An underground pipeline detection mapping and geographic information intelligent management system

Abstract

This invention relates to an intelligent management system for underground pipeline detection, mapping, and geographic information, comprising a data acquisition terminal, edge computing nodes, and a cloud platform. The data acquisition terminal collects raw observation data. The edge computing nodes perform preliminary processing of the raw observation data on-site. The cloud platform includes an access layer, a computing layer, a quality layer, and a results layer. The access layer performs data unpacking, format conversion, coordinate transformation, metadata registration, and evidence storage. The computing layer performs fine processing and target identification on the full GPR (Geographic Information Mapping) data, fusion and optimization on multi-source alignments, and defect identification and structuring on CCTV (Gas Video Recording). The quality layer outputs quality inspection reports and review work orders. The results layer stores approved objects in a spatial database and establishes version and temporal records, forming a pipeline asset database. This invention effectively improves the efficiency and reliability of underground pipeline detection, mapping, and management.

Description

Technical Field

[0001] This invention relates to the field of information and data management, and in particular to an intelligent management system for underground pipeline detection, mapping, and geographic information. Background Technology

[0002] With the acceleration of urbanization and the increasing density of underground space utilization, underground pipeline detection and mapping face increasingly severe technical challenges and application demands. Traditional underground pipeline detection mainly relies on single technologies such as electromagnetic induction, ground penetrating radar (GPR), and pipeline endoscopic inspection (CCTV), but these are constrained by bottlenecks such as limited detection accuracy, low automation of data processing, and difficulty in multi-source information fusion. While electromagnetic induction is highly sensitive to metallic pipelines, it is ineffective for non-metallic pipelines and is prone to interference and misjudgment in complex electromagnetic environments. GPR technology can detect pipelines of various materials, but signal interpretation heavily relies on professional experience, the accuracy of automated target identification is low, and its penetration ability is limited in soils with high water content. CCTV inspection can directly reflect the internal condition of pipelines, but the detection efficiency is low, and defect identification mainly relies on manual interpretation, resulting in strong subjectivity and a high rate of missed detections. In addition, the existing technical system lacks unified data standards and quality control mechanisms, making it difficult to effectively integrate and maintain data generated by different equipment and projects in the long term. This leads to problems such as redundant investment, data silos, and low utilization of results, urgently requiring the construction of integrated and intelligent technical solutions. Summary of the Invention

[0003] To address the aforementioned problems, the present invention aims to provide an intelligent management system for underground pipeline detection, mapping, and geographic information, which effectively improves the efficiency and reliability of underground pipeline detection, mapping, and management.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: An intelligent management system for underground pipeline detection, mapping, and geographic information includes a data acquisition terminal, edge computing nodes, and a cloud platform. The data acquisition terminal consists of detection equipment, measurement and positioning equipment, and a field application (APP) for collecting and forming raw observation data. The edge computing nodes are used for preliminary processing of the raw observation data on-site. The cloud platform includes an access layer, a computing layer, a quality layer, and a results layer. The access layer receives the observation data after preliminary processing by the edge computing nodes and completes data unpacking, format conversion, coordinate conversion, metadata registration, and evidence storage. The computing layer performs fine processing and target identification on the full GPR, fusion and optimization on multi-source alignments, and defect identification and structuring on CCTV. The quality layer outputs quality inspection reports and review work orders through a rule engine and AI anomaly detection. The results layer solidifies the approved pipeline segments, appurtenances, and manhole objects into a spatial database and establishes version and temporal records to form a pipeline asset database.

[0005] Furthermore, the detection equipment includes electromagnetic pipeline detectors, GPR, CCTV inspection equipment, and probe equipment; the measurement and positioning equipment includes GNSS RTK, total station, and SLAM; the system binds the output of each device with a unified timestamp, coordinate frame, and task number to ensure that the data can be integrated and traced.

[0006] Furthermore, the field application is task-centric, supporting task assignment, regional segmentation, offline base map download and offline data collection, and providing standardized forms, trajectory recording, photo / video capture, voice-to-text recording, and attachment management. The field application supports Bluetooth / serial port connection to devices, enabling automatic backfilling of observation values. It also has built-in real-time quality gating rules, providing immediate prompts on positioning quality, attribute integrity, geometric rationality, and consistency with existing data. For high-risk categories or low-confidence sections, it generates retesting suggestions, including supplementing data collection points, switching to active mode, adding GPR cross-sectional verification, and adding excavation verification points.

