A spatial computing-based system, medium, and electronic equipment for identifying power grid facility damage.

By planning non-overlapping shooting angles before drone flight and combining a visual-point cloud fusion strategy, the problems of multiple tower obstruction and data fusion in post-disaster power grid facilities were solved, enabling rapid and accurate disaster damage identification and assessment, outputting disaster damage assessment reports, and supporting emergency repair decisions.

CN123071701APending Publication Date: 2026-11-10CHINA ELECTRIC POWER RESEARCH INSTITUTE CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611034797.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-13
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing drone inspection technology has several problems in identifying power grid facilities after disasters, including difficulty in separating overlapping towers, incomplete coverage of disaster damage types, reliance on pre-installed sensors for deformation detection, inability to transmit data due to post-disaster communication interruptions, and failure to effectively integrate visual and lidar data.

Method used

Based on spatial computing methods, the existing tower coordinates of the power grid company are used to proactively plan non-overlapping shooting angles before flight. Combined with a visual-point cloud fusion strategy, the sensor strategy is adaptively selected for data collection and analysis, realizing pixel-level segmentation and multi-dimensional feature extraction of the towers. Disaster damage is determined through visual-point cloud fusion.

Benefits of technology

It enables rapid and accurate disaster damage identification of power grid facilities after disasters within the edge computing unit of a drone, solves the problem of multiple tower obstruction, detects tower head detachment without the need for pre-disaster historical data, and establishes a visual-point cloud fusion disaster damage analysis architecture, improving identification efficiency and accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123071701A_ABST
    Figure CN123071701A_ABST
Patent Text Reader

Abstract

This invention discloses a method and system for identifying power grid facility damage based on spatial computing. The method includes: performing occlusion analysis on the target equipment based on its precise coordinates, and selecting a preset shooting angle in the unoccluded area; collecting target equipment data based on the preset shooting angle according to an adaptive selection strategy, and obtaining pixel-level target equipment segmentation regions for each target equipment by combining an edge instance segmentation model; extracting multi-dimensional feature data of the target equipment based on the target equipment segmentation regions from at least two shooting angles; the multi-dimensional feature data includes the three corners of the target equipment, one of which is extracted from two angles; and identifying the damage to the target equipment based on the multi-dimensional feature data through visual-point cloud fusion judgment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of power grid disaster emergency assessment technology, and more specifically, to a spatial computing-based system, medium, and electronic equipment for identifying power grid facility damage. Background Technology

[0002] Power grid facilities are highly vulnerable to structural damage during extreme disasters such as typhoons, floods, earthquakes, and fires. Rapidly and accurately identifying the location and extent of damage is crucial for post-disaster emergency repairs and resource allocation. Current power grid facility damage identification technologies have the following shortcomings: (1) Although UAV inspection has been widely used in daily power inspection, the existing solutions are mostly pre-programmed flight routes, lacking the ability to adapt autonomously to post-disaster scenarios and unable to perform intelligent fine-grained analysis of disaster-damaged areas. (2) In densely populated areas, multiple towers overlap in the drone footage. Existing visual methods are difficult to effectively separate the towers, making it impossible to conduct disaster damage analysis on individual towers. Existing solutions usually passively accept the overlap phenomenon and lack a proactive shooting strategy to avoid overlap. (3) In the existing UAV solutions equipped with lidar, lidar data and visual image data are usually processed independently, lacking a mechanism for adaptive switching and deep fusion based on target distance: when the tower is far away, the lidar point cloud is sparse and cannot be effectively utilized, and when the distance is close, the visual image is difficult to accurately obtain the three-dimensional angle relationship between the tower and the ground. The two sensors each have their own strengths but fail to work together to complement each other. Summary of the Invention

[0003] This invention proposes a method and system for identifying power grid facility damage based on spatial computing, aiming to solve the problems in rapid post-disaster assessment of power grid facility damage, such as difficulty in separating overlapping towers, incomplete coverage of damage types, reliance on pre-installed sensors for deformation detection, inability to transmit data due to post-disaster communication interruptions, and failure to effectively integrate visual and lidar data.

[0004] To address the aforementioned problems, according to one aspect of the present invention, a method for identifying power grid facility damage based on spatial computation is provided, the method comprising: Based on the precise coordinates of the target device, an occlusion analysis is performed on the target device, and a preset shooting angle is selected in the unoccluded area; According to the adaptive selection strategy, target device data is collected based on the preset shooting angle, and combined with the edge instance segmentation model, pixel-level target device segmentation region of each target device is obtained; Based on the target device segmentation region from at least two shooting angles, multidimensional feature data of the target device is extracted; the multidimensional feature data includes the three corners of the target device, one of which is extracted from both angles. Based on the multidimensional feature data, the damage to the target equipment is identified through visual-point cloud fusion.

[0005] Preferably, the step of performing occlusion analysis on the target device based on its precise coordinates and selecting a preset shooting angle within the unoccluded area includes: Map the precise coordinates of the target device to the actual geographic space; Using the coordinates of the current target device after mapping as the origin, calculate the azimuth and distance of adjacent devices relative to the current target device; Each adjacent device is projected onto a 360° azimuth circle centered on the current target device according to its corresponding azimuth angle and distance, and each adjacent device forms a corresponding fan-shaped occlusion zone. Find the sector-shaped occlusion intervals corresponding to all adjacent devices of the current target device, and superimpose the sector-shaped occlusion intervals of all adjacent devices of the current target device to obtain the complete occlusion azimuth interval of the current target device; Within the remaining unobstructed azimuth range, the direction in which the angle between two planar azimuth angles differs by a preset angle is selected as the preset shooting angle.

[0006] Preferably, the adaptive selection strategy includes: Long-range layer: When the LiDAR cannot effectively cover the target device, it automatically switches to pure vision mode and uses only the image data of the gimbal camera to obtain the pixel-level segmentation area of ​​the target device; Mid-range level: When the LiDAR acquisition range can completely cover the main body of the target device, the vision-point cloud deep fusion strategy is enabled; the LiDAR point cloud data assists the visual segmentation model to determine the pixel-level segmentation area of ​​the target device; Close-range local layer: When the lidar can only cover a local area of ​​the target device, the lidar scanning angle is actively adjusted to obtain the pixel-level segmentation area of ​​the target device.

[0007] Preferably, the extraction of multidimensional feature data of the target device includes: For a segmented region of a target device corresponding to at least two shooting angles, the multidimensional feature data includes three corners of the target device, one of which is extracted from both angles; morphological thinning algorithms are used to extract skeleton lines; and the outer contour skeletal curves representing the main structure of the target device are selected from the skeleton lines. If the lidar completely covers the main body of the target device, the surface features and three-dimensional contour curves of the main body of the target device can be extracted from the multi-view lidar point cloud data to obtain the three-dimensional deformation information of each facade of the target device. If the lidar only covers a local area of ​​the target equipment, collect point cloud data of the area where the target equipment tower foot meets the ground, and extract the target equipment tower foot point cloud data and the ground point cloud data respectively; calculate the angle between the main body direction of the target equipment and the ground normal vector through the tower foot point cloud data and the ground point cloud data to generate the tilt geometric feature data of the target equipment. The multidimensional feature data and point cloud data are visually fused to obtain fused three-dimensional geometric data; the backbone curve of the target device is extracted from the fused three-dimensional geometric data. The four visual backbone curves corresponding to the two shooting angles are mapped to the same three-dimensional reference coordinate system through projection transformation, and the multi-angle backbone curves are normalized.

