A method for detecting and positioning external wall leakage defects based on three-dimensional point cloud modeling

Abstract

The application discloses a kind of outer wall leakage defect detection and positioning method based on three-dimensional point cloud modeling, comprising the following steps: data acquisition, pre-processing generates outer wall point cloud dataset;Visible light image data and infrared thermal image data are pre-processed;Color information and temperature information are mapped to corresponding point cloud, and outer wall fusion color point cloud data are obtained;Carry out surface reconstruction and gridding processing, generate three-dimensional surface data;Three-dimensional surface data is automatically segmented into outer wall wall surface unit;For each outer wall wall surface unit, construct outer wall multi-source feature set;Carry out feature fusion and classification, output outer wall leakage defect identification result;Carry out spatial superposition analysis, generate outer wall leakage risk distribution data;Extract three-dimensional coordinate value, generate three-dimensional visualization result and outer wall leakage detection report.The application fuses three-dimensional point cloud modeling and deep learning, realizes outer wall leakage intelligent detection, with the advantages of high precision, accurate positioning, strong visualization.

Description

Technical Field

[0001] This invention relates to the field of intelligent building structure detection technology, and in particular to a method for detecting and locating external wall leakage defects based on three-dimensional point cloud modeling. Background Technology

[0002] Currently, the detection of leakage defects in building exterior walls mainly relies on manual inspections, infrared thermography, or two-dimensional image recognition. These methods are greatly affected by ambient lighting, climate conditions, and the complex structure of the exterior walls. The detection results are easily influenced by subjective factors, making it difficult to achieve comprehensive coverage and precise location of complex areas on the building facade. Traditional image or thermography analysis methods can only provide two-dimensional surface information, lacking spatial depth data support, and cannot quantify the true location and extent of defects, resulting in low reliability of the detection results.

[0003] While existing 3D point cloud-based building structure analysis technologies can acquire the geometry of exterior walls, they typically lack effective integration with visible light images and infrared thermal imaging data, making it difficult to comprehensively reflect the color changes and temperature anomalies of the exterior walls. Existing algorithms often rely on fixed parameters or traditional classifiers, failing to perform joint learning and adaptive recognition of multi-source features in complex environments, and struggling to accurately distinguish between defect types such as leaks, hollow areas, and cracks. These issues result in insufficient accuracy in exterior wall leak detection, inaccurate localization, and a lack of quantitative basis for risk assessment.

[0004] Therefore, how to provide a method for detecting and locating external wall leakage defects based on 3D point cloud modeling is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a method for detecting and locating external wall leakage defects based on three-dimensional point cloud modeling. This invention integrates three-dimensional point cloud modeling and deep learning to achieve intelligent detection of external wall leakage, and has the advantages of high accuracy, accurate positioning, and strong visualization.

[0006] A method for detecting and locating external wall leakage defects based on three-dimensional point cloud modeling according to an embodiment of the present invention includes the following steps:

[0007] Collect raw point cloud data of the building facade, visible light image data and infrared thermal image data of the corresponding location, and preprocess the data to generate an exterior wall point cloud dataset.

[0008] Preprocessing of visible light image data and infrared thermal image data;

[0009] Under a unified spatial coordinate system, the color information of the image and the temperature information of the infrared thermal image are mapped to the corresponding point cloud to obtain the fused color point cloud data of the exterior wall;

[0010] The exterior wall is integrated with color point cloud data for surface reconstruction and meshing to generate three-dimensional surface data.

[0011] The three-dimensional surface data is automatically divided into multiple independent and continuous exterior wall surface units;

[0012] For each exterior wall unit, geometric features, color histogram and texture features, temperature gradient and heat distribution features are extracted to construct a multi-source feature set for the exterior wall.

[0013] An improved PointNeXt network is used to perform feature fusion and classification on the multi-source feature set of the external wall, and output the identification results of the external wall leakage defects.

[0014] Spatial overlay analysis is performed on the identification results of external wall leakage defects to generate external wall leakage risk distribution data;

[0015] Extract the three-dimensional coordinates of each leakage defect point in the high-risk area of ​​the exterior wall, generate a three-dimensional visualization result of the exterior wall leakage, and automatically generate an exterior wall leakage detection report.

[0016] Optionally, obtaining the color point cloud data of the exterior wall includes:

[0017] Under a unified spatial coordinate system, the external wall point cloud dataset is spatially clipped to remove point clouds from non-external wall areas, retaining the effective external wall point cloud area of ​​the building facade, and constructing the external wall point cloud modeling area based on the boundary range of the external wall.

[0018] The point cloud data of the exterior wall within the modeling area is processed into a grid using a surface reconstruction method based on neighborhood point fitting, and a weighted average strategy is used to generate continuous three-dimensional surface grid data of the exterior wall.

[0019] Geometric optimization is performed on the 3D curved surface mesh data of the exterior wall to obtain smooth 3D curved surface data of the exterior wall;

[0020] The color information of visible light image data and the temperature information of infrared thermal image data are mapped to the corresponding grid cells of smooth three-dimensional curved surface data of exterior walls using a spatial projection matrix.

[0021] When the attribute data of multiple exterior wall points are mapped to the same exterior wall surface mesh unit, interpolation fusion calculation is performed. Based on the weighted average result, the color value, temperature value and normal vector direction of the mesh node are recalculated to obtain exterior wall fusion color three-dimensional surface data that simultaneously contains geometric information, color information and temperature information.

[0022] Optionally, the generation of the three-dimensional surface data includes:

[0023] Quality screening is performed on the color 3D curved surface data of the exterior wall to remove isolated points, noise points and low-confidence point cloud data, and adaptive resampling of the exterior wall point cloud is performed based on the density distribution characteristics of the exterior wall point cloud;

[0024] The resampled point cloud data of the exterior wall is processed into a grid using a surface reconstruction method based on neighborhood point fitting to generate preliminary three-dimensional surface data of the exterior wall.

[0025] Normal vector estimation is performed on the preliminary three-dimensional surface data of the exterior wall. The normal vector direction of each node is calculated within the local neighborhood of each exterior wall grid node. By calculating the weighted average of the spatial distance weight between the neighborhood point and the center point and the normal vector direction difference weight, a smooth and continuous normal vector distribution is obtained.

[0026] A geometric optimization algorithm combining Laplace smoothing and Gaussian weights is used to iteratively update the spatial coordinates of adjacent external wall grid nodes on the three-dimensional surface data of the exterior wall estimated by the normal vector to obtain geometrically smoothed and optimized three-dimensional surface data of the exterior wall.

[0027] Residual detection and local correction are performed on the geometrically smoothed 3D surface data of the exterior wall to finally generate smooth 3D surface data of the exterior wall with high geometric accuracy and surface continuity.

