Transformer substation screen cabinet pressing plate state intelligent checking method based on machine vision

Abstract

The invention belongs to the technical field of substation secondary equipment state checking, and particularly relates to a substation screen cabinet pressing plate state intelligent checking method based on machine vision. Comprising the following steps: S1, acquiring and marking image data of a protection pressing plate, and establishing a pressing plate data set; s2, improving a YOLO V5 algorithm based on a protection pressing plate state detection requirement; s3, carrying out the logic and layout design of the front-end interaction page of the pressing plate state checking application program; s4, the server receives the pressing plate image from the front end, an improved YOLO V5 pressing plate detection algorithm is executed through a built-in interface, and a detection result is returned to the front end and stored in a pressing plate database; according to the transformer substation screen cabinet pressing plate state intelligent checking method based on machine vision, relay protection pressing plate information is automatically checked through artificial intelligence image recognition, the field pressing plate checking accuracy is improved, the secondary equipment state checking time is remarkably shortened, and the operation and maintenance efficiency is improved.

Description

Technical Field

[0001] This invention belongs to the technical field of substation secondary equipment status verification, specifically involving a machine vision-based intelligent verification method for the status of substation panel pressure plates. Background Technology

[0002] With the accelerated development of power grids, the number and scale of substations are constantly increasing, leading to a corresponding increase in the number of secondary equipment within them. Relay protection devices, as crucial secondary equipment reflecting the operating status of primary equipment within substations, play a vital role in the safe and stable operation of the power system. Given the large number of secondary equipment within substations, related inspection work involves verifying the information and status of numerous circuit boards. On-site verification often relies on manual checks by maintenance personnel, and any human error can pose significant risks to the safe operation of the power grid.

[0003] Substation secondary equipment status verification includes key aspects such as protection device settings, operating setting zones, the activation and deactivation of soft and hard switchboards, and the operating status of the station's AC / DC auxiliary power supply system. Based on the content of the substation secondary equipment status verification work, the entire verification process can be divided into setting verification, setting zone number checking, switchboard position checking, device alarm checking, and communication status checking. Daily statistics show that switchboard position checking accounts for a large proportion of the time spent in each stage of secondary equipment inspection, and is a significant factor contributing to the long verification time for substation secondary equipment.

[0004] Based on the workflow for checking the status and position of pressure plates, the verification process can be divided into several aspects: opening cabinet doors station by station and screen by screen, checking the status of pressure plates in and out, comparing with the correct position, and recording the results. The steps of checking the status of pressure plates in and out and comparing with the correct position are the key factors affecting the long verification time for the status of secondary equipment in substations. Shortening the time for checking and comparing the status and position of pressure plates can greatly improve the efficiency of on-site secondary equipment status verification. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a machine vision-based intelligent verification method for the status of substation panel switchboards. By using artificial intelligence image recognition, the method automatically verifies the relay protection switchboard information, improves the accuracy of on-site switchboard verification, significantly shortens the time for secondary equipment status verification, and improves operation and maintenance efficiency.

[0006] The objective of this invention is achieved as follows: a machine vision-based intelligent verification method for the status of substation panel pressure plates, comprising the following steps: S1. Collect and annotate image data of the protective pressure plate, and establish a pressure plate dataset; S2. Improve the YOLO V5 algorithm based on the requirements for detecting the status of the protective pressure plate; S3. Perform the logic and layout design of the application's front-end interactive page to check the pressure plate status; S4. The server receives the pressure plate image from the front end, executes the improved YOLO V5 pressure plate detection algorithm through the built-in interface, returns the detection result to the front end, and saves it into the pressure plate database.

[0007] Furthermore, the protective pressure plate in step S1 includes two types: pressure plate type and continuous plate type. The image data acquisition of the protective pressure plate includes taking pictures of each type of protective pressure plate under various working conditions at different angles and with different light intensities. The image data annotation of the protective pressure plate includes inputting the annotation information of the captured protective pressure plate pictures along with the pictures into the network during network model training.

[0008] Furthermore, the annotation information of the protective pressure plate image includes the category ID of the target protective pressure plate and the coordinate information of the four corner points of the bounding box. According to the different types of protective pressure plates and their working states, the protective pressure plates are divided into 5 categories, including pressure plate-in, pressure plate-out, continuous plate-in, continuous plate-standby, and continuous plate-out.

[0009] Furthermore, step S2, which improves the YOLO V5 algorithm based on the requirements for detecting the state of the protective pressure plate, includes: Improvements were made to the YOLO V5 algorithm for small object detection and for single-label non-overlapping objects. The improvement for small object detection addresses the limitation of increasing image resolution during image data acquisition for protective pressure plates by enhancing feature extraction and reducing feature loss, thus improving small object detection performance. The improvement for single-label non-overlapping objects addresses the inconsistency between the design of the YOLO V5 object detection algorithm for multi-label multi-classification and occlusion problems and the actual characteristics of pressure plate detection by modifying the non-maximum suppression algorithm.

