Intelligent patrol method, device and equipment for ground feature risk of power transmission channel and storage medium

Abstract

The invention relates to the technical field of power transmission safety, in particular to an intelligent patrol method, device and equipment for ground feature risks of a power transmission channel and a storage medium, and the method comprises the steps: collecting a high-resolution satellite remote sensing image of a power transmission line in a target region, and taking the satellite remote sensing image as a reference image; using the unmanned aerial vehicle to obtain an unmanned aerial vehicle image of the target area and a ground monitoring device image; through feature extraction and matching of a multi-source image, exterior point elimination and model estimation are carried out on a matching result, and identification and classification of ground features are realized. Extracting ground feature distance information, physical information, path and environment information, constructing a real-time risk index, and carrying out quantitative evaluation on ground feature risks; and determining an inspection strategy according to a ground feature risk assessment result. According to the method, the satellite remote sensing image, the unmanned aerial vehicle image and the ground monitoring image are organically combined, and the situation that all inspection means operate independently is broken, so that panoramic observation of the power transmission channel is achieved, and multi-source data support is provided for fine-grained power grid risk monitoring.

Description

Technical Field

[0001] The present invention relates to the technical field of transmission safety, and particularly to an intelligent inspection method, device, equipment and storage medium for ground object risks in a transmission channel. Background Art

[0002] During the traditional inspection process of transmission lines, the challenges mainly focus on complex terrain, high labor costs and low efficiency. For areas with complex terrain, such as mountains, canyons and near rivers, traditional inspection methods often have difficulty reaching the target location quickly and effectively. This not only increases the operation difficulty, but also may be unable to conduct a comprehensive inspection due to terrain restrictions, resulting in missed detection of potential safety hazards. In addition, the cost of manual inspection is also a heavy burden. With the rising cost of human resources, coupled with the need for a large number of people to carry out long-distance and high-risk on-site operations, the overall operation and maintenance costs continue to climb. At the same time, the efficiency of manual inspection is relatively low. Due to relying on manual visual judgment and manual recording, the speed of data processing and analysis is slow, and it is difficult to achieve real-time monitoring and rapid response.

[0003] In the prior art, there are methods of using drones for monitoring the state of the transmission line body and ground monitoring and shooting equipment for monitoring the state of the transmission channel. However, the drone inspection is restricted by airspace, weather and cost, and can only conduct state control of the transmission line in a small area. Although the ground monitoring and shooting equipment can conduct real-time online monitoring and shooting, it can only focus on the line protection area and cannot cover the area outside the protection area. Summary of the Invention

[0004] The present invention provides an intelligent inspection method, device, equipment and storage medium for ground object risks in a transmission channel, thus effectively solving the problems in the background art.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is: an intelligent inspection method for ground object risks in a transmission channel, including the following steps: Collect high-resolution satellite remote sensing images of the transmission line in the target area, use the satellite remote sensing images as reference images, and use drones to obtain drone images and ground monitoring and shooting equipment images of the target area; Through feature extraction and matching of multi-source images, outlier rejection and model estimation are performed on the matching results to achieve the recognition and classification of ground objects; Extract ground object distance information, physical information, path and environmental information, construct real-time risk indicators, and quantitatively evaluate the ground object risks; Determine the inspection strategy according to the ground object risk assessment result.

[0006] Further, the feature extraction and matching of the multi-source images include the following steps: Based on a deep learning neural network, the Litong feature extraction algorithm extracts feature points and calculates feature descriptors for the satellite remote sensing images, UAV images, and ground surveillance device images respectively; The feature matching algorithm is used to match the feature points in the UAV images and ground surveillance device images with the feature descriptors in the satellite remote sensing images; Based on the resolution, color, and size characteristics of multi-source images, a multi-view stereo matching algorithm is adopted to generate the spatial point cloud of the transmission line corridor in the target area.

[0007] Further, in the process of using the feature matching algorithm to match the feature points in the UAV images and ground surveillance device images with the feature descriptors in the satellite remote sensing images, it includes: The UAV images and satellite remote sensing images use the nearest neighbor matching. The distance between each feature descriptor in the UAV images and the feature descriptors in the satellite remote sensing images is measured, and the feature descriptor with the closest distance is selected as the matching result; The ground surveillance device images and satellite remote sensing images use the spatial matching algorithm, and each feature descriptor in the ground surveillance images is classified and matched with the feature descriptors in the satellite remote sensing images based on spatial reconstruction.