[0028] Optionally, the division of the exterior wall unit includes:

[0029] Extract the normal vector direction, curvature value and spatial coordinate information of each surface mesh node of the exterior wall from the smooth three-dimensional surface data of the exterior wall, and establish the surface feature set of the exterior wall;

[0030] Initial plane detection is performed on the feature set of the exterior wall surface, and a random sampling consistency plane fitting algorithm is used to obtain multiple candidate plane regions of the exterior wall;

[0031] Using the candidate planar region of the exterior wall as the initial seed for region growth, a region growth algorithm based on normal vector consistency, curvature change and spatial connectivity is used to expand the wall boundary and form the initial wall unit of the exterior wall;

[0032] The initial wall surface units of the exterior wall are smoothed and their connectivity is optimized to form a set of exterior wall surface units that are geometrically continuous, have smooth boundaries, and are stable in connectivity.

[0033] The optimized exterior wall units are subjected to integrity checks and redundancy removal to form a set of exterior wall units with continuous spatial structure, stable normal vector distribution, and complete geometric features.

[0034] Optionally, the construction of the multi-source feature set of the exterior wall includes:

[0035] Each exterior wall unit was selected as the feature extraction object in sequence, and the corresponding exterior wall fusion color 3D curved surface data, preprocessed visible light image data and preprocessed infrared thermal image data were loaded in a unified spatial coordinate system.

[0036] Geometric features are extracted from the exterior wall fusion color 3D curved surface data based on the range of exterior wall surface units, and the geometric feature matrix of the exterior wall surface units is obtained by synthesis;

[0037] Color histograms and texture features are extracted from the preprocessed visible light image data based on the image regions corresponding to the exterior wall units.

[0038] Temperature gradient and heat distribution features are extracted from the preprocessed infrared thermal image data based on the thermal image region corresponding to the exterior wall unit.

[0039] The geometric features, color histograms, texture features, temperature gradients, and heat distribution features are standardized and uniformly encoded. A numerical normalization algorithm is used to linearly map each feature value to the range of zero to one. The normalized feature data are indexed and identified according to the unique number of the exterior wall unit. Geometric features, color features, and temperature features under the same number are associated accordingly. The multi-source feature data of each exterior wall unit are fused and encoded. A feature vector group is formed by weighted feature concatenation. The multi-dimensional feature vectors of all exterior wall units are stacked and combined to construct a multi-source feature set for the exterior wall.

[0040] Optionally, the generation of the external wall leakage defect identification result includes:

[0041] Load the corresponding multi-source feature set of the external wall under a unified spatial coordinate system;

[0042] The multi-source feature set of the exterior wall is input into the feature fusion layer of the improved PointNeXt network, and adaptive neighborhood radius search calculation, normal vector embedding calculation and spatial attention weighting calculation are performed to obtain the fused comprehensive feature vector.

[0043] The improved PointNeXt network includes a feature fusion layer and an external wall leakage defect classification layer;

[0044] The fused integrated feature vector is input into the external wall leakage defect classification layer, and multi-source joint feature learning and classification calculation are performed. The geometric information, color information and temperature information contained in the integrated feature vector are fused and analyzed, and the external wall leakage defect identification result is output.

[0045] Optionally, the generation of the external wall leakage risk distribution data includes:

[0046] Under a unified three-dimensional coordinate system, the identified external wall leakage defect category data are loaded to establish a three-dimensional spatial dataset of external wall leakage defects;

[0047] In the 3D spatial dataset of external wall leakage defects, spatial location registration and overlap detection are performed on discolored areas, crack and hollow areas and temperature abnormal areas, and the spatial overlap between different defect types is calculated.

[0048] Based on the feature response values ​​output by the improved PointNeXt network, color change feature weights, geometric feature weights, and temperature feature weights are assigned to the discoloration region, crack and hollow region, and temperature anomaly region, respectively, to construct a multi-source feature weighted calculation model.

[0049] In the multi-source feature weighted calculation model, the spatial overlap of the three types of external wall leakage defect areas is weighted and fused with their corresponding feature weights to calculate the weighted spatial overlap index.

[0050] After obtaining the weighted spatial overlap index, the spatial consistency, overlapping area ratio and adjacent connectivity of the external wall leakage defect area are comprehensively judged, and the connectivity strength between each area is evaluated. When the weighted spatial overlap index reaches the set leakage judgment threshold or exceeds the threshold of the corresponding level in the graded leakage risk threshold set, the spatial location is marked as a high-risk area for external wall leakage.

[0051] High-risk areas for external wall leakage are classified and recorded. The identified high-risk areas for external wall leakage are correlated with the results of external wall leakage defect identification. The three-dimensional coordinates, feature type combinations, and weight contributions of each feature of the high-risk areas are recorded to generate external wall leakage risk distribution data.

[0052] Optionally, the generation of the external wall leakage detection report includes:

[0053] Under a unified spatial coordinate system, extract the set of external wall leakage defect points in high-risk areas, obtain the three-dimensional coordinate values ​​of each external wall leakage defect point, and establish a spatial coordinate set of external wall leakage defects.

[0054] Based on the spatial coordinate set of external wall leakage defects, a combination of point cloud geometric fitting and projection calculation is used to calculate the geometric parameters of external wall leakage defect points in each high-risk area of ​​external wall leakage, and to calculate the area, length and distribution density of external wall leakage defects.

[0055] Based on the grading standards in the set of external wall leakage risk thresholds, external wall leakage defects are divided into high-risk, medium-risk and low-risk levels, and the risk level labels are associated with the three-dimensional coordinates of the external wall leakage defect points to form external wall leakage defect risk attribute data.

[0056] Based on the risk attribute data of external wall leakage defects, a three-dimensional visualization model of external wall leakage is generated, which maps the type, risk level and spatial distribution information of external wall leakage defects to a three-dimensional coordinate system. At the same time, geometric anomalies such as cracks and hollows are marked in linear and planar form to reflect the spatial characteristics and morphological distribution of external wall leakage defects.

[0057] Based on the 3D visualization model of external wall leakage and the risk attribute data of external wall leakage defects, an external wall leakage detection report is automatically generated.

[0058] The beneficial effects of this invention are:

[0059] This invention constructs a method for detecting and locating exterior wall leakage defects based on 3D point cloud modeling. It achieves fusion modeling and intelligent identification of multi-source information on building facades, effectively overcoming the problems of low accuracy in existing 2D image detection, susceptibility to lighting and environmental influences, and inaccurate defect location. By accurately registering and interpolating laser point cloud data, visible light image data, and infrared thermal image data in a unified spatial coordinate system, a fused color point cloud data of the exterior wall is formed, simultaneously containing geometric, color, and temperature information. This solves the problems of spatial misalignment and low information utilization in traditional methods, providing complete 3D data support for subsequent intelligent identification.

