Power transmission line tower automatic identification method, system and device based on laser radar and storage medium

Abstract

The invention discloses a power transmission line tower automatic identification method, system and device based on a laser radar and a storage medium, and belongs to the technical field of power system intelligent inspection and three-dimensional point cloud processing, and the method comprises the steps: collecting three-dimensional point cloud data of a region along a power transmission line, carrying out the preprocessing, obtaining a structural point cloud region, and carrying out the space division; according to the geometric height and the normal vector direction of each region, screening candidate targets meeting a preset condition; performing point cloud classification on the candidate targets to obtain point cloud regions belonging to target categories; based on the point cloud region, extracting upper and lower boundary endpoints, and constructing a structure direction vector connecting the two endpoints; and calculating a direction deviation angle through an included angle between the structure direction vector and the vertical direction, determining the attitude classification, and outputting corresponding spatial position and structure vector information. The problems that an existing power transmission line inspection technology is low in recognition accuracy in a complex environment, depends on manual judgment and cannot automatically obtain tower space attitude information are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system intelligent inspection and three-dimensional point cloud processing, and particularly relates to a power transmission line tower automatic identification method, system and device based on a laser radar and a storage medium. BACKGROUND

[0002] The power transmission line tower is an important infrastructure of the power system, and bears the important responsibility of supporting the power transmission conductor and maintaining the safety distance of the power transmission line. With the continuous expansion of the power grid, the number of towers increases dramatically, and factors such as extreme weather, geological disasters, and environmental aging also significantly increase the risk of tower damage, tilting, and collapse. The traditional inspection method mainly relies on manual on-site visual inspection or analysis through photographs, which not only has the problems of high operation danger, high labor cost, and long cycle, but also often has difficulty in carrying out large-scale investigation in a timely manner after natural disasters.

[0003] In recent years, the power transmission line detection method based on unmanned aerial vehicle aerial images has been widely studied. Using deep learning target detection networks such as the YOLO series, Faster R-CNN, etc. for image recognition can improve the inspection efficiency to a certain extent. However, this kind of method is limited by environmental factors such as light changes, weather conditions, and angle obstructions, and is prone to false detection and missed detection in complex backgrounds. In addition, relying solely on two-dimensional image information makes it difficult to accurately determine the three-dimensional attitude change of the tower.

[0004] As an active three-dimensional space data acquisition means, the laser radar (LiDAR) has the advantages of high precision, all-weather, and anti-light interference, providing a new technical approach for tower detection. However, the current point cloud-based tower detection method still has significant shortcomings, such as simply relying on height, density, and other threshold screening, which cannot effectively distinguish towers from other tall objects (such as tall trees and communication towers), and most of them require a large amount of manual intervention and processing, with low automation level, making it difficult to apply to post-disaster emergency inspection and other high-timeliness demand scenarios. Therefore, there is an urgent need for an intelligent tower identification method that can fully utilize the three-dimensional features of laser radar point clouds, has high accuracy, strong environmental adaptability, and high automation level. SUMMARY

[0005] To solve the above technical problems, a power transmission line tower automatic identification method based on a laser radar is proposed, which comprises collecting three-dimensional point cloud data of the area along the power transmission line to obtain a spatial point set, and preprocessing the spatial point set to obtain a structure point cloud region.

[0006] The structure point cloud region is spatially divided, and according to the geometric height and normal vector direction of each region, point cloud segments that meet the preset conditions are selected as candidate targets.

[0007] Candidate targets are classified into point clouds to obtain point cloud regions belonging to the target category;

[0008] Based on the point cloud region, extract the upper and lower boundary endpoints and construct the structural direction vector connecting the two endpoints;

[0009] The orientation offset angle is calculated by the angle between the structural orientation vector and the vertical direction, the attitude classification is determined, and the corresponding spatial position and structural vector information are output.

[0010] As a preferred embodiment of the automatic identification method for transmission line towers based on lidar described in this invention, the step of collecting three-dimensional point cloud data of the area along the transmission line, obtaining a spatial point set, and preprocessing it to obtain a structural point cloud region includes...

[0011] The original point cloud is cleaned to remove point sets that are below the range of the main structure in the height dimension. Point cloud regions that do not meet the continuous structural features are identified based on local point density and spatial arrangement direction. Noise points in the point cloud regions are removed to obtain three-dimensional point cloud data with continuous structural features, forming structural point cloud regions.

