A substation protection panel inspection method and device and medium

Abstract

The application discloses a kind of substation protection pressboard inspection method, device and medium, environment data are collected by laser radar, and environment map is established according to environment data, based on environment map and screen cabinet coordinate data, after moving to target inspection point, by multiple different preset height image sensors, image acquisition is carried out to target screen cabinet, to obtain corresponding original image and carry out data preprocessing, the corresponding target image obtained is input into the target identification model pre-trained, to obtain the identification result including identification state and pressboard position coordinates output by target identification model, so as to identify the state of protection pressboard by the image collected at different heights, improve the identification effect and flexibility, and reduce the identification cost.

Description

Technical Field

[0001] This invention relates to the field of substation operation and maintenance technology, specifically to a method, device, and medium for inspecting substation protection pressure plates. Background Technology

[0002] The protection switch is a key connecting element in the substation relay protection system, serving as a bridge between the protection device and the external wiring. The working status of the protection switch directly affects the switching of protection functions and the effectiveness of the action output circuit. Therefore, the protection switch needs to be inspected regularly.

[0003] Existing inspection solutions for protection pressure plates include traditional manual inspection and automatic inspection. Traditional manual inspection is time-consuming, has a high rate of missed and false detections, and poor identification results. Automatic inspection involves installing sensors on the pressure plates or using a dedicated mobile inspection robot to identify the status of each pressure plate in the substation. However, using sensors for pressure plate status identification is costly, and using a mobile inspection robot, which only uses a single camera for identification and data collection, suffers from poor identification results and low flexibility.

[0004] Therefore, the existing solution suffers from drawbacks such as high cost, poor recognition effect and low flexibility in identifying the status of the protective pressure plate. Summary of the Invention

[0005] The technical problem to be solved by the present invention is that the existing solutions have the defects of high cost, poor identification effect and poor flexibility in identifying the status of protection pressure plates. The purpose is to provide a method, device and medium for inspecting substation protection pressure plates, which solves the problems of high cost, poor identification effect and poor flexibility in identifying the status of protection pressure plates in the existing solutions.

[0006] This invention is achieved through the following technical solution:

[0007] Firstly, this application provides a method for inspecting substation protection circuit boards, including:

[0008] Environmental data is collected by LiDAR, and an environmental map is built based on the environmental data. Based on the environmental map and the cabinet coordinate data, the system moves to the target inspection point. The target inspection point is used to indicate the location of the target cabinet. The cabinet coordinate data includes multiple pre-collected cabinet coordinates.

[0009] The target cabinet is captured by multiple image sensors at different preset heights to obtain corresponding raw images. The raw images are then preprocessed to obtain the corresponding target images.

[0010] Each of the target images is input into a pre-trained target recognition model to obtain the recognition result output by the target recognition model. The recognition result includes the recognition status and the coordinates of the pressure plate position. The recognition status is used to indicate whether each pressure plate in the target cabinet is in the input state or the output state.

[0011] In one possible implementation, the step of acquiring environmental data via lidar and creating an environmental map based on the environmental data includes:

[0012] The LiDAR is used to scan the indoor environment of the substation to obtain the corresponding raw point cloud data. The raw point cloud data is then subjected to dynamic object filtering, ground segmentation, and downsampling to obtain the target point cloud data.

[0013] Based on the target point cloud data, a corresponding octree map is generated as the environment map.

[0014] In one possible implementation, the movement to the target inspection point based on the environmental map and cabinet coordinate data includes:

[0015] Based on the environmental map, obtain the initial coordinate information of the current initial position of the mobile patrol platform;

[0016] Among the multiple screen cabinet coordinates in the screen cabinet coordinate data, the screen cabinet coordinates that are closest to the initial coordinate information are taken as the target coordinates, and the screen cabinet corresponding to the target coordinates is taken as the target screen cabinet;

[0017] Based on the initial coordinate information and the target coordinates, a navigation path is established, and based on the navigation path, the mobile inspection platform is controlled to move to the target inspection point, and then the target coordinates are updated.