[0008] Further, the outlier rejection and model estimation of the matching results include: The random sample consensus (RANSAC) algorithm is used to estimate the model by randomly selecting a set of feature point pairs, and the projection error between other feature points and the model is calculated; A threshold is set to determine inliers and outliers, and finally the model with the most inliers is selected as the optimal model.

[0009] Further, after the outlier rejection and model estimation of the matching results, it further includes: According to the optimal model and multi-source images, coordinate transformation is performed to restore the three-dimensional structure of the target area; A three-dimensional digital model of the transmission line corridor is constructed. The three-dimensional digital model of the transmission line corridor is divided into different subgraphs. Each subgraph represents a specific ground object or ground object component. Each subgraph is matched with the ground objects in the ground object database to realize the identification and classification of ground objects.

[0010] Further, the coordinate transformation to restore the three-dimensional structure of the target area includes: According to the optimal model and the geographical information of the satellite remote sensing images, combined with the attitude of the UAV, camera parameters, and the shooting parameter information of multi-view and multi-modal ground surveillance devices, geometric calculations are performed; Infer the coordinates in three-dimensional space and use deep learning methods to restore the three-dimensional structure of the target area.

[0011] Further, the construction of the three-dimensional digital model of the transmission corridor and the division of the three-dimensional digital model of the transmission corridor into different subgraphs include: Construct a three-dimensional digital model of the transmission corridor based on the generated spatial point cloud and the information of the transmission line environment ledger in the target area; Based on the spatial model subgraph segmentation and labeling technology, divide the three-dimensional digital model of the transmission corridor into different subgraphs.

[0012] Further, the construction of the three-dimensional digital model of the transmission corridor further includes: Utilize the results of multi-source image ground object risk identification and the information of the risk history database, analyze the color differences between images from different sources, and perform color correction and matching for the three-dimensional digital model of the transmission corridor; Extract the texture features of the three-dimensional digital model of the transmission corridor, and perform texture correction and synthesis; Perform shape correction on the three-dimensional digital model of the transmission corridor through shape matching and deformation correction algorithms.

[0013] Further, after constructing the three-dimensional digital model of the transmission corridor, it further includes: According to the control requirements of the transmission corridor, design a timing update mechanism, and regularly update the three-dimensional digital model of the transmission corridor according to the preset update period or time interval.

[0014] Further, the extraction of ground object distance information, physical information, path and environment information, and the construction of real-time risk indicators include: Extract the coordinates of the ground object and the coordinates of the two nearest transmission towers near the ground object, and calculate the nearest distance D from the ground object to the transmission line; Extract the physical information, path and environment information H of the ground object; Obtain the meteorological information Q of the area where the ground object is located in real time; Obtain the ground object status detection and evaluation information G according to the feature extraction results; Construct a real-time risk indicator: R = a1D + a2H + a3Q + a4G; In the formula, a 1、 a 2、 a 3、 a4 represents the weight values of each factor.

[0015] Further, the quantitative assessment of the ground object risk includes: Collect the historical fault information of the transmission corridor, analyze the consequences of various faults through simulation, confirm the risk indicator weights, and realize the quantitative assessment of the ground object risk R; According to the quantitative assessment value of the risk, grade the risk levels from small to large.

[0016] The present invention also includes an intelligent inspection device for ground object risks in a power transmission channel, which uses the method as described above. The device includes: An acquisition unit, configured to acquire high-resolution satellite remote sensing images of a power transmission line in a target area, use the satellite remote sensing images as reference images, and use an unmanned aerial vehicle to acquire unmanned aerial vehicle images and ground monitoring device images of the target area; An identification unit, configured to perform outlier rejection and model estimation on the matching results by feature extraction and matching of multi-source images, so as to realize the identification and classification of ground objects; A risk assessment unit, configured to extract ground object distance information, physical information, path and environmental information, construct real-time risk indicators, and quantitatively evaluate the ground object risks; A strategy formulation unit, configured to determine an inspection strategy according to the ground object risk assessment result.

[0017] The present invention also includes a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the method as described above is implemented.

[0018] The present invention also includes a storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method as described above is implemented.