[0060] At the algorithm level, this invention addresses the problems of uneven distribution of point clouds on exterior walls, complex surface features, and difficulties in multi-source data fusion. It structurally improves the PointNeXt network by introducing an adaptive neighborhood radius search mechanism, enabling adaptive sampling of point clouds in regions with different densities and improving the stability of feature extraction. Through a normal vector embedding mechanism, the surface normal information of the exterior wall is jointly encoded with spatial coordinates, enhancing the model's ability to express the geometric morphology and structural anomalies of the exterior wall surface. Furthermore, through a spatial attention weighting mechanism, a dynamic weight allocation relationship is established between geometric features, color features, and temperature features, achieving adaptive fusion of multi-source features, thereby significantly improving the accuracy and robustness of exterior wall leakage defect identification.

[0061] Furthermore, based on the identification results, this invention constructs a multi-feature weighted spatial overlay analysis model to comprehensively determine the spatial consistency, overlapping area ratio, and connectivity of different types of exterior wall leakage defect areas, establishing a weighted spatial overlap index to achieve a three-dimensional quantitative assessment of high-risk exterior wall leakage areas. Unlike traditional two-dimensional detection, this process can classify and locate exterior wall defects in three-dimensional space, accurately calculate the area, length, and distribution density of defects, and generate an exterior wall leakage detection report containing risk level and coordinate information. Through these innovative steps, this invention not only automates and digitizes the entire process of exterior wall leakage defect detection from detection to assessment, but also provides high-precision, visualized technical support for building maintenance and urban digital twin systems, demonstrating significant engineering application value and promising prospects for widespread adoption. Attached Figure Description

[0062] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0063] Figure 1 This is an overall flowchart of a method for detecting and locating external wall leakage defects based on three-dimensional point cloud modeling proposed in this invention;

[0064] Figure 2 This is a schematic diagram of the external wall leakage defect identification structure based on the improved PointNeXt network, which is proposed in this invention as a method for detecting and locating external wall leakage defects based on three-dimensional point cloud modeling. Detailed Implementation

[0065] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0066] refer to Figure 1 and Figure 2 A method for detecting and locating external wall leakage defects based on 3D point cloud modeling includes the following steps:

[0067] Collect raw point cloud data of the building facade, visible light image data and infrared thermal image data of the corresponding location, and preprocess the data to generate an exterior wall point cloud dataset.

[0068] Preprocessing of visible light image data and infrared thermal image data;

[0069] Under a unified spatial coordinate system, the color information of the image and the temperature information of the infrared thermal image are mapped to the corresponding point cloud to obtain the fused color point cloud data of the exterior wall;

[0070] The exterior wall is integrated with color point cloud data for surface reconstruction and meshing to generate three-dimensional surface data.

[0071] The three-dimensional surface data is automatically divided into multiple independent and continuous exterior wall surface units;

[0072] For each exterior wall unit, geometric features, color histogram and texture features, temperature gradient and heat distribution features are extracted to construct a multi-source feature set for the exterior wall.

[0073] An improved PointNeXt network is used to perform feature fusion and classification on the multi-source feature set of the external wall, and output the identification results of the external wall leakage defects.

[0074] Spatial overlay analysis is performed on the identification results of external wall leakage defects to generate external wall leakage risk distribution data;

[0075] Extract the three-dimensional coordinates of each leakage defect point in the high-risk area of ​​the exterior wall, generate a three-dimensional visualization result of the exterior wall leakage, and automatically generate an exterior wall leakage detection report.

[0076] In this embodiment, obtaining the color point cloud data of the exterior wall fusion includes:

[0077] Under a unified spatial coordinate system, the external wall point cloud dataset is spatially clipped to remove point clouds from non-external wall areas, retaining the effective external wall point cloud area of ​​the building facade, and constructing the external wall point cloud modeling area based on the boundary range of the external wall.

[0078] The point cloud data of the exterior wall within the modeling area is meshed using a surface reconstruction method based on neighborhood point fitting. The surface reconstruction method includes: selecting several neighboring points within the neighborhood of each exterior wall point, calculating the local surface equation through least squares fitting, determining the surface node position based on the normal vector direction and spatial position of each point, and generating continuous three-dimensional surface mesh data of the exterior wall using a weighted average strategy. The weighted average strategy introduces three weighting factors: spatial distance, normal vector direction, and point density, to achieve a dynamic balance between smoothness and geometric accuracy in the surface reconstruction of the exterior wall point cloud, so that the generated three-dimensional surface of the exterior wall achieves the optimal balance between detail preservation and overall continuity.

[0079] Geometric optimization of the 3D curved surface mesh data of the exterior wall is performed by adopting a surface optimization algorithm based on weighted smoothing. By repeatedly updating the spatial coordinates and normal vector directions of adjacent mesh nodes with weighted iterations, the change in the angle between the normal vectors of adjacent nodes is minimized, thereby improving the continuity and geometric accuracy of the exterior wall surface and obtaining smooth 3D curved surface data of the exterior wall.

[0080] The spatial projection matrix is ​​used to map the color information of visible light image data and the temperature information of infrared thermal image data to the corresponding grid cells of smooth three-dimensional curved surface data of the exterior wall. The spatial projection matrix consists of camera intrinsic parameter matrix and extrinsic parameter matrix, which is used to establish the projection relationship between three-dimensional spatial coordinates and two-dimensional image pixels, so as to realize the spatial correspondence between color and temperature attributes.

[0081] When the attribute data of multiple exterior wall points are mapped to the same exterior wall surface mesh unit, interpolation fusion calculation is performed. The color value, temperature value and normal vector direction of the mesh node are recalculated based on the weighted average result to obtain exterior wall fusion color three-dimensional surface data that simultaneously contains geometric information, color information and temperature information.

[0082] The generation of the fused color 3D curved surface data of the exterior wall includes: after the color information of the visible light image and the temperature information of the infrared thermal image are mapped onto the smooth 3D curved surface data of the exterior wall, if the attribute data of multiple exterior wall points fall into the same exterior wall surface grid cell, a unique fusion result is generated through interpolation fusion calculation; the interpolation fusion is based on geometric information weight, color information weight, and temperature information weight, and a weighted average is applied to the mapped data. The geometric information weight is determined according to the spatial distance between the mapped point and the center of the grid cell, with a greater weight for closer distances; the color information weight is determined according to the consistency between the color of the mapped point and the color of the grid cell, with a greater weight for more similar colors; the temperature information weight is determined according to the consistency between the temperature value of the mapped point and the color of the grid cell. The similarity of the average temperature of the wall area is determined, with greater weight given to closer temperatures. In the fusion calculation, the color value of each mapping point is multiplied by its color information weight, summed, and then divided by the total color weight to obtain the average color value of the grid cell. The temperature value is multiplied by the temperature information weight, summed, and then divided by the total temperature weight to obtain the average temperature value. The normal vector direction is multiplied by the geometric information weight, summed, and then divided by the total geometric weight to obtain the smoothed normal vector direction. Through weighted averaging calculation, each external wall surface grid cell retains only one set of color value, temperature value, and normal vector direction, ultimately forming fused color 3D curved surface data of the external wall that simultaneously contains geometric, color, and temperature information.

