteristics of the signal, effectively removing noise generated by the surrounding electromagnetic environment. For electromagnetic wave reflection data, wavelet filtering algorithm is used to analyze and process signals of different frequency components to remove high-frequency noise and low-frequency interference; For sound wave propagation data, a bandpass filtering algorithm is used to retain only the signal within the effective frequency range and remove noise at irrelevant frequencies; The working principle and beneficial effects of the above technical solution are as follows: The collected electromagnetic signal data, electromagnetic wave reflection data, and sound wave propagation data are filtered separately to remove noise interference and improve data quality. For example, for electromagnetic signal data, an adaptive filtering algorithm is used to automatically adjust the filtering parameters according to the characteristics of the signal, effectively removing noise generated by the surrounding electromagnetic environment; for electromagnetic wave reflection data, a wavelet filtering algorithm is used to analyze and process signals of different frequency components, removing high-frequency noise and low-frequency interference; for sound wave propagation data, a bandpass filtering algorithm is used to retain only signals within the effective frequency range and remove noise at irrelevant frequencies. The filtered data is then normalized to convert data of different types and dimensions into a unified numerical range, facilitating subsequent data fusion and analysis.

[0025] In one embodiment, a multi-source data fusion model is constructed to perform feature-level fusion of target data to generate a fused feature vector, including: A multi-source data fusion model is established to fuse preprocessed electromagnetic signal data, electromagnetic wave reflection data, and sound wave propagation data; During the fusion process, a feature-level fusion method is adopted to extract feature vector information of various types of data, and the extracted feature vector information is combined to form a fusion feature vector containing multiple aspects of underground pipeline information, thereby achieving a comprehensive description of underground pipeline information. The working principle and beneficial effects of the above technical solution are as follows: A multi-source data fusion model is established to fuse preprocessed electromagnetic signal data, electromagnetic wave reflection data, and sound wave propagation data. A feature-level fusion method is adopted to extract feature information from each type of data, such as the magnetic field strength variation characteristics of electromagnetic signal data, the reflected wave waveform characteristics of electromagnetic wave reflection data, and the sound velocity and sound attenuation characteristics of sound wave propagation data. The extracted feature information is combined to form a fused feature vector containing multifaceted information about underground pipelines, achieving a comprehensive description of underground pipeline information.

[0026] In one embodiment, a fusion method for feature-level fusion of target data using a multi-source data fusion model includes: Obtain the preprocessed target data; Three pre-constructed independent feature extraction sub-networks are used to extract features from electromagnetic signal data, electromagnetic wave reflection data, and sound wave propagation data in the target data, respectively, to obtain feature vector information of the three types of data; the three independent feature extraction sub-networks include electromagnetic signal feature extraction sub-network, electromagnetic wave reflection feature extraction sub-network, and sound wave feature extraction sub-network. The attention weights among the three types of feature vectors are calculated based on the modal attention mechanism. The attention weights are then used to enhance each feature vector. The enhanced feature vectors are then concatenated and compressed through a fully connected layer to obtain a fused feature vector. The working principle and beneficial effects of the above technical solution are as follows: Under normal circumstances, since the data is collected simultaneously, only denoising and standardization processing of the multi-source data is required. To further improve the fusion accuracy, spatiotemporal alignment can also be performed on the data to obtain the target data. The spatiotemporal alignment includes time alignment and spatial alignment. Time alignment uses the sampling time of the electromagnetic signal data as a reference, and interpolates and completes the electromagnetic wave reflection data and sound wave propagation data to unify the time step. Spatial alignment binds the spatial grid of the electromagnetic wave reflection data with the spatial labels of the electromagnetic signal data and sound wave propagation data. Features are extracted from the three types of preprocessed data through three independent feature extraction sub-networks to obtain electromagnetic signal feature vectors, electromagnetic wave reflection feature vectors, and sound wave feature vectors. Preferably, the electromagnetic signal feature extraction sub-network adopts a combination structure of 1D-CNN and LSTM, the electromagnetic wave reflection feature extraction sub-network adopts a 2D-CNN structure, and the sound wave feature extraction sub-network adopts a combination structure of 2D-CNN and 1D-CNN. The combined 1D-CNN and LSTM structure includes two convolutional blocks and one bidirectional LSTM layer. The convolutional blocks use 1D convolutional layers with a kernel size of 5 and the ReLU activation function, and the bidirectional LSTM has 64 hidden units. The 2D-CNN structure includes three convolutional blocks and a global max pooling layer. The convolutional blocks use 3×3 convolutional kernels and the ReLU activation function. The combined 2D-CNN and 1D-CNN structure includes one 2D convolutional block with a 3×3 kernel and one 1D convolutional block with a 5×1 kernel. Finally, attention weights are calculated between the three types of feature vectors based on a modal attention mechanism. These attention weights are used to weight and enhance each feature vector. The enhanced feature vectors are then concatenated and compressed through a fully connected layer to obtain a fused feature vector. Through the above technical solution, data collected by multiple technologies such as electromagnetic induction, ground penetrating radar, and acoustic detection are integrated, giving full play to the advantages of each technology and overcoming the limitations of a single technology. This enables the acquisition of more comprehensive and accurate information about underground pipelines, effectively improving the accuracy of detection.