[0012] As a preferred embodiment of the automatic identification method for transmission line towers based on lidar described in this invention, the step of spatially dividing the structural point cloud region and selecting point cloud segments that meet preset conditions as candidate targets based on the geometric height and normal vector direction of each region includes:

[0013] The point cloud data in the structural point cloud region is grouped according to spatial proximity, and the height range and normal vector direction distribution characteristics are calculated for each group. By judging whether the height range meets the first condition and whether the main direction of the normal vector is located in the preset vertical vector interval, the point cloud groups that do not meet the conditions are filtered out, and the remaining groups are retained as candidate target regions.

[0014] As a preferred embodiment of the automatic identification method for transmission line towers based on lidar described in this invention, the step of classifying candidate targets into point clouds to obtain point cloud regions belonging to the target category includes:

[0015] A multi-scale spatial neighborhood is constructed based on the point cloud data of the candidate target, and the spatial distribution relationship of the points in the neighborhood is feature-encoded. The spatial structure expression of the point cloud region is formed by local feature extraction and global feature aggregation, and classification judgment is performed to output the point cloud region that is judged as the target category.

[0016] As a preferred embodiment of the automatic identification method for transmission line towers based on lidar described in this invention, the step of extracting the upper and lower boundary endpoints based on the point cloud region and constructing the structural direction vector connecting the two endpoints includes:

[0017] Extract the height coordinates of all points from the target point cloud region, and select point sets within the intervals set before the maximum height value and after the minimum height value respectively. Calculate the geometric centroid of each point as the top endpoint and bottom endpoint, and construct the structural direction vector based on the spatial positional relationship between the top endpoint and the bottom endpoint.

[0018] As a preferred embodiment of the automatic identification method for transmission line towers based on lidar described in this invention, the step of calculating the directional offset angle by the angle between the structural direction vector and the vertical direction, determining the attitude classification, and outputting the corresponding spatial position and structural vector information includes:

[0019] The structural direction vector is multiplied by the unit vertical vector in the three-dimensional coordinate system. The unit vertical vector is the unit vector in the positive z-axis direction of the coordinate axis, which represents the vertical reference direction. The angle between the two vectors is calculated by combining the magnitude of the structural direction vector.

[0020] The system determines the level based on the included angle value and the set attitude classification threshold, and outputs the corresponding attitude category label, structural orientation vector, and spatial coordinates of the point cloud region.

[0021] As a preferred embodiment of the automatic identification method for transmission line towers based on lidar described in this invention, the output of the corresponding spatial location and structural vector information further includes,

[0022] Spatial boundary calculation is performed on the point cloud region that is identified as the target category, and spatial position coordinates are extracted based on the geometric centroid of the point set in the current region. The result output object is constructed by combining the structural orientation vector and the attitude classification label, forming a unified output structure that includes structural orientation, spatial position and attitude information.

[0023] Another objective of this invention is to provide an automatic identification system for transmission line towers based on lidar. This invention solves the problems of low identification accuracy, reliance on manual judgment, and inability to automatically acquire tower spatial attitude information in existing transmission line inspection technologies under complex environments.

[0024] As a preferred embodiment of the automatic identification system for transmission line towers based on lidar described in this invention, it is characterized by including a point cloud data processing module, a candidate target extraction and point cloud classification module, a structural orientation construction and attitude analysis module, and an output and structural information generation module.

[0025] The point cloud data processing module receives the raw 3D point cloud data collected by the lidar device, cleans and preprocesses the data, removes ground points with a height lower than the tower structure, vegetation points with a density higher than the threshold and arranged in a disordered manner, and isolated noise points formed at the edges, and outputs a subset of point cloud with continuity, directionality and geometric integrity to form a structured point cloud region.

[0026] The candidate target extraction and point cloud classification module is based on the spatial division of the structural point cloud region, and selects candidate regions with tower geometric features from it, while completing the point cloud classification operation.

[0027] First, the structural point cloud is grouped by spatial proximity, and the geometric height range and normal vector direction distribution of each group are calculated. Point cloud clusters that do not conform to the tower characteristics are filtered out. The filtered candidate target point cloud is input into the classification model to extract local and global structural features and output the point cloud region belonging to the tower category.

[0028] The structure orientation construction and attitude analysis module performs structural modeling and attitude state analysis on point cloud regions identified as tower categories: extracting the upper and lower boundary points of the tower, calculating the geometric centroids of the top and bottom within a specified height range, and constructing a structural orientation vector connecting the two ends; calculating the angle between this vector and the unit vertical vector in three-dimensional space to obtain the structural offset angle; and comparing the angle value with a preset attitude classification threshold to determine the tower state.