[0019] The beneficial effects of the present invention are as follows: By proposing a multi-source image fusion analysis method, satellite remote sensing images, unmanned aerial vehicle images and ground monitoring device images are organically combined, breaking the situation where each inspection means operates in isolation, realizing the optimization of the efficiency and effect of inspection resources, thereby realizing the panoramic observation of the power transmission channel, and providing multi-source data support for fine-grained power grid risk monitoring. Establish a ground object risk assessment mechanism for the power transmission channel, dynamically monitor the ground object risk status of the power transmission channel, provide limited support for the intelligent control of ground object risks, reduce the occurrence of power grid failures and accidents, improve the reliability and stability of the power grid, reduce the power outage risk, and ensure power supply. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0021] Figure 1 It is a flowchart of the method of the present invention; Figure 2 It is a structural schematic diagram of the device of the present invention; Figure 3 It is a structural schematic diagram of the computer device of the present invention. Specific implementation mode

[0022] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments.

[0023] As Figure 1 shown: An intelligent inspection method for ground object risks in a transmission channel includes the following steps: Collect high-resolution satellite remote sensing images of the transmission line in the target area. Using the satellite remote sensing image as a reference image, use a drone to obtain the drone image and the ground monitoring device image in the target area; Through feature extraction and matching of multi-source images, outlier rejection and model estimation are performed on the matching results to achieve the recognition and classification of ground objects; Extract the distance information, physical information, path and environmental information of the ground object, construct a real-time risk index, and quantitatively evaluate the ground object risk; Determine the inspection strategy according to the ground object risk assessment result.

[0024] By proposing a multi-source image fusion analysis method, the satellite remote sensing image, the drone image and the ground monitoring device image are organically combined, breaking the situation where each inspection method runs in isolation, realizing the optimization of the efficiency and effect of inspection resources, so as to achieve a panoramic view of the transmission channel and provide multi-source data support for fine-grained power grid risk monitoring. Establish a ground object risk assessment mechanism for the transmission channel, dynamically monitor the ground object risk status of the transmission channel, provide limited support for the intelligent control of ground object risks, reduce the occurrence of power grid failures and accidents, improve the reliability and stability of the power grid, reduce the power outage risk, and ensure power supply.

[0025] In this embodiment, the feature extraction and matching of multi-source images include the following steps: Based on a deep learning neural network, use the feature extraction algorithm to extract feature points and calculate feature descriptors for the satellite remote sensing image, the drone image and the ground monitoring device image respectively; Use the feature matching algorithm to match the feature points in the drone image and the ground monitoring device image with the feature descriptors in the satellite remote sensing image; Based on the resolution, color and size characteristics of multi-source images, adopt a multi-view stereo matching algorithm to generate a spatial point cloud of the transmission channel in the target area.

[0026] Among them, when using the feature matching algorithm to match the feature points in the drone image and the ground monitoring device image with the feature descriptors in the satellite remote sensing image, it includes: For the UAV images and satellite remote sensing images, the nearest neighbor matching is used. The distance metric is performed between each feature descriptor in the UAV images and the feature descriptors in the satellite remote sensing images, and the feature descriptor with the closest distance is selected as the matching result. For the ground surveillance device images and MicroStar remote sensing images, the spatial matching algorithm is used. Based on spatial reconstruction, each feature descriptor in the ground surveillance images is classified and matched with the feature descriptors in the satellite remote sensing images.

[0027] To provide better feature matching for image pairs, corner matching is adopted for quantitative measurement. Corners are good matching features. When the viewing point changes, the corner features are stable. In addition, there are intensity mutations in the neighborhood of corners. The corner detection algorithm is used to detect corners in the images. The corner detection algorithms include Harris corner detection algorithm, SIFT feature point detection algorithm ((Scale Invariant Feature Transform)), FAST algorithm corner detection algorithm, and SURF feature point detection algorithm (Speeded-up robust feature).

[0028] Harris corner detection algorithm: The Harris algorithm is a point feature extraction algorithm based on the Moravec algorithm. In 1988, C. Harris and M.J Stephens designed a local image detection window. By moving a small window in different directions, the average change in intensity can be determined. Corners can be easily identified by observing the intensity values within the small window. When moving the window, flat areas do not show intensity changes in all directions. Edge areas do not have intensity changes along the edge direction. For corners, significant intensity changes occur in all directions. The Harris corner detector provides a mathematical method for detecting flat areas, edges, and corners. The features detected by Harris are numerous, with rotational invariance and scale variability.

