Heat supply pipe network leakage high-precision detection and abnormity intelligent diagnosis integrated system

Abstract

The invention relates to the technical field of heat supply network detection, and discloses a heat supply network leakage high-precision detection and abnormity intelligent diagnosis integrated system. Comprising an air-ground collaborative inspection platform for collecting infrared and depth data, a multi-modal data space-time alignment module for executing heterogeneous data space-time registration, a three-dimensional apparent heat flow form construction module for establishing geometry-temperature coupling representation, a heterogeneous vector field analysis module for screening anomalies based on gradient-normal coupling, and a data processing module for processing the heterogeneous vector field. The thermal geometry self-adaptive reconstruction module is used for executing local high-precision recursive sampling; the multi-view radiation verification module is used for eliminating artifacts by utilizing radiation directivity; and the reverse tracking and positioning module is used for inverting the three-dimensional space position of a leakage source. Accurate mapping of thermal-geometric data is realized by constructing a three-dimensional visual heat flow space; eliminating environmental interference and artifacts by utilizing the coupling characteristics of the thermal gradient and the geometric normal vector and the multi-view radiation difference; self-adaptive high-density sampling is carried out based on thermal geometry saliency, and the leakage source burial depth is inversed by using a reverse heat conduction model.

Description

Technical Field

[0001] This invention relates to the field of heating pipeline network detection technology, specifically an integrated system for high-precision detection and intelligent diagnosis of leaks and anomalies in heating pipeline networks. Background Technology

[0002] Urban centralized heating networks are a crucial component of urban infrastructure, and their operational safety directly impacts energy efficiency and public safety. Due to their long-term underground burial, these pipelines are susceptible to corrosion, aging, and geological subsidence, leading to frequent leaks. Currently, infrared thermal imaging technology, with its advantages of being non-contact, having a fast response time, and capable of large-area scanning, has become the primary method for detecting leaks in heating networks. Routine inspections typically utilize drones or ground inspection vehicles equipped with infrared thermal imagers to scan the network's coverage area, acquiring images of surface heat distribution. Then, through manual or automated algorithms, abnormal areas with significantly higher surface temperatures than the background environment are identified, allowing for the inference of underground pipeline leak conditions.

[0003] First, existing methods primarily rely on two-dimensional planar thermal images for analysis, lacking the ability to perceive the three-dimensional geometry of the Earth's surface. This makes it difficult to correct projection distortions caused by tilted shooting angles or ground undulations, resulting in positional deviations when thermal anomaly areas are mapped to the geographic coordinate system. Second, traditional anomaly detection logic is mostly based on a single temperature difference threshold, failing to effectively distinguish the physical source of thermal anomalies. In real-world environments, temperature rise on sun-facing surfaces due to solar radiation, specular reflection artifacts from highly reflective surface materials, and heat accumulation caused by terrain slope can all easily create interference features similar to leaks on thermal images, leading to frequent false alarms and severely impacting the accuracy of detection results.

[0004] In large-scale inspections, lower resolution can easily miss early, weak leak signals, while high-precision sampling across the entire area can lead to data redundancy and inefficiency. More critically, current detection methods typically only provide two-dimensional surface coordinates of suspected leak points, failing to effectively infer the specific burial depth of underground leak sources using the gradient distribution characteristics of the surface temperature field. This lack of quantitative depth information results in a lack of precise basis for subsequent excavation and maintenance, often requiring large-scale exploratory excavations based on experience, increasing maintenance costs and construction risks. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an integrated system for high-precision detection and intelligent diagnosis of leaks in heating pipe networks. This system solves the problems of existing technologies, such as the separation of thermal imaging data and spatial geometric information in complex surface environments, susceptibility to environmental interference leading to false alarms, difficulty in inverting the depth of leak sources, and the difficulty in balancing inspection efficiency and resolution.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution: a high-precision detection and intelligent diagnosis system for heating pipeline leaks, including an air-ground collaborative inspection platform, a multimodal data spatiotemporal alignment module, a three-dimensional apparent thermal manifold construction module, a heterogeneous vector field analysis module, a thermal geometry adaptive reconstruction module, a multi-viewpoint radiation verification module, and a reverse tracking and positioning module.

[0007] The air-ground collaborative inspection platform is equipped with an RGB-D depth camera, an infrared thermal imager, and a positioning and navigation unit. It is used to scan and inspect the heating pipeline network coverage area, and collect and output visible light images, depth images, and infrared thermal image data.

[0008] The multimodal data spatiotemporal alignment module connects to the aforementioned platform and is responsible for handling the temporal and spatial synchronization of heterogeneous sensor data. Based on preset extrinsic calibration parameters, this module performs timestamp synchronization and spatial coordinate mapping on the data from the RGB-D depth camera and the infrared thermal imager, establishing a rigid transformation relationship between the two coordinate systems. Specifically, the intrinsic parameter matrix of the RGB-D depth camera and the pinhole camera's inverse projection model are used to restore the depth image pixels to three-dimensional spatial coordinates. Then, using extrinsic rotation matrices and translation vectors, these coordinates are transformed to the infrared thermal imager coordinate system and reprojected onto the imaging plane. Subsequently, the system uses a bilinear interpolation algorithm to calculate the grayscale values ​​of the projected points and combines this with radiometric calibration curves to deduce the physical temperature, generating a point data set with both accurate geometric structure and thermophysical properties.

[0009] The 3D apparent thermal manifold construction module utilizes a visual SLAM algorithm to process continuous keyframes, generating a 3D geometric mesh model. Aligned temperature data is then mapped onto the model surface, thus constructing a 3D apparent thermal manifold space that integrates geometric structure and temperature distribution. During construction, the module performs surface reconstruction on the global point cloud map, generating a continuous triangular mesh. A weighted fusion algorithm determines the temperature attributes of the mesh vertices, where the weights are calculated considering both the observation viewpoint and distance, prioritizing data from vertical observations and closer proximity. Finally, barycentric interpolation assigns temperature values ​​to the interior of the triangular facets, establishing a continuous temperature scalar field attached to the geometric surface.