[0029] The output and structural information generation module extracts the boundaries of each target point cloud region identified as a tower and calculates the spatial geometric centroid of the current region as the current position coordinates; it encapsulates the attitude classification label and structural direction vector together into a result output object, forming unified output information containing spatial position, structural direction and attitude state.

[0030] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the automatic identification method for transmission line towers based on lidar.

[0031] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the aforementioned automatic identification method for transmission line towers based on lidar.

[0032] The beneficial effects of this invention are as follows: By employing point cloud spatial feature extraction and multi-scale classification methods, this invention effectively improves the accuracy of pole and tower recognition in complex backgrounds. Compared to image-dependent recognition methods, this invention is unaffected by lighting, occlusion, and vegetation interference, and can stably distinguish poles and towers from surrounding tall objects, resulting in more reliable recognition performance.

[0033] This invention has the ability to automatically analyze the attitude state of towers. By extracting the upper and lower boundary points to construct the structural direction vector and calculating the angle between it and the vertical direction, it can accurately determine whether the tower is tilted or abnormal, which helps to detect potential hazards such as deformation and collapse in a timely manner.

[0034] This invention automates the entire process from data preprocessing to structure judgment, has high computational efficiency, is suitable for deployment on drones or portable devices for independent operation, and is particularly suitable for scenarios where communication is limited after disasters and inspection timeliness is required, thus possessing good engineering application value. Attached Figure Description

[0035] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Fig. 1 The above is a flowchart of an automatic identification method for transmission line towers based on lidar, provided as an embodiment of the present invention.

[0037] Fig. 2 A deep learning flowchart of an automatic identification method for transmission line towers based on lidar provided in one embodiment of the present invention. Detailed Implementation

[0038] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0039] Example 1, referring to Figs. 1-2 This is the first embodiment of the present invention, which provides an automatic identification method for transmission line towers based on lidar, including:

[0040] Existing methods for detecting transmission line towers mainly rely on manual inspections, target detection techniques based on two-dimensional images, or lidar point cloud analysis based on simple geometric rules. These methods have many shortcomings in practical applications. For example, manual inspections are highly dangerous and inefficient. Image-based detection methods are easily affected by environmental factors such as changes in lighting, occlusion, and weather conditions, resulting in low accuracy and a lack of three-dimensional structural information, making it difficult to accurately determine tower attitude changes. Existing lidar point cloud-based methods mostly rely on simple rules such as height thresholds and volume filtering, failing to effectively distinguish towers from surrounding tall vegetation and buildings, leading to high false detection rates. Furthermore, they typically require significant manual post-processing, resulting in low overall automation and insufficient processing efficiency, making it difficult to meet the high-timeliness and high-accuracy application requirements of post-disaster emergency inspections and intelligent operation and maintenance.

[0041] To address the aforementioned issues, this invention proposes an automatic identification method for transmission line towers based on lidar, aiming to solve the following technical problems: overcoming the low detection accuracy of existing image recognition methods in complex environments, fully utilizing the advantages of lidar's three-dimensional spatial data to achieve complete extraction and identification of the three-dimensional structural features of the towers; solving the problem that traditional methods based on simple geometric rules cannot accurately distinguish between towers and tall non-tower targets (such as trees, communication towers, etc.), improving the accuracy and robustness of tower identification by extracting multi-scale point cloud features through deep learning; addressing the difficulty of existing methods in judging the health status of towers (such as tilting, deformation, collapse) in real time, proposing a tilt angle calculation method based on endpoint extraction and centerline generation to achieve intelligent analysis of tower attitude; reducing reliance on manual post-processing, improving the automation level of the entire tower identification process, enabling the method to be deployed on front-end platforms such as drones and vehicle-mounted equipment, meeting the application needs of large-scale rapid inspection, post-disaster emergency detection, and intelligent operation and maintenance of the power grid.

[0042] S1. Collect three-dimensional point cloud data of the area along the power transmission line, and preprocess the spatial point set to obtain the structural point cloud region.

[0043] The original point cloud is cleaned to remove point sets that are below the range of the main structure in the height dimension. Point cloud regions that do not meet the continuous structural features are identified based on local point density and spatial arrangement direction. Noise points in the point cloud regions are removed to obtain three-dimensional point cloud data with continuous structural features, forming structural point cloud regions.

[0044] (1) LiDAR point cloud data acquisition:

[0045] During the data acquisition phase, drones or ground-based mobile platforms equipped with lidar are used to fly or move along the power transmission lines to acquire high-density 3D point cloud data covering the towers and surrounding environment. To ensure the effectiveness of subsequent processing, the acquisition system must have sufficient spatial resolution, such as a point cloud sampling density of no less than 300 points / square meter.