[0029] SIFT corner detection algorithm: The SIFT algorithm is a scale-invariant feature point detection algorithm, which can be used to identify similar targets in other images. The image features of SIFT are represented as key-point-descriptors. When checking image matching, two sets of key-point-descriptors are provided as input to the Nearest Neighbor Search (NNS), and a closely matching key-point-descriptors is generated.

[0030] FAST Algorithm: FAST is a corner detection algorithm created by Trajkovic and Hedley in 1998. For FAST, corner detection is superior to edge detection because corners have two-dimensional intensity changes and are easily distinguishable from neighboring points. It is suitable for real-time image processing applications. The FAST corner detector should meet the following requirements: 1. The detected positions should be consistent, insensitive to noise changes, and not move for multiple images of the same scene. 2. The detected corners should be as close as possible to the correct positions. 3. The corner detector should be fast enough. To speed up the FAST algorithm, a corner response function (CRF) is usually used. This function gives a numerical value of the corner intensity based on the image intensity of the local neighborhood. The CRF is calculated for the image, and the local maximum of the CRF is taken as the corner. The multi-grid technique improves the calculation speed of the algorithm and suppresses the detected false corners. FAST is an accurate and fast algorithm with good positioning (position accuracy) and high point reliability. The algorithmic difficulty of FAST corner detection lies in the selection of the optimal threshold.

[0031] 2) Image Registration After the feature points are detected, they need to be associated in some way. The correspondence can be determined by methods such as NCC or SDD (Sum of Squared Difference). Normalized cross correlation (NCC). The working principle of cross correlation is to analyze the pixel window around each point in the first image and associate them with the pixel window around each point in the second image. The points with the maximum two-way correlation are taken as the corresponding pairs.

[0032] Calculate the similarity between the "windows" of each displacement (shifts) in the two images based on the image intensity values.

[0033] ; where, is the average value image of the window. are the two pictures respectively. The displacement or offset calculated by the NCC coefficient. The range of the NCC coefficient is. The displacement parameter corresponding to the NCC peak represents the geometric transformation between the two images. The advantage of this method is simple calculation, but it is extremely slow. In addition, such algorithms require significant overlap between the source images.

[0034] 3) Calculate the Homography Matrix Homography matrix estimation is the third step of image stitching. In homography matrix estimation, the unwanted corners that do not belong to the overlapping area are deleted. The RANSAC algorithm is used for homography.

[0035] The RANSAC algorithm fits a mathematical model from an observation dataset that may contain outliers and is an iterative method for robust parameter estimation. This algorithm is nondeterministic because it only produces a reasonable result with a certain probability, which increases as more iterations are performed. The RANSAC algorithm is used to fit a model in a robust manner in the presence of a large number of data outliers. The RANSAC algorithm has many applications in computer vision.

[0036] Randomly select a set of data from the dataset and consider it as valid data (inliers) to determine the model with undetermined parameters. Test all the data in the dataset with this model. The data that satisfies the model becomes inliers, and the rest are outliers (usually points of noise, incorrect measurement, or incorrect data). Iterate until the number of inliers obtained by a certain parameter model is the largest, and then this model is the optimal model. After performing RANSAC, only correct matches can be seen in the image because RANSAC finds a homography matrix related to most points and discards incorrect matches as outliers.

[0037] Given two sets of related points, the next step is to establish the transformation relationship between the two sets of points, that is, the image transformation relationship. Homography is a mapping between two spaces and is often used to represent the correspondence between two images of the same scene. It can match most related feature points and can achieve image projection, enabling one image to overlap with another over a large area through projection. Let the matching points of the two images be respectively...

[0038] 4) Image deformation and fusion The last step is to deform and fuse all input images into a conforming output image. Basically, all input images can be simply deformed onto a plane called the composite panoramic plane.

[0039] Image fusion methods for transition smoothing include feathering, pyramid, and gradient. The last step is to fuse the pixel colors in the overlapping area to avoid seams. The simplest available form is to use feathering, which uses weighted average color values to fuse overlapping pixels. Usually, an alpha factor is used, often called the alpha channel, which has a value of 1 at the central pixel and linearly decreases to 0 at the boundary pixels. When there are at least two overlapping images in the output stitched image, the following alpha value is used to calculate the color at one pixel: Assume that two images overlap in the output image; the coordinates of each pixel point in the image are (x, y), where (R, G, B) is the color value of the pixel. The weighted average color value of the overlapping pixels in the stitched output image is used to calculate the pixel value of (x, y).

