A substation electrical protection pressboard detection method, device and processing equipment

Abstract

This application provides a method, apparatus, and processing equipment for detecting electrical protection pressure plates in a substation. It is used to perform granular individual identification of electrical protection pressure plates within a dense cluster, thereby providing accurate detection targets for intelligent activation / deactivation status detection of electrical protection pressure plates. This enables granular intelligent activation / deactivation status detection of electrical protection pressure plates within the cluster. The method includes: acquiring an actual image I collected from the electrical protection pressure plate cabinet in the substation, where different electrical protection pressure plates are configured in the form of panels; and using a semantic segmentation model M... 语义分割 The predictive mask for the electrical protection pressure plate panel is identified from the actual image I. This predictive mask indicates the region where the electrical protection pressure plate panel is segmented from the actual image I. Based on the segmentation distance between different electrical protection pressure plates, the predictive mask for the electrical protection pressure plate panel is divided into individual predictive masks for different electrical protection pressure plate instances.

Description

Technical Field

[0001] This application relates to the field of substations, specifically to a method, device, and processing equipment for detecting electrical protection pressure plates in substations. Background Technology

[0002] With the continuous expansion of the power grid, the number of electrical protection circuit boards on the transmission lines of substations has increased dramatically, leading to a corresponding increase in the cost of manual inspection. To break free from the constraints of traditional manual inspection, prevent misoperation of electrical protection circuit boards, and solve the problem of automated management of current substation electrical protection circuit board information, it is necessary to use an intelligent on / off status detection method suitable for substation electrical protection circuit boards.

[0003] Currently, intelligent status detection of substation electrical protection plates can be divided into two categories: image feature matching-based detection technology and neural network object detection-based detection technology. Image feature matching-based electrical protection plate activation / deactivation status detection originated in the 1980s. This method involves two steps: first, using feature point sampling technology to extract features from the electrical protection plate image; then, matching these features with features in a pre-built dataset to determine the activation / deactivation status of the electrical protection plate. However, due to limitations of the feature point sampling algorithm, this method has extremely high requirements regarding camera angle and ambient light levels when capturing the electrical protection plate image. Neural network object detection-based methods use models such as SSD, MaskR-CNN, and YOLO as a foundation, training the models with large-scale datasets to achieve robust activation / deactivation status detection for electrical protection plates. However, while object detection algorithms perform well for scenarios with discretely distributed electrical protection plates, their performance is less ideal for densely distributed electrical protection plates. Furthermore, the current two types of electrical protection pressure plate activation / deactivation status detection methods have not thoroughly studied the correspondence between electrical protection pressure plates and ledgers, making it difficult to meet the requirements for automated management of the corresponding information.

[0004] It can be found that the existing intelligent status detection schemes for substation electrical protection pressure plates have poor detection accuracy in practical application scenarios where electrical protection pressure plates are densely packed and environmental conditions are harsh. Summary of the Invention

[0005] This application provides a method, apparatus, and processing equipment for detecting electrical protection pressure plates in substations. It is used to perform individual identification of electrical protection pressure plates at the granular level in a dense cluster of electrical protection pressure plates, thereby providing accurate detection targets for intelligent activation / deactivation status detection of electrical protection pressure plates. This enables the intelligent activation / deactivation status detection of electrical protection pressure plates at the granular level to be completed in the cluster of electrical protection pressure plates.

[0006] Firstly, this application provides a method for testing electrical protection pressure plates in substations, the method comprising:

[0007] Acquire actual image I from the electrical protection pressure plate cabinet of the substation, wherein the electrical protection pressure plate cabinet is configured with different electrical protection pressure plates in the form of a panel;

[0008] Using semantic segmentation model M 语义分割 Identify the predictive mask for the electrical protection pressure plate panel from actual image I. Among them, electrical protection pressure plate panel predictive mask Used to indicate the area where the electrical protection pressure plate panel is located, segmented from the actual image I;

[0009] Based on the dividing distance between different electrical protection pressure plates, predict the mask of the electrical protection pressure plate panel. Individual predictive masks for different electrical protection pressure plate instances

[0010] In conjunction with the first aspect of this application, in a first possible implementation of the first aspect of this application, the electrical protection pressure plate panel prediction mask is prepared according to the preset dividing distance between different electrical protection pressure plates. Individual predictive masks for different electrical protection pressure plate instances Subsequently, the method also includes:

[0011] Predictive masks based on actual image I and individual examples of different electrical protection pressure plates. And image classification model M 图像分类 Identify the activation / deactivation status of each individual electrical protection pressure plate instance.

[0012] Through optical character recognition model M OCR The predicted coordinates of the upper left corner of each electrical protection pressure plate label were identified from the actual image I. Predicted coordinates in the lower right corner And text prediction based on labels;

[0013] Predict the upper left corner coordinates of each electrical protection pressure plate label. Predicted coordinates in the lower right corner And the predicted text of the labels, and the corresponding activation / deactivation status of each individual electrical protection pressure plate instance. By performing correlation, information about the electrical protection pressure plate can be obtained.

[0014] In conjunction with the first possible implementation of the first aspect of this application, in the second possible implementation of the first aspect of this application, a predictive mask is generated using actual image I and individual examples of different electrical protection pressure plates. And image classification model M 图像分类Identify the activation / deactivation status of each individual electrical protection pressure plate instance. include:

[0015] Predictive mask based on individual examples of different electrical protection pressure plates. Sub-images I corresponding to each individual electrical protection pressure plate instance are divided from the actual image I. i ;

[0016] Image classification model M 图像分类 Identify each subgraph I i Corresponding surrender / withdrawal status

[0017] In conjunction with the first possible implementation of the first aspect of this application, in the third possible implementation of the first aspect of this application, the predicted coordinates of the upper left corner corresponding to each electrical protection pressure plate label are... Predicted coordinates in the lower right corner And the predicted text of the labels, and the corresponding activation / deactivation status of each individual electrical protection pressure plate instance. By performing correlation, information about the electrical protection pressure plate is obtained, including:

[0018] Calculate individual predictive masks for different electrical protection pressure plate instances pixel center point C i ;

[0019] Calculate the predicted coordinates of the upper left corner of each electrical protection pressure plate label. and the predicted coordinates in the lower right corner To the center point C of each pixel i Average distance D (i,j) ;

[0020] Using the latest algorithm, the predicted text of the label corresponding to each electrical protection pressure plate label is compared with the activation / deactivation status of each individual electrical protection pressure plate instance. By performing correlation, information about the electrical protection pressure plate can be obtained.

[0021] In conjunction with the first possible implementation of the first aspect of this application, the fourth possible implementation of the first aspect of this application further includes:

[0022] Acquire sample image I from the electrical protection pressure plate cabinet.

[0023] Content enhancement was performed on sample image I using random brightness adjustment, random dynamic blur, random flipping, and random rotation as data augmentation measures.