[0010] Furthermore, in step S3, the layout of the front-end interactive page of the pressure plate status verification application is designed by writing WXXL templates and WXSS style files to design the page appearance of the application. The front-end interactive page consists of four parts: the main page, the pressure plate preview page, the QR code preview page, and the result feedback page. The main page has two container views with different icons: an input image icon and an input QR code icon. It also has a "Start Detection" button and a general navigation bar at the bottom, which provides an interface for jumping between the main page and the results feedback page.

[0011] Furthermore, in step S3, the logic design of the front-end interactive page of the pressure plate status verification application utilizes the built-in interface of existing software and programs to implement JavaScript script functions. The user starts the image upload function through the input image icon on the main page, and the main page executes the jump logic to jump to the pressure plate preview page. The platen preview page calls relevant program interfaces such as wx.chooseImage to retrieve the platen image from the local device and displays the selected image for user confirmation. After user confirmation, the platen status verification application front end uses relevant program interfaces such as wx.uploadFile to upload the image to the server for platen target detection. The server returns the platen status, location, and binary information of the image marked with the target platen, which is then sent to the main page, and the user is redirected back to the main page.

[0012] Furthermore, the detection results in step S4 are saved in txt text format, and the results include the pressure plate category, the horizontal and vertical coordinates of the center point of the pressure plate prediction frame, and the width and height of the pressure plate prediction frame.

[0013] Furthermore, the pressure plate database in step S4 stores the pressure plate dataset established in step S1, and relevant prior information of the pressure plate is obtained from the pressure plate database when performing the protection pressure plate status verification work. The pressure plate database uses the screen cabinet as the basic unit to store and manage the pressure plate data. Then, all screen cabinets are serialized, and each screen cabinet has only one unique code. The relevant prior information is associated through the screen cabinet code. Finally, the screen cabinet code is converted into a QR code and pasted on the screen cabinet. By simply scanning the QR code, the screen cabinet code can be automatically obtained. This brings great convenience from both on-site practice and management perspectives.

[0014] Furthermore, in step S4, when performing the status verification of the protective pressure plates, the coordinate information of each protective pressure plate is used to determine its position in the matrix. Then, based on this, a one-to-one mapping relationship between the protective pressure plates and the pressure plate database is established, and the status verification function of the pressure plates is finally realized.

[0015] Further, in step S4, determining the position of each protective pressure plate in the matrix based on its coordinate information includes: S41. Verify the number of detected pressure plates by combining the database information to determine whether there are any missed detections. S42. Use perspective transformation to correct visual distortions in the image; S43. Locate the position of the current protective pressure plate in the matrix arrangement based on the coordinates of the remaining pressure plates; S44. Generate and output the result object. The object attributes include the pressure plate name, pressure plate position (position:(i,j)), pressure plate preset state, and comparison result.

[0016] The beneficial effects of this invention are as follows: The intelligent verification method for the status of substation panel pressure plates based on machine vision includes the following steps: S1, acquiring and annotating image data of the protective pressure plates to establish a pressure plate dataset; S2, improving the YOLO V5 algorithm based on the requirements for protective pressure plate status detection; S3, designing the logic and layout of the front-end interactive page of the pressure plate status verification application; S4, the server receives the pressure plate images from the front end, executes the improved YOLO V5 pressure plate detection algorithm through the built-in interface, returns the detection results to the front end, and saves them into the pressure plate database. Through the above steps, the intelligent verification method for the status of substation panel pressure plates based on machine vision of this invention constructs and annotates a dataset by acquiring images covering different shooting angles, lighting, shadows, and pressure plate types, and then improves the YOLO V5 algorithm. The algorithm addresses the issues of missed or incorrect detections and overlapping detection frames in small target detection on pressure plates. It also designs a front-end program containing modules such as a homepage and a pressure plate preview page, along with a corresponding back-end system. A database is built with the cabinet as the unit and associated with the prior information of the pressure plates. By correcting image distortion through perspective transformation and locating the position of the pressure plate matrix through gridded calculation, the algorithm enables automatic identification and verification of the pressure plate status, improving the accuracy of on-site pressure plate verification, significantly shortening the time for secondary equipment status verification, and improving operation and maintenance efficiency. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a machine vision-based intelligent verification method for the status of substation panel pressure plates.

[0019] Figure 2 This is a schematic diagram illustrating the operational logic relationships within and between pages in an embodiment of the present invention.

[0020] Figure 3 This is a schematic diagram of the system interaction of the pressure plate status detection application in an embodiment of the present invention.