[0008] Preferably, the step of identifying the damage to the target device based on the multi-dimensional feature data through visual-point cloud fusion includes: The tower head endpoint height of the normalized target equipment backbone curve is compared laterally with the tower head endpoint height of adjacent similar equipment on the same line. When the height difference exceeds the preset tower head detachment threshold, the target equipment is determined to have experienced a tower head detachment fault.

[0009] Preferably, the identification of damage to the target device includes: The overall trend of the four normalized backbone curves is used to determine whether the target equipment is tilted; if there is a consistent deflection angle in the relative vertical direction, the tilt direction of the target equipment is initially determined. When the point cloud data completely covers the main body of the target equipment, cross-validation is performed by combining the main body surface features in the three-dimensional deformation information of each facade of the target equipment; when the point cloud data of the target equipment tower foot and the ground point cloud data are obtained, the tilt geometric feature data of the target equipment is used as a precise tilt measurement, and compared with the preset tilt tolerance threshold, the severity of the tilt of the target equipment is determined.

[0010] Preferably, the identification of damage to the target device includes: Under normal operating conditions, determine whether the structural morphology of the four normalized backbone curves remains consistent; if any of the following abnormalities occur in the backbone curves: abrupt change in local curvature, discontinuous curve breakage, or significant deviation between the direction and morphology of a single curve and the other curves, determine that asymmetric deformation or local fracture has occurred at the corresponding location of the target equipment; when point cloud data covers the fracture area, use the three-dimensional deformation information of each facade of the target equipment to verify the fracture location and quantify the severity of the fracture deformation of the target equipment.

[0011] According to another aspect of the present invention, the present invention provides a power grid facility disaster identification system based on spatial computing, the system comprising: The selection unit is used to perform occlusion analysis on the target device based on the precise coordinates of the target device, and select a preset shooting angle in the unoccluded area; The segmentation unit is used to collect target device data based on the preset shooting angle according to the adaptive selection strategy, and combine it with the edge instance segmentation model to obtain the pixel-level target device segmentation region for each target device; An extraction unit is used to segment the target device region based on at least two shooting angles and extract multidimensional feature data of the target device. The identification unit is used to identify the damage to the target device based on the multidimensional feature data and through visual-point cloud fusion.

[0012] Preferably, the selection unit is further configured to: Map the precise coordinates of the target device to the actual geographic space; Using the coordinates of the current target device after mapping as the origin, calculate the azimuth and distance of adjacent devices relative to the current target device; Each adjacent device is projected onto a 360° azimuth circle centered on the current target device according to its corresponding azimuth solution and distance, and each adjacent device forms a corresponding fan-shaped occlusion zone; Find the sector-shaped occlusion intervals corresponding to all adjacent devices of the current target device, and superimpose the sector-shaped occlusion intervals of all adjacent devices of the current target device to obtain the complete occlusion azimuth interval of the current target device; Within the remaining unobstructed azimuth range, the direction in which the angle between two planar azimuth angles differs by a preset angle is selected as the preset shooting angle.

[0013] Preferably, the adaptive selection strategy includes: Long-range layer: When the LiDAR cannot effectively cover the target device, it automatically switches to pure vision mode and uses only the image data of the gimbal camera to obtain the pixel-level segmentation area of ​​the target device; Mid-range level: When the LiDAR acquisition range can completely cover the main body of the target device, the vision-point cloud deep fusion strategy is enabled; the LiDAR point cloud data assists the visual segmentation model to determine the pixel-level segmentation area of ​​the target device; Close-range local layer: When the lidar can only cover a local area of ​​the target device, the lidar scanning angle is actively adjusted to obtain the pixel-level segmentation area of ​​the target device.

[0014] Preferably, the extraction unit is further configured to: For a segmented region of a target device corresponding to at least two shooting angles, the multidimensional feature data includes three corners of the target device, one of which is extracted from both angles; morphological thinning algorithms are used to extract skeleton lines; and the outer contour skeletal curves representing the main structure of the target device are selected from the skeleton lines. If the lidar completely covers the main body of the target device, the surface features and three-dimensional contour curves of the main body of the target device can be extracted from the multi-view lidar point cloud data to obtain the three-dimensional deformation information of each facade of the target device. If the lidar only covers a local area of ​​the target equipment, collect point cloud data of the area where the target equipment tower foot meets the ground, and extract the target equipment tower foot point cloud data and the ground point cloud data respectively; calculate the angle between the main body direction of the target equipment and the ground normal vector through the tower foot point cloud data and the ground point cloud data to generate the tilt geometric feature data of the target equipment. The multidimensional feature data and point cloud data are visually fused to obtain fused three-dimensional geometric data; the backbone curve of the target device is extracted from the fused three-dimensional geometric data. The four visual backbone curves corresponding to the two shooting angles are mapped to the same three-dimensional reference coordinate system through projection transformation, and the multi-angle backbone curves are normalized.

[0015] Preferably, the identification unit is further configured to: The tower head endpoint height of the normalized target equipment backbone curve is compared laterally with the tower head endpoint height of adjacent similar equipment on the same line. When the height difference exceeds the preset tower head detachment threshold, the target equipment is determined to have experienced a tower head detachment fault.

[0016] Preferably, the identification unit is further configured to: The overall trend of the four normalized backbone curves is used to determine whether the target equipment is tilted; if there is a consistent deflection angle in the relative vertical direction, the tilt direction of the target equipment is initially determined. When the point cloud data completely covers the main body of the target equipment, cross-validation is performed by combining the main body surface features in the three-dimensional deformation information of each facade of the target equipment; when the point cloud data of the target equipment tower foot and the ground point cloud data are obtained, the tilt geometric feature data of the target equipment is used as a precise tilt measurement, and compared with the preset tilt tolerance threshold, the severity of the tilt of the target equipment is determined.

[0017] Preferably, the identification unit is further configured to: Under normal operating conditions, determine whether the structural morphology of the four normalized backbone curves remains consistent; if any of the following abnormalities occur in the backbone curves: abrupt change in local curvature, discontinuous curve breakage, or significant deviation between the direction and morphology of a single curve and the other curves, determine that asymmetric deformation or local fracture has occurred at the corresponding location of the target equipment; when point cloud data covers the fracture area, use the three-dimensional deformation information of each facade of the target equipment to verify the fracture location and quantify the severity of the fracture deformation of the target equipment.

[0018] In another aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a method for identifying power grid facility damage based on spatial computing.

[0019] According to another aspect of the present invention, the present invention provides an electronic device, comprising: The aforementioned computer-readable storage medium; and One or more processors for executing a program in the computer-readable storage medium.

[0020] This invention provides a method and system for identifying power grid facility damage based on spatial computing. The method includes: performing occlusion analysis on target equipment based on its precise coordinates, and selecting a preset shooting angle within the unoccluded area; collecting target equipment data based on the preset shooting angle using an adaptive selection strategy, and obtaining pixel-level target equipment segmentation regions for each target equipment by combining an edge instance segmentation model; extracting multi-dimensional feature data of the target equipment based on the segmented regions from at least two shooting angles; the multi-dimensional feature data includes the three corners of the target equipment, with one corner being extracted from two angles; and identifying the damage to the target equipment through visual-point cloud fusion based on the multi-dimensional feature data. This invention utilizes existing tower coordinate records to proactively plan non-overlapping shooting angles for each tower before flight, fundamentally solving the multi-tower occlusion problem; it introduces a lateral comparison mechanism for towers on the same line, enabling detection of tower head detachment without pre-disaster historical data; and it establishes a visual-point cloud fusion disaster analysis architecture, adaptively selecting sensor strategies and performing deep fusion analysis based on the target tower distance. This invention is fully automated within the edge computing unit of an unmanned aerial vehicle (UAV) and does not rely on public network communication. Attached Figure Description

[0021] Exemplary embodiments of the present invention can be more fully understood by referring to the following figures: Figure 1 This is a flowchart of a power grid facility disaster identification method based on spatial computing according to an embodiment of the present invention; Figure 2 This is a flowchart of a power grid facility disaster identification method based on spatial computing according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a three-level vision-point cloud fusion architecture according to an embodiment of the present invention; and Figure 4 This is a structural diagram of a power grid facility damage identification system based on spatial computing according to an embodiment of the present invention. Detailed Implementation

[0022] Exemplary embodiments of the invention will now be described with reference to the accompanying drawings. However, the invention may be embodied in many different forms and is not limited to the embodiments described herein. These embodiments are provided to fully and completely disclose the invention and to fully convey its scope to those skilled in the art. The terminology used in the exemplary embodiments illustrated in the drawings is not intended to limit the invention. In the drawings, the same units / elements are referred to by the same reference numerals.

