Intelligent surveying equipment and method for traditional wading heritage facilities

Abstract

The present disclosure provides a traditional wading heritage facility intelligent surveying device and method, relating to the technical field of heritage facility surveying and protection. The method comprises: deploying multi-source sensing devices in the target facility area, collecting data including spatial geometric structure, surface image and wall humidity, and performing fusion processing to form a unified fusion data set; inputting the fusion data set into a pre-trained hidden danger identification model to comprehensively analyze multi-dimensional features to identify cracks, leakage and other hidden dangers; based on the hidden danger distribution characteristics and the fusion data, constructing a facility reconstruction three-dimensional model, which contains geometric structure, seepage information and hidden danger spatial position; and determining the structure risk level and potential instability position in combination with the model. The technical scheme can improve the data integrity and defect identification accuracy, make the risk assessment more reliable and refined, and effectively improve the intelligent level and decision support capability of the wading heritage facility surveying.

Description

Technical Field

[0001] This disclosure relates to the field of heritage facility surveying and protection technology, and more specifically, to a smart surveying device and method for traditional water-related heritage facilities. Background Technology

[0002] Traditional water-related heritage facilities, as an important component of historical hydraulic architecture and regional culture, are diverse in form and structure, ancient in age, and buried in complex environments. They are often located underground or in semi-enclosed spaces, such as ancient wells, karez wells, ancient canals, dams, and underground irrigation ditches. These facilities have long been subjected to the combined effects of water erosion, soil seepage, climate change, and geological movements, making them highly susceptible to structural damage such as cracks, leakage, and settlement, which in turn affects their overall stability and the safety of cultural relic preservation. Due to their unique spatial form and historical sensitivity, the surveying and testing of these facilities has always been a technical challenge in the field of cultural relic protection and structural diagnosis.

[0003] Currently, the surveying of water-related heritage facilities largely relies on manual on-site inspections or single technical methods, such as collecting data solely through geometric measurements or surface observations. This approach often fails to comprehensively capture the overall condition of the facilities, resulting in limited data coverage and a lack of effective correlation between data from different sources, thus affecting the accuracy of risk assessments. Furthermore, current surveying processes typically focus on local feature analysis, failing to efficiently integrate multiple physical parameters, leading to biases in the identification and location of structural defects. Moreover, the construction of 3D models often ignores the dynamic changes of key risk factors. With the increasing prominence of facility aging issues, existing technologies are no longer sufficient to meet the demands for refined and intelligent surveying of water-related heritage sites. A more comprehensive solution is urgently needed to improve the reliability and efficiency of surveying water-related heritage facilities.

[0004] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this disclosure is to provide an intelligent surveying device and method for traditional water-related heritage facilities based on multi-source heterogeneous data, thereby improving the accuracy of structural defect identification and spatial positioning, and realizing refined assessment and intelligent diagnosis of the structural status of water-related heritage facilities.

[0006] Other features and advantages of this disclosure will become apparent from the following detailed description, or may be learned in part from practice of this disclosure.

[0007] According to a first aspect of the present disclosure, a method for intelligent surveying of traditional water-related heritage facilities based on multi-source heterogeneous data is provided, comprising:

[0008] Multi-source sensing devices are deployed within the survey area of ​​the target water-related heritage facility, and multi-source heterogeneous datasets of the target water-related heritage facility are collected through the multi-source sensing devices. The multi-source heterogeneous datasets include at least spatial geometric structure data, surface image data, and wall humidity data.

[0009] The multi-source heterogeneous datasets are fused to form a fused dataset;

[0010] The fused dataset is input into a pre-trained hazard identification model to identify risks and hazards and determine the hazard distribution characteristics.

[0011] Based on the hazard distribution feature data and the fused dataset, a three-dimensional model for facility reconstruction is constructed. The three-dimensional model for facility reconstruction includes the geometric structure, seepage data, and spatial distribution location of the target water-related heritage facility.

[0012] The structural risk level and potential instability location of the target water heritage facility were determined by combining the reconstructed 3D model of the facility.

[0013] In some example embodiments of this disclosure, based on the foregoing scheme, the method further includes:

[0014] Obtain external environmental parameters for extreme weather conditions, including at least rainfall intensity parameters, surface runoff parameters, and groundwater level change parameters;

[0015] Based on the external environmental parameters and the reconstructed three-dimensional model of the facility, dynamic seepage simulation is performed to predict the internal water accumulation distribution and structural instability trend of the target water-related heritage facility under extreme weather conditions.

[0016] Emergency protection task data is generated based on the internal water accumulation distribution and the structural instability trend.

[0017] In some example embodiments of this disclosure, based on the foregoing scheme, the hazard identification model includes at least a convolutional neural network and a graph neural network; the step of inputting the fused dataset into the pre-trained hazard identification model for risk and hazard identification, and determining hazard distribution characteristic data, includes:

[0018] The convolutional neural network extracts crack and erosion features from the surface image data, and the graph neural network analyzes the spatial relationship between humidity and seepage nodes in the wall humidity data to identify potential leakage channels.

[0019] In the hazard identification model, the crack and erosion features and the potential leakage channel features are subjected to multimodal feature fusion processing to generate preliminary hazard identification results;

[0020] The distribution characteristics of the hazards are determined based on the preliminary hazard identification results.

[0021] In some example embodiments of this disclosure, based on the foregoing scheme, determining the hazard distribution characteristic data according to the preliminary hazard identification results includes:

[0022] Spatial clustering operations are performed on the preliminary hazard identification results to form a set of hazard areas;

[0023] Based on the crack length, seepage intensity, and settlement depth extracted from the set of potential hazard areas, the potential hazard distribution feature data corresponding to the fused dataset is generated.

[0024] In some example embodiments of this disclosure, based on the foregoing scheme, the step of constructing a 3D reconstruction model of the facility based on the hazard distribution characteristic data and the fused dataset includes:

[0025] The fused dataset is divided into voxels and then subjected to 3D point cloud reconstruction to generate a geometric mesh model of the target water-related heritage facility.

[0026] The seepage data of the target water-related heritage facility is determined based on the spatial geometry data, the surface image data, and the wall humidity data.

[0027] The seepage data and the hazard distribution characteristic data are superimposed on the geometric mesh model to construct a three-dimensional reconstruction model of the facility with seepage distribution attributes.

[0028] In some example embodiments of this disclosure, based on the foregoing scheme and combined with the three-dimensional model of the facility reconstruction, the structural risk level and potential instability location of the target water-related heritage facility are determined, including:

[0029] The reconstructed 3D model of the facility is input into a pre-trained physical constraint neural network to determine the stress and strain field distributions of the reconstructed 3D model of the facility through the physical constraint neural network.

[0030] Based on the stress field and strain field distribution, the stress concentration and humidity gradient change data of the target water-related heritage facility are determined;

[0031] By combining the stress concentration and humidity gradient change data at each location, the risk level of each location of the target water-related heritage facility is classified, and the structural risk level corresponding to each location is determined.

[0032] The location region where the structural risk level is greater than or equal to a preset level threshold is identified as a potential instability location.

[0033] In some exemplary embodiments of this disclosure, based on the foregoing scheme, the step of performing dynamic seepage simulation based on the external environmental parameters and the reconstructed three-dimensional model of the facility to predict the internal water accumulation distribution and structural instability trend of the target water-related heritage facility under extreme weather conditions includes:

[0034] The geometric structure corresponding to the reconstructed 3D model of the facility is discretized by meshing, and the seepage data and wall humidity data are mapped to generate a seepage solution mesh corresponding to the geometric structure of the reconstructed 3D model of the facility.

[0035] Rainfall intensity parameters, surface runoff parameters, and groundwater level change parameters are set on the seepage solution grid to form dynamic boundary conditions, and a dynamic seepage numerical model is established based on the finite element method or the finite volume method.

[0036] Based on the dynamic seepage numerical model, the water volume distribution and wall permeability stress data at each time point are calculated in time series, so as to determine the internal water distribution and structural instability trend of the target water-related heritage facility under extreme weather conditions.

[0037] In some example embodiments of this disclosure, based on the foregoing scheme, corresponding emergency protection task data is generated based on the internal water accumulation distribution and the structural instability trend, including:

[0038] Based on the internal water accumulation distribution and the structural instability trend, identify the key protection areas corresponding to the target water-related heritage facilities;

[0039] Based on the regional parameters of the key protection area, the corresponding emergency protection measures are selected from the preset emergency protection task database. The emergency protection measures include at least one or more combinations of drainage measures, reinforcement measures, and sealing measures.

[0040] Based on the emergency protection measures and the key protection areas, emergency protection task data including protection location, priority, and execution order is generated.

[0041] In some example embodiments of this disclosure, based on the foregoing scheme, the multi-source sensing device includes at least a lidar, a close-range image acquisition unit, and a soil moisture sensor; the acquisition of the multi-source heterogeneous dataset of the target water-related heritage facility through the multi-source sensing device includes:

[0042] The spatial geometric structure data of the target water-related heritage facility is collected, wherein the spatial geometric structure data is composed of spatial point cloud information output by the laser scanning module;

[0043] Surface image data of the target water-related heritage facility is collected, wherein the surface image data consists of texture image information output by the close-up photography module;

[0044] Collect wall humidity data of the target water-related heritage facility, wherein the wall humidity data consists of humidity and seepage pressure information output by the seepage humidity sensing module;

[0045] The attitude sensing unit of the multi-source sensing device performs synchronous time stamping and attitude calibration on the spatial geometric structure data, the surface image data, and the wall humidity data to construct the multi-source heterogeneous dataset.

[0046] In some example embodiments of this disclosure, based on the foregoing scheme, the multi-source heterogeneous dataset is fused to form a fused dataset, including:

[0047] A unified spatial coordinate framework is established based on the aforementioned spatial geometric structure data;

[0048] Based on the unified spatial coordinate framework, the surface image data is spatially registered with the spatial geometric structure data through a feature point matching algorithm, and the wall humidity data is spatially interpolated and mapped to associate the humidity value and seepage pressure value with the corresponding geometric node in the spatial coordinate framework to form a fused dataset.

[0049] According to a second aspect of the present disclosure, an intelligent surveying device for traditional water-related heritage facilities based on multi-source heterogeneous data is provided, comprising a data acquisition platform, a control processing unit, and a communication unit, wherein:

[0050] The data acquisition platform integrates a lidar, a close-range camera, and a humidity sensor, and is equipped with an attitude sensing component to provide attitude and time synchronization information.

[0051] The communication unit is used for data interaction and command issuance with external terminals;

[0052] The control processing unit is configured to:

[0053] The data acquisition platform is controlled to collect multi-source heterogeneous datasets within the survey area of ​​the target water-related heritage facility. The multi-source heterogeneous datasets include at least spatial geometric structure data, surface image data, and wall humidity data.

[0054] The multi-source heterogeneous datasets are fused to obtain a fused dataset;

[0055] The fused dataset is input into a pre-trained hazard identification model to identify risks and hazards, and to determine the hazard distribution characteristics data.

[0056] Based on the hazard distribution feature data and the fused dataset, a 3D model of facility reconstruction is constructed. The 3D model of facility reconstruction includes the geometric structure, seepage data, and spatial distribution location of the target water-related heritage facility.

[0057] Based on the reconstructed 3D model of the facility, the structural risk level and potential instability location of the target water-related heritage facility were determined;

[0058] The fused dataset, the hazard distribution characteristic data, the facility reconstruction 3D model, and the structural risk level and potential instability location are transmitted and / or stored through the communication unit.