[0007] Furthermore, the raw observation data undergoes preliminary processing, specifically as follows: on the GPR side, denoising, gain adjustment, time-zero correction, mileage alignment, and coarse extraction of candidate targets are performed; on the EM side, observation cleaning, outlier removal, centerline fitting, and depth curve smoothing are performed, and preliminary confidence levels are provided.

[0008] Furthermore, the GPR side completes denoising, gain, time-zero correction, mileage alignment, and coarse extraction of candidate targets, as follows: The GPR side takes the original A-scan and B-scan data as input and first performs denoising and enhancement, including DC drift, background removal, bandpass filtering and power frequency interference removal, time-varying gain to compensate for deep energy attenuation, and normalization of amplitude between different survey lines; then, time-zero correction and time-depth conversion preparation are performed; in terms of mileage alignment, radar sampling triggering and GNSS trajectory are synchronized, and the profile is resampled and equidistantized; multiple survey lines are spliced ​​and managed according to task number and survey line direction to generate a quick-viewable profile quick view; in terms of coarse extraction of candidate targets, local peak tracking is used to mark suspected pipe reflections / regions, and candidate points and corresponding evidence fragment indices are output.

[0009] Furthermore, the EM side completes observation cleaning, outlier removal, centerline fitting, and depth curve smoothing, and provides preliminary confidence levels, as follows: The EM side uses the initial observation sequence as input, first performing observation cleaning and standardization: unifying units and fields, removing missing values, saturated readings, and obviously unreasonable depths; combining GNSS quality indicators to remove suspected positioning anomalies; centerline fitting is implemented robustly on the edge side: segmented fitting and smoothing are performed according to the sampling point sequence, inflection points are extracted, and continuous line positions are generated; depth curve smoothing uses a sliding window method; preliminary confidence levels are provided by multi-factor fusion, including: signal quality, multi-frequency consistency, observation density and repeatability, and positioning quality; edge nodes output the confidence level of each pipeline segment and its dominant cause label.

[0010] Furthermore, the full GPR is subjected to fine-grained processing and target identification, as follows: Standardized signal preprocessing is performed on the full GPR data, including denoising and background removal, filtering and gain compensation, time-zero calibration, odometer alignment and equidistant resampling; The denoising and background removal are performed by eliminating the DC component through in-channel mean subtraction and using moving average background estimation. ; Where B(t) is the moving average or smoothed signal value at time t; M is the length of the moving window; i is the index position of the current data point; k is the summation variable, representing the index of each data point within the window; s k (t) represents the value of the original signal with index k at time t; iM / 2 to i+M / 2 is a symmetrical window range centered on the current point i, with M / 2 points before and after it; Remove horizontal stripes and combine wavelet transform or frequency domain filtering to suppress high-frequency noise and power frequency interference; Filtering and gain compensation employ a Butterworth bandpass filter to preserve the effective frequency band, and utilize a time-varying gain function. G(t) is the function value at time t; G0 is the baseline value of the initial value; a is the growth rate constant; e is the base of the natural logarithm; compensating for the exponential attenuation of electromagnetic waves in the medium to ensure the detectability of deep targets; time-zero calibration by finding the direct wave calibration system delay, mileage alignment and equidistant resampling resample the non-uniformly acquired profiles to the standard spatial grid; An intelligent recognition method combining segmentation model and feature matching is employed to automatically identify suspected pipe echo features and output structured target elements in the preprocessed GPR profile. The segmentation model is based on a convolutional neural network architecture and uses the normalized contrast-enhanced profile image I... enhanced Given (x,t) as input, the bounding box (x) is output through end-to-end training. i ,ti ,w i ,h i ,c i ,p i ), which includes the target location (x) i ,t i ), size (w) i ,h i Category c) i With model confidence p i The feature matching method uses a hyperbola template. ; Where x is the spatial coordinate variable; t is the time coordinate variable; x0 is the spatial reference point; t0 is the time reference point; and a is the scale parameter, which controls the spatiotemporal coupling strength. The template variance controls the broadening of the kernel function; It is a composite measure of spatiotemporal differences; Template matching is performed with normalized cross-correlation, and target extraction converts the identification results into standardized elements, including spatial location, morphological indicators, intensity indicators, category probability and comprehensive confidence. Finally, an evidence fragment index is established, which associates each candidate target with the specific interval of the original profile, key frame snapshot and processing parameter record, forming a complete structured output that includes target ID, geometric attributes, signal features, classification results, confidence assessment, uncertainty quantification and evidence chain tracing.