[0023] Unless otherwise stated, the terms used herein (including technical terms) have their common meaning as understood by one of ordinary skill in the art. Furthermore, it is understood that terms defined in commonly used dictionaries should be understood to have a meaning consistent with the context of their relevant field, and not to be interpreted as having an idealized or overly formal meaning.

[0024] Figure 1 This is a flowchart of a power grid facility disaster identification method based on spatial computing according to an embodiment of the present invention.

[0025] To address the challenges in rapid post-disaster damage assessment of power grid facilities, such as the difficulty in separating overlapping towers, incomplete coverage of damage types, reliance on pre-installed sensors for deformation detection leading to data loss due to post-disaster communication interruptions, and ineffective fusion of visual and lidar data, this invention provides a spatial computing-based method for identifying and locating power grid facility damage. This invention utilizes existing tower coordinate records from power grid companies to proactively plan non-overlapping shooting angles for each tower before flight, fundamentally resolving the issue of multiple tower obstruction. It introduces a lateral comparison mechanism for towers along the same line, enabling the detection of tower head detachment without requiring pre-disaster historical data. A visual-point cloud fusion damage analysis architecture is established, adaptively selecting sensor strategies based on the distance to the target tower and performing deep fusion analysis. The entire method is fully automated within the edge computing unit of an unmanned aerial vehicle (UAV) and does not rely on public network communication.

[0026] This invention provides proactive planning of non-overlapping shooting angles for pole coordinates: Utilizing the power grid company's existing precise pole coordinate ledger, before drone flight, the azimuth angle and distance of adjacent poles are calculated with each pole as the origin. The adjacent poles are projected onto a 360° azimuth circle, marking the obstructed azimuth intervals. Within the unobstructed intervals, two directions with a planar angle difference of approximately 90° are selected as shooting angles. During drone flight, the gimbal orientation is proactively adjusted according to the pre-calculated angles to ensure that each pole is independently photographed from two non-overlapping perspectives. This invention provides a vision-point cloud fusion disaster damage analysis architecture: Sensor strategies are adaptively selected based on the distance between the target tower and the UAV. When the tower is far away and the LiDAR cannot effectively cover it, the system automatically switches to pure vision mode for image acquisition and segmentation analysis. When the tower is close and the LiDAR's acquisition range can cover the tower body, point cloud assistance is enabled—using the depth information of the LiDAR point cloud to assist the visual segmentation model in distinguishing the target tower from the background tower, while simultaneously identifying damage based on the tower body's surface and curves at different angles. When the LiDAR can only cover a portion of the tower, the LiDAR angle is adjusted to focus on acquiring the area where the tower meets the ground. The point cloud is used to accurately calculate the angular relationship between the tower and the ground, and then fused with the visually extracted backbone curves for analysis. This invention provides a method for detecting tower head detachment without prior knowledge by comparing towers along the same line laterally: using adjacent towers along the same line as a reference, the tower head height of the target tower is extracted and compared laterally with adjacent towers of the same type along the same line. If the tower head height of the target tower is significantly lower than that of the target tower, it is determined that tower head detachment has occurred. This method does not require pre-disaster tower height data; it only requires a lateral comparison between towers along the same line to detect anomalies. This invention provides a comprehensive disaster damage assessment method that combines visual backbone curves and point cloud surfaces: Based on the backbone curves extracted from visual images, when lidar data is available, the main surface features of the tower obtained from the point cloud and the angle relationship between the tower and the ground junction are fused to achieve a comprehensive assessment of three types of disaster damage: tower tilt, breakage, and tower head detachment, and output a disaster damage assessment report.

[0027] like Figure 1 As shown, this invention provides a method for identifying power grid facility damage based on spatial computing. The method includes: Step 101: Perform occlusion analysis on the target device based on its precise coordinates, and select a preset shooting angle in the unoccluded area; Preferably, based on the precise coordinates of the target device, an occlusion analysis is performed on the target device, and a preset shooting angle is selected in the unoccluded area, including: Map the precise coordinates of the target device to the actual geographic space; Using the coordinates of the current target device after mapping as the origin, calculate the azimuth and distance of adjacent devices relative to the current target device; Each adjacent device is projected onto a 360° azimuth circle centered on the current target device according to its corresponding azimuth angle and distance. Each adjacent device forms a fan-shaped occlusion zone. The width of the fan-shaped occlusion zone is determined by the physical size of the adjacent devices and the distance between the adjacent devices and the current target device. Find the sector-shaped occlusion intervals corresponding to all adjacent devices of the current target device, and superimpose the sector-shaped occlusion intervals of all adjacent devices of the current target device to obtain the complete occlusion azimuth interval of the current target device; Within the remaining unobstructed azimuth range, the direction in which the angle between two planar azimuth angles differs by a preset angle is selected as the preset shooting angle.

[0028] This invention provides geographic registration and non-overlapping shooting angle planning based on tower register coordinates.

[0029] (1) Load the existing accurate coordinate ledger of power grid company towers (including tower number, latitude and longitude coordinates, tower type and design height, etc.) into the local edge computing unit of UAV.

[0030] (2) By matching the features of the ground top view taken by the UAV with the high-resolution satellite image, and by integrating the GPS / IMU real-time positioning and attitude data, pixel-level high-precision georegistration is achieved, and the coordinates of the ledger are mapped to the actual geographic space.

[0031] (3) Active planning of non-overlapping shooting angles for each tower: Using the current tower's coordinates in the ledger as the origin, calculate the azimuth and distance of all adjacent towers relative to the current tower; project each adjacent tower onto a 360° azimuth circle centered on the current tower according to its azimuth and distance, with each adjacent tower occupying a fan-shaped obstruction zone; take the union of all obstruction zones of adjacent towers to obtain all obstructed azimuth zones; in the unobstructed azimuth zones, select two directions with a plane angle difference of approximately 90° as the shooting angle of that tower. This planning is entirely based on the ledger coordinates and can be completed offline before flight.

[0032] Step 102: Based on the adaptive selection strategy, collect target device data based on the preset shooting angle, and combine it with the edge instance segmentation model to obtain the pixel-level target device segmentation region for each target device; Preferably, the adaptive selection strategy includes: Long-range layer: When the LiDAR cannot effectively cover the target device, it automatically switches to pure vision mode and uses only the image data of the gimbal camera to obtain the pixel-level segmentation area of ​​the target device; Mid-range level: When the LiDAR acquisition range can completely cover the main body of the target device, the vision-point cloud deep fusion strategy is enabled; the LiDAR point cloud data assists the visual segmentation model to determine the pixel-level segmentation area of ​​the target device; Close-range local layer: When the lidar can only cover a local area of ​​the target device, the lidar scanning angle is actively adjusted to obtain the pixel-level segmentation area of ​​the target device.

