Intelligent detection method and system for defects of power distribution equipment based on visual recognition

Abstract

This invention relates to the field of defect detection technology, and particularly to a method and system for intelligent defect detection of distribution network equipment based on visual recognition. The method includes: locating the distribution network equipment requiring defect detection; generating an aerial photography task using the equipment as a waypoint; activating a pre-stored attitude parameter set in the drone after it reaches the waypoint; collecting aerial photography data; capturing snapshots containing the distribution network equipment; creating a unique identifier group corresponding to each equipment; identifying the location identifiers in the snapshots; and performing preprocessing to obtain the target outline. This invention, by comparing and evaluating rules and determining the presence of equipment defects, enables precise imaging of defect locations and automated detection of distribution network equipment defects without human intervention, greatly improving the consistency of detection results and enhancing the intelligence and efficiency of distribution network equipment inspection.

Description

Technical Field

[0001] This invention relates to the field of defect detection technology, and in particular to a method and system for intelligent defect detection of power distribution network equipment based on visual recognition. Background Technology

[0002] Distribution network equipment refers to all kinds of equipment used in the distribution network. During routine inspections, maintenance personnel need to visually inspect the equipment to determine if there are any abnormalities, such as damage or foreign objects attached. With the development of drone technology, inspection methods are gradually shifting from manual inspection to intelligent and automated inspection. Drones can conduct high-altitude inspections of distribution network equipment according to preset routes and quickly obtain images of the equipment's operating status and temperature distribution information by carrying high-definition camera equipment or infrared imaging equipment.

[0003] In existing technologies, when using drones to inspect power distribution network equipment, the process typically involves simply collecting image data and uploading it to a backend system for manual or semi-automatic analysis after returning to base. If a problem is found, the drone needs to return to the site for re-inspection or remediation. Some inspection methods also use mobile networks to upload the collected image data to the cloud for manual or big data model analysis. However, in practical applications, drones cannot perform real-time analysis and require multiple round trips for repeated shooting, while also placing high demands on network transmission speed and stability, thus affecting overall inspection efficiency.

[0004] Therefore, "how to make a preliminary judgment on image data in a drone and automatically trigger reshoot when an anomaly is detected" is the technical problem that this invention needs to solve. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for intelligent detection of defects in power distribution equipment based on visual recognition, so as to solve the problem mentioned in the background art of "how to make preliminary judgments on image data in UAVs and automatically trigger reshoots when abnormalities are detected".

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A visual recognition-based intelligent detection method for defects in power distribution network equipment, the method comprising:

[0008] The system locates the distribution network equipment that needs defect detection, generates an aerial photography task using the distribution network equipment as a waypoint, and activates the attitude parameter set pre-stored in the drone when the drone arrives at the waypoint. It then collects aerial photography data, captures a snapshot containing the distribution network equipment, and creates an identifier group that corresponds one-to-one with the distribution network equipment. The identifier group consists of a positioning identifier and a calibration identifier. The system identifies the positioning identifier in the snapshot and performs preprocessing to obtain the target contour. The preprocessing includes at least image segmentation, RGB image conversion, and contour extraction.

[0009] Retrieve the baseline contour of the positioning marker, and when the target contour is the same as the baseline contour, define the corresponding snapshot as the target snapshot. From the target snapshot, traverse the pixel coordinates of the calibration marker to obtain the actual coordinates. Determine whether the actual coordinates are the same as the standard coordinates pre-written in the UAV. If so, retrieve the normal operating condition image of the distribution network equipment pre-stored in the UAV, compare the target snapshot with the normal operating condition image, and determine whether there is a pixel difference. If so, obtain the position information of the calibration marker, set several shooting points, locate the real-time position parameters of the UAV, generate the offset route, and extract several monitoring images from the aerial photography data collected by the UAV at the shooting points according to the preset time step.

[0010] Once the drone has returned to base, all monitoring images are extracted and input into the image recognition model. The output is semantic features, which are compared with pre-built evaluation rules to determine whether there are any equipment defects.

[0011] Furthermore, the steps for locating the distribution network equipment requiring defect detection, and generating aerial photography tasks using the distribution network equipment as a waypoint, include:

[0012] Obtain the initial and final positions of the drone, use the nearest neighbor algorithm to connect all waypoints, generate a flight route, and send it to the drone;

[0013] Set the drone's shooting parameters at the points it passes through, mark them on the flight path, and generate an aerial photography mission.