[0011] Furthermore, the multi-source alignment is fused and optimized, unifying the EM centerline and depth curve, GPR candidate targets, high-confidence nodes such as well points, existing pipeline network and as-built drawing data, trajectory and terrain constraints to the same coordinate reference. The three-dimensional centerline and burial depth curve are generated through graph optimization solution, and the segmented confidence and uncertainty corridor are output. Conflict labels and review suggestions are generated for conflict segments. The system performs defect identification and structured processing on CCTV data, completes video decoding and quality assessment, timecode-mileage code alignment and pipe segment mapping, and uses video detection and time-series classification models to identify defects such as cracks, leaks, deformations, misalignments, deposits, and tree root intrusion. It outputs defect type, level, occurrence mileage range, keyframe evidence and confidence level, and writes them into structured defect records according to a unified dictionary encoding, which are then associated with the corresponding pipe segment objects to support operation and maintenance and risk analysis.

[0012] Furthermore, the quality inspection report and review work order are output through the rule engine and AI anomaly detection, as follows: The quality layer receives structured data produced by the computing layer, including complete information such as pipeline geometry, attribute parameters, confidence assessment, uncertainty quantification, and evidence chain index. Complementary quality inspection is achieved through parallel processing of the two engines: The rule engine performs deterministic verification based on a predefined constraint rule base, including quantifiable quality dimensions such as data integrity, coding standards, geometric topology, engineering standards, and evidence integrity; The AI ​​anomaly detection engine uses machine learning methods to identify implicit anomaly patterns and statistical deviations. The outputs of the two engines are weighted and fused and thresholded to generate graded quality inspection conclusions of pass, require rectification, or reject. Structured quality inspection reports and actionable review work orders are automatically generated based on the problem type, severity, and spatial distribution.

[0013] Furthermore, the deliverables layer receives graded quality inspection conclusions from the quality layer, indicating whether the items have passed, require rectification, or have been rejected. For items that have passed, the layer performs data storage and solidification: writing 2D / 3D geometry, attributes, accuracy level and confidence level, uncertainty parameters, evidence chain references, and source and processing link identifiers into the spatial database; simultaneously establishing version and temporal records, generating version_id and validity period fields for each addition, change, or obsolescence, supporting historical backtracking, difference comparison, and rollback; the deliverables layer solidifies object relationships, including the association between pipe sections and appurtenances / well chambers, pipeline topology relationships and segmentation rules, and establishes stable associations between structured defect records and pipe sections to support the operation and maintenance closed loop.

[0014] The present invention has the following beneficial effects: 1. This invention adopts a three-layer architecture of acquisition end-edge-cloud to realize intelligent collaboration of the entire process of underground pipeline detection from data acquisition to results application. It has significant advantages in data quality assurance and processing efficiency. Multi-device fusion acquisition overcomes the limitations of single detection methods through the unified scheduling of multi-detection devices such as GPR, electromagnetic induction, and CCTV and high-precision measurement and positioning devices, and achieves full coverage detection of pipelines of different materials and burial depths, significantly improving detection accuracy. 2. The multi-source pipeline fusion optimization of this invention unifies the modeling and robust optimization of multiple information sources such as GPR target identification, EM pipeline detection, well point measurement, and existing data. Through a mathematical framework combining strong and soft constraints, it effectively solves the data conflict and inconsistency problem between different detection methods, outputs a three-dimensional pipeline position with confidence quantification, and significantly improves the accuracy and reliability of the position. The AI-driven automatic identification adopts a hybrid method combining deep learning and traditional algorithms to automatically identify and extract pipeline features in GPR profiles and defects in CCTV videos, which greatly improves data processing efficiency and reduces the subjectivity and false negative rate of manual interpretation. 3. This invention employs a dual-engine quality control approach, combining the deterministic verification of a rule engine with the statistical learning capabilities of AI anomaly detection. This enables comprehensive identification and tiered handling of data quality issues, ensuring the engineering usability and legal validity of the resulting data. Attached Figure Description