[0018] In one possible implementation, updating the target coordinates includes:

[0019] Obtain the coordinates of the next unidentified cabinet that is closest to the current location of the mobile inspection platform as the coordinates to be confirmed;

[0020] Obtain a user confirmation signal, which indicates whether the user has moved to the coordinates to be confirmed;

[0021] If the user confirmation signal is yes, the coordinates to be confirmed will be used as the new target coordinates;

[0022] If the user confirmation signal is negative, the coordinates to be confirmed are skipped, and the coordinates of the cabinets closest to the current location of the mobile inspection platform are taken as the new target coordinates from the remaining unidentified cabinet coordinates.

[0023] In one possible implementation, the step of acquiring images of the target cabinet using multiple image sensors at different preset heights to obtain corresponding original images includes:

[0024] A first raw image is acquired using a first image sensor, a second raw image is acquired using a second image sensor, and a third raw image is acquired using a third image sensor.

[0025] The first preset height of the first image sensor is greater than the second preset height of the second image sensor, and the second preset height is greater than the third preset height of the third image sensor.

[0026] In one possible implementation, the data preprocessing of each of the original images to obtain the corresponding target image includes:

[0027] A general preprocessing is performed on the first original image, the second original image, and the third original image to obtain the corresponding first stable image, the second stable image, and the third stable image. The general preprocessing includes electronic image stabilization, white balance calibration, and timestamp alignment.

[0028] Each stable image undergoes dedicated preprocessing to obtain the corresponding first target image, second target image, and third target image.

[0029] In one possible implementation, the specific preprocessing of each stable image to obtain the corresponding first target image, second target image, and third target image includes:

[0030] The first stable image is subjected to perspective pre-correction and shadow removal processing to obtain the first target image that can avoid perspective distortion;

[0031] The second stable image is subjected to text sharpening and color enhancement processing to obtain the second target image that can achieve screen text recognition;

[0032] The third stable image is subjected to ground reflection suppression and edge distortion correction processing to obtain the third target image that can avoid interference from the reflection of the metal pressure plate.

[0033] In one possible implementation, inputting each of the target images into a pre-trained target recognition model includes:

[0034] The corner points of each target image are obtained, and based on each corner point, the perspective transformation of each target image is performed through the homography matrix to obtain the corresponding corrected image;

[0035] Each of the corrected images is input into the target recognition model, wherein the target recognition model is obtained by training the YOLO algorithm.

[0036] Secondly, this application provides a substation protection pressure plate inspection device, comprising:

[0037] The acquisition module is used to collect environmental data through lidar and build an environmental map based on the environmental data. Based on the environmental map and the cabinet coordinate data, it moves to the target inspection point. The target inspection point is used to indicate the location of the target cabinet. The cabinet coordinate data includes multiple pre-collected cabinet coordinates.

[0038] The processing module is used to acquire images of the target cabinet through multiple image sensors at different preset heights to obtain corresponding raw images, and to preprocess each raw image to obtain the corresponding target image;

[0039] The recognition module is used to input each of the target images into a pre-trained target recognition model to obtain the recognition result output by the target recognition model. The recognition result includes the recognition status and the coordinates of the pressure plate position. The recognition status is used to indicate whether each pressure plate in the target cabinet is in the input state or the output state.

[0040] Thirdly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the above-described method for inspecting substation protection pressure plates.

[0041] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0042] This application provides a method, device, and medium for inspecting substation protection pressure plates. It collects environmental data using lidar and establishes an environmental map based on this data. Using the environmental map and the pressure plate coordinates, the device moves to the target inspection point and uses multiple image sensors at different preset heights to acquire images of the target pressure plate. The acquired raw images are then pre-processed, and the images are input into a pre-trained target recognition model to obtain the model's output, which includes the recognition status and pressure plate position coordinates. This allows for status recognition of the protection pressure plates using images acquired at different heights, improving recognition effectiveness and flexibility while reducing recognition costs. Attached Figure Description

[0043] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0044] Figure 1 A flowchart illustrating the substation protection pressure plate inspection method provided in this application embodiment. Figure 1 ;

[0045] Figure 2 This is a schematic diagram of the structure of the mobile inspection platform provided in the embodiments of this application;

[0046] Figure 3 A flowchart illustrating the substation protection pressure plate inspection method provided in this application embodiment. Figure 2 ;

[0047] Figure 4 This is a schematic diagram of the substation protection pressure plate inspection device provided in the embodiments of this application. Detailed Implementation

[0048] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0049] The protection pressure plate acts as a bridge between the protection device and the external wiring, and is directly related to the switching of the protection function and the effectiveness of the action output circuit.

