Distribution network tower material identification method based on spatial three-dimensional segmentation and point cloud group structure

Abstract

The invention discloses a distribution network tower material identification method based on spatial three-dimensional segmentation and a point cloud group structure, and the method comprises the steps: collecting and preprocessing distribution network tower and environment three-dimensional point cloud data, and constructing a material feature database; calculating the space distance between the points, and constructing an undirected point cloud picture; selecting seed points according to density, distribution or rules in the graph, wherein each seed point corresponds to a material category; carrying out random walk clustering by taking the seed point as a starting point, and completing space segmentation according to a probability maximum principle; constructing an improved three-dimensional Gaussian mixture model, training parameters, and generating fine-grained clustering labels; the clustering labels and the feature database are combined to complete type calibration, quantity statistics and spatial positioning; and outputting a structured result for material inventory, inspection and management. According to the method, the random walk clustering and the improved three-dimensional Gaussian mixture model are fused, so that high-precision identification and spatial positioning of the distribution network tower materials are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent distribution network material identification, and in particular to a distribution network tower material identification method based on spatial stereo segmentation and point cloud group structure. Background Art

[0002] With the continuous advancement of intelligence and automation in the power industry, efficient management and accurate identification of distribution towers and their associated materials have become critical components of smart distribution network construction. Traditionally, the identification, counting, and inspection of distribution tower materials rely primarily on manual inspections or two-dimensional image-based recognition methods. These methods are limited by significant subjectivity, poor environmental adaptability, and insufficient spatial information acquisition, making large-scale, accurate, and efficient asset management difficult. The rapid development of three-dimensional point cloud technology has provided new insights into the spatial modeling and intelligent identification of power facilities. Three-dimensional point cloud data captured by devices such as LiDAR and structured light sensors can fully reflect the spatial distribution and geometry of towers and surrounding materials, becoming a crucial data foundation for intelligent distribution network material identification.

[0003] However, existing three-dimensional point cloud material recognition methods generally have several technical bottlenecks. On the one hand, in view of the diverse types of materials, dense distribution, occlusion and overlap in complex distribution network scenarios, traditional methods based on single clustering, simple segmentation or rule matching are difficult to effectively separate various types of materials, and are prone to misclassification, omission or blurred boundaries. On the other hand, existing algorithms such as Gaussian mixture models and cluster segmentation mostly focus only on the spatial coordinates of point clouds, and fail to fully integrate spatial topological structures, local attribute differences and multi-scale hierarchical information, resulting in insufficient recognition capabilities for heterogeneous materials, fine-grained structures and spatial hierarchies. Most methods lack adaptability and innovation in model structure, making it difficult to take into account global and local, macro and micro feature modeling, which limits recognition accuracy and engineering promotion in practical applications.

[0004] Therefore, how to provide a distribution network tower material identification method based on spatial stereo segmentation and point cloud group structure is a problem that technicians in this field urgently need to solve. Summary of the Invention

[0005] One objective of this invention is to propose a distribution network tower material identification method based on spatial stereo segmentation and point cloud clustering. This method integrates technologies such as 3D point cloud data processing, spatial segmentation, random walk clustering, and a multi-structure improved 3D Gaussian mixture model. It describes in detail how to achieve automated, high-precision identification and spatial positioning of distribution network towers and their associated materials through spatial modeling, fine-grained clustering, and intelligent comparison of normalized point cloud data. This method boasts the advantages of fully utilizing spatial information, high recognition accuracy, strong adaptability to complex scenarios, and a high degree of automation. It can significantly improve the efficiency and intelligence of distribution network asset inventory, intelligent inspection, and operation and maintenance management.

[0006] A distribution network tower material identification method based on spatial stereo segmentation and point cloud group structure according to an embodiment of the present invention includes: Collect 3D point cloud data of distribution network towers and their surroundings, pre-process the 3D point cloud data to obtain a normalized point cloud dataset, and build a distribution network material feature database; Based on the normalized point cloud dataset, the spatial distance between each point and other points in the neighborhood is calculated to construct an undirected point cloud graph. Each point is used as a node of the undirected point cloud graph, the neighborhood relationship is used as the edge, and the edge weight is defined by the spatial distance. According to the spatial density, geometric distribution or manually set rules of the points in the undirected point cloud, a group of seed points are selected as the starting nodes of random walk clustering, and various sub-points correspond to different material categories; Starting from the seed point, random walk clustering is performed on the undirected point cloud map. The random walk probability of each unclassified point to the seed point of each category is calculated, and each point is assigned to the category with the largest probability value to obtain the spatial segmentation result. For each spatial segmentation area in the spatial segmentation result, a corresponding improved three-dimensional Gaussian mixture model is established, the parameters of the three-dimensional Gaussian mixture model are trained, and fine-grained clustering labels are output; Based on fine-grained clustering labels and combined with the distribution network material feature database, the point cloud data of each spatial segmentation area is calibrated, the quantity is counted, and the spatial positioning is performed to obtain the material category, quantity, and spatial location information; The material category, quantity and spatial location information are output in the form of structured results for inventory, inspection and asset management of distribution network tower materials.

[0007] Optionally, the three-dimensional point cloud data specifically includes spatial coordinate information, reflection intensity information and color information of the distribution network tower and surrounding materials in three-dimensional space.

[0008] Optionally, the preprocessing of the three-dimensional point cloud data specifically includes noise removal, downsampling, point cloud registration and coordinate normalization.

[0009] Optionally, the construction of the distribution network material feature database specifically refers to combining manual labeling results and historical material information to extract geometric features, analyze spatial distribution and organize attribute information of different distribution network materials, summarize and classify the typical three-dimensional shapes, dimensional parameters, spatial structure characteristics and related attributes of various materials, and form a distribution network material feature database that can be used for material identification and comparison.