[0015] Figure 1 This is a system architecture diagram of the present invention. Detailed Implementation

[0016] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments: refer to Figure 1 In this embodiment, an intelligent management system for underground pipeline detection, mapping, and geographic information is provided, including a data acquisition terminal, edge computing nodes, and a cloud platform. The data acquisition terminal consists of detection equipment, measurement and positioning equipment, and a field application, used to collect and form raw observation data. The edge computing nodes are used to perform preliminary processing of the raw observation data on-site. The cloud platform includes an access layer, a computing layer, a quality layer, and a results layer. The access layer receives the observation data after preliminary processing by the edge computing nodes and completes data unpacking, format conversion, coordinate conversion (including parameter management and recording), metadata registration, and evidence storage. The computing layer performs fine processing and target identification on the full GPR, fusion optimization on multi-source alignments, and defect identification and structuring on CCTV. The quality layer outputs quality inspection reports and review work orders through a rule engine and AI anomaly detection. The results layer solidifies the approved pipeline segments, appurtenances, and manhole objects into a spatial database and establishes version and temporal records to form a queryable, analyzable, and traceable pipeline asset database.

[0017] In this embodiment, the detection equipment includes an electromagnetic pipeline detector (active / passive mode, supporting multi-frequency, multi-coil attitude and signal strength / phase observation outputs), a GPR (frequency band selection based on depth and resolution, supporting odometer / triggered sampling), a CCTV inspection device (video and odometer encoding, defect labeling), and a probe device (point depth verification); the measurement and positioning equipment includes GNSS RTK (fixed solution status, DOP, base station information recording), a total station (obstructed area and high-precision control), and SLAM (trajectory and point cloud-assisted positioning in indoor / tunnel / obstructed environments); the system binds the outputs of each device with a unified timestamp, coordinate frame, and task number to ensure data fusion and traceability.

[0018] In this embodiment, the field application is task-centric, supporting task assignment, area segmentation, offline base map download and offline data acquisition. It provides standardized forms (pipe type, material, diameter, ownership, laying method, burial depth definition, detection method, etc.), trajectory recording (detection route, sampling sequence), photo / video acquisition (automatically writing EXIF ​​extended information such as time, location, and direction), voice-to-text recording, and attachment management (as-built drawings, on-site records). The field application supports Bluetooth / serial port connection to devices, enabling automatic backfilling of observation values ​​(such as EM depth, signal indicators, automatic association of GPR files with acquisition trajectories, CC...). TV video is synchronized with mileage, reducing errors from manual transcription; the field APP also has built-in real-time quality gating rules, which provide immediate prompts on positioning quality (RTK fixed solution, DOP threshold, number of satellites, intermittent positioning), attribute integrity (required fields, coding legality), geometric rationality (point jumps, abnormal polylines, depth abrupt changes), and consistency with existing data (significant deviation from the bottom reservoir pipeline, suspected crossing conflict); for high-risk pipelines or low-confidence sections, retest suggestions are generated, including supplementing sampling points, switching to active mode, adding GPR cross-section verification, and adding excavation verification points to achieve on-site correction and reduce rework rate.

[0019] In this embodiment, the raw observation data undergoes preliminary processing as follows: on the GPR side, denoising, gain, time-zero correction, mileage alignment, and coarse extraction of candidate targets are performed; on the EM side, observation cleaning, outlier removal, centerline fitting, and depth curve smoothing are performed, and a preliminary confidence level is given.