[0010] The heterogeneous vector field analysis module calculates the geometric normal vector field and thermal gradient vector field on the 3D apparent thermal manifold surface, and filters suspected leakage areas by analyzing their coupling characteristics. This module calculates the unit geometric normal vector for each triangular facet to characterize the surface orientation, and uses the finite element method to calculate the gradient of the temperature scalar field on the tangent plane, obtaining the thermal gradient vector reflecting the direction and intensity of temperature changes. By calculating the discrete Laplace divergence of the thermal gradient field at the grid vertices, the system first identifies high-divergence connected regions with divergence values ​​exceeding a threshold. To eliminate terrain interference, the module further calculates the cosine similarity between the thermal gradient vector and the direction of maximum slope (i.e., the opposite direction of the geometric normal vector projected onto the horizontal plane); if the thermal gradient direction of most facets is highly correlated with the slope direction, it is identified as an environmental thermal anomaly and eliminated; conversely, if the thermal gradient is radially distributed and decoupled from the local terrain, it is marked as a suspected area caused by underground leakage.

[0011] The thermal geometry adaptive reconstruction module is responsible for establishing a "perception-analysis" closed loop, dynamically adjusting the sampling strategy by calculating the thermal geometric saliency index of suspected areas. The saliency index is derived by weighting the thermal gradient vector magnitude (characterizing the intensity of the thermal field) and the discrete average geometric curvature (characterizing the complexity of the terrain). When the average saliency index of an area exceeds a threshold, the system schedules the inspection platform to perform close-range high-density sampling. Simultaneously, the module performs local recursive subdivision of the 3D mesh of suspected areas: detecting the projected area of ​​triangular facets at the optimal observation distance; if it exceeds the texture mapping threshold, the facets are split and new vertices are inserted; the new vertices are corrected in position by projecting onto a high-resolution depth map. This process is repeated recursively until the mesh density matches the high-resolution thermal image, and the thermal gradient vector field is recalculated on the subdivided high-precision mesh, achieving super-resolution reconstruction of the data.

[0012] The multi-viewpoint radiation verification module plans multi-angle observation paths around the suspected area center, acquires temperature measurement sequences, calculates variance, and uses the bidirectional reflectance distribution function to eliminate false heat sources. Specifically, the module calculates the cosine of the angle between the effective observation line of sight and the surface normal vector, and calculates the coefficient of variation of the temperature sequence. If the coefficient of variation is below a threshold, it indicates that the target has isotropic radiation characteristics and is identified as a real leak point; if the coefficient of variation is high, and the temperature observation value shows a strong correlation with the cosine of the incident angle (i.e., the Pearson correlation coefficient exceeds the threshold), it indicates that there is obvious directional reflection in the area and is identified as a false anomaly caused by highly reflective surface materials.

[0013] The reverse tracking and positioning module constructs an underground detection space beneath the surface and retrieves leakage source parameters using a reverse heat conduction model. The module first extracts the lateral surface temperature distribution curve perpendicular to the thermal ridge and establishes a steady-state heat conduction model of the underground linear heat source in a semi-infinite soil medium based on the Green's function method. The steady-state heat conduction model describes the surface temperature distribution as the sum of the ambient temperature and a temperature rise term including parameters such as burial depth and lateral distance. The system uses a nonlinear least squares method to construct the target loss function and solves for the burial depth value that minimizes the fitting error between the theoretical model and the observed data through an iterative optimization algorithm. Furthermore, the module introduces a medium correction coefficient determined based on surface texture classification (such as soil moisture content and cover type) to physically correct the estimated burial depth. Finally, combining the surface geometric center coordinates, it outputs quantitative detection results including three-dimensional coordinates, burial depth, and confidence level.

[0014] This invention provides an integrated system for high-precision detection of leaks and intelligent diagnosis of anomalies in heating pipe networks. It offers the following advantages:

[0015] 1. This invention constructs a three-dimensional apparent thermal manifold space, accurately mapping spatiotemporally aligned infrared temperature data onto a three-dimensional geometric mesh surface, establishing a continuous temperature scalar field coupled with a geometric structure model. This method overcomes the shortcomings of traditional two-dimensional infrared detection, which ignores the influence of terrain undulations on thermal imaging projection, eliminates spatial positioning errors caused by viewing angle differences, achieves accurate correspondence between temperature anomaly areas and physical geographic coordinates, and significantly improves positioning accuracy in complex terrain environments.

[0016] 2. This invention utilizes a heterogeneous vector field analysis and multi-viewpoint radiation verification mechanism. By calculating the divergence and directional coupling characteristics of the thermal gradient vector and the surface geometric normal vector, and combining this with the characteristics of the two-way reflection distribution function, it analyzes the variation of radiation intensity with the observation angle. This mechanism can effectively distinguish, from a physical perspective, the true temperature rise generated by underground heat sources from false thermal anomalies caused by solar radiation, terrain slope, or highly reflective surface materials, significantly reducing the false alarm rate caused by environmental interference.

[0017] 3. This invention combines adaptive thermal geometry reconstruction and reverse heat conduction tracing technology, enabling automatic scheduling of inspection resources for localized high-density sampling based on saliency indicators, and parameter inversion of the surface temperature profile based on a steady-state heat conduction model. This design ensures high efficiency across a large area while achieving refined perception of suspected areas, and can further calculate the underground depth and three-dimensional spatial location of the leak point, providing quantitative depth reference information for subsequent excavation and repair. Attached Figure Description

[0018] Figure 1 This is a system framework diagram of the present invention;

[0019] Figure 2This is a schematic diagram of the multimodal data acquisition and spatiotemporal alignment process of the present invention;

[0020] Figure 3 This is a schematic diagram illustrating the principle of three-dimensional apparent thermal manifold space construction and scalar field mapping of the present invention.

[0021] Figure 4 This is a schematic diagram illustrating the principle of heterogeneous vector field analysis and anomaly region screening in this invention.

[0022] Figure 5 This is a flowchart of the saliency-based adaptive resolution reconstruction and dynamic sampling process of the present invention;

[0023] Figure 6 This is a flowchart of the multi-viewpoint radiation consistency verification and authenticity anomaly identification of the present invention;

[0024] Figure 7 This is a flowchart of the leakage source depth inversion based on reverse thermal flow ray tracing of the present invention.

[0025] Among them, 100 is the air-ground collaborative inspection platform; 200 is the multimodal data spatiotemporal alignment module; 300 is the three-dimensional visual thermal manifold construction module; 400 is the heterogeneous vector field analytical module; 500 is the thermal geometry adaptive reconstruction module; 600 is the multi-viewpoint radiation consistency verification module; and 700 is the reverse tracking and positioning module. Detailed Implementation

[0026] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] See attached document Figure 1 , Figure 1 This is a structural block diagram of an integrated system for high-precision detection and intelligent diagnosis of leaks in heating pipe networks according to an embodiment of the present invention. The present invention provides an integrated system for high-precision detection and intelligent diagnosis of leaks in heating pipe networks, comprising:

[0028] The air-ground collaborative inspection platform 100 is equipped with an RGB-D depth camera, an infrared thermal imager, and a positioning and navigation unit. It is used to perform scanning and inspection tasks in the heating pipeline network coverage area and output visible light images, depth images, and infrared thermal image data.