[0027] In one embodiment, a deep learning-based underground pipeline identification model is constructed to intelligently analyze and identify fused feature vectors, resulting in the identification of underground pipelines, including: We obtained several labeled underground pipeline data as training samples, and used a convolutional neural network as the basic architecture to build an underground pipeline identification model based on deep learning. Input the fused feature vector into the underground pipeline identification model, and output the identification results of the underground pipeline features, so as to accurately identify the type, location, direction, burial depth and diameter of the underground pipeline; The working principle and beneficial effects of the above technical solution are as follows: A deep learning-based underground pipeline identification model is constructed, using a convolutional neural network (CNN) as its basic architecture. The fused feature vectors are input into the identification model, and through model training and learning, the feature patterns of underground pipelines are automatically extracted, achieving accurate identification of information such as the type (metallic pipeline, non-metallic pipeline, etc.), location, direction, burial depth, and pipe diameter of underground pipelines. During model training, a large amount of labeled underground pipeline data is used as training samples, and the model parameters are continuously adjusted to improve the model's identification accuracy.

[0028] The identification process of the deep learning-based underground pipeline identification model includes: The fused feature vector from multiple sources is used as input, and the fused feature vector is converted from a 1D vector into a 2D feature map of a preset size through a feature reconstruction layer. It is worth noting that the total dimension of the 2D feature map is consistent with the dimension of the fused feature vector. Then, a convolutional neural network backbone is used to extract features from the 2D feature map, and convolution, batch normalization, activation and pooling are performed sequentially through at least two convolutional blocks. Finally, a globally shared feature vector is output. Each convolutional block described above contains at least two convolutional layers, and the convolutional layers preferably use 3*3 convolutional kernels. Finally, based on the globally shared feature vector, the feature recognition results of underground pipelines are output through multiple independent task branches; In this embodiment, the number of independent task branches is preferably 5, including location identification branch, direction identification branch, burial depth identification branch, pipe diameter identification branch and pipeline type identification branch; Among them, the location recognition branch preferably uses a fully connected layer and a linear activation function to output the planar coordinates of the pipeline in a preset coordinate system; The preferred branch for pipeline orientation identification is obtained by using a fully connected layer and a Sigmoid activation function to output the angle between the pipeline orientation and the due north direction. The burial depth identification branch is preferably obtained by using a fully connected layer and a linear activation function to output the pipeline burial depth value; The pipe diameter identification branch preferably uses a fully connected layer and a linear activation function to output the pipe diameter value; The pipeline type identification branch preferably uses a fully connected layer and a Softmax activation function to output the classification probability of the pipeline type; It is worth noting that the model loss optimization used in the above recognition process preferably adopts a weighted joint loss function for training optimization, and the weighted joint loss function is constructed based on the number and type of independent task branches.