[0024] The sample image I is labeled with the corresponding sample electrical protection pressure plate panel prediction mask M and sample electrical protection pressure plate instance individual prediction mask. Sample submission / rejection status Predicted coordinates of the top left corner of the sample Predicted coordinates of the bottom right corner of the sample After predicting the text using sample labels, the semantic segmentation model M is trained using the sample image I and the sample electrical protection pressure plate panel prediction mask M. 语义分割 The mask is predicted using sample image I and sample electrical protection pressure plate instance. and sample submission / return status To train image classification model M 图像分类 The predicted coordinates are obtained from the actual image I of the sample and the top left corner of the sample. Predicted coordinates of the bottom right corner of the sample And sample labels predict text to train the optical character recognition model M OCR .

[0025] In conjunction with any possible implementation of the first aspect of this application, in the fifth possible implementation of the first aspect of this application, after obtaining the electrical protection pressure plate information, the method further includes:

[0026] The electrical protection pressure plate information is archived in the system according to the ledger information organization method.

[0027] In conjunction with the first aspect of this application, in the sixth possible implementation of the first aspect of this application, the electrical protection pressure plate panel prediction mask is prepared according to the dividing distance between different electrical protection pressure plates. Individual predictive masks for different electrical protection pressure plate instances Previously, the methods also included:

[0028] Convert the actual image I into a grayscale image;

[0029] Based on the grayscale image, the division distance between different electrical protection pressure plate panels is calculated in both the horizontal and vertical directions using an adaptive thresholding strategy.

[0030] Secondly, this application provides a substation electrical protection pressure plate detection device, the device comprising:

[0031] The acquisition unit is used to acquire the actual image I collected from the electrical protection pressure plate cabinet of the substation, wherein the electrical protection pressure plate cabinet is configured with different electrical protection pressure plates in the form of a panel;

[0032] Semantic segmentation unit, used to segment semantic segmentation model M 语义分割 Identify the predictive mask for the electrical protection pressure plate panel from actual image I. Among them, electrical protection pressure plate panel predictive mask Used to indicate the area where the electrical protection pressure plate panel is located, segmented from the actual image I;

[0033] Dividing units are used to divide the electrical protection pressure plate panel predictive mask according to the dividing distance between different electrical protection pressure plates. Individual predictive masks for different electrical protection pressure plate instances

[0034] In conjunction with the second aspect of this application, in a first possible implementation of the second aspect of this application, the apparatus further includes:

[0035] The activation / deactivation status recognition unit is used to predict the mask based on the actual image I and individual instances of different electrical protection pressure plates. And image classification model M 图像分类 Identify the activation / deactivation status of each individual electrical protection pressure plate instance.

[0036] Optical character recognition unit, used to recognize optical characters through optical character recognition model M OCR The predicted coordinates of the upper left corner of each electrical protection pressure plate label were identified from the actual image I. Predicted coordinates in the lower right corner And text prediction based on labels;

[0037] The association unit is used to predict the coordinates of the upper left corner of each electrical protection pressure plate label. Predicted coordinates in the lower right corner And the predicted text of the labels, and the corresponding activation / deactivation status of each individual electrical protection pressure plate instance. By performing correlation, information about the electrical protection pressure plate can be obtained.

[0038] In conjunction with the first possible implementation of the second aspect of this application, in the second possible implementation of the second aspect of this application, the engagement / disengagement status identification unit is specifically used for:

[0039] Predictive mask based on individual examples of different electrical protection pressure plates. Sub-images I corresponding to each individual electrical protection pressure plate instance are divided from the actual image I. i ;

[0040] Image classification model M 图像分类 Identify each subgraph I i Corresponding surrender / withdrawal status

[0041] In conjunction with the first possible implementation of the second aspect of this application, in the third possible implementation of the second aspect of this application, the associated unit is specifically used for:

[0042] Calculate individual predictive masks for different electrical protection pressure plate instances pixel center point C i ;

[0043] Calculate the predicted coordinates of the upper left corner of each electrical protection pressure plate label. and the predicted coordinates in the lower right corner To the center point C of each pixel i Average distance D (i,j) ;

[0044] Using the latest algorithm, the predicted text of the label corresponding to each electrical protection pressure plate label is compared with the activation / deactivation status of each individual electrical protection pressure plate instance. By performing correlation, information about the electrical protection pressure plate can be obtained.

[0045] In conjunction with the first possible implementation of the second aspect of this application, in the fourth possible implementation of the second aspect of this application, the apparatus further includes a model training unit, used for:

[0046] Acquire sample image I from the electrical protection pressure plate cabinet.

[0047] Content enhancement was performed on sample image I using random brightness adjustment, random dynamic blur, random flipping, and random rotation as data augmentation measures.

[0048] The sample image I is labeled with the corresponding sample electrical protection pressure plate panel prediction mask M and sample electrical protection pressure plate instance individual prediction mask. Sample submission / rejection status Predicted coordinates of the top left corner of the sample Predicted coordinates of the bottom right corner of the sample After predicting the text using sample labels, the semantic segmentation model M is trained using the sample image I and the sample electrical protection pressure plate panel prediction mask M. 语义分割 The mask is predicted using sample image I and sample electrical protection pressure plate instance. and sample submission / return status To train image classification model M 图像分类 The predicted coordinates are obtained from the actual image I of the sample and the top left corner of the sample. Predicted coordinates of the bottom right corner of the sample And sample labels predict text to train the optical character recognition model M OCR .

[0049] In conjunction with any possible implementation of the second aspect of this application, in the fifth possible implementation of the second aspect of this application, the apparatus further includes an archiving unit, used for:

[0050] The electrical protection pressure plate information is archived in the system according to the ledger information organization method.

[0051] In conjunction with the second aspect of this application, in the sixth possible implementation of the second aspect of this application, dividing the units is further used for:

[0052] Convert the actual image I into a grayscale image;

[0053] Based on the grayscale image, the division distance between different electrical protection pressure plate panels is calculated in both the horizontal and vertical directions using an adaptive thresholding strategy.

[0054] Thirdly, this application provides a processing device, including a processor and a memory, wherein a computer program is stored in the memory, and when the processor invokes the computer program in the memory, it executes the method provided by the first aspect of this application or any possible implementation of the first aspect of this application.

[0055] Fourthly, this application provides a computer-readable storage medium storing a plurality of instructions adapted for loading by a processor to perform the method provided in the first aspect of this application or any possible implementation thereof.

[0056] From the above, it can be concluded that this application has the following beneficial effects:

[0057] To address the need for intelligent on / off status detection of electrical protection circuit breakers in substations, after acquiring the actual image I collected from the electrical protection circuit breaker cabinet in the substation, this application, compared to the existing technology that directly identifies the on / off status of the electrical protection circuit breaker through image processing based on the actual image I, uses a semantic segmentation model M. 语义分割 Identify the predictive mask for the electrical protection pressure plate panel from actual image I. Then, based on the dividing distance between different electrical protection pressure plates, the predictive mask for the electrical protection pressure plate panel is created. Individual predictive masks for different electrical protection pressure plate instances At this point, the individual predictive mask for the electrical protection pressure plate instance is shown. Then, the image of the electrical protection pressure plate instance can be segmented from the actual image I. The individual recognition of the electrical protection pressure plate at the granular level can be performed in the dense electrical protection pressure plate cluster, thereby providing accurate detection objects for the intelligent activation and deactivation status detection of the electrical protection pressure plate. In this way, the intelligent activation and deactivation status detection of the electrical protection pressure plate at the granular level can be completed in the electrical protection pressure plate cluster. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1This is a flowchart illustrating one method for testing electrical protection pressure plates in substations according to this application.