[0033] This invention provides image segmentation technology. Image segmentation is a fundamental technology in the field of computer vision, aiming to separate different target objects in an image from the background and assign independent pixel-level regions to each target. The image segmentation technology involved in this invention mainly includes the following two aspects: Instance segmentation: Unlike semantic segmentation, which only distinguishes between "foreground / background," instance segmentation requires distinguishing each individual pixel within the same category. For example, in an image containing multiple towers, instance segmentation not only identifies "these pixels belong to the towers," but also distinguishes "pixel region A belongs to tower #023, and pixel region B belongs to tower #024." This invention deploys a lightweight instance segmentation model on the edge computing unit of a UAV to achieve pixel-level accurate separation of tower targets.

[0034] This invention provides pole instance segmentation and point cloud assistance based on edge deployment.

[0035] (1) Deploy the lightweight image segmentation model on the local edge computing unit of the UAV. The UAV flies to the vicinity of each tower according to the non-overlapping shooting angle planned in step one, and the gimbal camera automatically turns to the pre-calculated angle to shoot.

[0036] (2) Determine the acquisition range status of the lidar based on the real-time distance between the UAV and the target tower: If the towers are far apart and the lidar cannot effectively cover them, switch to pure vision mode and use only image data for segmentation. If the LiDAR acquisition range can cover the main body of the tower, then point cloud-assisted segmentation is enabled—using the three-dimensional spatial position information of the LiDAR point cloud to assist the visual segmentation model in distinguishing the target tower from other towers in the background, thereby improving the segmentation accuracy in dense scenes. If the lidar can only cover a portion of the tower, adjust the lidar scanning angle to focus on collecting data at the junction of the tower and the ground.

[0037] (3) The edge segmentation model segments each image into tower instances, combining ledger location information and point cloud auxiliary information to extract independent pixel-level segmentation regions for each tower. The entire segmentation process is completed locally on the UAV, without relying on public network communication. Figure 2 As shown.

[0038] Step 103: Segment the target device region based on at least two shooting angles and extract multidimensional feature data of the target device; Preferably, the multidimensional feature data of the target device is extracted, including: For the segmented regions of the target device corresponding to at least two shooting angles, the multidimensional feature data includes the three corners of the target device, one of which is extracted from both angles; morphological thinning algorithms are used to extract skeleton lines; and the outer contour skeletal curves representing the main structure of the target device are selected from the skeleton lines. If the lidar completely covers the main body of the target device, the surface features and three-dimensional contour curves of the main body of the target device can be extracted from the multi-view lidar point cloud data to obtain the three-dimensional deformation information of each facade of the target device. If the lidar only covers a local area of ​​the target equipment, collect point cloud data of the area where the target equipment tower foot meets the ground, and extract the target equipment tower foot point cloud data and the ground point cloud data respectively; calculate the angle between the main body direction of the target equipment and the ground normal vector through the tower foot point cloud data and the ground point cloud data to generate the tilt geometric feature data of the target equipment. Visual-point cloud fusion is performed between multidimensional feature data and point cloud data to obtain fused 3D geometric data; the backbone curve of the target device is extracted from the fused 3D geometric data. The four visual backbone curves corresponding to the two shooting angles are mapped to the same three-dimensional reference coordinate system through projection transformation, and the multi-angle backbone curves are normalized.

[0039] This invention provides a skeleton extraction method: after obtaining the segmented regions of a tower instance, the skeleton lines of that region are extracted using a morphological thinning algorithm. The skeleton lines are one-dimensional centerline representations of the target shape, preserving the target's topological structure and main extension direction. From the skeleton lines, the outermost continuous curves extending from the tower legs to the tower head—i.e., the peripheral backbone curves—are further filtered out. Crossarms and other auxiliary structures are excluded during the skeleton extraction process, ensuring that the backbone curves only reflect the structural orientation of the tower's main body.

[0040] This invention provides point cloud recognition and segmentation technology. A point cloud is a collection of three-dimensional spatial data points collected by sensors such as LiDAR. Each point contains its three-dimensional coordinates (x, y, z) in space, and some point clouds also contain information such as reflection intensity. The point cloud processing technology involved in this invention mainly includes the following three aspects: (1) Point cloud target recognition and segmentation: Identify and separate point cloud clusters belonging to poles from the original point cloud data. Since point clouds naturally contain three-dimensional spatial location information, they can help determine the spatial relationship between different targets in the scene. For example, the three-dimensional spatial location information of point clouds can help the visual segmentation model distinguish the target pole from other poles in the background, thereby improving the segmentation accuracy in dense scenes.

[0041] (2) Point cloud feature extraction: Structural features are extracted from the point cloud of the tower, including the surface features of each face of the tower body and the three-dimensional contour curve. These features can be used for cross-validation with the visual backbone curve to provide accurate three-dimensional information on local deformation.

[0042] This invention provides visual backbone curve extraction and point cloud feature acquisition.

[0043] (1) Visual skeleton curve extraction: Skeleton lines were extracted from the segmented regions of each tower at two shooting angles using a morphological thinning algorithm; two outermost continuous curves were selected from the skeleton lines—extending from the tower leg to the tower head—representing the outer contour skeleton of the tower from that perspective, while auxiliary structures such as crossarms were excluded. Two skeleton curves were obtained for each angle, for a total of four curves from both angles. The tower head endpoint height of each skeleton curve was also extracted.

[0044] (2) Point cloud feature acquisition (when LiDAR is available): If the lidar completely covers the main body of the tower: extract the curved features and three-dimensional contour curves of the main body of the tower from the point cloud data collected from different angles, and obtain the three-dimensional deformation information of each surface of the tower; If the lidar only covers a local area of ​​the tower: extract the point cloud data of the tower base and the ground from the point cloud data of the tower-ground junction area, and calculate the angle between the main direction of the tower and the ground normal vector. For example... Figure 3 As shown.

[0045] Step 104: Based on multi-dimensional feature data, identify the damage to the target equipment through visual-point cloud fusion.

[0046] Preferably, based on multi-dimensional feature data, visual-point cloud fusion is used to identify damage to target equipment, including determination of tower head detachment. The tower head endpoint height of the normalized target equipment backbone curve is compared laterally with the tower head endpoint height of adjacent similar equipment on the same line. When the height difference exceeds the preset tower head detachment threshold, the target equipment is determined to have experienced a tower head detachment fault.

[0047] Preferably, based on multi-dimensional feature data, visual-point cloud fusion is used to identify damage to target equipment, including equipment tilt fusion determination. The overall trend of the four normalized backbone curves is used to determine whether the target equipment is tilted; if there is a consistent deflection angle in the relative vertical direction, the tilt direction of the target equipment is initially determined. When the point cloud data completely covers the main body of the target equipment, cross-validation is performed by combining the main body surface features in the three-dimensional deformation information of each facade of the target equipment; when the point cloud data of the target equipment tower foot and the ground point cloud data are obtained, the tilt geometric feature data of the target equipment is used as a precise tilt measurement, and compared with the preset tilt tolerance threshold, the severity of the tilt of the target equipment is determined.

[0048] Preferably, based on multi-dimensional feature data, visual-point cloud fusion is used to identify damage to target equipment, including equipment breakage fusion determination. Under normal operating conditions, determine whether the structural morphology of the four normalized backbone curves remains consistent; if any of the following abnormalities occur in the backbone curves: abrupt change in local curvature, discontinuous curve breakage, or significant deviation between the direction and morphology of a single curve and the other curves, determine that asymmetric deformation or local fracture has occurred at the corresponding location of the target equipment; when point cloud data covers the fracture area, use the three-dimensional deformation information of each facade of the target equipment to verify the fracture location and quantify the severity of the fracture deformation of the target equipment.