[0050] Protection circuit boards physically connect or disconnect functional circuits, and their status—whether engaged or disengaged—must strictly comply with the requirements of the power grid's operating mode. When the actual status of the protection circuit board differs from the power grid's demand status, false activation or deactivation will occur, potentially causing the protection system to malfunction or fail to operate, leading to widespread power outages or even major power grid accidents such as equipment damage.

[0051] Traditional manual inspection methods face three major challenges: First, the verification workload is large, with a full inspection of a single substation taking 2-4 hours and consuming a large amount of human resources; second, the rate of missed and false inspections is high, with manual visual inspections having an error rate under long-term repetitive work, which has become a major hidden danger to the safe operation of the power grid. It is necessary to develop automatic inspection of pressure plates to solve the hidden dangers of errors in manual verification.

[0052] The existing automatic inspection of protection pressure plates is mainly achieved by adding sensors such as microswitches to the pressure plates. This requires power outages to modify the equipment and lay dedicated lines. The cost of modification per station is high and the construction is complicated. During maintenance, it is susceptible to electromagnetic interference, which can lead to false alarms. It can only monitor the on / off status and cannot identify defects such as loose pressure plates or corrosion. It is not economical and safe.

[0053] Existing automatic inspection solutions for protection pressure plates also include mobile inspection robots, which identify the status of each pressure plate in the substation. However, such mobile inspection robots need to be based on fixed inspection routes and sequences to inspect each protection pressure plate in turn, and cannot change the phased inspection objectives, resulting in poor flexibility. Furthermore, since they only use a single camera for identification and data collection, the identification effect is poor.

[0054] Therefore, this application provides a method for inspecting substation protection pressure plates. Environmental data is collected using lidar, and an environmental map is created based on this data. After moving to the target inspection point, multiple image sensors at different preset heights are used to acquire images of the target pressure plate. The original images are then pre-processed, and the acquired target images are input into a pre-trained target recognition model to obtain the model's output, which includes the recognition status and pressure plate position coordinates. This method uses images acquired at different heights to identify the status of the protection pressure plates, improving recognition effectiveness and flexibility while reducing recognition costs.

[0055] The technical solutions of this application and how they solve the aforementioned technical problems are described in detail below using specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0056] Example 1

[0057] Figure 1 A flowchart illustrating the substation protection pressure plate inspection method provided in this application embodiment. Figure 1 Image recognition unit applied to mobile inspection platforms Figure 2 This is a schematic diagram of the structure of the mobile inspection platform provided in the embodiments of this application, combined with... Figure 1 and Figure 2 As shown, the method includes:

[0058] S101. Collect environmental data using lidar and create an environmental map based on the environmental data.

[0059] Specifically, when scanning the indoor environment of the substation using a 16-line lidar, data from the IMU (Inertial Measurement Unit) and wheeled odometer are simultaneously fused to perform multi-sensor fusion, thereby obtaining the corresponding raw point cloud data. This achieves electromagnetic shielding to ensure stability, and the raw point cloud data undergoes dynamic object filtering, ground segmentation, and downsampling processing to obtain target point cloud data.

[0060] Furthermore, based on the target point cloud data, the target point cloud data is optimized through NDT (Normal Distributions Transform) algorithm for frame matching, loop closure detection, and pose graph to generate a corresponding octree map as a high-precision environment map.

[0061] S102. Based on the environmental map and cabinet coordinate data, move to the target inspection point.

[0062] Specifically, after obtaining the environmental map, based on the high-precision environmental map, the initial coordinate information of the current patrol mobile platform is obtained. Among the multiple screen cabinet coordinates in the screen cabinet coordinate data, the screen cabinet coordinates that are closest to the initial coordinate information are taken as the target coordinates, and the screen cabinet corresponding to the target coordinates is taken as the target screen cabinet.