[0014] Furthermore, the step of retrieving the reference contour of the positioning identifier and defining the corresponding snapshot as the target snapshot when the target contour is the same as the reference contour includes:

[0015] Determine the matching index of the baseline contour, calculate the similarity between the target contour and the baseline contour, and define the corresponding snapshot as the target snapshot when the similarity is greater than the threshold.

[0016] Within the target snapshot, a coordinate system is constructed, and the actual coordinates of the calibration identifier are read.

[0017] Furthermore, the method also includes:

[0018] A multi-dimensional evaluation index for configuring positioning identifiers, wherein the multi-dimensional evaluation index includes at least: contour and color features;

[0019] The target snapshot is adjusted based on the aforementioned multidimensional evaluation metrics.

[0020] Furthermore, the step of determining whether the actual coordinates are the same as the standard coordinates pre-written in the drone includes:

[0021] When the actual coordinates differ from the standard coordinates, update the similarity and reselect the target snapshot.

[0022] Configure attribute data for a normal operating condition image, wherein the attribute data includes at least: device number and location information, and insert labels generated from the attribute data into the normal operating condition image.

[0023] Furthermore, the step of comparing the equipment against pre-built evaluation rules to determine whether a defect exists includes:

[0024] Determine the risk level of each evaluation rule, edit the emergency response plan, and establish the correspondence between the emergency response plan and the evaluation rule;

[0025] The system integrates monitoring images, evaluation rules, and emergency response plans to generate defect detection results, which are then sent to a pre-set terminal.

[0026] Furthermore, the system includes:

[0027] The preprocessing module is used to locate the distribution network equipment that needs to be defect-detected, generate aerial photography tasks with the distribution network equipment as the waypoint, and activate the attitude parameter set pre-stored in the drone when the drone arrives at the waypoint, collect aerial photography data, capture snapshots containing the distribution network equipment, and create an identification group that corresponds one-to-one with the distribution network equipment. The identification group consists of positioning identification and calibration identification. The positioning identification in the snapshot is identified and preprocessed to obtain the target contour. The preprocessing includes at least: image segmentation, RGB image conversion and contour extraction.

[0028] The judgment module is used to retrieve the reference contour of the positioning mark, and when the target contour is the same as the reference contour, the corresponding snapshot is defined as the target snapshot. From the target snapshot, the pixel coordinates of the calibration mark are traversed to obtain the actual coordinates. It is judged whether the actual coordinates are the same as the standard coordinates pre-written in the UAV. If so, the normal working condition image of the distribution network equipment pre-stored in the UAV is retrieved, and the target snapshot is compared with the normal working condition image to determine whether there is a pixel difference. If so, the position information of the calibration mark is obtained, several shooting points are set, the real-time position parameters of the UAV are located, the offset route is generated, and several monitoring images are extracted from the aerial photography data collected by the UAV at the shooting points according to the preset time step.

[0029] The output module is used to extract all monitoring images after the drone returns to base, input them into the image recognition model, output semantic features, compare them with the pre-built evaluation rules, and determine whether there are any equipment defects.

[0030] Furthermore, the preprocessing module includes:

[0031] The sending unit is used to obtain the initial and final positions of the UAV, use the nearest neighbor algorithm to connect all the waypoints, generate the flight route, and send it to the UAV.

[0032] The setting unit is used to set the shooting parameters of the drone at the points it passes through, mark them on the flight path, and generate aerial photography missions.

[0033] Furthermore, the determination module includes:

[0034] Define a unit to determine the matching index of the baseline contour, calculate the similarity between the target contour and the baseline contour, and define the corresponding snapshot as the target snapshot when the similarity is greater than the threshold.

[0035] The reading unit is used to construct a coordinate system in the target snapshot and read the actual coordinates of the calibration mark;

[0036] The update unit is used to update the similarity and reselect the target snapshot when the actual coordinates are different from the standard coordinates;

[0037] An insertion unit is used to configure attribute data for a normal operating condition image, wherein the attribute data includes at least: device number and location information, and to insert labels generated from the attribute data into the normal operating condition image.

