Power pipeline point cloud semantic recognition and three-dimensional modeling method based on PointNet+

Abstract

The application discloses a power pipeline point cloud semantic recognition and three-dimensional modeling method based on PointNet++, and belongs to the technical field of three-dimensional laser point cloud data processing, deep learning semantic segmentation and three-dimensional modeling of electric power engineering. The method builds a full-process automatic processing system from original point cloud input to complete three-dimensional model output, and completes modeling through four links of point cloud preprocessing, semantic recognition, parameterized fitting and topological reconstruction. In the point cloud preprocessing link, statistical filtering is combined with radius filtering denoising, FPFH feature matching and ICP precise registration, and voxel downsampling technology is used to generate a standardized point cloud dataset. The application solves the problems of low automation degree, poor precision and weak scene adaptability of three-dimensional modeling of electric power pipelines in the prior art, can reduce manual intervention by more than 80%, the average semantic segmentation precision of components reaches 95.2%, is fully suitable for various electric power pipeline scenes such as transformer substations, cable pipe corridors and power transmission lines, and provides technical support for digital management of the whole life cycle of electric power pipelines.

Description

Technical Field

[0001] This invention relates to the interdisciplinary fields of 3D laser point cloud data processing, deep learning semantic segmentation, and 3D modeling of power engineering, specifically to a method for semantic recognition and parametric 3D modeling of multiple components of power pipelines based on an improved PointNet++. Background Technology

[0002] As a core component of power grid infrastructure, high-precision 3D models of power pipelines are crucial foundational data for power engineering design, construction management, operation and maintenance, and digital twin construction. With the rapid transformation of my country's power grid construction towards intelligence and digitalization, the number of power pipelines in various scenarios such as substations, cable corridors, and transmission lines has increased significantly, and their structures have become increasingly complex. This places higher demands on the efficiency, accuracy, and automation of 3D modeling of power pipelines.

[0003] Currently, the 3D modeling technology for power pipelines mainly suffers from the following three types of technical defects: For example, Chinese patent CN121883893A discloses a method for extracting power lines. This method can only achieve basic geometric fitting of a single power line through Euclidean clustering and catenary models, and cannot semantically distinguish various pipeline components such as cables, supports, joints, and valves. Another example is Chinese patent CN119151924B, which discloses a method for detecting inspection anomalies. This method only segments specific components such as insulator chains and is easily affected by point cloud noise, component occlusion, and mixed arrangement of different components, resulting in low modeling accuracy and poor model standardization, which makes it difficult to meet the requirements of refined modeling in power engineering.

[0004] For example, Chinese patent CN120107602B discloses a semantic segmentation method for high-voltage transmission lines. This method uses a network structure that combines kernel convolution and graph edge convolution with attention modules, but it is only designed for transmission line scenarios and cannot be adapted to complex indoor scenarios such as substations and cable corridors. Another example is Chinese patent CN117593750A, which discloses a multi-source feature extraction method. Although it improves the point cloud feature extraction capability, it does not combine the prior geometric knowledge of power pipeline engineering, making it difficult to achieve accurate parametric modeling of power-specific components. Furthermore, it does not provide a fully automated solution from the original point cloud input to the final 3D model output.

[0005] In summary, existing technologies cannot simultaneously meet the requirements of high automation, high precision, strong robustness, and wide scene adaptability for 3D modeling of power pipelines. Therefore, there is an urgent need to develop a fully automated method for semantic recognition and 3D modeling of power pipeline point clouds. Summary of the Invention

[0006] To address the technical deficiencies in the existing technologies, this invention provides a point cloud semantic recognition and 3D modeling method for power pipelines based on PointNet++. By integrating improved deep learning semantic recognition technology with parametric modeling technology driven by prior geometric knowledge in power engineering, the method achieves full automation of the 3D modeling process for power pipelines, significantly improving modeling efficiency and accuracy, and enhancing adaptability to complex scenarios.

[0007] Specifically, the present invention includes four functional modules connected in sequence: point cloud preprocessing module, semantic recognition module, parametric fitting module, and topology reconstruction module.

[0008] The point cloud preprocessing module is used to denoise, register multiple site points, and downsample the original point cloud of power pipelines to generate a standardized point cloud dataset. The semantic recognition module is used to perform multi-component semantic segmentation on the standardized point cloud based on the improved PointNet++ deep learning network to obtain point cloud data with category labels. The parameter fitting module is used to combine the prior geometric knowledge base of power pipelines to perform component-level parameter fitting and optimization on the segmented component point clouds to generate a geometric model that conforms to power engineering standards. The topology reconstruction module is used to construct the connection relationship between various components according to the power engineering topology specifications, generate a complete 3D model of power pipelines, and support export in multiple common formats.