[0040] In this embodiment, outlier rejection and model estimation are performed on the matching results, including: Using the Random Sample Consensus (RANSAC) algorithm, a set of feature point pairs is randomly selected for model estimation, and the projection error between other feature points and the model is calculated; A threshold is set to determine inliers and outliers, and finally, the model with the most inliers is selected as the optimal model.

[0041] After outlier rejection and model estimation are performed on the matching results, it also includes: According to the optimal model and multi-source images, coordinate transformation is performed to restore the three-dimensional structure of the target area; A three-dimensional digital model of the transmission channel is constructed, and the three-dimensional digital model of the transmission channel is segmented into different subgraphs. Each subgraph represents a specific ground object or ground object component, and each subgraph is matched with the ground objects in the ground object database to achieve the recognition and classification of ground objects.

[0042] Among them, performing coordinate transformation to restore the three-dimensional structure of the target area includes: According to the optimal model and the geographic information of the satellite remote sensing image, combined with the attitude of the unmanned aerial vehicle, camera parameters, and the shooting parameter information of multi-view and multi-modal ground monitoring devices, geometric calculations are performed; Infer the coordinates in three-dimensional space, and use deep learning methods to restore the three-dimensional structure of the target area.

[0043] Using the ground object image data of the transmission channel, ground object images with multiple views and multiple modalities are obtained, and a multi-view feature extraction algorithm is studied. By analyzing the features such as the shape, texture, and color of ground objects from different views, a multi-view feature description subset is constructed, and multi-modal feature extraction is performed according to the ground object type and physical characteristics; Based on the image rotation invariant feature extraction algorithm, by performing rotation transformation on the ground object image, a rotation invariant feature representation is extracted, the local pixel extreme value features are analyzed, the representative local extreme points in the ground object image are explored, local pixel extreme value features are constructed, and a feature vector representation for ground object recognition is established by combining image rotation invariance and local pixel extreme value features; Using the multi-view and multi-modal ground object image features, a ground object recognition and positioning algorithm is designed to fuse the feature information from different views and modalities to improve the accuracy and robustness of ground object recognition. Considering the multi-form changes of ground objects, accurate recognition and positioning of ground objects at different angles and forms are achieved through feature matching and positioning algorithms; Combined with satellite images and visualization data, using information such as ground object boundaries and feature points, registration and alignment of satellite images and visualization data are performed, a spatial matching algorithm is developed, and the ground object recognition results in satellite images are matched with visualization data to improve the accuracy and reliability of ground object recognition.

[0044] Construct a three-dimensional digital model of the power transmission channel, and divide the three-dimensional digital model of the power transmission channel into different subgraphs, including: Construct a three-dimensional digital model of the power transmission channel based on the generated spatial point cloud and the ledger information of the power transmission line environment in the target area; Based on the spatial model subgraph segmentation and marking technology, divide the three-dimensional digital model of the power transmission channel into different subgraphs.

[0045] As an optimization of the above embodiment, constructing the three-dimensional digital model of the power transmission channel further includes: Utilize the multi-source image ground object risk recognition results and the risk history database information to analyze the color differences between images from different sources, and perform color correction and matching for the three-dimensional digital model of the power transmission channel; Extract the texture features of the three-dimensional digital model of the power transmission channel, and perform texture correction and synthesis; Perform shape correction on the three-dimensional digital model of the power transmission channel through shape matching and deformation correction algorithms.

[0046] After constructing the three-dimensional digital model of the power transmission channel, it further includes: According to the management and control requirements of the power transmission channel, design a timing update mechanism, and regularly update the three-dimensional digital model of the power transmission channel according to the preset update period or time interval.