[0029] The multimodal data spatiotemporal alignment module 200 is configured to perform time stamp synchronization and spatial coordinate mapping on the data collected by the RGB-D depth camera and the infrared thermal imager based on preset extrinsic calibration parameters, establish a rigid transformation relationship between the depth camera coordinate system and the infrared thermal imager coordinate system, and output spatiotemporally aligned multimodal data frames.

[0030] The 3D apparent thermal manifold construction module 300 is configured to process continuous keyframe sequences based on the visual SLAM algorithm to generate a 3D geometric mesh model, and map the aligned temperature data as a scalar field texture onto the surface of the 3D geometric mesh model to construct a 3D apparent thermal manifold space containing geometric structure and temperature distribution.

[0031] The heterogeneous vector field analysis module 400 is configured to calculate the geometric normal vector field and the thermal gradient vector field on the surface of the three-dimensional apparent thermal manifold space, calculate the divergence characteristics of the thermal gradient vector field, and screen suspected leakage areas based on the divergence threshold and the coupling relationship between the thermal gradient and the geometric normal vector.

[0032] The thermal geometry adaptive reconstruction module 500 is configured to calculate the thermal geometry significance index of the suspected leak area, schedule the air-ground collaborative inspection platform 100 to perform close-range high-density sampling based on the significance index, and perform local recursive subdivision of the three-dimensional mesh of the suspected leak area to update the geometric normal vector and thermal gradient data.

[0033] The multi-viewpoint radiation verification module 600 is configured to plan a multi-angle observation path around the center of the suspected leak area, obtain temperature measurement sequences at different observation angles, calculate the variance of the observed temperature sequences, and determine whether the suspected leak area is a real underground heat source based on the bidirectional reflection distribution function characteristics.

[0034] The reverse tracking and positioning module 700 is configured to construct an underground three-dimensional voxel space below the surface, project virtual heat flow rays into the underground along the opposite direction of the surface thermal gradient, and determine the three-dimensional coordinates and burial depth of the leak point based on the intersection density of the rays in the voxel space.

[0035] To more clearly illustrate the technical solutions of the embodiments of the present invention, the modules and their specific working principles in the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0036] See attached document Figure 2 , Figure 2 A schematic diagram of the multimodal data acquisition and spatiotemporal alignment process according to an embodiment of the present invention is shown. The air-ground collaborative inspection platform 100 and the multimodal data spatiotemporal alignment module 200 work together through the following steps to provide basic data with accurate geometric and physical property correspondence for subsequent three-dimensional thermal manifold construction.

[0037] The air-ground collaborative inspection platform 100 includes an industrial-grade drone as the flight platform and an inspection robot as the ground mobile platform. The data acquisition payload comprises a physically rigidly connected RGB-D depth camera and a high-resolution infrared thermal imager. The RGB-D depth camera is configured to output visible light and depth images of the scene, while the infrared thermal imager is configured to output long-wave infrared radiation images from the same viewpoint. During physical installation, the two sensors maintain parallel optical axes or a fixed angle. The relative position and rotation angle between their optical centers have been precisely calibrated before system operation, forming a fixed extrinsic parameter rotation matrix. Translation vector .

[0038] The specific processing procedure executed by the multimodal data spatiotemporal alignment module 200 is as follows:

[0039] Step S201: Time synchronization of multimodal data streams. The multimodal data spatiotemporal alignment module 200 receives depth image frames output from the RGB-D depth camera. thermal image frames output by infrared thermal imager To address the differences in data link transmission delays among different sensors, the multimodal data spatiotemporal alignment module 200 employs a combined hardware and software timestamp alignment strategy. At the hardware level, a unified hardware pulse signal triggers the shutters of each sensor; at the software level, the timestamps of depth image frames are used... Based on this, search for timestamps in the thermal image cache queue. Select the absolute value of the time difference The smallest frame that is similar to the preset synchronization threshold is used as the matching frame to establish a one-to-one correspondence in the time dimension.

[0040] Step S202: Spatial point cloud reconstruction based on the pinhole model. For the synchronized depth image The multimodal data spatiotemporal alignment module 200 first performs distortion correction on the image using pre-stored distortion coefficients to eliminate geometric distortion caused by lens distortion. Then, the multimodal data spatiotemporal alignment module 200 iterates through the effective pixels in the image. Based on the internal parameters of the RGB-D depth camera, and using the pinhole camera inverse projection model, the two-dimensional pixel coordinates are restored to three-dimensional spatial coordinates in the depth camera coordinate system. The calculation process follows the equation below:

[0041] ;

[0042] in, For pixels The corresponding depth measurement value, This is the intrinsic parameter matrix of the RGB-D depth camera. For pixels The homogeneous coordinate vector.

[0043] Step S203: Rigid coordinate transformation and reprojection between heterogeneous sensors. To assign temperature information to the aforementioned three-dimensional spatial coordinates, the multimodal data spatiotemporal alignment module 200 uses pre-calibrated extrinsic parameter data to transform the points in the depth camera coordinate system... Transform to the coordinate system of the infrared thermal imager and project onto the imaging plane of the thermal imager. Multimodal data spatiotemporal alignment module: 200 calculation points. In thermal images Projected pixel coordinates :

[0044] ;

[0045] in, This is the intrinsic parameter matrix of the infrared thermal imager. Let be the homogeneous coordinates of the projection point. As a scale factor, To calculate the three-dimensional position of the point in the infrared thermal imager coordinate system.

[0046] Step S204: Interpolation and temperature inversion of thermal radiation data. Due to the calculated projected coordinates... For non-integer values, the multimodal data spatiotemporal alignment module 200 employs a bilinear interpolation algorithm to calculate the interpolated grayscale value at a given location based on the grayscale values ​​of the four neighboring pixels surrounding the projected coordinates. Subsequently, the multimodal data spatiotemporal alignment module 200 converts the interpolated grayscale value into a physical temperature value using the radiometric calibration curve of the infrared thermal imager or a lookup table (LUT). (Celsius or Kelvin). This calibration process takes into account corrections for ambient temperature, emissivity, and atmospheric transmittance parameters to ensure that the obtained temperature value reflects the true radiative temperature of the object's surface.