[0038] Furthermore, the output module includes:

[0039] Establish units to determine the risk level of each evaluation rule, edit emergency response plans, and establish the correspondence between emergency response plans and evaluation rules;

[0040] The sending unit is used to integrate monitoring images, evaluation rules and emergency response plans, generate defect detection results, and send them to a preset terminal.

[0041] Compared with the prior art, the beneficial effects of the present invention are:

[0042] By setting positioning markers at distribution network equipment, drones can quickly lock onto the equipment's location during flight, reducing errors caused by relying solely on GPS positioning and improving positioning efficiency. Setting calibration markers ensures that the drone's defect detection location is consistent each time, guaranteeing image consistency and facilitating comparative analysis of images before and after, thus enabling drone-based defect detection. By comparing actual coordinates with standard coordinates, the deviation between the drone's previous shots and the actual coordinates can be quantified, enabling automatic calibration of the shooting position and ensuring spatial accuracy. By comparing normal operating condition images with target snapshots, rapid anomaly detection of distribution network equipment can be achieved on the drone side, avoiding the generation of large amounts of data transmission. By generating offset routes, the drone can gradually approach the calibration markers, further improving the accuracy and stability of defect identification. Without requiring drone reciprocating flights, and by comparing evaluation rules and determining the presence of equipment defects, precise imaging of defect locations and automated detection of distribution network equipment defects can be achieved without human intervention, greatly improving the consistency of detection results and enhancing the intelligence and efficiency of distribution network equipment inspection. Attached Figure Description

[0043] Figure 1 A flowchart illustrating the intelligent detection method for defects in distribution network equipment based on visual recognition provided in an embodiment of the present invention;

[0044] Figure 2 This is a first sub-flowchart of the intelligent detection method for defects in distribution network equipment based on visual recognition provided in an embodiment of the present invention;

[0045] Figure 3 This is a second sub-flowchart of the intelligent detection method for defects in distribution network equipment based on visual recognition provided in an embodiment of the present invention;

[0046] Figure 4 This is a third sub-flow diagram of the intelligent detection method for defects in distribution network equipment based on visual recognition provided in an embodiment of the present invention;

[0047] Figure 5 A block diagram illustrating the composition of a visual recognition-based intelligent detection system for defects in distribution network equipment, as provided in an embodiment of the present invention.

[0048] Figure 6 This is a block diagram of the preprocessing module in the intelligent detection system for defects in distribution network equipment based on visual recognition provided in an embodiment of the present invention.

[0049] Figure 7 This is a block diagram of the judgment module in the intelligent detection system for defects in distribution network equipment based on visual recognition provided in an embodiment of the present invention.

[0050] Figure 8This is a block diagram of the output module in the intelligent detection system for defects in distribution network equipment based on visual recognition provided in an embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0052] In Example 1, Figure 1 The implementation flow of the intelligent detection method for defects in distribution network equipment based on visual recognition provided in an embodiment of the present invention is illustrated below:

[0053] S100: Locate the distribution network equipment that needs to be defect-detected, generate an aerial photography task using the distribution network equipment as a waypoint, and activate the attitude parameter set pre-stored in the drone when the drone arrives at the waypoint, collect aerial photography data, capture a snapshot containing the distribution network equipment, and create an identification group that corresponds one-to-one with the distribution network equipment. The identification group consists of a positioning identifier and a calibration identifier. Identify the positioning identifier in the snapshot and perform preprocessing to obtain the target contour. The preprocessing includes at least: image segmentation, RGB image conversion, and contour extraction.

[0054] The distribution network equipment requiring defect detection within the area is identified. This equipment can include distribution transformers, ring main units, switchgear, and metering boxes, among others. There are multiple distribution network devices. Using a GIS geographic information system, historical inspection records, or pre-stored location data, the location data of each device is determined. Using the device's location as a waypoint, and combining the initial and final positions of the drone, an aerial photography flight path covering all distribution network devices is generated. The photography requirements for each device are determined, and corresponding attitude parameter sets, such as flight attitude, flight altitude, gimbal angle, shooting distance, and image resolution, are configured for each waypoint. This generates an aerial photography task, which involves using a drone to collect and acquire status information of the distribution network equipment during its operation. Once the drone reaches the waypoint, it retrieves the attitude parameter set pre-stored in the drone. Under the constraints of the attitude parameters, it quickly adjusts the drone's flight attitude, altitude, and gimbal orientation, activates the onboard camera, and captures image data at the waypoint to obtain aerial photography data. The collected images are then preliminarily filtered, and image frames containing the power distribution equipment are extracted to obtain snapshots. A snapshot refers to a single frame of image data containing the power distribution equipment.