[0049] This invention provides precise local point cloud measurement: when the point cloud only covers a local area of ​​the tower, the LiDAR scanning angle can be adjusted to focus on the junction area between the tower and the ground. By collecting point cloud data of the contact area between the tower feet and the ground, the angle between the tower's main direction and the ground normal vector can be accurately calculated, providing the most direct geometric measure of the tower's tilt.

[0050] This invention provides a comprehensive disaster damage assessment based on visual-point cloud fusion.

[0051] (1) The four visual backbone curves from two angles are mapped to the same three-dimensional reference coordinate system through projection transformation.

[0052] (2) Tilt determination: Calculate the overall orientation of the four backbone curves in the projection space. If there is a consistent deflection angle with the vertical direction, the tilt direction is initially determined. When point cloud data is available, the information of the main surface of the tower obtained by fusing the point cloud is cross-validated. When the point cloud of the tower-ground junction point is available, the angle between the main direction of the tower and the ground normal vector is accurately calculated using the point cloud. This angle is used as a measure of the accurate angular tilt relationship between the tower and the ground to determine whether the degree of tilt exceeds the allowable range.

[0053] (3) Fracture Judgment: Under normal circumstances, the four backbone curves should maintain structural consistency in the projection space. Fracture is determined if any of the following abnormal features are present: obvious bending (abrupt curvature) occurs at a certain point in the backbone curve; discontinuity (break) occurs in the backbone curve; obvious inconsistency occurs between the four backbone curves (such as a significant deviation in direction or shape between one curve and the other three curves), indicating that the tower structure has undergone asymmetrical deformation or local fracture at that location. When the point cloud data covers the fracture area, the three-dimensional depth information of the point cloud is used to verify the fracture location and extent.

[0054] (4) Determination of Tower Head Detachment – ​​Based on Lateral Comparison along the Same Line: Extract the height of the tower head endpoint of the target tower and compare it with the height of the tower head endpoint of adjacent towers of the same type along the same line. The design height of towers along the same line is consistent, and under normal circumstances, the tower head heights should be approximately equal. If the height of the tower head endpoint of the target tower is significantly lower than that of the adjacent tower (the difference exceeds the preset threshold), then it is determined that tower head detachment has occurred.

[0055] (5) Based on the combined visual backbone curve analysis results and point cloud feature analysis results, output the damage type (no damage / tilt / breakage / tower head detachment) and severity parameters of the tower.

[0056] This invention provides a disaster damage assessment report output.

[0057] This invention summarizes the disaster damage assessment results of all poles and outputs a disaster damage assessment report, including: a pole and tower statistical list (number, location, disaster damage type, severity, and data source labeling), a damaged pole and tower ranking table (by severity), and a disaster damage situation map (labeling the location and disaster damage type of each damaged pole and tower), supporting decisions on emergency repair priorities and resource allocation.

[0058] The following are illustrative examples of embodiments of the present invention.

[0059] Taking a typhoon disaster as an example, after the typhoon, many power transmission towers along a certain area tilted, broke, and had their tower heads detached. The disaster caused widespread damage to communication base stations, resulting in a public network outage. The power grid company possessed a precise coordinate register of all the towers in the area. The drone, equipped with a high-definition gimbal camera, LiDAR, and edge computing units, executed the entire operation autonomously.

[0060] Phase 1: Geographic Registration and Non-Overlapping Shooting Angle Planning. The UAV loads the tower coordinates from the existing database onto the local machine. During flight, it captures ground-view video, performs feature matching with pre-disaster satellite imagery, and fuses GPS / IMU data to achieve geographic registration. The system performs overlap analysis on each tower—for example, tower #023's adjacent towers #022 and #024 are located 120 meters northeast (35°) and 130 meters southwest (210°), respectively. Using #023 as the origin, the system projects #022 and #024 onto a 360° azimuth circle, selecting two 90° apart directions (165° and 255°) within the unobstructed area as shooting angles. Angle planning is completed for all 20 towers one by one.

[0061] Phase Two: Distance Judgment and Adaptive Sensor Selection. The drone flies along the route, calculating the distance to each tower in real time. Tower #010, 280 meters away (long distance): The lidar point cloud is too sparse at this distance, so it automatically switches to pure visual mode and only uses the gimbal camera to take pictures at the planned angle. Tower #023, 120 meters away (medium distance): The LiDAR's acquisition range can completely cover the main body of the tower, enabling the visual-point cloud deep fusion mode. The gimbal camera captures images at two angles, 165° and 255°, while the LiDAR simultaneously acquires the point cloud of the main body of the tower, extracting the surface features of each facet of the tower; Tower #031, 60 meters away (close range): The lidar can only cover the lower part of the tower. The system automatically adjusts the lidar scanning angle to focus on the area where the tower foot meets the ground, accurately collecting point clouds of the tower foot and the ground.

[0062] Phase 3: Visual Segmentation and Backbone Curve Extraction. Instance segmentation and backbone curve extraction are performed on the images of all towers. Taking #023 as an example, in images taken at 165° and 255° angles, the 3D position information of the LiDAR point cloud assists the segmentation model in eliminating interference from other towers in the background, extracting the precise pixel-level region of #023. Two peripheral backbone curves are extracted from the segmented region at each angle, for a total of four curves across both angles, and the height of the tower head endpoint is recorded.

[0063] Phase 4: Visual-point cloud fusion disaster assessment.

[0064] Tower #010 (purely visual): The overall direction of the four main curves deviates from the vertical direction by 1.9°, which is determined to be an overall tilt of 1.9%.

[0065] Tower #023 (visual-point cloud deep fusion): The visual backbone curve shows an overall deviation of 2.1°; the main curved surface features of the tower extracted from the point cloud show an abnormal bend in the middle of the tower body, consistent with the bending position of the visual backbone curve, indicating a break, with the break location approximately 22 meters above the ground. The point cloud data further confirmed the three-dimensional morphology of the break surface.

[0066] Tower #031 (point cloud focusing on ground junction): The visual backbone curve shows a slight tilt. The point cloud computing results of the tower foot-ground junction point accurately collected by the lidar show that the angle between the main direction of the tower and the ground normal vector is 2.4°, which exceeds the maximum allowable tilt of 1.5%, and is judged to be an overall tilt of 2.4% (based on the accurate measurement value of the point cloud).

[0067] Tower #045 (visual comparison only + horizontal comparison): The height of the tower head is 32.5 meters. Adjacent towers #044 and #046 on the same line are of the same model, with tower head heights of 41.2 meters and 40.8 meters respectively. The difference of approximately 8.5 meters exceeds the preset threshold, indicating a tower head detachment.

[0068] The remaining 16 towers are normal.

[0069] Final output. After the drone returns, the analysis results are exported, generating a disaster damage assessment report: Of the 20 poles, 2 are tilted (#010, 1.9%; #031, 2.4%), 1 is broken (#023, at 22 meters), 1 has a detached tower top (#045), and 16 are intact. The data source for each pole is marked (pure vision / vision-point cloud fusion / precise point cloud measurement), which the emergency command center uses to formulate repair priorities and resource allocation plans.

