Satellite image-based high-voltage transmission line three-dimensional model reconstruction method and system

Abstract

This invention relates to the field of power engineering identification technology, specifically to a method and system for reconstructing a 3D model of a high-voltage transmission line based on satellite imagery. The method includes: acquiring satellite imagery; identifying tower types; acquiring geographic information data and, in conjunction with the satellite imagery and tower types, performing tower height inversion to obtain the tower height; and reconstructing a 3D model of the high-voltage transmission line using a constructed parametric model library based on the satellite imagery, tower types, and tower height. This solution can efficiently and stably construct a 3D model of a high-voltage transmission line using satellite imagery, thus solving the problem of existing methods relying on expensive data sources and being difficult to apply on a large scale.

Description

Technical Field

[0001] This invention relates to the field of power engineering identification technology, specifically to a method and system for reconstructing a three-dimensional model of a high-voltage transmission line based on satellite imagery. Background Technology

[0002] As a key supporting structure for transmission lines, power transmission towers' 3D models are of significant value in power system analysis, operation, maintenance, and planning. Traditional 3D modeling primarily relies on design drawings, using software such as AutoCAD and 3ds Max to construct models manually or semi-automatically. This method is inefficient, costly, and suffers from lagging model data updates, making it difficult to reflect real-time on-site conditions. To address these limitations, various automated 3D reconstruction technologies have emerged in recent years, primarily including image-based 3D modeling and 3D laser scanning modeling as the two mainstream methods.

[0003] Image-based 3D modeling techniques have advantages such as wide applicability and strong scalability. For example, the model-driven method based on tilted UAV imagery: by using a predefined parametric tower library and deformable component model, the quadrangular prism structure of the tower body is fitted based on symmetry and coplanar constraints, and the tower head parameters are optimized using a Markov chain Monte Carlo (MCMC) sampler. For complex structures and occlusion problems, the Neural Radiation Field (NeRF) technique is introduced: by improving the segmentation model, combining Canny edge detection and binarization processing to accurately separate the hollow areas of the tower, the spatial geometry is modeled using signed distance field (SDF), and the surface reconstruction is achieved by combining the volume rendering formula.

[0004] With the development of Light Detection and Ranging (LiDAR) technology, its advantage of rapidly and accurately acquiring massive amounts of 3D point cloud data has gradually made it the mainstream method.

[0005] To address the challenges of complex and diverse high-voltage power tower structures, researchers proposed a data-driven 3D line fitting method based on RANSAC to reconstruct the four main legs. This method combines a shape context algorithm to identify the tower head type and estimates key parameters using a Metropolis-Hastings sampler and simulated annealing. Later, by introducing a strategy combining Transformer and geometric statistical features, the tower base and frame were reconstructed based on prior structural knowledge and a data-driven approach. A standard parameter model library for the tower head was established, and a hybrid driving strategy of Consistent Point Drift (CPD) and simulated annealing was used to reconstruct the tower head. Other researchers have proposed fusing 3D point cloud and 2D image data using a generalized template structure and constructing an insulator model using line-tower endpoint matching. Alternatively, they have proposed registering image point clouds and lidar point clouds in a unified coordinate system and using voxel grid resampling to accelerate the ICP algorithm, enabling the simultaneous reconstruction of insulators and other small components such as lightning protection wires.

[0006] In summary, existing technologies either focus on detailed local modeling of a single object, or only extract parameters of the transmission line without constructing a 3D model, or rely on non-satellite data sources and existing vector data to build a 3D model. There is a lack of a systematic method that can comprehensively utilize high-resolution satellite imagery to achieve automated, low-cost 3D model reconstruction of large-scale high-voltage transmission lines.

[0007] Specifically, current research and model building of transmission lines have the following shortcomings: 1. Limited data collection coverage: Most current 3D modeling methods rely on UAV low-altitude aerial photography or LiDAR point cloud data collection. Due to limitations such as flight permits, battery life, and terrain and weather conditions, it is difficult to achieve rapid and complete data collection over long distances and along multiple routes. 2. High cost; The overall investment in drone flights, lidar equipment procurement and maintenance is huge, and the acquisition of large-scale line data relies on a multi-flight, regional collection strategy, which makes it difficult to reduce the overall cost and has obvious shortcomings in terms of economy and security. 3. Difficult to apply to large-scale scenarios; Most research based on point cloud and low-altitude image data focuses on the detailed modeling of individual towers or substations, while research based on satellite imagery focuses more on the planning and positioning of transmission lines, and neither has formed a large-scale line modeling system.

