CN-121904594-B - Distribution network power equipment inspection method, device, equipment and medium

CN121904594BCN 121904594 BCN121904594 BCN 121904594BCN-121904594-B

Abstract

The application relates to a method, a device, equipment and a medium for inspecting distribution network power equipment, which relate to the technical field of power equipment inspection and comprise the steps of synchronously collecting point cloud data and visible light image data and preprocessing; the method comprises the steps of constructing a multi-dimensional dynamic clearance threshold system based on a distribution network digital twin model and combining environment parameters, obstacle semantic types and voltage levels, achieving clearance safety distance calculation and obstacle risk grading elimination, reversely projecting the distribution network equipment three-dimensional model to an image determination boundary frame, identifying key components through YOLOv multi-mode feature fusion target detection models, extracting three-dimensional geometric features and two-dimensional texture features of the key components, constructing geometric-texture dual-guide factors, and achieving defect feature accurate enhancement by adaptively adjusting enhancement algorithm parameters. The application can realize the synergy of clearance inspection and defect detection, improve the fusion precision of multidimensional sensing data, and guide image enhancement by utilizing geometric information so as to improve the recognition accuracy in a complex environment.

Inventors

HU BING
WU JIAHAO
YU NANNAN
LONG JIAWEN
LIN HAO
LONG ZEYU
TANG KAI
LIU WEIYAN
YU SHI
QIU ZHENYU
SHEN HONGJIE
Liang Guiyu
XU CHANG

Assignees

江西科晨洪兴信息技术有限公司

Dates

Publication Date: 20260512
Application Date: 20260323

Claims (10)