[0047] Step S205: Generate aligned multimodal data frames. After the above processing, the system output contains spatial geometric information. Corresponding physical temperature attribute This dataset is a collection of point data. It eliminates parallax caused by sensor installation position deviations, ensuring that the geometry and thermal features of each 3D spatial point originate from the same physical location. For points whose calculation results exceed the infrared thermal imager's field of view or whose depth values ​​are invalid, the multimodal data spatiotemporal alignment module 200 marks them as invalid data and removes them in subsequent processing.

[0048] See attached document Figure 3 , Figure 3A schematic diagram illustrating the principle of three-dimensional apparent thermal manifold space construction and scalar field mapping according to an embodiment of the present invention is shown. The three-dimensional apparent thermal manifold construction module 300 receives the spatiotemporally aligned multimodal data frame output from the previous stage, constructs a geometric surface with a continuous topological structure through the following steps, and maps discrete temperature measurements to continuous scalar field functions on this geometric surface to establish a three-dimensional apparent thermal manifold space, including the following steps:

[0049] Step S301: Global pose optimization and fusion of multi-frame point clouds. The 3D visual thermal manifold construction module 300 uses a visual SLAM algorithm to process the continuously input keyframe sequence. For each frame of data, the 3D visual thermal manifold construction module 300 calculates the current camera position in the global world coordinate system through feature point matching. pose matrix below Based on this pose matrix, the 3D thermal manifold construction module 300 transforms the color point cloud and temperature data in each frame's local coordinate system to the global world coordinate system. The 3D thermal manifold construction module 300 then overlays the transformed point cloud data from multiple frames and performs voxel filtering to remove overlapping and redundant points, forming a dense global point cloud map. During this process, the 3D thermal manifold construction module 300 minimizes reprojection errors and eliminates accumulated drift by detecting loop closures and performing global bundle adjustment, ensuring the consistency of global geometry. The specific feature extraction and nonlinear optimization processes in the SLAM algorithm are standard techniques in the field of computer vision and will not be elaborated upon here.

[0050] Step S302: Generation of the 3D geometric manifold surface. The 3D apparent thermal manifold construction module 300 performs surface reconstruction on the global point cloud map to generate a continuous triangular mesh model. :

[0051] ;

[0052] in, Represents the set of grid vertices. This represents the set of triangular faces that connect vertices. The total number of grid vertices. This represents the total number of triangular facets. The 3D visual thermal manifold construction module 300 uses a Poisson surface reconstruction algorithm or a truncated signed distance function fusion algorithm to calculate the implicit surface function of the point cloud and extract isosurfaces to generate a mesh. This process can automatically fill local holes caused by viewpoint occlusion and smooth outliers, ensuring that the generated geometric surface is topologically closed or locally continuous. Each vertex All contain defined three-dimensional spatial coordinates .

[0053] Step S303: Thermal Scalar Field Mapping Based on Multi-Observation Fusion. Since the same spatial point may be observed by multiple data frames from different perspectives and at different times, the 3D apparent thermal manifold construction module 300 determines the mesh vertices through a weighted fusion algorithm. final temperature property The 3D visual thermal manifold construction module 300 first performs projection data association:

[0054] For any vertex in the grid Using the world-camera transformation matrix of each keyframe The camera intrinsic parameters are then projected back onto the image plane of each keyframe. If the projected point falls within the image boundary and its depth value matches the depth of the Zbuffer buffer, it is determined that the frame is not occluded. The 3D visual thermal manifold construction module 300 then determines that the keyframe is a valid observation frame and adds it to the set. .

[0055] For sets Each observation frame in The 3D visual thermal manifold construction module 300 extracts the observed temperature value of the corresponding pixel. And calculate the observation confidence weights. Confidence weight The calculation comprehensively considers the influence of observation angle and observation distance, and follows the following fusion equation:

[0056] ;

[0057] in, For set Index number of the observed frame, weight Defined as ;

[0058] In the formula, As vertex The unit geometric normal vector at that location, To point from vertex to the first The unit direction vector of the optical center of the frame camera (the weight is reset to 0 when the dot product is less than 0). Let be the Euclidean distance from the vertex to the camera. To control the attenuation coefficient of distance sensitivity, the reliability weight calculation logic is based on the physical characteristics of infrared thermometry: the closer the observation angle is to the normal direction and the closer the observation distance, the more accurate the received radiation energy, and therefore a higher fusion weight is assigned.

[0059] Step S304: Smoothing and manifold construction of the thermal scalar field are completed. After the above fusion calculation, the three-dimensional apparent thermal manifold construction module 300 is a mesh model. each vertex It was assigned a unique temperature scalar value. For triangular facets any point inside Its temperature value The temperature is obtained by re-interpolating the temperature values ​​at the three vertices of the facet. Based on this, the system establishes a continuously distributed temperature scalar field on the surface of the 3D geometric mesh model, realizing a one-to-one mapping between temperature values ​​and surface points in 3D space, thereby constructing a 3D apparent thermal manifold that integrates geometric structure and thermophysical properties. This manifold space unifies the geometric structure data and thermodynamic distribution data of the heating pipe network cover surface in the same mathematical model, supporting subsequent gradient calculations and divergence analysis directly on the surface.

[0060] See attached document Figure 4 , Figure 4 This is a schematic diagram illustrating the principle of heterogeneous vector field analysis and initial screening of abnormal regions according to an embodiment of the present invention. The heterogeneous vector field analysis module 400 receives the three-dimensional apparent thermal manifold data constructed in the previous stage, simultaneously calculates the geometric and thermal field characteristics on the manifold surface using discrete differential geometry methods, and screens suspected leakage regions by analyzing the coupling relationship between the two, including the following steps:

[0061] Step S401: Construct the geometric normal vector field. The heterogeneous vector field analysis module traverses the 3D apparent thermal manifold mesh model 400 times. Each triangular facet in For vertices The heterogeneous vector field analysis module 400 obtains the unit geometric normal vector of a triangular facet formed in a counter-clockwise order by calculating and normalizing the cross product of two edge vectors. The geometric normal vector represents the spatial orientation of the Earth's surface within this local micro-element. The set of normal vectors for all patches is defined as the geometric normal field, which describes the topographic relief structure of the monitored area.