[0008] Therefore, there is an urgent need for a method and system for reconstructing three-dimensional models of high-voltage transmission lines based on satellite imagery, which can efficiently and stably construct three-dimensional models of high-voltage transmission lines through satellite imagery, in order to solve the problem that existing methods rely on expensive data sources and are difficult to apply on a large scale. Summary of the Invention

[0009] One of the objectives of this invention is to provide a method for reconstructing three-dimensional models of high-voltage transmission lines based on satellite imagery. This method can efficiently and stably construct three-dimensional models of high-voltage transmission lines using satellite imagery, thereby solving the problem that existing methods rely on expensive data sources and are difficult to apply on a large scale.

[0010] The basic solution provided by this invention is a method for reconstructing a three-dimensional model of a high-voltage transmission line based on satellite imagery, comprising: S1. Acquire satellite imagery, identify tower types, and obtain tower types; S2. Obtain geographic information data and combine it with satellite imagery and tower type to perform tower height inversion and obtain the tower height; S3. Based on satellite imagery, tower type, and tower height, a three-dimensional model of the high-voltage transmission line is reconstructed using a constructed parametric model library.

[0011] Further, S1 includes: S101. Locate the towers in the satellite imagery and obtain the tower area; S102. Extract the preset area shadow in the tower area; S103. Based on the extracted preset area shadow, the pole type is identified through the constructed type recognition module; the type recognition model is used to input the extracted preset area shadow and output the probability of the pole type; the pole type with the highest probability value is obtained as the recognition result.

[0012] Furthermore, S2 includes: S201, extracting and constructing a multi-dimensional feature vector from satellite imagery, tower type, and geographic information data, including: attribute features, geometric features, and environmental features; S202. Based on the feature vector, the tower height is obtained by performing height inversion through the constructed regression model.

[0013] Furthermore, the regression model is constructed using the LightGBM framework; The regression model is constructed in a forward stepwise manner, and the prediction accuracy is improved by gradually fitting the residuals of the current model. The total number of iterations is , No. Wheel model output : in This represents the input feature vector. This is the output of the previous round of predictions. For the first The decision tree trained in the round, The learning rate is used to control the contribution of each tree to the final model. The LightGBM framework employs gradient-based one-sided sampling and mutually exclusive feature binding techniques during training. Its objective function ℒ is defined in each iteration as: in The loss function; This is a regularization term used to control model complexity; The number of training samples, For the first The true height of each sample To predict the height for the model.

[0014] Furthermore, S3 includes: S301. Based on the attribute features and geometric features extracted from the satellite image, the model is located and oriented. According to the tower type, the corresponding tower reference model is retrieved from the constructed parametric model library, and the height of the reference model is geometrically adjusted according to the tower height. S302. Based on the preset conductor connection points in the insulator model corresponding to the benchmark model, automatically generate three-dimensional conductors between towers to obtain a three-dimensional model of the high-voltage transmission line.

[0015] Furthermore, the positioning includes: According to the latitude and longitude of the tower ground elevation Calculate its position in a three-dimensional Cartesian coordinate system. : in The radius of curvature of the circle is denoted as . Let the square of the ellipsoid's eccentricity be... The semi-major axis of the reference ellipsoid.

[0016] Furthermore, the orientation includes: The angle of the tower in the horizontal plane To determine spatial orientation, the tower's orientation angle in the horizontal plane. This is determined by its function in the line and its relationship with adjacent positions; the current tower The plane coordinates are The coordinates of its preceding and following adjacent towers are respectively , The orientation angle is calculated as follows: like For a straight-line tower, its crossarm should be perpendicular to the line connecting adjacent towers. like For corner towers, the long arm of the crossarm should point in the direction of the outer bisector of the line corner: in from Measurement is performed clockwise in the positive direction of the axis. , They are vectors exist and Components in direction.

[0017] Furthermore, the benchmark model includes: the standard dimensions, geometric frame, crossarm layout and topological connection relationship of the corresponding tower model, and two types of insulator string models, tension type and straight type, are constructed according to the main morphology of the insulator under stress. The reference model has preset conductor connection points, and the number of insulator discs is adjusted according to the voltage level; The insulator string model serves as an auxiliary component, serving as a reference model alongside the tower model.