[0046] (2) Point cloud preprocessing:

[0047] The purpose of cloud data preprocessing is to remove irrelevant points (such as ground, vegetation, and noise), retain sparse structures with vertical features, and provide clean and structurally clear input data for subsequent pole identification.

[0048] First, ground point data removal is performed. Since the ground typically occupies a large amount of low-altitude point cloud data and lacks the height characteristics of towers, a simple height threshold filtering method is used to remove points with heights below a set threshold z. threshold Specifically, for each point in the point cloud, if its z-coordinate satisfies:

[0049] z i <z threshold

[0050] If the point is identified as a ground point, it is deleted. This step significantly reduces the amount of data processed subsequently and eliminates ground interference.

[0051] Next, vegetation points are removed. Since vegetation (especially shrubs and low trees) appears as a dense, irregularly distributed set of small volume points in a 3D point cloud, it can be identified by calculating local point density and normal vector distribution characteristics. Calculate the vegetation point p for each point. i The number of points N in the neighborhood i and volume V i The point density is defined as:

[0052]

[0053] Point sets with a density exceeding a set threshold and lacking obvious vertical alignment features are identified as vegetation noise and removed. The vertical component of the normal vector can assist in identifying vertical structural features; its calculation formula is as follows:

[0054]

[0055] in, Let be the normal vector of the point. It is a unit vector in the vertical direction with a small verticality (e.g., d). z <d threshold Dense point clouds (d) are more likely to be vegetation. threshold The threshold representing verticality.

[0056] Finally, isolated points and small noise clusters are removed. Since lidar systems are prone to generating isolated noise points at edges or long distances, a clustering algorithm based on Euclidean distance (such as DBSCAN) is used to remove points with a number less than a set minimum threshold N. min Small clusters.

[0057] S2. Divide the space in the structural point cloud region, and select point cloud segments that meet the preset conditions as candidate targets based on the geometric height and normal vector direction of each region.

[0058] The point cloud data in the structural point cloud region is grouped according to spatial proximity, and the height range and normal vector direction distribution characteristics are calculated for each group. By judging whether the height range meets the first condition and whether the main direction of the normal vector is located in the preset vertical vector interval, the point cloud groups that do not meet the conditions are filtered out, and the remaining groups are retained as candidate target regions.

[0059] In the region candidate extraction stage, a coarse screening is performed based on the physical characteristics of the tower structure. For a point cloud patch, its overall height h is calculated, and the normal vector of each local point is extracted. When the region height h > h min And verticality d z When the area exceeds a preset threshold, it is retained as a candidate area for tower construction. This process effectively eliminates most low-lying objects and tilted structures, ensuring the integrity of the tower.

[0060] S3. Classify the candidate targets into point clouds to obtain point cloud regions belonging to the target category.

[0061] A multi-scale spatial neighborhood is constructed based on the point cloud data of the candidate target, and the spatial distribution relationship of the points in the neighborhood is feature-encoded. The spatial structure expression of the point cloud region is formed by local feature extraction and global feature aggregation, and classification judgment is performed to output the point cloud region that is judged as the target category.

[0062] In a preferred embodiment of this invention, point cloud classification of candidate targets involves a deep learning-based point cloud feature extraction and classification method to improve the accuracy and robustness of pole identification in complex environments. The preprocessed and initially screened candidate region point clouds are input into a deep learning network. The input data includes the spatial coordinates (x, y, z) of each point and optional additional features (such as reflection intensity).

[0063] The candidate target extraction and point cloud classification module is powered by a deep learning network, implemented using a classification model based on an improved PointNet++ network structure. The model structure includes sub-modules for sampling and grouping, local feature extraction, multi-scale feature fusion, global feature encoding, and classification output. These modules are used to extract multi-scale spatial structure features from the candidate target point cloud and output classification results, such as... Fig. 2 As shown.

[0064] In the sampling and grouping stage, representative center points are selected using the farthest point sampling (FPS) algorithm, and a neighborhood is defined for each center point to form a local subset. Subsequently, in the local feature extraction stage, a shared multilayer perceptron (Shared MLP) is used to extract fine-grained features of each point within the local neighborhood, and local descriptors are generated through pooling operations (such as max pooling). To further consider local information at different scales, this invention introduces a multi-scale grouping (MSG) strategy in the feature extraction process, that is, features are extracted separately within different radius ranges, and multi-scale features are fused, effectively improving the modeling capability of tower structures at different scales.