[0059] According to a third aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory storing computer-readable instructions, which, when executed by the processor, implement the conventional intelligent surveying method for water-related heritage facilities based on multi-source heterogeneous data as described above.

[0060] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the conventional intelligent surveying method for water-related heritage facilities based on multi-source heterogeneous data as described in any of the preceding claims.

[0061] The technical solutions provided in this disclosure may have the following beneficial effects:

[0062] The intelligent survey method for traditional water-related heritage facilities based on multi-source heterogeneous data in the example embodiments of this disclosure has two main advantages. First, by deploying multi-source sensing devices to collect heterogeneous data including spatial geometry, surface images, and wall humidity, and fusing this data to form a unified fused dataset, the survey data coverage of water-related heritage facilities can be more comprehensive. Furthermore, effective correlations are established between data from different sources, overcoming the information gaps caused by single or isolated data in traditional methods, and providing a more complete and consistent data foundation for subsequent analysis. Second, by inputting the fused dataset into a pre-trained hazard identification model for risk and hazard identification, multiple physical parameters can be efficiently integrated to achieve accurate identification of structural damage such as cracks and leaks. The identification and localization of defects avoids identification biases caused by local feature analysis or parameter dispersion, thereby improving the accuracy of defect detection. Furthermore, the 3D model constructed based on the distribution characteristics of hidden dangers and the fusion dataset not only covers geometric structural information but also incorporates seepage data and the spatial distribution of hidden dangers. This allows the model to dynamically reflect changes in key risk factors, avoiding the limitations of traditional 3D modeling that ignores the dynamic nature of risks, thus enhancing the model's ability to represent the actual state of the facility. Finally, by combining this 3D model to determine the structural risk level and potential instability location, a more reliable and refined risk assessment can be provided for water-related heritage facilities. This comprehensively addresses the problems of low reliability and insufficient efficiency in related technologies, significantly improving the intelligence level and decision support capabilities of the survey process.

[0063] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0064] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0065] Figure 1 The illustration schematically shows a flowchart of a conventional intelligent survey method for water-related heritage facilities based on multi-source heterogeneous data, according to some embodiments of the present disclosure.

[0066] Figure 2 The illustration shows a schematic diagram of a process for generating emergency protection mission data according to some embodiments of the present disclosure.

[0067] Figure 3 The illustration shows a flowchart of determining the structural risk level and potential instability location according to some embodiments of the present disclosure.

[0068] Figure 4 The illustration shows a schematic diagram of the composition of a conventional intelligent surveying device for water-related heritage facilities based on multi-source heterogeneous data, according to some embodiments of the present disclosure.

[0069] Figure 5 The schematic diagram illustrates the structural schematic of a computer system of an electronic device according to some embodiments of the present disclosure.

[0070] Figure 6 A schematic diagram of a computer-readable storage medium according to some embodiments of the present disclosure is shown.

[0071] In the accompanying drawings, the same or corresponding reference numerals indicate the same or corresponding parts. Detailed Implementation

[0072] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this specification. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this specification as detailed in the appended claims.

[0073] It should be understood that although the terms first, second, third, etc., may be used in this specification to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this specification, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0074] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0075] Furthermore, the accompanying drawings are for illustrative purposes only and are not necessarily drawn to scale. The block diagrams shown in the drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0076] In this example embodiment, a smart survey method for traditional water heritage facilities based on multi-source heterogeneous data is first provided. This smart survey method for traditional water heritage facilities based on multi-source heterogeneous data can be applied to terminal devices or servers. This embodiment does not make any special limitations on this, and the following description will take the server executing the method as an example. Figure 1 The illustration schematically depicts a flowchart of a conventional intelligent survey method for water-related heritage facilities based on multi-source heterogeneous data, according to some embodiments of this disclosure. Reference Figure 1 As shown, this intelligent surveying method for traditional water-related heritage facilities based on multi-source heterogeneous data may include the following steps:

[0077] Step S110: Deploy multi-source sensing devices within the survey area of ​​the target water-related heritage facility, and collect multi-source heterogeneous datasets of the target water-related heritage facility through the multi-source sensing devices. The multi-source heterogeneous datasets include at least spatial geometric structure data, surface image data, and wall humidity data.

[0078] Step S120: The multi-source heterogeneous dataset is fused to form a fused dataset;

[0079] Step S130: Input the fused dataset into the pre-trained hazard identification model to identify risks and hazards, and determine the hazard distribution characteristic data;

[0080] Step S140: Based on the hazard distribution feature data and the fused dataset, construct a three-dimensional model for facility reconstruction. The three-dimensional model for facility reconstruction includes the geometric structure, seepage data, and spatial distribution location of the target water-related heritage facility.

[0081] Step S150: Determine the structural risk level and potential instability location of the target water heritage facility by combining the reconstructed three-dimensional model of the facility.

[0082] According to the intelligent survey method for traditional water-related heritage facilities based on multi-source heterogeneous data in this example embodiment, on the one hand, by deploying multi-source sensing devices to collect multi-source heterogeneous data including spatial geometry, surface images, and wall humidity, and fusing these data to form a unified fused dataset, the survey data coverage of water-related heritage facilities can be more comprehensive, and effective correlations can be established between data from different sources. This overcomes the information loss problem caused by single or isolated data in traditional methods, providing a more complete and consistent data foundation for subsequent analysis. On the other hand, by inputting the fused dataset into a pre-trained hazard identification model for risk and hazard identification, multiple physical parameters can be efficiently integrated to achieve accurate identification of structural damage such as cracks and leaks. The identification and localization of defects avoids identification biases caused by local feature analysis or parameter dispersion, thereby improving the accuracy of defect detection. Furthermore, the 3D model constructed based on the distribution characteristics of hidden dangers and the fusion dataset not only covers geometric structural information but also incorporates seepage data and the spatial distribution of hidden dangers. This allows the model to dynamically reflect changes in key risk factors, avoiding the limitations of traditional 3D modeling that ignores the dynamic nature of risks, thus enhancing the model's ability to represent the actual state of the facility. Finally, by combining this 3D model to determine the structural risk level and potential instability location, a more reliable and refined risk assessment can be provided for water-related heritage facilities. This comprehensively addresses the problems of low reliability and insufficient efficiency in related technologies, significantly improving the intelligence level and decision support capabilities of the survey process.

[0083] The following will further explain the intelligent survey method for traditional water-related heritage facilities based on multi-source heterogeneous data in this example embodiment.

[0084] In step S110, multi-source sensing devices are deployed within the survey area of ​​the target water heritage facility, and multi-source heterogeneous datasets of the target water heritage facility are collected through the multi-source sensing devices. The multi-source heterogeneous datasets include at least spatial geometric structure data, surface image data, and wall humidity data.

[0085] In one example embodiment of this disclosure, a multi-source sensing device refers to a multi-sensor combination system used to simultaneously detect different physical properties of a target facility within the same time period. For example, the multi-source sensing device may include a laser scanning module, a close-range image acquisition module, a humidity and seepage measurement module, and an attitude detection module mounted on the same support platform. The laser scanning module may employ a rotating multi-line lidar (Light Detection and Ranging, LiDAR), which works by emitting laser pulses and receiving reflected signals to determine the distance to spatial target points, thereby generating spatial point cloud data of the facility surface. This point cloud data constitutes spatial geometric structure data. The module achieves 360° rotation scanning via motor drive, and works with an inertial measurement unit (IMU) to provide attitude parameter compensation, thereby improving the spatial continuity and geometric accuracy of the point cloud data.

[0086] The close-range image acquisition module is used to acquire high-resolution image data of the surface of the target water-related heritage facility. This surface image data is acquired using an industrial camera, lens assembly, and adaptive illumination compensation device, maintaining image clarity and contrast even in low-light or highly reflective environments. The image data includes color, texture, and surface morphology information, which can be used to reflect surface cracks, erosion marks, and weathering characteristics. This module shares a time synchronization signal with the laser scanning module to ensure consistency between image frames and point cloud frames at the time of acquisition.

[0087] The humidity and seepage measurement module is used to collect humidity and seepage pressure data of the walls of the target water-related heritage facility, i.e., the wall humidity data. This module includes a capacitive humidity sensor and a miniature pressure sensor array to sense the internal water content and pore water pressure of the material. The humidity sensor obtains the humidity signal by detecting changes in the dielectric constant, while the pressure sensor measures local seepage stress based on the piezoresistive effect. The sensing module can employ a structure combining wired and wireless node transmission methods to adapt to data transmission requirements under different depths and material conditions.

[0088] The attitude detection module includes an accelerometer and a gyroscope, used to measure the tilt angle, rotation, and displacement of the multi-source sensing devices in space, thereby providing attitude correction parameters for various data in subsequent data fusion processing. By calibrating the attitude of the laser scanning module and the image acquisition module through this module, the acquired spatial geometric structure data, surface image data, and wall humidity data can be kept consistent in time and space, avoiding spatial misalignment caused by device deflection or vibration.

[0089] In practical applications, multi-source sensing devices can be mounted on ground tripod structures for surveying open channels, dams, or larger spaces. They can also be mounted on tracked mobile platforms for automated mobile measurements in confined spaces such as karez wells and underground canals. For locations lacking manual operation capabilities, drones or extendable robotic arms can be used for targeted data collection in localized areas. By simultaneously collecting these three types of data within the same spatial coordinate system, a spatiotemporal correspondence can be established between the spatial structure, surface features, and internal humidity status of the target water-related heritage facility at the data level. This ensures the integrity and comparability of the physical attribute data, providing a unified reference basis for subsequent data fusion and risk analysis.

[0090] In step S120, the multi-source heterogeneous dataset is fused to form a fused dataset.

[0091] In one example embodiment of this disclosure, the fusion processing refers to the computational process of spatial registration, temporal alignment, and interpolation mapping of spatial geometric structure data, surface image data, and wall humidity data within a unified coordinate framework, enabling point-to-point correspondence and information overlay of different types of data sources within the same reference system. Specifically, firstly, using the spatial point cloud output by the laser scanning module as the global coordinate reference, edge features and feature planes of the point cloud are extracted using a feature point recognition algorithm; then, using the feature texture points in the surface image data as a reference, feature matching and geometric registration are performed using a Scale-Invariant Feature Transform (SIFT) algorithm or a Speeded Up Robust Features (SURF) algorithm, ensuring a one-to-one correspondence between image pixels and point cloud coordinates. After registration, humidity and seepage pressure data are mapped to the corresponding point cloud node positions using a spatial interpolation algorithm (such as Kriging interpolation or inverse distance weighted interpolation), thus ensuring that each spatial node simultaneously contains multi-dimensional attribute parameters such as geometric position, image texture, and humidity value.

[0092] In this process, to eliminate the impact of differences in sampling frequencies and time delays between different sensors, a timestamp synchronization mechanism can be used to perform linear interpolation or time window averaging of the data acquisition time. For scenes with large variations in ambient lighting, brightness normalization and gamma correction can be used to preprocess the image data to improve the stability of texture feature matching. For noise interference in the humidity sensor signal, median filtering or Kalman filtering algorithms can be used for smoothing to improve the continuity of humidity distribution data. Through the above fusion processing, multidimensional coupling of spatial geometry, imagery, and humidity information can be achieved in a single data structure, thereby maintaining topological consistency and physical correlation between data in subsequent calculations.