[0018] Furthermore, the geometric adjustment of the reference model height based on the tower height includes: Through the overall scaling factor Geometric adjustments are made to the baseline model, where The tower height is predicted based on a regression model. This is the original height of the reference model; the scaling operation is applied simultaneously to the tower body and insulator components.

[0019] The second objective of this invention is to provide a satellite imagery-based 3D model reconstruction system for high-voltage transmission lines, which can efficiently and stably construct 3D models of high-voltage transmission lines using satellite imagery, thereby solving the problem that existing methods rely on expensive data sources and are difficult to apply on a large scale.

[0020] The present invention provides a second basic solution: a three-dimensional model reconstruction system for high-voltage transmission lines based on satellite imagery, used to execute the above-mentioned three-dimensional model reconstruction method for high-voltage transmission lines based on satellite imagery.

[0021] Beneficial effects: This solution utilizes satellite imagery, specifically high-resolution commercial satellite imagery. Based on this single data source, the 3D model of high-voltage transmission lines is reconstructed, which is easy to obtain and offers broad satellite imagery coverage. Specifically, this solution identifies tower types using satellite imagery, then combines this with geographic information data, and finally, using the satellite imagery and tower type data, performs tower height inversion to obtain the tower height. Finally, based on the satellite imagery, tower type, and tower height, the 3D model of the high-voltage transmission line is reconstructed using a pre-built parametric model library.

[0022] This solution is designed to address the characteristics of high-resolution satellite imagery and the inherent limitations of this data source. Compared to existing solutions, this solution offers the following advantages: 1. Wide coverage and low cost model acquisition: For the first time, it has been achieved to complete the three-dimensional reconstruction of a 100-kilometer-level high-voltage transmission line by relying solely on satellite imagery that is wide coverage, easy to obtain and low cost, breaking through the fundamental limitations of airborne lidar and other methods in terms of coverage, cost, terrain, meteorology and airspace. 2. It has achieved an end-to-end automated processing loop from satellite imagery to 3D models. Addressing the core shortcomings of low information density and single perspective in satellite imagery, it integrates deep learning, machine learning, and parametric modeling technologies to achieve end-to-end automatic generation of 3D models from satellite imagery. This significantly reduces manual intervention and improves the efficiency and stability of line modeling. It makes rapid modeling and dynamic updating of cross-regional and long-distance transmission lines feasible and has the potential for large-scale business applications.

[0023] This solution addresses the problem of traditional methods relying on expensive data sources and being difficult to apply on a large scale. It proposes an automated processing workflow based on satellite imagery, which achieves complete reconstruction from two-dimensional remote sensing images to three-dimensional line models through the organic combination of type recognition, height inversion, and model-driven 3D reconstruction.

[0024] In summary, this solution can efficiently and stably construct three-dimensional models of high-voltage transmission lines using satellite imagery, thus solving the problem that existing methods rely on expensive data sources and are difficult to apply on a large scale. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating an embodiment of the method for reconstructing a three-dimensional model of a high-voltage transmission line based on satellite imagery according to the present invention. Figure 2 This is a satellite image of the tower head shadow of the gantry tower in an embodiment of the three-dimensional model reconstruction method for high-voltage transmission lines based on satellite imagery of the present invention; Figure 3 This is a satellite image of the tower head shadow of the drum-shaped tower in an embodiment of the three-dimensional model reconstruction method for high-voltage transmission lines based on satellite imagery of the present invention; Figure 4 This is a satellite image of the tower head shadow of the wine glass tower in an embodiment of the high-voltage transmission line three-dimensional model reconstruction method based on satellite imagery of the present invention; Figure 5 This is a satellite image of the tower head shadow of the cat-head tower in an embodiment of the high-voltage transmission line three-dimensional model reconstruction method based on satellite imagery of the present invention; Figure 6 This is a schematic diagram of the reference model of the pylon in an embodiment of the three-dimensional model reconstruction method for high-voltage transmission lines based on satellite imagery of the present invention. Figure 7This is a schematic diagram of the reference model of the drum-shaped tower in an embodiment of the three-dimensional model reconstruction method for high-voltage transmission lines based on satellite imagery of the present invention; Figure 8 This is a schematic diagram of the reference model of the wine glass tower in an embodiment of the three-dimensional model reconstruction method for high-voltage transmission lines based on satellite imagery of the present invention. Figure 9 This is a schematic diagram of the reference model of the cat-head tower in an embodiment of the three-dimensional model reconstruction method for high-voltage transmission lines based on satellite imagery of the present invention. Figure 10 This is a schematic diagram of a three-dimensional model of a high-voltage transmission line in an embodiment of the method for reconstructing a three-dimensional model of a high-voltage transmission line based on satellite imagery according to the present invention. Detailed Implementation