[0055] At each distribution network device, an identification group is set up. This group primarily provides an "identifiable, reproducible, and calibrable" reference benchmark for drone inspections. Simply put, the identification group ensures that the drone's shooting position and angle remain consistent during each inspection. The advantage of this method is that it improves the consistency and comparability of image data; by comparing snapshots taken at different times, it is possible to determine whether there are any anomalies in the distribution network device. The identification group consists of positioning markers and calibration markers. Positioning markers occupy a larger spatial area, providing the drone with a fast and accurate spatial reference during inspections. Calibration markers, on the other hand, occupy a smaller spatial area, guiding the drone to automatically adjust its flight attitude and camera parameters, ensuring that the acquired images remain consistent in shooting angle, distance, and composition compared to previous images. For example, a positioning marker could be the outer area of ​​a circular target, while a calibration marker could be the center point of the circular target.

[0056] By using feature matching, snapshots containing location markers are found, and these snapshots are preprocessed. The preprocessing includes image segmentation, RGB image conversion, and contour extraction. Image segmentation refers to separating the region containing the location marker from the background to reduce interference from irrelevant information. RGB image conversion refers to calculating the RGB value of each pixel block. Using a contour extraction algorithm combined with an edge detection algorithm, the boundary contour of the location marker is extracted, and the obtained boundary contour is defined as the target contour, which is a boundary contour containing color information.

[0057] S200: Retrieve the baseline contour of the positioning marker, and when the target contour is the same as the baseline contour, define the corresponding snapshot as the target snapshot. From the target snapshot, traverse the pixel coordinates of the calibration marker to obtain the actual coordinates. Determine whether the actual coordinates are the same as the standard coordinates pre-written in the UAV. If so, retrieve the normal operating condition image of the distribution network equipment pre-stored in the UAV, compare the target snapshot with the normal operating condition image, and determine whether there is a pixel difference. If so, obtain the position information of the calibration marker, set several shooting points, locate the real-time position parameters of the UAV, generate the offset route, and extract several monitoring images from the aerial photography data collected by the UAV at the shooting points according to the preset time step.

[0058] The reference outline of the positioning marker in the power distribution equipment identification group is read from the drone. The reference outline refers to the outline of the positioning marker in a standard state (standard state means the specified position, flight altitude, and pitch angle are all set values). It is then determined whether the target outline is the same as the reference outline. If they are the same, it means the drone has reached the same reference point as previous snapshots, and the snapshot containing the power distribution equipment captured by the drone is defined as the target snapshot. In this embodiment, the power distribution equipment refers to the area within the power distribution equipment that needs to be detected by the drone, such as the bushings, heat sinks, or oil tank casing of a power distribution transformer. In the target snapshot, the area where the calibration marker is located is marked, and the corresponding pixel coordinates are read. For example, the calibration marker includes pixels in the intervals of row 1, column 30-row 1, column 51, row 2, column 12-row 2, column 33, etc., and such pixel intervals are defined as the actual coordinates. The system checks whether the actual coordinates are the same as the standard coordinates set during previous shooting. If they are the same, it means that the drone has reached the same position as in previous shooting, and the camera and drone parameters are the same as in previous shooting. At this time, it retrieves the normal operating condition image corresponding to the power distribution equipment from the drone and compares the normal operating condition image with the target snapshot point by point at the corresponding pixel position to determine whether they are completely identical. If they are identical, there is no pixel difference, which means that there is no defect in the power distribution equipment. If they are different, it means that the status of the power distribution equipment is different from that in previous shooting. The system reads the position information of the calibration mark from the drone and selects several shooting points around the position information. The shooting points are selected by the power distribution equipment maintenance personnel. The shooting points are closer to the power distribution equipment than the shooting position of the target snapshot. The drone's current position parameters are obtained in real time through the positioning module on the drone and compared with the shooting points. The spatial deviation between the drone's current position and the shooting points is calculated to generate an offset route. The offset route is used to guide the drone to the shooting points. Once the drone reaches the shooting location, it extracts several monitoring images from the aerial data according to a preset time step, where the preset step can be 1 second, 2 seconds, etc.