[0010] Optionally, the method of calculating the spatial distance between each point and other points in the neighborhood based on the normalized point cloud dataset, constructing an undirected point cloud graph, using each point as a node of the undirected point cloud graph, using neighborhood relationships as edges, and defining edge weights using spatial distances includes: For each point in the normalized point cloud dataset, the neighborhood point set of each point is determined by the radius neighborhood algorithm; For each point and other points in the neighborhood, calculate the Euclidean distance between the three-dimensional space coordinates. The Euclidean distance is the square root of the sum of the squares of the distance differences in the three-dimensional coordinate axis direction, reflecting the actual distance between the two points in three-dimensional space; Each point in the normalized point cloud dataset is used as a node of the undirected point cloud graph, and each pair of points that satisfy the neighborhood relationship is used as an edge of the undirected point cloud graph. According to the calculated Euclidean distance, an edge weight is assigned to each edge by setting a distance decay rule. The closer the distance, the greater the edge weight, and the farther the distance, the smaller the edge weight. This completes the construction of the nodes, edges and edge weights of the undirected point cloud graph.

[0011] Optionally, a group of seed points are selected as starting nodes of random walk clustering according to the spatial density, geometric distribution or manually set rules of the points in the undirected point cloud image, and various sub-points correspond to different material categories, including: Based on the undirected point cloud graph, the spatial density of each node is calculated. The spatial density is the number of neighboring nodes within a preset radius centered on the node. Sort all nodes from high to low according to spatial density, and preliminarily select nodes with spatial density greater than the set threshold as candidate nodes; Screen all candidate nodes for spatial uniformity; According to manually set rules or historical recognition experience, the seed points after spatial uniformity screening are combined with spatial position and point cloud features to calibrate the corresponding material categories. Each material category contains at least one seed point, and the seed point corresponding to each material category is used as the starting node of random walk clustering.

[0012] Optionally, obtaining the spatial segmentation result includes: The seed points corresponding to the obtained material categories are used as the starting nodes, and the category labels of each node are initialized on the undirected point cloud graph. Initially, only the seed points are assigned category labels, and the other nodes are in an unclassified state. On the undirected point cloud graph, for each unclassified node, the random walk probability to the seed point of each category is calculated based on the transition probability defined by the edge weight; For each unclassified node, set the category label to the category corresponding to the maximum random walk probability to complete the initial space segmentation of the current round; The preliminary spatial segmentation results are evaluated through self-supervised feedback. Based on the local consistency of the node category, neighborhood confidence, and boundary uncertainty, nodes with fuzzy boundaries and uncertain classification are automatically identified and temporarily marked as pseudo-labeled seed points. Append the pseudo-labeled seed points as a new seed set, combine the original seed points, and re-perform random walk clustering on the undirected point cloud. For all unclassified nodes and pseudo-labeled seed points, recalculate the random walk probability to each category seed point and update the category label. After the preliminary spatial segmentation results are corrected through self-supervised feedback, the pseudo-labeled seed points are appended as a new seed set and combined with the original seed points. Random walk clustering is performed again on the undirected point cloud graph, and the category assignment process is continuously iterated until the category labels of all nodes remain stable or the preset maximum number of iterations is reached, and the spatial segmentation results are finally output.

[0013] Optionally, the output fine-grained clustering labels include: For each spatial segmentation region in the spatial segmentation result, extracting point cloud data within the spatial segmentation region, wherein the point cloud data includes three-dimensional spatial coordinate information as input data of the improved three-dimensional Gaussian mixture model; The input point cloud data is input into the topological Gaussian module, the heterogeneous covariance module and the hierarchical fusion module respectively. The topological Gaussian module includes a topological feature extraction unit and a feature fusion unit, the heterogeneous covariance module includes a covariance type discrimination unit and a covariance parameter generation unit, and the hierarchical fusion module includes a multi-scale initialization unit and a hierarchical probability fusion unit. The three modules serve as substructures of the improved three-dimensional Gaussian mixture model. In the topological Gaussian module, the topological feature extraction unit takes point cloud data and radius neighborhood connection information as input to calculate the local topological features of each point. The feature fusion unit fuses the local topological features with the spatial coordinate information of the point and outputs a topological fusion feature set. In the heterogeneous covariance module, the covariance type discrimination unit takes the topological fusion feature set as input and combines the spatial characteristics of the point cloud distribution to discriminate the covariance type corresponding to each Gaussian component. The covariance parameter generation unit automatically generates covariance parameters based on the covariance type and related point set, and outputs the covariance type and parameter set of each Gaussian component. In the hierarchical fusion module, the multi-scale initialization unit takes the covariance type and parameter set and point cloud data as input to initialize Gaussian components at different spatial scales. The hierarchical probability fusion unit fuses the component parameters and attribution probabilities of each spatial scale and outputs a multi-scale Gaussian component parameter set. Taking the multi-scale Gaussian component parameter set as the initial parameters, combined with the topological fusion features and covariance parameters, the expectation maximization algorithm is used within the overall framework of the three-dimensional Gaussian mixture model to iteratively optimize the mean, covariance, and weight of each component, and output the final converged Gaussian component parameter set; According to the final converged Gaussian component parameter set, for each point cloud data point in the spatial segmentation area, the attribution probability under all Gaussian components is calculated, and fine-grained clustering labels are assigned according to the maximum probability principle. The fine-grained clustering labels and corresponding Gaussian component parameters of each point in the spatial segmentation area are output for fine material differentiation and category calibration.

[0014] Optionally, obtaining the material category, quantity, and spatial location information includes: According to the output fine-grained cluster labels, the point cloud data in the spatial segmentation area are grouped according to the cluster labels, and each group corresponds to a Gaussian component or fine-grained category; For each fine-grained category, the spatial coordinate features and distribution features of the point cloud and the standard material geometric features, spatial structure features and attribute information in the distribution network material feature database are extracted to construct the point cloud feature vector and the database feature vector; Calculate the similarity between the point cloud feature vector and the database feature vector, the similarity is based on cosine similarity, and select the one with the largest similarity as the category matching basis; Each fine-grained category is matched with the most similar material category in the database, the material type of the point cloud category is calibrated, and the number of point cloud data and the centroid coordinates or boundary range in three-dimensional space are recorded as the basis for quantity statistics and spatial positioning; The material type, quantity and spatial location information of all fine-grained categories are summarized to obtain the structured results of material category, quantity and spatial location information within the spatial segmentation area.