[0026] The following detailed description illustrates the specific implementation method: The markings in the accompanying drawings include: Example 1 This embodiment provides a method for reconstructing a three-dimensional model of a high-voltage transmission line based on satellite imagery, as shown in the attached figure. Figure 1 As shown, it includes the following: S1. Acquire satellite imagery, identify tower types, and obtain tower types; In the 3D reconstruction of transmission lines, tower type is a crucial factor affecting the accuracy of the 3D model. Due to the limitations of the overhead view and resolution of high-resolution satellite imagery, the fine structure of the tower itself is often difficult to identify directly. Classification based solely on tower images is challenging and inaccurate. However, at specific solar altitude angles, the shadow outline cast by the tower (especially the tower head) on the ground is strongly correlated with its 3D structure, and the shadow characteristics differ significantly among different tower types. Therefore, this scheme uses tower shadows as the primary information source for identifying tower types under satellite imagery conditions. By constructing a dedicated shadow image dataset and a deep learning model, effective classification is achieved even with insufficient image features. The specific process is as follows: S101. Locate the towers in the satellite imagery and obtain the tower area; This solution is based on deep learning principles to locate towers and obtain tower areas, specifically including: Construct a Faster R-CNN architecture to detect pole targets in high-resolution satellite imagery and compute feature representations for each candidate region; A Region Proposal Network (RPN) is used to generate multi-scale anchor boxes and extract candidate tower locations. Multi-level semantic information is fused through a Feature Pyramid Network (FPN) to address tower targets of different sizes. Redundant detection results are removed by non-maximum suppression (NMS) algorithm to obtain the final tower detection bounding box, which is used as the tower region.

[0027] To obtain the geographic coordinates of the tower, the center point of the bottom edge of the tower detection bounding box is used as its representative position on the image plane, and combined with the georeferenced information from satellite imagery, it is converted into latitude and longitude coordinates. This completes the geographical location of the tower.

[0028] S102. Extract the preset area shadow in the tower area; Based on tower positioning, this scheme uses the tower head shadow (i.e., the preset area shadow is the tower head shadow) as the center and uniformly crops the satellite image into 244×244 pixel sub-images. To enhance shadow features and reduce background interference, the sub-images undergo grayscale conversion and contrast enhancement processing. Histogram equalization is used to improve the visual effect of the preset area shadow. An example of the preset area shadow in the original satellite image is shown below. Figure 2 , 3 Figures 4 and 5 show the tower heads of the four types of towers: the Dry Pyramid, the Drum Pyramid, the Goblet Pyramid, and the Cat Head Pyramid, respectively. Through the above processing, a tower head shadow dataset is constructed for the four types of towers (Dry Pyramid, Drum Pyramid, Goblet Pyramid, and Cat Head Pyramid), providing training and validation samples for subsequent classification models.

[0029] S103. Based on the extracted preset area shadow, the tower type is identified through the constructed type recognition module.

[0030] The type recognition model takes the extracted shadow area (i.e., the sub-image after grayscale and contrast enhancement processing, a 244×244 grayscale image) as input and outputs the probability of the tower type; the tower type with the highest probability value is obtained as the recognition result. The type recognition model consists of several convolutional layers and fully connected layers, and uses the ReLU activation function, local response normalization (LRN) layers, and Dropout layers. In this embodiment, it consists of 5 convolutional layers and 3 fully connected layers, and uses ReLU activation function, local response normalization (LRN), and Dropout techniques to enhance generalization ability.

[0031] The AlexNet deep network architecture is used to build and train the type recognition model. The specific process includes: The type recognition model is initialized using ImageNet pre-trained weights and transfer learning is performed on the Tower Head Shadow dataset; During training, the objective is to minimize the multi-class cross-entropy loss, as shown in the following formula: in For batch size, For the number of categories, For the sample In the The true label of the class, This represents the normalized probability predicted by the network. The parameters are updated using the Adam optimizer, with the initial learning rate set to... When the validation set loss no longer decreases over five consecutive training epochs, the learning rate is reduced by a scaling factor of 0.5, with a lower bound of [value missing]. Meanwhile, an EarlyStopping strategy is employed to terminate training early to prevent overfitting if the validation set accuracy or loss does not show significant improvement within 10 epochs.