[0029] like Figure 2 , Figure 3As shown, in one embodiment, the underground pipeline detection method based on multi-source data fusion and intelligent analysis further includes: S101. Collection of underground pipeline change events based on collective intelligence perception, which involves collecting multi-source perception data from corresponding level data sources within the target area through data source classification rules; wherein, the multi-source perception data includes text information and image information; S102. Perform hierarchical cleaning and standardization on the multi-source sensing data to obtain preprocessed data; among which, hierarchical cleaning is to select highly relevant data based on the data source level, and standardization is to unify the geographic description format and image coordinate system. S103. Natural language processing and computer vision processing technologies are used to extract underground pipeline information from the preprocessed data to obtain target data, which is used to generate a prediction map of underground pipeline information distribution. S104. Optimize the identification results of underground pipelines by using the underground pipeline information distribution prediction map to obtain complete distribution information of underground pipelines in the target area; The complete distribution information is marked and displayed on an electronic map in a visual way; The working principle of the above technical solution is as follows: In actual situations, due to the complexity of the underground environment, electromagnetic induction, ground-penetrating radar, and acoustic detection can usually only perform a rough detection of information such as the distribution of important underground pipeline paths. This is suitable for general construction scenarios. However, when the work area is limited or the underground pipeline network in the current area is particularly complex and requires fine-grained operations, this rough detection needs to be repeated as the work continues to be carried out to achieve accurate identification of pipeline information and avoid damage to underground pipelines during construction. However, this method greatly affects the construction progress, making the construction plan need to be adjusted according to changes in underground pipeline information, and there is even a risk of accidentally touching underground pipelines. In order to solve these problems that may exist in actual fine-grained operations, this application proposes a swarm intelligence sensing module, which is used to integrate the underground pipeline data (network recorded data) obtained by swarm intelligence sensing with the actual detection data (identification results) and the network recorded data to reasonably optimize and complete the original underground pipeline identification results, thereby obtaining complete distribution information of underground pipelines in the target area. This enables the reduction of construction waiting time and improvement of construction efficiency when carrying out fine-grained operations in the target area. Specifically, firstly, data source classification rules need to be established based on the credibility of the data sources. Then, multi-source sensing data of underground pipelines related to the target area are collected from the corresponding level of data sources according to these classification rules. Based on the actual construction, acceptance, and inclusion of underground pipelines into the management system, the collection time span is preferably 6 months. The collected multi-source sensing data includes text information and image information. Among them, the data source classification rules preferably divide the data sources into first-level data sources, second-level data sources, and third-level data sources according to their credibility. First-level data sources include official pipeline reporting systems (such as the underground pipeline integrated management information system) and data shared by authorized construction units. Second-level data sources include industry communities and government new media interaction data. Third-level data sources include public content on social media platforms and historical archive OCR (Optical Character Recognition) data. Meanwhile, targeted data collection optimization includes: real-time data synchronization between primary and secondary data sources via API integration or compliant web crawlers; and dual-filtering data collection from tertiary data sources through keyword retrieval and geofencing. After collecting complete and fixed multi-source sensing data, according to the pre-defined hierarchical orientation, different levels of cleaning rules are used to clean the corresponding multi-source sensing data to ensure high reliability and high relevance in data screening at each level. The hierarchical cleaning and optimization of multi-source sensing data includes filtering data from primary data sources for format errors using format anomaly detection methods; and filtering out content from secondary and tertiary data sources lacking location information or pipeline keywords using location and keyword filtering rules. Finally, the TextCNN (Text Convolutional Neural Network) text classification model and ResNet (Residual...) are used. The Neural Network image classification model filters highly relevant data with a confidence level ≥ 0.7; the filtered data is then standardized to obtain uniform data for subsequent processing; through the process of "data source hierarchical - targeted collection - hierarchical cleaning - standardization processing", the system integrates multi-source data (text, images, etc.) with different confidence levels, avoiding the problems of disorder and inconsistent confidence levels of multi-source data; at the same time, the application of natural language processing and computer vision technology enables effective information extraction from unstructured data (such as text records and acceptance images), maximizing the value of historical data recorded on the network and providing rich support for pipeline information completion; After obtaining unified preprocessed data on spatial coordinates and geographic descriptions (e.g., a point-line-attribute triple structure), underground pipeline-related information is extracted using natural language processing (NLP) and computer vision techniques, based on the data types within the preprocessed data. Specifically, NLP extracts pipeline entity information and spatial relationships from textual information, and computer vision identifies pipeline features and scene auxiliary information from image information, integrating them to obtain target data. The target data is then used to generate a distribution prediction map of underground pipeline information. It is worth noting that when generating the distribution prediction map using the target data, duplicate pipeline information is not filtered using the existing method of selecting only one element. Instead of using a fixed formula, the method generates weights based on the number of repetitions. These weights are then used to highlight relevant pipeline information in the generated distribution prediction map. For example, as the weight increases, the color level of the corresponding identifier is gradually increased to distinguish it from other surrounding pipeline information. This color level also corresponds to its credibility, providing reliable support for subsequent fusion and completion. For repeated pipeline information, a "weighted + color-level highlighting" approach is used to intuitively reflect the information's credibility while avoiding bias from single data points, making the final complete distribution information result more reliable and reducing the risk of misinterpretation. Apart from the novel filtering method described above, the other steps in generating the underground pipeline information distribution prediction map using the target data are common knowledge to those skilled in the art and will not be elaborated here. Finally, using the safety standards for underground pipeline construction in various cities (such as the planning and handling of underground pipeline intersections, the planning and setting requirements for overhead lines, and the requirements for pipelines passing through bridges and tunnels) as constraints, and combining the data fusion model generated from the training data, the underground pipeline information distribution prediction map and the underground pipeline identification results are fused and optimized to obtain reasonable and accurate complete distribution information of underground pipelines in the target area. The complete distribution information is then sent to the result output module for visualization and marking on the electronic map. It is worth noting that during the final marking and display, the aforementioned weights are highlighted accordingly, and the original identification results are specially marked, enabling construction personnel to quickly understand the pipeline layout and risk level, facilitating advance planning of work paths, avoiding high-risk areas, reducing the blindness of on-site decision-making, and further improving the accuracy and planning of precision operations. The beneficial effects of the above technical solution are as follows: By integrating actual detection data with network-recorded collective sensing data through the collective sensing module, the original coarse identification results are directly optimized and supplemented, avoiding the drawbacks of "repeated detection during construction" in traditional methods. It eliminates the need to suspend construction due to unclear pipeline information while waiting for detection results or the possibility of misjudgment, reducing frequent adjustments to the construction plan, thereby shortening the overall construction cycle and improving the efficiency of precision operations. Simultaneously, for precision operation scenarios with complex underground environments or limited work areas, the solution uses multi-source sensing data fusion (actual detection data + network-recorded collective sensing data) to "reasonably optimize and supplement" the original coarse identification results. Compared to single detection methods, the fused information retains the authenticity of the actual detection site while supplementing historical or global data recorded on the network, ultimately obtaining "complete distribution information." This solves the problem of "information fragmentation" in complex areas under traditional coarse detection, meeting the needs of precision operations for pipeline details.