[0021] Figure 4 This is a schematic diagram illustrating the specific execution logic for obtaining the pressure plate status and position information in an embodiment of the present invention.

[0022] Figure 5This is a schematic diagram illustrating the implementation process of the pressure plate status verification function in an embodiment of the present invention.

[0023] Figure 6 This is a schematic diagram of the matrix arrangement process of the pressure plates in an embodiment of the present invention.

[0024] Figure 7 This is a schematic diagram showing how the coordinates of the four vertex pressure plates of the distortion group are obtained in an embodiment of the present invention.

[0025] Figure 8 This is a schematic diagram of gridded computing in an embodiment of the present invention. Detailed Implementation

[0026] The present invention will now be further described with reference to the accompanying drawings.

[0027] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] It should be noted that in the embodiments of the present invention, all directional indications (such as up-down-left-right-forward-backward...) are only used to explain the relative positional relationship and movement between the components in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly. The connection can be a direct connection or an indirect connection.

[0030] like Figure 1 As shown, the intelligent verification method for the status of substation panel pressure plates based on machine vision of the present invention includes the following steps: S1. Collect and annotate image data of the protective pressure plate, and establish a pressure plate dataset; S2. Improve the YOLO V5 algorithm based on the requirements for detecting the status of the protective pressure plate; S3. Perform the logic and layout design of the application's front-end interactive page to check the pressure plate status; S4. The server receives the pressure plate image from the front end, executes the improved YOLO V5 pressure plate detection algorithm through the built-in interface, returns the detection result to the front end, and saves it into the pressure plate database.

[0031] To make the samples in the pressure plate dataset more representative, this invention considers the following four environmental factors that may affect the accuracy of pressure plate recognition: shooting angle, light changes, shadow and reflection superposition, and pressure plate type.

[0032] When manually photographing the protective pressure plates, it's impossible to guarantee that the image is always taken from the front; in actual operation, certain shooting angles will occur. In an image taken from above, the first row of pressure plates will appear significantly larger than the last row; in images taken from the side, one side of the switch will appear larger than the other. These different shooting angles cause perspective distortion in the images.

[0033] Considering that the intensity of light on site is often unpredictable, daytime lighting conditions are generally good, resulting in brighter overall images. Conversely, on cloudy days or under artificial lighting, the overall image will be darker. Therefore, when capturing images of the pressure plate, it is essential to consider different scenarios, including suitable lighting, strong lighting, and dim lighting.

[0034] Uneven brightness in images is not simply a matter of overall brightening or darkening. When shadows obscure the image, some pressure plates may appear very bright, while others are significantly darker. Some pressure plate cabinets also have glass partitions; the glass reflects the images of objects in the scene, and when photographed manually, human figures can also be reflected on the glass. These reflections of people and their surroundings superimposed on the pressure plate photos can significantly interfere with detection and recognition algorithms.

[0035] Furthermore, in one embodiment, the protective pressure plate in step S1 includes two types: pressure plate type and continuous plate type. The image data acquisition of the protective pressure plate includes taking pictures of each type of protective pressure plate at different angles and with different light intensities under various working conditions. The image data annotation of the protective pressure plate includes inputting the annotation information of the captured protective pressure plate pictures along with the pictures into the network during network model training.

[0036] Furthermore, in one embodiment, the annotation information of the protective pressure plate image includes the category ID of the target protective pressure plate and the coordinate information of the four corner points of the bounding box. According to the different types of protective pressure plates and their working states, the protective pressure plates are divided into 5 categories, including pressure plate-in, pressure plate-out, continuous plate-in, continuous plate-standby, and continuous plate-out.

[0037] In object detection, there are two ways to define small objects: a) based on absolute size, objects smaller than 32*32 pixels are considered small objects; b) based on relative size, objects smaller than 0.12% of the original image size are considered small objects.

[0038] Due to the limitations of the YOLO V5 object detection algorithm on the input image size, after adaptive scaling, the size of a single pressure plate in the protective plate image often falls below the reference threshold for small targets, especially when encountering images with densely packed pressure plates, such as those of main transformer protection devices. For example, a protective pressure plate image originally 2520 pixels * 3360 pixels becomes 480 pixels * 640 pixels after scaling. At this point, the pixel size of a single pressure plate is approximately 22 pixels * 34 pixels, which is clearly smaller than the reference thresholds for the two small target definitions mentioned above. Therefore, the protective pressure plate detection problem also falls under the category of small target object detection.