[0020] In this embodiment, the GPR side performs denoising, gain, time-zero correction, odometer alignment, and coarse extraction of candidate targets, as follows: The GPR side takes the original A-scan and B-scan data as input and first performs denoising and enhancement, including DC drift, background removal (eliminating horizontal stripes and system noise), bandpass filtering and power frequency interference removal, time-varying gain (such as SEC / AGC) to compensate for deep energy attenuation, and normalizes the amplitude between different survey lines to improve comparability; then, time-zero correction (calibrating the transmit-receive system delay and unifying the zero-time position) and time depth conversion are performed. Preparation (recording initial values ​​of dielectric constant and velocity model; complex velocity inversion is generally not performed on the edge side, but parameter locations are retained); for mileage alignment, radar sampling triggering is synchronized with GNSS trajectory, and the profile is resampled and equidistantized to eliminate lateral stretching caused by velocity unevenness; multiple survey lines are spliced ​​and managed according to task number and survey line direction to generate a quick-viewable profile quick view; for coarse extraction of candidate targets, local peak tracking is used to mark suspected pipe reflections / regions, and candidate points (mileage, two-way time / estimated depth, target intensity index) and corresponding evidence fragment index are output.

[0021] In this embodiment, the EM side completes observation cleaning, outlier removal, centerline fitting, and depth curve smoothing, and provides an initial confidence level, as follows: The EM side takes multi-frequency, multi-mode (active / passive) initial observation sequences as input, and first performs observation cleaning and standardization: unifying units and fields, removing missing values, saturation readings, and obviously unreasonable depths; combined with GNSS quality indicators (fixed solution, DOP, velocity mutation) to remove suspected positioning anomalies. The anomaly removal adopts a rule + statistical joint strategy, such as identifying signal strength / phase mutations, depth jumps (exceeding thresholds and lacking corresponding turning points / well point evidence), and points with contradictory directions to adjacent points, and retaining the reasons for removal and a traceable list; centerline fitting is implemented robustly on the edge side: segmented fitting and smoothing are performed according to the sampling point sequence (e.g., RANS). AC / weighted least squares + curvature constraints are used to extract inflection points and generate continuous alignments. The depth curve smoothing adopts a sliding window method to suppress jitter while preserving the true trend, and allows for local constraint enhancement near roads and well points. The preliminary confidence level is given by multi-factor fusion, including: signal quality (amplitude, phase stability, peak sharpness), multi-frequency consistency (the degree of consistency between alignment and depth in different frequencies / modes), observation density and repeatability (consistency in round trips or cross-validation of the same road segment), and positioning quality (proportion of GNSS fixed solutions, trajectory continuity). The edge nodes output the confidence level of each pipeline segment and its dominant cause label (such as "high multi-frequency consistency", "average positioning quality", "strong suspected interference"), and automatically mark low-confidence road segments as key sections that need to be retested or recommended to be converted to GPR / excavation verification.

[0022] In this embodiment, the full GPR is finely processed and target is identified, as follows: Standardized signal preprocessing is performed on the full GPR data, including denoising and background removal, filtering and gain compensation, time-zero calibration, odometer alignment and equidistant resampling, to eliminate system noise, DC drift, horizontal stripe interference and compensate for deep signal attenuation; The denoising and background removal are performed by eliminating the DC component through in-channel mean subtraction and using moving average background estimation. ; Where B(t) is the moving average or smoothed signal value at time t; M is the length of the moving window; i is the index position of the current data point; k is the summation variable, representing the index of each data point within the window; s k (t) represents the value of the original signal with index k at time t; iM / 2 to i+M / 2 is a symmetrical window range centered on the current point i, with M / 2 points before and after it; Remove horizontal stripes and combine wavelet transform or frequency domain filtering to suppress high-frequency noise and power frequency interference; Filtering and gain compensation employ a Butterworth bandpass filter to retain the effective frequency band of 100MHz-1000MHz, and utilize a time-varying gain function. G(t) is the function value at time t; G0 is the baseline value of the initial value; a is the growth rate constant; e is the base of the natural logarithm; compensating for the exponential attenuation of electromagnetic waves in the medium ensures the detectability of deep targets; time-zero calibration, by finding the direct wave calibration system delay, mileage alignment, and equidistant resampling, resamples the non-uniformly acquired profiles to a standard spatial grid, laying the data foundation for subsequent unified processing and target identification; An intelligent recognition method combining segmentation model and feature matching is employed to automatically identify suspected pipe echo features and output structured target elements in the preprocessed GPR profile. The segmentation model is based on a convolutional neural network architecture and uses the normalized contrast-enhanced profile image I... enhanced Given (x,t) as input, the bounding box (x) is output through end-to-end training. i ,t i ,w i ,h i ,c i ,p i ), which includes the target location (x) i ,t i ), size (w) i ,h i Category c) i With model confidence p i The feature matching method uses a hyperbola template. ; Where x is the spatial coordinate variable; t is the time coordinate variable; x0 is the spatial reference point; t0 is the time reference point; and a is the scale parameter, which controls the spatiotemporal coupling strength. The template variance controls the broadening of the kernel function; It is a composite measure of spatiotemporal differences; Template matching is performed with normalized cross-correlation, and target extraction converts the identification results into standardized elements, including spatial location (mileage coordinates and depth), morphological indicators (aspect ratio, area, compactness), intensity indicators (signal-to-noise ratio, peak energy), class probability (Softmax output), and comprehensive confidence. Finally, an evidence fragment index is established, which associates each candidate target with the specific interval of the original profile, key frame snapshots, and processing parameter records, forming a complete structured output that includes target ID, geometric attributes, signal features, classification results, confidence assessment, uncertainty quantification, and evidence chain tracing. This provides a standardized data interface for subsequent multi-source fusion, quality inspection, and result consolidation.