[0060] Figure 2 This is a schematic diagram of a scene where actual image I was acquired in this application;

[0061] Figure 3 This is a schematic diagram of a scenario for data enhancement measures in this application;

[0062] Figure 4 This is a schematic diagram illustrating one scenario of annotation processing in this application;

[0063] Figure 5 The semantic segmentation model M in this application 语义分割 A schematic diagram of an exemplary model structure;

[0064] Figure 6 For the image classification model M in this application 图像分类 A schematic diagram of an exemplary model structure;

[0065] Figure 7 The optical character recognition model M of this application OCR A flowchart illustrating optical character recognition;

[0066] Figure 8 This is a schematic diagram of a substation electrical protection pressure plate detection device according to this application;

[0067] Figure 9 This is a schematic diagram of one type of processing equipment used in this application. Detailed Implementation

[0068] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0069] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules is not necessarily limited to those explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices. The naming or numbering of steps appearing in this application does not imply that the steps in the method flow must be performed in the chronological / logical order indicated by the naming or numbering. The execution order of named or numbered process steps can be changed according to the desired technical purpose, as long as the same or similar technical effect is achieved.

[0070] The module division described in this application is a logical division. In practical applications, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the coupling or direct coupling or communication connection between modules shown or discussed may be through some interfaces, and the indirect coupling or communication connection between modules may be electrical or other similar forms, none of which are limited in this application. Furthermore, the modules or sub-modules described as separate components may or may not be physically separated, may or may not be physical modules, or may be distributed in multiple circuit modules. Some or all of the modules may be selected to achieve the purpose of the solution in this application according to actual needs.

[0071] Before introducing the substation electrical protection pressure plate testing method provided in this application, the background content involved in this application will be introduced first.

[0072] The substation electrical protection pressure plate detection method, device, and computer-readable storage medium provided in this application can be applied to processing equipment to perform individual identification of electrical protection pressure plates at the granular level in a dense cluster of electrical protection pressure plates, thereby providing accurate detection objects for intelligent activation / deactivation status detection of electrical protection pressure plates, and thus enabling intelligent activation / deactivation status detection of electrical protection pressure plates at the granular level in the cluster of electrical protection pressure plates.

[0073] The substation electrical protection pressure plate detection method mentioned in this application can be implemented by a substation electrical protection pressure plate detection device, or by different types of processing devices such as servers, physical hosts, or user equipment (UE) that integrate the substation electrical protection pressure plate detection device. The substation electrical protection pressure plate detection device can be implemented in hardware or software. The UE can be a terminal device such as a smartphone, tablet, laptop, desktop computer, or personal digital assistant (PDA). The processing devices can be configured in a device cluster.

[0074] For example, the processing equipment can be related equipment in the substation that performs the task of automated management of electrical protection pressure plate information, such as equipment that performs the task of intelligent activation / deactivation status detection of electrical protection pressure plates, or personal equipment of staff. It can be understood that the processing equipment can be configured into different equipment forms according to different application scenarios, so this application does not make specific limitations here.

[0075] The following section introduces the substation electrical protection pressure plate testing method provided in this application.

[0076] First, refer to Figure 1 , Figure 1 This paper illustrates a flowchart of a substation electrical protection pressure plate testing method according to the present application. The substation electrical protection pressure plate testing method provided by the present application may specifically include the following steps S101 to S103:

[0077] Step S101: Obtain actual image I from the electrical protection pressure plate cabinet of the substation, wherein the electrical protection pressure plate cabinet is configured with different electrical protection pressure plates in the form of a panel;

[0078] It is understandable that when using image processing to detect the intelligent activation / deactivation status of electrical protection pressure plates in substations, the first step is to collect images of the electrical protection pressure plates from the substation site.

[0079] In practical applications, voltage protection pressure plates exist in clusters. Specifically, a large number of voltage protection pressure plates can be deployed in the field in the form of voltage protection pressure plate cabinets. Therefore, when collecting field images of voltage protection pressure plates, the images show a large number of voltage protection pressure plates deployed on the cabinet and displayed on the panel.

[0080] The electrical protection pressure plate cabinet can be filmed using a fixed on-site camera device (including a camera or a camera directly), or on-site staff can film the electrical protection pressure plate cabinet using a handheld camera device (including a camera). For example, staff can film it using the rear camera of a smartphone or a digital camera.

[0081] As an example, see reference Figure 2 The illustration shown is a scene diagram of actual image I collected in this application. The requirements for taking the picture are: the electrical protection pressure plate cabinet is at a 90° angle to the ground, the camera / camera uses a 35-50mm standard lens, the optical axis of the lens is consistent with the eye level (or center point) of the electrical protection pressure plate, the shooting distance is between 0.5-3m, and the obtained picture of the electrical protection pressure plate should include the complete electrical protection pressure plate panel.

[0082] Furthermore, it should be understood that the acquisition of the actual image I in step S101 can be either real-time acquisition of the image, such as calling or triggering the camera to perform image acquisition, or image retrieval processing, that is, retrieving the image acquired by the previously completed image acquisition work.

[0083] Step S102, using semantic segmentation model M 语义分割 Identify the predictive mask for the electrical protection pressure plate panel from actual image I. Among them, the electrical protection pressure plate panel predictive mask Used to indicate the area where the electrical protection pressure plate panel is located, segmented from the actual image I;

[0084] It is understandable that the initial identification of the electrical protection pressure plate cluster in the actual image I is achieved through semantic segmentation in this application. Specifically, this can be handled by the semantic segmentation model M pre-trained in this application. 语义分割 The semantic segmentation it performs is accomplished by predicting the mask using the electrical protection pressure plate panel. This is used to indicate the electrical protection pressure plate cluster identified in the input image. It can be understood that in practical applications, this application uses the area where the electrical protection pressure plate panel is located as the area where the electrical protection pressure plate cluster is located, which has the characteristics of being lightweight and can be quickly and accurately identified.

[0085] Specifically, the semantic segmentation process here does not require the identification of specific features of the electrical protection pressure plate. Instead, it uses an indirect method to detect the electrical protection pressure plate cluster by detecting the electrical protection pressure plate panel.

[0086] Step S103: Based on the dividing distance between different electrical protection pressure plates, predict the mask of the electrical protection pressure plate panel. Individual predictive masks for different electrical protection pressure plate instances

[0087] It is understandable that obtaining a predictive mask for the electrical protection pressure plate panel that can indicate the electrical protection pressure plate cluster (composed of different individual electrical protection pressure plate instances) is crucial. Then, it can be further divided into finer granular levels.

[0088] It is understood that, in accordance with the application requirements of this application, when deploying voltage protection pressure plates in a voltage protection pressure plate cabinet, a regular density can be adopted. In this way, multiple voltage protection pressure plates on the panel can present a certain regular distance, or interval, which can specifically be the interval in the vertical and horizontal directions.

[0089] Therefore, in this application, the concept of distance division can be introduced. Based on the pre-set or real-time set division distance between different electrical protection pressure plates, the overall electrical protection pressure plate panel predictive mask can be applied. Decomposed into individual predictive masks for different electrical protection pressure plate instances Individual predictive mask for each electrical protection pressure plate instance For a given electrical protection pressure plate instance, predict the mask based on that instance. Then an image of an individual electrical protection pressure plate instance can be segmented from the actual image I.