[0065] In the feature aggregation stage, all local descriptors are stacked using a shared multilayer perceptron, and a fixed-length global feature vector is generated through global pooling (such as global max pooling or average pooling) to fully characterize the spatial structure of the entire candidate point cloud region. Finally, in the classification output stage, two or three fully connected layers (FC) are used with Dropout regularization to output the classification probability. The cross-entropy loss function is used during classifier training, defined as:

[0066]

[0067] Among them, y i For real category labels, This is the probability that the network predicts the pole / tower category. By minimizing this loss function, the network continuously optimizes its parameters during training, improving classification accuracy.

[0068] During the inference phase, the network performs forward inference on the point cloud of each candidate region, outputting a probability value indicating whether it belongs to the tower category, and performs binary classification based on a set threshold. Point cloud regions identified as towers are then fed into subsequent structure reconstruction and tilt angle detection modules. This invention's deep learning classification method fully utilizes multi-scale fusion of local and global spatial features, possessing excellent environmental adaptability and efficient inference speed. It is particularly suitable for large-scale, automated inspection tasks of transmission lines, significantly improving the accuracy of tower identification, reducing false positives and false negatives, and providing a reliable foundation for subsequent structural state analysis.

[0069] In one optional embodiment of this invention, point cloud classification of candidate targets involves employing a classification method based on traditional statistical features and rule template matching to achieve preliminary discrimination of candidate point cloud regions. In this embodiment, firstly, geometric feature indicators are calculated for each candidate target point cloud region, including overall point cloud height, width, length, volume estimation, mean density distribution, and principal direction distribution of the normal vector. Subsequently, the extracted multidimensional features are compared with the preset tower standard template feature vector for similarity measurement, commonly using methods such as Euclidean distance or cosine similarity calculation. When the similarity exceeds a set threshold, the candidate region is determined to be a tower.

[0070] Regarding its integration with the workflow of this invention, this alternative approach is also executed after the point cloud preprocessing and candidate region selection steps, and its output will be directly used as input data for the structural orientation modeling and attitude angle calculation modules. Therefore, it can seamlessly integrate with the entire workflow of this invention in terms of data flow and processing nodes.

[0071] However, compared with the spatial feature automatic extraction method based on deep learning models in the preferred embodiment, this alternative solution has the following shortcomings:

[0072] This method relies on manually set template rules and feature thresholds, lacks adaptive learning capabilities, and struggles to cope with complex scenarios involving diverse tower types, large scale differences, and significant changes in perspective. In situations with strong background interference in the natural environment (such as large trees or similar communication tower structures), traditional statistical features are prone to overlap, leading to a significant increase in misclassification and missed classification rates. When faced with changes in point cloud acquisition quality (such as uneven point density or sampling defects), fixed rules struggle to maintain classification stability, resulting in low overall robustness. Although the inference efficiency is high, it cannot provide a deep understanding of spatial structure and cannot improve recognition performance through training optimization.

[0073] In contrast, the preferred embodiment of the present invention introduces an improved PointNet++ network structure to automatically model local and global multi-scale spatial features, which has stronger nonlinear modeling capabilities and environmental adaptability. In particular, it can significantly improve the identification accuracy and robustness of towers in complex backgrounds and structurally similar scenarios, making it suitable for deployment in high-reliability application scenarios such as automated inspection and rapid disaster response, and has higher engineering practical value.

[0074] S4. Based on the point cloud region, extract the upper and lower boundary endpoints and construct the structural direction vector connecting the two endpoints.

[0075] Extract the height coordinates of all points from the target point cloud region, and select point sets within the intervals set before the maximum height value and after the minimum height value respectively. Calculate the geometric centroid of each point as the top endpoint and bottom endpoint, and construct the structural direction vector based on the spatial positional relationship between the top endpoint and the bottom endpoint.

[0076] Tower endpoint extraction: Tower endpoint extraction is based on the following basic assumptions: the overall structural height of the tower is significantly higher than the surrounding environment, and the tower point cloud has clear upper and lower boundaries in the vertical direction. This invention proposes an endpoint extraction method that combines point cloud height statistics and local clustering analysis.

[0077] For each sub-block of tower points identified through deep learning classification, the height coordinates (i.e., z-axis coordinates) of all points within it are extracted and denoted as the set {z... i Within this set, select the maximum value z. max The corresponding point is taken as the top point P of the tower. top Select the minimum value z min The corresponding point is taken as the bottom point P of the tower. bottom Specifically defined as:

[0078]

[0079] in, p represents the set of all points in the pole point cloud sub-block. i =(x i ,y i ,z i ) represents the position vector of a single point.