[0059] S300: After the drone returns to base, all monitoring images are extracted and input into the image recognition model. The output is semantic features, which are compared with the pre-built evaluation rules to determine whether there are any equipment defects.

[0060] After collecting the monitoring images, the drone begins its return journey. Upon reaching its destination, the monitoring images are extracted from the drone and input into a trained image recognition model. This model, built on deep learning algorithms, converts pixel information in the image into meaningful semantic features. The monitoring image is input into the model, and the semantic features are output. These features are then matched and compared with pre-built evaluation rules to determine if the distribution network equipment has defects. For example, by recognizing the monitoring image, the output semantic feature might be "crack," and the corresponding area in the monitoring image might be "distribution network equipment insulator." A certain evaluation rule might be: if the insulator has cracks, it is considered a defective equipment.

[0061] In Example 2, Figure 2 The first sub-flowchart of the intelligent detection method for defects in distribution network equipment based on visual recognition provided in an embodiment of the present invention is shown. The following details the steps of locating the distribution network equipment that needs to be detected for defects, using the distribution network equipment as a waypoint, and generating an aerial photography task:

[0062] S101: Obtain the initial and final positions of the drone, use the nearest neighbor algorithm to connect all waypoints, generate a flight route, and send it to the drone.

[0063] Determine the initial and final positions of the drone. Using the nearest neighbor algorithm, start from the initial position and select the nearest unvisited path point as the next node to visit. Gradually expand the path until the final position is reached, generate a flight route, and send this flight route to the drone.

[0064] S102: Set the drone's shooting parameters at the points it passes through, mark them on the flight path, and generate an aerial photography mission.

[0065] A set of shooting parameters is set for each waypoint. These shooting parameters are the heading angle, pitch angle, and lens focal length at the corresponding positions of previous target snapshots. Each waypoint corresponds to a set of shooting parameters. The shooting parameters are marked on the corresponding flight path to generate an aerial photography mission.

[0066] In Example 3, Figure 3 The second sub-flowchart of the intelligent detection method for distribution network equipment defects based on visual recognition provided in this embodiment of the invention is shown. The following details the step of retrieving the reference contour of the positioning identifier and defining the corresponding snapshot as the target snapshot when the target contour is the same as the reference contour:

[0067] S201: Determine the matching index of the baseline contour, calculate the similarity between the target contour and the baseline contour, and when the similarity is greater than the threshold, define the corresponding snapshot as the target snapshot.

[0068] The matching indices for the baseline contour are determined, including contour area and orientation. Using Euclidean distance algorithm and index weighting, the similarity between the target contour and the baseline contour under each matching index is calculated. If the similarity between the two under each matching index is greater than the threshold, the snapshot corresponding to the target contour is defined as the target snapshot.

[0069] S202: In the target snapshot, construct a coordinate system and read the actual coordinates of the calibration mark.

[0070] Select the origin in the target snapshot and construct a coordinate system to determine the coordinates of the pixels corresponding to the calibration markers, and summarize them to obtain the actual coordinates. For example, if the upper left corner of the target snapshot is taken as the origin and a coordinate system is constructed, then the coordinates of the pixel in the first row and first column are (1,1), the coordinates of the pixel in the first row and second column are (1,2), and so on.

[0071] In Example 4, unlike Example 1, the method further includes:

[0072] A multi-dimensional evaluation index for configuring positioning identifiers, wherein the multi-dimensional evaluation index includes at least: contour and color features;

[0073] The target snapshot is adjusted based on the aforementioned multidimensional evaluation metrics.

[0074] When traversing all snapshots and searching for location markers, a multi-dimensional evaluation index is determined for each location marker. This multi-dimensional evaluation index includes the location marker's color and brightness characteristics in the snapshot, as well as specific outline features. Snapshots with the same multi-dimensional evaluation index as previous snapshots are identified, thus initially determining the target snapshot.

[0075] In Example 5, Figure 3 The second sub-flowchart of the intelligent detection method for defects in distribution network equipment based on visual recognition provided in an embodiment of the present invention is shown. The following details the step of determining whether the actual coordinates are the same as the standard coordinates pre-written in the UAV:

[0076] S203: When the actual coordinates are different from the standard coordinates, update the similarity and reselect the target snapshot.