[0093] In step S130, the fused dataset is input into a pre-trained hazard identification model to identify risks and hazards and determine the hazard distribution characteristics data.

[0094] In one example embodiment of this disclosure, the hazard identification model is a multimodal data analysis model built based on deep learning technology. Its input receives geometric features, image features, and humidity features provided by a fused dataset, and its output generates spatial distribution results for various structural hazards. The model employs a joint structure of a Convolutional Neural Network (CNN) and a Graph Neural Network (GNN). The former is used to extract local texture and morphological features, while the latter is used to analyze the topological relationships and correlation patterns between humidity distribution nodes. During model operation, the CNN automatically learns the local texture gradient features of the image through kernel sliding operations to identify surface anomalies such as cracks and erosion; the GNN calculates the spatial correlation of humidity distribution based on node features and edge weight matrices, thereby identifying seepage paths and areas of concentrated humidity anomalies. The two types of features are concatenated and weighted through a multimodal fusion layer to generate a hazard discrimination vector. During the training phase, the model learns the feature distribution patterns of different hazard types using a large amount of sample data. During the inference phase, it can achieve rapid mapping between input data and hazard types, thereby generating preliminary hazard identification results.

[0095] When determining the distribution characteristics of potential hazards, the model can further perform spatial clustering and feature summarization of hazard areas. By extracting indicators such as crack length, seepage intensity, and humidity gradient changes through cluster centers and boundary conditions, a parameter set characterizing each hazard type can be formed. For heritage facilities with different materials or environmental conditions, adaptive expansion can be achieved through model transfer learning or parameter fine-tuning, thus maintaining identification accuracy without changing the network structure. The application of the hazard identification model can perform hazard identification tasks under the multidimensional constraints of fused data, ensuring spatial continuity and parameter consistency in the identification results, which helps to accurately locate and classify structural risk areas.

[0096] In step S140, a three-dimensional model of facility reconstruction is constructed based on the hazard distribution feature data and the fused dataset. The three-dimensional model of facility reconstruction includes the geometric structure, seepage data, and spatial distribution location of the target water heritage facility.

[0097] In one example embodiment of this disclosure, the construction process of the 3D model for facility reconstruction involves reconstructing and visualizing the fused dataset and hazard characteristic data within the same 3D coordinate system. Specifically, firstly, the point cloud is voxelized using the spatial geometric structure data in the fused dataset. A dense point cloud is generated using a voxel fusion algorithm and converted into a polygonal mesh model, which characterizes the geometric shape and spatial continuity of the facility. Secondly, the humidity and seepage parameters extracted from the fused dataset are mapped to the mesh nodes, forming a seepage distribution field corresponding to the geometric structure. Thirdly, the spatial coordinates corresponding to the hazard distribution characteristic data are embedded into the mesh structure, storing the hazard location and characteristic parameters in the model as a marker layer or attribute layer. Through the above process, the 3D model not only contains geometric morphology information but also a multi-layered data structure reflecting the material humidity state and hazard characteristics, enabling multi-parameter joint queries and regional risk analysis.

[0098] In terms of model implementation, an octree-based spatial indexing structure can be used to improve the efficiency of point cloud reconstruction and attribute mapping; alternatively, a volumetric grid-based rendering method can be employed to support high-resolution seepage visualization. For facilities requiring long-term monitoring, the model can be dynamically refreshed by updating hazard characteristic data, enabling the 3D model to reflect the risk evolution trend over time. The establishment of this model allows subsequent risk calculations to be based on multidimensional data with real spatial correspondences, thereby improving the accuracy of calculations and the consistency of spatial analysis.

[0099] In step S150, the structural risk level and potential instability location of the target water heritage facility are determined by combining the reconstructed three-dimensional model of the facility.

[0100] In one example embodiment of this disclosure, the structural risk level refers to a grading index obtained by comprehensively and quantitatively assessing the stability of different areas or structural units within a target water-related heritage facility under stress, seepage, and humidity. It reflects the facility's safety margin and degree of structural deterioration under specific environmental conditions. Essentially, it involves multi-dimensional statistical analysis of geometric parameters, humidity gradient change data, and stress concentration results in the reconstructed 3D model of the facility to calculate the risk score corresponding to each structural node or grid unit. This score is then divided into multiple levels based on a preset threshold range, such as low-risk, medium-risk, and high-risk levels. This grading system can achieve numerical mapping based on fuzzy comprehensive evaluation, analytic hierarchy process (AHP), or machine learning regression models. The structural risk level not only quantitatively describes the differences in safety status between different areas but also provides a basis for prioritizing subsequent reinforcement, drainage, and sealing measures.

[0101] Potential instability locations refer to spatial areas within the overall structure of a facility reconstruction 3D model where risk assessment calculations indicate stress concentration exceeding safety limits or humidity gradient change rates exceeding critical thresholds. These areas are considered warning locations where structural slippage, deformation, or collapse may occur. The determination of these locations relies on a Physically Informed Neural Network (PINN) to calculate and analyze the stress and strain fields of the model nodes. The results are then compared with threshold conditions in the risk level assessment system to select a set of spatial nodes that meet the instability criteria. Potential instability locations are typically represented by spatial coordinates and their corresponding risk levels, and can be visualized in the 3D model to guide on-site structural reinforcement or monitoring point deployment.

[0102] By simultaneously calculating structural risk levels and potential instability locations in the 3D model of facility reconstruction, a two-tiered judgment relationship can be established at the spatial level, where the risk level distribution constrains the location of instability areas. This transforms the safety assessment of water-related heritage facilities from a macro-level classification process to a refined spatial identification process. Structural risk levels provide a global risk distribution framework, while potential instability locations offer specific coordinates of local anomalies. The combination of these two elements ensures that facility condition diagnosis is continuous, quantitative, and operational. This definition system ensures that the risk assessment results not only have numerical basis but also spatial mappability, providing crucial input for subsequent dynamic seepage prediction and emergency response task generation.

[0103] The following provides a detailed explanation of the traditional intelligent survey method for water-related heritage facilities based on multi-source heterogeneous data in steps S110 to S150.

[0104] In one example embodiment of this disclosure, it can be achieved through Figure 2 The steps in the document are used to generate emergency protection task data. (Refer to...) Figure 2 As shown, it can specifically include:

[0105] Step S210: Obtain external environmental parameters for extreme weather conditions, including at least rainfall intensity parameters, surface runoff parameters, and groundwater level change parameters;

[0106] Step S220: Based on the external environmental parameters and the reconstructed three-dimensional model of the facility, perform dynamic seepage simulation to predict the internal water accumulation distribution and structural instability trend of the target water-related heritage facility under extreme weather conditions;

[0107] Step S230: Generate corresponding emergency protection task data based on the internal water accumulation distribution and the structural instability trend.

[0108] Among them, the rainfall intensity parameter refers to the meteorological input used to describe the variation of precipitation per unit time. It reflects the instantaneous intensity and duration of the impact of rainfall events on the external load of water-related heritage facilities. The rainfall intensity parameter can be obtained in real time by rain gauges or meteorological monitoring stations set up in the catchment area of ​​heritage facilities. For example, tipping bucket rain gauges or laser raindrop spectrometers can be used, or continuous time series data can be obtained by meteorological radar inversion and extrapolation of short-term precipitation models.

[0109] Surface runoff parameters refer to the volume fraction or velocity distribution of flowing water formed on the surface of a facility and its upstream catchment area under rainfall or snowmelt conditions. They are used to reflect the infiltration pressure and fluid shear effect of external surface runoff on the outer wall or foundation of the structure. Surface runoff parameters can be estimated by deploying open channel flow meters, ultrasonic flow meters, or watershed runoff algorithms based on digital elevation models and land cover parameters.

[0110] Groundwater level variation parameters refer to the dynamic information of the water level of the groundwater layer in the facility foundation area changing over time. This information can be measured using pressure level gauges, vibrating wire piezometers, or capacitive level gauges, or obtained through a regional groundwater dynamic monitoring network. When acquiring external environmental parameters, rainfall intensity, runoff, and groundwater level data must be sampled and timestamped according to a unified time reference. Abnormal impulses should be eliminated using Kalman filtering or Empirical Mode Decomposition (EMD) to ensure the continuity and comparability of the data in the time domain.

[0111] In different survey scenarios, external environmental parameters can also be extracted through regional meteorological database interfaces or Geographic Information Systems (GIS). This embodiment does not limit the specific method of obtaining external environmental parameters. Through this series of synchronous acquisition and correction processes, a consistent spatiotemporal constraint relationship can be established between the external load characteristics under extreme weather conditions and the boundary response of the facility's three-dimensional model, providing accurate input conditions for subsequent dynamic seepage simulation.

[0112] Dynamic seepage simulation refers to the process of calculating the seepage path, pressure field, and time-varying distribution of water content in a porous media structure under the influence of a three-dimensional geometric model of the facility and external boundary conditions. The seepage solution mesh generated after geometric discretization of the reconstructed three-dimensional model serves as the basis of the computational domain. Seepage data and wall moisture data are mapped to the mesh nodes to form an initial hydraulic parameter field, including permeability coefficient, porosity, specific storage capacity, and residual water content. Time-dependent boundary conditions can be set based on rainfall intensity parameters, surface runoff parameters, and groundwater level variation parameters under extreme weather conditions. A variable flux or variable head boundary is applied at the top, an outflow boundary or impermeable boundary is set at the bottom, and the lateral boundary is set with a hydraulic gradient based on the terrain slope. The seepage governing equations can be variations of the Richards Equation or Darcy's Law, solved discretizedly using the Finite Element Method (FEM) or the Finite Volume Method (FVM). To improve convergence stability, a dynamic damping factor can be introduced in the unsaturated section and a Newton-Raphson iterative solution can be used. During extreme rainfall, to maintain numerical stability, an adaptive control algorithm combined with an implicit integral scheme can be employed for the time step. The calculation results can include the head field, pore water pressure field, and seepage velocity field at each time point. The accumulated water volume can be obtained through element volume integration, and the wall permeability stress is calculated from the combination of Darcy velocity and viscosity coefficient. In an optional implementation, the dynamic seepage simulation can be weakly coupled with the temperature or stress field to reflect the nonlinear response of material properties to changes in water content.

[0113] The critical protection area refers to the spatial region where the water volume exceeds a set threshold or the wall seepage stress exceeds a safety limit in dynamic seepage simulation, reflecting the spatial region where potential risks to the facility are concentrated under extreme weather conditions. By jointly determining the water volume and seepage stress of each calculation unit in the 3D model, a set of mesh units that meet the conditions is extracted to form a critical protection area dataset. Area parameters may include spatial coordinate range, material type, accessibility level, and historical hazard labels, but this embodiment is not limited to these.

[0114] The emergency protection task database is a pre-established set of task templates, recording the execution strategies, resource requirements, and applicable conditions corresponding to different protection types. Based on the parameters of key protection areas, corresponding emergency protection measures are matched in the database. For example, emergency protection measures may include drainage measures, reinforcement measures, and sealing measures, such as setting up emergency drainage pumps, adding temporary supports, or using anti-seepage coatings to seal leaking areas. When multiple key protection areas coexist, task priorities can be calculated based on a regional weight function, which can be formed by combining indicators such as water accumulation growth rate, seepage stress growth rate, risk level, and personnel access constraints. The three-dimensional coordinates of the protection location, task type, priority, and execution order can be output in a structured format for use by the on-site protection system. In an optional implementation, task ranking can be achieved in real time using an algorithm based on a Learning to Rank (LTR) model. Through the automated transformation from physical simulation to emergency task generation, protection decisions are based on quantitative prediction, enabling pre-planning of key protection actions before extreme weather events occur, improving the emergency response efficiency and protection reliability of water-related heritage facilities.