[0009] As a further improvement of the present invention, the point cloud preprocessing module specifically performs the following steps: Step 1, Point cloud denoising: First, a statistical filtering algorithm is used to calculate the average distance between each point and its k-neighboring points, and outliers with a distance greater than a preset threshold are removed. Then, a radius filtering algorithm is used to remove noise points with excessively low local density. Step 2, Multi-site point cloud registration: The FPFH feature points of each site cloud are extracted for initial matching to obtain an initial transformation matrix. Then, the ICP algorithm is used for fine registration to control the registration error within 3mm. Step 3, Point cloud downsampling: A voxel downsampling method is used, with the voxel size set to 0.05m. While fully preserving the geometric features of the point cloud, the amount of point cloud data is reduced to 10%-20% of the original data to generate a standardized point cloud dataset.

[0010] As a further improvement of the present invention, the semantic recognition module specifically performs the following steps: Step 4, Training Dataset Construction: Collect point cloud data of power pipelines in three typical scenarios: substations, cable corridors, and transmission lines. Label the category labels of 12 typical components, including cables, pipes, supports, joints, valves, elbows, flanges, insulators, switchgear, transformers, grounding electrodes, and busbars. Construct a multi-scenario, multi-component training dataset containing 5,000 labeled samples. Step 5, PointNet++ Network Improvement and Training: Introduce channel attention and spatial attention mechanisms into the SA layer of the original PointNet++ network to enhance the network's feature extraction capabilities for small components and complex structures. At the same time, adopt a multi-scale grouping strategy to improve the network's adaptability to components of different sizes. The network training uses the Adam optimizer with an initial learning rate of 0.001, a batch size of 16, and 100 training epochs. The cross-entropy loss function combined with the Dice loss function is used to solve the class imbalance problem. Step 6, Point Cloud Semantic Reasoning: Input the preprocessed standardized point cloud into the trained improved PointNet++ network and output a semantic segmentation point cloud with category labels.

[0011] As a further improvement of the present invention, the parametric fitting module specifically performs the following steps: Step 7, Geometric prior knowledge base construction: storing the standard geometric shape, size range, connection method and material attribute information of various pipeline components specified in the power industry standards DL / T5445-2010 "Design Standard for Power Engineering Cables" and GB50217-2018 "Design Standard for Power Engineering Cables"; Step 8, Component point cloud clustering: using the density-based DBSCAN algorithm, setting the neighborhood radius to 0.1m and the minimum number of samples to 10, the semantic segmentation point cloud of the same category is segmented into a single independent component point cloud; Step 9, Parametric model fitting and optimization: calling the corresponding geometric model from the geometric prior knowledge base according to the component category, wherein the pipeline adopts a cylindrical model, the support adopts a cuboid model, the joint adopts a sphere or torus model, and the elbow adopts an arc model, the least squares method is used to solve the model parameters, and the parameters are constrained and optimized in combination with the size range in the geometric prior knowledge base to generate a component-level parametric model that conforms to the engineering standard.

[0012] As a further improvement of the present invention, the topology reconstruction module specifically performs the following steps: Step 10, Component connection relationship identification: Based on the spatial position, geometric features, and connection rules in the geometric prior knowledge base of the components, the connection relationship between each component is determined; wherein, the connection between the pipe and the joint must meet the condition that the axis coincides and the end face distance is less than 0.1m, the connection between the pipe and the support must meet the condition that the pipe axis is perpendicular to the top surface of the support and the distance is less than 0.05m, and the connection between the cable and the switch cabinet must meet the condition that the cable end is located in the terminal area of ​​the switch cabinet; Step 11, Topology network construction: The pipeline topology is represented by a directed graph structure, with components as nodes and connection relationships as edges. The attributes of the edges include connection type, connection direction, connection strength, and other information to construct a complete pipeline topology network; Step 12, Model integration and export: The parametric models of each component are spatially spliced ​​according to the topological relationship, and attribute information such as material, color, number, and specification is added to generate a complete three-dimensional model of the power pipeline, which supports export to common three-dimensional formats such as IFC, OBJ, STL, and DWG.