[0077] If the actual coordinates of the calibration mark are different from the standard coordinates, it means that the shooting position of the corresponding snapshot is different from the shooting position of the previous period. In this case, the drone position is adjusted, the similarity is recalculated, and a new target snapshot is selected.

[0078] S204: Configure attribute data for a normal operating condition image, wherein the attribute data includes at least: device number and location information, and insert labels generated from the attribute data into the normal operating condition image.

[0079] Determine the attribute data for each normal operating condition image. The attribute data includes the device number of the distribution network equipment corresponding to the normal operating condition image and the location of the corresponding distribution network equipment. Use the attribute data to generate tags and insert the tags into the normal operating condition image. The tags are displayed in the form of watermarks or notes.

[0080] In Example 6, Figure 4 The diagram shows the third sub-process flowchart of the intelligent detection method for distribution network equipment defects based on visual recognition provided in this embodiment of the invention. The following details the step of comparing the pre-constructed evaluation rules to determine whether equipment defects exist:

[0081] S301: Determine the risk level of each evaluation rule, edit the emergency response plan, and establish the correspondence between the emergency response plan and the evaluation rule.

[0082] Edit several evaluation rules, set the risk level for each evaluation rule, including high, medium and low risk levels, and write the corresponding emergency response plan into each evaluation rule. The emergency response plan is the specific handling measures, such as notifying maintenance personnel to handle the situation.

[0083] S302: Integrate monitoring images, evaluation rules, and emergency response plans to generate defect detection results and send them to the preset terminal.

[0084] Once it is determined through monitoring images that there is indeed a defect in the distribution network equipment, the corresponding monitoring images, evaluation rules, and emergency response plans are summarized to generate a defect detection result, which is then sent to a preset terminal, namely the distribution network equipment management personnel terminal.

[0085] Figure 5 The diagram illustrates the structural block diagram of a visual recognition-based intelligent detection system for distribution network equipment defects provided in an embodiment of the present invention. The visual recognition-based intelligent detection system for distribution network equipment defects 1 includes:

[0086] Preprocessing module 11 is used to locate the distribution network equipment that needs to be defect-detected, generate an aerial photography task with the distribution network equipment as the waypoint, activate the attitude parameter set pre-stored in the drone when the drone arrives at the waypoint, collect aerial photography data, capture a snapshot containing the distribution network equipment, create an identification group corresponding to the distribution network equipment, wherein the identification group consists of positioning identification and calibration identification, identify the positioning identification in the snapshot, and perform preprocessing to obtain the target contour, wherein the preprocessing includes at least: image segmentation, RGB image conversion and contour extraction;

[0087] The judgment module 12 is used to retrieve the reference contour of the positioning mark, and when the target contour is the same as the reference contour, the corresponding snapshot is defined as the target snapshot. From the target snapshot, the pixel coordinates of the calibration mark are traversed to obtain the actual coordinates. It is judged whether the actual coordinates are the same as the standard coordinates pre-written in the UAV. If so, the normal working condition image of the distribution network equipment pre-stored in the UAV is retrieved. The target snapshot and the normal working condition image are compared to determine whether there is a pixel difference. If so, the position information of the calibration mark is obtained, several shooting points are set, the real-time position parameters of the UAV are located, the offset route is generated, and several monitoring images are extracted from the aerial photography data collected by the UAV at the shooting points according to the preset time step.

[0088] The output module 13 is used to extract all monitoring images after the UAV returns to base, input them into the image recognition model, output semantic features, compare them with the pre-built evaluation rules, and determine whether there are any equipment defects.

[0089] Figure 6 This diagram illustrates the composition of a preprocessing module 11 in a vision-based intelligent detection system for distribution network equipment defects, as provided in an embodiment of the present invention. The preprocessing module 11 includes:

[0090] The sending unit 111 is used to obtain the initial and final positions of the UAV, use the nearest neighbor algorithm to connect all the waypoints, generate the flight route, and send it to the UAV.

[0091] Setting unit 112 is used to set the shooting parameters of the drone at the passing points, mark them on the flight route, and generate aerial photography tasks.