[0080] To improve the robustness of endpoint extraction and prevent extraction errors caused by noise points, this invention does not directly select the highest and lowest points as endpoints in practice. Instead, it selects all points whose height falls within the range of the highest 5% and the lowest 5%, and calculates their geometric centers (centroids) as the top and bottom endpoints. The centroid calculation formula is as follows:

[0081]

[0082] Among them, z 95％ and z 5％ These are the 95th and 5th percentiles of the z-axis height, respectively. top N bottom This corresponds to the number of points. By employing a method based on local statistics, the interference of single-point outliers on endpoint extraction is effectively suppressed, improving the stability and accuracy of vertices and bottom points. After completing endpoint extraction, this invention further generates a tower centerline vector by connecting the top and bottom points. The definition is as follows:

[0083]

[0084] The centerline vector not only reflects the overall spatial orientation of the tower but also provides a precise basis for subsequent tilt angle calculations. This is achieved by aligning the centerline with the standard vertical direction. The angle calculation can determine whether the tower is tilted and classify its health status accordingly.

[0085] S5. Calculate the directional offset angle by the angle between the structural direction vector and the vertical direction, determine the attitude classification, and output the corresponding spatial position and structural vector information.

[0086] The structural direction vector is multiplied by the unit vertical vector in the three-dimensional coordinate system. The unit vertical vector is the unit vector in the positive z-axis direction of the coordinate axis, which represents the vertical reference direction. The angle between the two vectors is calculated by combining the magnitude of the structural direction vector.

[0087] The system determines the level based on the included angle value and the set attitude classification threshold, and outputs the corresponding attitude category label, structural orientation vector, and spatial coordinates of the point cloud region.

[0088] Spatial boundary calculation is performed on the point cloud region that is identified as the target category, and spatial position coordinates are extracted based on the geometric centroid of the point set in the current region. The result output object is constructed by combining the structural orientation vector and the attitude classification label, forming a unified output structure that includes structural orientation, spatial position and attitude information.

[0089] In a preferred embodiment of the present invention, determining the attitude classification is as follows:

[0090] The vector represents the ideal vertical direction, that is, the standard direction the tower should point in under no-tilt conditions. Calculate the centerline vector. unit vector in the vertical direction The angle θ between the two vectors. Since the angle between vectors can be calculated from the dot product of the two vectors and their respective magnitudes, the formula for the tilt angle θ is defined as:

[0091]

[0092] Where "·" represents the vector dot product operation. The Euclidean modulus of the centerline vector is calculated as follows:

[0093]

[0094] The formula for calculating the dot product is:

[0095]

[0096] Because of the unit vector Since the x and y components are 0, only the z-direction component participates in the dot product. The tilt angle θ can be directly obtained using the above formula. The tilt angle θ reflects the degree of deviation between the tower's centerline and the standard vertical direction; the smaller the value, the closer the tower is to an ideal vertical state; the larger the value, the more severe the tower tilt.

[0097] Based on the magnitude of the tilt angle θ, a health status classification standard is set: when the tilt angle θ is less than the normal threshold θ... normal When the tilt angle θ is between θ, it is judged as a normal tower; when the tilt angle θ is between θ, it is judged as a normal tower. normal With danger threshold θ dangerous When the angle of inclination is between θ and θ', it is judged as a tilted tower; when the tilt angle θ exceeds the danger threshold θ'', it is considered a tilted tower. dangerous At that time, it was determined to be a severely tilted or collapsed pole.

[0098] In one optional embodiment of the present invention, determining the attitude classification is as follows:

[0099] An attitude determination method is adopted based on the ratio of the projected length of the structural orientation vector on the vertical axis. In this embodiment, it is not necessary to calculate the angle between the structural orientation and the vertical direction. Instead, the z-axis component of the structural orientation vector is extracted, and its ratio to the magnitude of the structural orientation vector is used as the "direction verticality coefficient". When this coefficient is close to 1, it indicates that the structure tends to be vertical; when the value is significantly lower than 1, it indicates that the structure has obvious skewness.

[0100] In the actual execution process, the constructed structural direction vector is first obtained, and its component form is (v x ,v y ,v z ); Calculate its modulus.

[0101] Then calculate v z / |v| serves as the verticality coefficient. Segmented thresholds are set based on this coefficient. For example, when the value is greater than the set threshold α (e.g., 0.98), it is judged as a normal tower; when the value is within the interval (β, α) (e.g., 0.95~0.98), it is judged as a slightly tilted tower; when the value is lower than β, it is judged as severely tilted or collapsed.