[0013] Beneficial effects The PointNet++-based semantic recognition and 3D modeling method for power pipeline point clouds described in this invention integrates deep learning semantic recognition technology with parametric modeling technology driven by prior geometric knowledge in power engineering. This constructs a fully automated processing system from raw point cloud input to complete 3D model output, achieving the following significant technical advantages: The level of automation has been greatly improved. Compared with traditional manual modeling methods, this invention reduces manual intervention by more than 80%, shortening the modeling cycle of a single 110kV substation cable layer from 15 days to less than 3 days, a reduction of more than 70%, significantly reducing the manpower and time costs of 3D modeling of power pipelines.

[0014] The modeling accuracy has reached an industry-leading level. The improved PointNet++ network achieves an average semantic segmentation accuracy of 95.2% for 12 typical components of power pipelines, and the dimensional error of the parametric fitting model is controlled within ±5mm, fully meeting the accuracy requirements of the DL / T5445-2010 standard for power engineering.

[0015] The robustness of the algorithm is significantly enhanced. Through multi-stage denoising, power engineering geometric prior constraints, and model parameter optimization, this invention effectively solves the modeling challenges caused by point cloud noise, component occlusion, and mixed arrangement of different components. Even with a point cloud missing rate of 30%, it can still maintain a modeling success rate of over 90%.

[0016] Comprehensive scenario adaptability. This invention can be widely applied to various power pipeline scenarios such as substations, cable corridors, and transmission lines. It also supports point cloud data processing from multiple data sources, including terrestrial 3D laser scanners, UAV LiDAR, and mobile measurement systems, making it widely applicable.

[0017] This invention fills a technological gap in the industry. It is the first to propose a full-process parametric modeling method based on PointNet++ point cloud semantic recognition combined with power engineering geometric prior driving, which solves the technical deficiency of existing technologies that can only achieve modeling of single components or single scenes, and provides core technical support for the digital construction of power pipelines. Attached Figure Description

[0018] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a schematic diagram of the overall process of the point cloud semantic recognition and 3D modeling method for power pipelines based on PointNet++ described in this invention. Figure 2 This is a detailed flowchart of the semantic recognition module and the parameterized fitting module described in this invention; Figure 3 This is a schematic diagram of the directed graph topology network structure of the topology reconstruction module described in this invention.

[0019] In the attached diagram: Point cloud preprocessing module 1, semantic recognition module 2, parametric fitting module 3, and topology reconstruction module 4. Detailed Implementation

[0020] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the invention, and should not be construed as limiting the invention. Rather, embodiments of the invention include all variations, modifications, and equivalents falling within the spirit and scope of the appended claims.

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

[0022] The point cloud semantic recognition and 3D modeling method for power pipelines based on PointNet++ described in this invention can be applied to 3D modeling of various power pipeline scenarios such as substations, cable corridors, and transmission lines, and can provide reliable 3D data support for power engineering design, construction, operation and maintenance, and digital twin construction.

[0023] like Figure 1 and Figure 2 As shown, this invention mainly includes four core parts: a point cloud preprocessing module, a semantic recognition module, a parametric fitting module, and a topology reconstruction module. These modules are connected in sequence to complete the automated processing from the original point cloud to a complete 3D model.

[0024] The core function of the point cloud preprocessing module is to remove noise and redundant data from the original point cloud, achieve accurate stitching of multi-site point clouds, and ultimately generate a standardized point cloud dataset. In practical applications, the specific execution process of this module is as follows: Step 1: Point Cloud Denoising: First, a statistical filtering algorithm is used to calculate the average distance between each point and its 20 neighboring points. The distance threshold is set to the average distance plus twice the standard deviation, and outliers with distances greater than this threshold are removed. Then, a radius filtering algorithm is used, setting the search radius to 0.1m and the minimum number of neighboring points to 5, to remove noise points with excessively low local density. Step 2: Multi-site Point Cloud Registration: The FPFH feature points of each site cloud are extracted, and feature matching is performed using the RANSAC algorithm to obtain the initial transformation matrix. Then, the ICP algorithm is used for fine registration, controlling the final registration error within 3mm. Step 3: Point Cloud Downsampling: A voxel downsampling method is used, setting the voxel size to 0.05m. While fully preserving the geometric features of the point cloud, the amount of point cloud data is reduced to 10%-20% of the original data, generating a standardized point cloud dataset.