[0083] In this embodiment, the generation of the three-dimensional surface data includes:

[0084] The quality of the fused color 3D curved surface data of the exterior wall is screened to remove isolated points, noise points and low confidence point cloud data. Based on the density distribution characteristics of the exterior wall point cloud, adaptive resampling is performed on the exterior wall point cloud. The adaptive resampling calculates the local point density based on the spatial distance between points and automatically adjusts the sampling step size to obtain exterior wall resampled point cloud data with uniform spatial distribution and complete structure.

[0085] The resampled point cloud data of the exterior wall is processed into a grid using a surface reconstruction method based on neighborhood point fitting. The surface reconstruction method includes: extracting a set of neighboring points within the neighborhood of each exterior wall point; calculating the local surface equation by least squares fitting; determining the spatial position of the exterior wall grid nodes based on the spatial distribution of neighboring points and the direction of the normal vector; and controlling the node generation density and smoothing constraints using a weighted comprehensive value of spatial distance, normal vector angle, and point density to generate preliminary three-dimensional surface data of the exterior wall.

[0086] Normal vector estimation is performed on the preliminary three-dimensional surface data of the exterior wall. A normal vector estimation algorithm based on the weighted neighborhood method is adopted. The normal vector direction of each node is calculated within the local neighborhood of each exterior wall grid node. By calculating the weighted average of the spatial distance weight between the neighboring point and the center point and the difference weight of the normal vector direction, a smooth and continuous normal vector distribution is obtained.

[0087] A geometric optimization algorithm combining Laplace smoothing and Gaussian weighting is used to optimize the three-dimensional surface data of the exterior wall based on the normal vector estimation. The spatial coordinates of adjacent exterior wall grid nodes are updated iteratively multiple times to obtain geometrically smoothed three-dimensional surface data of the exterior wall. Specifically, in each iteration, a weighted average coordinate is calculated based on the spatial distance weight between adjacent nodes, the normal vector difference weight, and the node curvature weight. The surface smoothness is controlled by constraining the node position offset, gradually eliminating noise and local abrupt changes on the exterior wall surface, and ensuring that the exterior wall surface achieves overall continuity and smoothness while maintaining its geometric characteristics.

[0088] Residual detection and local correction are performed on the geometrically smoothed 3D surface data of the exterior wall. A threshold-driven residual correction algorithm is used to calculate the fitting error of each exterior wall grid node. When the local error exceeds the preset tolerance threshold, neighborhood point weighted reconstruction is performed. The spatial coordinates and normal vector directions of abnormal nodes are iteratively updated until the global residual is less than the set threshold. Finally, smooth 3D surface data of the exterior wall with high geometric accuracy and surface continuity is generated.

[0089] In this embodiment, the division of the exterior wall unit includes:

[0090] The normal vector direction, curvature value and spatial coordinate information of each curved mesh node of the exterior wall are extracted from the smooth three-dimensional surface data of the exterior wall to establish the exterior wall surface feature set for exterior wall unit segmentation.

[0091] Initial plane detection is performed on the feature set of the outer wall surface. The random sampling consistency plane fitting algorithm is adopted. Several node sample point sets are randomly selected in the smooth three-dimensional surface data of the outer wall. The local plane equation is calculated by the least squares method, and the distance residual from each sample point to the plane is calculated. Nodes with residuals less than the plane fitting threshold are marked as interior points, and nodes with residuals greater than the threshold are marked as exterior points. The iteration is repeated until the residuals converge, and multiple candidate plane regions of the outer wall are obtained.

[0092] Using the candidate planar region of the outer wall as the initial seed for region growth, a region growth algorithm based on normal vector consistency, curvature change and spatial connectivity is adopted to expand the wall boundary. Specifically, when the angle between the normal vector of the adjacent node and the current region is less than the angle threshold, the curvature difference is less than the curvature threshold, and the distance between the node and the current region boundary is less than the connectivity threshold, the adjacent node is merged into the current region. This process continues until the nodes in the region no longer meet the growth conditions, thus forming the initial wall unit of the outer wall.

[0093] The initial wall surface units of the exterior wall are subjected to boundary smoothing and connectivity optimization. A boundary adjustment method based on minimum energy constraint is adopted. By calculating the spatial distance, normal vector direction difference and node density difference between adjacent exterior wall surface units, the spatial position of the boundary nodes is adjusted to make the wall boundary line continuous and smooth. The difference in normal vector direction between adjacent exterior wall surface units is judged. When the difference is less than the connectivity threshold, the two units are merged into the same exterior wall surface unit. Otherwise, they are kept separate, forming a set of geometrically continuous, boundary smooth and connectivity stable exterior wall surface units.

[0094] Integrity checks and redundancy removal are performed on the optimized exterior wall units. The area, number of nodes, and normal vector distribution variance of each exterior wall unit are calculated. When the area is less than the minimum wall threshold or the normal vector distribution variance exceeds the set upper limit, the exterior wall unit is removed, forming a set of exterior wall units with continuous spatial structure, stable normal vector distribution, and complete geometric features.

[0095] In this embodiment, the construction of the multi-source feature set of the external wall includes:

[0096] Each exterior wall unit was selected as the feature extraction object in sequence, and the corresponding exterior wall fusion color 3D curved surface data, preprocessed visible light image data and preprocessed infrared thermal image data were loaded in a unified spatial coordinate system.

[0097] Geometric features are extracted from the exterior wall fusion color 3D curved surface data based on the range of exterior wall surface units. The geometric features include normal vector change, curvature and surface undulation. The normal vector change is obtained by calculating the root mean square value of the angle between the normal vectors of adjacent point cloud nodes. The curvature is calculated by fitting the principal curvature value of the node based on a local quadratic surface. The surface undulation is obtained by calculating the local height variance to characterize the flatness of the exterior wall surface. The geometric feature matrix of the exterior wall surface unit is obtained by combining the features.

[0098] Color histograms and texture features are extracted from the preprocessed visible light image data based on the image regions corresponding to the exterior wall units. The color histograms are calculated from the pixel distribution of each color channel in the image region, and the texture features are calculated from the contrast, energy, and entropy values ​​of the gray-level co-occurrence matrix of the image region.

[0099] Temperature gradient and heat distribution features are extracted from the preprocessed infrared thermal image data based on the thermal image region corresponding to the outer wall unit. The temperature gradient is calculated from the temperature difference between adjacent thermal image pixels, and the heat distribution features are calculated from the temperature mean and temperature variance of the thermal image region.