[0023] In this embodiment, the multi-source alignment is fused and optimized, and the constraints such as the EM centerline and depth curve, GPR candidate targets, well points and other high-confidence nodes, existing pipeline network and as-built drawing data, trajectory and terrain are unified to the same coordinate reference. The three-dimensional centerline and burial depth curve are generated through graph optimization solution, and the segmented confidence and uncertainty corridor are output at the same time. Conflict labels and review suggestions are generated for conflict segments. The system performs defect identification and structured processing on CCTV data, completes video decoding and quality assessment, timecode-mileage code alignment and pipe segment mapping, and uses video detection and time-series classification models to identify defects such as cracks, leaks, deformations, misalignments, deposits, and tree root intrusion. It outputs defect type, level, occurrence mileage range, keyframe evidence and confidence level, and writes them into structured defect records according to a unified dictionary encoding, which are then associated with the corresponding pipe segment objects to support operation and maintenance and risk analysis.

[0024] In this embodiment, a quality inspection report and a review work order are output through a rule engine and AI anomaly detection, as follows: The quality layer receives structured data produced by the computation layer, including complete information such as pipeline geometry, attribute parameters, confidence assessment, uncertainty quantification, and evidence chain index. Complementary quality inspection is achieved through parallel processing of dual engines: The rule engine performs deterministic verification based on a predefined constraint rule base, including quantifiable quality dimensions such as data integrity, coding standards, geometric topology, engineering standards, and evidence integrity; The AI ​​anomaly detection engine uses machine learning methods to identify implicit anomaly patterns and statistical deviations. The dual engine outputs are weighted and fused and thresholded to generate graded quality inspection conclusions of pass, require rectification, or reject. Based on the problem type, severity, and spatial distribution, a structured quality inspection report and an actionable review work order are automatically generated.

[0025] In this embodiment, the results layer receives graded quality inspection conclusions from the quality layer, indicating whether the quality layer has passed, requires rectification, or rejected. For objects that pass, the layer performs database storage and solidification: writing 2D / 3D geometry, attributes, accuracy level and confidence level, uncertainty parameters, evidence chain references, and source and processing link identifiers into the spatial database; simultaneously establishing version and temporal records, generating version_id and validity period fields for each addition, change, or obsolescence, supporting historical retrospection, difference comparison, and rollback; the results layer solidifies object relationships, including the association between pipe segments and appurtenances / well chambers, pipeline topology relationships and segmentation rules, and establishes stable associations between structured defect records and pipe segments to support closed-loop operation and maintenance; the results layer provides query and analysis capabilities, including filtering by ownership / pipe type / age / risk, spatial retrieval, net distance and conflict analysis, excavation impact analysis, and emergency impact range simulation, and publishes these to the outside world using standard GIS services and business APIs, while ensuring configurable access control and desensitization strategies to ensure that sensitive pipelines meet safety and compliance requirements in shared scenarios.