[0092] Figure 7 This diagram illustrates the structural composition of the judgment module 12 in the intelligent detection system for defects in distribution network equipment based on visual recognition provided in an embodiment of the present invention. The judgment module 12 includes:

[0093] Definition unit 121 is used to determine the matching index of the reference contour, calculate the similarity between the target contour and the reference contour, and define the corresponding snapshot as the target snapshot when the similarity is greater than the threshold.

[0094] The reading unit 122 is used to construct a coordinate system in the target snapshot and read out the actual coordinates of the calibration mark;

[0095] Update unit 123 is used to update the similarity and reselect the target snapshot when the actual coordinates are different from the standard coordinates;

[0096] The insertion unit 124 is used to configure attribute data of the normal operating condition image, wherein the attribute data includes at least: device number and location information, and inserts labels generated by the attribute data into the normal operating condition image.

[0097] Figure 8 This diagram illustrates the structural block diagram of the output module 13 in the intelligent detection system for defects in distribution network equipment based on visual recognition provided in an embodiment of the present invention. The output module 13 includes:

[0098] Unit 131 is established to determine the risk level of each evaluation rule, edit emergency response plans, and establish the correspondence between emergency response plans and evaluation rules;

[0099] The sending unit 132 is used to integrate monitoring images, evaluation rules and emergency response plans, generate defect detection results, and send them to a preset terminal.

[0100] The preprocessing module 11 is mainly used to complete step S100, the judgment module 12 is mainly used to complete step S200, and the output module 13 is mainly used to complete step S300.

[0101] The issuing unit 111 is mainly used to complete step S101, and the setting unit 112 is mainly used to complete step S102;

[0102] The definition unit 121 is mainly used to complete step S201, the reading unit 122 is mainly used to complete step S202, the update unit 123 is mainly used to complete step S203, and the insertion unit 124 is mainly used to complete step S204.

[0103] The establishment unit 131 is mainly used to complete step S301, and the sending unit 132 is mainly used to complete step S302.

[0104] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent detection of defects of a device in a network configuration based on visual recognition, characterized in that, The method includes: The system locates the distribution network equipment that needs defect detection, generates an aerial photography task using the distribution network equipment as a waypoint, and activates the attitude parameter set pre-stored in the drone when the drone arrives at the waypoint. It then collects aerial photography data, captures a snapshot containing the distribution network equipment, and creates an identifier group that corresponds one-to-one with the distribution network equipment. The identifier group consists of a positioning identifier and a calibration identifier. The system identifies the positioning identifier in the snapshot and performs preprocessing to obtain the target contour. The preprocessing includes at least image segmentation, RGB image conversion, and contour extraction. Retrieve the baseline contour of the positioning marker, and when the target contour is the same as the baseline contour, define the corresponding snapshot as the target snapshot. From the target snapshot, traverse the pixel coordinates of the calibration marker to obtain the actual coordinates. Determine whether the actual coordinates are the same as the standard coordinates pre-written in the UAV. If so, retrieve the normal operating condition image of the distribution network equipment pre-stored in the UAV, compare the target snapshot with the normal operating condition image, and determine whether there is a pixel difference. If so, obtain the position information of the calibration marker, set several shooting points, locate the real-time position parameters of the UAV, generate the offset route, and extract several monitoring images from the aerial photography data collected by the UAV at the shooting points according to the preset time step. Once the drone has returned to base, all monitoring images are extracted and input into the image recognition model. The output is semantic features, which are compared with pre-built evaluation rules to determine whether there are any equipment defects.

2. The visual recognition-based intelligent detection method for device defect in network distribution according to claim 1, characterized in that, The steps for locating the distribution network equipment that needs to be defect-detected, and generating an aerial photography mission using the distribution network equipment as a waypoint, include: Obtain the initial and final positions of the drone, use the nearest neighbor algorithm to connect all waypoints, generate a flight route, and send it to the drone; Set the drone's shooting parameters at the points it passes through, mark them on the flight path, and generate an aerial photography mission.

3. The intelligent detection method for defects in distribution network equipment based on visual recognition according to claim 1, characterized in that, The step of retrieving the reference contour of the positioning marker and defining the corresponding snapshot as the target snapshot when the target contour is the same as the reference contour includes: Determine the matching index of the baseline contour, calculate the similarity between the target contour and the baseline contour, and define the corresponding snapshot as the target snapshot when the similarity is greater than the threshold. Within the target snapshot, a coordinate system is constructed, and the actual coordinates of the calibration identifier are read.