[0100] The geometric features, color histograms, texture features, temperature gradients, and thermal distribution features are standardized and uniformly encoded. A numerical normalization algorithm is used to linearly map each feature value to the range of zero to one. The normalized feature data are indexed and identified according to the unique number of the exterior wall unit. Geometric features, color features, and temperature features under the same number are associated accordingly. The multi-source feature data of each exterior wall unit are fused and encoded. A feature vector group is formed by weighted feature concatenation. The weighted feature concatenation is achieved by assigning weight coefficients to geometric features, color features, and temperature features respectively, resulting in a multi-dimensional feature vector of the exterior wall unit that includes normal vector change, curvature, surface undulation, color channel statistical parameters, texture matrix parameters, temperature gradient, and temperature variance. The multi-dimensional feature vectors of all exterior wall units are stacked and combined to construct a multi-source feature set of the exterior wall. The multi-source feature set of the exterior wall contains the geometric information, color information, and temperature information corresponding to each exterior wall unit.

[0101] In this embodiment, the generation of the external wall leakage defect identification result includes:

[0102] Load the corresponding multi-source feature set of the external wall under a unified spatial coordinate system;

[0103] The multi-source feature set of the exterior wall is input into the feature fusion layer of the improved PointNeXt network, and adaptive neighborhood radius search calculation, normal vector embedding calculation and spatial attention weighting calculation are performed to obtain the fused comprehensive feature vector.

[0104] The improved PointNeXt network includes a feature fusion layer and an external wall leakage defect classification layer;

[0105] The adaptive neighborhood radius search calculation includes: for each external wall point in the multi-source feature set of the external wall, calculating the average spatial distance between the external wall point and its neighboring points, and calculating the average point spacing of all external wall points. The initial neighborhood radius is multiplied by the ratio of the global average point spacing to the local average point spacing to obtain the adaptive neighborhood radius of the external wall point. When the local point density is higher than the global average density, the neighborhood radius automatically shrinks; when the local point cloud density is lower than the global average density, the neighborhood radius automatically increases, thereby ensuring that the number of external wall points in different regions of the multi-source feature set of the external wall is consistent. Within the determined adaptive neighborhood range, the spatial coordinate difference between the neighboring points and the center point is averaged to generate a basic feature vector reflecting the local geometric changes of the external wall.

[0106] The execution of the normal vector embedding calculation includes: jointly encoding the spatial coordinates, normal vector direction, and basic feature vector generated by adaptive neighborhood radius search for each external wall point in the multi-source feature set of the external wall; combining the three types of data to form an enhanced geometric input vector; multiplying the input vector with weight parameters and adding a bias term; and then processing it through a nonlinear mapping function to obtain a high-dimensional embedding feature; through this calculation process, the network simultaneously learns the positional distribution, surface direction, and local geometric change features of the external wall points in the feature space, thereby enhancing its response capability to cracks, voids, and normal protrusions;

[0107] The execution of spatial attention weighted calculation includes: using the high-dimensional geometric embedding features generated by normal vector embedding calculation as the geometric channel input, and using the color features and temperature features from the multi-source feature set of the exterior wall as the color channel and temperature channel input, respectively. Joint weighting is performed on the three types of input channels, calculating the response intensity of each channel and normalizing it to obtain the corresponding weight coefficients. Then, each channel feature is multiplied by its corresponding weight coefficient and summed to obtain the fused comprehensive feature vector. In this process, features with high response intensity are given higher weights, thereby highlighting local feature areas of exterior wall discoloration, cracks, and temperature anomalies, while weakening the feature response of normal wall areas, thus achieving dynamic weighted fusion of multi-source information.

[0108] The fused comprehensive feature vector is input into the external wall leakage defect classification layer. The external wall leakage defect classification layer is the output layer structure in the improved PointNeXt network. It adopts a combination of multi-layer linear transformation and feature normalization to perform feature mapping and category determination on the input comprehensive feature vector, perform multi-source joint feature learning and classification calculation, and perform fusion analysis on the geometric information, color information and temperature information contained in the comprehensive feature vector to output the external wall leakage defect identification result. The external wall leakage defect identification result includes discoloration areas based on color changes, crack and hollow areas based on geometric features, and temperature anomaly areas based on temperature distribution.

[0109] The multi-source joint feature learning and classification calculation includes: inputting the comprehensive feature vector after adaptive neighborhood radius search, normal vector embedding, and spatial attention weighting into the classification layer; performing multi-level linear transformation and feature normalization operations on the comprehensive feature vector of each exterior wall point to extract high-level semantic features; establishing a multi-source joint feature learning function to jointly model the correlation between geometric information, color information, and temperature information, and calculating the contribution of each feature dimension in the identification of exterior wall leakage defects; inputting the learned feature representation into the classification function, and calculating the probability value of each exterior wall point belonging to different leakage defect categories; in the probability calculation, normalization operations are used to make the sum of the probabilities of each category equal to one, and the category with the highest probability value is taken as the final identification result of the exterior wall point.

[0110] In this embodiment, the generation of the external wall leakage risk distribution data includes:

[0111] Under a unified three-dimensional coordinate system, the identified external wall leakage defect category data are loaded, including discoloration areas based on color changes, crack and hollow areas based on geometric features, and temperature anomaly areas based on temperature distribution, to establish a three-dimensional spatial dataset of external wall leakage defects.

[0112] In the three-dimensional spatial dataset of external wall leakage defects, spatial location registration and overlap detection are performed on discolored areas, crack and hollow areas, and temperature abnormal areas. The spatial overlap between different defect types is calculated. The spatial overlap is determined by the ratio of the number of overlapping points to the total number of detection points.

[0113] Based on the feature response values ​​output by the improved PointNeXt network, color change feature weights, geometric feature weights, and temperature feature weights are assigned to the discoloration region, crack and hollow region, and temperature anomaly region, respectively. The color change feature weight is determined based on the feature response intensity of the color change channel, the geometric feature weight is determined based on the average amplitude of normal vector perturbation and curvature change, and the temperature feature weight is determined based on the combination value of temperature gradient change rate and temperature variance. A multi-source feature weighted calculation model is constructed.

[0114] In the multi-source feature weighted calculation model, the spatial overlap of three types of external wall leakage defect areas is weighted and fused with their corresponding feature weights to calculate a weighted spatial overlap index. The weighted spatial overlap index is obtained by multiplying the color change feature weight, geometric feature weight, and temperature feature weight by their corresponding spatial overlap, summing them, and then normalizing them. It is used to comprehensively characterize the degree of composite overlap of external wall leakage defect areas in three-dimensional space.

[0115] After obtaining the weighted spatial overlap index, the spatial consistency, overlapping area ratio and adjacent connectivity of the external wall leakage defect area are comprehensively judged. The spatial connectivity analysis algorithm based on neighborhood correlation degree is adopted. By calculating the distance matrix and the angle matrix between the normal vectors of adjacent areas, the connectivity strength between each area is evaluated. When the weighted spatial overlap index reaches the set leakage judgment threshold or exceeds the threshold of the corresponding level in the graded leakage risk threshold set, the spatial location is marked as a high-risk area for external wall leakage.