[0015] Optionally, the material category, quantity and spatial location information are output in the form of structured results for inventory, inspection and asset management of distribution network tower materials. Specifically, the name, corresponding quantity, three-dimensional spatial coordinates or spatial range, category label and related attribute information of each type of material that has been automatically identified are organized into a standardized data format, and output as a table, database record or interface data for direct call by automatic inventory, inspection scheduling and asset management.

[0016] The beneficial effects of the present invention are: This invention improves the automated identification and management of distribution network towers and their materials in complex scenarios by integrating spatial segmentation, random walk clustering, and a structurally innovative 3D Gaussian mixture model. Compared to existing approaches that rely solely on spatial clustering or traditional Gaussian mixture models, this invention not only fully exploits the spatial topological relationships and multi-scale structural characteristics of 3D point clouds, but also achieves collaborative identification of local details and overall distribution of materials through modular innovation, effectively addressing issues such as difficulty distinguishing dense, overlapping, and heterogeneous materials, blurred boundaries, and insufficient spatial hierarchy.

[0017] The method of the present invention can output the category, quantity and spatial position of distribution network towers and their ancillary materials in a high-precision and structured manner without requiring a large amount of manual intervention, thereby improving the automation and accuracy of asset inventory and inspection.

[0018] The present invention has good adaptability and engineering scalability, and can provide a strong data foundation and technical support for the operation and maintenance, asset management and intelligent decision-making of smart distribution networks. It has significant practical application value and promotion prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 This is a flow chart of the distribution network tower material identification method based on spatial stereo segmentation and point cloud group structure proposed by the present invention; Figure 2 This is a schematic structural diagram of a three-dimensional Gaussian mixture model for the distribution network tower material identification method based on spatial stereo segmentation and point cloud group structure proposed by the present invention; Figure 3 This is a schematic diagram of the point cloud segmentation and recognition results of the distribution network tower material recognition method based on spatial stereo segmentation and point cloud group structure proposed in the present invention. DETAILED DESCRIPTION

[0020] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0021] refer to Figure 1 、 Figure 2 and Figure 3 , a distribution network tower material recognition method based on spatial stereo segmentation and point cloud group structure, including: Collect 3D point cloud data of distribution network towers and their surroundings, pre-process the 3D point cloud data to obtain a normalized point cloud dataset, and build a distribution network material feature database; Based on the normalized point cloud dataset, the spatial distance between each point and other points in the neighborhood is calculated to construct an undirected point cloud graph. Each point is used as a node of the undirected point cloud graph, the neighborhood relationship is used as the edge, and the edge weight is defined by the spatial distance. According to the spatial density, geometric distribution or manually set rules of the points in the undirected point cloud, a group of seed points are selected as the starting nodes of random walk clustering, and various sub-points correspond to different material categories; Starting from the seed point, random walk clustering is performed on the undirected point cloud map. The random walk probability of each unclassified point to the seed point of each category is calculated, and each point is assigned to the category with the largest probability value to obtain the spatial segmentation result. For each spatial segmentation area in the spatial segmentation result, a corresponding improved three-dimensional Gaussian mixture model is established, the parameters of the three-dimensional Gaussian mixture model are trained, and fine-grained clustering labels are output; Based on fine-grained clustering labels and combined with the distribution network material feature database, the point cloud data of each spatial segmentation area is calibrated, the quantity is counted, and the spatial positioning is performed to obtain the material category, quantity, and spatial location information; The material category, quantity and spatial location information are output in the form of structured results for inventory, inspection and asset management of distribution network tower materials.

[0022] In this embodiment, the three-dimensional point cloud data specifically includes spatial coordinate information, reflection intensity information and color information of the distribution network tower and surrounding materials in three-dimensional space.

[0023] In this embodiment, the preprocessing of the three-dimensional point cloud data specifically includes noise removal, downsampling, point cloud registration and coordinate normalization.

[0024] In this embodiment, the construction of the distribution network material feature database specifically refers to combining manual labeling results and historical material information to extract geometric features, analyze spatial distribution and organize attribute information of different distribution network materials, summarize and classify the typical three-dimensional shapes, dimensional parameters, spatial structure characteristics and related attributes of various materials, and form a distribution network material feature database that can be used for material identification and comparison.

[0025] In this embodiment, based on the normalized point cloud dataset, the spatial distance between each point and other points in the neighborhood is calculated to construct an undirected point cloud graph. Each point is used as a node of the undirected point cloud graph, the neighborhood relationship is used as an edge, and the edge weight is defined by the spatial distance, including: For each point in the normalized point cloud dataset, the neighborhood point set of each point is determined by the radius neighborhood algorithm, specifically: Set a predetermined neighborhood radius threshold for each point; For each point, traverse all other points in the normalized point cloud dataset and calculate the Euclidean distance between the three-dimensional space coordinates of the point and other points; All points whose Euclidean distance to the point is not greater than the neighborhood radius threshold are classified into the neighborhood point set of the point; For each point and other points in the neighborhood, calculate the Euclidean distance between the three-dimensional space coordinates. The Euclidean distance is the square root of the sum of the squares of the distance differences in the three-dimensional coordinate axis direction, reflecting the actual distance between the two points in three-dimensional space; Each point in the normalized point cloud dataset is used as a node of the undirected point cloud graph, and each pair of points that satisfy the neighborhood relationship is used as an edge of the undirected point cloud graph. According to the calculated Euclidean distance, an edge weight is assigned to each edge by setting a distance decay rule. The closer the distance, the greater the edge weight, and the farther the distance, the smaller the edge weight. This completes the construction of the nodes, edges and edge weights of the undirected point cloud graph.