[0030] In one embodiment, the identification results of underground pipelines are optimized using an underground pipeline information distribution prediction map to obtain complete distribution information of underground pipelines within the target area, including: The rule base of safety standards for the construction of urban underground pipelines is obtained as a constraint condition, and a data fusion model is generated using training data. This model is used to fuse and optimize the underground pipeline information distribution prediction map based on the identification results of underground pipelines, so as to obtain the complete distribution information of underground pipelines in the target area. The training steps for this data fusion model include: S1: Obtain multiple sets of training sample data; each set of training sample data corresponds to baseline data and detailed data within a target area. The baseline data is the identification result of underground pipelines within the target area, and the detailed data is the distribution prediction map data of underground pipeline information within the target area. It is worth noting that the detailed data has been graded, cleaned, and standardized. S2: Construct a dual-branch feature extraction module containing a baseline branch and a detail branch. The baseline branch transforms the recognition results into low-dimensional vectors that preserve topological relationships through a fully connected layer. The detail branch includes text feature extraction and image feature extraction. Text feature extraction uses a pre-trained BERT (Bidirectional Encoder Representations from Transformers) model to semantically encode textual information (such as pipe diameter 300mm, burial depth 1.2m, etc.) in the detail data, outputting a 512-dimensional text feature vector. Image feature extraction uses a CNN (Convolutional Neural Network) model (preferably ResNet-18) to perform coordinate mapping on image information (including spatial relationships and coordinate-unified image information) in the detail data, extracting spatial coordinate features of pipeline endpoints and inflection points, outputting a 256-dimensional coordinate feature vector. The text feature vector and coordinate feature vector are concatenated (dimension unification is required during concatenation) and input into a GNN (Graph Neural Network). The Networks model (preferably a graph attention network model) learns pipeline detail features (including but not limited to precise coordinates of inflection points, pipe diameter, burial depth, and branch connection relationships), and outputs a pipeline detail feature vector. The text feature vector and coordinate feature vector are concatenated to transform it into a graph structure that can be processed by GNN. Then, the GNN captures the topological relationships between nodes, and finally generates a vector containing pipeline detail features, which is used for subsequent comparison and fusion with the low-dimensional vector that preserves the topological relationships obtained above. S3: Load the rule library of urban underground pipeline construction safety standards containing quantitative constraint indicators. These quantitative constraint indicators include, but are not limited to, the minimum spacing between intersecting pipes, the minimum burial depth of pipelines, and the minimum distance between pipelines and buildings. The specific constraint conditions are limited according to the regulations issued by the city where the underground pipeline is actually constructed. The pipeline detail feature vectors output by the detailed branches are screened for safety compliance through the corresponding quantitative constraint indicators. The pipeline detail feature vectors that comply with the urban underground pipeline construction safety standards are retained and recorded as compliance feature vectors. It is worth noting that during the safety compliance screening, both the detailed features of a single pipeline and the associated detailed features of multiple intersecting pipelines need to be judged simultaneously. If the associated detailed features of multiple intersecting pipelines do not meet the corresponding urban underground pipeline construction safety standards, the pipeline detailed features corresponding to the time node with the furthest difference from the current time node are identified as non-compliant feature vectors and removed. The associated detailed features of the remaining intersecting pipelines are then re-verified until the associated detailed features of the remaining multiple intersecting pipelines meet the urban underground pipeline construction safety standards. The pipeline detailed feature vectors that meet the urban underground pipeline construction safety standards are recorded as compliant feature vectors. By using a step-by-step removal method, the precision of the pipeline detailed feature vectors is ensured as much as possible, providing reliable data support for subsequent pipeline information supplementation. S4: Construct a fusion module. Using the low-dimensional vector output by the baseline branch (containing the coordinates and topological relationships of underground pipelines) as anchor points, calculate