[0115] In one example embodiment of this disclosure, the hazard identification model may include at least a convolutional neural network and a graph neural network. The steps involved in inputting a fused dataset into the pre-trained hazard identification model for risk identification include:

[0116] Crack and erosion features in surface image data can be extracted using convolutional neural networks, and the spatial relationship between humidity and seepage nodes in wall humidity data can be analyzed using graph neural networks to identify potential leakage channels. In the hazard identification model, multimodal feature fusion processing is performed on crack and erosion features and potential leakage channel features to generate preliminary hazard identification results. Hazard distribution feature data are determined based on the preliminary hazard identification results.

[0117] Convolutional Neural Networks (CNNs) are deep learning models based on the principles of local receptive fields and weight sharing. They extract image features step-by-step through convolutional layers, pooling layers, and fully connected layers. For example, the input surface image data is first preprocessed through standardization and normalization to eliminate the influence of illumination differences and color deviations on feature recognition. Then, in the first convolutional layer, 3×3 or 5×5 convolutional kernels are used to extract local texture gradients, forming a primary feature map. In the intermediate convolutional layers, a multi-scale convolutional kernel structure is employed to capture the morphological features of cracks of different sizes, while pooling operations are combined to reduce spatial dimensions and enhance the model's translation invariance. In higher convolutional layers, nonlinear mapping is performed using activation functions (e.g., modified ReLU) to extract deep features of crack edge contours, erosion patch boundaries, and surface discontinuities. The output feature map of the CNN can then be used by a Softmax layer or a Sigmoid classification layer to generate a crack region probability distribution map, thereby obtaining mask information for areas in the image that may contain structural damage. In some alternative implementations, the convolutional neural network can employ a residual network (ResNet) or a feature pyramid network (FPN) structure to adapt to variations in image resolution or crack morphology complexity. By extracting deep features from surface image data using convolutional neural networks, high feature recognition robustness can be maintained even under complex lighting, reflection, or surface contamination conditions, ensuring stable spatial continuity and boundary clarity in crack detection results of the image layer.

[0118] Graph Neural Networks (GNNs) are a form of neural network capable of performing feature propagation and relation modeling on graph-structured data. Their input consists of a node feature matrix and an adjacency matrix. The node feature matrix can contain the humidity value, seepage pressure, and material permeability coefficient of each measuring point in the wall humidity data, while the adjacency matrix describes the spatial adjacency relationships and distance weights between measuring points. Through the message passing mechanism of the GNN, feature information between nodes is continuously aggregated in multi-layer propagation, ensuring that each node's representation includes not only its own humidity characteristics but also the humidity gradient and flow trend information of its neighboring regions. In model implementation, Graph Convolutional Networks (GCNs) or Graph Attention Networks (GATs) structures can be used. Convolutional or attention-weighted operations are employed to aggregate and update the features of neighboring nodes, thereby identifying the connectivity features of areas with abnormal humidity and potential leakage channels. To improve computational efficiency, a sparse adjacency strategy can be introduced into the GNN, retaining only the edge weights of highly relevant nodes to reduce unnecessary computation. In some alternative implementations, node weight updates can employ gated recurrent units (GRUs) to achieve dynamic modeling of time-series humidity data, enabling the model to reflect seepage trends over time. Through the structured feature propagation mechanism of graph neural networks, the spatial correlation patterns of leakage channels, water accumulation paths, and abnormal seepage areas can be automatically learned from the spatial distribution of humidity, improving the ability to identify hidden seepage risks.

[0119] Multimodal feature fusion refers to the process of mapping, fusing, and weighting feature vectors from different types of sensor sources within a unified feature space. For example, the fusion layer can be implemented using feature concatenation or feature weighted fusion: feature concatenation directly concatenates image features and humidity features along the feature dimension, then compresses them to a unified dimension via a fully connected layer; feature weighted fusion assigns adaptive weights to different modal features through an attention mechanism to enhance the contribution of key modalities in the judgment. Optionally, the multimodal fusion module can introduce a self-attention mechanism or a multi-head attention mechanism to capture the deep dependency between image texture and humidity gradient changes. The fused features are input into a classification subnetwork after nonlinear transformation, and preliminary hazard identification results are generated through threshold judgment, including hazard type labels (such as cracks, leaks, settlement) and their spatial coordinates. Through this fusion process, the model can achieve cross-validation of structural anomalies in the image layer and seepage anomalies in the humidity layer, avoiding feature loss and misjudgment caused by a single modality.

[0120] Hazard distribution feature data refers to a multidimensional dataset representing the spatial location, type, and intensity parameters of hazards, which can be obtained through spatial clustering algorithms and regional statistical analysis. Spatial clustering algorithms can employ density-based spatial clustering of applications with noise (DBSCAN) or mean-shift methods to group adjacent hazard points into hazard regions. For each clustered region, characteristic parameters such as crack length, seepage intensity, and settlement depth are calculated, and the results are normalized to form a standardized hazard distribution feature matrix. In an optional implementation, the boundaries of hazard regions can be smoothed using region growing algorithms or morphological dilation-erosion operations to obtain continuous spatial boundaries. The hazard distribution feature data is ultimately bound to the three-dimensional coordinate system corresponding to the fused dataset, providing complete input data for subsequent construction of a three-dimensional model for facility reconstruction. Through a multi-stage processing flow of feature data extraction, fusion, and identification, the hazard identification model can achieve a unified transformation from multi-source physical information to spatialized risk features, ensuring high consistency and reliability of the identification results at both the texture and physical layers.

[0121] In one example embodiment of this disclosure, the determination of hazard distribution characteristic data based on preliminary hazard identification results can be achieved through the following steps, specifically including:

[0122] Spatial clustering operations can be performed on the preliminary hazard identification results to form a set of hazard areas; based on the crack length, seepage intensity and settlement depth in the extracted hazard area set, hazard distribution feature data corresponding to the fusion dataset can be generated.

[0123] Spatial clustering refers to grouping potential hazard points based on spatial proximity and feature similarity within a unified spatial coordinate framework of the fused dataset, thereby forming a continuous set of hazard regions. For example, the hazard point set in the preliminary hazard identification results first undergoes noise removal and coordinate standardization. Noise removal can employ the three-standard-deviation method or a Local Outlier Factor (LOF) algorithm to exclude isolated misidentified points. Clustering algorithms can use density-based spatial clustering of applications with noise (DBSCAN), which aggregates hazard points with continuous density into a single hazard region by setting a spatial neighborhood radius and a minimum sample size threshold. For linear crack features, an improved direction-weighted DBSCAN algorithm can be used to incorporate the directional gradient information of crack points into the similarity metric to maintain the continuity of the slender structure. For seepage or settlement feature regions, clustering methods based on mean drift or Gaussian Mixture Model (GMM) can be selected to adaptively capture hazard patches of different morphologies. After spatial clustering, morphological dilation, erosion, and boundary smoothing operations can be used to generate continuous boundary surfaces for the hazard areas, making the area contours more closely match the actual geometry of the facilities. After three-dimensional connectivity analysis, the clustering results yield a set of hazard area data, with each hazard area containing its spatial extent, hazard type, average confidence level, and geometric center coordinates. This spatial clustering operation transforms discrete hazard identification points into spatially interpretable risk units, ensuring the spatial distribution of hazard areas is continuous, closed, and computable.

[0124] Crack length refers to the length of the main crack framework or the total crack propagation length within the hazard area, reflecting the geometric scale of the damage to the continuity of the structural surface. Crack length calculation is based on the principles of framework extraction and curve fitting. First, a framework extraction algorithm is performed within the hazard area, simplifying the crack region to a centerline of a single pixel width. Then, the geodesic distance of the framework is calculated using a shortest path algorithm or spline curve fitting method. For multi-branch cracks, the length of each branch can be calculated separately and summed with weighted coefficients to obtain the total crack length of the region. Crack direction information can be extracted through Principal Component Analysis (PCA) and stored as an auxiliary parameter.

[0125] Seepage intensity refers to the magnitude of water flow formed under the combined effects of the wall humidity gradient and local permeability within a potential hazard area, used to characterize the degree of seepage activity within the material. Seepage intensity is calculated based on wall humidity data and seepage parameter fields from a fused dataset, using Darcy's law to calculate the seepage flow. For unsaturated areas, empirical formulas can be used to convert humidity values ​​into effective permeability coefficients. The calculated seepage intensity is then used to determine the mean and maximum values ​​within the region, and its rate of change over time is recorded.

[0126] Settlement depth refers to the vertical displacement within a hazardous area relative to the surrounding stable area, reflecting the deformation trend of the structural foundation or wall. Settlement depth is obtained by subtracting the spatial geometric data of the hazardous area from a reference surface, which can be a least-squares plane fit or an average surface based on historical point clouds. Calculation methods include Model-to-Model Distance (M3C2) or point-by-point vertical projection difference. To improve accuracy, the calculated results of crack length, seepage intensity, and settlement depth can undergo robust statistical processing, such as median and quantile filtering to suppress the influence of outliers. Finally, the hazardous area distribution characteristic data is indexed by spatial nodes in the fused dataset, binding and storing parameters such as crack length, seepage intensity, and settlement depth for each hazardous area, forming a comprehensive data structure that includes spatial location, hazardous type, and physical characteristics. This process enables a complete transformation of hazardous areas from identification and clustering to parameterized representation, giving hazardous information geometric, physical, and spatial attributes, thus improving the accuracy of hazardous feature description and data correlation.

[0127] In an example embodiment of this disclosure, a 3D reconstruction model of a facility can be constructed based on hazard distribution characteristic data and a fused dataset through the following steps, specifically including:

[0128] The fused dataset can be divided into voxels and then subjected to 3D point cloud reconstruction to generate a geometric mesh model of the target water-related heritage facility. The seepage data of the target water-related heritage facility can be determined based on spatial geometric structure data, surface image data, and wall humidity data. The seepage data and hazard distribution characteristic data can be superimposed on the geometric mesh model to construct a 3D reconstruction model of the facility with seepage distribution attributes.

[0129] Voxel partitioning refers to the process of dividing continuous point cloud data into regular volumetric units in 3D space to achieve data sparsity, noise filtering, and spatial structure discretization. A voxel is a spatial volume element with a cube or cuboid as its basic unit, and each voxel represents a fixed volume region in the geometric space of a facility. By setting the voxel resolution, a trade-off can be made between reconstruction accuracy and computational efficiency; for example, a voxel side length of 0.01m can be used for high-precision wall reconstruction, while 0.05m can be used for rapid reconstruction of the overall geometric shape. Voxel partitioning can be based on an octree data structure to achieve adaptive hierarchical management, that is, high-density partitioning is used in complex geometric regions, and low-density partitioning is used in smooth regions, thereby reducing the number of redundant points and preserving local geometric details. After partitioning, a weighted average or principal component analysis is performed on the point cloud within each voxel to determine the position and normal vector of the representative point. Voxelized point cloud data can significantly reduce noise interference and data volume, thereby reducing the computational complexity of subsequent 3D point cloud reconstruction algorithms and maintaining geometric consistency.