[0063] Furthermore, based on the initial coordinate information and target coordinates, a navigation path is established using the A* algorithm and dynamic window method. Based on the navigation path, the real-time pose of the inspection mobile platform is controlled, and it moves to the target inspection point. The real-time pose ensures that the inspection mobile platform, i.e. the robot, can accurately stop at the designated position and orientation in front of each inspection point, i.e., each cabinet.

[0064] Specifically, when controlling the motion of the mobile inspection platform, the linear velocity is 0-1.0 m / s, the angular velocity is 0-0.8 rad / s, the PID control docking accuracy is position error ±1 cm, and the heading angle is ±0.5°.

[0065] Furthermore, after controlling the mobile inspection platform to move to the target inspection point and completing the identification of the protective pressure plate status of the current target cabinet, the target coordinates are updated.

[0066] To update the target coordinates, after obtaining the coordinates of the next unidentified cabinet that is closest to the current location of the mobile inspection platform as the coordinates to be confirmed, the user confirmation signal input by the back-end personnel is obtained. The user confirmation signal is used to indicate whether to move to the coordinates to be confirmed.

[0067] If the user confirms the signal, the coordinates to be confirmed are used as the new target coordinates. If the user confirms the signal, the coordinates to be confirmed are skipped, and among the remaining unidentified cabinet coordinates, the coordinates of the cabinet closest to the current location of the mobile inspection platform are selected as the new target coordinates. Based on the new target coordinates, the mobile inspection platform is controlled to move to the corresponding inspection point. This changes the phased inspection target and improves the flexibility of the inspection.

[0068] S103. Using multiple image sensors at different preset heights, images of the target cabinet are acquired to obtain the corresponding original images.

[0069] Specifically, such as Figure 2 As shown, after the mobile inspection platform moves to the target inspection point, it acquires images through a vertically fixed camera pan-tilt unit, which includes a first image sensor, a second image sensor, and a third image sensor.

[0070] A first raw image is acquired using a high-angle first image sensor, such as a camera; a second raw image is acquired using a mid-angle second image sensor; and a third raw image is acquired using a low-angle third image sensor.

[0071] The first preset height of the first image sensor is greater than the second preset height of the second image sensor, and the second preset height is greater than the third preset height of the third image sensor.

[0072] Furthermore, the coverage height range of the first image sensor at a high position is 1.5m to 1.8m, with a slightly downward shooting angle. The target area of ​​the cabinet that the first image sensor focuses on is the upper protection area of ​​the cabinet. The coverage height range of the second image sensor at a mid-position is 1.2m to 1.5m, with a horizontal shooting angle. The target area of ​​the cabinet that the second image sensor focuses on is the middle protection area of ​​the cabinet. The coverage height range of the third image sensor at a low position is 0.9m to 1.2m, with a slightly upward shooting angle. The target area of ​​the cabinet that the third image sensor focuses on is the lower pressure plate area of ​​the cabinet.

[0073] Furthermore, after the mobile inspection platform, i.e. the robot, stops moving, a 1.2-second shooting delay is set to eliminate residual mechanical vibration and ensure that the original images acquired are clear and stable.

[0074] S104. Perform data preprocessing on each of the original images to obtain the corresponding target image.

[0075] Specifically, after acquiring the corresponding raw images through image sensors at different heights, the raw images are first uniformly preprocessed, and then targeted special preprocessing is performed on each raw image after the general preprocessing to obtain the first target image, the second target image, and the third target image after different preprocessing.

[0076] S105. Input each of the target images into the pre-trained target recognition model to obtain the recognition result output by the target recognition model.

[0077] Specifically, after acquiring the first target image, the second target image, and the third target image, the corner points of each target image are obtained. If the number of corner points in each target image is four, the tilted viewing angle of each target image is projected onto a standard frontal view based on each corner point using a homography matrix to perform perspective transformation correction, thereby obtaining a corrected image that eliminates the deformation caused by the shooting angle. If the number of corner points in each target image is not four, the corner points of the corresponding target image are predicted and obtained using a backup corner point prediction algorithm, and the corresponding corrected image is obtained based on the predicted corner points.