the spatial Euclidean distance between the pipeline coordinates in the compliance feature vector output by the detail branch and the pipeline coordinates in the baseline data. If the spatial Euclidean distance is less than a preset distance (set according to the accuracy error range of multi-source data, with an initial value of 0.5m), it is determined to be a high-matching region. The detail data in the compliance feature vector is used to supplement the baseline data. (In a high-matching region, it indicates that the correlation between the detail data and the baseline data is strong. When there is a coordinate error (less than 0.5m) between the detail data and the baseline data for the same pipeline information, the detail data is used to correct the baseline data. When the detail data and the baseline data are for different pipeline information, if a certain pipeline information exists only in the detail data, and the spatial Euclidean distance between it and any pipeline coordinate in the baseline data satisfies the high-matching condition...) The criteria for determining the area are as follows: while keeping the baseline data unchanged, the detailed data is used to complete the baseline data to generate an accurate and complete underground pipeline map. It is worth noting that when performing pipeline completion operations, the execution logic of first correcting and then completing must be followed. Otherwise, it is determined as a low-matching area, and only the baseline data is retained. (In a low-matching area, it means that the correlation between the detailed data and the baseline data is weak. When the detailed data and the baseline data are for the same pipeline information, there is a large coordinate error (greater than or equal to 0.5m), so only the baseline data is retained. When the detailed data and the baseline data are for different pipeline information, if the pipeline information exists only in the detailed data and the spatial Euclidean distance between it and any pipeline coordinate in the baseline data does not meet the criteria for a high-matching area; or if the pipeline information exists only in the baseline data but not in the detailed data, then only the baseline data is retained.) S5: Integrate the completion results from high-matching regions and the retained baseline data from low-matching regions to obtain preliminary fusion results; form the complete framework of the fusion model; S6: Set the loss function (preferably including baseline consistency loss, detail completion loss and constraint satisfaction loss), and train the complete framework of the fusion model by combining multiple sets of training sample data (divided into training set, validation set and test set). This includes first pre-training the GNN model based on the training set to learn the pipeline topology relationship, then fine-tuning the dual-branch feature extraction module and fusion module end-to-end, and iterating repeatedly until the preset training conditions are met, and then outputting the data fusion model. Furthermore, for the repetitive pipeline information in the detailed data, the weights are quantized according to the number of repetitions to obtain weighted target data, which is recorded as auxiliary data and used as auxiliary input for training the data fusion model; During the detailed feature vector generation stage, the weight values ​​carried by the auxiliary data are added as feature dimensions to the detailed feature vector of the corresponding pipeline to form an auxiliary feature vector with weight labels. Meanwhile, when designing the loss function, weight matching loss is added, and cross-entropy is used to calculate the deviation between the pipeline confidence output by the model and the corresponding weight quantization value. Compared to the aforementioned data fusion model, the final generated data fusion model can output not only basic information such as pipeline coordinates, type, and burial depth, but also credibility color-coded information corresponding to the basic pipeline information, helping construction personnel to quickly understand the pipeline risk level. The beneficial effects of the above technical solution are as follows: The above technical solution provides a data fusion model that meets the data fusion requirements of underground pipelines by combining benchmark data anchoring, detailed data completion, and safety constraint error correction. Through the above technical solution, the measured pipeline identification results are used as the benchmark, and the topological relationship is preserved through the dual-branch module, laying a reliable foundation for subsequent optimization; the detailed branch uses BERT, CNN, GNN multimodal technology to accurately extract pipeline details (precise coordinates, pipe diameter, etc.), and completes the benchmark after dimensional unification to meet the accuracy requirements of fine operation; combined with quantitative safety standards, single and intersecting pipelines are screened in both directions, and non-compliant data is also removed by time, which greatly reduces the risk of accidental contact during construction; the fusion module completes the data by matching spatial distance to avoid benchmark tampering.