1. The inspection method for the distribution network power equipment is characterized by comprising the following steps of: Synchronously acquiring point cloud data and visible light image data of a laser radar of a power distribution network area, and preprocessing the point cloud data and the visible light image data; Constructing a distribution network digital twin model based on the preprocessed point cloud data and the preprocessed visible light image data, wherein the distribution network digital twin model comprises a digital surface model, a digital elevation model, a normalized digital surface model and a distribution network equipment semantic model; Performing clearance analysis on the power distribution network area based on the distribution network digital twin model to identify power wires and obstacles in the power distribution network area, and constructing a multi-dimensional dynamic clearance threshold system by combining environment perception data, obstacle semantic features and distribution network voltage levels to calculate clearance safety distances between the power wires and the obstacles; Carrying out risk classification based on the clearance safety distance and the semantic features of the obstacles, removing non-electric obstacles in the electric power distribution network area, extracting electric power towers of the electric power distribution network area, establishing a topological connection relationship between the electric power wires and the electric power towers through space topological relationship analysis, and constructing a distribution network electric power equipment three-dimensional model containing electric power equipment space position information and topological relationship information; Reversely projecting the spatial position information of the power equipment in the three-dimensional model of the distribution network power equipment to the visible light image data, determining an image boundary box of a key component, and identifying the key component image in the image boundary box by using a pre-trained multi-modal feature fused deep learning target detection model, wherein the multi-modal feature fused deep learning target detection model is YOLOv detection model fusing three-dimensional geometric features and two-dimensional texture features; And extracting three-dimensional geometric features and two-dimensional texture features corresponding to the key component image, constructing a geometric-texture double-guide factor, and performing defect feature self-adaptive enhancement processing on the key component image by using the geometric-texture double-guide factor to obtain the key component image after defect enhancement.
2. The method according to claim 1, wherein preprocessing the point cloud data specifically comprises: calculating the average distance from each point in the point cloud data to a plurality of nearest neighbors by adopting a statistical outlier removal algorithm, and removing noise points with the average distance exceeding a preset Gaussian distribution range in the point cloud data; and filtering and downsampling the denoised point cloud data by adopting a voxel grid downsampling algorithm, and performing interframe registration and fusion processing on the filtered and downsampled point cloud data to obtain the processed point cloud data.
3. The method according to claim 1, wherein preprocessing the visible light image data comprises: adopting a Brown-Conrady distortion model, and carrying out distortion correction processing on the visible light image data by utilizing camera internal parameters and distortion coefficients which are obtained through calibration in advance; Performing contrast enhancement processing on the corrected visible light image data by adopting a limited contrast self-adaptive histogram equalization algorithm so as to improve the overall contrast of the visible light image data; And carrying out inter-frame deblurring processing on the visible light image data subjected to the contrast enhancement processing so as to optimize image texture details of the visible light image data.
4. The method according to claim 1, wherein the step of performing a headroom analysis on the power distribution network area based on the distribution network digital twin model to identify power conductors and obstacles in the power distribution network area, and constructing a multi-dimensional dynamic headroom threshold system by combining environmental awareness data, obstacle semantic features and distribution network voltage levels to calculate a headroom safety distance between the power conductors and the obstacles, specifically comprises: based on the elevation characteristic value of the normalized digital surface model and the semantic model of the distribution network equipment, extracting a wire candidate region in the power distribution network region by adopting a sliding window and semantic segmentation combined algorithm, and fitting to obtain the three-dimensional coordinates of the electric wires in the wire candidate region by utilizing a random sampling consistency algorithm introducing wire curvature constraint; Based on the distribution network digital twin model, fusing three-dimensional geometric features and semantic features of the obstacle, identifying a region which has an elevation feature value larger than 0 and is not conductive in the power distribution network region as the obstacle, and carrying out semantic classification on the obstacle to obtain a semantic classification result of the obstacle, wherein the semantic classification result at least comprises one of tree, building and construction machinery classification; dynamically adjusting a clearance safety threshold according to the distribution network voltage level, the obstacle semantic classification result and the environment sensing data, wherein the environment sensing data comprises wind speed, temperature, humidity and ice coating thickness; And calculating a three-dimensional minimum Euclidean distance between the electric power lead and the obstacle according to the three-dimensional coordinates of the lead, the coordinates of the surface points of the obstacle and the environmental perception data, comparing the three-dimensional minimum Euclidean distance with the clearance safety threshold, and determining the final clearance safety distance.
5. The method according to claim 1, wherein the risk classification is performed based on the clearance safety distance and the obstacle semantic feature, non-electric obstacles in the electric power distribution network area are removed, electric power towers of the electric power distribution network area are extracted, a topological connection relation between the electric power wires and the electric power towers is established through space topological relation analysis, and a three-dimensional model of the distribution network electric power equipment including electric power equipment space position information and topological relation information is constructed, specifically including: performing surface object classification processing on the normalized digital surface model by using a classifier integrating geometric features and semantic features so as to distinguish power equipment objects from non-power obstacle objects; Based on the classification result, combining the geometric features and semantic features of the electric power tower, extracting the tower outline of the electric power tower by adopting a combined algorithm of morphological expansion operation, outline detection and semantic verification, and calculating to obtain three-dimensional bounding box coordinates of the electric power tower; Carrying out space association and semantic binding on the electric power wires and the electric power towers through space topological relation analysis, and establishing a topological connection relation of the wires and the towers; Classifying risks of the non-electric obstacle objects based on the clearance safety distance and semantic features of the non-electric obstacle objects, and eliminating low-priority non-electric obstacle objects in the electric power distribution network area according to the risk classes; And generating a distribution network power equipment three-dimensional model based on the power equipment objects reserved in the power distribution network area, the three-dimensional bounding box coordinates and the topological connection relation, wherein the distribution network power equipment three-dimensional model comprises spatial position information and topological relation information of towers, wires and accessory parts.