[0026] The present invention constructs an undirected point cloud map using a radius neighborhood algorithm based on a normalized point cloud dataset, which not only achieves fine modeling of spatial structure but also improves the accuracy of material segmentation and clustering. Unlike traditional point cloud connection methods that are based only on global distance or fixed adjacency relationships, this method sets a predetermined neighborhood radius for each point and dynamically screens point pairs with close relationships in the actual space, ensuring that the neighborhood connection can not only reflect the true physical structure of the object but also filter out irrelevant redundant connections, thereby enhancing the spatial expression capability of the point cloud map. The Euclidean distance and distance decay rules are used to assign weights to edges, which quantifies the strength of spatial relationships and enables clustering and subsequent segmentation to fully utilize local geometric features. The present invention not only introduces spatial constraints in the selection of nodes and edges, but also provides a solid data foundation and a controllable structural framework for spatial segmentation, clustering and material identification through a flexible edge weight allocation mechanism, improving the accuracy and robustness of material boundary identification and spatial topology modeling in complex environments, and effectively supporting the automated intelligent identification of multiple categories of materials in point cloud scenarios.

[0027] In this embodiment, a group of seed points are selected as the starting nodes of random walk clustering according to the spatial density, geometric distribution or manually set rules of the points in the undirected point cloud image. Various sub-points correspond to different material categories, including: Based on the undirected point cloud graph, the spatial density of each node is calculated. The spatial density is the number of neighboring nodes within a preset radius centered on the node. Sort all nodes from high to low according to spatial density, and preliminarily select nodes with spatial density greater than the set threshold as candidate nodes; All candidate nodes are screened for spatial uniformity, specifically: Traverse the candidate nodes in sequence, select the first candidate node as the seed point, and then only retain the candidate nodes whose 3D Euclidean distance to the selected seed point is greater than the preset minimum distance. This is done until all candidate nodes are traversed, ensuring that the selected seed points are evenly distributed in space and are spaced apart by more than the minimum distance. According to manually set rules or historical recognition experience, the seed points after spatial uniformity screening are combined with spatial position and point cloud features to calibrate the corresponding material categories. Each material category contains at least one seed point, and the seed points corresponding to each material category are used as the starting nodes of random walk clustering. The manually set rules refer to pre-defining the typical distribution positions, shape characteristics or size thresholds of different material categories in space based on the actual distribution specifications or industry standards of distribution network pole tower materials. For example, insulators are usually located at both ends of crossarms and guy wires are commonly found near the tower base. Experience is used to assist in determining the category to which the seed points belong. Historical recognition experience refers to the accumulated knowledge of the association between material categories and spatial characteristics and point cloud attributes based on the manual annotation results of a large amount of previous point cloud data or verified recognition cases. For example, the statistical laws of the spatial density, geometric characteristics and reflection intensity characteristics of a certain type of hardware are known. Combined with the distribution patterns of similar materials in historical data, the newly identified seed points are calibrated.

[0028] In this embodiment, obtaining the spatial segmentation result includes: The seed points corresponding to the obtained material categories are used as the starting nodes, and the category labels of each node are initialized on the undirected point cloud graph. Initially, only the seed points are assigned category labels, and the other nodes are in an unclassified state. On the undirected point cloud graph, for each unclassified node, the random walk probability to each category seed point is calculated based on the transition probability defined by the edge weight. The random walk probability is the probability distribution of starting from the unclassified node and following the path with the largest edge weight to finally reach each category seed point. The random walk probability to each category seed point is calculated as follows: According to the edge weights between each node in the undirected point cloud graph, the edge weights of each pair of adjacent nodes are normalized to obtain the transition probability matrix between nodes. The probability of each node transferring to a neighboring node is equal to the ratio of the edge weight to the sum of all outgoing edge weights. Assign labels of corresponding categories to all seed points, initialize the random walk probability of the seed points to 1, and the random walk probability of other unclassified nodes to 0; For each unclassified node, the cumulative probability of each node to each category seed point is continuously updated in an iterative manner according to the transition probability matrix and the probability distribution of known nodes until convergence or the preset number of iterations is reached, and finally the probability distribution of each unclassified node to each category seed point is obtained; For each unclassified node, set the category label to the category corresponding to the maximum random walk probability to complete the initial space segmentation of the current round; The preliminary spatial segmentation results are evaluated through self-supervisory feedback. Based on the local consistency of the node category, neighborhood confidence, and boundary uncertainty, nodes with fuzzy boundaries and uncertain classification are automatically identified and temporarily marked as pseudo-label seed points. The local consistency of the node category refers to the degree to which the node belongs to the same category as most nodes in the spatial neighborhood in the spatial segmentation result. If the category label of the node is consistent with that of most nodes in the neighborhood, the local consistency of the node is considered to be high. Conversely, if there are many nodes of different categories in the neighborhood, the local consistency of the node is low. The neighborhood confidence refers to the proportion of the category to which the node belongs that is consistent with the category label of the nodes in the neighborhood, which is used to measure the reliability of the node category determination. The boundary uncertainty refers to the fact that the probability distribution of the node's belonging category in spatial segmentation is relatively close or changes frequently, making it difficult to clearly determine the category to which it belongs, reflecting the uncertainty of the node in the boundary area of ​​different categories. Append the pseudo-labeled seed points as a new seed set, combine the original seed points, and re-perform random walk clustering on the undirected point cloud. For all unclassified nodes and pseudo-labeled seed points, recalculate the random walk probability to each category seed point and update the category label. After the preliminary spatial segmentation results are corrected through self-supervised feedback, the pseudo-labeled seed points are appended as a new seed set and combined with the original seed points. Random walk clustering is performed again on the undirected point cloud graph, and the category assignment process is continuously iterated until the category labels of all nodes remain stable or the preset maximum number of iterations is reached, and the spatial segmentation results are finally output.