[0047] (1) Spatial modeling method of power transmission channel based on multi-source images Considering the resolution, color, and size characteristics of multi-source images, adopt a multi-view stereo matching algorithm to generate the spatial point cloud of the ground objects inside the power transmission channel. Extract and match the features of the multi-source images, and restore the three-dimensional structure of the ground objects through triangulation or deep learning methods. Based on the generated spatial point cloud, construct a three-dimensional mesh model of the ground objects, and considering the ground object hidden danger information, propose a three-dimensional mesh generation technology oriented to ground object hidden dangers; Utilize the multi-source image ground object recognition results and the ground object database information to study the differences and correction methods of the color, texture, and shape of the power transmission channel spatial model. Analyze the color differences between images from different sources, and perform color correction and matching to improve the consistency and realism of the spatial model. Extract the texture features of the ground object model, and perform texture correction and synthesis to increase the details and texture of the model. Perform shape correction on the power transmission channel spatial model through shape matching and deformation correction algorithms to improve the accuracy and restoration degree of modeling; Based on the spatial model subgraph segmentation and labeling technology, the decomposition and recognition of submodels such as poles, lines, and ground objects in the transmission channel spatial model are realized. A segmentation algorithm is designed to divide the transmission channel spatial model into different subgraphs, and each subgraph represents a specific ground object or ground object component. The subgraph labeling task is carried out to match each subgraph with the ground objects in the ground object database, realizing the recognition and classification of ground objects. Combining the ground object hidden danger information, the submodel is analyzed and evaluated to improve the discovery and early warning ability of potential hidden dangers.

[0048] (2)Quantitative analysis method of ground object hidden dangers based on multiple perspectives and multiple moments Based on the event-triggered driven update mechanism of ground object state changes, by monitoring the changes in the ground object state, the intelligent update of the transmission channel model is realized. An event-triggered algorithm is developed to identify the events that trigger the update according to the characteristics and rules of the ground object state changes, and trigger the corresponding update operations. A timed update mechanism is designed to regularly update the transmission channel model according to the preset update period or time interval to keep it consistent with the actual ground object state; Using multi-perspective image data, the update technology of the ground object hidden danger state is studied to realize the multi-form normalization and merging recognition of ground object hidden dangers. The feature extraction and matching of multi-perspective images are carried out. By analyzing the feature changes of ground object hidden dangers from different perspectives, different forms of hidden dangers are identified and merged. Analyze the distribution and risk of ground object hidden dangers at different spatial position coordinates, consider the risk assessment of the ground object position, and update the state information of the ground object hidden dangers; Considering the physical properties and path information of the hidden danger ground object, combined with the physical properties of the hidden danger ground object, such as material, structure, etc., and the environmental conditions of the path where it is located, a real-time risk index is constructed to quantify the risk level of the ground object hidden danger. Predict the ground object path and evaluate the probability of its causing transmission channel failures. By simulating and analyzing the consequences of various failures, the risk quantitative assessment based on the temporal position changes of the hidden danger object is realized; In this embodiment, the ground object distance information, physical information, path and environmental information are extracted to construct a real-time risk index, including: Extract the ground object coordinates and the coordinates of the two nearest poles near the ground object, and calculate the nearest distance D from the ground object to the transmission line; Extract the ground object physical information, path and environmental information H; Obtain the meteorological information Q of the area where the ground object is located in real time; According to the feature extraction results, obtain the ground object state detection and evaluation information G; Construct a real-time risk index: R = a1D + a2H + a3Q + a4G; In the formula, a 1、 a 2、 a 3、 a4 represents the weight values of each factor.

[0049] Among them, the quantitative assessment of ground object risks includes: Collect historical fault information of the transmission channel, analyze the consequences of various faults through simulation, confirm the weight of risk indicators, and realize the quantitative assessment R of ground object risks; According to the quantitative assessment value of risks, the risk levels are classified from low to high.

[0050] Space-air-ground collaborative inspection strategy; 1) Based on the ground object risk assessment results, all ground objects that are not without impact trigger the on-site inspection process, and the early warning information of the ground objects will be scrolled and displayed on the human-machine interaction system. The early warning information includes the longitude and latitude information of the ground object risk, coupling equipment information, risk level, ground object type, and the identified ground object image, etc. The equipment information includes tower number, line sag information, and conductor arrangement method, etc. The identified ground object image refers to the result after the recognition of remote sensing images, UAV images, and sub-images of ground monitoring equipment images.

[0051] 2) The monitoring personnel dispatch work orders according to the early warning information. When dispatching work orders, the monitoring images of the ground monitoring device at this moment and the on-site inspection information of the nearby executable UAVs are automatically retrieved. If the target ground object exists in the monitoring image, the monitoring personnel will conduct remote verification of the ground object status. If the ground object status is good, the work order will be closed; if the ground object status is not good, manual on-site disposal will be carried out. If the target ground object does not exist in the monitoring image, but there is an on-site verifiable UAV, the UAV on-site verification will be automatically triggered. The UAV on-site verification takes pictures of the ground object status and automatically transmits the images back to the human-machine interaction system. The monitoring personnel judge the ground object status according to the information transmitted back by the UAV. If the ground object status is good, the work order will be closed; if the ground object status is not good, manual on-site verification will be carried out. If the target ground object does not exist in the monitoring image and there is no on-site verifiable UAV, manual on-site disposal needs to be carried out through the mobile app. The good ground object status means that the ground object will not endanger the normal operation of the nearby transmission line due to its own reasons or external force factors.