[0026] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0027] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will 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 program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0028] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0029] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0030] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An intelligent management system for underground pipeline detection, mapping, and geographic information, characterized in that, The system comprises a data acquisition terminal, edge computing nodes, and a cloud platform. The data acquisition terminal consists of detection equipment, measurement and positioning equipment, and a field application, used to collect and generate raw observation data. The edge computing nodes are used for preliminary processing of the raw observation data on-site. The cloud platform includes an access layer, a computing layer, a quality layer, and a results layer. The access layer receives the observation data after preliminary processing by the edge computing nodes and performs data unpacking, format conversion, coordinate transformation, metadata registration, and evidence storage. The computing layer performs fine processing and target identification on the full GPR data, fusion optimization on multi-source alignments, and defect identification and structuring on CCTV data. The quality layer outputs quality inspection reports and review work orders through a rule engine and AI anomaly detection. The results layer will solidify the approved pipeline segments, appurtenances, and manhole objects into the spatial database and establish version and temporal records to form a pipeline asset library.

2. The intelligent management system for underground pipeline detection, mapping, and geographic information according to claim 1, characterized in that, The detection equipment includes electromagnetic pipeline detectors, GPR, CCTV detection equipment, and probe equipment; the measurement and positioning equipment includes GNSSRTK, total station, and SLAM; the system binds the output of each device with a unified timestamp, coordinate frame, and task number to ensure that the data can be integrated and traced.

3. The intelligent management system for underground pipeline detection, mapping, and geographic information according to claim 2, characterized in that, The field application is task-centric, supporting task assignment, regional segmentation, offline base map download and offline data collection, and providing standardized forms, trajectory recording, photo / video capture, voice-to-text recording and attachment management. The application supports Bluetooth / serial port connection to devices, enabling automatic backfilling of observation values. It also includes built-in real-time quality gating rules, providing immediate prompts on positioning quality, attribute integrity, geometric rationality, and consistency with existing data. For high-risk categories or low-confidence sections, it generates retesting suggestions, including supplementing data collection points, switching to active mode, adding GPR cross-sectional verification, and adding excavation verification points.

4. The intelligent management system for underground pipeline detection, mapping, and geographic information according to claim 1, characterized in that, The initial processing of the raw observation data is as follows: on the GPR side, denoising, gain, time-zero correction, mileage alignment and coarse extraction of candidate targets are performed; on the EM side, observation cleaning, outlier removal, centerline fitting and depth curve smoothing are performed, and a preliminary confidence level is given.

5. The intelligent management system for underground pipeline detection, mapping, and geographic information according to claim 4, characterized in that, The GPR side performs denoising, gain, time-zero correction, mileage alignment and coarse extraction of candidate targets, as follows: The GPR side takes the original A-scan and B-scan data as input, and first performs denoising and enhancement: including DC drift, background removal, bandpass filtering and power frequency interference removal, time-varying gain to compensate for deep energy attenuation, and normalizes the amplitude between different measurement lines. Then, preparations for time zero correction and time depth conversion are carried out; In terms of mileage alignment, radar sampling triggering is synchronized with GNSS trajectory, and the profile is resampled and isochronized. Multiple survey lines are spliced ​​and managed according to task number and survey line direction to generate quick profile views that can be viewed quickly; in terms of coarse extraction of candidate targets, local peak tracking is used to mark suspected pipe reflections and output candidate point locations and corresponding evidence fragment indexes.

6. The intelligent management system for underground pipeline detection, mapping, and geographic information according to claim 4, characterized in that, The EM side completes observation cleaning, outlier removal, centerline fitting, and depth curve smoothing, and provides a preliminary confidence level, as follows: The EM side takes the initial observation sequence as input and first performs observation cleaning and consistency: unifying units and fields, removing missing values, saturated readings, and obviously unreasonable depths; combining GNSS quality indicators to remove suspected positioning anomalies; centerline fitting is implemented robustly on the edge side: segmented fitting and smoothing are performed according to the sampling point sequence, inflection points are extracted, and continuous line positions are generated; depth curve smoothing adopts the sliding window method; the preliminary confidence level is given by multi-factor fusion, including factors such as signal quality, multi-frequency consistency, observation density and repeatability, and positioning quality; the edge nodes output the confidence level of each pipeline segment and its dominant cause label.