[0029] This invention, by introducing a random walk-based spatial segmentation and self-supervised feedback mechanism on undirected point cloud graphs, achieves automatic material classification and boundary optimization, improving the accuracy and intelligence of point cloud segmentation. Unlike traditional spatial segmentation methods, this method not only utilizes edge weight normalization and a transition probability matrix to efficiently simulate information transfer between points, enabling the orderly propagation of category labels within the spatial structure, but also enhances the ability to model complex spatial relationships through the cumulative iteration of random walk probabilities. Using the triple criteria of local consistency, neighborhood confidence, and boundary uncertainty, it automatically filters nodes with ambiguous classification boundaries and uncertain category determination, forms pseudo-label seed points, and dynamically expands clustering starting points, enabling self-correction and refinement of segmentation labels. The self-supervised feedback correction mechanism effectively addresses the misclassification and omission issues that traditional methods are prone to in dense multi-material environments, overlapping boundaries, and weak categories, improving the sensitivity and robustness of segmentation results to boundaries and details. The overall process possesses excellent adaptability and automation capabilities, providing more reliable basic data support for material identification and spatial analysis, with significant engineering practical value and technological innovation.

[0030] In this embodiment, the outputting of fine-grained cluster labels includes: For each of the spatial segmentation regions in the spatial segmentation result, point cloud data in the spatial segmentation region is extracted, the point cloud data including three-dimensional space coordinate information, as input data of the improved three-dimensional Gaussian mixture model; The input point cloud data is respectively input to a topological Gaussian module, a heterogeneous covariance module and a hierarchical fusion module, the topological Gaussian module including a topological feature extraction unit and a feature fusion unit, the heterogeneous covariance module including a covariance type discrimination unit and a covariance parameter generation unit, and the hierarchical fusion module including a multi-scale initialization unit and a hierarchical probability fusion unit, the three modules serving as substructures of the improved three-dimensional Gaussian mixture model; In the topological Gaussian module, the topological feature extraction unit calculates local topological features of each point with the point cloud data and radius neighborhood connection information as input, and the feature fusion unit fuses the local topological features with the spatial coordinate information of the point to output a topological fusion feature set, and the calculation of the local topological features of each point specifically includes: A local subgraph is constructed based on the radius neighborhood of each point, and first-order and high-order spectral features are calculated through a graph Laplacian operator to capture the local structure complexity and connectivity of the point in space and obtain a basic topological description of the point cloud; On the basis of the spectral features, a principal component analysis is used to obtain a main direction of the neighborhood, and a small perturbation is made to the position of the neighborhood point, and the change amplitude of the main direction and the structural characteristics before and after the perturbation is recorded to measure the stability and sensitivity of the local topological structure under spatial changes; On the basis of the extraction of the multi-order neighborhood graph spectral features, the main direction and the perturbation stability features, the distribution of points of various categories in the neighborhood of each point is counted under multiple neighborhood scales, the information entropy is calculated, the diversity and redundancy of the spatial structure around the node under different scales are obtained, and the distribution uniformity and local complexity of the node under different spatial levels are reflected to enhance the recognition ability of complex structures and boundary regions; In the heterogeneous covariance module, the covariance type discrimination unit discriminates the covariance type corresponding to each Gaussian component with the topological fusion feature set as input in combination with the spatial characteristics of the point cloud distribution, and the covariance parameter generation unit automatically generates covariance parameters according to the covariance type and the related point set to output the covariance type and parameter set of each Gaussian component, and the discrimination of the covariance type corresponding to each Gaussian component in combination with the spatial characteristics of the point cloud distribution refers to selecting the most suitable covariance structure type to describe the distribution characteristics of the component according to the spatial distribution form, density and directionality features of the point cloud in each component, including diagonal type, full matrix type and sparse type, and the covariance parameter generation unit automatically generates covariance parameters according to the covariance type and the related point set, specifically including: For the identified covariance type, determine the specific parameter elements that need to be calculated. The diagonal type only calculates the variance of each coordinate axis, the full matrix type calculates the covariance between all coordinate axes, and the sparse type calculates the preset key parameters. For the relevant point set assigned to the Gaussian component, calculate the distance distribution of all points to the component center, and then extract the required spatial feature values ​​to provide basic data for covariance parameter calculation; According to the covariance type and spatial characteristic values, the corresponding statistical method is used to automatically generate covariance parameters, and the generated parameter set is output as a description of the spatial distribution of Gaussian components; In the hierarchical fusion module, the multi-scale initialization unit takes the covariance type and parameter set and point cloud data as input, initializes the Gaussian components at different spatial scales, and the hierarchical probability fusion unit fuses the component parameters and attribution probabilities of each spatial scale and outputs a multi-scale Gaussian component parameter set. The hierarchical probability fusion unit fuses the component parameters and attribution probabilities of each spatial scale by weighted integration of the Gaussian component parameters obtained at different spatial scales and their respective point attribution probabilities to generate a Gaussian component parameter set and attribution probability distribution that comprehensively reflects the multi-scale structural characteristics. Taking the multi-scale Gaussian component parameter set as the initial parameters, combined with the topological fusion features and covariance parameters, the expectation maximization algorithm is used within the overall framework of the three-dimensional Gaussian mixture model to iteratively optimize the mean, covariance, and weight of each component, and output the final converged Gaussian component parameter set; According to the final converged Gaussian component parameter set, for each point cloud data point in the spatial segmentation area, the attribution probability under all Gaussian components is calculated, and fine-grained cluster labels are assigned according to the maximum probability principle. The fine-grained cluster labels and corresponding Gaussian component parameters of each point in the spatial segmentation area are output for fine material differentiation and category calibration. The calculation of the attribution probability under all Gaussian components is specifically as follows: Based on the spatial distance between each point and the center of each Gaussian component and the spatial distribution parameters of the component, the relative similarity of the point belonging to each Gaussian component is evaluated; The relative similarity of each Gaussian component is weighted with the mixing weight to obtain the weighted attribution score of each point to each Gaussian component; The weighted attribution scores of all Gaussian components are normalized to obtain the final attribution probability distribution of each point under all Gaussian components, and the fine-grained clustering label of the point is determined with the maximum probability.