[0032] After training, the best-performing type recognition model on the validation set is selected for testing. Overall accuracy (OA) and average accuracy (AA) are used to evaluate the model's accuracy. During the inference phase, preprocessed sub-images are input into the trained model, and the type recognition model outputs score vectors for each category. The probability is converted to class probability using the Softmax function: Ultimately Determine the category label of the tower to provide reliable category information for subsequent height inversion and 3D reconstruction.

[0033] S2. Obtain geographic information data and combine it with satellite imagery and tower type to perform tower height inversion and obtain the tower height; Tower height is a key geometric parameter that determines the vertical scale accuracy of a model in 3D reconstruction. Traditional height estimation methods based on shadow length or single-image photogrammetry are significantly affected by factors such as changes in solar altitude angle, shadow occlusion, and image side-view distortion when applied to large-scale power transmission corridors with significant terrain undulations, resulting in large errors and poor stability.

[0034] This scheme recognizes that the design height of transmission line towers is not an isolated parameter, but a complex function determined by multiple factors, including their structural attributes (tower type, function), the spatial geometry of the line (span, location), and the surrounding environment (topography, land cover). To address the difficulty of accurately depicting this complex relationship using a single feature or simple model, this scheme proposes and constructs for the first time a multi-dimensional collaborative feature system of "attributes, geometry, and environment" specifically for tower height inversion from satellite imagery. This system systematically integrates multi-source information extracted from satellite imagery and establishes a high-precision inversion model based on advanced machine learning algorithms, thereby achieving robust and reliable estimation of tower height under satellite imagery conditions. The specific process is as follows: S201. Extract and construct multi-dimensional feature vectors from satellite imagery, tower type, and geographic information data; Features in three dimensions are extracted and constructed from tower type, satellite imagery and geographic information data to form a comprehensive feature vector, which serves as the input to the height inversion model.

[0035] Detailed features for each dimension are shown in Table 1. Among them, attribute features directly reflect the design specifications of the tower itself; geometric features describe the absolute position, relative spatial relationship, and direct visible scale of the tower in the image; environmental features depict the potential impact of the local environment on engineering design and safety margin. The synergistic use of these features is the core of this scheme to overcome the scarcity of satellite imagery information and achieve high-precision inversion.

[0036] Table 1: Feature Dimension Table S202. Based on the feature vector, the tower height is obtained by performing height inversion through the constructed regression model.

[0037] To efficiently learn the complex nonlinear mapping relationship between the above multidimensional features and the actual height of the tower, this scheme uses the LightGBM framework (gradient boosting framework) to construct a regression model.

[0038] The regression model is constructed using a forward stepwise approach, improving prediction accuracy by progressively fitting the residuals of the current model. Let the total number of iterations be... , No. Wheel model output It is given by the following formula: in This represents the input feature vector. This is the output of the previous round of predictions. For the first The decision tree trained in the round, The learning rate controls the contribution of each tree to the final model.

[0039] The LightGBM framework employs gradient-based one-sided sampling and mutually exclusive feature binding techniques during training to improve training efficiency and reduce memory usage. Its objective function ℒ is defined in each iteration as: in For the loss function, mean squared error (MSE) is selected in this embodiment: ; This is a regularization term used to control model complexity; The number of training samples, For the first The true height of each sample The model is used to predict height. During training, the weights of the leaf nodes of each tree are optimized using gradient descent, and an early stopping strategy and learning rate decay mechanism are combined to ensure model convergence and good generalization ability.

[0040] After the model is trained, its performance is evaluated on an independent test set, and the root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination are calculated. : ; ; in This represents the average height of the actual heights in the test set.

[0041] S3. Based on multi-dimensional data of high-voltage transmission lines (satellite images, tower types, and tower heights), a three-dimensional model of the high-voltage transmission line is reconstructed using a constructed parametric model library.

[0042] While the above steps can extract some attributes and geometric parameters such as the spatial location, type, and estimated height of power transmission lines from satellite imagery, they still cannot obtain the complete and detailed three-dimensional geometric structure of power equipment such as power transmission lines and insulators. Therefore, relying solely on this data source to achieve efficient and stable reconstruction of transmission lines still faces the severe challenge of recovering complex three-dimensional structures and topological relationships from sparse and incomplete two-dimensional information.