[0090] As is easily understood, for the actual image I, this application introduces a two-level processing. The first level is semantic segmentation, which indirectly detects the region where the electrical protection pressure plate cluster is located by using the electrical protection pressure plate panel. The second level is distance partitioning, which also indirectly separates the predictive mask for each individual electrical protection pressure plate instance by partitioning the distance. This method enables individual identification of electrical protection pressure plates at the granular level, facilitating subsequent one-to-one intelligent activation / deactivation status detection. It effectively solves the detection problem of densely distributed electrical protection pressure plates, possesses extremely high detail description capabilities, and fully meets the data requirements for electrical protection pressure plate status detection and even other subsequent detections (such as pressure plate wear, pressure plate model, etc.).

[0091] In summary, regarding the intelligent activation / deactivation status detection requirement for electrical protection circuit breakers in substations, after acquiring the actual image I collected from the electrical protection circuit breaker cabinet in the substation, compared to the existing technology that directly identifies the activation / deactivation status of the electrical protection circuit breaker through image processing based on the actual image I, this application uses a semantic segmentation model M... 语义分割 Identify the predictive mask for the electrical protection pressure plate panel from actual image I. Then, based on the dividing distance between different electrical protection pressure plates, the predictive mask for the electrical protection pressure plate panel is created. Individual predictive masks for different electrical protection pressure plate instances At this point, the individual predictive mask for the electrical protection pressure plate instance is shown. Then, the image of the electrical protection pressure plate instance can be segmented from the actual image I. The individual recognition of the electrical protection pressure plate at the granular level can be performed in the dense electrical protection pressure plate cluster, thereby providing accurate detection objects for the intelligent activation and deactivation status detection of the electrical protection pressure plate. In this way, the intelligent activation and deactivation status detection of the electrical protection pressure plate at the granular level can be completed in the electrical protection pressure plate cluster.

[0092] Furthermore, we will elaborate on the possible implementation methods of the above scheme in practical applications and the possible extension schemes.

[0093] For the individual prediction mask of the electrical protection pressure plate instance. The segmentation distance involved in the process, as mentioned earlier, can be either preset or configured in real time. It can be understood that the segmentation distance determined in real time is more targeted and matches the actual image I being processed.

[0094] As a practical implementation, the real-time processing of the partition distance can include the following:

[0095] The actual image;

[0096] Based on the grayscale image, the division distance between different electrical protection pressure plate panels is calculated in both the horizontal and vertical directions using an adaptive thresholding strategy.

[0097] It is understood that this application believes that, in order to facilitate the automatic recognition of distance division based on images, the actual image I can be converted into a grayscale image first, thereby reducing its image content and presenting image content that is more relevant to the spacing of the electrical protection pressure plate panel.

[0098] Next, this application introduces a preset adaptive thresholding strategy. Based on the grayscale image content that is more related to the spacing of the electrical protection pressure plate panels, the application automatically calculates the division distance between different electrical protection pressure plate panels that are adapted to the current image content, thereby achieving a real-time processing effect of high matching and high precision division distance.

[0099] Furthermore, as mentioned earlier, the individual identification of the electrical protection pressure plate granularity is for the subsequent one-to-one intelligent activation / deactivation status detection of the electrical protection pressure plate. Therefore, in practical applications, as another specific implementation method, this application may also include intelligent activation / deactivation status detection after step S103, namely:

[0100] Predictive masks based on actual image I and individual examples of different electrical protection pressure plates. And image classification model M 图像分类 Identify the activation / deactivation status of each individual electrical protection pressure plate instance.

[0101] Through optical character recognition model M OCR The predicted coordinates of the upper left corner of each electrical protection pressure plate label were identified from the actual image I. Predicted coordinates in the lower right corner And text prediction based on labels;

[0102] Predict the upper left corner coordinates of each electrical protection pressure plate label. Predicted coordinates in the lower right corner And the predicted text of the labels, and the corresponding activation / deactivation status of each individual electrical protection pressure plate instance. By performing correlation, information about the electrical protection pressure plate can be obtained.

[0103] As can be seen from the above, for the intelligent activation / deactivation status detection of individual electrical protection pressure plates, this application can involve not only specific activation / deactivation statuses. The identification can also be further extended to the identification of the label text of individual electrical protection pressure plate instances, thus enabling richer detection of the working status of electrical protection pressure plates.

[0104] This label text can be understood as a textual description of an individual electrical protection pressure plate instance in practical applications, and is presented on the label of the individual electrical protection pressure plate instance through textual descriptions such as sticky notes.

[0105] In the process of intelligently detecting the on / off status of individual electrical protection pressure plates, a pre-trained image classification model M can be used. 图像分类 To accomplish this, it can be understood that in practical applications, the voltage protection plate's on / off state can specifically include two states: on and off. Therefore, it can be configured to identify whether the on or off state is matched using a classification model to achieve intelligent on / off state detection.

[0106] In the process of detecting the intelligent label text of individual electrical protection pressure plate instances, a pre-trained optical character recognition model M can be used. OCR To complete, with image classification model M 图像分类 Similarly, in practical applications, the specific label content displayed on the label of an individual electrical protection pressure plate instance can be determined using an optical character recognition model (OCR). OCR It is achieved through Optical Character Recognition (OCR).

[0107] Furthermore, the process of optical character recognition may also involve the detection of label positions, such as predicting the coordinates of the top left corner of the label. Predicted coordinates in the lower right corner It is understood that, corresponding to this application, in practical applications, the label at each instance of the electrical protection pressure plate can be configured as a regular rectangular shape. Thus, its position in the image can be indicated by two diagonal coordinates, where the coordinates can be predicted from the upper left corner. Predicted coordinates in the lower right corner The indication is understandable; by introducing location information, the information content of the electrical protection switchboard can be further enhanced. Furthermore, this enhances the processing of electrical protection switchboard information, specifically its associated activation / deactivation status. In the process of predicting text by tags, the accuracy of association processing can be further ensured by leveraging the positioning effect of location information.

[0108] Specifically, as another practical implementation method, the location-based operation involved in the association processing of electrical protection pressure plate information can include the following:

[0109] Calculate individual predictive masks for different electrical protection pressure plate instances pixel center point C i ;

[0110] Calculate the predicted coordinates of the upper left corner of each electrical protection pressure plate label. and the predicted coordinates in the lower right corner To the center point C of each pixel i Average distance D (i,j) ;

[0111] Using the latest algorithm, the predicted text of the label corresponding to each electrical protection pressure plate label is compared with the activation / deactivation status of each individual electrical protection pressure plate instance. By performing correlation, information about the electrical protection pressure plate can be obtained.

[0112] It is understandable that, in practical operation, this application uses the pixel center point C as the reference. i Based on this, combined with the predicted coordinates of the upper left corner and the predicted coordinates in the lower right corner To calculate the average distance D (i,j) Then, based on the nearest matching principle of the nearest algorithm, each group corresponds to the minimum average distance D. (i,j) Predicting text and surrender status using tags The content of the electrical protection pressure plate information as an individual electrical protection pressure plate instance.