7. The intelligent management system for underground pipeline detection, mapping, and geographic information according to claim 1, characterized in that, The detailed processing and target identification of the full GPR are as follows: Standardized signal preprocessing is performed on the full GPR data, including denoising and background removal, filtering and gain compensation, time-zero calibration, odometer alignment and equidistant resampling; An intelligent recognition method combining segmentation model and feature matching is adopted to automatically recognize the suspected pipe echo features in the pre-processed GPR profile and output structured target elements. The segmentation model is based on a convolutional neural network architecture, which takes the normalized contrast-enhanced profile image I enhanced (x,t) as input and outputs the bounding box(x i ,t i ,w i ,h i ,c i ,p i ) through end-to-end training, where it contains the target position(x i ,t i ), size(w i ,h i ), category c i and model confidence p i ; the feature matching method uses hyperbolic curve template and normalized cross-correlation for template matching, and the target extraction converts the recognition results into standardized elements, including spatial position, morphological index, intensity index, category probability and comprehensive confidence. Finally, the evidence segment index is established, which associates each candidate target to the specific interval, key frame snapshot and processing parameter record of the original profile, forming a complete structured output containing target ID, geometric attribute, signal feature, classification result, confidence evaluation, uncertainty quantification and evidence chain traceability.

8. The intelligent management system for underground pipeline detection, mapping, and geographic information according to claim 7, characterized in that, The multi-source alignment fusion optimization unifies the EM centerline and depth curve, GPR candidate targets, high-confidence nodes such as well points, existing pipeline network and as-built drawing data, trajectory and terrain constraints to the same coordinate reference. The three-dimensional centerline and burial depth curve are generated through graph optimization solution. At the same time, segmented confidence and uncertainty corridors are output, and conflict labels and review suggestions are generated for conflict segments. The system performs defect identification and structured processing on CCTV data, completes video decoding and quality assessment, timecode-mileage code alignment and pipe segment mapping, and uses video detection and time-series classification models to identify defects such as cracks, leaks, deformations, misalignments, deposits, and tree root intrusion. It outputs defect type, level, occurrence mileage range, keyframe evidence and confidence level, and writes them into structured defect records according to a unified dictionary encoding, which are then associated with the corresponding pipe segment objects to support operation and maintenance and risk analysis.

9. The intelligent management system for underground pipeline detection, mapping, and geographic information according to claim 1, characterized in that, The process of outputting quality inspection reports and review work orders through a rule engine and AI anomaly detection is as follows: The quality layer receives structured data produced by the computation layer, including complete information such as pipeline geometry, attribute parameters, confidence assessment, uncertainty quantification, and evidence chain index. Complementary quality inspection is achieved through parallel processing by two engines: The rule engine performs deterministic verification based on a predefined constraint rule base, including quantifiable quality dimensions such as data integrity, coding standards, geometric topology, engineering standards, and evidence integrity; The AI ​​anomaly detection engine uses machine learning methods to identify implicit anomaly patterns and statistical deviations. The dual-engine outputs are weighted, fused, and thresholded to generate graded quality inspection conclusions (pass, require rectification, or reject). Based on the problem type, severity, and spatial distribution, a structured quality inspection report and an actionable review work order are automatically generated.

10. The intelligent management system for underground pipeline detection, mapping, and geographic information according to claim 1, characterized in that, The results layer receives graded quality inspection conclusions from the quality layer, indicating whether the items have passed, require rectification, or have been rejected. For items that have passed, the layer performs data storage and solidification: writing 2D / 3D geometry, attributes, accuracy level and confidence level, uncertainty parameters, evidence chain references, source and processing link identifiers into the spatial database; simultaneously establishing version and temporal records, generating version_id and validity period fields for each addition, change, or obsolescence, supporting historical backtracking, difference comparison, and rollback; the results layer solidifies object relationships, including the association between pipe sections and appurtenances / well chambers, pipeline topology relationships and segmentation rules, and establishes stable associations between structured defect records and pipe sections to support the operation and maintenance closed loop.

Images