[0078] Furthermore, each corrected image is input into the target recognition model to obtain the recognition result output by the target recognition model. The recognition result includes the recognition status and the coordinates of the pressure plate position. The recognition status is used to indicate whether each pressure plate in the target cabinet is in the input or output state, and the pressure plate position coordinates are indicated by the output bounding box.

[0079] Based on the corrected image, preprocessing such as size scaling, pixel normalization, and contrast enhancement is performed. Then, a target recognition model driven by a lightweight YOLOv8n algorithm is used to detect and locate all pressure plates, output the bounding box coordinates of the pressure plates, and identify the status of each protective pressure plate.

[0080] Furthermore, to identify the status of each protective pressure plate, the pressure plate area is cropped by 10% based on the positioning coordinates to extract the ROI (Region of Interest). Image segmentation is performed using methods such as adaptive threshold binarization and morphological closing operation to fill holes. The Zhang-Suen algorithm is used to extract the center line of the rotation axis with a single pixel width. For each pressure plate that is located, the angle of its rotation axis is analyzed.

[0081] Specifically, when the rotation angle of the rotating shaft is within the range of 80° to 90°, the identification status of the protective pressure plate is in the engaged state; when the rotation angle of the rotating shaft is within the range of -5° to 5°, the identification status of the protective pressure plate is in the disengaged state.

[0082] Furthermore, the target recognition model is trained using the YOLO algorithm. The target recognition model is trained on a dedicated dataset containing 5,000 images of substation pressure plates. The dedicated dataset covers common on-site interference scenarios such as blurring, reflection, and occlusion, ensuring the robustness of the target recognition model.

[0083] This application provides a method for inspecting substation protection pressure plates. It collects environmental data using lidar and creates an environmental map based on this data. Using the environmental map and the pressure plate coordinates, the system moves to the target inspection point and uses multiple image sensors at different preset heights to capture images of the target pressure plate. The system obtains the corresponding raw images and performs data preprocessing. The acquired target images are then input into a pre-trained target recognition model to obtain the model's output, which includes the recognition status and pressure plate position coordinates. This method uses images collected at different heights to identify the status of the protection pressure plates, improving recognition effectiveness and flexibility while reducing recognition costs.

[0084] like Figure 2 As shown, the mobile inspection platform includes an image sensor, namely a camera fixed gimbal, a chassis unit, and an image recognition unit. The camera fixed gimbal is located on the upper part of the mobile inspection platform, the chassis unit is located below the camera fixed gimbal, and the image recognition unit is located inside the chassis unit and is mechanically fixed to the chassis unit to avoid vibration during the movement from affecting the processing stability.

[0085] For the fixed camera pan-tilt unit, a vertically distributed three-dimensional visual acquisition area is formed. The multi-height camera achieves all-round visual coverage and solves the problem of blind spots from a single viewpoint. The fixed camera pan-tilt unit includes three 2-megapixel industrial cameras, which are vertically mounted on fixed brackets at heights of 1.2m, 1.5m, and 1.8m, respectively, covering the inspection range from different vertical perspectives.

[0086] For the chassis unit, as the robot's mobile carrier, the chassis unit achieves high-precision autonomous mobile positioning and path planning through a combination of LiDAR + UWB + differential drive, ensuring the robot's stable operation in complex inspection environments such as factory workshops and substations. The chassis unit includes a 2D LiDAR, two differential drive wheels, and a UWB positioning module. The LiDAR is installed at the front of the chassis for environmental obstacle detection, map building, and real-time obstacle avoidance. The differential drive wheels are symmetrically distributed on both sides of the chassis, achieving steering through speed differences, and working with a high-precision encoder to provide feedback on motion status and improve movement accuracy. The UWB positioning module is integrated in the center of the chassis, achieving high-precision indoor and outdoor positioning through ultra-wideband wireless technology, and optimizing the navigation path by fusing data with LiDAR.

[0087] The image recognition unit is equipped with a lightweight YOLOv8 algorithm, which is adapted to edge computing scenarios to achieve efficient and real-time image analysis. It can complete real-time image recognition under low power consumption conditions, meeting the dual requirements of response speed and computing power in inspection scenarios. The image recognition unit includes a controller, which is responsible for implementing the above-mentioned substation protection pressure plate inspection method and uploading the recognition results to the cloud platform or local storage.