[0116] High-risk areas for external wall leakage are classified and recorded. The identified high-risk areas for external wall leakage are correlated with the results of external wall leakage defect identification. The three-dimensional coordinates, feature type combinations, and weight contributions of each feature of the high-risk areas are recorded to generate external wall leakage risk distribution data.

[0117] In this embodiment, the generation of the external wall leakage detection report includes:

[0118] Under a unified spatial coordinate system, extract the set of external wall leakage defect points in high-risk areas, obtain the three-dimensional coordinate values ​​of each external wall leakage defect point, and establish a spatial coordinate set of external wall leakage defects.

[0119] Based on the spatial coordinate set of external wall leakage defects, a combination of point cloud geometric fitting and projection calculation is used to calculate the geometric parameters of external wall leakage defect points in each high-risk area of ​​external wall leakage, calculate the area, length and distribution density of external wall leakage defects, project the defect point set onto the plane of the external wall surface, calculate the projected area of ​​external wall leakage defects based on the boundary contour of the point cloud, reconstruct the point sequence of external wall leakage defect points of crack and hollow type along the connected path, calculate the length of external wall leakage defects based on the cumulative Euclidean distance between adjacent points, and calculate the distribution density of external wall leakage defects by counting the number of external wall leakage defect points within the area of ​​the wall unit.

[0120] Based on the grading standards in the set of external wall leakage risk thresholds, external wall leakage defects are divided into high-risk, medium-risk and low-risk levels, and the risk level labels are associated with the three-dimensional coordinates of the external wall leakage defect points to form external wall leakage defect risk attribute data.

[0121] Based on the risk attribute data of external wall leakage defects, a three-dimensional visualization model of external wall leakage is generated. The type, risk level and spatial distribution information of external wall leakage defects are mapped to a three-dimensional coordinate system. The risk classification is visualized through color and transparency. High-risk areas are shown in red, medium-risk areas are shown in orange, and low-risk areas are shown in yellow. At the same time, geometric anomalies such as cracks and hollow areas are marked with lines and areas to reflect the spatial characteristics and morphological distribution of external wall leakage defects.

[0122] Based on the 3D visualization model of external wall leakage and the risk attribute data of external wall leakage defects, an external wall leakage detection report is automatically generated. The report includes the 3D coordinate data, area, length, distribution density and corresponding risk level information of external wall leakage defects, and statistically analyzes the leakage risk ratio and defect distribution of each external wall unit to form a quantitative assessment result of external wall leakage.

[0123] Example 1:

[0124] To verify the feasibility of this invention in practice, it was applied to the detection and location of exterior wall leaks in a large commercial complex. Located in a humid coastal region, this building has a large exterior wall area, numerous floors, and a complex structure. Long-term exposure to the humid climate and wind erosion has resulted in water seepage, discoloration, and surface bulging in some facade areas. Traditional manual inspection methods are limited by perspective and height, making it impossible to fully grasp the overall condition of the exterior walls. This is especially true for high-rise facades, curtain wall joints, and areas where decorative layers meet; manual inspection is time-consuming, inaccurate, and poses safety risks. Therefore, this building was chosen as the verification scenario to utilize the exterior wall leak defect detection and location method based on 3D point cloud modeling proposed in this invention to comprehensively assess the distribution of exterior wall leak risks and achieve high-precision, visualized exterior wall health diagnosis.

[0125] In the implementation process, ground-based laser scanning equipment and drones equipped with lidar are used in conjunction to collect 3D point cloud data of the building facade. Simultaneously, corresponding high-resolution visible light images and infrared thermal images are acquired. After preprocessing and noise removal, the collected data forms a complete exterior wall point cloud dataset, which is then spatially registered to a common coordinate system. The key advantage of this invention lies in the fusion processing of multi-source data. It can map the color information of visible light images and the temperature distribution information of infrared thermal images to corresponding point cloud nodes, thereby generating fused color point cloud data of the exterior wall that simultaneously contains geometric, color, and thermal information, providing a comprehensive data foundation for subsequent intelligent recognition.

[0126] The generated color point cloud data of the exterior walls is subjected to surface reconstruction and meshing. Through normal vector estimation and geometric optimization, the surface model of the exterior walls is made more continuous and smooth in detail, and the three-dimensional shape of the walls is realistically reflected. Then, using a segmentation algorithm based on normal consistency, curvature change and spatial connectivity, the point cloud of the entire building's exterior walls is divided into multiple independent wall units. This step realizes regional management of complex facades, so that subsequent analysis can perform feature extraction and defect identification for each wall individually, which significantly improves processing efficiency and spatial resolution.

[0127] In the data analysis phase, geometric features, color histograms and texture features, as well as temperature gradients and heat distribution features, were extracted from each exterior wall unit to construct a multi-source feature set. Geometric features reflect wall deformation and surface undulations, color and texture features reflect discoloration and water stains, and temperature features reveal areas of abnormal heat conduction. This multi-source feature set was input into an exterior wall leakage defect identification model based on an improved PointNeXt network. This model introduces an adaptive neighborhood radius search mechanism to dynamically adjust the sampling range of point cloud features in areas of different densities, ensuring balanced feature extraction. Through a normal vector embedding mechanism, the model can learn the directional change features of the exterior wall surface in a high-dimensional feature space. A spatial attention weighting mechanism enables the model to automatically focus on geometric anomalies, color changes, and temperature anomalies, thus achieving adaptive fusion of multi-source information. The final output can distinguish discolored areas, cracked and hollow areas, and temperature anomaly areas, providing a reliable basis for intelligent identification of exterior wall leaks.

[0128] After identifying external wall leakage defects, the identification results are spatially overlaid to perform registration and overlap calculations on abnormal areas from different feature sources in three-dimensional space. By analyzing the spatial consistency and overlap of color change, geometric anomaly, and temperature anomaly areas, and combining feature response weights, a multi-source feature weighted calculation model is constructed. When the comprehensive weighted result reaches a preset threshold, the area can be identified as a high-risk area for external wall leakage. In this way, the system can automatically identify the leakage risk distribution in three-dimensional space, achieving a quantitative expression of the leakage risk level of different parts of the external wall. Unlike previous methods that relied solely on infrared images to determine thermal anomaly areas or on manual observation of color changes, this invention can achieve joint reasoning of multi-source features in three-dimensional space, greatly improving the accuracy and stability of risk assessment.

[0129] Finally, a 3D visualization model of external wall leakage is generated using the identified high-risk areas. Different types of defects are labeled using color-coded layers, and the 3D coordinates, area, length, and risk level of each defect point are recorded in a digital report. The visualization model can be freely rotated, zoomed in, and measured on a computer, providing building maintenance personnel with an intuitive leakage distribution map and quantitative data support. The inspection report can be directly output to the building management system, realizing the archiving and digital management of leakage monitoring results.