[0113] Furthermore, as another practical implementation method, the activation / deactivation status of each individual electrical protection pressure plate instance is identified. In the specific process, compared to the image classification model M 图像分类 Trained to directly input actual image I and individual prediction masks for different electrical protection pressure plate instances In addition to the method for identifying the engagement / disengagement status, this application also has another preferred method to achieve a more lightweight engagement / disengagement status detection, which may include the following:

[0114] Predictive mask based on individual examples of different electrical protection pressure plates. Sub-images I corresponding to each individual electrical protection pressure plate instance are divided from the actual image I. i ;

[0115] Image classification model M 图像分类 Identify each subgraph I i Corresponding surrender / withdrawal status

[0116] It is understandable that, through image classification model M 图像分类 When detecting the activation / deactivation status, the input image of the electrical protection pressure plate particle size can be directly input, that is, the mask can be predicted through individual instances of the electrical protection pressure plate. The corresponding sub-image I divided from the actual image I i Thus, while ensuring that the input image contains only one instance of an electrical protection pressure plate, the image classification model M is improved. 图像分类 It can easily perform image recognition of the activation / deactivation status of input images without considering the matching problem between different activation / deactivation statuses and different electrical protection pressure plate instances, achieving more visual, lightweight, and stable image processing, which not only improves recognition efficiency but also ensures recognition accuracy.

[0117] Furthermore, regarding the three models mentioned above, namely semantic segmentation model M... 语义分割 Image classification model M 图像分类 And optical character recognition model M OCR In practice, it may also involve model training.

[0118] Specifically, it may also include the following model training content:

[0119] Acquire sample image I from the electrical protection pressure plate cabinet.

[0120] Content enhancement was performed on sample image I using random brightness adjustment, random dynamic blur, random flipping, and random rotation as data augmentation measures.

[0121] The sample image I is labeled with the corresponding sample electrical protection pressure plate panel prediction mask M and sample electrical protection pressure plate instance individual prediction mask. Sample submission / rejection status Predicted coordinates of the top left corner of the sample Predicted coordinates of the bottom right corner of the sample After predicting the text using sample labels, the semantic segmentation model M is trained using the sample image I and the sample electrical protection pressure plate panel prediction mask M. 语义分割 The mask is predicted using sample image I and sample electrical protection pressure plate instance. and sample submission / return status To train image classification model M 图像分类 The predicted coordinates are obtained from the actual image I of the sample and the top left corner of the sample. Predicted coordinates of the bottom right corner of the sample And sample labels predict text to train the optical character recognition model M OCR .

[0122] It is understandable that the input images involved in the specific training process of the model correspond to the specific image processing it implements, so no specific explanation will be given.

[0123] In layman's terms, the training process can be understood as follows: the sample information used to train the model is input into the initial model, the model performs image recognition to complete one forward propagation, and then the loss function is calculated based on the image recognition results output by the model and the corresponding image recognition annotation results. The model parameters are optimized based on the loss function calculation results to complete one backpropagation. After meeting the training requirements such as the number of training sessions, training time, and recognition accuracy, the model training is completed, and the model can then be put into practical use.

[0124] Using various data augmentation techniques such as random brightness adjustment, random dynamic blur, random flipping, and random rotation on sample image I can effectively increase the amount of data samples, thereby improving the robustness of the model and making the trained model have a more stable and accurate recognition effect.

[0125] Random brightness refers to adding electrical protection pressure plate images under various lighting scenarios by adjusting the V color channel of the HSV color model of the image.

[0126] Random dynamic blur refers to processing images using dynamic blur operators with arbitrary convolution kernel sizes to simulate scenes with slight shaking when the image is captured, thereby enhancing the adaptability of the final trained model to image shaking.

[0127] Random flipping and random rotation refer to images used to add various shooting angles to a picture.

[0128] For more details, please refer to Figure 3 The illustration shown provides a more vivid understanding of one scenario of the data enhancement measures proposed in this application.

[0129] The specific training process for different models can be understood by referring to the following example.

[0130] First, use a data annotation tool to annotate the information in sample image I (hereinafter referred to as electrical protection pressure plate image I), referring to... Figure 4 The illustration shown represents a scenario of annotation processing in this application. After annotation, the following information can be obtained:

[0131] 1. Image I of electrical protection pressure plate;

[0132] 2. Electrical protection pressure plate panel mask M;

[0133] 3. Example of an electrical protection pressure plate: Individual mask M i ;

[0134] 4. Individual Status Information S of Electrical Protection Pressure Plate Example i ∈[Entry, Exit];

[0135] 5. The top left corner of the individual label for the electrical protection pressure plate example has the following coordinates in the pixel coordinate system:

[0136] 6. The label in the lower right corner of the individual label for the electrical protection pressure plate example is in pixel coordinates.

[0137] 7. Example of an electrical protection pressure plate: Individual label text T i .

[0138] 1) Semantic segmentation model M 语义分割 Training processing

[0139] An image-mask pair is formed using the electrical protection pressure plate image I and the individual mask M of the electrical protection pressure plate instance, which is then used to train the semantic segmentation model M. 语义分割 During training, the numerical value of the semantic segmentation model loss function L is calculated and recorded. 语义分割 .

[0140] like Figure 5 The semantic segmentation model M of this application is shown. 语义分割 A schematic diagram of an exemplary model structure, which can be used to design semantic segmentation models using encoding and decoding structures:

[0141] The encoding structure consists of three convolutional modules used to extract image features;

[0142] The decoding structure consists of three upsampling modules, which are used to convert features into semantic segmentation masks;

[0143] In the encoding and decoding structures, the convolutional and upsampling modules complete the cross-module transmission of information through skip connections.

[0144] The convolutional module consists of two convolutional layers and one average pooling layer. The convolutional modules all use 3×3 convolutional kernels with a stride of 1. The ReLU algorithm is used at the end of the convolutional layers to perform non-linear activation on the output features. The average pooling layer uses 3×3 pooling convolutional kernels with a stride of 2.

[0145] The role of the convolution module is to extract high-dimensional features and reduce the feature resolution to improve the computational efficiency of the model.

[0146] The upsampling module fuses two input features. Input feature 1 is processed through an upsampling layer to adjust its resolution to match that of input feature 2. Then, the two features are superimposed pixel by pixel, passing through two convolutional layers with 3×3 kernels and a stride of 1, to obtain the output feature of the upsampling module. The function of the upsampling module is to represent the inherent information of high-dimensional features using low-dimensional features.

[0147] The loss function can be the cross-entropy loss function, whose mathematical definition is as follows:

[0148]

[0149] In the formula, i = 0, 1, 2, ..., N represents the pixel number, c = 0, 1, 2, ..., M represents the semantic segmentation category number, and y (i,c) This represents the one-hot encoded value of the annotation data for pixel i. If the true class of pixel i is equal to c, then y (i,c) =1, otherwise y (i,c) =0, p (i,c) This represents the predicted probability of pixel number i.