[0088] Example 2

[0089] Based on the above embodiment one, step S103 will be further explained. Figure 3 A flowchart illustrating the substation protection pressure plate inspection method provided in this application embodiment. Figure 2 ,like Figure 3 As shown, the method includes:

[0090] S201. Perform general preprocessing on the first original image, the second original image, and the third original image to obtain the corresponding stable image.

[0091] Specifically, a general preprocessing is performed on the first original image, the second original image, and the third original image to obtain the corresponding first stable image, the second stable image, and the third stable image. The general preprocessing includes electronic image stabilization, white balance calibration, and timestamp alignment.

[0092] S202. Perform dedicated preprocessing on each stable image to obtain the corresponding first target image, second target image and third target image.

[0093] Specifically, the first stable image undergoes perspective pre-correction and shadow removal processing to obtain a first target image that avoids perspective distortion; the second stable image undergoes text sharpening and color enhancement processing to obtain a second target image that enables screen text recognition; and the third stable image undergoes ground reflection suppression and edge distortion correction processing to obtain a third target image that avoids interference from metal pressure plate reflections.

[0094] This application provides a method for inspecting substation protection pressure plates. It collects environmental data using lidar and creates an environmental map based on this data. Using the environmental map and the pressure plate coordinates, the system moves to the target inspection point and uses multiple image sensors at different preset heights to capture images of the target pressure plate. The system obtains the corresponding raw images and performs data preprocessing. The acquired target images are then input into a pre-trained target recognition model to obtain the model's output, which includes the recognition status and pressure plate position coordinates. This method uses images collected at different heights to identify the status of the protection pressure plates, improving recognition effectiveness and flexibility while reducing recognition costs.

[0095] In summary, the substation protection pressure plate inspection method provided in this application, applied to a mobile inspection platform, i.e., an inspection robot, comprehensively covers protection pressure plates at different heights of the cabinet through three vertically fixed cameras; it uses LiDAR to scan the environment in real time and autonomously construct a high-precision map with a positioning accuracy of ±1cm, without any auxiliary markings; the collected images are automatically analyzed by the optimized YOLOv8 algorithm, which can accurately identify the pressure plate's engagement / disengagement status.

[0096] In specific deployments, the mobile inspection platform automatically enters mapping mode after powering on, autonomously moves and scans the relay protection room environment, and completes high-precision map construction within 5 minutes; tasks can be started with one click via the backend system or the button on the robot body; the robot autonomously navigates to each protection cabinet, and three vertically fixed cameras automatically capture images from different angles; the backend displays the pressure plate status analysis results and abnormal alarms in real time; and it supports querying historical data by time, cabinet number, and other dimensions.

[0097] Therefore, this invention achieves full-coverage, blind-spot-free inspection through an innovative architecture of a three-camera fixed gimbal + laser navigation + lightweight YOLOv8: the three cameras are vertically distributed (1.2m / 1.5m / 1.8m), covering the pressure plate and protection device with a full height range of 0.9-1.8m, eliminating the visual blind spot of >35% in traditional single-view solutions; centimeter-level precise positioning: laser SLAM navigation positioning accuracy ±1cm (better than magnetic navigation ±10cm), compensation ensures that the vertical angle deviation between the camera and the cabinet is <0.5°, and the perspective distortion rate is reduced to <2% (15-25% in traditional existing solutions); power-off-free construction: no need to modify the pressure plate body as in sensor solutions.

[0098] Figure 4 This is a schematic diagram of the substation protection pressure plate inspection device provided in the embodiments of this application, as shown below. Figure 4 As shown, the device 400 includes:

[0099] The acquisition module 401 is used to collect environmental data through lidar and build an environmental map based on the environmental data. Based on the environmental map and the cabinet coordinate data, it moves to the target inspection point. The target inspection point is used to indicate the location of the target cabinet. The cabinet coordinate data includes multiple pre-collected cabinet coordinates.

[0100] Processing module 402 is used to acquire images of the target cabinet through multiple image sensors of different preset heights to obtain corresponding original images, and to perform data preprocessing on each of the original images to obtain corresponding target images;

[0101] The recognition module 403 is used to input each of the target images into a pre-trained target recognition model to obtain the recognition result output by the target recognition model. The recognition result includes the recognition status and the coordinates of the pressure plate position. The recognition status is used to indicate whether each pressure plate in the target cabinet is in the input state or the output state.