[0039] The original YOLO V5 object detection algorithm was used to predict large, medium, and small targets using feature maps of different scales: 80*80, 40*40, and 20*20. The 80*80 feature map was the largest of the three, roughly revealing the shape and outline of the pressure plate and containing a significant amount of shallow feature information. In the 40*40 feature map, the network-extracted features were further abstracted, enhancing representativeness and making the contrast between the pressure plate target and the background more pronounced. The 20*20 feature map, obtained after depthwise convolution, was clearly more suitable for large target images; a single pressure plate object was difficult to discern with the naked eye, while the overall regional outline of the pressure plate in its "retired" state was more prominent. However, significant shallow feature information was lost; the pressure plate area in its "engaged" state was almost completely integrated into the background, making it difficult to distinguish.

[0040] Analysis revealed the following three reasons for the poor performance in small target detection: (1) Small targets have less feature information Because small target objects generally have low resolution and a small pixel area ratio, the feature information that can be extracted from them is not sufficient, and they are easily affected by noise in the image, which leads to the network's inability to accurately locate them.

[0041] (2) Loss of key information in deep feature maps caused by multi-layer convolution As the network depth increases, the size of the output feature map will continuously shrink after multiple convolution operations. This causes the downsampling stride of subsequent convolution operations to be larger than the size of the small target, ultimately resulting in the loss of small target feature information contained in the feature map during the downward transmission process, i.e., feature loss.

[0042] (3) Incomplete and insufficient extraction of image feature information In the original YOLO V5 object detection algorithm, feature extraction mainly relies on convolution operations. To capture as much edge detail and other feature information as possible, YOLO V5 primarily uses 3x3 and 1x1 convolution kernels. As a trade-off, the network lacks a larger receptive field during feature extraction, resulting in image feature information being limited to a small area and failing to establish connections between broader feature information. Furthermore, the extracted feature information lacks data correlation across different dimensions such as channels and space. These uncorrelated, localized, small-scale feature information are easily lost during depthwise convolution, ultimately leading to incomplete and insufficient extracted feature information. Therefore, to improve small object detection performance without increasing image resolution, this invention will improve the algorithm by increasing feature extraction and reducing feature loss.

[0043] The original YOLO V5 algorithm, as a general-purpose object detection algorithm, was designed to handle object detection tasks in various complex scenarios, including multi-label multi-classification (an object may belong to multiple different categories simultaneously) and occlusion problems. For example, in a traffic scenario: 1) A target's label category may simultaneously belong to both motor vehicles and cars; 2) When observing two vehicles driving side by side from one angle, there is partial overlap and obstruction.

[0044] In response to the above situation, YOLO V5 has adopted corresponding technical measures to cope with multi-label and multi-classification scenarios and overcome the problems of target overlap and occlusion.

[0045] However, in actual detection scenarios for protective pressure plates, the pressure plates, as the objects being detected, do not present issues of multiple labels, multiple classifications, or occlusion. Furthermore, pressure plate detection is clearly a single-label, multi-classification problem, meaning a pressure plate target can only belong to one label category. In addition, the distribution and arrangement of pressure plates on the protective cabinet are very regular: they are evenly distributed on the same spatial plane and arranged at equal intervals, thus eliminating the possibility of local overlap or occlusion. Therefore, overlapping detection boxes in pressure plate detection scenarios are unreasonable. Consequently, the original YOLO V5 object detection algorithm's design for addressing multiple labels, multiple classifications, and occlusion issues is inconsistent with the actual characteristics of pressure plate detection.

[0046] Furthermore, in one embodiment, step S2, which improves the YOLO V5 algorithm based on the requirement for detecting the state of the protective pressure plate, includes: Improvements were made to the YOLO V5 algorithm for small object detection and for single-label non-overlapping objects. The improvement for small object detection addresses the limitation of increasing image resolution during image data acquisition for protective pressure plates by enhancing feature extraction and reducing feature loss, thus improving small object detection performance. The improvement for single-label non-overlapping objects addresses the inconsistency between the design of the YOLO V5 object detection algorithm for multi-label multi-classification and occlusion problems and the actual characteristics of pressure plate detection by modifying the non-maximum suppression algorithm.

[0047] Furthermore, in one embodiment, the layout of the front-end interactive page of the pressure plate status verification application in step S3 is designed by writing WXXL templates and WXSS style files to design the page appearance of the application. The front-end interactive page consists of four parts: the main page, the pressure plate preview page, the QR code preview page, and the result feedback page. The main page has two container views with different icons: an input image icon and an input QR code icon. It also has a "Start Detection" button and a general navigation bar at the bottom, which provides an interface for jumping between the main page and the results feedback page.

[0048] The front-end terminal in this invention is equipped with a Snapdragon 6300 5G processor, supporting dynamic memory expansion technology of 8GB (+8GB) + 256GB, optimized storage, and can easily handle daily use and high-performance games. It also features the latest Android 14 system, offering greater personalization, security, and operability. Employing 5G dual-mode, dual-carrier aggregation, and 5G intelligent networking technology, it boasts broader and higher outdoor networking capabilities. In terms of hardware, it can be equipped with a 6.78-inch 120Hz refresh rate 2K screen, providing not only smoother and more stable visuals but also enhanced scratch and drop resistance.