[0052] 3) The on-site inspection personnel receive the on-site verification work order through the mobile app, conduct inspections in order from urgent to slow according to the ground object risk level, and feedback the ground object status and disposal situation on-site to the human-machine interaction system through the mobile app.

[0053] 4) The monitoring personnel review and correct the manual on-site verification situation. If the ground object status is good after disposal, the work order will be closed; if the ground object status is not good, the work order will be re-dispatched for on-site disposal until completion.

[0054] As Figure 2 shown, this embodiment also includes an intelligent inspection device for ground object risks in the transmission channel. Using the method as described above, the device includes: The acquisition unit is used to acquire high-resolution satellite remote sensing images of the transmission lines in the target area. Taking the satellite remote sensing images as reference images, it uses drones to obtain drone images and ground monitoring device images of the target area. The recognition unit is used to extract and match the features of multi-source images, remove outliers from the matching results, and perform model estimation to achieve the recognition and classification of ground objects. The risk assessment unit is used to extract the distance information, physical information, path and environmental information of ground objects, construct real-time risk indicators, and quantitatively evaluate the risks of ground objects. The strategy formulation unit is used to determine the inspection strategy based on the ground object risk assessment results.

[0055] Please refer to Figure 3 the structural schematic diagram of the computer device provided by the embodiment of the present application shown. A computer device 400 provided by an embodiment of the present application includes: a processor 410 and a memory 420. The memory 420 stores a computer program executable by the processor 410. When the computer program is executed by the processor 410, it executes the above method.

[0056] An embodiment of the present application also provides a storage medium 430. A computer program is stored on the storage medium 430. When the computer program is run by the processor 410, it executes the above method.

[0057] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. An intelligent inspection method for ground object risks in a power transmission channel, characterized in that, It includes the following steps: Collect high-resolution satellite remote sensing images of the transmission lines in the target area. Using the satellite remote sensing images as reference images, use drones to obtain drone images of the target area and images captured by ground monitoring equipment; Through feature extraction and matching of multi-source images, perform outlier rejection and model estimation on the matching results to achieve the recognition and classification of ground objects; Extract the distance information, physical information, path and environmental information of ground objects, construct real-time risk indicators, and quantitatively evaluate the risks of ground objects; Determine the inspection strategy according to the risk assessment results of ground objects.

2. The intelligent inspection method for ground object risks in a power transmission corridor according to claim 1, wherein The feature extraction and matching of multi-source images includes the following steps: Based on a deep learning neural network, use a feature extraction algorithm to extract feature points and calculate feature descriptors for the satellite remote sensing images, drone images, and images captured by ground monitoring equipment respectively; Use a feature matching algorithm to match the feature points in the drone images and the images captured by ground monitoring equipment with the feature descriptors in the satellite remote sensing images; Based on the resolution, color, and size characteristics of multi-source images, adopt a multi-view stereo matching algorithm to generate a spatial point cloud of the transmission channel in the target area.

3. The intelligent inspection method for ground object risks of a power transmission channel according to claim 2, wherein, In the process of using the feature matching algorithm to match the feature points in the drone images and the images captured by ground monitoring equipment with the feature descriptors in the satellite remote sensing images, it includes: For the drone images and satellite remote sensing images, use the nearest neighbor matching. Measure the distance between each feature descriptor in the drone images and the feature descriptors in the satellite remote sensing images, and select the feature descriptor with the closest distance as the matching result; For the images captured by ground monitoring equipment and satellite remote sensing images, use a spatial matching algorithm to perform classification matching of each feature descriptor in the ground monitoring images and the feature descriptors in the satellite remote sensing images based on spatial reconstruction.

4. The intelligent inspection method for ground object risks of a power transmission channel according to claim 2, wherein The outlier rejection and model estimation of the matching results include: Use the Random Sample Consensus (RANSAC) algorithm. Randomly select a set of feature point pairs for model estimation, and calculate the projection error of other feature points with respect to this model; Set a threshold to determine inliers and outliers, and finally select the model with the most inliers as the optimal model.