[0062] Step S402: Calculate the thermal gradient vector field of the manifold surface. The heterogeneous vector field analysis module 400 is based on the finite element method, calculating the thermal gradient vector field for each triangular facet. Calculate the temperature scalar field on the tangent plane. The gradient of temperature. This gradient is a tangent vector defined on the surface of a non-Euclidean manifold, representing the direction and intensity of the most dramatic temperature change along the surface. Thermal gradient vector. The calculation formula is as follows:

[0063] ;

[0064] in, The representation is defined on the manifold surface. Discrete gradient operator on;

[0065] Triangular facet The area;

[0066] The index of the vertex of the face;

[0067] For the first Temperature scalar values ​​of each vertex;

[0068] For the first The opposite edge vectors of each vertex (e.g., vertex) The corresponding opposite edge is ;

[0069] This is the unit geometric normal vector calculated in step S401. This formula ensures the calculated thermal gradient vector. It is located strictly within the tangent plane of the grid surface.

[0070] Step S403: Calculate the discrete divergence characteristics of the thermal gradient field. To identify thermal anomalies with diffusion source characteristics, the heterogeneous vector field analysis module 400 calculates the divergence of the thermal gradient field at each grid vertex. The thermal field formed on the ground surface by heating network leaks typically exhibits a radial distribution with a high center and low periphery, corresponding to the positive divergence extrema of the gradient field. The heterogeneous vector field analysis module 400 uses the discrete Laplace operator to calculate the vertices. Discrete divergence at point :

[0071] ;

[0072] in, As vertex The corresponding Voronoi control area;

[0073] For the vertex The set of adjacent facets;

[0074] Geometric weighting elements, For the edge The corresponding diagonal angle;

[0075] From the vertex Edge vectors pointing to adjacent vertices.

[0076] Step S404: Initial anomaly screening based on heterogeneous field coupling characteristics. The heterogeneous vector field analysis module 400 performs logical discrimination based on the heterogeneous field characteristics of thermal gradient and geometric normal vector to distinguish between real underground heat sources and surface environmental disturbances. The heterogeneous vector field analysis module 400 first filters out all divergence values. Greater than the preset positive threshold The connected regions. For each selected high-divergence region, the heterogeneous vector field analysis module 400 calculates the thermal gradient direction consistency index within the region. .

[0077] Specifically, the heterogeneous vector field analysis module 400 calculates the thermal gradient vectors of all patches within the calculation region. The vector of the maximum slope direction of the surface The dot product cosine similarity (i.e., the opposite direction of the geometric normal vector projected onto the horizontal plane). If the thermal gradient direction of more than a preset proportion (e.g., 80%) of the areas within the region is highly positively correlated with the direction of the maximum slope (cosine similarity greater than 0.9), then the thermal anomaly is determined to be environmental heat caused by the terrain slope (e.g., heat on a sunny slope) and is removed. Conversely, if the thermal gradient direction exhibits a radial distribution and its correlation with the local terrain slope direction is lower than a preset threshold, i.e., it shows a decoupling between thermal distribution and geometric structure, then it is determined to be a real anomaly caused by underground leakage.

[0078] Step S405: Output the set of suspected leakage regions. After the above screening steps, the heterogeneous vector field analysis module 400 marks the mesh connectivity regions that meet the high divergence condition and pass the geometric decoupling verification as suspected leakage regions, and extracts the geometric center coordinates and coverage radius of the region as the input target for the next stage of adaptive reconstruction.

[0079] See attached document Figure 5 , Figure 5 A flowchart illustrating the saliency-based adaptive resolution reconstruction and dynamic sampling process according to an embodiment of the present invention is shown. The thermal geometry adaptive reconstruction module 500, for the suspected leakage areas screened in the previous stage, establishes a closed-loop feedback mechanism between the calculation results and the acquisition hardware to achieve adaptive switching from coarse-grained scanning to fine-grained sensing. The processing flow executed by the thermal geometry adaptive reconstruction module 500 specifically includes the following steps:

[0080] Step S501: Calculation of thermal-geometric saliency index. The thermal geometry adaptive reconstruction module 500 quantitatively evaluates the information richness of each grid cell within the suspected leak area and defines the thermal-geometric saliency index. This thermal-geometric significance index characterizes whether the current region requires higher resolution data support. For each vertex within the suspected region... The thermal geometry adaptive reconstruction module 500 calculates the significance value according to the following formula:

[0081] ;

[0082] in, The magnitude of the thermal gradient vector calculated in the previous step characterizes the degree of drastic change in the thermal field; and These are the preset normalized weighting coefficients.

[0083] As vertex The discrete average geometric curvature at a given location is used to characterize the local complexity of the terrain surface's unevenness; its calculation formula is as follows:

[0084] ;

[0085] In the formula, As vertex Area of ​​the Voronoi control region on the grid surface;

[0086] For the vertex The set of directly adjacent vertices in a ring neighborhood;

[0087] The vertex index within the neighborhood;

[0088] To connect vertices With neighboring vertices The edge vector, The magnitude (i.e., side length) of the edge vector.

[0089] and They are the edges The angle values ​​of the two vertices opposite the edge in two adjacent triangular facets shared by the region. This formula shows that the more steep the thermal gradient or the greater the curvature of the surface geometry, the more significant the effect.

[0090] Step S502: Active perception path planning and sampling based on saliency. When the average saliency index of the region... When the preset segmentation threshold is exceeded, the thermal geometry adaptive reconstruction module 500 triggers an active sensing mechanism. The thermal geometry adaptive reconstruction module 500 then determines the location of the geometric center of the suspected region. and the region average normal vector Calculate the optimal hovering point for the air-ground collaborative inspection platform 100. :

[0091] ;

[0092] in, The three-dimensional coordinates of the geometric center of the suspected leak area;

[0093] It is the average unit vector of the normal vectors of all patches in the region, representing the main orientation of the region;

[0094] The optimal observation distance is calculated based on the infrared thermal imager's field of view (FOV) and the required spatial resolution (e.g., 5 meters for millimeter-level resolution). The thermal geometry adaptive reconstruction module 500 will... The coordinate commands are sent to the flight control unit, which drives the inspection platform to adjust its attitude and approach the location. In the approach and hovering state, the sensors acquire high-resolution infrared and depth image sequences of the local area at a high frame rate, achieving a physical-level increase in data sampling density.