[0049] Furthermore, in one embodiment, the logic design of the front-end interactive page of the pressure plate status verification application in step S3 utilizes the built-in interface of existing software and programs to implement JavaScript script functions. The user starts the image upload function through the input image icon on the main page, and the main page executes the jump logic to jump to the pressure plate preview page. The platen preview page calls relevant program interfaces such as wx.chooseImage to retrieve the platen image from the local device and displays the selected image for user confirmation. After user confirmation, the platen status verification application front end uses relevant program interfaces such as wx.uploadFile to upload the image to the server for platen target detection. The server returns the platen status, location, and binary information of the image marked with the target platen, which is then sent to the main page, and the user is redirected back to the main page.

[0050] The front-end interactive interface consists of four parts: the homepage, the plate preview page, the QR code preview page, and the result feedback page. The operational logic within and between the pages is as follows: Figure 2 As shown.

[0051] (1) Acquisition of pressure plate status and positioning information When a user initiates the "Upload Image" function on the homepage, the main page executes a redirection logic, leading to the platen preview page. According to the design, this page automatically calls the `wx.chooseImage` interface during initialization to retrieve the platen image from the local device and displays the selected image for user confirmation. After user confirmation, the frontend uses the `wx.uploadFile` interface to upload the image to the server for platen target detection. Then, it sends the platen status, location, and binary information of the image marked with the target platen back from the server to the homepage and redirects the user back to the homepage.

[0052] (2) Acquisition of prior information of the cabinet Users initiate the "Upload QR Code" function on the homepage. The frontend then redirects to a QR code preview page, during which the `wx.scanCode` interface is called. This interface activates the local camera to capture the QR code on the display cabinet and automatically decodes it. The decoded code is then displayed on the preview page for user confirmation. After user confirmation, the frontend calls the `wx.request` interface to upload the decoded code to the server. Finally, the server returns relevant information about the display cabinet to the homepage, which then redirects the user back to the homepage.

[0053] (3) Checking the condition of the pressure plate After successfully obtaining the pressure plate category, location, and related prior information, when the user activates the "Start Verification" function on the homepage, the front-end will begin name matching and status verification based on the relative positions of each pressure plate. Once the results are generated, the front-end will automatically redirect to the results feedback page. Next, the front-end will display images of the pressure plates with prediction boxes on the results feedback page, and show prompts based on the detection results.

[0054] Finally, the front-end also provides two optional features: a) Image saving; save the image of the pressure plate with the prediction box marker to the local machine.

[0055] b) Download the test results in tabular form to view the specific test details. The tab includes the name of the pressure plate, its location, the test status, and the preset status.

[0056] Similar to web development, WeChat Mini Programs can utilize WXML and WXSS languages ​​for interface development. The former determines the page layout and related human-computer interaction feedback, such as component placement, data binding, and conditional rendering. The latter determines the style of each module component, such as color, shape, and size.

[0057] (1) Main page First is the main page, whose page components include: two container views with different icons, a "Start Detection" button, and a general navigation bar at the bottom, which provides a jump interface between the home page and the results feedback page.

[0058] (2) Preview page of the pressure plate image After clicking the "Input Image" icon on the main page, the front end will redirect to the pressure plate prediction page. While the page is loading, the system will prompt you to select a pressure plate image. The pressure plate image display area supports horizontal scrolling preview to view multiple images. After clicking the "Confirm" button, the page will trigger the pressure plate detection function, upload the selected image to the server for detection, and store the received server-returned data in a global variable before redirecting back to the main page.

[0059] (3) QR code preview page Next, clicking the "Enter QR Code" icon on the main page will redirect you to a QR code preview page. While the page is loading, the system will use the API interface to activate the camera function and capture an image of the QR code on the display cabinet. The page will then display the decoded display cabinet code. Clicking the "Confirm" button below will trigger the function to retrieve the prior information of the display cabinet and redirect you back to the main page.

[0060] (4) Results Feedback Page Clicking the "Start Detection" button on the main page will trigger the pressure plate verification function and redirect you to the results feedback page. On the results feedback page, the front end will display the predicted pressure plate image with bounding box markers, which also supports scrolling preview.

[0061] In addition, the front-end will also provide prompts regarding the correctness of the detection results. If the detection results are normal and consistent with the preset state, a prompt box with a correct mark will appear; otherwise, a warning mark will be displayed. Due to the limited clarity of the preview image, the front-end also provides an interface for saving the predicted image. Clicking the "Save Image" button will save the image to local storage and provide a prompt. Clicking the "View Details" button will provide the detection results in a table format for saving and downloading. If you want to input a new image for detection, click the "Detection and Recognition" button in the bottom navigation bar to return to the main page and start again.