[0130] The principle of 3D point cloud reconstruction is to find topological adjacency relationships in the point cloud data and construct facet connections to achieve topological closure of points, edges, and faces. Common methods include Poisson Surface Reconstruction and Delaunay Triangulation. Poisson Surface Reconstruction can obtain a smooth and continuous implicit surface by solving the Poisson equation based on the global least squares principle; Delaunay Triangulation can generate triangular faces by determining the normal consistency of the local neighborhood of the point cloud. For areas with gaps or sparse sampling, surface hole filling algorithms and voxel interpolation methods can be used for mesh repair. The resulting geometric mesh model consists of node coordinates, facet elements, and normal information, possessing a high-fidelity spatial topological structure that can be used to represent the surface and internal geometry of facilities. In optional implementations, the reconstruction algorithm can introduce normal constraints or a normal consistency-based filtering mechanism to improve the surface smoothness and edge sharpness of the model. By performing voxel partitioning and 3D point cloud reconstruction operations, the original discrete spatial geometric structure data can be converted into a continuous and computable geometric mesh model, providing a standardized spatial carrier for the subsequent overlay of seepage data and hazard distribution data.

[0131] Seepage data refers to information on the movement and moisture content of water within or on the surface of a facility, used to characterize the permeability, humidity gradient, and potential leakage paths of materials. The process of determining seepage data includes three stages: parameter extraction, spatial mapping, and dynamic correction. First, a moisture content distribution field is established based on wall humidity data. Spatial interpolation algorithms such as Kriging interpolation or Inverse Distance Weighting (IDW) are used to calculate the humidity values ​​of unmeasured nodes, forming a continuous humidity distribution within a unified coordinate system of the fused dataset. Second, the local slope and pore connectivity of the material are calculated by combining spatial geometric data, transforming these structural features into a seepage direction vector field. Then, the humidity distribution is constrained and corrected by combining crack orientation and boundary morphology extracted from surface image data, locally enhancing the seepage path at crack locations. Alternatively, seepage data can also be obtained by solving the steady-state head distribution field in a local region using the finite difference method. The final seepage data includes node humidity values, seepage direction vectors, permeability coefficients, and local head differences. This data can be directly embedded into the geometric mesh model as a physical property layer to describe moisture migration and water accumulation trends within the facility. Through these steps, the 3D model can acquire physical seepage characteristics on top of its spatial geometry, achieving dynamic coupling between the geometric structure and environmental parameters.

[0132] The overlay process refers to the spatial binding of physical parameter fields and structural risk characteristics within the same three-dimensional coordinate system. First, the node coordinates in the seepage data are matched one-to-one with the geometric grid nodes, and humidity values, permeability coefficients, and seepage direction vector information are added to the node attributes. For hazard distribution characteristic data, the coordinate range of the hazard area is spatially overlapped with the grid cells, and the grid cells covering the hazard area are determined through Boolean operations or volume weighting. After overlay, each grid node or cell simultaneously possesses geometric attributes, seepage attributes, and hazard characteristic attributes, allowing for differentiation of different data sources through attribute layer identification. To improve data visualization and analysis efficiency, seepage data can employ continuous chromatographic mapping, and hazard characteristic data can be encoded using category labels or confidence gradients. Optionally, the overlay process can be completed in a Geographic Information System (GIS) environment, using a three-dimensional spatial analysis module for automatic registration; alternatively, it can be automated through script import in a Computer-Aided Design (CAD) platform. The resulting 3D reconstruction model of the facility, after being superimposed, establishes a unified data association between the geometric morphology layer and the physical property layer. This allows the model to not only reflect the spatial structural characteristics of the facility but also demonstrate the coupling relationship between seepage status and hazard distribution. The model supports subsequent mechanical calculations and risk analysis, and provides fundamental input data for dynamic seepage simulation under extreme weather conditions, thereby achieving the synergistic effect of multi-source information in structural safety assessment.

[0133] In one example embodiment of this disclosure, it can be achieved through Figure 3 The steps in the process involve combining a 3D model of the facility reconstruction to determine the structural risk level and potential instability location of the target water-related heritage facility, with reference to... Figure 3 As shown, it can specifically include:

[0134] Step S310: Input the reconstructed 3D model of the facility into a pre-trained physical constraint neural network to determine the stress field and strain field distribution of the reconstructed 3D model of the facility through the physical constraint neural network;

[0135] Step S320: Determine the stress concentration and humidity gradient change data of the target water-related heritage facility based on the stress field and strain field distribution;

[0136] Step S330: Combining the stress concentration and humidity gradient change data at each location, the risk level of each location of the target water-related heritage facility is classified, and the structural risk level corresponding to each location is determined.

[0137] Step S340: The location area where the structural risk level is greater than or equal to the preset level threshold is determined as the potential instability location.

[0138] The Physically Constrained Neural Network (PCN) is a deep learning model that combines data-driven feature learning with the constraint solution of physical equations. Its core idea is to introduce governing equations, boundary conditions, and material constitutive relations as constraints during neural network training, ensuring that the output satisfies both data characteristics and mechanical laws. The network's input includes the geometric node coordinates of the reconstructed 3D model of the facility, material properties (such as elastic modulus and Poisson's ratio), and seepage parameters in the hazard distribution area; the output is the stress and strain components at each node. The network structure typically employs a Multi-Layer Perceptron (MLP), using fully connected layers to perform nonlinear mapping of the input features. The loss function consists of two parts: a data fitting error term, used to constrain the consistency between measured or simulated data and the network's prediction results; and a physical residual term, used to ensure that the prediction results satisfy the equilibrium equations, geometric equations, and constitutive equations. The PCN calculates the residual gradient and backpropagates it using gradient descent and automatic differentiation techniques to simultaneously minimize data errors and physical residuals.

[0139] In an optional implementation, the physical constraint neural network can adopt a hierarchical regional structure, establishing sub-networks for different material regions in the 3D model of the facility reconstruction and applying continuity constraints at the boundaries. For complex seepage coupling problems, it can be extended to a fluid-structure interaction network (FSIN) to simultaneously solve the simultaneous equations of seepage pressure and structural stress. Through this process, the stress and strain field distributions of the facility structure under different stress and seepage environments can be accurately obtained under limited data conditions, ensuring that the results possess both physical consistency and spatial continuity.

[0140] Stress concentration refers to the degree of abnormal stress accumulation in localized areas within a structure, reflecting potential damage initiation locations or structural weaknesses. Stress concentration is calculated through spatial gradient analysis and normalization of the stress field results. Specifically, the principal stress directions and amplitudes are calculated at the mesh nodes of the facility reconstruction 3D model. The Von Mises equivalent stress criterion is used to reduce the multidimensional stress state to a scalar field. Then, stress concentration areas are determined through local peak detection or sliding window standard deviation calculation. Humidity gradient change data originates from seepage data embedded in the facility reconstruction 3D model. The humidity gradient field is obtained by calculating the gradient vector of node humidity in the three-dimensional coordinate direction. To reflect the rate of change, the difference ratio of the humidity gradient can be calculated over time to form humidity gradient change rate data.

[0141] The coupling relationship between humidity gradient changes and stress concentration can be determined through correlation analysis, such as calculating the Pearson correlation coefficient or mutual information, to identify the degree of influence of humidity changes on structural stress distribution. This step allows for the acquisition of spatial coupling characteristics between physical loading and environmental humidity within the facility, providing quantitative parameters for subsequent risk level classification.

[0142] Risk level classification maps the structural state from continuous physical quantities to discrete safety levels, visually reflecting the differences in structural stability across different areas. The determination of risk levels is based on multi-parameter comprehensive evaluation principles; for example, a weighted evaluation model or a fuzzy comprehensive evaluation model can be used. The weighted evaluation model calculates the comprehensive risk value by setting weights for stress concentration and humidity gradient changes; the fuzzy comprehensive evaluation model transforms continuous variables into fuzzy sets using membership functions, and then obtains the discrete level output through fuzzy inference and defuzzification. The risk level classification can include three levels: low risk, medium risk, and high risk, corresponding to different structural safety state intervals. The level thresholds can be determined based on historical survey data or standard documents (such as structural safety assessment specifications); this embodiment does not impose special limitations on this. To enhance the continuity of spatial representation, the risk level classification results can be used to form a continuous risk level field through spatial interpolation or Kriging smoothing.

[0143] The preset risk level threshold refers to the critical value used to distinguish between a safe state and an unstable state. It can be set according to the structure type, material strength, and safety factor. For example, the preset risk level threshold can be 0.75. When the comprehensive risk value is greater than or equal to 0.75, the area can be considered to be in a high-risk state and defined as a potential instability location. Of course, the specific value of the preset risk level threshold can be customized according to the actual situation, and this embodiment does not impose any special limitations on it. The process of extracting potential instability locations can include threshold screening and spatial connectivity analysis. Threshold screening involves traversing all nodes in the risk level field and marking nodes with a level not lower than the threshold as instability candidate points. Spatial connectivity analysis is based on the three-dimensional eight-neighbor or six-neighbor rule to identify mutually adjacent candidate points and aggregate them into instability regions.

[0144] In some optional implementations, to enhance boundary accuracy, morphological dilation-erosion operations or region-growing-based boundary reconstruction methods can be introduced to smoothly correct the boundaries of potential instability zones. For multi-layered structures, potential instability zones can be calculated independently at different levels, and inter-layer correspondences can be established through vertical projection to identify hidden instability layers within the structure. The final potential instability locations are stored in three-dimensional spatial coordinates and corresponding risk levels, and can be displayed through a three-dimensional visualization system. These steps enable the transformation from quantitative assessment to spatial location, allowing potential instability zones to be accurately identified in the reconstructed three-dimensional model of the facility, thus providing direct spatial guidance for structural reinforcement, drainage intervention, and real-time monitoring.

[0145] In one example embodiment of this disclosure, dynamic seepage simulation based on external environmental parameters and a reconstructed 3D model of the facility can be achieved through the following steps to predict the internal water accumulation distribution and structural instability trend of the target water-related heritage facility under extreme weather conditions. Specifically, this may include:

[0146] The system can discretize the geometry of the 3D model of the reconstructed facility by meshing it and mapping seepage data with wall humidity data to generate a seepage solution mesh corresponding to the geometry of the 3D model. Rainfall intensity parameters, surface runoff parameters, and groundwater level change parameters are set on the seepage solution mesh to form dynamic boundary conditions, and a dynamic seepage numerical model is established based on the finite element method or finite volume method. Based on the dynamic seepage numerical model, the water volume distribution and wall permeability stress data at each time point are calculated in a time series to determine the internal water distribution and structural instability trend of the target water-related heritage facility under extreme weather conditions.

[0147] Mesh discretization refers to the process of transforming the geometric structure of a continuous 3D model of a facility into a finite number of regularly shaped discrete elements, enabling them to be used for numerical solutions of the seepage field equations. Mesh elements can be tetrahedral, hexahedral, or prismatic, with the specific selection determined by the geometric complexity of the facility and the required accuracy of the seepage analysis; this embodiment does not impose any special limitations. When the geometric model contains curved surfaces or irregular cross-sections, a tetrahedral unstructured mesh is used to maintain shape fit; for more regular walls or dams, a hexahedral structured mesh can be used to improve solution stability. During mesh generation, boundary surface meshes are first generated based on the facility's geometric boundaries, and then internal mesh elements are generated using a volumetric meshing algorithm. In refined regions, such as around cracks or areas with significant humidity gradients, a local densification strategy is used to increase element density, allowing the model to more accurately reflect local seepage characteristics. To ensure computational accuracy, mesh quality assessment is performed, including aspect ratio, tilt, and minimum angle detection, and geometric correction is performed using a smoothing algorithm. The mapping between seepage data and wall humidity data is completed through spatial registration and interpolation.