[0031] This invention proposes an underground pipeline detection system based on multi-source data fusion and intelligent analysis, comprising: The data acquisition module includes an electromagnetic induction detector, a ground-penetrating radar, and a sound wave detector, used to collect electromagnetic signal data, electromagnetic wave reflection data, and sound wave propagation data of the target area; The data preprocessing module is used to filter and normalize the collected data to generate target data; The data fusion module is used to perform feature-level fusion on preprocessed data using a multi-source data fusion model to generate a fused feature vector. The intelligent analysis module is used to analyze and identify fused feature vectors using a deep learning-based underground pipeline identification model, and obtain the identification results to determine relevant information about underground pipelines. The crowd intelligence sensing module is used to generate a predicted map of the distribution of underground pipeline information, optimize the identification results of underground pipelines, and obtain complete distribution information of underground pipelines in the target area. The results output module is used to visually label and display the identification results or complete distribution information on an electronic map.

[0032] The working principle and beneficial effects of the proposed technical solution have been explained in the method section, and will not be repeated here.

[0033] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.

Claims

1. A method for detecting underground pipelines based on multi-source data fusion and intelligent analysis, characterized in that, include: Collect multi-source data of the target area; the multi-source data includes electromagnetic signal data, electromagnetic wave reflection data and sound wave propagation data; filter and normalize the multi-source data to obtain the target data; A multi-source data fusion model is constructed to perform feature-level fusion of target data and generate a fused feature vector. A deep learning-based underground pipeline identification model is constructed to intelligently analyze and identify fused feature vectors, thereby obtaining the identification results of underground pipelines. The recognition results are marked and displayed on an electronic map in a visual way.

2. The underground pipeline detection method based on multi-source data fusion and intelligent analysis according to claim 1, characterized in that, Collect multi-source data for the target area, including: Electromagnetic induction detectors are used to detect the target area and collect electromagnetic signal data of underground metal pipelines to obtain rough location and direction information of the underground metal pipelines. Ground-penetrating radar is used to scan the target area and collect electromagnetic wave reflection data of underground pipelines to obtain information on the burial depth and diameter of underground pipelines. Acoustic wave detectors are used to conduct acoustic wave detection in the target area and collect acoustic wave propagation data of the soil medium around underground pipelines. This data is used to obtain information on the differences between underground pipelines and the surrounding soil medium, providing auxiliary information for the location of non-metallic underground pipelines.