[0025] The core function of the semantic recognition module is to achieve accurate semantic segmentation of various components of power pipelines. In a specific embodiment of the present invention, the specific execution process of this module is as follows: Step 4, Training Dataset Construction: Collect point cloud data of power pipelines in three typical scenarios: substations, cable corridors, and transmission lines. Label the category labels of 12 typical components, including cables, pipes, supports, joints, valves, elbows, flanges, insulators, switchgear, transformers, grounding electrodes, and busbars. Construct a multi-scenario, multi-component training dataset containing 5000 sets of labeled samples, with the ratio of training set, validation set, and test set being 7:2:1. Step 5, PointNet++ Network Improvement and Training: Introduce channel attention and spatial attention mechanisms into the SA layer of the original PointNet++ network. The channel attention mechanism enhances the network's ability to extract key features by adaptively learning the weights of each feature channel; the spatial attention mechanism enhances the network's attention to small components by adaptively learning the weights of each spatial location. At the same time, a multi-scale grouping strategy is adopted, setting three different scale grouping radii of 0.1m, 0.2m, and 0.4m to improve the network's adaptability to components of different sizes. The network training uses the Adam optimizer with an initial learning rate of 0.001, which decays to 0.5 times every 20 epochs. The batch size is 16, and the training epochs are 100. The loss function combines cross-entropy loss and Dice loss with a 1:1 weight ratio to address class imbalance during training. After training, the network achieves an average segmentation accuracy of 95.2% for each type of component. Step 6: Point Cloud Semantic Inference: The preprocessed standardized point cloud is input into the trained improved PointNet++ network, outputting a semantically segmented point cloud with class labels.

[0026] The core function of the parametric fitting module is to convert the segmented point cloud into a parametric geometric model that conforms to power engineering specifications. In a specific embodiment of the present invention, the specific execution process of this module is as follows: Step 7, Geometric Prior Knowledge Base Construction: Store information such as the standard geometric shape, size range, connection method, and material properties of various pipeline components specified in the power industry standards DL / T5445-2010 "Design Standard for Power Engineering Cables" and GB50217-2018 "Design Standard for Power Engineering Cables". For example, the diameter range of DN100 cable pipe is 100mm±5mm, and the height range of the support is 200mm-1000mm. Step 8, Component Point Cloud Clustering: Use the density-based DBSCAN algorithm, set the neighborhood radius to 0.1m, and the minimum number of samples to 10, to segment the semantically segmented point clouds of the same category into individual independent component point clouds. Step 9: Parametric Model Fitting and Optimization: Based on the component category, the corresponding geometric model is retrieved from the geometric prior knowledge base. For example, pipes use a cylindrical model, and the axis and radius of the cylinder are solved using the least squares method; supports use a cuboid model, and the length, width, height, and spatial position of the cuboid are solved using the least squares method; joints use a sphere or torus model; and elbows use an arc model. After obtaining the model parameters, the parameters are constrained and optimized using the size range in the geometric prior knowledge base to ensure that the generated model dimensions conform to power engineering standards.

[0027] The core function of the topology reconstruction module is to construct the correct connection relationships between various components and generate a complete 3D model of the power pipeline. In a specific embodiment of the present invention, the specific execution process of this module is as follows: Step 10, Component connection relationship identification: Based on the spatial position, geometric features, and connection rules in the geometric prior knowledge base of the components, the connection relationships between the components are determined. Among them, the connection between the pipe and the joint must meet the condition that the axis coincides and the end face distance is less than 0.1m; the connection between the pipe and the support must meet the condition that the pipe axis is perpendicular to the top surface of the support and the distance is less than 0.05m; the connection between the cable and the switch cabinet must meet the condition that the cable end is located in the terminal area of ​​the switch cabinet. Step 11, Topology network construction: The pipeline topology is represented by a directed graph structure, with each pipeline component as a node and the connection relationship between the components as an edge. The attributes of the edge include information such as connection type, connection direction, and connection strength, to construct a complete pipeline topology network. Step 12, Model Integration and Export: Spatially stitch together the parametric models of each component according to the topological relationship, add attribute information such as material, color, number, specification, manufacturer, and installation date, generate a complete 3D model of the power pipeline, and support exporting to common 3D formats such as IFC, OBJ, STL, and DWG.

[0028] To more clearly illustrate the technical solution of the present invention, the specific implementation steps of the present invention will be described in detail below using the modeling of power pipelines in the indoor cable layer of a 110kV substation as an example: Data acquisition: The indoor cable layer of the substation was scanned using a FAROFocus S350 terrestrial 3D laser scanner. A total of 12 stations were acquired, with a total of approximately 230 million points and a scanning accuracy of ±2mm.