[0070] This invention provides a method for identifying and locating power grid facility damage based on spatial computing. The main process includes: (1) Load the existing accurate coordinate ledger of the power grid company onto the local of the UAV, and map the ledger coordinates to the actual geographic space through georeferencing; based on the ledger coordinates, actively plan the non-overlapping shooting angle for each tower: take the current tower coordinates as the origin, calculate the azimuth and distance of adjacent towers, project the adjacent towers onto the 360° azimuth circle to mark the occluded azimuth interval, and select two directions with a plane angle difference of about 90° in the unoccluded interval as the shooting angle; (2) The UAV flies to the vicinity of each pole at the planned angle and adaptively selects the sensor strategy according to the real-time distance between the target pole and the UAV: ​​at long distance, it automatically switches to pure vision mode; at medium distance, it enables LiDAR point cloud-assisted visual segmentation; at close distance, it adjusts the LiDAR focus on the pole-ground junction area; and the pole instance is segmented by a lightweight image segmentation model deployed on the edge computing unit to extract the independent pixel-level area of ​​each pole. (3) Extract skeleton lines from the segmented regions of the two shooting angle images of each tower, filter out the outer skeleton curves from the tower leg to the tower head, exclude auxiliary structures such as crossarms, and extract two skeleton curves for each angle, for a total of four skeleton curves for the two angles; when lidar data is available, extract point cloud features simultaneously - including the main surface features of the tower and the angle data of the tower-ground junction point; (4) Fusion analysis of visual backbone curves and point cloud features: The four backbone curves are mapped to the same three-dimensional reference coordinate system through projection transformation. If the overall direction has a consistent deflection angle with the vertical direction, it is determined to be tilted. The pole-ground angle of the point cloud is used as a measure of the precise angle tilt relationship between the pole and the ground to determine whether the tilt degree exceeds the allowable range. Under normal circumstances, the four backbone curves should maintain a consistent structure in the projection space. If there is a bend, discontinuity or obvious inconsistency between the four curves, it is determined to be broken. The three-dimensional information of the point cloud is used to verify the breakage location. The height of the tower head end point of the target pole is extracted and compared laterally with the height of the tower head end point of the adjacent similar poles on the same line. If the height of the tower head of the target pole is significantly lower, it is determined to be that the tower head has fallen off. (5) Output a disaster loss assessment report, including a list of tower statistics, disaster loss type, severity, data source labeling, and disaster loss situation map.

[0071] The non-overlapping shooting angle planning method in this invention specifically includes: using the current tower coordinates as the origin, calculating the azimuth and distance of all adjacent towers relative to the current tower; projecting each adjacent tower onto a 360° azimuth circle centered on the current tower according to its azimuth and distance, with each adjacent tower occupying a fan-shaped obstruction zone, the width of which is determined by the size and distance of the adjacent towers; taking the union of all obstruction zones of adjacent towers to obtain all obstructed azimuth zones; and selecting two directions with a plane angle difference of approximately 90° as shooting angles within the unobstructed azimuth zones. This planning is entirely based on the tower coordinates and can be completed offline before flight and executed directly during flight.

[0072] The visual-point cloud adaptive fusion strategy in this invention includes three levels: At long distances, when lidar cannot effectively cover the tower, it automatically switches to pure vision mode, using only image data from the gimbal camera for segmentation and analysis. At the mid-range level, when the LiDAR acquisition range can completely cover the main body of the tower (i.e., the point cloud data can completely identify the distance of the tower's curved surfaces at different angles), visual-point cloud deep fusion is enabled: the three-dimensional spatial position information of the LiDAR point cloud assists the visual segmentation model in distinguishing the target tower from the background tower, while extracting the main body curved surface features of the tower from the point cloud and cross-validating them with the visual backbone curve. For close-range / local-level applications, when the lidar can only cover a portion of the tower, adjust the lidar scanning angle to focus on collecting data at the junction of the tower's base and the ground. Use point cloud data to accurately calculate the angle between the tower's main direction and the ground normal vector.

[0073] The tower head detachment determination method in this invention utilizes the inherent characteristic of consistent design heights among towers on the same line: It extracts the tower head endpoint height of the target tower's backbone curve and compares it laterally with the tower head endpoint heights of adjacent towers of the same type on the same line. Since adjacent towers on the same line are usually of the same tower type and design height, their tower head heights should normally be approximately equal. If the difference between the tower head endpoint height of the target tower and the adjacent tower's tower head endpoint height exceeds a preset threshold, then tower head detachment is determined to have occurred. This method requires no pre-disaster historical data or standard tower models; it relies solely on lateral comparisons between towers on the same line to complete a priori detection.

[0074] The tilt determination method described in this invention integrates visual and point cloud data: the overall direction of the visual backbone curve is used as the initial tilt direction determination; when point cloud data is available, the angle between the main body direction of the tower and the ground normal vector is accurately calculated using the point cloud data of the tower-ground combination collected by the lidar, which is used to determine whether the precise angular tilt relationship between the tower and the ground exceeds the allowable range.

[0075] The core calculations of the entire method of this invention (shooting angle planning, sensor adaptive selection, image segmentation, backbone curve extraction, point cloud feature extraction, visual-point cloud fusion analysis, and disaster damage assessment) are all completed automatically in the local edge computing unit of the UAV. No manual interaction is required during flight, it does not rely on public network communication, and it is not affected by communication interruptions after a disaster.

[0076] The present invention provides a proactive planning method for non-overlapping shooting angles based on tower coordinates, fundamentally solving the problem of the inability to analyze each tower individually in densely populated areas due to multiple overlapping towers. This planning is entirely based on tower coordinates and can be calculated offline before flight, representing an innovative proactive shooting strategy for post-disaster emergency scenarios.

[0077] The vision-point cloud adaptive fusion architecture provided by this invention automatically switches sensor strategies based on the target distance. At long distances, pure vision provides a large coverage area; at medium distances, deep vision-point cloud fusion offers high accuracy; and at close distances, focusing on the tower-ground junction enables precise angle calculation, fully leveraging the complementary advantages of the two sensors.

[0078] This invention utilizes the three-dimensional spatial information of LiDAR point clouds to assist visual segmentation models in distinguishing target towers from background towers, thus solving the problem that pure visual segmentation in dense scenes is easily affected by background interference.

[0079] This invention uses lidar to accurately collect data at the junction of the tower and the ground, and calculates the angle between the tower's main direction and the ground normal vector, providing the most direct and accurate geometric measurement of the tower's tilt, unaffected by shooting angle or image distortion.

[0080] The tower head detachment detection method provided by this invention, which involves lateral comparison of towers along the same power line, can complete the detection without requiring pre-disaster historical data or standard models. Utilizing the inherent characteristic of consistent design height among towers along the same power line for lateral comparison, it is particularly suitable for scenarios where communication is interrupted after a disaster and historical databases are inaccessible.

[0081] The unified detection framework for three types of disaster damage—tilt, breakage, and tower head detachment—provided by this invention covers the most common structural damage types to towers after disasters such as typhoons.

[0082] This invention enables fully automated local edge computing units on drones, without relying on public network communication, without the need for pre-installed sensors, and unaffected by post-disaster communication interruptions.

[0083] Figure 4 This is a structural diagram of a power grid facility damage identification system based on spatial computing according to an embodiment of the present invention.

[0084] like Figure 4 As shown, this invention provides a power grid facility damage identification system based on spatial computing. The system includes: The selection unit 401 is used to perform occlusion analysis on the target device based on the precise coordinates of the target device, and select a preset shooting angle in the unoccluded area; Preferably, the selection unit 401 is used to perform occlusion analysis on the target device based on the precise coordinates of the target device, select a preset shooting angle in the unoccluded area, and is also used to: Map the precise coordinates of the target device to the actual geographic space; Using the coordinates of the current target device after mapping as the origin, calculate the azimuth and distance of adjacent devices relative to the current target device; Each adjacent device is projected onto a 360° azimuth circle centered on the current target device according to its corresponding azimuth angle and distance, and each adjacent device forms a corresponding fan-shaped occlusion zone. Find the sector-shaped occlusion intervals corresponding to all adjacent devices of the current target device, and superimpose the sector-shaped occlusion intervals of all adjacent devices of the current target device to obtain the complete occlusion azimuth interval of the current target device; Within the remaining unobstructed azimuth range, the direction in which the angle between two planar azimuth angles differs by a preset angle is selected as the preset shooting angle.