[0130] In this implementation scenario, the present invention effectively solves the problems of numerous blind spots, strong subjectivity of manual judgment, and non-reusable data in traditional detection methods. Through multi-source data fusion and deep learning recognition mechanisms, the present invention achieves automation and intelligence in the detection of external wall leakage defects, allowing the true state of the external wall structure to be presented intuitively in a three-dimensional visualization. The detection process is efficient, the data is comprehensive, and the results are accurate, providing a scientific basis and technical support for subsequent external wall maintenance and long-term structural health monitoring. The application of the present invention significantly improves the level of intelligence in external wall leakage detection and provides a scalable technical path for the digital transformation of building maintenance and urban space management.

[0131] Table 1. Performance Comparison Analysis of Different External Wall Leakage Detection Methods

[0132] Detection methods Recognition accuracy (%) Positioning error (cm) False negative rate (%) Testing time (min / 100㎡) Visualization Manual inspection method 72.4 15.8 21.3 95 Low Infrared thermography 80.7 10.2 17.6 63 middle Traditional point cloud modeling method 85.9 7.6 12.8 48 Medium and high Method of the present invention 94.6 3.2 5.7 26 high

[0133] As shown in Table 1, the proposed method for detecting and locating external wall leakage defects based on 3D point cloud modeling significantly outperforms traditional methods in terms of detection accuracy, location accuracy, missed detection rate, and detection efficiency. Compared with manual inspection, the recognition accuracy is improved by 22.2%, and the detection time is shortened by approximately 70%. This is mainly due to the introduction of multi-source data fusion and deep learning automatic recognition mechanisms, which avoid the uncertainty of human judgment. Compared with single infrared thermal imaging, the proposed method shows significant improvements in both recognition accuracy and missed detection rate. This is because infrared detection is highly dependent on ambient temperature differences, while multi-source fusion technology effectively compensates for the shortcomings of thermal imaging under low-contrast conditions. Compared with traditional point cloud modeling methods, this invention achieves deep joint learning of geometric features, color features, and temperature features through an improved PointNeXt network, resulting in an approximately 10% improvement in recognition accuracy and a reduction in location error to about 3 cm. Furthermore, this method can generate 3D visualized detection results and automated reports, which are significantly superior to traditional methods in terms of visualization. The detection results are intuitive and easy to maintain and use for decision-making.

[0134] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for detecting and locating external wall leakage defects based on 3D point cloud modeling, characterized in that, Includes the following steps: Collect raw point cloud data of the building facade, visible light image data and infrared thermal image data of the corresponding location, and preprocess the data to generate an exterior wall point cloud dataset. Preprocessing of visible light image data and infrared thermal image data; Under a unified spatial coordinate system, the color information of the image and the temperature information of the infrared thermal image are mapped to the corresponding point cloud to obtain the fused color point cloud data of the exterior wall; The exterior wall is integrated with color point cloud data for surface reconstruction and meshing to generate three-dimensional surface data. The three-dimensional surface data is automatically divided into multiple independent and continuous exterior wall surface units; For each exterior wall unit, geometric features, color histogram and texture features, temperature gradient and heat distribution features are extracted to construct a multi-source feature set for the exterior wall. An improved PointNeXt network is used to perform feature fusion and classification on the multi-source feature set of the external wall, and output the identification results of the external wall leakage defects. The specific steps of using the improved PointNeXt network for feature fusion and classification of multi-source feature sets of external walls include: Load the corresponding multi-source feature set of the external wall under a unified spatial coordinate system; The improved PointNeXt network includes a feature fusion layer and an external wall leakage defect classification layer; The multi-source feature set of the exterior wall is input into the feature fusion layer of the improved PointNeXt network, and adaptive neighborhood radius search calculation, normal vector embedding calculation and spatial attention weighting calculation are performed to obtain the fused comprehensive feature vector. The adaptive neighborhood radius search calculation includes: for each external wall point in the multi-source feature set of the external wall, calculating the local average point distance between the external wall point and its neighboring points, and calculating the global average point distance between the external wall points. The initial neighborhood radius is multiplied by the ratio of the global average point distance to the local average point distance to obtain the adaptive neighborhood radius of the external wall point. When the local point density is higher than the global average density, the neighborhood radius automatically decreases; when the local point cloud density is lower than the global average density, the neighborhood radius automatically increases. The execution of normal vector embedding calculation includes: jointly encoding the spatial coordinates, normal vector direction, and basic feature vector generated by adaptive neighborhood radius search for each external wall point in the multi-source feature set of the external wall; combining the three types of data to form an enhanced geometric input vector; multiplying the input vector with the weight parameters and adding a bias term; and then processing it through a nonlinear mapping function to obtain high-dimensional embedding features. The execution of spatial attention weighted calculation includes: taking the high-dimensional embedding features generated by normal vector embedding calculation as geometric channel input, taking the color features and temperature features in the multi-source feature set of the exterior wall as color channel and temperature channel input respectively, performing joint weighting processing on the three types of input channels, calculating the response intensity of each channel and performing normalization processing to obtain the corresponding weight coefficient; and multiplying each channel feature by the corresponding weight coefficient and adding them together to obtain the fused comprehensive feature vector. The fused comprehensive feature vector is input into the external wall leakage defect classification layer, and multi-source joint feature learning and classification calculation are performed. The geometric information, color information and temperature information contained in the comprehensive feature vector are fused and analyzed, and the external wall leakage defect identification result is output. Spatial overlay analysis is performed on the identification results of external wall leakage defects to generate external wall leakage risk distribution data; Extract the three-dimensional coordinates of each leakage defect point in the high-risk area of ​​the exterior wall, generate a three-dimensional visualization result of the exterior wall leakage, and automatically generate an exterior wall leakage detection report.

2. The method for detecting and locating external wall leakage defects based on three-dimensional point cloud modeling according to claim 1, characterized in that, The obtained external wall fusion color point cloud data includes: Under a unified spatial coordinate system, the external wall point cloud dataset is spatially clipped to remove point clouds from non-external wall areas, retaining the effective external wall point cloud area of ​​the building facade, and constructing the external wall point cloud modeling area based on the boundary range of the external wall. The point cloud data of the exterior wall within the modeling area is processed into a grid using a surface reconstruction method based on neighborhood point fitting, and a weighted average strategy is used to generate continuous three-dimensional surface grid data of the exterior wall. Geometric optimization is performed on the 3D curved surface mesh data of the exterior wall to obtain smooth 3D curved surface data of the exterior wall; The color information of visible light image data and the temperature information of infrared thermal image data are mapped to the corresponding grid cells of smooth three-dimensional curved surface data of exterior walls using a spatial projection matrix. When the attribute data of multiple exterior wall points are mapped to the same exterior wall surface mesh unit, a weighted fusion calculation is performed. Based on the weighted average result, the color value, temperature value, and normal vector direction of the mesh node are recalculated to obtain exterior wall fused color point cloud data that simultaneously contains geometric, color, and temperature information.