[0043] To address this, this embodiment employs a hybrid 3D reconstruction strategy driven by both prior models and image data. The core of this strategy lies in constructing a parameterized transmission line component model library as prior knowledge to compensate for the lack of fine structural information in the images. Simultaneously, geometric and attribute data extracted from satellite imagery are used to drive the instantiation, spatial registration, and conductor generation of the standard model. Ultimately, this achieves a reliable conversion from 2D remote sensing information to a 3D physical scene.

[0044] The specific steps are as follows: S301. Based on the attribute features and geometric features extracted from the satellite image, the model is located and oriented. According to the tower type, the corresponding tower reference model is retrieved from the constructed parametric model library, and the height of the reference model is geometrically adjusted according to the tower height. The model location and orientation are used to determine the position and orientation of each reference model in the 3D map. Based on the transmission line design specifications and typical design drawings issued by the State Grid Corporation of China, parametric benchmark models were constructed in a modeling environment using a human-computer interaction approach for the four main tower types (T-shaped tower, drum-shaped tower, goblet-shaped tower, and cat-head tower). Each benchmark model defines the standard dimensions, geometric frame, crossarm layout, and topological connection relationships of the corresponding tower type. Simultaneously, based on the main forms of insulators under stress, tension-type (…) Figure 6 and Figure 7 (Middle blue model) and linear ( Figure 8 and Figure 9 Two types of insulator string models (yellow model in the middle). The reference model has pre-set precise conductor connection points (…). Figure 6 , 7 (Red dots in 8, 9, etc.), and adjust the number of insulator discs according to the voltage level. The above insulator models, as auxiliary components, are integrated and managed with the tower models to form a complete parametric model library of "tower-insulator"; Figure 6 , 7 Figures 8 and 9 show the baseline models after four types of tower-type integrated insulator strings.

[0045] The model instantiation process is driven collaboratively by attribute features and geometric features extracted from satellite images, as detailed below: First, positioning and orientation are performed; spatial positioning is achieved through geographic coordinate transformation; the latitude and longitude of the tower are then... ground elevation Substitute into the following formula to calculate its position in the three-dimensional Cartesian coordinate system. This ensures that the spatial distribution of the model is consistent with the real geographical environment; in The radius of curvature of the circle is denoted as . Let the square of the ellipsoid's eccentricity be... The semi-major axis of the reference ellipsoid.

[0046] Spatial orientation is determined by the orientation angle of the tower in the horizontal plane. The orientation angle of the tower in the horizontal plane is as follows. This is determined by its function in the line and its adjacent position relationship. Let the current tower be... The plane coordinates are The coordinates of its preceding and following adjacent towers are respectively , The orientation angle is calculated as follows: like For a straight-line tower, its crossarm should be perpendicular to the line connecting adjacent towers. like For corner towers, the long arm of the crossarm should point in the direction of the outer bisector of the line corner: in from Measurement is performed clockwise in the positive direction of the axis. , They are vectors exist and Components in direction.

[0047] Secondly, the baseline model is acquired and its height is adjusted (instantiated); based on the determination of the model's spatial position and orientation, the corresponding baseline model is retrieved from the parametric model library according to the tower type identification results; Through the overall scaling factor Geometric adjustments are made to the baseline model, where The tower height is predicted based on a regression model. This represents the original height of the benchmark model. The scaling operation is applied simultaneously to the tower body and insulator components, thereby adapting its geometric dimensions to the predicted actual height while maintaining the overall structural proportions and topological relationships of the model.

[0048] S302. Based on the preset conductor connection points in the insulator model corresponding to the benchmark model, automatically generate three-dimensional conductors between towers to obtain a three-dimensional model of the high-voltage transmission line. After completing the positioning, orientation, and instantiation of all towers, continuous conductors are still missing in the line scenario. Therefore, in this embodiment, based on the predefined conductor splicing points on the insulator models of adjacent towers, the catenary algorithm is used to simulate the natural shape of the conductor under its own weight and tension, automatically generating a three-dimensional conductor. The final output is a three-dimensional model of a high-voltage transmission line with a complete structure and accurate spatial topology. The reconstruction result is as follows: Figure 10 As shown.