[0150] Set the expected threshold T of the loss function for the semantic segmentation model. 语义分割 If L 语义分割 <T 语义分割 If L = , it means the semantic segmentation model can correctly divide the area where the electrical protection pressure plate panel is located in the image, that is, the semantic segmentation model meets the requirements; if L 语义分割 >T 语义分割 If the result is negative, it means the semantic segmentation model cannot correctly classify the area where the electrical protection pressure plate panel is located in the image, meaning the semantic segmentation model does not yet meet the requirements. Training should continue until L... 语义分割 <T 语义分割 .

[0151] 2) Image classification model M 图像分类 Training processing

[0152] Example of using an electrical protection pressure plate: Individual mask Mi Extract the sub-image I corresponding to each individual instance from image I of the electrical protection pressure plate. i The individual sub-figure of the electrical protection pressure plate example I i Individual status information S of electrical protection pressure plate instance i Image-label pairs are formed to train the image classification model M. 图像分类 During training, the numerical value of the image classification model loss function L is calculated and recorded. 图像分类 .

[0153] like Figure 6 The image classification model M shown in this application 图像分类 The diagram illustrates an exemplary model structure. The main structure of the image classification model can be formed by stacking four residual modules, followed by a convolutional layer with a 1×1 kernel and a stride of 1, and a global average pooling layer to transform image features into image classification prediction categories. The residual modules are used to extract features, and their structure consists of two cascaded convolutional layers. The corresponding convolutional layers use 3×3 kernels and a stride of 1. After feature extraction by the convolutional layers, skip connections are used to superimpose the input features and the extracted features pixel by pixel to form the output features.

[0154] Image classification models can also use the cross-entropy loss function; please refer to the semantic segmentation model M mentioned earlier. 语义分割 The details described in the text will not be repeated here.

[0155] Set the expected threshold T of the loss function for the image classification model. 图像分类 If L 图像分类 <T 图像分类 If L is true, it means the image classification model can correctly distinguish the current on / off state of the electrical protection pressure plate instance in the sub-image, i.e., the image classification model meets the requirements; if L 图像分类 >T 图像分类 If the image classification model fails to correctly distinguish the current on / off state of the electrical protection pressure plate instance in the sub-image, it means the image classification model does not yet meet the requirements. Training should continue until L... 图像分类 <T 图像分类 .

[0156] 3) Optical Character Recognition Model M OCR Training processing

[0157] Using image I of an electrical protection pressure plate, and the top left corner of the individual label for an example of an electrical protection pressure plate, the coordinates in the pixel coordinate system are as follows: The label for each electrical protection pressure plate instance is located in the lower right corner in pixel coordinates. and electrical protection pressure plate example individual label text T iImage-detection box pairs are used to train the optical character recognition model M. OCR During the training process, the numerical value of the OCR model loss function L is calculated and recorded. OCR .

[0158] like Figure 7 The optical character recognition model M of this application is shown. OCR A flowchart illustrating optical character recognition, optical character recognition model M OCR A two-stage design is adopted. In the first stage, the YOLOv5 model can be used to detect text regions in the image. In the second stage, the CRNN model is used, taking the text regions selected in the first stage as input, and performing sequence processing through bidirectional LSTM to convert the image within the text regions into a string.

[0159] Set the expected threshold T of the loss function for the OCR model. OCR If L OCR <T OCR If L is true, it means the OCR model can correctly identify the area where the individual label of the electrical protection pressure plate instance is located and the corresponding text information in the label, that is, the OCR model meets the requirements; if L OCR >T OCR If the result is negative, it means the OCR model cannot correctly identify the area where the individual label of the electrical protection pressure plate instance is located and the corresponding text information in the label. In other words, the OCR model does not yet meet the requirements and training should continue until L is reached. OCR <T OCR .

[0160] Furthermore, as another practical implementation method, after obtaining the electrical protection pressure plate information through correlation processing, it is also possible to trigger corresponding archiving processing based on the application scenario of electrical protection pressure plate management in substations, i.e.:

[0161] The electrical protection pressure plate information is archived in the system according to the ledger information organization method.

[0162] It is understandable that, corresponding to the ledger management method for electrical protection pressure plates in the substation, the electrical protection pressure plate information can be updated to achieve the granular ledger management accuracy of electrical protection pressure plates in the system, thereby facilitating high-precision status monitoring and maintenance of each electrical protection pressure plate in the substation.

[0163] The above is an introduction to the substation electrical protection pressure plate testing method provided in this application. In order to facilitate better implementation of the substation electrical protection pressure plate testing method provided in this application, this application also provides a substation electrical protection pressure plate testing device from the perspective of functional modules.

[0164] See Figure 8 , Figure 8This is a schematic diagram of a substation electrical protection pressure plate testing device according to this application. In this application, the substation electrical protection pressure plate testing device 800 may specifically include the following structure:

[0165] The acquisition unit 801 is used to acquire the actual image I collected from the electrical protection pressure plate cabinet of the substation, wherein the electrical protection pressure plate cabinet is configured with different electrical protection pressure plates in the form of a panel;

[0166] Semantic segmentation unit 802 is used to segment semantic segmentation model M 语义分割 Identify the predictive mask for the electrical protection pressure plate panel from actual image I. Among them, electrical protection pressure plate panel predictive mask Used to indicate the area where the electrical protection pressure plate panel is located, segmented from the actual image I;

[0167] The dividing unit 803 is used to divide the electrical protection pressure plate panel prediction mask according to the dividing distance between different electrical protection pressure plates. Individual predictive masks for different electrical protection pressure plate instances

[0168] In yet another exemplary implementation, the apparatus further includes:

[0169] The deployment / retraction status recognition unit 804 is used to predict the mask based on the actual image I and individual instances of different electrical protection pressure plates. And image classification model M 图像分类 Identify the activation / deactivation status of each individual electrical protection pressure plate instance.

[0170] Optical character recognition unit 805, used for optical character recognition model M OCR The predicted coordinates of the upper left corner of each electrical protection pressure plate label were identified from the actual image I. Predicted coordinates in the lower right corner And text prediction based on labels;

[0171] The association unit 806 is used to predict the upper left corner coordinates of each electrical protection pressure plate label. Predicted coordinates in the lower right corner And the predicted text of the labels, and the corresponding activation / deactivation status of each individual electrical protection pressure plate instance. By performing correlation, information about the electrical protection pressure plate can be obtained.

[0172] In yet another exemplary implementation, the deployment / retreat status identification unit 804 is specifically used for:

[0173] Predictive mask based on individual examples of different electrical protection pressure plates. Sub-images I corresponding to each individual electrical protection pressure plate instance are divided from the actual image I. i ;

[0174] Image classification model M 图像分类 Identify each subgraph I i Corresponding surrender / withdrawal status

[0175] In yet another exemplary implementation, the association unit 806 is specifically used for:

[0176] Calculate individual predictive masks for different electrical protection pressure plate instances pixel center point C i ;

[0177] Calculate the predicted coordinates of the upper left corner of each electrical protection pressure plate label. and the predicted coordinates in the lower right corner To the center point C of each pixel i Average distance D (i,j) ;

[0178] Using the latest algorithm, the predicted text of the label corresponding to each electrical protection pressure plate label is compared with the activation / deactivation status of each individual electrical protection pressure plate instance. By performing correlation, information about the electrical protection pressure plate can be obtained.