3. The method for detecting and locating external wall leakage defects based on three-dimensional point cloud modeling according to claim 1, characterized in that, The generation of the three-dimensional surface data includes: Quality screening is performed on the color point cloud data of the exterior wall to remove isolated points, noisy points and low-confidence point cloud data. Adaptive resampling of the exterior wall point cloud is performed based on the density distribution characteristics of the exterior wall point cloud. The resampled point cloud data of the exterior wall is processed into a grid using a surface reconstruction method based on neighborhood point fitting to generate preliminary three-dimensional surface data of the exterior wall. Normal vector estimation is performed on the preliminary three-dimensional surface data of the exterior wall. The normal vector direction of each node is calculated within the local neighborhood of each exterior wall grid node. By calculating the weighted average of the spatial distance weight between the neighborhood point and the center point and the normal vector direction difference weight, a smooth and continuous normal vector distribution is obtained. A geometric optimization algorithm combining Laplace smoothing and Gaussian weights is used to iteratively update the spatial coordinates of adjacent external wall grid nodes on the three-dimensional surface data of the exterior wall estimated by the normal vector to obtain geometrically smoothed and optimized three-dimensional surface data of the exterior wall. Residual detection and local correction are performed on the geometrically smoothed 3D surface data of the exterior wall to finally generate smooth 3D surface data of the exterior wall with high geometric accuracy and surface continuity.

4. The method for detecting and locating external wall leakage defects based on three-dimensional point cloud modeling according to claim 3, characterized in that, The division of the exterior wall unit includes: Extract the normal vector direction, curvature value and spatial coordinate information of each surface mesh node of the exterior wall from the smooth three-dimensional surface data of the exterior wall, and establish the surface feature set of the exterior wall; Initial plane detection is performed on the feature set of the exterior wall surface, and a random sampling consistency plane fitting algorithm is used to obtain multiple candidate plane regions of the exterior wall; Using the candidate planar region of the exterior wall as the initial seed for region growth, a region growth algorithm based on normal vector consistency, curvature change and spatial connectivity is used to expand the wall boundary and form the initial wall unit of the exterior wall; The initial wall surface units of the exterior wall are smoothed and their connectivity is optimized to form a set of exterior wall surface units that are geometrically continuous, have smooth boundaries, and are stable in connectivity. The optimized exterior wall units are subjected to integrity checks and redundancy removal to form a set of exterior wall units with continuous spatial structure, stable normal vector distribution, and complete geometric features.

5. The method for detecting and locating external wall leakage defects based on three-dimensional point cloud modeling according to claim 1, characterized in that, The construction of the multi-source feature set of the exterior wall includes: Each exterior wall unit was selected as the feature extraction object in sequence, and the corresponding exterior wall fused color point cloud data, preprocessed visible light image data and preprocessed infrared thermal image data were loaded in a unified spatial coordinate system. Geometric features are extracted from the color point cloud data of the exterior wall based on the range of the exterior wall unit, and the geometric feature matrix of the exterior wall unit is obtained by combining them. Color histograms and texture features are extracted from the preprocessed visible light image data based on the image regions corresponding to the exterior wall units. Temperature gradient and heat distribution features are extracted from the preprocessed infrared thermal image data based on the thermal image region corresponding to the exterior wall unit. The geometric features, color histograms, texture features, temperature gradients, and heat distribution features are standardized and uniformly encoded. A numerical normalization algorithm is used to linearly map each feature value to the range of zero to one. The normalized feature data are indexed and identified according to the unique number of the exterior wall unit. Geometric features, color features, and temperature features under the same number are associated accordingly. The multi-source feature data of each exterior wall unit are fused and encoded. A feature vector group is formed by weighted feature concatenation. The multi-dimensional feature vectors of all exterior wall units are stacked and combined to construct a multi-source feature set for the exterior wall.

6. The method for detecting and locating external wall leakage defects based on three-dimensional point cloud modeling according to claim 1, characterized in that, The generation of the external wall leakage risk distribution data includes: Under a unified three-dimensional coordinate system, the identified external wall leakage defect category data are loaded to establish a three-dimensional spatial dataset of external wall leakage defects; In the 3D spatial dataset of external wall leakage defects, spatial location registration and overlap detection are performed on discolored areas, crack and hollow areas and temperature abnormal areas, and the spatial overlap between different defect types is calculated. Based on the feature response values ​​output by the improved PointNeXt network, color change feature weights, geometric feature weights, and temperature feature weights are assigned to the discoloration region, crack and hollow region, and temperature anomaly region, respectively, to construct a multi-source feature weighted calculation model. In the multi-source feature weighted calculation model, the spatial overlap of the three types of external wall leakage defect areas is weighted and fused with their corresponding feature weights to calculate the weighted spatial overlap index. After obtaining the weighted spatial overlap index, the spatial consistency, overlapping area ratio and adjacent connectivity of the external wall leakage defect area are comprehensively judged, and the connectivity strength between each area is evaluated. When the weighted spatial overlap index reaches the set leakage judgment threshold or exceeds the threshold of the corresponding level in the graded leakage risk threshold set, the spatial location is marked as a high-risk area for external wall leakage. High-risk areas for external wall leakage are classified and recorded. The identified high-risk areas for external wall leakage are correlated with the results of external wall leakage defect identification. The three-dimensional coordinates, feature type combinations, and weight contributions of each feature of the high-risk areas are recorded to generate external wall leakage risk distribution data.

7. The method for detecting and locating external wall leakage defects based on three-dimensional point cloud modeling according to claim 1, characterized in that, The generation of the external wall leakage detection report includes: Under a unified spatial coordinate system, extract the set of external wall leakage defect points in high-risk areas, obtain the three-dimensional coordinate values ​​of each external wall leakage defect point, and establish a spatial coordinate set of external wall leakage defects. Based on the spatial coordinate set of external wall leakage defects, a combination of point cloud geometric fitting and projection calculation is used to calculate the geometric parameters of external wall leakage defect points in each high-risk area of ​​external wall leakage, and to calculate the area, length and distribution density of external wall leakage defects. Based on the grading standards in the set of external wall leakage risk thresholds, external wall leakage defects are divided into high-risk, medium-risk and low-risk levels, and the risk level labels are associated with the three-dimensional coordinates of the external wall leakage defect points to form external wall leakage defect risk attribute data. Based on the risk attribute data of external wall leakage defects, a three-dimensional visualization model of external wall leakage is generated, which maps the type, risk level and spatial distribution information of external wall leakage defects to a three-dimensional coordinate system. At the same time, geometric anomalies such as cracks and hollows are marked in linear and planar form to reflect the spatial characteristics and morphological distribution of external wall leakage defects. Based on the 3D visualization model of external wall leakage and the risk attribute data of external wall leakage defects, an external wall leakage detection report is automatically generated.

Images