[0049] This solution uses satellite imagery to efficiently and stably construct 3D models of high-voltage transmission lines, addressing the problem that existing methods rely on expensive data sources and are difficult to apply on a large scale. Compared to existing methods for acquiring and modeling transmission line information, which mainly rely on airborne lidar data, satellite imagery, as a wide-area, low-cost data source, can systematically solve the problems existing in current technologies.

[0050] Furthermore, the design addresses the inherent limitations of this data source, solving the problem of how to efficiently and stably construct large-scale 3D scene models of transmission lines suitable for macroscopic visualization and business management of power grids, relying solely on a single data source of high-resolution commercial satellite imagery. Specifically, this includes: To address the problem of reliably extracting pole type information from single-view, limited-resolution satellite imagery; existing research mostly relies on multi-view or three-dimensional information provided by UAV imagery or lidar point clouds, while the top-down perspective of satellite imagery makes it difficult to directly distinguish the fine structure of poles. This solution extracts shadows from a preset area through image processing, and then identifies the type of pole by constructing a type recognition model. The type recognition model is built and trained using the AlexNet deep network architecture, which can accurately identify the pole and provides reliable category information for subsequent height inversion and 3D reconstruction. This solution breaks through existing limitations and enables automatic and accurate identification of key information such as pole type from satellite imagery.

[0051] This addresses the problem of reliably retrieving the 3D height of power transmission towers from only 2D satellite imagery; traditional methods based on shadow-based length measurements or simple geometric relationships suffer from large errors and poor stability in large-scale, complex terrain scenarios. This solution establishes a machine learning model (regression model) that integrates features from multiple image sources to achieve a reliable and high-precision mapping from two-dimensional planar information to the three-dimensional height parameters of the tower. Specifically, this solution constructs a multi-dimensional feature system of "attribute-geometry-environment", which integrates exclusive engineering features derived from the design specifications and spatial topology of transmission line engineering, such as "tower type", "span", "crossarm projection length", and "corner function", thereby achieving reliable estimation of its height under single-view optical image conditions.

[0052] This addresses the problem of efficient generation of structured 3D models driven by sparse image features; existing 3D reconstruction methods for transmission lines are mostly driven by rich point cloud / multi-view image data. This solution creates a hybrid 3D reconstruction strategy driven by prior models and image data. Specifically, it uses sparse but crucial parameters (location, height, type, and orientation) extracted from the imagery to drive a prior parametric model library, thereby efficiently generating 3D models of 100-kilometer-scale lines that are structurally sound, topologically accurate, and precisely matched to the real geographical environment. In particular, the driving force behind this solution's type identification lies in using prior structural knowledge of tower types to compensate for the scarcity of structural information in satellite imagery 3D reconstruction. Furthermore, it incorporates unique compensatory designs, such as automatically calculating the orientation of corner towers based on the spatial relationship between adjacent tower locations. These designs leverage the engineering knowledge inherent in each type of model to compensate for the lack of direct observation in the imagery data. The aim is to utilize the engineering prior knowledge inherent in the models to compensate for the inherent deficiencies in direct observation from satellite imagery.

[0053] This solution effectively overcomes the inherent defect of low information density in satellite imagery by using a technical process of shadow feature recognition, engineering feature fusion and inversion, and parameter and model hybrid-driven reconstruction, and achieves a stable transformation from wide-area, low-cost data to directly applicable 3D scenes.

[0054] This embodiment also provides a system for reconstructing a three-dimensional model of a high-voltage transmission line based on satellite imagery, used to execute the above-mentioned method for reconstructing a three-dimensional model of a high-voltage transmission line based on satellite imagery, including: The type recognition module is used to acquire satellite imagery, identify the tower type, and obtain the tower type. The height analysis module is used to acquire geographic information data and, in combination with satellite imagery and tower type, perform tower height inversion to obtain the tower height. The model generation module is used to reconstruct three-dimensional models of high-voltage transmission lines based on satellite imagery, tower type, and tower height, using a constructed parametric model library.

[0055] The above descriptions are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A method for reconstructing a three-dimensional model of a high-voltage transmission line based on satellite imagery, characterized in that, include: S1. Acquire satellite imagery, identify tower types, and obtain tower types; S2. Obtain geographic information data and, in conjunction with satellite imagery and tower type, perform tower height inversion to obtain the tower height; S3. Based on satellite imagery, tower type, and tower height, a three-dimensional model of the high-voltage transmission line is reconstructed using a constructed parametric model library.