6. The method of claim 1, wherein the construction of the multi-modal feature fused deep learning target detection model specifically comprises: a point cloud feature projection fusion module is additionally arranged in a backbone network of YOLOv framework, wherein the point cloud feature projection fusion module is configured to project three-dimensional geometric features of a three-dimensional model of the distribution network power equipment to a two-dimensional image feature layer so as to realize multi-mode fusion of the three-dimensional geometric features and the two-dimensional texture features; Introducing an attention mechanism fusion layer in the Neck network of the YOLOv framework, wherein the attention mechanism fusion layer is configured to perform attention weighting processing on the multi-mode fusion features so as to enhance the feature expression weight of the key component image; and adding a geometric constraint loss term into the YOLOv architecture loss function, wherein the geometric constraint loss term is configured to construct a constraint term based on the deviation of the three-dimensional geometric feature and participate in back propagation calculation so as to optimize the regression capability and the category discrimination capability of the model on the target space position.
7. The method according to claim 1, wherein the extracting the three-dimensional geometric feature and the two-dimensional texture feature corresponding to the critical component image, constructing a geometric-texture dual-guide factor, and performing defect feature adaptive enhancement processing on the critical component image by using the geometric-texture dual-guide factor, so as to obtain a critical component image after defect enhancement, specifically including: Extracting three-dimensional geometric features of a three-dimensional point cloud cluster corresponding to the key component image, and extracting two-dimensional texture features of the key component image, wherein the three-dimensional geometric features comprise a surface normal map, a local curvature map, a relative height map and a geometric roughness map, and the two-dimensional texture features comprise a texture gradient map, an edge detection map, a gray level co-occurrence matrix feature map and a defect texture sensitivity map; Calculating a geometric guide factor by adopting a parameter self-adaptive Sigmoid function based on the relative height map and the geometric roughness map so as to distinguish an electric equipment body region from a background region and mark a geometric abnormal region; Calculating texture guide factors by adopting a Gaussian weighting algorithm based on the texture gradient map and the defect texture sensitivity map so as to distinguish a defect texture region from a normal texture region and mark a texture abnormal region; Carrying out weighted fusion on the geometric guide factor and the texture guide factor to obtain the geometric-texture double guide factor, wherein the weighted fusion weight is configured to be dynamically adjusted according to the illumination condition or the background complexity of the inspection scene; And based on the geometric-texture double-guide factor, adaptively adjusting the filter kernel size, the enhancement intensity and the texture retention coefficient of a Retinex image enhancement algorithm, and executing the processing of enhancing the contrast of the defect texture region, retaining the normal texture region and inhibiting the interference of the background region on the key component image to obtain the key component image after defect enhancement.
8. Join in marriage net power equipment and patrol and examine device, characterized by, include: the acquisition module is used for synchronously acquiring point cloud data and visible light image data of the laser radar in the power distribution network area and preprocessing the point cloud data and the visible light image data; The first construction module is used for constructing a distribution network digital twin model based on the preprocessed point cloud data and the preprocessed visible light image data, wherein the distribution network digital twin model comprises a digital surface model, a digital elevation model, a normalized digital surface model and a distribution network equipment semantic model; The calculation module is used for carrying out clearance analysis on the power distribution network area based on the distribution network digital twin model so as to identify power wires and obstacles in the power distribution network area, and constructing a multi-dimensional dynamic clearance threshold system by combining environment perception data, obstacle semantic features and distribution network voltage levels so as to calculate clearance safety distance between the power wires and the obstacles; the second construction module is used for carrying out risk classification based on the clearance safety distance and the obstacle semantic characteristics, removing non-electric obstacles in the electric power distribution network area, extracting electric power towers of the electric power distribution network area, establishing a topological connection relation between the electric power wires and the electric power towers through space topological relation analysis, and constructing a distribution network electric power equipment three-dimensional model containing electric power equipment space position information and topological relation information; The identification module is used for reversely projecting the spatial position information of the power equipment in the three-dimensional model of the distribution network power equipment to the visible light image data, determining an image boundary box of a key component, and identifying the key component image in the image boundary box by utilizing a pre-trained multi-mode feature fused deep learning target detection model, wherein the multi-mode feature fused deep learning target detection model is a YOLOv detection model fusing three-dimensional geometric features and two-dimensional texture features; And the enhancement module is used for extracting the three-dimensional geometric features and the two-dimensional texture features corresponding to the key component image, constructing a geometric-texture double-guide factor, and performing defect feature self-adaptive enhancement processing on the key component image by using the geometric-texture double-guide factor to obtain the key component image after defect enhancement.
9. An electronic device comprising a storage medium, a processor and a computer program stored on the storage medium and executable on the processor, characterized in that the processor implements the method of any one of claims 1 to 7 when executing the computer program.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the method of any of claims 1 to 7.