[0062] Furthermore, in one embodiment, the detection results in step S4 are saved in txt text format, and the results include the pressure plate category, the horizontal and vertical coordinates of the center point of the pressure plate prediction frame, and the width and height of the pressure plate prediction frame.

[0063] Furthermore, in one embodiment, the pressure plate database in step S4 stores the pressure plate dataset established in step S1, and relevant prior information of the pressure plate is obtained from the pressure plate database when performing the protection pressure plate status verification work. The pressure plate database uses the screen cabinet as the basic unit to store and manage the pressure plate data. Then, all screen cabinets are serialized, and each screen cabinet has only one unique code. The relevant prior information is associated through the screen cabinet code. Finally, the screen cabinet code is converted into a QR code and pasted on the screen cabinet. By simply scanning the QR code, the screen cabinet code can be automatically obtained. This brings great convenience from both on-site practice and management perspectives.

[0064] Furthermore, in one embodiment, in step S4, when performing the status verification of the protective pressure plates, the position of each protective pressure plate in the matrix is ​​determined based on its coordinate information. Then, using this as a link, a one-to-one mapping relationship between the protective pressure plates and the pressure plate database is established, and the status verification function of the pressure plates is finally realized.

[0065] Furthermore, in one embodiment, step S4, determining the position of each protective pressure plate in the matrix based on its coordinate information, includes: S41. Verify the number of detected pressure plates by combining the database information to determine whether there are any missed detections. S42. Use perspective transformation to correct visual distortions in the image; S43. Locate the position of the current protective pressure plate in the matrix arrangement based on the coordinates of the remaining pressure plates; S44. Generate and output the result object. The object attributes include the pressure plate name, pressure plate position (position:(i,j)), pressure plate preset state, and comparison result.

[0066] The system interaction diagram of the entire pressure plate status detection application is as follows: Figure 3 As shown, the specific execution logic for obtaining the pressure plate status and position information is as follows: Figure 4 As shown.

[0067] First, the server receives the platen image from the front end, and then executes the platen detection algorithm based on the improved YOLO V5 through the built-in interface. The detection results are saved in txt text format. The first column is the platen category, the second and third columns are the horizontal and vertical coordinates of the center point of the platen prediction box, and the fourth and fifth columns are the width and height of the platen prediction box, respectively.

[0068] Then, a file manager was used to read the txt result file. The string function `split` was used, with spaces and newline characters "\n" as delimiters. Each line represented a pressure plate, and the first three columns (the pressure plate category and the center coordinates of the prediction box) were retained. The txt text was then converted into a Python list object. Next, the states of each pressure plate were divided according to its category, as shown in Table 1. New pressure plate object variables were then constructed, replacing the original category with the state.

[0069] Table 1 Pressure Plate Status Mapping Table Pressure plate category Category Number Pressure plate state Status number yb-on 1 Investment 1 lp-on 2 Investment 1 yb-off 3 quit 0 lp-off 4 quit 0 lp-off-sl 5 quit 0 Finally, due to the format limitations of data transmission between the front-end and back-end, the pressure plate object variable also needs to be converted into a JSON string. It's worth noting that the JSON data returned by the back-end includes not only the pressure plate status and position information, but also a detection result image with a prediction bounding box.

[0070] Since the verification of the pressure plate status requires some prior information about the pressure plate, this information is stored in the server's database for easy access and management. A good database can not only greatly improve retrieval efficiency but also significantly reduce data maintenance and management costs.

[0071] The database design uses control cabinets as the basic unit to store and manage pressure plate data. Then, all control cabinets are serialized, with each cabinet having a unique code. This code is used to link related prior information. Finally, the control cabinet code is converted into a QR code and affixed to the cabinet. Simply scanning the QR code automatically retrieves the control cabinet code, providing significant convenience from both on-site implementation and management perspectives.

[0072] The flowchart for implementing the pressure plate status verification function is as follows: Figure 5 As shown. Based on the arrangement characteristics of the pressure plates on the protective cabinet, namely that all pressure plates are arranged in a matrix on the same plane, evenly distributed, and without spatial overlap, the position of each pressure plate in the matrix is ​​determined according to its coordinate information. Then, using this as a link, a one-to-one mapping relationship between the pressure plates and the database is established, ultimately realizing the status verification function of the pressure plates. Therefore, how to accurately locate the pressure plates in the matrix is ​​the key to the status verification function.

[0073] The flowchart of the matrix arrangement of the pressure plates is as follows: Figure 6 As shown. First, the number of detected pressure plates is verified using database information to determine if any were missed. If any were missed, the program terminates and reports the error to the system; otherwise, the next step is executed.