[0095] Step S503: Local recursive subdivision of the mesh model. In conjunction with the density enhancement of physical sampling, the thermal geometry adaptive reconstruction module 500 performs super-resolution reconstruction of the 3D mesh model at the algorithmic level. The thermal geometry adaptive reconstruction module 500 detects each triangular facet within the suspected region. Calculate its optimal observation distance The projected area of ​​the screen is calculated. If the projected pixel area of ​​the facet is greater than the set single texture mapping threshold, it indicates that the mesh geometry accuracy is lower than the image sampling accuracy. The thermal geometry adaptive reconstruction module 500 executes the Loop subdivision or 1-4 facet splitting algorithm: in the triangular facet... A new vertex is inserted on each of the three edges, splitting the original patch into four smaller patches. The coordinates of the newly inserted vertices are not simply interpolated linearly, but rather corrected for position by projecting the new vertices onto the nearest high-resolution depth map surface to restore realistic micro-topographical geometry. This process is performed recursively until the mesh density matches the pixel density of the high-resolution thermal image.

[0096] Step S504: High-fidelity thermal manifold update and gradient recalculation. After mesh subdivision, the thermal geometry adaptive reconstruction module 500 uses the high-resolution image data acquired in step S502 to remap the temperature texture of all subdivided vertices. Due to the improved mesh resolution, the thermal geometry adaptive reconstruction module 500 can accurately map the minute temperature difference details in the thermal image onto micro-topographic features. Subsequently, the thermal geometry adaptive reconstruction module 500 re-executes the thermal gradient vector field calculation on the subdivided local high-precision mesh. The reconstructed thermal gradient field has a higher signal-to-noise ratio and spatial resolution, which can accurately delineate the subtle thermal textures formed by underground heat flow on the surface, providing a sub-meter-level data foundation for subsequent precise positioning.

[0097] See attached document Figure 6 , Figure 6A flowchart illustrating multi-viewpoint radiation consistency verification and anomaly identification according to an embodiment of the present invention is shown. The multi-viewpoint radiation consistency verification module 600, for suspected leak areas output in previous steps, utilizes multi-angle observation data acquired during air-ground collaborative inspections, and based on the physical characteristics of the bidirectional reflectance distribution function (BRDF), eliminates false anomalies caused by solar reflection or environmental stray radiation, retaining genuine underground heat sources with isotropic radiation characteristics. The multi-viewpoint radiation consistency verification module 600 specifically performs the following steps:

[0098] Step S601: Construction and occlusion removal of the multi-angle observation set. The multi-viewpoint radiation consistency verification module 600 traverses each surface point to be verified within the suspected leak area. The multi-viewpoint radiometric consistency verification module 600 searches the spatiotemporal index database to obtain a set of keyframes covering all visible points on the surface to be verified. To ensure the validity of the observation data, the multi-viewpoint radiometric consistency verification module 600 performs a depth buffer test (Z-buffer-Test) or ray projection detection: for any candidate keyframe... Constructing from the camera optical center to surface point The gaze vector. If the gaze vector does not intersect with other geometric meshes in the scene, then the keyframe is determined to be a point. Unobstructed, it is included in the effective observation set. For each observation frame in the set, the multi-viewpoint radiometric consistency verification module 600 reads the corresponding temperature observation value through reprojection. .

[0099] Step S602: Calculation of the correlation between the observation angle and the surface normal. Real heat leakage sources are typically approximated as Lambertian radiators, whose radiation intensity varies little with the observation angle; while sunlight reflection or glass curtain wall reflection usually exhibits strong directionality. To distinguish between these two physical properties, the multi-viewpoint radiation consistency verification module 600 calculates the cosine of the incident angle for each valid observation. For point... and its related first The calculation formula is as follows: (Number of observation frames)

[0100] ;

[0101] in, For surface points The unit geometric normal vector at that location;

[0102] For the first The camera optical center coordinates of each keyframe;

[0103] These are the three-dimensional coordinates of a point on the surface.

[0104] This constitutes the observation line-of-sight vector;

[0105] To observe the angle between the line of sight and the surface normal;

[0106] It represents the magnitude of the vector.

[0107] Step S603: Calculation of the coefficient of variation of radiation observations. Multi-viewpoint radiation consistency verification module 600 statistically analyzes the valid observation set. The temperature data distribution is analyzed to calculate the dispersion of radiation observations. To eliminate the influence of the baseline temperature magnitude, the multi-viewpoint radiation consistency verification module 600 uses the coefficient of variation. As a quantitative indicator for measuring radiation consistency:

[0108] ;

[0109] in, To effectively observe the total number of keyframes;

[0110] For the first Keyframes Temperature observations;

[0111] This represents the average temperature at that point across all viewing angles. The coefficient of variation is defined by this formula. It characterizes the sensitivity of temperature observations to the observation angle.

[0112] Step S604: Authenticity and Anomaly Identification Based on BRDF Characteristics. The multi-viewpoint radiation consistency verification module 600 identifies true and false anomalies based on the calculated coefficient of variation. Execute logical judgment. If the coefficient of variation... Less than the preset consistency threshold (For example, 0.05) indicates that the radiation energy in the anomalous area remains stable under different observation angles, consistent with the isotropic radiation characteristics (i.e., Lambertian characteristics) of an underground heat source conducted through the soil medium. The multi-viewpoint radiation consistency verification module 600 then identifies it as a true leak point. Conversely, if the coefficient of variation is... Greater than the threshold The multi-viewpoint radiation consistency verification module 600 further calculates the temperature observation sequence. With the cosine sequence of the incident angle Pearson correlation coefficient between .like If the radiation intensity is greater than the preset correlation threshold (e.g., 0.7), it indicates that the radiation intensity in the area fluctuates regularly with the change of the observation viewpoint, which is consistent with the directional reflection characteristics of highly reflective surface materials (such as metal manhole covers and waterlogged roads). The multi-viewpoint radiation consistency verification module 600 judges it as a false anomaly and removes it from the suspected list.

[0113] Step S605: Output the final leak detection result. After the above verification, the multi-viewpoint radiation consistency verification module 600 maps the retained real abnormal areas back to the global 3D map, generating a final detection report containing the precise 3D coordinates of the leak point, coverage area, and confidence score. This process realizes the correction of the limitations of single-frame thermal infrared images by utilizing multi-viewpoint redundant information.

[0114] See attached document Figure 7 , Figure 7 A flowchart illustrating the leakage source depth inversion process based on reverse heat flow ray tracing according to an embodiment of the present invention is shown. The leakage source depth inversion module 700, for the actual leakage area determined through multi-viewpoint verification, utilizes the spatial attenuation characteristics of the surface temperature field to inversely deduce the burial depth and intensity of the underground heat source through physical modeling. This process mathematically reverses the forward diffusion process of heat conduction, specifically executing the following steps:

[0115] Step S701: Constructing the thermal anomaly profile coordinate system and extracting features. The leak source depth inversion module 700 first determines the thermal ridge line of the suspected leak area, i.e., the line connecting the maximum temperatures along the pipeline network. The leak source depth inversion module 700 selects several sampling points on the thermal ridge line and constructs a local profile coordinate system perpendicular to the tangent direction of the thermal ridge line for each sampling point. On this profile, the origin is the projection point of the thermal ridge line. Extracting surface temperature as a function of lateral distance Distribution curve of change To eliminate the influence of surface noise, the leak source depth inversion module 700 performs Gaussian smoothing or B-spline curve fitting on the extracted discrete temperature data points to obtain a continuous temperature distribution profile.