Description

Distribution network power equipment inspection method, device, equipment and medium Technical Field The application relates to the technical field of power equipment inspection, in particular to an inspection method, device, equipment and medium for distribution network power equipment. Background Along with the advanced promotion of the construction of a novel power system and the intelligent upgrading of a power distribution network, the coverage range of the power distribution network line is gradually wide, the scene is also gradually complex, equipment is also greatly increased in advance, and more manpower and material resources are required for inspection of equipment such as towers, cables, insulators and the like. The manual inspection is restricted by topography (such as steep road sections in mountain areas and river barriers), efficiency bottlenecks, hidden danger of close-range contact of safety risk high-voltage equipment and the like, the electric unmanned aerial vehicle can effectively avoid the safety risks of climbing towers and close-range contact of the high-voltage equipment by virtue of strong functional advantages of the electric unmanned aerial vehicle, plays an important role in distribution network inspection, can remarkably improve the manual efficiency, and has become a core means for monitoring distribution network equipment. The core targets of the power distribution network inspection are to synchronously guarantee ' headroom safety ' and ' equipment health ', the influence of the headroom safety ' and the ' equipment health ' on the operation stability of a power grid is very large, the wire discharge and short-circuit tripping are easy to occur due to insufficient headroom, and from the viewpoint of the equipment health dimension, the faults such as insulator breakage, cable skin cracking, pole tower corrosion and the like can cause accidents such as insulation breakdown, wire breakage, even power failure and the like if the faults are not found or treated in time. Therefore, improvement of clearance monitoring and defect detection is important to reduce fault occurrence rate and ensure power supply reliability. However, the existing unmanned aerial vehicle inspection technology still has some problems, and improvement in automation and precision is needed. First, clearance safety assessment (measuring the safety distance between a wire and a tree, house) and equipment defect detection (identifying insulator breakage, cable cracking, etc.) are two independent processes, and not only is the cost increased, but also the labor efficiency is reduced. And secondly, the single sensor has insufficient data precision, unreliable clearance analysis, difficulty in distinguishing 'power equipment' from 'trees, houses and the like' by using a laser radar, manual screening of targets and easy missed judgment, incapability of accurately calculating the three-dimensional distance between a wire and an obstacle by using a visible light image, lack of quantitative basis in clearance evaluation, and incapability of meeting the requirements of safety inspection. Finally, the anti-interference performance is weak under the complex environment, the data quality is poor, the defect omission ratio is increased due to the real conditions such as outdoor illumination change, meteorological interference, background confusion and the like, and the clearance calculation accuracy is reduced. In addition, most of the existing image enhancement technologies (such as traditional Retinex) are global or enhancement based on two-dimensional features, real defects cannot be distinguished from background textures by utilizing three-dimensional geometric information, excessive background enhancement is easily caused in a complex scene, and defect recognition is disturbed. In summary, how to realize the synergy of headroom inspection and defect detection, improve the fusion precision of multidimensional sensing data, and guide image enhancement by utilizing geometric information to improve the recognition accuracy in a complex environment is a technical problem to be solved in the current inspection technology of the unmanned aerial vehicle of the distribution network toward automation and accurate upgrading. Disclosure of Invention In view of the above, the application provides a method, a device, equipment and a medium for inspecting distribution network power equipment, which mainly aims to solve the technical problems of how to realize the synergy of headroom inspection and defect detection, improve the fusion precision of multidimensional sensing data, and guide image enhancement by utilizing geometric information so as to improve the recognition accuracy in a complex environment. In a first aspect, the present application provides a method for inspecting power equipment in a distribution network, including: Synchronously acquiring point cloud data and visible light image data of a laser radar of a power di