3. The underground pipeline detection method based on multi-source data fusion and intelligent analysis according to claim 1, characterized in that, Filtering and normalization preprocessing of multi-source data yields the target data, including: The collected electromagnetic signal data, electromagnetic wave reflection data, and sound wave propagation data are filtered to remove noise interference. The filtered data is normalized to convert data of different types and dimensions into a unified numerical range to obtain the target data. For electromagnetic signal data, an adaptive filtering algorithm is used to automatically adjust the filtering parameters according to the characteristics of the signal, effectively removing noise generated by the surrounding electromagnetic environment. For electromagnetic wave reflection data, wavelet filtering algorithm is used to analyze and process signals of different frequency components to remove high-frequency noise and low-frequency interference; For sound wave propagation data, a bandpass filtering algorithm is used to retain only the signal within the effective frequency range and remove noise at irrelevant frequencies.

4. The underground pipeline detection method based on multi-source data fusion and intelligent analysis according to claim 1, characterized in that, A multi-source data fusion model is constructed to perform feature-level fusion of target data, generating a fused feature vector, including: A multi-source data fusion model is established to fuse preprocessed electromagnetic signal data, electromagnetic wave reflection data, and sound wave propagation data; During the fusion process, a feature-level fusion method is adopted to extract feature vector information of various types of data, and the extracted feature vector information is combined to form a fusion feature vector containing multiple aspects of underground pipeline information, thereby achieving a comprehensive description of underground pipeline information.

5. The underground pipeline detection method based on multi-source data fusion and intelligent analysis according to claim 4, characterized in that, Fusion methods that utilize multi-source data fusion models to perform feature-level fusion of target data include: Obtain the preprocessed target data; Three pre-constructed independent feature extraction sub-networks are used to extract features from electromagnetic signal data, electromagnetic wave reflection data, and sound wave propagation data in the target data, respectively, to obtain feature vector information of the three types of data; the three independent feature extraction sub-networks include electromagnetic signal feature extraction sub-network, electromagnetic wave reflection feature extraction sub-network, and sound wave feature extraction sub-network. The attention weights among the three types of feature vectors are calculated based on the modal attention mechanism. The attention weights are then used to enhance each feature vector. The enhanced feature vectors are then concatenated and compressed through a fully connected layer to obtain a fused feature vector.

6. The underground pipeline detection method based on multi-source data fusion and intelligent analysis according to claim 1, characterized in that, A deep learning-based underground pipeline identification model is constructed to intelligently analyze and identify fused feature vectors, yielding the identification results of underground pipelines, including: We obtained several labeled underground pipeline data as training samples, and used a convolutional neural network as the basic architecture to build an underground pipeline identification model based on deep learning. The fused feature vector is input into the underground pipeline identification model, and the identification results of the underground pipeline features are output, so as to accurately identify the type, location, direction, burial depth and diameter of the underground pipeline.

7. The underground pipeline detection method based on multi-source data fusion and intelligent analysis according to claim 1, characterized in that, Also includes: The collection of underground pipeline change events based on crowd-sensing involves collecting multi-source sensing data from corresponding-level data sources within the target area through data source hierarchical rules; the multi-source sensing data includes text information and image information. Multi-source sensing data is subjected to hierarchical cleaning and standardization to obtain preprocessed data; hierarchical cleaning involves filtering highly relevant data based on the data source level, and standardization involves unifying the geographic description format and image coordinate system. Natural language processing and computer vision processing technologies are used to extract underground pipeline-related information from the preprocessed data to obtain target data, which is then used to generate a predicted map of underground pipeline information distribution. By using the underground pipeline information distribution prediction map, the identification results of underground pipelines are optimized to obtain complete distribution information of underground pipelines in the target area; The complete distribution information is marked and displayed on an electronic map in a visual way.