[0074] However, in actual testing, the images of the pressure plates often have varying degrees of shooting tilt angles, causing visual distortion of the image content. That is, points closer to the camera appear larger, while those farther away appear smaller. The overall arrangement of the pressure plates is wedge-shaped, which destroys the true distance and positional relationship between the pressure plates in the image. This will bring a large deviation to the subsequent meshing calculation.

[0075] To address the aforementioned issues, perspective transformation is employed to correct visual distortions in the image. Correcting distorted images using perspective transformation requires obtaining the coordinates of four points in the distorted image and their corresponding corrected coordinates. A coordinate transformation system is then established using these two sets of coordinates. Since the characteristics of the pressure plate matrix are well-defined, the coordinates of the four vertex pressure plates are selected as the distortion group data. Considering that the horizontal and vertical distances between pressure plates are equal and fixed in reality, and that the correction goal is only to restore the pressure plates to an equidistant, regular matrix arrangement without needing to know their actual coordinates, the row width and column height of the pressure plate matrix in the ideal target image can be set to fixed values. This yields the coordinates of the four vertices of the ideal group, allowing the calculation of the perspective transformation matrix. The remaining pressure plate coordinates are then corrected using the perspective transformation matrix.

[0076] The coordinates of the four vertex pressure plates of the distortion group are obtained as follows: Figure 7 As shown, the quadrilateral region represents the arrangement of the pressure plates. First, calculate the Euclidean distance from each pressure plate to the origin o, and find the maximum and minimum distances to obtain points B and D. Since the distorted image of the pressure plates under tilt still exhibits symmetry, the midpoint of B and D is the geometric center of the pressure plate arrangement region. Then, based on this, divide the blue region into four equal parts. The point in each region with the greatest Euclidean distance from the center point is the vertex located within that region.

[0077] The purpose of meshing is to locate the remaining pressure plates within the matrix arrangement based on their coordinates. For example, for a pressure plate panel arranged in 4 rows and 9 columns, the meshing process will construct a 3*8 mesh based on the coordinates of the center points of the four pressure plates at their vertices. Figure 8 As shown, each node in the grid represents the coordinates of the ideal center point of the pressure plate.

[0078] Taking the pressure plate mapping at the 2nd row and 2nd column of the matrix as an example, calculate the Euclidean distance from each corrected pressure plate coordinate to the point (2, 2). The one with the smallest distance is considered to belong to that point. Repeat the above operation to confirm the pressure plate corresponding to each node in the mesh. Then update the pressure plate object information to the form of state (state: 0 / 1) & node position (position: (i, j)).

[0079] When performing state comparison and outputting results, the node position is used as the link. First, the detected pressure plate objects are mapped one by one with the pressure plate data in the database. Then, the state comparison is performed, and a new result object is generated for output. The object attributes include: pressure plate name, pressure plate position (position:(i,j)), pressure plate preset state, and comparison result (if the detected pressure plate state is consistent with the preset state, the result is 1, otherwise it is 0).

[0080] In summary, the intelligent verification method for the status of substation panel pressure plates based on machine vision of the present invention includes the following steps: S1, acquiring and annotating image data of the protective pressure plates to establish a pressure plate dataset; S2, improving the YOLO V5 algorithm based on the requirements for protective pressure plate status detection; S3, designing the logic and layout of the front-end interactive page of the pressure plate status verification application; S4, the server receives the pressure plate images from the front end, executes the improved YOLO V5 pressure plate detection algorithm through the built-in interface, returns the detection results to the front end, and saves them into the pressure plate database. Through the above steps, the intelligent verification method for the status of substation panel pressure plates based on machine vision of the present invention constructs and annotates a dataset by acquiring images covering different shooting angles, lighting, shadows, and pressure plate types, and then improves the YOLO V5 algorithm. The algorithm addresses the issues of missed or incorrect detections and overlapping detection frames in small target detection on pressure plates. It also designs a front-end program containing modules such as a homepage and a pressure plate preview page, along with a corresponding back-end system. A database is built with the cabinet as the unit and associated with the prior information of the pressure plates. By correcting image distortion through perspective transformation and locating the position of the pressure plate matrix through gridded calculation, the algorithm enables automatic identification and verification of the pressure plate status, improving the accuracy of on-site pressure plate verification, significantly shortening the time for secondary equipment status verification, and improving operation and maintenance efficiency.

[0081] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0082] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A machine vision-based intelligent verification method for the status of substation switchgear pressure plates, characterized in that, Includes the following steps: S1. Collect and annotate image data of the protective pressure plate, and establish a pressure plate dataset; S2. Improve the YOLO V5 algorithm based on the requirements for detecting the status of the protective pressure plate; S3. Perform the logic and layout design of the application's front-end interactive page to check the pressure plate status; S4. The server receives the pressure plate image from the front end, executes the improved YOLO V5 pressure plate detection algorithm through the built-in interface, returns the detection result to the front end, and saves it into the pressure plate database.