[0102] This optional solution maintains the same overall process as the present invention, still proceeding after the structural orientation vector extraction step, and outputting the corresponding attitude classification label and structural information result object. Its advantage lies in a simpler calculation process, avoiding inverse cosine calculation, and making it suitable for front-end terminal devices with limited resources or relatively low accuracy requirements.

[0103] However, compared to the attitude classification method based on the complete included angle calculation in the preferred embodiment, this scheme has the following shortcomings:

[0104] The accuracy of the judgment depends on the empirical parameter settings, is not sensitive to small structural direction angles, and has a blurred classification boundary; it is difficult to deal with some special structures that are not vertical but have reasonable postures, such as non-standard towers; it cannot accurately quantify the offset angle, lacks traceability, and is difficult to use as a reliable quantitative indicator for operation and maintenance judgment.

[0105] In contrast, the preferred embodiment of the present invention accurately obtains the true angle between the structural direction and the vertical direction by calculating the vector dot product and the modulus. It has good geometric interpretation and physical consistency, clear classification boundaries, and the results are quantifiable and comparable. It is suitable for key scenarios such as fault early warning, trend monitoring and rapid post-disaster response in power inspection, and has higher accuracy and practical value.

[0106] Example 2 is the second embodiment of the present invention, which differs from the previous embodiment in that:

[0107] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0108] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0109] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0110] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0111] Example 3 is the third embodiment of the present invention. This embodiment provides an automatic identification system for transmission line towers based on lidar, including a point cloud data processing module, a candidate target extraction and point cloud classification module, a structural orientation construction and attitude analysis module, and an output and structural information generation module.

[0112] The point cloud data processing module receives the raw 3D point cloud data collected by the lidar device, cleans and preprocesses the data, removes ground points with a height lower than the tower structure, vegetation points with a density higher than the threshold and arranged in a disordered manner, and isolated noise points formed at the edges, and outputs a subset of point cloud with continuity, directionality and geometric integrity to form a structured point cloud region.

[0113] The candidate target extraction and point cloud classification module is based on the spatial division of the structural point cloud region, and selects candidate regions with tower geometric features from it, while completing the point cloud classification operation.

[0114] First, the structural point cloud is grouped by spatial proximity, and the geometric height range and normal vector direction distribution of each group are calculated. Point cloud clusters that do not conform to the tower characteristics are filtered out. The filtered candidate target point cloud is input into the classification model to extract local and global structural features and output the point cloud region belonging to the tower category.

[0115] The structure orientation construction and attitude analysis module performs structural modeling and attitude state analysis on point cloud regions identified as tower categories: extracting the upper and lower boundary points of the tower, calculating the geometric centroids of the top and bottom within a specified height range, and constructing a structural orientation vector connecting the two ends; calculating the angle between this vector and the unit vertical vector in three-dimensional space to obtain the structural offset angle; and comparing the angle value with a preset attitude classification threshold to determine the tower state.

[0116] The output and structural information generation module extracts the boundaries of each target point cloud region identified as a tower and calculates the spatial geometric centroid of the current region as the current position coordinates; it encapsulates the attitude classification label and structural direction vector together into a result output object, forming unified output information containing spatial position, structural direction and attitude state.

[0117] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An automatic identification method for transmission line towers based on lidar, characterized in that: include, Collect three-dimensional point cloud data of the area along the power transmission line, and preprocess the spatial point set to obtain the structural point cloud region. The structural point cloud region is spatially divided, and point cloud segments that meet preset conditions are selected as candidate targets based on the geometric height and normal direction of each region. Candidate targets are classified into point clouds to obtain point cloud regions belonging to the target category; Based on the point cloud region, extract the upper and lower boundary endpoints and construct the structural direction vector connecting the two endpoints; The orientation offset angle is calculated by the angle between the structural orientation vector and the vertical direction, the attitude classification is determined, and the corresponding spatial position and structural vector information are output.

2. The automatic identification method for transmission line towers based on lidar as described in claim 1, characterized in that: The method involves collecting three-dimensional point cloud data along the transmission line area, obtaining a spatial point set, and preprocessing it to obtain a structural point cloud region, including... The original point cloud is cleaned to remove point sets that are below the range of the main structure in the height dimension. Point cloud regions that do not meet the continuous structural features are identified based on local point density and spatial arrangement direction. Noise points in the point cloud regions are removed to obtain three-dimensional point cloud data with continuous structural features, forming structural point cloud regions.