[0148] In the seepage data, the humidity, permeability coefficient, and seepage direction vector of each node are mapped to the corresponding grid node or cell center. Inverse Distance Weighting (IDW) or Kriging Interpolation is used to supplement the numerical values ​​of non-overlapping areas. Humidity data can be synchronized with coordinates through timestamps to achieve spatiotemporal mapping, preserving dynamic changes in the grid domain. Through this mapping process, the seepage solution grid possesses both geometric topological properties and physical parameter properties, achieving an integrated representation of facility geometry and seepage information.

[0149] Rainfall intensity parameters, surface runoff parameters, and groundwater level variation parameters together constitute the time-varying input of the external hydrodynamic environment of the facility. Rainfall intensity parameters reflect the amount of water acting on the facility surface per unit time, usually expressed in millimeters per hour (mm / h), and can be input based on meteorological data or rainfall curves generated by extreme weather forecast models. Surface runoff parameters represent the water flow convergence characteristics on the facility surface, and the flow boundary distribution can be determined by calculating surface slope, roughness coefficient, and flow direction. Groundwater level variation parameters describe the dynamic boundaries of the aquifers at the bottom and surrounding areas of the facility, and can be provided by measured data from observation wells or regional groundwater numerical models. Establishing dynamic boundary conditions requires mapping these parameters to the boundary nodes of the seepage solution grid. Rainfall intensity parameters act on the top boundary of the model to form a time-dependent flux boundary condition (Neumann boundary), surface runoff parameters are converted to surface flow boundaries using the Manning formula, and groundwater level variation parameters are applied as head boundary conditions (Dirichlet boundary) to the bottom and lateral boundaries.

[0150] The establishment of a dynamic seepage numerical model can be based on the seepage control equation, that is, in the saturated region, a combination of Darcy's law and the continuity equation can be used. For example, the dynamic seepage numerical model can be expressed as the following relationship:

[0151] ;

[0152] in, Let S be the porosity, S be the saturation, ρ be the water density, and q be the seepage velocity vector; the relationship between q and the hydraulic head h is given by... Given that K is the permeability tensor, This represents the head gradient. In the unsaturated region, a nonlinear relationship also needs to be introduced. Where Ks is the saturated permeability coefficient and n is an empirical exponent. When using the finite element method, the governing equations are discretized spatially into element integral form, and the nodal head is solved using shape function interpolation. When using the finite volume method, the equations are integrated over the control volume, and the flow rate is solved using a high-resolution flux reconstruction method. For time discretization, an implicit difference scheme can be used to ensure stability, and the Newton-Raphson iteration can be used to solve the nonlinear terms. To handle unsteady seepage under heavy rainfall, adaptive time step control can be introduced, dynamically adjusting the time step based on residual changes. Model calculations can obtain the head distribution at each node at different times by solving the linear equation system, thereby obtaining the seepage velocity and flow rate. Through this dynamic solution based on the finite element method or the finite volume method, the transient seepage behavior of the facility under extreme weather conditions can be accurately described, capturing the temporal evolution of water accumulation and discharge processes.

[0153] The calculation of water volume distribution is achieved by integrating regions in the model mesh where the water head is higher than the height of the surface nodes. Specifically, the free water surface position is calculated at each time step, used as the integration boundary, and volumetric elements beyond this boundary are accumulated to obtain the water volume distribution curves at different time points. The calculation of wall seepage stress data is based on the seepage stress coupling principle, by solving the seepage force term. Where ρ is the density of water and g is the acceleration due to gravity. The water head gradient is used to superimpose the seepage force onto the total stress field at the structural nodes, forming the time-varying distribution data of the wall seepage stress.

[0154] In an optional implementation, the wall permeation stress can be dynamically interacted with the structural displacement field through a fluid-structure interaction model, synchronizing stress changes with seepage evolution. For water accumulation and diffusion areas, a flow direction analysis algorithm based on flux conservation can be introduced to determine the water accumulation path and calculate the local water volume growth rate and recession rate to assess the facility's drainage capacity. Through time series calculations, a dynamic evolution map of water volume distribution and wall permeation stress can be generated; by performing correlation analysis, the region of stress concentration and enhanced seepage coupling in the structure can be identified through the stress gradient change trend over time, which is the potential instability development zone. Through the implementation of the above steps, the dynamic mapping of internal hydrodynamics and structural stress under extreme weather conditions can be achieved at the numerical simulation level, allowing for a quantitative characterization of water accumulation distribution and instability trends.

[0155] In one example embodiment of this disclosure, emergency protection task data can be generated based on the internal water accumulation distribution and structural instability trend through the following steps, which may specifically include:

[0156] Based on the distribution of internal water accumulation and the trend of structural instability, the key protection areas corresponding to the target water-related heritage facilities can be identified; based on the regional parameters of the key protection areas, corresponding emergency protection measures can be selected from the preset emergency protection task database. The emergency protection measures include at least one or more combinations of drainage measures, reinforcement measures, and sealing measures; based on the emergency protection measures and the key protection areas, emergency protection task data containing protection location, priority, and execution order can be generated.

[0157] Critical protection areas refer to localized spatial regions within a facility structure that, under extreme weather conditions, experience concentrated water accumulation or significantly increased seepage stress, exhibiting a potential tendency towards instability. The identification process uses dynamic seepage numerical models and structural risk level assessments as inputs, employing spatial overlay analysis and risk clustering algorithms. For example, firstly, the spatial overlap between water volume distribution data and wall seepage stress data is extracted from the 3D model of the reconstructed facility, and the comprehensive risk index for each grid cell is calculated. Subsequently, clustering algorithms such as K-means or density-based spatial clustering of applications with noise (DBSCAN) are used to aggregate high-risk cells into continuous protection areas, and their spatial connectivity is checked to ensure the geometric integrity of the protection areas. The identified critical protection areas can be further calculated for their center coordinates, spatial boundaries, and area to determine their spatial extent and priority sequence. Optionally, a time weighting term can be introduced into the identification process, assigning higher weights to areas where risk persists across different time steps, thereby improving the stability of the dynamic risk assessment.

[0158] The emergency protection task database is a pre-built standardized database of protection measures, recording the applicable conditions, construction methods, and execution parameters for different types of protection tasks. Regional parameters can include the spatial location, area, depth, risk level, seepage type, and structural characteristics of the protected area, used to retrieve the most suitable protection solution. Emergency protection measures in the database can be categorized into at least three main types: drainage measures, reinforcement measures, and sealing measures. Drainage measures are mainly used to address protected areas with a significant increase in water volume distribution; for example, this may include setting up drainage ditches, pumping devices, or temporary drainage holes. Reinforcement measures are mainly applied to areas with high seepage stress and high structural risk levels; for example, this may include grouting reinforcement, steel reinforcement, and partial concrete encapsulation. Sealing measures are used to prevent the expansion of seepage channels or backflow of water; for example, this may involve using expansive waterproofing materials, geomembranes, or grouting sealing techniques. During the search, the database can match the applicable tags of measures based on the type tags of the key protected area (e.g., "water accumulation type," "seepage type," "fissure type") with the applicable tags of the measures, and select one or more combined measures to form a protection solution according to the risk level priority rule. In some alternative implementations, rule-based expert systems or machine learning classifiers (such as random forests) can be used to quickly reason about and match protective measures. This database-driven approach to selecting protective strategies can significantly improve the speed of emergency decision-making and the adaptability of measures, ensuring that protective strategies are consistent with the characteristics of on-site risks.

[0159] Emergency protection task data is a structured set of task instructions used to guide personnel or intelligent devices in performing corresponding emergency operations. The generation process includes task parameter extraction, priority ranking, and time-series scheduling. In the task parameter extraction stage, the task execution location and the scope of the protected object are determined based on the three-dimensional coordinates of the protected area and the type of protection measures. In the priority ranking stage, a comprehensive priority index is calculated based on the risk level, instability trend rate, and environmental impact factors (such as rainfall duration). Tasks with higher priority indices are prioritized for execution. The execution order is determined based on a task dependency graph. If there is spatial overlap or operational dependency between protection tasks, such as "drainage" needing to be executed before "reinforcement," an optimal task sequence is generated using a topological sorting algorithm.

[0160] In some optional implementations, a task scheduling algorithm based on reinforcement learning (RL) can be employed to optimize the protection sequence through a dynamic reward mechanism to minimize the overall risk residual. The final generated emergency protection task data includes fields such as task number, 3D coordinates of the protection location, task type, priority, execution order, and execution duration, and can be output to an emergency management platform or mobile terminal for command and dispatch or automated control. Through this process, emergency protection work can achieve a closed-loop linkage from risk identification to task generation, making protective measures executable and adaptable in real time, ensuring the safety protection of critical areas can be completed in the shortest possible response time under extreme weather conditions.

[0161] In one example embodiment of this disclosure, the multi-source sensing device includes at least a lidar, a close-range image acquisition unit, and a soil moisture sensor; the acquisition of a multi-source heterogeneous dataset of the target water-related heritage facility can be achieved through the following steps, specifically including:

[0162] It can collect spatial geometric structure data of the target water-related heritage facilities, which consists of spatial point cloud information output by the laser scanning module; collect surface image data of the target water-related heritage facilities, which consists of texture image information output by the close-range photography module; collect wall humidity data of the target water-related heritage facilities, which consists of humidity and seepage pressure information output by the seepage humidity sensing module; and use the attitude sensing unit of the multi-source sensing device to perform synchronous time stamping and attitude calibration on the spatial geometric structure data, surface image data and wall humidity data to construct a multi-source heterogeneous dataset.

[0163] Multi-source sensing equipment refers to a comprehensive sensing device deployed within the survey area of ​​the target water-related heritage facility to simultaneously acquire spatial geometric data, surface image data, and wall moisture data. For example, multi-source sensing equipment can consist of a lidar module, a close-range image acquisition unit, a soil moisture sensor, and an attitude sensing unit, and achieve spatiotemporal correspondence of multimodal data through a unified time synchronization system.

[0164] The lidar module is used for high-precision scanning of the spatial geometry of heritage facilities. Its working principle is based on the Time-of-Flight (ToF) ranging principle, which calculates the target distance by emitting laser pulses and measuring the round-trip time of the reflected signals. To adapt to underground or confined spaces, the lidar module can employ a 360° rotating scanning head or a line laser scanning structure to acquire omnidirectional point cloud data of the facility's interior walls. In implementation, the lidar module can be a phase-shift lidar or a pulse lidar; the former offers high precision but is suitable for short-range scanning, while the latter is suitable for long-range, large-area scanning. During the scanning process, a self-calibration algorithm corrects geometric errors between scan points in real time, and attitude angle data provided by the attitude sensing unit is used for coordinate correction, ensuring strict alignment of the point cloud data with the actual spatial coordinate system. The lidar module outputs spatial geometric data, which represents the three-dimensional morphology of the facility's surface in point cloud form, offering high spatial resolution and geometric accuracy.