[0029] Point cloud preprocessing: Steps 1 and 2 are executed sequentially to denoise the point cloud, removing approximately 12 million noise points; fine registration of 12 point cloud points is completed, with a final registration error of 2.1 mm; Step 3 is executed to perform voxel downsampling, reducing the amount of point cloud data to approximately 21 million.

[0030] Semantic recognition: The preprocessed point cloud is input into the trained improved PointNet++ network, and step 6 is performed to perform semantic reasoning, resulting in semantic segmentation results for 12 types of components. The overall segmentation accuracy is 94.8%, of which the pipe segmentation accuracy is 96.2%, the bracket segmentation accuracy is 95.7%, and the joint segmentation accuracy is 93.5%.

[0031] Parametric fitting: Step 7 is executed to construct a geometric prior knowledge base containing information on commonly used pipeline components of the substation; Step 8 is executed to perform component clustering on the segmented point cloud, identifying a total of 326 independent components; Step 9 is executed to perform parametric fitting and optimization on each component, with an average error of 3.2 mm for parametric fitting.

[0032] Topology Reconstruction: Step 10 identifies the connection relationships between the components, identifying a total of 412 connection edges; Step 11 constructs the pipeline topology network; Step 12 integrates all component models according to the topological relationships to generate a complete 3D model of the cable layer power pipeline, and exports it as an IFC format file, with a file size of approximately 12MB.

[0033] Understandably, this invention addresses the practical needs and pain points of existing technologies in power pipeline modeling. It constructs a fully automated modeling solution based on three core objectives: improving automation, enhancing modeling accuracy, and increasing scenario adaptability. This invention can be applied not only to the as-built modeling of power pipelines in newly built substations, rapidly generating 3D models that meet engineering standards; but also to the operation and maintenance modeling of in-service cable tunnels, enabling digital management of pipeline assets; and further, to the modeling of transmission line towers and pipelines, providing foundational data for digital twin construction.

[0034] It should be noted that in practical applications, parameters such as the point cloud downsampling voxel size, semantic segmentation threshold, and parametric fitting accuracy can be flexibly adjusted according to specific modeling needs to adapt to power pipeline modeling with different accuracy requirements and scenarios. Furthermore, this invention can also be extended to support the processing of multiple data sources such as UAV LiDAR point clouds and mobile measurement system point clouds, further improving the flexibility of data acquisition and modeling efficiency.

[0035] In summary, by integrating deep learning semantic recognition technology with parametric modeling technology driven by geometric priors in power engineering, this invention effectively solves the problems of low automation, poor accuracy, and weak robustness in existing 3D modeling of power pipelines, and realizes full-process automation of 3D modeling of power pipelines, which has broad engineering application value.

[0036] It should be noted that in the description of this specification, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0037] In the description of this specification, the references to "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0038] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for semantic recognition and 3D modeling of power pipeline point clouds based on PointNet++, characterized in that, It includes a point cloud preprocessing module (1), a semantic recognition module (2), a parameterized fitting module (3), and a topology reconstruction module (4) connected in sequence; the point cloud preprocessing module (1) is used to complete the denoising, multi-site cloud registration and downsampling of the original point cloud of the power pipeline, and generate a standardized point cloud dataset; The semantic recognition module (2) is used to perform multi-component semantic segmentation on standardized point clouds based on the improved PointNet++ deep learning network to obtain point cloud data with category labels; The parametric fitting module (3) is used to combine the prior knowledge base of power pipeline geometry to perform component-level parametric fitting and optimization on the segmented component point clouds, and generate a geometric model that conforms to the power engineering standard; the topology reconstruction module (4) is used to construct the connection relationship between each component according to the power engineering topology specification, generate a complete three-dimensional model of power pipeline and support multi-format export.

2. The method for semantic recognition and 3D modeling of power pipeline point clouds based on PointNet++ according to claim 1, characterized in that, The point cloud preprocessing module (1) specifically performs the following steps: Step 1, Point Cloud Denoising: A combination of statistical filtering and radius filtering is used to remove outliers and local noise points from the original point cloud. Step 2, Multi-site cloud registration: Extract FPFH feature points from each site cloud to perform initial matching and obtain the initial transformation matrix. Then, use the ICP algorithm for fine registration to control the registration error within 3mm. Step 3, Point Cloud Downsampling: Using the voxel downsampling method, the voxel size is set to 0.05m. While preserving the geometric features of the point cloud, the data volume is reduced to 10%-20% of the original data to generate a standardized point cloud dataset.