3. The automatic identification method for transmission line towers based on lidar as described in claim 2, characterized in that: The step of spatially dividing the structural point cloud region and selecting point cloud segments that meet preset conditions as candidate targets based on the geometric height and normal vector direction of each region includes: The point cloud data in the structural point cloud region is grouped according to spatial proximity, and the height range and normal vector direction distribution characteristics are calculated for each group. By judging whether the height range meets the first condition and whether the main direction of the normal vector is located in the preset vertical vector interval, the point cloud groups that do not meet the conditions are filtered out, and the remaining groups are retained as candidate target regions.

4. The automatic identification method for transmission line towers based on lidar as described in claim 3, characterized in that: The step of classifying candidate targets into point clouds to obtain point cloud regions belonging to the target category includes, A multi-scale spatial neighborhood is constructed based on the point cloud data of the candidate target, and the spatial distribution relationship of the points in the neighborhood is feature-encoded. The spatial structure expression of the point cloud region is formed by local feature extraction and global feature aggregation, and classification judgment is performed to output the point cloud region that is judged as the target category.

5. The automatic identification method for transmission line towers based on lidar as described in claim 4, characterized in that: The step of extracting the upper and lower boundary endpoints based on the point cloud region and constructing the structural direction vector connecting the two endpoints includes... Extract the height coordinates of all points from the target point cloud region, and select point sets within the intervals set before the maximum height value and after the minimum height value respectively. Calculate the geometric centroid of each point as the top endpoint and bottom endpoint, and construct the structural direction vector based on the spatial positional relationship between the top endpoint and the bottom endpoint.

6. The automatic identification method for transmission line towers based on lidar as described in claim 5, characterized in that: The process of calculating the directional offset angle by using the angle between the structural direction vector and the vertical direction, determining the corresponding attitude classification, and outputting the corresponding spatial position and structural vector information includes: The structural direction vector is multiplied by the unit vertical vector in the three-dimensional coordinate system. The unit vertical vector is the unit vector in the positive z-axis direction of the coordinate axis, which represents the vertical reference direction. The angle between the two vectors is calculated by combining the magnitude of the structural direction vector. The system determines the level based on the included angle value and the set attitude classification threshold, and outputs the corresponding attitude category label, structural orientation vector, and spatial coordinates of the point cloud region.

7. The automatic identification method for transmission line towers based on lidar as described in claim 6, characterized in that: The spatial location and structural vector information corresponding to the output also includes, Spatial boundary calculation is performed on the point cloud region that is identified as the target category, and spatial position coordinates are extracted based on the geometric centroid of the point set in the current region. The result output object is constructed by combining the structural orientation vector and the attitude classification label, forming a unified output structure that includes structural orientation, spatial position and attitude information.

8. An automatic identification system for transmission line towers based on lidar, employing the automatic identification method for transmission line towers based on lidar as described in any one of claims 1 to 7, characterized in that, It includes: a point cloud data processing module, a candidate target extraction and point cloud classification module, a structure orientation construction and attitude analysis module, and an output and structure information generation module; The point cloud data processing module receives the raw 3D point cloud data collected by the lidar device, cleans and preprocesses the data, removes ground points with a height lower than the tower structure, vegetation points with a density higher than the threshold and arranged in a disordered manner, and isolated noise points formed at the edges, and outputs a subset of point cloud with continuity, directionality and geometric integrity to form a structured point cloud region. The candidate target extraction and point cloud classification module is based on the spatial division of the structural point cloud region, and selects candidate regions with tower geometric features from it, while completing the point cloud classification operation. First, the structural point cloud is grouped by spatial proximity, and the geometric height range and normal vector direction distribution of each group are calculated. Point cloud clusters that do not conform to the tower characteristics are filtered out. The filtered candidate target point cloud is input into the classification model to extract local and global structural features and output the point cloud region belonging to the tower category. The structure orientation construction and attitude analysis module performs structural modeling and attitude state analysis on point cloud regions identified as tower categories: extracting the upper and lower boundary points of the tower, calculating the geometric centroids of the top and bottom within a specified height range, and constructing a structural orientation vector connecting the two ends; calculating the angle between this vector and the unit vertical vector in three-dimensional space to obtain the structural offset angle; and comparing the angle value with a preset attitude classification threshold to determine the tower state. The output and structural information generation module extracts the boundary of each target point cloud region identified as a tower and calculates the spatial geometric centroid of the current region as the current position coordinates. The attitude classification label and the structural orientation vector are encapsulated together as a result output object, forming a unified output information that includes spatial location, structural orientation, and attitude state.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the automatic identification method for transmission line towers based on lidar as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the automatic identification method for transmission line towers based on lidar as described in any one of claims 1 to 7.

Images