[0031] The present invention integrates the topological Gaussian module, heterogeneous covariance module and hierarchical fusion module for point cloud data in spatial segmentation areas, achieving innovation and functional breakthroughs in the structure of the three-dimensional Gaussian mixture model. By introducing multivariate topological features such as local graph spectrum, main direction perturbation and multi-scale information entropy, a fine characterization of the complex spatial structure of the point cloud is achieved, enabling the model to fully perceive local connectivity, stability and spatial diversity, and effectively improving the resolution of material boundaries, fine-grained structures and multi-category overlapping scenes; the heterogeneous covariance module adaptively selects the covariance type according to the distribution morphology of the point cloud, and automatically generates parameters based on different spatial features, thereby improving the model's adaptability to spatial anisotropy and complex morphological materials, and avoiding the problems of traditional Gaussian models being sensitive to single structure and noise; the hierarchical fusion module enables the model to capture both local details and the overall pattern of point cloud data through weighted fusion of multi-scale Gaussian components and probabilities, thereby enhancing the clustering robustness in multi-level structures. The overall process supports full-parameter self-learning and structural adaptation, improving the automation, intelligence and engineering scalability of material clustering and category calibration, and providing higher accuracy, stronger generalization capabilities and a better data foundation for intelligent material identification and asset management in complex power distribution network environments.

[0032] In this embodiment, obtaining the material category, quantity and spatial location information includes: According to the output fine-grained cluster labels, the point cloud data in the spatial segmentation area are grouped according to the cluster labels, and each group corresponds to a Gaussian component or fine-grained category; For each fine-grained category, the spatial coordinate features and distribution features of the point cloud and the standard material geometric features, spatial structure features and attribute information in the distribution network material feature database are extracted to construct the point cloud feature vector and the database feature vector; Calculate the similarity between the point cloud feature vector and the database feature vector, the similarity is based on cosine similarity, and select the one with the largest similarity as the category matching basis; Each fine-grained category is matched with the most similar material category in the database, the material type of the point cloud category is calibrated, and the number of point cloud data and the centroid coordinates or boundary range in three-dimensional space are recorded as the basis for quantity statistics and spatial positioning; The material type, quantity and spatial location information of all fine-grained categories are summarized to obtain the structured results of material category, quantity and spatial location information within the spatial segmentation area.

[0033] In this embodiment, the material category, quantity and spatial location information are output in the form of structured results for inventory, inspection and asset management of distribution network tower materials. Specifically, the name, corresponding quantity, three-dimensional spatial coordinates or spatial range, category label and related attribute information of each type of material that has been automatically identified are organized into a standardized data format, and output as a table, database record or interface data for direct call by automatic inventory, inspection scheduling and asset management.

[0034] Example 1: In order to verify the feasibility of the present invention in implementation, the present invention was applied to a distribution network tower inspection and material automatic identification project at the junction of a certain urban area and suburbs. The power transmission and distribution lines in this area are densely distributed, with a total of 24 towers. The surrounding environment of the towers is complex, with green belts, construction sites, and construction debris piles. In the past, the manual inspection method not only required a large amount of manpower and material resources, but was also easily affected by factors such as occlusion and stacking during the statistics and spatial positioning of tower materials, resulting in inaccurate statistics, missed materials, and unclear descriptions of spatial distribution. Especially at night or in severe weather conditions, manual efficiency and accuracy are even lower. In response to these industry problems, this project applied the distribution network tower material identification method based on spatial stereo segmentation and point cloud group structure of the present invention on a large scale in this area for the first time, comprehensively improving the automation and intelligence level of distribution network material management.

[0035] During actual application, the on-site inspection team uses an unmanned aerial vehicle system equipped with high-precision lidar to collect all-round three-dimensional point cloud data for each tower and its surrounding 5-meter radius. The point cloud data volume for a single inspection of each tower is approximately 430,000 to 530,000 points, and all data is uploaded to the cloud in real time. The system first removes noise, downsamples, aligns, and coordinates the raw point cloud data to generate a normalized point cloud dataset. Subsequently, an undirected point cloud map is constructed based on spatial neighborhood relationships, the point cloud density distribution is automatically analyzed, representative seed points are extracted, and a random walk clustering algorithm is guided through multiple rounds of self-supervised feedback to complete preliminary spatial segmentation. For each segmented area, the system automatically constructs an innovative three-dimensional Gaussian mixture model that integrates topological features, covariance adaptation, and multi-scale probability. It performs fine-grained clustering of point cloud materials and performs multi-attribute matching with standard templates in the distribution network material feature database to automatically determine the material category, quantity, and spatial location.

[0036] Table 1 Comparative data between the method of the present invention and manual inspection in the identification of distribution network tower materials Tower number Total materials (labor) Total number of materials (present invention) Recognition accuracy (manual,%) Recognition accuracy (in this invention,%) Recognition time (manual, min) Identification time (inventive, min) Spatial positioning error (manual, m) Spatial positioning error (in this invention, m) Tower number A01 188 187 98.4 99.5 24.3 2.7 0.42 0.13 A01 A02 170 171 97.6 99.4 23.1 2.3 0.48 0.15 A02 A03 193 193 100.0 100.0 25.7 2.6 0.36 0.11 A03 A06 182 179 94.0 98.3 21.9 2.2 0.53 0.17 A06 A08 183 182 97.3 99.5 22.8 2.5 0.46 0.12 A08 A09 176 174 92.6 97.9 24.8 2.4 0.55 0.18 A09 A11 189 189 100.0 100.0 23.5 2.6 0.38 0.12 A11 A12 187 187 100.0 100.0 23.3 2.3 0.41 0.10 A12 average 183.5 183.0 96.8 99.3 23.7 2.5 0.45 0.14 average According to the comparative data in Table 1, it can be clearly seen that the “distribution network tower material identification method based on spatial stereo segmentation and point cloud group structure” proposed in the present invention is significantly superior to the traditional manual inspection method in multiple key performance indicators.