[0116] Step S702: Establish a semi-infinite medium heat conduction inversion model. The leakage source depth inversion module 700, based on the Green's function method, establishes a steady-state heat conduction model of a subsurface linear heat source in a semi-infinite soil medium. The steady-state heat conduction model assumes that heat originates from the subsurface depth... A linear source diffuses towards the Earth's surface, creating a temperature field that follows a Lorentz distribution. The leakage source depth inversion module 700 defines the theoretical surface temperature distribution function. as follows:

[0117] ;

[0118] in, Horizontal distance from the center of the heat source The theoretical temperature value at that location;

[0119] The background ambient temperature, which is far from the thermal anomaly area, is obtained by the leakage source depth inversion module 700 automatically sampling the edge temperature of the area;

[0120] This represents the peak temperature at the thermal ridge.

[0121] The depth of the leak source to be inverted;

[0122] Let be the lateral distance variable on the cross-section. This formula establishes the relationship between the lateral temperature difference distribution at the Earth's surface and the underground burial depth. The nonlinear geometric mapping relationship between them.

[0123] Step S703: Inverse parameter optimization based on least squares method. The leakage source depth inversion module 700 uses nonlinear least squares method to construct the target loss function. The goal is to find the optimal burial depth that minimizes the fitting error between the theoretical model and the observed data. The formula for calculating the target loss function is as follows:

[0124] ;

[0125] in, The number of effective sampling points on the cross-section;

[0126] For the first The lateral distance coordinates of each sampling point;

[0127] For the first The actual observed temperature at each sampling point;

[0128] This represents the peak temperature rise. The leakage source depth inversion module 700 uses the Levenberg-Marquardt iterative algorithm to solve the above objective function, updating it iteratively. The value is calculated until the loss function converges or the preset number of iterations is reached, thereby obtaining the optimal estimated burial depth corresponding to that profile. .

[0129] Step S704: Multi-parameter coupling correction and confidence assessment. Considering the influence of soil moisture content and surface cover type on thermal conductivity, the leakage source depth inversion module 700 introduces a medium correction coefficient. The estimated burial depth is physically corrected, and the final burial depth is calculated. :

[0130] ;

[0131] Among them, the correction coefficient The material parameters are determined by the system's preset material parameter library based on the visible light texture features obtained in step S203.

[0132] Specifically, for dry soil or asphalt pavement, The value ranges from 0.9 to 1.1; for areas with moist soil or vegetation cover, the increased thermal conductivity due to moisture leads to wider heat diffusion. The value ranges from 0.7 to 0.9; for concrete pavements, The value ranges from 1.1 to 1.3. Meanwhile, the leakage source depth inversion module 700 calculates the root mean square error (RMSE) of the fitting residual as a confidence index of the inversion result. If the RMSE is greater than the preset threshold, the inversion result is marked as low confidence and manual review is prompted.

[0133] Step S705: Generate three-dimensional leakage source spatial information. The leakage source depth inversion module 700 will use the calculated burial depth... The coordinates of the geometric center of the Earth's surface obtained in step S405 By combining these factors, the estimated underground absolute coordinates of the leak source can be generated. The leak source depth inversion module 700 encapsulates the three-dimensional coordinates, leak intensity level, and confidence level information into a standardized abnormal event data packet, which is then output to the user interactive terminal or geographic information system (GIS) to complete quantitative detection.

Claims

1. A high-precision integrated system for detecting leaks and intelligently diagnosing anomalies in heating pipe networks, characterized in that: include: The air-ground collaborative inspection platform is used to collect visible light, depth and infrared image data of the heating pipeline network area by carrying an RGB-D depth camera and an infrared thermal imager. The multimodal data spatiotemporal alignment module is connected to the air-ground collaborative inspection platform and is used to perform timestamp synchronization and spatial coordinate mapping on the collected data based on the external parameter calibration parameters to generate spatiotemporally aligned multimodal data frames. A three-dimensional apparent thermal manifold construction module is connected to the multimodal data spatiotemporal alignment module. It is used to generate a three-dimensional geometric mesh based on visual SLAM and map temperature data onto the mesh surface to construct a three-dimensional apparent thermal manifold space. The heterogeneous vector field analysis module, connected to the three-dimensional apparent thermal manifold construction module, is used to calculate the geometric normal vector field and thermal gradient vector field of the space, and to screen suspected leakage areas based on the thermal gradient divergence characteristics and the coupling relationship between the thermal gradient and the geometric normal vector. The thermal geometry adaptive reconstruction module, connected to the heterogeneous vector field analysis module, is used to schedule the platform to perform close sampling based on the thermal geometry significance index of the suspected leak area, and to recursively subdivide the local mesh to update the vector field data. The multi-viewpoint radiation verification module, connected to the thermal geometry adaptive reconstruction module, is used to acquire multi-angle temperature observation sequences of the suspected area and determine the authenticity of the leakage source based on the bidirectional reflectance distribution function characteristics and the variance of the observation sequence. The reverse tracking and positioning module, connected to the multi-viewpoint radiation verification module, is used to construct a reverse heat conduction model of the surface temperature field and the underground heat source, and to determine the three-dimensional coordinates and burial depth of the leak point by reverse optimization of the heat source parameters.

2. The integrated system for high-precision detection and intelligent diagnosis of leaks in heating pipe networks according to claim 1, characterized in that, The multimodal data spatiotemporal alignment module is configured to perform the following operations when establishing rigid transformation relationships: By using the intrinsic parameter matrix of an RGB-D depth camera and the pinhole camera inverse projection model, the pixel coordinates in the depth image are restored to three-dimensional spatial coordinates in the depth camera coordinate system. Using a pre-calibrated extrinsic rotation matrix and translation vector, the three-dimensional spatial coordinates in the depth camera coordinate system are transformed to the infrared thermal imager coordinate system and projected onto the imaging plane of the infrared thermal imager to obtain the projected pixel coordinates. The grayscale value at the coordinates of the projected pixel is calculated using a bilinear interpolation algorithm, and the grayscale value is inverted into a physical temperature value according to the radiometric calibration curve, thereby generating a set of point data containing spatial geometric information and corresponding physical temperature attributes.