[0102] Furthermore, the acquisition module 401 is specifically used to scan the indoor environment of the substation using the lidar to obtain the corresponding raw point cloud data, and to perform dynamic object filtering, ground segmentation and downsampling processing on the raw point cloud data to obtain target point cloud data.

[0103] Based on the target point cloud data, a corresponding octree map is generated as the environment map.

[0104] Furthermore, the acquisition module 401 is specifically used to acquire the initial coordinate information of the current initial position of the mobile patrol platform based on the environmental map;

[0105] Among the multiple screen cabinet coordinates in the screen cabinet coordinate data, the screen cabinet coordinates that are closest to the initial coordinate information are taken as the target coordinates, and the screen cabinet corresponding to the target coordinates is taken as the target screen cabinet;

[0106] Based on the initial coordinate information and the target coordinates, a navigation path is established, and based on the navigation path, the mobile inspection platform is controlled to move to the target inspection point, and then the target coordinates are updated.

[0107] Furthermore, the acquisition module 401 is specifically used to acquire the coordinates of the next unidentified cabinet that is closest to the current location of the mobile inspection platform as the coordinates to be confirmed;

[0108] Obtain a user confirmation signal, which indicates whether the user has moved to the coordinates to be confirmed;

[0109] If the user confirmation signal is yes, the coordinates to be confirmed will be used as the new target coordinates;

[0110] If the user confirmation signal is negative, the coordinates to be confirmed are skipped, and the coordinates of the cabinets closest to the current location of the mobile inspection platform are taken as the new target coordinates from the remaining unidentified cabinet coordinates.

[0111] Furthermore, the processing module 402 is specifically used to acquire a first original image through a first image sensor, acquire a second original image through a second image sensor, and acquire a third original image through a third image sensor;

[0112] The first preset height of the first image sensor is greater than the second preset height of the second image sensor, and the second preset height is greater than the third preset height of the third image sensor.

[0113] Further, the processing module 402 is specifically used to perform general preprocessing on the first original image, the second original image and the third original image to obtain the corresponding first stable image, the second stable image and the third stable image, wherein the general preprocessing includes electronic image stabilization, white balance calibration and timestamp alignment;

[0114] Each stable image undergoes dedicated preprocessing to obtain the corresponding first target image, second target image, and third target image.

[0115] Further, the processing module 402 is specifically used to perform perspective pre-correction and shadow removal processing on the first stable image to obtain the first target image that can avoid perspective distortion;

[0116] The second stable image is subjected to text sharpening and color enhancement processing to obtain the second target image that can achieve screen text recognition;

[0117] The third stable image is subjected to ground reflection suppression and edge distortion correction processing to obtain the third target image that can avoid interference from the reflection of the metal pressure plate.

[0118] Furthermore, the recognition module 403 is specifically used to acquire the corner points of each target image, and based on each corner point, to perform perspective transformation correction on each target image through the homography matrix to obtain the corresponding corrected image;

[0119] Each of the corrected images is input into the target recognition model, wherein the target recognition model is obtained by training the YOLO algorithm.

[0120] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the substation protection pressure plate inspection method described above.