8. The underground pipeline detection method based on multi-source data fusion and intelligent analysis according to claim 7, characterized in that: The data source classification rules categorize data sources into Tier 1, Tier 2, and Tier 3 data sources based on their reliability. Primary data sources include official pipeline reporting systems and shared data from authorized construction units; Secondary data sources include industry communities and government new media interaction data; Level 3 data sources include public content from social media platforms and historical archive OCR data; Targeted data collection includes: Real-time data synchronization between primary and secondary data sources is achieved through API integration or compliant web crawling. Data is collected through a dual-filtering approach using keyword search and geofencing to access three levels of data sources. Hierarchical cleaning of multi-source sensing data includes: The data collected from the primary data source is filtered for format errors using a format anomaly detection method. By using location and keyword filtering rules, content without location information or pipeline keywords in secondary and tertiary data sources is filtered out. Then, the TextCNN text classification model and ResNet image classification model are used to filter highly relevant data with a confidence level ≥ 0.

7. The target data to be extracted includes: Natural language processing technology is used to extract pipeline entity information and spatial relationships from text information, and computer vision processing technology is used to identify pipeline features and scene auxiliary information from image information, and the target data is obtained by integrating them.

9. The underground pipeline detection method based on multi-source data fusion and intelligent analysis according to claim 7, characterized in that, By using the underground pipeline information distribution prediction map, the identification results of underground pipelines are optimized to obtain complete distribution information of underground pipelines within the target area, including: The rule base of safety standards for the construction of urban underground pipelines is obtained as a constraint condition, and a data fusion model is generated using training data. This model is used to fuse and optimize the underground pipeline information distribution prediction map based on the identification results of underground pipelines, so as to obtain the complete distribution information of underground pipelines in the target area. The training steps for a data fusion model include: Acquire multiple sets of training sample data; each set of training sample data corresponds to the baseline data and detailed data within a target area. The baseline data is the identification result of underground pipelines within the target area, and the detailed data is the predicted distribution map data of underground pipeline information within the target area. A dual-branch feature extraction module consisting of a baseline branch and a detail branch is constructed. The baseline branch transforms the recognition results into low-dimensional vectors that preserve topological relationships through a fully connected layer. The detail branch includes text feature extraction and image feature extraction. Text feature extraction uses a pre-trained BERT model to semantically encode the text information in the detail data and outputs a text feature vector. Image feature extraction uses a CNN model to perform coordinate mapping on the image information in the detail data and extracts the spatial coordinate features of pipeline endpoints and inflection points, outputting a coordinate feature vector. The text feature vector and the coordinate feature vector are concatenated and input into a GNN model to learn the pipeline detail features, outputting the pipeline detail feature vector. Load the rule library of urban underground pipeline construction safety standards containing quantitative constraint indicators, perform safety compliance screening on the pipeline detail feature vectors output by the detail branches, and retain the pipeline detail feature vectors that comply with the urban underground pipeline construction safety standards, which are denoted as compliance feature vectors. Using the low-dimensional vector output by the baseline branch as the anchor point, calculate the spatial Euclidean distance between the pipeline coordinates in the compliance feature vector output by the detail branch and the pipeline coordinates in the baseline data. If the spatial Euclidean distance is less than the preset distance, it is determined to be a high-matching region, and the baseline data is supplemented with the detail data in the compliance feature vector; otherwise, it is determined to be a low-matching region, and only the baseline data is retained. By integrating the completion results from high-matching regions and the retained baseline data from low-matching regions, preliminary fusion results are obtained, forming a complete framework for the fusion model. Set a loss function and train the complete framework of the fusion model by combining multiple sets of training sample data until the preset training conditions are met, and then output the data fusion model.

10. An underground pipeline detection system based on multi-source data fusion and intelligent analysis, characterized in that, include: The data acquisition module includes an electromagnetic induction detector, a ground-penetrating radar, and a sound wave detector, used to collect electromagnetic signal data, electromagnetic wave reflection data, and sound wave propagation data of the target area; The data preprocessing module is used to filter and normalize the collected data to generate target data; The data fusion module is used to perform feature-level fusion on preprocessed data using a multi-source data fusion model to generate a fused feature vector. The intelligent analysis module is used to analyze and identify fused feature vectors using a deep learning-based underground pipeline identification model, and obtain the identification results to determine relevant information about underground pipelines. The results output module is used to visually label and display the recognition results on an electronic map.

Images