[0165] The close-range image acquisition unit is an imaging module used to acquire image data of the facility's surface. Its basic components include a high-resolution image sensor, a lens assembly, an exposure control unit, and an image storage module. This unit typically employs a close-range photogrammetry method based on computer vision principles, achieving high-fidelity capture of texture features through multi-view imaging. The image sensor can be a complementary metal-oxide-semiconductor (CMOS) or charge-coupled device (CCD) type, and the lens focal length can be adaptively adjusted according to the facility's space size; this embodiment does not impose any special limitations on this. During the shooting process, the close-range image acquisition unit achieves panoramic coverage through multi-angle movement shooting or mechanical gimbal control, ensuring the continuity and overlap of image data. Image exposure is dynamically adjusted by an automatic exposure algorithm based on lighting conditions to cope with changes in light in underground or dimly lit spaces. Optionally, the close-range image acquisition unit can also integrate an infrared supplementary lighting module or a structured light emitting unit to improve imaging quality in low-light environments. After image acquisition is completed, the image data is spatially registered using spatial pose information (including position coordinates and rotation angle) provided by the attitude sensing unit to ensure accurate positioning of each frame in the global coordinate system. The surface image data output by the close-range image acquisition unit includes texture information, crack distribution, and surface erosion conditions of the facility surface, providing basic data support for subsequent visual feature extraction and hazard identification.

[0166] Soil moisture sensors are sensing components used to measure moisture data in facility walls. Their function is to sense the water content and seepage pressure of the soil inside or around the structural material. Soil moisture sensors can operate using time-domain reflectometry (TDR) or frequency-domain reflectometry (FDR) principles, reflecting the water content of the medium by measuring the propagation speed of electromagnetic waves or changes in impedance. For wall materials, a drilled moisture sensing probe can be inserted into the wall to collect moisture profiles at different depths. For foundations or dams, embedded soil moisture sensors can be deployed at specific depths to monitor moisture changes over a long period. The sensor sampling frequency can be adjusted according to the rate of environmental change; a high-frequency sampling mode is used during rainfall or seepage monitoring, while a low-frequency mode is used in stable environments to extend service life. Moisture and seepage pressure data are digitized by an analog-to-digital converter (ADC) and transmitted to the main control unit for recording and analysis via a wireless or wired communication module. In alternative implementations, the humidity sensor can also incorporate resistive, capacitive, or dielectric constant measurement principles to adapt to heritage structures made of different materials. The collected wall humidity data not only reflects the moisture content of the facility materials but also indirectly characterizes the direction and velocity of seepage, providing important physical evidence for the analysis of the facility's seepage status.

[0167] The attitude sensing unit is a key module in multi-source sensing devices used for data synchronization, time stamping, and attitude calibration. It consists of an Inertial Measurement Unit (IMU), a Global Positioning System (GPS) or a local positioning system, and a time synchronization module. The IMU, composed of a three-axis gyroscope and a three-axis accelerometer, measures the device's attitude changes and motion state; the GPS or local positioning system provides a spatial position reference. The time synchronization module ensures the consistency of data acquisition time across all sensors through a timestamp mechanism, achieving microsecond-level synchronization accuracy using either Network Time Protocol (NTP) or Precision Time Protocol (PTP). In confined underground spaces where satellite signals are unavailable, the attitude sensing unit can maintain attitude continuity through geomagnetic navigation or inertial integration algorithms. During calibration, the system first registers the lidar point cloud, close-range imagery, and humidity sensor data to the device coordinate system, then transforms them to a unified spatial coordinate system using an attitude matrix. This time-attitude integrated calibration mechanism ensures that data collected by different sensors remain consistent in both time and space, thus forming a complete multi-source heterogeneous dataset.

[0168] When collecting spatial geometric data of the target water-related heritage facilities, the lidar module emits a high-frequency laser beam to scan the surface of the facilities, measuring point cloud data and calculating the three-dimensional coordinates of each point. The spatial geometric data is optimized through point cloud filtering, noise removal, and ground separation algorithms (such as Statistical Outlier Removal, SOR) to ensure data accuracy and integrity. When collecting surface image data, the near-field image acquisition unit images the facility surface from multiple angles, acquiring multiple overlapping images. The overlap rate is typically controlled between 60% and 80% to facilitate subsequent image stitching and feature registration. When collecting wall moisture data, soil moisture sensors are deployed at multiple depths to collect moisture and seepage pressure values, and the data is labeled to a unified spatial coordinate system. Through time synchronization and spatial calibration by the attitude sensing unit, different types of data are fused under a unified time axis and spatial coordinate system to form a complete multi-source heterogeneous dataset. These steps ensure spatial consistency between different physical quantities, enabling the establishment of quantitative correlations between geometric, image, and moisture information in subsequent fusion analysis, thereby improving the integrity and spatial accuracy of heritage facility survey data.

[0169] In an optional embodiment of this disclosure, the fusion processing of multi-source heterogeneous datasets to form a fused dataset can be achieved through the following steps, specifically including:

[0170] A unified spatial coordinate framework can be established based on spatial geometric structure data. On the basis of the unified spatial coordinate framework, surface image data is spatially registered with spatial geometric structure data through feature point matching algorithm, and spatial interpolation mapping is performed on wall humidity data. The humidity value and seepage pressure value are associated with the corresponding geometric nodes in the spatial coordinate framework to form a fused dataset.

[0171] The spatial coordinate frame refers to a coordinate reference system used for geometric alignment, spatial calibration, and unified projection of data from different sensor sources. It primarily provides a unified spatial reference for spatial geometric structure data, surface image data, and wall humidity data. The process of establishing a spatial coordinate frame can include three steps: coordinate system definition, point cloud reference construction, and spatial reference transformation. In the coordinate system definition stage, a right-handed Cartesian coordinate system can be used as the mathematical basis for the unified spatial coordinate frame. The origin can be set at the center of the survey area or at a reference measurement point, and the coordinate axes are oriented according to the main extension direction of the facility or geographic north. In the point cloud reference construction stage, an initial point cloud reference coordinate system is established using feature points from the spatial geometric structure data acquired by the lidar module. This coordinate system is then corrected using the attitude matrix (including pitch, yaw, and roll angles) provided by the attitude sensing unit and registered with the Geographic Coordinate System (GCS). In optional implementations, if the facility is located in a closed environment or underground space, a fusion positioning algorithm using an inertial measurement unit and a geomagnetic sensor can be used to maintain coordinate system stability. In the spatial reference transformation stage, point cloud data collected from multiple time periods or devices are globally aligned using least squares registration or Iterative Closest Point (ICP) algorithms to form a unified, continuous, and non-overlapping spatial geometric framework. This spatial coordinate framework serves as the geometric reference for subsequent mapping of image data and humidity data, ensuring consistent coordinate reference relationships between different data sources in space and guaranteeing the spatial accuracy and consistency of subsequent fusion calculations.

[0172] Feature point matching algorithms are key technologies used to establish geometric relationships between image information and 3D point clouds. Feature point matching achieves spatial alignment based on image feature extraction and 3D geometric constraints. For example, firstly, stable local feature points can be extracted from surface image data. For instance, Scale-Invariant Feature Transform (SIFT), Speeded-Up Robust Features (SURF), or Histogram of Oriented Gradients (HOG) algorithms can be used for feature extraction to obtain salient points such as texture edges, corners, and crack features. Secondly, corresponding 3D geometric feature points are extracted from the spatial geometric structure data. These features may include curvature change points, normal abrupt change points, or local minimum curvature points. Using the principle of projection mapping, image feature points are projected onto 3D space through the camera's intrinsic and extrinsic parameter matrices, and the matching error between them and point cloud feature points is calculated. Matching optimization can use a random sampling consensus algorithm to filter out erroneous matching point pairs and employ bundle adjustment for global optimization, thereby ensuring registration accuracy at the millimeter level. In some alternative implementations, for environments with uneven lighting and repetitive textures, deep learning-based feature matching networks (such as SuperGlue or D2-Net) can be used to improve the robustness of feature matching. Through this spatial registration process, surface image data is accurately superimposed on spatial geometric structure data, realizing the spatial integration of two-dimensional image information and three-dimensional geometric information, enabling the surface texture, cracks, and geometric morphology of the facility to be represented synchronously in a unified spatial domain.

[0173] Spatial interpolation mapping refers to the process of using known measurement point data to infer the spatial distribution of humidity and seepage pressure at unknown points, thus forming a continuous physical field in three-dimensional space from discrete measurements. Specifically, firstly, the coordinates of the humidity data sampling points on the wall are matched with the node coordinates of the spatial geometric structure data to construct a distance matrix between nodes. Then, the Inverse Distance Weighting (IDW) algorithm is used to calculate the humidity value for each unmeasured node. For areas with drastic geological changes or significant humidity gradients, the Kriging Interpolation method can be used to model spatial autocorrelation using a semivariance function, thereby obtaining a humidity distribution field that better reflects the actual seepage trend. The mapping process for seepage pressure values ​​is similar to that for humidity; a continuous seepage pressure field is obtained through interpolation of pressure measurement point data. To ensure consistency between humidity and seepage pressure values, the interpolation results can be jointly corrected based on experience. After interpolation, the humidity and seepage pressure values ​​are respectively bound to the corresponding geometric node attribute tables in the spatial coordinate frame. Each node has a unique humidity value, seepage pressure value, and its time label. In some alternative implementations, if the humidity data has time-series characteristics, spatio-temporal kriging or Kalman filtering methods can be used for dynamic mapping to obtain a continuous distribution of humidity over time. Through this spatial interpolation mapping process, the wall humidity data is transformed from discrete measurement points into three-dimensional field data with spatial continuity and physical interpretability, enabling a quantitative coupling relationship between the seepage state and the geometric structure within a unified coordinate framework.

[0174] It should be noted that although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0175] Furthermore, in this example embodiment, an intelligent surveying device for traditional water-related heritage facilities based on multi-source heterogeneous data is also provided. (Refer to...) Figure 4 As shown, the intelligent surveying equipment 400 for traditional water-related heritage facilities based on multi-source heterogeneous data includes: a data acquisition platform 410, a control and processing unit 420, and a communication unit 430. Wherein:

[0176] The data acquisition platform 410 integrates a lidar, a close-range camera, and a humidity sensor, and is equipped with an attitude sensing component to provide attitude and time synchronization information.

[0177] The communication unit 430 is used for data interaction and command issuance with an external terminal;

[0178] The control processing unit 420 is configured to:

[0179] The data acquisition platform is controlled to collect multi-source heterogeneous datasets within the survey area of ​​the target water-related heritage facility. The multi-source heterogeneous datasets include at least spatial geometric structure data, surface image data, and wall humidity data.

[0180] The multi-source heterogeneous datasets are fused to obtain a fused dataset;

[0181] The fused dataset is input into a pre-trained hazard identification model to identify risks and hazards, and to determine the hazard distribution characteristics data.

[0182] Based on the hazard distribution feature data and the fused dataset, a 3D model of facility reconstruction is constructed. The 3D model of facility reconstruction includes the geometric structure, seepage data, and spatial distribution location of the target water-related heritage facility.

[0183] Based on the reconstructed 3D model of the facility, the structural risk level and potential instability location of the target water-related heritage facility were determined;

[0184] The fused dataset, the hazard distribution characteristic data, the facility reconstruction 3D model, and the structural risk level and potential instability location are transmitted and / or stored through the communication unit.

[0185] The specific details of each module of the intelligent survey equipment for traditional water-related heritage facilities based on multi-source heterogeneous data mentioned above have been described in detail in the corresponding intelligent survey method for traditional water-related heritage facilities based on multi-source heterogeneous data, so they will not be repeated here.