[0121] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for inspecting substation protection pressure plates, characterized in that, The method is applied to a mobile inspection platform and includes: Environmental data is collected by lidar, and an environmental map is built based on the environmental data. Based on the environmental map and the cabinet coordinate data, the system moves to the target inspection point. The target inspection point is used to indicate the location of the target cabinet. The cabinet coordinate data includes multiple pre-collected cabinet coordinates. The target cabinet is imaged by multiple image sensors at different preset heights to obtain corresponding raw images. The raw images are then preprocessed to obtain the corresponding target images. Each of the target images is input into a pre-trained target recognition model to obtain the recognition result output by the target recognition model. The recognition result includes the recognition status and the coordinates of the pressure plate position. The recognition status is used to indicate whether each pressure plate in the target cabinet is in the input state or the output state. The step of acquiring images of the target cabinet using multiple image sensors at different preset heights to obtain corresponding original images includes: acquiring a first original image using a first image sensor, acquiring a second original image using a second image sensor, and acquiring a third original image using a third image sensor; wherein the first preset height of the first image sensor is greater than the second preset height of the second image sensor, and the second preset height is greater than the third preset height of the third image sensor. The step of preprocessing each of the original images to obtain the corresponding target image includes: The first original image, the second original image, and the third original image are subjected to general preprocessing to obtain the corresponding first stable image, the second stable image, and the third stable image. The general preprocessing includes electronic image stabilization, white balance calibration, and timestamp alignment. Each stable image is subjected to dedicated preprocessing to obtain the corresponding first target image, the second target image, and the third target image. The specific preprocessing of each stable image to obtain the corresponding first target image, second target image, and third target image includes: performing perspective pre-correction and shadow elimination processing on the first stable image to obtain a first target image that avoids perspective distortion; performing text sharpening and color enhancement processing on the second stable image to obtain a second target image that enables screen text recognition; and performing ground reflection suppression and edge distortion correction processing on the third stable image to obtain a third target image that avoids interference from metal pressure plate reflections.

2. The substation protection pressure plate inspection method according to claim 1, characterized in that, The step of collecting environmental data via lidar and creating an environmental map based on the environmental data includes: The laser radar is used to scan the indoor environment of the substation to obtain the corresponding raw point cloud data. The raw point cloud data is then subjected to dynamic object filtering, ground segmentation, and downsampling to obtain the target point cloud data. Based on the target point cloud data, a corresponding octree map is generated as the environment map.

3. The substation protection pressure plate inspection method according to claim 1, characterized in that, The process of moving to the target inspection point based on the environmental map and cabinet coordinate data includes: Based on the environmental map, obtain the initial coordinate information of the current initial position of the mobile patrol platform; Among the multiple screen cabinet coordinates in the screen cabinet coordinate data, the screen cabinet coordinates that are closest to the initial coordinate information are taken as the target coordinates, and the screen cabinet corresponding to the target coordinates is taken as the target screen cabinet; Based on the initial coordinate information and the target coordinates, a navigation path is established, and based on the navigation path, the mobile inspection platform is controlled to move to the target inspection point, and then the target coordinates are updated.

4. The substation protection pressure plate inspection method according to claim 3, characterized in that, Updating the target coordinates includes: Obtain the coordinates of the next unidentified cabinet that is closest to the current location of the mobile inspection platform as the coordinates to be confirmed; Obtain a user confirmation signal, which indicates whether the user has moved to the coordinates to be confirmed; If the user confirmation signal is yes, the coordinates to be confirmed will be used as the new target coordinates; If the user confirmation signal is negative, the coordinates to be confirmed are skipped, and the coordinates of the cabinets closest to the current location of the mobile inspection platform are selected as the new target coordinates from the remaining unidentified cabinet coordinates.

5. The substation protection pressure plate inspection method according to claim 1, characterized in that, The step of inputting each of the target images into the pre-trained target recognition model includes: The corner points of each target image are obtained, and based on each corner point, the perspective transformation of each target image is performed through the homography matrix to obtain the corresponding corrected image; Each of the corrected images is input into the target recognition model, wherein the target recognition model is obtained by training the YOLO algorithm.

6. A substation protection pressure plate inspection device, characterized in that, The device is applied in the substation protection pressure plate inspection method according to any one of claims 1-5, comprising: The acquisition module is used to collect environmental data through lidar and build an environmental map based on the environmental data. Based on the environmental map and the cabinet coordinate data, it moves to the target inspection point. The target inspection point is used to indicate the location of the target cabinet. The cabinet coordinate data includes multiple pre-collected cabinet coordinates. The processing module is used to acquire images of the target cabinet through multiple image sensors at different preset heights to obtain corresponding raw images, and to perform data preprocessing on each raw image to obtain the corresponding target image; The recognition module is used to input each of the target images into a pre-trained target recognition model to obtain the recognition result output by the target recognition model. The recognition result includes the recognition status and the coordinates of the pressure plate position. The recognition status is used to indicate whether each pressure plate in the target cabinet is in the input state or the output state.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 5.

Images