3. The method for semantic recognition and 3D modeling of power pipeline point clouds based on PointNet++ according to claim 1, characterized in that, The semantic recognition module (2) specifically performs the following steps: Step 4: Training Dataset Construction: Collect point cloud data of power pipelines in three typical scenarios: substations, cable corridors, and transmission lines. Label the category labels of 12 typical components and construct a multi-scenario, multi-component training dataset containing 5,000 sets of labeled samples. Step 5, PointNet++ Network Improvement and Training: Introduce channel attention and spatial attention mechanisms into the SA layer of the original PointNet++ network, adopt a multi-scale grouping strategy to improve feature extraction capabilities, and train the network using the constructed training dataset. Step 6, Point Cloud Semantic Reasoning: Input the preprocessed standardized point cloud into the trained improved PointNet++ network, and output a semantic segmentation point cloud with category labels.

4. The method for semantic recognition and 3D modeling of power pipeline point clouds based on PointNet++ according to claim 3, characterized in that, In step 5, the network training uses a combination of cross-entropy loss function and Dice loss function with a weight ratio of 1:1; the Adam optimizer is used, with an initial learning rate of 0.001, which decays to 0.5 times every 20 rounds; the batch size is 16, and the training rounds are 100.

5. The method for semantic recognition and 3D modeling of power pipeline point clouds based on PointNet++ according to claim 1, characterized in that, The parameterized fitting module (3) specifically performs the following steps: Step 7: Construction of the geometric prior knowledge base: Store the standard geometric shapes, size ranges, connection methods, and material properties of various pipeline components specified in power industry standards; Step 8, Component point cloud clustering: The density-based DBSCAN algorithm is used, with a neighborhood radius of 0.1m and a minimum number of samples of 10, to segment semantic segmentation point clouds of the same category into individual component point clouds; Step 9: Parametric Model Fitting and Optimization: Based on the component category, the corresponding geometric model is retrieved from the geometric prior knowledge base. The least squares method is used to solve the model parameters, and the parameters are optimized in combination with geometric prior constraints to generate a component-level parametric model that conforms to engineering standards.

6. The method for semantic recognition and 3D modeling of power pipeline point clouds based on PointNet++ according to claim 5, characterized in that, In step 9, the geometric model includes a cylindrical model, a cuboid model, a sphere model, a toroidal model, and an arc model, which correspond to various power pipeline components such as pipes, supports, joints, flanges, and elbows, respectively.

7. The method for semantic recognition and 3D modeling of power pipeline point clouds based on PointNet++ according to claim 1, characterized in that, The topology reconstruction module (4) specifically performs the following steps: Step 10: Component connection relationship identification: Based on the spatial position, geometric features, and connection rules in the geometric prior knowledge base of the components, determine the connection relationship between each component; Step 11: Topology network construction: The pipeline topology is represented by a directed graph structure, with components as nodes and the connection relationships between components as edges. The attributes of the edges include connection type, connection direction, and connection strength information to construct a complete pipeline topology network. Step 12, Model Integration and Export: Spatially stitch together the parametric models of each component according to the topological relationship, add attribute information, generate a complete 3D model of the power pipeline, and support exporting to various general 3D formats.

8. The method for semantic recognition and 3D modeling of power pipeline point clouds based on PointNet++ according to claim 7, characterized in that, In step 10, the connection between the pipe and the joint must meet the conditions that the axes coincide and the distance between the end faces is less than 0.1m, the connection between the pipe and the support must meet the conditions that the pipe axis is perpendicular to the top surface of the support and the distance is less than 0.05m, and the connection between the cable and the switch cabinet must meet the condition that the cable end is located in the terminal area of ​​the switch cabinet.

9. The method for semantic recognition and 3D modeling of power pipeline point clouds based on PointNet++ according to claim 7, characterized in that, In step 12, the supported 3D formats for export include IFC, OBJ, STL, and DWG.

10. The method for semantic recognition and 3D modeling of power pipeline point clouds based on PointNet++ according to claim 1, characterized in that, The method supports the processing of multiple data sources, including terrestrial 3D laser scanners, UAV lidar, and mobile measurement systems.

Images