[0186] It should be noted that although several modules or units of intelligent surveying equipment for traditional water-related heritage facilities based on multi-source heterogeneous data have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0187] Furthermore, in an exemplary embodiment of this disclosure, an electronic device is also provided that can implement the above-described intelligent survey method for traditional water-related heritage facilities based on multi-source heterogeneous data.

[0188] Those skilled in the art will understand that various aspects of this disclosure can be implemented as a system, method, or program product. Therefore, various aspects of this disclosure can be embodied in the following forms: a completely hardware embodiment, a completely software embodiment (including firmware, microcode, etc.), or an embodiment combining hardware and software aspects, collectively referred to herein as a "circuit," "module," or "system."

[0189] The following reference Figure 5 To describe an electronic device 500 according to such an embodiment of the present disclosure. Figure 5 The electronic device 500 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.

[0190] like Figure 5 As shown, the electronic device 500 is presented in the form of a general-purpose computing device. The components of the electronic device 500 may include, but are not limited to: at least one processing unit 510, at least one storage unit 520, a bus 530 connecting different system components (including storage unit 520 and processing unit 510), and a display unit 540.

[0191] The storage unit stores program code that can be executed by the processing unit 510, causing the processing unit 510 to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure. For example, the processing unit 510 can perform actions such as... Figure 1 Step S110, as shown, involves deploying multi-source sensing devices within the survey area of ​​the target water-related heritage facility and collecting a multi-source heterogeneous dataset of the facility. This dataset includes at least spatial geometric data, surface image data, and wall humidity data. Step S120 involves fusing the multi-source heterogeneous dataset to form a fused dataset. Step S130 involves inputting the fused dataset into a pre-trained hazard identification model for risk and hazard identification, determining hazard distribution characteristic data. Step S140 involves constructing a 3D reconstruction model of the facility based on the hazard distribution characteristic data and the fused dataset. This 3D reconstruction model includes the geometric structure of the target water-related heritage facility, seepage data, and the spatial distribution location of hazards. Step S150 involves combining the 3D reconstruction model to determine the structural risk level and potential instability location of the target water-related heritage facility.

[0192] Storage unit 520 may include readable media in the form of volatile storage units, such as random access memory (RAM) 521 and / or cache memory 522, and may further include read-only memory (ROM) 523.

[0193] Storage unit 520 may also include a program / utility 524 having a set (at least one) program module 525, such program module 525 including but not limited to: operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0194] Bus 530 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0195] Electronic device 500 can also communicate with one or more external devices 570 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with electronic device 500, and / or with any device that enables electronic device 500 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 550. Furthermore, electronic device 500 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 560. As shown, network adapter 560 communicates with other modules of electronic device 500 via bus 530. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 500, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0196] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, terminal device, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0197] In exemplary embodiments of this disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the methods described above is stored. In some possible embodiments, various aspects of this disclosure may also be implemented as a program product including program code that, when the program product is run on a terminal device, causes the terminal device to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure.

[0198] refer to Figure 6 As shown, a program product 600 for implementing the above-described intelligent surveying method for conventional water heritage facilities based on multi-source heterogeneous data, according to embodiments of the present disclosure, is described. It may employ a portable compact disk read-only memory (CD-ROM) and include program code, and can run on a terminal device, such as a personal computer. However, the program product of this disclosure is not limited thereto. In this document, the readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0199] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0200] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting programs for use by or in conjunction with an instruction execution system, apparatus, or device.

[0201] The program code contained on the readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0202] Program code for performing the operations of this disclosure can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0203] Furthermore, the above figures are merely illustrative of the processes included in the method according to exemplary embodiments of this disclosure and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0204] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0205] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

[0206] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for intelligent surveying of traditional water-related heritage facilities based on multi-source heterogeneous data, characterized in that, include: Multi-source sensing devices are deployed within the survey area of ​​the target water-related heritage facility, and multi-source heterogeneous datasets of the target water-related heritage facility are collected through the multi-source sensing devices. The multi-source heterogeneous datasets include at least spatial geometric structure data, surface image data, and wall humidity data. The multi-source heterogeneous datasets are fused to form a fused dataset; The fused dataset is input into a pre-trained hazard identification model to identify risks and hazards and determine the hazard distribution characteristics. Based on the hazard distribution feature data and the fused dataset, a three-dimensional model for facility reconstruction is constructed. The three-dimensional model for facility reconstruction includes the geometric structure, seepage data, and spatial distribution location of the target water-related heritage facility. Based on the reconstructed 3D model of the facility, the structural risk level and potential instability location of the target water-related heritage facility were determined; The step of constructing a 3D model for facility reconstruction based on the hazard distribution characteristic data and the fused dataset includes: The fused dataset is divided into voxels and then subjected to 3D point cloud reconstruction to generate a geometric mesh model of the target water-related heritage facility. The seepage data of the target water-related heritage facility is determined based on the spatial geometry data, the surface image data, and the wall humidity data. The seepage data and the hazard distribution characteristic data are superimposed on the geometric mesh model to construct a three-dimensional reconstruction model of the facility with seepage distribution attributes.

2. The method according to claim 1, characterized in that, The method further includes: Obtain external environmental parameters for extreme weather conditions, including at least rainfall intensity parameters, surface runoff parameters, and groundwater level change parameters; Based on the external environmental parameters and the reconstructed three-dimensional model of the facility, dynamic seepage simulation is performed to predict the internal water accumulation distribution and structural instability trend of the target water-related heritage facility under extreme weather conditions. Emergency protection task data is generated based on the internal water accumulation distribution and the structural instability trend.

3. The method according to claim 1, characterized in that, The hazard identification model includes at least a convolutional neural network and a graph neural network; the step of inputting the fused dataset into the pre-trained hazard identification model for risk and hazard identification, and determining the hazard distribution characteristic data, includes: The convolutional neural network extracts crack and erosion features from the surface image data, and the graph neural network analyzes the spatial relationship between humidity and seepage nodes in the wall humidity data to identify potential leakage channels. In the hazard identification model, the crack and erosion features and the potential leakage channel features are subjected to multimodal feature fusion processing to generate preliminary hazard identification results; The distribution characteristics of the hazards are determined based on the preliminary hazard identification results.

4. The method according to claim 3, characterized in that, The step of determining the hazard distribution characteristic data based on the preliminary hazard identification results includes: Spatial clustering operations are performed on the preliminary hazard identification results to form a set of hazard areas; Based on the crack length, seepage intensity, and settlement depth extracted from the set of potential hazard areas, the potential hazard distribution feature data corresponding to the fused dataset is generated.

5. The method according to claim 1, characterized in that, Based on the reconstructed 3D model of the facility, the structural risk level and potential instability location of the target water-related heritage facility were determined, including: The reconstructed 3D model of the facility is input into a pre-trained physical constraint neural network to determine the stress and strain field distributions of the reconstructed 3D model of the facility through the physical constraint neural network. Based on the stress field and strain field distribution, the stress concentration and humidity gradient change data of the target water-related heritage facility are determined; By combining the stress concentration and humidity gradient change data at each location, the risk level of each location of the target water-related heritage facility is classified, and the structural risk level corresponding to each location is determined. The location region where the structural risk level is greater than or equal to a preset level threshold is identified as a potential instability location.

6. The method according to claim 2, characterized in that, The step of performing dynamic seepage simulation based on the external environmental parameters and the reconstructed three-dimensional model of the facility to predict the internal water accumulation distribution and structural instability trend of the target water-related heritage facility under extreme weather conditions includes: The geometric structure corresponding to the reconstructed 3D model of the facility is discretized by meshing, and the seepage data and wall humidity data are mapped to generate a seepage solution mesh corresponding to the geometric structure of the reconstructed 3D model of the facility. Rainfall intensity parameters, surface runoff parameters, and groundwater level change parameters are set on the seepage solution grid to form dynamic boundary conditions, and a dynamic seepage numerical model is established based on the finite element method or the finite volume method. Based on the dynamic seepage numerical model, the water volume distribution and wall permeability stress data at each time point are calculated in time series, so as to determine the internal water distribution and structural instability trend of the target water-related heritage facility under extreme weather conditions.

7. The method according to claim 6, characterized in that, Based on the internal water accumulation distribution and the structural instability trend, corresponding emergency protection task data is generated, including: Based on the internal water accumulation distribution and the structural instability trend, identify the key protection areas corresponding to the target water-related heritage facilities; Based on the regional parameters of the key protection area, the corresponding emergency protection measures are selected from the preset emergency protection task database. The emergency protection measures include at least one or more combinations of drainage measures, reinforcement measures, and sealing measures. Based on the emergency protection measures and the key protection areas, emergency protection task data including protection location, priority, and execution order is generated.

8. The method according to claim 1, characterized in that, The multi-source sensing device includes at least a lidar, a close-range image acquisition unit, and a soil moisture sensor; the acquisition of the multi-source heterogeneous dataset of the target water-related heritage facility through the multi-source sensing device includes: The spatial geometric structure data of the target water-related heritage facility is collected, wherein the spatial geometric structure data is composed of spatial point cloud information output by the laser scanning module; Surface image data of the target water-related heritage facility is collected, wherein the surface image data consists of texture image information output by the close-up photography module; Collect wall humidity data of the target water-related heritage facility, wherein the wall humidity data consists of humidity and seepage pressure information output by the seepage humidity sensing module; The attitude sensing unit of the multi-source sensing device performs synchronous time stamping and attitude calibration on the spatial geometric structure data, the surface image data, and the wall humidity data to construct the multi-source heterogeneous dataset.

9. An intelligent surveying device for traditional water-related heritage facilities based on multi-source heterogeneous data, characterized in that, It includes a data acquisition platform, a control and processing unit, and a communication unit, wherein: The data acquisition platform integrates a lidar, a close-range camera, and a humidity sensor, and is equipped with an attitude sensing component to provide attitude and time synchronization information. The communication unit is used for data interaction and command issuance with external terminals; The control processing unit is configured to: The data acquisition platform is controlled to collect multi-source heterogeneous datasets within the survey area of ​​the target water-related heritage facility. The multi-source heterogeneous datasets include at least spatial geometric structure data, surface image data, and wall humidity data. The multi-source heterogeneous datasets are fused to obtain a fused dataset; The fused dataset is input into a pre-trained hazard identification model to identify risks and hazards, and to determine the hazard distribution characteristics data. Based on the hazard distribution feature data and the fused dataset, a 3D model of facility reconstruction is constructed. The 3D model of facility reconstruction includes the geometric structure, seepage data, and spatial distribution location of the target water-related heritage facility. Based on the reconstructed 3D model of the facility, the structural risk level and potential instability location of the target water-related heritage facility were determined; The fused dataset, the hazard distribution characteristic data, the facility reconstruction 3D model, and the structural risk level and potential instability location are transmitted and / or stored through the communication unit; The step of constructing a 3D model for facility reconstruction based on the hazard distribution characteristic data and the fused dataset includes: The fused dataset is divided into voxels and then subjected to 3D point cloud reconstruction to generate a geometric mesh model of the target water-related heritage facility. The seepage data of the target water-related heritage facility is determined based on the spatial geometry data, the surface image data, and the wall humidity data. The seepage data and the hazard distribution characteristic data are superimposed on the geometric mesh model to construct a three-dimensional reconstruction model of the facility with seepage distribution attributes.

Images