[0037] In terms of recognition accuracy, manual identification achieved an average accuracy of 96.8%, while the proposed method achieved 99.3%, significantly improving recognition precision. For some towers, such as A03, A11, and A12, both methods achieved 100% accuracy. However, for most other towers, such as A01, A06, and A09, the proposed method significantly outperformed manual methods, demonstrating that the algorithm possesses more stable material recognition capabilities, effectively reducing missed detections and misjudgments.

[0038] The comparison of recognition time is even more striking. The average time consumed by manual inspection is 23.7 minutes, while the method of the present invention takes only 2.5 minutes, which is about 9.5 times higher than the traditional method. This advantage is particularly critical for large-scale inspection tasks, which can significantly reduce time costs and improve on-site operation and maintenance efficiency.

[0039] In terms of spatial positioning error, the average error of manual methods is 0.45 meters, while the present invention is controlled within 0.14 meters, with a significant improvement in accuracy. The spatial recognition capability of the present invention is particularly valuable in scenarios requiring high-precision installation, re-inspection, or inventory management.

[0040] The present invention not only has high recognition and positioning accuracy, but also reduces labor and time costs. It is particularly suitable for complex structure towers, dense material mounting environments, and multi-tower continuous detection tasks, and can provide more reliable, fast, and efficient technical support for intelligent operation and maintenance of distribution networks.

[0041] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A distribution network tower material identification method based on spatial stereo segmentation and point cloud group structure is characterized by: include: Collect 3D point cloud data of distribution network towers and their surroundings, pre-process the 3D point cloud data to obtain a normalized point cloud dataset, and build a distribution network material feature database; Based on the normalized point cloud dataset, the spatial distance between each point and other points in the neighborhood is calculated to construct an undirected point cloud graph. Each point is used as a node of the undirected point cloud graph, the neighborhood relationship is used as the edge, and the edge weight is defined by the spatial distance. According to the spatial density, geometric distribution or manually set rules of the points in the undirected point cloud, a group of seed points are selected as the starting nodes of random walk clustering, and various sub-points correspond to different material categories; Starting from the seed point, random walk clustering is performed on the undirected point cloud map. The random walk probability of each unclassified point to the seed point of each category is calculated, and each point is assigned to the category with the largest probability value to obtain the spatial segmentation result. For each spatial segmentation area in the spatial segmentation result, a corresponding improved three-dimensional Gaussian mixture model is established, the parameters of the three-dimensional Gaussian mixture model are trained, and fine-grained clustering labels are output; Based on fine-grained clustering labels and combined with the distribution network material feature database, the point cloud data of each spatial segmentation area is calibrated, the quantity is counted, and the spatial positioning is performed to obtain the material category, quantity, and spatial location information; The material category, quantity and spatial location information are output in the form of structured results for inventory, inspection and asset management of distribution network tower materials.

2. The distribution network tower material identification method based on spatial stereo segmentation and point cloud group structure according to claim 1 is characterized in that: The three-dimensional point cloud data specifically includes spatial coordinate information, reflection intensity information and color information of the distribution network tower and surrounding materials in three-dimensional space.

3. The distribution network tower material identification method based on spatial stereo segmentation and point cloud group structure according to claim 1 is characterized in that: The preprocessing of the three-dimensional point cloud data specifically includes noise removal, downsampling, point cloud registration and coordinate normalization.

4. The distribution network tower material identification method based on spatial stereo segmentation and point cloud group structure according to claim 1 is characterized in that: The construction of the distribution network material feature database specifically refers to combining manual labeling results and historical material information to extract geometric features, analyze spatial distribution and organize attribute information of different distribution network materials, summarize and classify the typical three-dimensional shapes, size parameters, spatial structure characteristics and related attributes of various materials, and form a distribution network material feature database that can be used for material identification and comparison.

5. The distribution network tower material identification method based on spatial stereo segmentation and point cloud group structure according to claim 1 is characterized in that: Based on the normalized point cloud dataset, the spatial distance between each point and other points in the neighborhood is calculated to construct an undirected point cloud graph. Each point is used as a node of the undirected point cloud graph, the neighborhood relationship is used as an edge, and the edge weight is defined by spatial distance, including: For each point in the normalized point cloud dataset, the neighborhood point set of each point is determined by the radius neighborhood algorithm; For each point and other points in the neighborhood, calculate the Euclidean distance between the three-dimensional space coordinates. The Euclidean distance is the square root of the sum of the squares of the distance differences in the three-dimensional coordinate axis direction, reflecting the actual distance between the two points in three-dimensional space; Each point in the normalized point cloud dataset is used as a node of the undirected point cloud graph, and each pair of points that satisfy the neighborhood relationship is used as an edge of the undirected point cloud graph. According to the calculated Euclidean distance, an edge weight is assigned to each edge by setting a distance decay rule. The closer the distance, the greater the edge weight, and the farther the distance, the smaller the edge weight. This completes the construction of the nodes, edges and edge weights of the undirected point cloud graph.