[0179] In yet another exemplary implementation, the apparatus further includes a model training unit 807, for:

[0180] Acquire sample image I from the electrical protection pressure plate cabinet.

[0181] Content enhancement was performed on sample image I using random brightness adjustment, random dynamic blur, random flipping, and random rotation as data augmentation measures.

[0182] The sample image I is labeled with the corresponding sample electrical protection pressure plate panel prediction mask M and sample electrical protection pressure plate instance individual prediction mask. Sample submission / rejection status Predicted coordinates of the top left corner of the sample Predicted coordinates of the bottom right corner of the sample After predicting the text using sample labels, the semantic segmentation model M is trained using the sample image I and the sample electrical protection pressure plate panel prediction mask M. 语义分割 The mask is predicted using sample image I and sample electrical protection pressure plate instance. and sample submission / return status To train image classification model M 图像分类 The predicted coordinates are obtained from the actual image I of the sample and the top left corner of the sample. Predicted coordinates of the bottom right corner of the sample And sample labels predict text to train the optical character recognition model M OCR .

[0183] In yet another exemplary implementation, the apparatus further includes an archiving unit 808, for:

[0184] The electrical protection pressure plate information is archived in the system according to the ledger information organization method.

[0185] In yet another exemplary implementation, the partitioning unit 803 is further configured to:

[0186] Convert the actual image I into a grayscale image;

[0187] Based on the grayscale image, the division distance between different electrical protection pressure plate panels is calculated in both the horizontal and vertical directions using an adaptive thresholding strategy.

[0188] This application also provides a processing device from a hardware architecture perspective, see [link / reference]. Figure 9 , Figure 9 This diagram illustrates a structural schematic of the processing device of this application. Specifically, the processing device may include a processor 901, a memory 902, and an input / output device 903. The processor 901 executes the computer program stored in the memory 902 to implement, for example... Figure 1 The steps of the method in the corresponding embodiment; or, when the processor 901 executes the computer program stored in the memory 902, it implements as follows: Figure 8 Corresponding to the functions of each unit in the embodiment, the memory 902 is used to store the functions executed by the processor 901 as described above. Figure 1 The computer program required for the method in the corresponding embodiment.

[0189] For example, a computer program may be divided into one or more modules / units, one or more of which are stored in memory 902 and executed by processor 901 to complete this application. One or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in a computer device.

[0190] The processing device may include, but is not limited to, processor 901, memory 902, and input / output device 903. Those skilled in the art will understand that the illustrations are merely examples of the processing device and do not constitute a limitation on the processing device. It may include more or fewer components than illustrated, or combine certain components, or different components. For example, the processing device may also include network access devices, buses, etc., and processor 901, memory 902, input / output device 903, etc., are connected via a bus.

[0191] The processor 901 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the processing device, connecting various parts of the device through various interfaces and lines.

[0192] The memory 902 can be used to store computer programs and / or modules. The processor 901 implements various functions of the computer device by running or executing the computer programs and / or modules stored in the memory 902 and by calling data stored in the memory 902. The memory 902 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function, etc.; the data storage area may store data created according to the use of the processing device, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0193] When processor 901 executes a computer program stored in memory 902, it can specifically perform the following functions:

[0194] Acquire actual image I from the electrical protection pressure plate cabinet of the substation, wherein the electrical protection pressure plate cabinet is configured with different electrical protection pressure plates in the form of a panel;

[0195] Using semantic segmentation model M 语义分割 Identify the predictive mask for the electrical protection pressure plate panel from actual image I. Among them, the electrical protection pressure plate panel predictive mask Used to indicate the area where the electrical protection pressure plate panel is located, segmented from the actual image I;

[0196] Based on the dividing distance between different electrical protection pressure plates, predict the mask of the electrical protection pressure plate panel. Individual predictive masks for different electrical protection pressure plate instances

[0197] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the substation electrical protection pressure plate detection device, processing equipment, and its corresponding units described above can be found in [reference to...]. Figure 1 The description of the substation electrical protection pressure plate detection method in the corresponding embodiment will not be repeated here.

[0198] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0199] Therefore, this application provides a computer-readable storage medium storing a plurality of instructions that can be loaded by a processor to execute the present application. Figure 1 The steps of the substation electrical protection pressure plate detection method in the corresponding embodiment can be referred to as follows for specific operations. Figure 1 The description of the substation electrical protection pressure plate detection method in the corresponding embodiment will not be repeated here.

[0200] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0201] Because of the instructions stored in the computer-readable storage medium, the present application can be executed as described above. Figure 1 The steps of the substation electrical protection pressure plate detection method in the corresponding embodiment can therefore achieve the results of this application. Figure 1 The beneficial effects that the substation electrical protection pressure plate detection method can achieve in the corresponding embodiment are detailed in the preceding description and will not be repeated here.

[0202] The above provides a detailed description of the substation electrical protection pressure plate detection method, device, processing equipment, and computer-readable storage medium provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for testing electrical protection pressure plates in substations, characterized in that, The method includes: Acquire actual images from the electrical protection pressure plate cabinet of the substation. The electrical protection pressure plate cabinet is equipped with different electrical protection pressure plates in the form of a panel; Through semantic segmentation model From the actual image Predictive mask for electrical protection pressure plate panel identified in the middle The electrical protection pressure plate panel predictive mask is described above. Used to indicate the actual image The area where the electrical protection pressure plate panel is separated; Based on the dividing distance between different electrical protection pressure plates, the electrical protection pressure plate panel prediction mask is used. Individual predictive masks for different electrical protection pressure plate instances ; The electrical protection pressure plate panel prediction mask is prepared according to the division distance between different electrical protection pressure plates. Individual predictive masks for different electrical protection pressure plate instances Subsequently, the method further includes: Through the actual image Individual predictive masks for different electrical protection pressure plate examples and image classification models Identify the activation / deactivation status of each individual electrical protection pressure plate instance. ; Through optical character recognition model From the actual image The predicted coordinates of the upper left corner of each electrical protection pressure plate label were identified. Predicted coordinates in the lower right corner And text prediction based on labels; The predicted coordinates of the upper left corner corresponding to each electrical protection pressure plate label. The predicted coordinates of the lower right corner And the predicted text of the label, and the activation / deactivation status corresponding to each individual electrical protection pressure plate instance. By performing correlation, information about the electrical protection pressure plate can be obtained; For the semantic segmentation model Using an encoding / decoding structure design, we have: The encoding structure consists of three convolutional modules used to extract image features; The decoding structure consists of three upsampling modules, which are used to convert features into semantic segmentation masks; In the encoding and decoding structures, the convolution module and the upsampling module complete the cross-module transmission of information through skip connections. The convolutional module consists of two convolutional layers and one average pooling layer. The convolutional kernel has a stride of 1. The ReLU algorithm is used at the end of the convolutional layer to non-linearly activate the output features. The average pooling layer... Pooling convolution kernel, 2 convolution strides; The function of the convolution module is to extract high-dimensional features and reduce the feature resolution to improve the model's computational efficiency. The upsampling module fuses two input features. Input feature 1 is processed through an upsampling layer to adjust its resolution to be the same as that of input feature 2. Then, the two features are superimposed pixel by pixel, and the process is repeated through two layers. After the convolutional layer with a convolutional kernel and a stride of 1, the output features of the upsampling module are obtained. The function of the upsampling module is to represent the information contained in the high-dimensional features using low-dimensional features. For the image classification model ,have: The main structure is formed by stacking four layers of residual modules, followed by connecting one layer. Convolutional layers with convolutional kernels, a stride of 1, and global average pooling layers are used to transform image features into image classification prediction categories. The residual module is used to extract features. The structure consists of... The system consists of two convolutional layers with a convolutional kernel and a convolutional stride of 1. After feature extraction by the convolutional layers, skip connections are used to combine the input features and the extracted features in a pixel-by-pixel manner to form the output features. For the optical character recognition model ,have: A two-stage design is adopted. In the first stage, the YOLOv5 model is used to detect text regions in the image. In the second stage, the CRNN model is used, taking the text regions selected in the first stage as input, and performing sequence processing through bidirectional LSTM to convert the image within the text region into a string.