5. The intelligent inspection method for ground object risks of a power transmission channel according to claim 4, wherein After the outlier rejection and model estimation of the matching results, it also includes: According to the optimal model and multi-source images, perform coordinate transformation to restore the three-dimensional structure of the target area; Construct a three-dimensional digital model of the transmission channel, divide the three-dimensional digital model of the transmission channel into different subgraphs. Each subgraph represents a specific ground object or ground object component, and match each subgraph with the ground objects in the ground object database to achieve the recognition and classification of ground objects.

6. The intelligent inspection method for ground object risks in a power transmission channel according to claim 5, characterized in that The coordinate transformation to restore the three-dimensional structure of the target area includes: According to the optimal model and the geographical information of the satellite remote sensing images, combined with the attitude of the drone, camera parameters, and the shooting parameter information of multi-view and multi-modal ground monitoring equipment, perform geometric calculations; Infer the coordinates in three-dimensional space, and use deep learning methods to restore the three-dimensional structure of the target area.

7. The intelligent inspection method for ground object risks in a power transmission channel according to claim 5, characterized in that The construction of the three-dimensional digital model of the transmission channel and the division of the three-dimensional digital model of the transmission channel into different subgraphs includes: Based on the generated spatial point cloud and the ledger information of the transmission line environment in the target area, construct a three-dimensional digital model of the transmission corridor. Based on the spatial model subgraph segmentation and labeling technology, divide the three-dimensional digital model of the transmission corridor into different subgraphs.

8. The intelligent inspection method for ground object risks of a power transmission channel according to claim 7, wherein The construction of the three-dimensional digital model of the transmission corridor further includes: Utilize the results of multi-source image ground object risk identification and the information in the risk history database, analyze the color differences between images from different sources, and perform color correction and matching for the three-dimensional digital model of the transmission corridor. Extract the texture features of the three-dimensional digital model of the transmission corridor, and perform texture correction and synthesis. Through the shape matching and deformation correction algorithms, perform shape correction on the three-dimensional digital model of the transmission corridor.

9. The intelligent inspection method for ground object risks of a power transmission channel according to claim 7, characterized in that After constructing the three-dimensional digital model of the transmission corridor, it further includes: According to the management and control requirements of the transmission corridor, design a timing update mechanism, and regularly update the three-dimensional digital model of the transmission corridor according to the preset update period or time interval.

10. The intelligent inspection method for ground object risks in a power transmission channel according to claim 1, characterized in that The extraction of the ground object distance information, physical information, path and environment information, and the construction of real-time risk indicators include: Extract the coordinates of the ground object and the coordinates of the two nearest transmission towers near the ground object, and calculate the nearest distance D from the ground object to the transmission line. Extract the physical information, path and environment information H of the ground object. Obtain the meteorological information Q of the area where the ground object is located in real time. According to the feature extraction results, obtain the ground object status detection and evaluation information G. Construct real-time risk indicators: R = a1D + a2H + a3Q + a4G; where a 1、 a 2、 a 3、 a4 represents the weight value of each factor.

11. The intelligent inspection method for ground object risks of a transmission channel according to claim 10, characterized in that, The quantitative risk assessment of the ground object includes: Collect the historical fault information of the transmission corridor, analyze the consequences of various faults through simulation, confirm the risk indicator weights, and realize the quantitative risk assessment R of the ground object. According to the quantitative risk assessment value, grade the risk levels from small to large.

12. An intelligent inspection device for ground object risks in a power transmission channel, characterized in that, When using the method described in any one of claims 1 to 11, the device includes: An acquisition unit, configured to acquire high-resolution satellite remote sensing images of the transmission line in the target area, use the satellite remote sensing images as reference images, and use unmanned aerial vehicles to obtain unmanned aerial vehicle images and ground monitoring device images of the target area. An identification unit, configured to perform outlier rejection and model estimation on the matching results through feature extraction and matching of multi-source images, and realize the identification and classification of ground objects. A risk assessment unit, configured to extract the ground object distance information, physical information, path and environment information, construct real-time risk indicators, and perform quantitative risk assessment on the ground object. A strategy formulation unit, configured to determine the inspection strategy according to the ground object risk assessment results.

13. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it realizes the method described in any one of claims 1-11.

14. A storage medium, on which a computer program is stored, characterized in that, When the computer program is executed by the processor, it realizes the method described in any one of claims 1-11.

Images