6. The distribution network tower material identification method based on spatial stereo segmentation and point cloud group structure according to claim 1 is characterized in that: According to the spatial density, geometric distribution or manually set rules of the points in the undirected point cloud, a group of seed points are selected as the starting nodes of the random walk clustering. Various sub-points correspond to different material categories, including: Based on the undirected point cloud graph, the spatial density of each node is calculated. The spatial density is the number of neighboring nodes within a preset radius centered on the node. Sort all nodes from high to low according to spatial density, and preliminarily select nodes with spatial density greater than the set threshold as candidate nodes; Screen all candidate nodes for spatial uniformity; According to manually set rules or historical recognition experience, the seed points after spatial uniformity screening are combined with spatial position and point cloud features to calibrate the corresponding material categories. Each material category contains at least one seed point, and the seed point corresponding to each material category is used as the starting node of random walk clustering.

7. The distribution network tower material identification method based on spatial stereo segmentation and point cloud group structure according to claim 1 is characterized in that: The obtaining of the spatial segmentation result includes: The seed points corresponding to the obtained material categories are used as the starting nodes, and the category labels of each node are initialized on the undirected point cloud graph. Initially, only the seed points are assigned category labels, and the other nodes are in an unclassified state. On the undirected point cloud graph, for each unclassified node, the random walk probability to the seed point of each category is calculated based on the transition probability defined by the edge weight; For each unclassified node, set the category label to the category corresponding to the maximum random walk probability to complete the initial space segmentation of the current round; The preliminary spatial segmentation results are evaluated through self-supervised feedback. Based on the local consistency of the node category, neighborhood confidence, and boundary uncertainty, nodes with fuzzy boundaries and uncertain classification are automatically identified and temporarily marked as pseudo-labeled seed points. Append the pseudo-labeled seed points as a new seed set, combine the original seed points, and re-perform random walk clustering on the undirected point cloud. For all unclassified nodes and pseudo-labeled seed points, recalculate the random walk probability to each category seed point and update the category label. After the preliminary spatial segmentation results are corrected through self-supervised feedback, the pseudo-labeled seed points are appended as a new seed set and combined with the original seed points. Random walk clustering is performed again on the undirected point cloud graph, and the category assignment process is continuously iterated until the category labels of all nodes remain stable or the preset maximum number of iterations is reached, and the spatial segmentation results are finally output.

8. The distribution network tower material identification method based on spatial stereo segmentation and point cloud group structure according to claim 1 is characterized in that: The output fine-grained clustering labels include: For each spatial segmentation region in the spatial segmentation result, extracting point cloud data within the spatial segmentation region, wherein the point cloud data includes three-dimensional spatial coordinate information as input data of the improved three-dimensional Gaussian mixture model; The input point cloud data is input into the topological Gaussian module, the heterogeneous covariance module and the hierarchical fusion module respectively. The topological Gaussian module includes a topological feature extraction unit and a feature fusion unit, the heterogeneous covariance module includes a covariance type discrimination unit and a covariance parameter generation unit, and the hierarchical fusion module includes a multi-scale initialization unit and a hierarchical probability fusion unit. The three modules serve as substructures of the improved three-dimensional Gaussian mixture model. In the topological Gaussian module, the topological feature extraction unit takes point cloud data and radius neighborhood connection information as input to calculate the local topological features of each point. The feature fusion unit fuses the local topological features with the spatial coordinate information of the point and outputs a topological fusion feature set. In the heterogeneous covariance module, the covariance type discrimination unit takes the topological fusion feature set as input and combines the spatial characteristics of the point cloud distribution to discriminate the covariance type corresponding to each Gaussian component. The covariance parameter generation unit automatically generates covariance parameters based on the covariance type and related point set, and outputs the covariance type and parameter set of each Gaussian component. In the hierarchical fusion module, the multi-scale initialization unit takes the covariance type and parameter set and point cloud data as input to initialize Gaussian components at different spatial scales. The hierarchical probability fusion unit fuses the component parameters and attribution probabilities of each spatial scale and outputs a multi-scale Gaussian component parameter set. Taking the multi-scale Gaussian component parameter set as the initial parameters, combined with the topological fusion features and covariance parameters, the expectation maximization algorithm is used within the overall framework of the three-dimensional Gaussian mixture model to iteratively optimize the mean, covariance, and weight of each component, and output the final converged Gaussian component parameter set; According to the final converged Gaussian component parameter set, for each point cloud data point in the spatial segmentation area, the attribution probability under all Gaussian components is calculated, and fine-grained clustering labels are assigned according to the maximum probability principle. The fine-grained clustering labels and corresponding Gaussian component parameters of each point in the spatial segmentation area are output.

9. The distribution network tower material identification method based on spatial stereo segmentation and point cloud group structure according to claim 1 is characterized in that: The material category, quantity and spatial location information obtained include: According to the output fine-grained cluster labels, the point cloud data in the spatial segmentation area are grouped according to the cluster labels, and each group corresponds to a Gaussian component or fine-grained category; For each fine-grained category, the spatial coordinate features and distribution features of the point cloud and the standard material geometric features, spatial structure features and attribute information in the distribution network material feature database are extracted to construct the point cloud feature vector and the database feature vector; Calculate the similarity between the point cloud feature vector and the database feature vector, the similarity is based on cosine similarity, and select the one with the largest similarity as the category matching basis; Each fine-grained category is matched with the most similar material category in the database, the material type of the point cloud category is calibrated, and the number of point cloud data and the centroid coordinates or boundary range in three-dimensional space are recorded as the basis for quantity statistics and spatial positioning; The material type, quantity and spatial location information of all fine-grained categories are summarized to obtain the structured results of material category, quantity and spatial location information within the spatial segmentation area.

10. The distribution network tower material identification method based on spatial stereo segmentation and point cloud group structure according to claim 1 is characterized in that: The material category, quantity and spatial location information are output in the form of structured results for inventory, inspection and asset management of distribution network tower materials. Specifically, the name, corresponding quantity, three-dimensional spatial coordinates or spatial range, category label and related attribute information of each type of material that has been automatically identified are organized into a standardized data format, and output as a table, database record or interface data for direct call by automatic inventory, inspection scheduling and asset management.

Images