4. The intelligent detection method for defects in distribution network equipment based on visual recognition according to claim 3, characterized in that, The method further includes: A multi-dimensional evaluation index for configuring positioning identifiers, wherein the multi-dimensional evaluation index includes at least: contour and color features; The target snapshot is adjusted based on the multidimensional evaluation metrics.

5. The intelligent detection method for defects in distribution network equipment based on visual recognition according to claim 3, characterized in that, The step of determining whether the actual coordinates are the same as the standard coordinates pre-written in the drone includes: When the actual coordinates differ from the standard coordinates, update the similarity and reselect the target snapshot. Configure attribute data for a normal operating condition image, wherein the attribute data includes at least: device number and location information, and insert labels generated from the attribute data into the normal operating condition image.

6. The intelligent detection method for defects in distribution network equipment based on visual recognition according to claim 1, characterized in that, The step of comparing the equipment against pre-built evaluation rules to determine whether there are any defects includes: Determine the risk level of each evaluation rule, edit the emergency response plan, and establish the correspondence between the emergency response plan and the evaluation rule; The system integrates monitoring images, evaluation rules, and emergency response plans to generate defect detection results, which are then sent to a pre-set terminal.

7. A visual recognition-based intelligent defect detection system for power distribution network equipment, characterized in that, The system includes: The preprocessing module is used to locate the distribution network equipment that needs to be defect-detected, generate aerial photography tasks with the distribution network equipment as the waypoint, and activate the attitude parameter set pre-stored in the drone when the drone arrives at the waypoint, collect aerial photography data, capture snapshots containing the distribution network equipment, and create an identification group that corresponds one-to-one with the distribution network equipment. The identification group consists of positioning identification and calibration identification. The positioning identification in the snapshot is identified and preprocessed to obtain the target contour. The preprocessing includes at least: image segmentation, RGB image conversion and contour extraction. The judgment module is used to retrieve the reference contour of the positioning mark, and when the target contour is the same as the reference contour, the corresponding snapshot is defined as the target snapshot. From the target snapshot, the pixel coordinates of the calibration mark are traversed to obtain the actual coordinates. It is judged whether the actual coordinates are the same as the standard coordinates pre-written in the UAV. If so, the normal working condition image of the distribution network equipment pre-stored in the UAV is retrieved, and the target snapshot is compared with the normal working condition image to determine whether there is a pixel difference. If so, the position information of the calibration mark is obtained, several shooting points are set, the real-time position parameters of the UAV are located, the offset route is generated, and several monitoring images are extracted from the aerial photography data collected by the UAV at the shooting points according to the preset time step. The output module is used to extract all monitoring images after the drone returns to base, input them into the image recognition model, output semantic features, compare them with the pre-built evaluation rules, and determine whether there are any equipment defects.

8. The intelligent detection system for defects in distribution network equipment based on visual recognition according to claim 7, characterized in that, The preprocessing module includes: The sending unit is used to obtain the initial and final positions of the UAV, use the nearest neighbor algorithm to connect all the waypoints, generate the flight route, and send it to the UAV. The setting unit is used to set the shooting parameters of the drone at the points it passes through, mark them on the flight path, and generate an aerial photography mission.

9. The intelligent detection system for defects in distribution network equipment based on visual recognition according to claim 7, characterized in that, The judgment module includes: Define a unit to determine the matching index of the baseline contour, calculate the similarity between the target contour and the baseline contour, and define the corresponding snapshot as the target snapshot when the similarity is greater than the threshold. The reading unit is used to construct a coordinate system in the target snapshot and read the actual coordinates of the calibration mark; The update unit is used to update the similarity and reselect the target snapshot when the actual coordinates are different from the standard coordinates; An insertion unit is used to configure attribute data for a normal operating condition image, wherein the attribute data includes at least: device number and location information, and to insert labels generated from the attribute data into the normal operating condition image.

10. The intelligent detection system for defects in distribution network equipment based on visual recognition according to claim 9, characterized in that, The output module includes: Establish units to determine the risk level of each evaluation rule, edit emergency response plans, and establish the correspondence between emergency response plans and evaluation rules; The sending unit is used to integrate monitoring images, evaluation rules and emergency response plans, generate defect detection results, and send them to a preset terminal.

Images