2. The method according to claim 1, characterized in that, The actual image Individual predictive masks for different electrical protection pressure plate examples and image classification models Identify the activation / deactivation status of each individual electrical protection pressure plate instance. ,include: Individual prediction masks based on the different electrical protection pressure plate examples From the actual image The sub-graphs corresponding to each individual example of the electrical protection pressure plate are divided into sections. ; Through the image classification model Identify each of the subgraphs. The corresponding deployment / retreat status .

3. The method according to claim 1, characterized in that, The predicted coordinates of the upper left corner corresponding to each electrical protection pressure plate label are then determined. The predicted coordinates of the lower right corner And the predicted text of the label, and the activation / deactivation status corresponding to each individual electrical protection pressure plate instance. By performing correlation, information about the electrical protection pressure plate is obtained, including: Calculate the individual predictive mask for the different electrical protection pressure plate instances. pixel center point ; Calculate the predicted coordinates of the upper left corner corresponding to each electrical protection pressure plate label. and the predicted coordinates of the lower right corner To the center point of each pixel average distance ; Using the nearest algorithm, the predicted text of the label corresponding to each electrical protection pressure plate label is compared with the activation / deactivation status corresponding to each individual electrical protection pressure plate instance. By performing correlation, the electrical protection pressure plate information is obtained.

4. The method according to claim 1, characterized in that, The method further includes: Acquire sample images from the electrical protection pressure plate cabinet. , The sample images were augmented using random brightness adjustment, random motion blur, random flipping, and random rotation. Enhance the content; The sample image to be obtained The corresponding sample electrical protection pressure plate panel prediction mask is labeled. Sample electrical protection pressure plate instance individual prediction mask Sample submission / rejection status Predicted coordinates of the top left corner of the sample Predicted coordinates of the bottom right corner of the sample After predicting text using sample labels, the sample images are used. and the sample electrical protection pressure plate panel prediction mask To train the semantic segmentation model Through the sample image The sample electrical protection pressure plate instance individual prediction mask and the sample submission / rejection status To train the image classification model Through the sample image The predicted coordinates of the upper left corner of the sample The predicted coordinates of the lower right corner of the sample The optical character recognition model is trained using the sample label predicted text. .

5. The method according to any one of claims 1 to 4, characterized in that, After obtaining the electrical protection pressure plate information, the method further includes: The electrical protection pressure plate information is archived in the system according to the ledger information organization method.

6. The method according to claim 1, characterized in that, Based on the division distance between different electrical protection pressure plates, the electrical protection pressure plate panel prediction mask is used. Individual predictive masks for different electrical protection pressure plate instances Previously, the method also included: The actual image Convert to grayscale image; Based on the grayscale image, the division distance between the different electrical protection pressure plates is calculated in both the horizontal and vertical directions using an adaptive thresholding strategy.

7. A substation electrical protection pressure plate detection device, characterized in that, The device includes: The acquisition unit is used to acquire actual images collected from the electrical protection pressure plate cabinet of the substation. The electrical protection pressure plate cabinet is equipped with different electrical protection pressure plates in the form of a panel; Semantic segmentation unit, used to perform semantic segmentation model From the actual image Predictive mask for electrical protection pressure plate panel identified in the middle The electrical protection pressure plate panel predictive mask is described above. Used to indicate the actual image The area where the electrical protection pressure plate panel is separated; A dividing unit is used to divide the electrical protection pressure plate panel prediction mask according to the dividing distance between different electrical protection pressure plates. Individual predictive masks for different electrical protection pressure plate instances ; The device further includes: The deployment / retreat status recognition unit is used to identify the actual image. Individual predictive masks for different electrical protection pressure plate examples and image classification models Identify the activation / deactivation status of each individual electrical protection pressure plate instance. ; Optical character recognition unit, used to recognize optical characters through an optical character recognition model From the actual image The predicted coordinates of the upper left corner of each electrical protection pressure plate label were identified. Predicted coordinates in the lower right corner And text prediction based on labels; The association unit is used to predict the upper left corner coordinates of each electrical protection pressure plate label. The predicted coordinates of the lower right corner And the predicted text of the label, and the activation / deactivation status corresponding to each individual electrical protection pressure plate instance. By performing correlation, information about the electrical protection pressure plate can be obtained; For the semantic segmentation model Using an encoding / decoding structure design, we have: The encoding structure consists of three convolutional modules used to extract image features; The decoding structure consists of three upsampling modules, which are used to convert features into semantic segmentation masks; In the encoding and decoding structures, the convolution module and the upsampling module complete the cross-module transmission of information through skip connections. The convolutional module consists of two convolutional layers and one average pooling layer. The convolutional kernel has a stride of 1. The ReLU algorithm is used at the end of the convolutional layer to non-linearly activate the output features. The average pooling layer... Pooling convolution kernel, 2 convolution strides; The function of the convolution module is to extract high-dimensional features and reduce the feature resolution to improve the model's computational efficiency. The upsampling module fuses two input features. Input feature 1 is processed through an upsampling layer to adjust its resolution to be the same as that of input feature 2. Then, the two features are superimposed pixel by pixel, and the process is repeated through two layers. After the convolutional layer with a convolutional kernel and a stride of 1, the output features of the upsampling module are obtained. The function of the upsampling module is to represent the information contained in the high-dimensional features using low-dimensional features. For the image classification model ,have: The main structure is formed by stacking four layers of residual modules, followed by connecting one layer. Convolutional layers with convolutional kernels, a stride of 1, and global average pooling layers are used to transform image features into image classification prediction categories. The residual module is used to extract features. The structure consists of... The system consists of two convolutional layers with a convolutional kernel and a convolutional stride of 1. After feature extraction by the convolutional layers, skip connections are used to combine the input features and the extracted features in a pixel-by-pixel manner to form the output features. For the optical character recognition model ,have: A two-stage design is adopted. In the first stage, the YOLOv5 model is used to detect text regions in the image. In the second stage, the CRNN model is used, taking the text regions selected in the first stage as input, and performing sequence processing through bidirectional LSTM to convert the image within the text region into a string.

8. A processing apparatus, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, and the processor executes the method as described in any one of claims 1 to 6 when it invokes the computer program in the memory.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions adapted for loading by a processor to perform the method of any one of claims 1 to 6.

Images