[0085] The segmentation unit 402 is used to collect target device data based on a preset shooting angle according to an adaptive selection strategy, and combine it with the edge instance segmentation model to obtain the pixel-level target device segmentation region for each target device; Preferably, the adaptive selection strategy includes: Long-range layer: When the LiDAR cannot effectively cover the target device, it automatically switches to pure vision mode and uses only the image data of the gimbal camera to obtain the pixel-level segmentation area of ​​the target device; Mid-range level: When the LiDAR acquisition range can completely cover the main body of the target device, the vision-point cloud deep fusion strategy is enabled; the LiDAR point cloud data assists the visual segmentation model to determine the pixel-level segmentation area of ​​the target device; Close-range local layer: When the lidar can only cover a local area of ​​the target device, the lidar scanning angle is actively adjusted to obtain the pixel-level segmentation area of ​​the target device.

[0086] Extraction unit 403 is used to segment the target device region based on at least two shooting angles and extract multidimensional feature data of the target device. Preferably, the extraction unit 403 is used to extract multidimensional feature data of the target device, and is also used for: For the segmented regions of the target device corresponding to at least two shooting angles, the multidimensional feature data includes the three corners of the target device, one of which is extracted from both angles; morphological thinning algorithms are used to extract skeleton lines; and the outer contour skeletal curves representing the main structure of the target device are selected from the skeleton lines. If the lidar completely covers the main body of the target device, the surface features and three-dimensional contour curves of the main body of the target device can be extracted from the multi-view lidar point cloud data to obtain the three-dimensional deformation information of each facade of the target device. If the lidar only covers a local area of ​​the target equipment, collect point cloud data of the area where the target equipment tower foot meets the ground, and extract the target equipment tower foot point cloud data and the ground point cloud data respectively; calculate the angle between the main body direction of the target equipment and the ground normal vector through the tower foot point cloud data and the ground point cloud data to generate the tilt geometric feature data of the target equipment. The multidimensional feature data and point cloud data are visually fused to obtain fused three-dimensional geometric data; the backbone curve of the target device is extracted from the fused three-dimensional geometric data. The four visual backbone curves corresponding to the two shooting angles are mapped to the same three-dimensional reference coordinate system through projection transformation, and the multi-angle backbone curves are normalized.

[0087] The identification unit 404 is used to identify the damage to the target equipment based on multi-dimensional feature data and through visual-point cloud fusion.

[0088] Preferably, the identification unit 404 is used to identify damage to the target equipment based on multi-dimensional feature data and through visual-point cloud fusion judgment, including determination of tower head detachment: The tower head endpoint height of the normalized target equipment backbone curve is compared laterally with the tower head endpoint height of adjacent similar equipment on the same line. When the height difference exceeds the preset tower head detachment threshold, the target equipment is determined to have experienced a tower head detachment fault.

[0089] Preferably, the identification unit 404 is used to identify the damage to the target equipment based on multi-dimensional feature data and through visual-point cloud fusion determination, including equipment tilt fusion determination: The overall trend of the four normalized backbone curves is used to determine whether the target equipment is tilted; if there is a consistent deflection angle in the relative vertical direction, the tilt direction of the target equipment is initially determined. When the point cloud data completely covers the main body of the target equipment, cross-validation is performed by combining the main body surface features in the three-dimensional deformation information of each facade of the target equipment; when the point cloud data of the target equipment tower foot and the ground point cloud data are obtained, the tilt geometric feature data of the target equipment is used as a precise tilt measurement, and compared with the preset tilt tolerance threshold, the severity of the tilt of the target equipment is determined.

[0090] Preferably, the identification unit 404 is used to identify damage to the target device based on multi-dimensional feature data and through visual-point cloud fusion determination, including device breakage fusion determination: Under normal operating conditions, determine whether the structural morphology of the four normalized backbone curves remains consistent; if any of the following abnormalities occur in the backbone curves: abrupt change in local curvature, discontinuous curve breakage, or significant deviation between the direction and morphology of a single curve and the other curves, determine that asymmetric deformation or local fracture has occurred at the corresponding location of the target equipment; when point cloud data covers the fracture area, use the three-dimensional deformation information of each facade of the target equipment to verify the fracture location and quantify the severity of the fracture deformation of the target equipment.

[0091] The present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a method for identifying power grid facility damage based on spatial computing.

[0092] This invention provides an electronic device, comprising: The aforementioned computer-readable storage medium; and One or more processors for executing a program in a computer-readable storage medium.

[0093] The present invention has been described with reference to a few embodiments. However, it will be apparent to those skilled in the art that other embodiments besides those disclosed above fall equivalently within the scope of the present invention.

[0094] Generally, all terms used in this invention are interpreted according to their ordinary meaning in the art, unless otherwise expressly defined herein. All references to “a / the / the [device, component, etc.]” ​​are openly interpreted as at least one instance of said device, component, etc., unless otherwise expressly stated. The steps of any method disclosed herein need not be performed in the exact order disclosed unless explicitly stated otherwise.

[0095] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0096] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0097] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0098] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0099] Finally, 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 the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for identifying power grid facility damage based on spatial computing, characterized in that, The method includes: Based on the precise coordinates of the target device, an occlusion analysis is performed on the target device, and a preset shooting angle is selected in the unoccluded area; According to the adaptive selection strategy, target device data is collected based on the preset shooting angle, and combined with the edge instance segmentation model, pixel-level target device segmentation region of each target device is obtained; Based on the target device segmentation region from at least two shooting angles, multidimensional feature data of the target device is extracted; the multidimensional feature data includes the three corners of the target device, one of which is extracted from both angles. Based on the multidimensional feature data, the damage to the target equipment is identified through visual-point cloud fusion.

2. The method according to claim 1, characterized in that, The step of performing occlusion analysis on the target device based on its precise coordinates, and selecting a preset shooting angle within the unoccluded area, includes: Map the precise coordinates of the target device to the actual geographic space; Using the coordinates of the current target device after mapping as the origin, calculate the azimuth and distance of adjacent devices relative to the current target device; Each adjacent device is projected onto a 360° azimuth circle centered on the current target device according to its corresponding azimuth angle and distance, and each adjacent device forms a corresponding fan-shaped occlusion zone. Find the sector-shaped occlusion intervals corresponding to all adjacent devices of the current target device, and superimpose the sector-shaped occlusion intervals of all adjacent devices of the current target device to obtain the complete occlusion azimuth interval of the current target device; Within the remaining unobstructed azimuth range, the direction in which the angle between two planar azimuth angles differs by a preset angle is selected as the preset shooting angle.

3. The method according to claim 1, characterized in that, The adaptive selection strategy includes: Long-range layer: When the LiDAR cannot effectively cover the target device, it automatically switches to pure vision mode and uses only the image data of the gimbal camera to obtain the pixel-level segmentation area of ​​the target device; Mid-range level: When the LiDAR acquisition range can completely cover the main body of the target device, the vision-point cloud deep fusion strategy is enabled; the LiDAR point cloud data assists the visual segmentation model to determine the pixel-level segmentation area of ​​the target device; Close-range local layer: When the lidar can only cover a local area of ​​the target device, the lidar scanning angle is actively adjusted to obtain the pixel-level segmentation area of ​​the target device.