2. The method for reconstructing a three-dimensional model of a high-voltage transmission line based on satellite imagery according to claim 1, characterized in that, S1 includes: S101. Locate the towers in the satellite imagery and obtain the tower area; S102. Extract the preset area shadow in the tower area; S103. Based on the extracted preset area shadow, the pole type is identified through the constructed type recognition module; the type recognition model is used to input the extracted preset area shadow and output the probability of the pole type; the pole type with the highest probability value is obtained as the recognition result.

3. The method for reconstructing a three-dimensional model of a high-voltage transmission line based on satellite imagery according to claim 1, characterized in that, The S2 includes: S201, extracting and constructing a multi-dimensional feature vector from satellite imagery, tower type and geographic information data, including: attribute features, geometric features and environmental features; S202. Based on the feature vector, the tower height is obtained by performing height inversion through the constructed regression model.

4. The method for reconstructing a three-dimensional model of a high-voltage transmission line based on satellite imagery according to claim 3, characterized in that, The regression model was constructed using the LightGBM framework. The regression model is constructed in a forward stepwise manner, and the prediction accuracy is improved by gradually fitting the residuals of the current model. The total number of iterations is , No. Wheel model output : in This represents the input feature vector. This is the output of the previous round of predictions. For the first The decision tree trained in the round, The learning rate is used to control the contribution of each tree to the final model. The LightGBM framework employs gradient-based one-sided sampling and mutually exclusive feature binding techniques during training. Its objective function ℒ is defined in each iteration as: in The loss function; This is a regularization term used to control model complexity; The number of training samples, For the first The true height of each sample To predict the height for the model.

5. The method for reconstructing a three-dimensional model of a high-voltage transmission line based on satellite imagery according to claim 3, characterized in that, The S3 includes: S301. Based on the attribute features and geometric features extracted from the satellite image, the model is located and oriented. According to the tower type, the corresponding tower reference model is retrieved from the constructed parametric model library, and the height of the reference model is geometrically adjusted according to the tower height. S302. Based on the preset conductor connection points in the insulator model corresponding to the benchmark model, automatically generate three-dimensional conductors between towers to obtain a three-dimensional model of the high-voltage transmission line.

6. The method for reconstructing a three-dimensional model of a high-voltage transmission line based on satellite imagery according to claim 5, characterized in that, The positioning includes: According to the latitude and longitude of the tower ground elevation Calculate its position in a three-dimensional Cartesian coordinate system. : in Let be the radius of curvature of the circle. Let the square of the eccentricity of the ellipsoid be . The semi-major axis of the reference ellipsoid.

7. The method for reconstructing a three-dimensional model of a high-voltage transmission line based on satellite imagery according to claim 5, characterized in that, The orientation includes: The orientation angle of the tower in the horizontal plane To determine spatial orientation, the tower's orientation angle in the horizontal plane. This is determined by its function in the line and its relationship with adjacent positions; the current tower The plane coordinates are The coordinates of its preceding and following adjacent towers are respectively , The orientation angle is calculated as follows: like For a straight-line tower, its crossarm should be perpendicular to the line connecting adjacent towers. like For corner towers, the long arm of the crossarm should point in the direction of the outer bisector of the line corner: in from Measurement is performed clockwise in the positive direction of the axis. , They are vectors exist and Components in direction.

8. The method for reconstructing a three-dimensional model of a high-voltage transmission line based on satellite imagery according to claim 6, characterized in that, The baseline model includes: the standard dimensions, geometric frame, crossarm layout and topological connection relationship of the corresponding tower model, and two types of insulator string models, tension type and straight type, are constructed according to the main morphology of the insulator under stress. The reference model has preset conductor connection points, and the number of insulator discs is adjusted according to the voltage level; The insulator string model serves as an auxiliary component, serving as a reference model alongside the tower model.

9. The method for reconstructing a three-dimensional model of a high-voltage transmission line based on satellite imagery according to claim 8, characterized in that, The geometric adjustment of the reference model height based on the tower height includes: Through the overall scaling factor Geometric adjustments are made to the baseline model, where The tower height is predicted based on a regression model. This is the original height of the reference model; the scaling operation is applied simultaneously to the tower body and insulator components.

10. A three-dimensional model reconstruction system for high-voltage transmission lines based on satellite imagery, characterized in that, Used to perform the method for reconstructing a three-dimensional model of a high-voltage transmission line based on satellite imagery as described in any one of claims 1-9.

Images