3. The integrated system for high-precision detection and intelligent diagnosis of leaks in heating pipe networks according to claim 1, characterized in that, The 3D apparent thermal manifold construction module is configured to perform the following operations when constructing the 3D apparent thermal manifold space: Surface reconstruction is performed on the global point cloud map to generate a continuous triangular mesh model consisting of mesh vertices and triangular faces; The final temperature attribute of the mesh vertex is determined by a weighted fusion algorithm. The confidence weight of the weighted fusion is calculated based on the observation viewpoint and observation distance. The closer the observation viewpoint is to the normal direction of the mesh vertex and the closer the observation distance is, the higher the fusion weight is assigned. The merged temperature values ​​are assigned to the mesh vertices, and the temperature values ​​at any point inside the triangular facet are obtained through centroid interpolation, thereby establishing a continuous temperature scalar field defined on the geometric surface.

4. The integrated system for high-precision detection and intelligent diagnosis of leaks in heating pipe networks according to claim 1, characterized in that, The heterogeneous vector field analysis module is configured to perform the following operations when screening suspected leakage areas: Traverse all triangular faces in the three-dimensional apparent thermal manifold space and calculate the unit geometric normal vector representing the orientation of the local micro-element space of the Earth's surface; The gradient of the temperature scalar field is calculated on the tangent plane of each triangular facet using the finite element method, thus obtaining the thermal gradient vector characterizing the direction and intensity of temperature change along the Earth's surface. The discrete divergence of the thermal gradient field at the grid vertices is calculated using the discrete Laplacian operator. Connected regions with divergence values ​​greater than a preset positive threshold are used as high-divergence regions for subsequent coupling relationship analysis.

5. The integrated system for high-precision detection and intelligent diagnosis of leaks in heating pipe networks according to claim 1, characterized in that, The heterogeneous vector field analysis module is specifically configured to screen suspected leakage areas based on the coupling relationship between thermal gradients and geometric normals as follows: Calculate the cosine similarity of the dot product between the thermal gradient vector of all patches in the high divergence region and the maximum slope direction vector of that patch, where the maximum slope direction vector is the opposite direction of the geometric normal vector projected onto the horizontal plane; If the cosine similarity between the thermal gradient direction and the maximum slope direction of a surface area exceeding a preset proportion in the region is greater than a preset threshold, then the region is determined to be an environmental thermal anomaly caused by the terrain slope and is removed. If the thermal gradient direction shows a radial distribution and its correlation with the local terrain slope direction is lower than a preset threshold, it is determined to be a suspected leakage area caused by underground leakage.

6. The integrated system for high-precision detection and intelligent diagnosis of leaks in heating pipe networks according to claim 1, characterized in that, The thermal geometry adaptive reconstruction module is configured to calculate the thermal geometry significance index as follows: For all vertices within the suspected leak area, the magnitude of the thermal gradient vector is calculated to characterize the degree of drastic change in the thermal field; Calculate the discrete average geometric curvature at the vertices to characterize the local concavity and convexity complexity of the terrain surface; The thermal gradient vector magnitude and discrete average geometric curvature are weighted and summed to obtain the thermal geometric saliency index of the vertex. If the average saliency index of the region exceeds the preset subdivision threshold, the air-ground collaborative inspection platform is triggered to perform close-range high-density sampling.

7. The integrated system for high-precision detection and intelligent diagnosis of leaks in heating pipe networks according to claim 1, characterized in that, The thermal geometry adaptive reconstruction module is configured to perform local recursive subdivision of the 3D mesh in the suspected leak area as follows: Detect the screen projection area of ​​each triangular facet within the suspected leak area at the optimal observation distance; If the projected pixel area of ​​the face is greater than the set single texture mapping threshold, a new vertex is inserted on each of the three sides of the triangular face, splitting the original face into four tiny faces. The newly inserted vertices are projected onto the surface of the high-resolution depth map for position correction, and this process is repeated recursively until the mesh density matches the pixel density of the high-resolution thermal image. The thermal gradient vector field is then recalculated on the subdivided local high-precision mesh.

8. The integrated system for high-precision detection and intelligent diagnosis of leaks in heating pipe networks according to claim 1, characterized in that, The multi-viewpoint radiation verification module is configured to: determine whether a suspected leak area is a real underground heat source. Calculate the cosine of the incident angle for all valid observation lines, which is the cosine of the angle between the observation line vector and the surface normal vector; Calculate the coefficient of variation of the observed temperature sequence. If the coefficient of variation is less than the preset consistency threshold, it is determined to be a real leak point with isotropic radiation characteristics. If the coefficient of variation is greater than the preset consistency threshold, the Pearson correlation coefficient between the temperature observation sequence and the incident angle cosine sequence is further calculated. If the absolute value of the correlation coefficient is greater than the preset correlation threshold, it indicates that the radiation intensity fluctuates regularly with the change of the observation angle, and is judged as a false anomaly with directional reflection characteristics.

9. The integrated system for high-precision detection and intelligent diagnosis of leaks in heating pipe networks according to claim 1, characterized in that, The reverse tracking and positioning module is configured to determine the three-dimensional coordinates and burial depth of the leak point as follows: Identify the thermal ridge line of the suspected leak area, and extract the surface temperature distribution curve as a function of lateral distance on a local profile perpendicular to the tangent direction of the thermal ridge line; A steady-state heat conduction model of an underground linear heat source in a semi-infinite soil medium is established based on the Green's function method. The model defines the theoretical surface temperature distribution function as the sum of the background ambient temperature and the temperature rise term, where the temperature rise term is related to the burial depth and lateral distance. The target loss function is constructed using the nonlinear least squares method, and the optimal estimated burial depth is obtained by using an iterative algorithm to minimize the fitting error between the theoretical model and the observed data.

10. The integrated system for high-precision detection and intelligent diagnosis of leaks in heating pipe networks according to claim 1, characterized in that, The reverse tracking and positioning module is also configured to perform multi-parameter coupling correction: A medium correction factor is introduced to physically correct the optimal estimated burial depth and calculate the final burial depth. The medium correction factor is determined based on the visible light texture characteristics of soil moisture content and surface cover type. The root mean square error of the fitting residuals is calculated as a confidence index of the inversion results, and the final burial depth is combined with the geometric center coordinates of the surface to generate the inferred underground absolute coordinates of the leakage source.

Images