4. The method according to claim 3, characterized in that, The extraction of multidimensional feature data from the target device, determined through visual-point cloud fusion, includes: For a segmented region of a target device corresponding to at least two shooting angles, the multidimensional feature data includes three corners of the target device, one of which is extracted from both angles; morphological thinning algorithms are used to extract skeleton lines; and the outer contour skeletal curves representing the main structure of the target device are selected from the skeleton lines. If the lidar completely covers the main body of the target device, the surface features and three-dimensional contour curves of the main body of the target device can be extracted from the multi-view lidar point cloud data to obtain the three-dimensional deformation information of each facade of the target device. If the lidar only covers a local area of ​​the target equipment, collect point cloud data of the area where the target equipment tower foot meets the ground, and extract the target equipment tower foot point cloud data and the ground point cloud data respectively; calculate the angle between the main body direction of the target equipment and the ground normal vector through the tower foot point cloud data and the ground point cloud data to generate the tilt geometric feature data of the target equipment. The multidimensional feature data and point cloud data are visually fused to obtain fused three-dimensional geometric data; the backbone curve of the target device is extracted from the fused three-dimensional geometric data. The four visual backbone curves corresponding to the two shooting angles are mapped to the same three-dimensional reference coordinate system through projection transformation, and the multi-angle backbone curves are normalized.

5. The method according to claim 4, characterized in that, The identification of damage to the target device includes: The tower head endpoint height of the normalized target equipment backbone curve is compared laterally with the tower head endpoint height of adjacent similar equipment on the same line. When the height difference exceeds the preset tower head detachment threshold, the target equipment is determined to have experienced a tower head detachment fault.

6. The method according to claim 4, characterized in that, The identification of damage to the target device includes: The overall trend of the four normalized backbone curves is used to determine whether the target equipment is tilted; if there is a consistent deflection angle in the relative vertical direction, the tilt direction of the target equipment is initially determined. When the point cloud data completely covers the main body of the target equipment, cross-validation is performed by combining the main body surface features in the three-dimensional deformation information of each facade of the target equipment; when the point cloud data of the target equipment tower foot and the ground point cloud data are obtained, the tilt geometric feature data of the target equipment is used as a precise tilt measurement, and compared with the preset tilt tolerance threshold, the severity of the tilt of the target equipment is determined.

7. The method according to claim 4, characterized in that, The identification of damage to the target device includes: Determine whether the structural morphology of the four normalized backbone curves remains consistent under normal operating conditions; If any of the following abnormalities occur in the backbone curve: a sudden change in local curvature, a broken or discontinuous curve, or a significant deviation in the direction and shape of a single curve from the other curves, it is determined that the target equipment has undergone asymmetrical deformation or local breakage at the corresponding location. When point cloud data covers the fracture area, the fracture location is verified by using the three-dimensional deformation information of each facade of the target equipment, and the severity of the fracture deformation of the target equipment is quantified.

8. A power grid facility damage identification system based on spatial computing, characterized in that, The system includes: The selection unit is used to perform occlusion analysis on the target device based on the precise coordinates of the target device, and select a preset shooting angle in the unoccluded area; The segmentation unit is used to collect target device data based on the preset shooting angle according to the adaptive selection strategy, and combine it with the edge instance segmentation model to obtain the pixel-level target device segmentation region for each target device; An extraction unit is used to segment the target device region based on at least two shooting angles and extract multidimensional feature data of the target device. The identification unit is used to identify the damage to the target device based on the multidimensional feature data and through visual-point cloud fusion.

9. The system according to claim 8, characterized in that, The selection unit is further configured to: Map the precise coordinates of the target device to the actual geographic space; Using the coordinates of the current target device after mapping as the origin, calculate the azimuth and distance of adjacent devices relative to the current target device; Each adjacent device is projected onto a 360° azimuth circle centered on the current target device according to its corresponding azimuth angle and distance, and each adjacent device forms a corresponding fan-shaped occlusion zone. Find the sector-shaped occlusion intervals corresponding to all adjacent devices of the current target device, and superimpose the sector-shaped occlusion intervals of all adjacent devices of the current target device to obtain the complete occlusion azimuth interval of the current target device; Within the remaining unobstructed azimuth range, the direction in which the angle between two planar azimuth angles differs by a preset angle is selected as the preset shooting angle.

10. The system according to claim 8, characterized in that, The adaptive selection strategy includes: Long-range layer: When the LiDAR cannot effectively cover the target device, it automatically switches to pure vision mode and uses only the image data of the gimbal camera to obtain the pixel-level segmentation area of ​​the target device; Mid-range level: When the LiDAR acquisition range can completely cover the main body of the target device, the vision-point cloud deep fusion strategy is enabled; the LiDAR point cloud data assists the visual segmentation model to determine the pixel-level segmentation area of ​​the target device; Close-range local layer: When the lidar can only cover a local area of ​​the target device, the lidar scanning angle is actively adjusted to obtain the pixel-level segmentation area of ​​the target device.

11. The system according to claim 10, characterized in that, The extraction unit is also used for: For a segmented region of a target device corresponding to at least two shooting angles, the multidimensional feature data includes three corners of the target device, one of which is extracted from both angles; morphological thinning algorithms are used to extract skeleton lines; and the outer contour skeletal curves representing the main structure of the target device are selected from the skeleton lines. If the lidar completely covers the main body of the target device, the surface features and three-dimensional contour curves of the main body of the target device can be extracted from the multi-view lidar point cloud data to obtain the three-dimensional deformation information of each facade of the target device. If the lidar only covers a local area of ​​the target equipment, collect point cloud data of the area where the target equipment tower foot meets the ground, and extract the point cloud data of the target equipment tower foot and the ground point cloud data respectively. By combining the tower foot point cloud data and the ground point cloud data, the angle between the main body direction of the target equipment and the ground normal vector is calculated, and the tilt geometric feature data of the target equipment is generated. The multidimensional feature data and point cloud data are visually fused to obtain fused three-dimensional geometric data; the backbone curve of the target device is extracted from the fused three-dimensional geometric data. The four visual backbone curves corresponding to the two shooting angles are mapped to the same three-dimensional reference coordinate system through projection transformation, and the multi-angle backbone curves are normalized.

12. The system according to claim 11, characterized in that, The identification unit is further configured to: The tower head endpoint height of the normalized target equipment backbone curve is compared laterally with the tower head endpoint height of adjacent similar equipment on the same line. When the height difference exceeds the preset tower head detachment threshold, the target equipment is determined to have experienced a tower head detachment fault.

13. The system according to claim 12, characterized in that, The identification unit is further configured to: The overall trend of the four normalized backbone curves is used to determine whether the target equipment is tilted; if there is a consistent deflection angle in the relative vertical direction, the tilt direction of the target equipment is initially determined. When the point cloud data completely covers the main body of the target device, cross-validation is performed by combining the main body surface features in the three-dimensional deformation information of each facade of the target device. When the target equipment tower foot point cloud data and ground point cloud data are acquired, the target equipment tilt geometric feature data is used as an accurate tilt measurement. The tilt severity of the target equipment is determined by comparing it with the preset tilt tolerance threshold.

14. The system according to claim 11, characterized in that, The identification unit is further configured to: Under normal operating conditions, determine whether the structural morphology of the four normalized backbone curves remains consistent; if any of the following abnormalities occur in the backbone curves: abrupt change in local curvature, discontinuous curve breakage, or significant deviation between the direction and morphology of a single curve and the other curves, determine that asymmetric deformation or local fracture has occurred at the corresponding location of the target equipment; when point cloud data covers the fracture area, use the three-dimensional deformation information of each facade of the target equipment to verify the fracture location and quantify the severity of the fracture deformation of the target equipment.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1-7.

16. An electronic device, characterized in that, include: The computer-readable storage medium as described in claim 15; as well as One or more processors for executing a program in the computer-readable storage medium.