2. The intelligent verification method for the status of substation cabinet pressure plates based on machine vision as described in claim 1, characterized in that, The protective pressure plate in step S1 includes two types: pressure plate type and continuous plate type. Image data acquisition of the protective pressure plate includes taking pictures of each type of protective pressure plate under various working conditions at different angles and with different light intensities. Image data annotation of the protective pressure plate includes inputting the annotation information of the captured protective pressure plate pictures along with the pictures into the network during network model training.

3. The intelligent verification method for the status of substation panel pressure plates based on machine vision as described in claim 2, characterized in that, The annotation information of the protective pressure plate image includes the category ID of the target protective pressure plate and the coordinate information of the four corner points of the bounding box. According to the different types of protective pressure plates and their working status, the protective pressure plates are divided into 5 categories, including pressure plate-in, pressure plate-out, continuous plate-in, continuous plate-standby, and continuous plate-out.

4. The intelligent verification method for the status of substation panel pressure plates based on machine vision as described in claim 1, characterized in that, Step S2, which improves the YOLO V5 algorithm based on the requirements for detecting the status of the protective pressure plate, includes: Improvements were made to the YOLO V5 algorithm for small object detection and for single-label non-overlapping objects. The improvement for small object detection addresses the limitation of increasing image resolution during image data acquisition for protective pressure plates by enhancing feature extraction and reducing feature loss, thus improving small object detection performance. The improvement for single-label non-overlapping objects addresses the inconsistency between the design of the YOLO V5 object detection algorithm for multi-label multi-classification and occlusion problems and the actual characteristics of pressure plate detection by modifying the non-maximum suppression algorithm.

5. The intelligent verification method for the status of substation panel pressure plates based on machine vision as described in claim 1, characterized in that, In step S3, the layout of the front-end interactive page of the pressure plate status check application is designed by writing WXXL templates and WXSS style files. The front-end interactive page consists of four parts: the main page, the pressure plate preview page, the QR code preview page, and the result feedback page. The main page has two container views with different icons: an input image icon and an input QR code icon. It also has a "Start Detection" button and a general navigation bar at the bottom, which provides an interface for jumping between the main page and the results feedback page.

6. The intelligent verification method for the status of substation panel pressure plates based on machine vision as described in claim 5, characterized in that, In step S3, the logic design of the front-end interactive page of the pressure plate status verification application utilizes the built-in interface of existing software and programs to implement JavaScript script functions. The user starts the image upload function through the input image icon on the main page, and the main page executes the jump logic to jump to the pressure plate preview page. The pressure plate preview page calls the relevant program interface to obtain the pressure plate image from the local device and displays the selected image. After the user confirms, the pressure plate status verification application front end uses the relevant program interface to upload the image to the server for pressure plate target detection. The server returns the pressure plate status, location, and binary information of the image marked with the target pressure plate, which is then sent to the main page and redirected back to the main page.

7. The intelligent verification method for the status of substation panel pressure plates based on machine vision as described in claim 1, characterized in that, The detection results in step S4 are saved in txt text format. The results include the pressure plate category, the horizontal and vertical coordinates of the center point of the pressure plate prediction frame, and the width and height of the pressure plate prediction frame.

8. The intelligent verification method for the status of substation panel pressure plates based on machine vision as described in claim 1, characterized in that, The pressure plate database in step S4 stores the pressure plate dataset established in step S1. When performing the protection pressure plate status verification work, the relevant prior information of the pressure plate is obtained from the pressure plate database. The pressure plate database uses the cabinet as the basic unit to store and manage the pressure plate data. Then, all cabinets are serialized, and each cabinet has only one unique code. The relevant prior information is associated through the cabinet code.

9. The intelligent verification method for the status of substation switchgear pressure plates based on machine vision as described in claim 8, characterized in that, In step S4, when performing the status verification of the protective pressure plates, the coordinate information of each protective pressure plate is used to determine its position in the matrix. Then, based on this, a one-to-one mapping relationship between the protective pressure plates and the pressure plate database is established, and the status verification function of the pressure plates is finally realized.

10. The intelligent verification method for the status of substation panel pressure plates based on machine vision as described in claim 9, characterized in that, In step S4, determining the position of each protective pressure plate in the matrix based on its coordinate information includes: S41. Verify the number of detected pressure plates by combining the database information to determine whether there are any missed detections. S42. Use perspective transformation to correct visual distortions in the image; S43. Locate the position of the current protective pressure plate in the matrix arrangement based on the coordinates of the remaining pressure plates; S44. Generate and output the result object. The object attributes include the pressure plate name, pressure plate position, pressure plate preset state, and comparison result.

Images