Substation pressing plate identification method and device, computer equipment and readable storage medium

Abstract

The invention relates to a transformer substation pressing plate identification method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: acquiring a real-time pressure plate image acquired from a target substation, and searching a reference pressure plate image corresponding to the real-time pressure plate image; performing pressing plate detection processing on the real-time pressing plate image through a preset pressing plate detection model to obtain a pressing plate detection frame image, and performing pressing plate detection processing on the reference pressing plate image through a preset reference detection model to obtain a reference detection frame image; performing clustering processing on the pressing plate detection frame image and the reference detection frame image to obtain first pressing plate position information and second pressing plate position information; and performing state comparison processing based on the first pressing plate position information and the second pressing plate position information, and identifying a pressing plate in an abnormal state in the real-time pressing plate image. According to the invention, the accuracy and robustness of the pressing plate identification and detection process can be effectively improved, so that the pressing plate state detection accuracy is effectively improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method, device, computer equipment, computer-readable storage medium and computer program product for identifying a pressure plate in a substation. Background Art

[0002] The transformer substation pressure plate switches are usually located on the transformer cabinet and transformer box, and usually have three states: open, closed, and standby. There are generally dozens of pressure plate switches arranged on a transformer cabinet. At any time, these pressure plate switches have their specific switch states to ensure the smooth operation of the power facilities. When the switch state needs to be adjusted, the staff will manually adjust the state of the corresponding switch according to the template.

[0003] However, due to visual fatigue of operators, some switches may be misoperated. Therefore, in order to reduce the probability of misoperation by operators and reduce workload, computer intelligent recognition technology has been introduced into the recognition scenario of pressure plate switches. At present, the mainstream technology mainly detects pressure plates based on traditional image processing technology. However, this method has a low detection accuracy and is difficult to ensure the abnormal recognition effect of the pressure plate status, which affects the normal operation of the substation. Summary of the invention

[0004] Based on this, it is necessary to provide a substation pressure plate identification method, device, computer equipment, computer readable storage medium and computer program product that can effectively improve the accuracy of pressure plate status detection in response to the above technical problems.

[0005] In a first aspect, the present application provides a substation pressure plate identification method, comprising:

[0006] Acquire a real-time pressure plate image collected from a target substation, and search for a reference pressure plate image corresponding to the real-time pressure plate image;

[0007] The real-time pressure plate image is subjected to pressure plate detection processing by a preset pressure plate detection model to obtain a pressure plate detection frame image, and the reference pressure plate image is subjected to pressure plate detection processing by a preset reference detection model to obtain a reference detection frame image, wherein the preset image detection model is obtained by training an initial YOLOv5 model with historical pressure plate images collected from the target substation, wherein the initial YOLOv5 model includes a small-scale detection head and uses EIoU as a coordinate loss function, and the preset reference detection model is obtained by training the initial YOLOv5 model with various types of reference pressure plate images;

[0008] Performing clustering processing on the pressure plate detection frame image and the reference detection frame image respectively to obtain first pressure plate position information corresponding to the pressure plate detection frame image and second pressure plate position information corresponding to the reference detection frame image;

[0009] A state comparison process is performed based on the first pressure plate position information and the second pressure plate position information to identify a pressure plate with an abnormal state in the real-time pressure plate image.

[0010] In one embodiment, performing pressure plate detection processing on the real-time pressure plate image by using a preset pressure plate detection model to obtain a pressure plate detection frame image includes:

[0011] Performing feature extraction processing on the real-time pressing plate image through a backbone network of a preset pressing plate detection model to obtain an original pressing plate image feature map;

[0012] Performing multi-scale feature fusion processing on the original pressure plate image feature map through the neck network of the preset pressure plate detection model to obtain a pressure plate image fusion feature map;

[0013] The pressure plate image fusion feature map is output predicted and processed by the detection head network of the preset pressure plate detection model to obtain a pressure plate detection frame image.

[0014] In one embodiment, clustering the platen detection frame image and the reference detection frame image respectively to obtain first platen position information corresponding to the platen detection frame image and second platen position information corresponding to the reference detection frame image includes:

[0015] Performing row-column clustering processing on the platen detection frame image by a density-based clustering algorithm to obtain a first row clustering result and a first column clustering result of the platen detection frame image, and performing row-column clustering processing on the reference detection frame image by a density-based clustering algorithm to obtain a second row clustering result and a second column clustering result of the reference detection frame image;

[0016] Based on the first row clustering result and the first column clustering result, the row position and column position of each pressure plate detection frame image in the real-time pressure plate image are determined to obtain the first pressure plate position information; based on the second row clustering result and the second column clustering result, the row position and column position of each reference detection frame image in the reference pressure plate image are determined to obtain the second pressure plate position information.

[0017] In one embodiment, performing state comparison processing based on the first pressure plate position information and the second pressure plate position information to identify a pressure plate with an abnormal state in the real-time pressure plate image includes:

[0018] Determine a platen detection frame image and a reference detection frame image having the same row position and column position based on the first platen position information and the second platen position information;

[0019] The state comparison processing is performed on the pressure plate detection frame image and the reference detection frame image with the same row position and column position, and the pressure plate detection frame image with a different state from the reference detection frame image is determined to obtain the abnormal state pressure plate in the real-time pressure plate image.

[0020] In one embodiment, the method further comprises:

[0021] Acquire historical pressure plate images collected from the target substation;

[0022] Resizing the historical platen image to obtain an original sample image with the same specifications;

[0023] Annotating the switch states of each pressing plate in the original sample image in the form of a rectangular frame to obtain a training sample image;

[0024] The initial YOLOv5 model is trained based on the training sample images to obtain a preset image detection model.

[0025] In one embodiment, the method further comprises:

[0026] Acquire a trained preset pressure plate detection model, and convert the preset pressure plate detection model into an open neural network exchange format;

[0027] Converting the preset platen detection model in the open neural network exchange format into a binary model file by a model conversion tool;

[0028] The binary model file is deployed on a mobile terminal for substation pressure plate identification.

[0029] In a second aspect, the present application also provides a substation pressure plate identification device, comprising:

[0030] An image acquisition module is used to acquire a real-time pressure plate image collected from a target substation and search for a reference pressure plate image corresponding to the real-time pressure plate image;

[0031] A pressure plate detection module, used to perform pressure plate detection processing on the real-time pressure plate image through a preset pressure plate detection model to obtain a pressure plate detection frame image, and perform pressure plate detection processing on the reference pressure plate image through a preset reference detection model to obtain a reference detection frame image, wherein the preset image detection model is obtained by training an initial YOLOv5 model with historical pressure plate images collected from the target substation, wherein the initial YOLOv5 model includes a small-scale detection head and uses EIoU as a coordinate loss function, and the preset reference detection model is obtained by training the initial YOLOv5 model with various types of reference pressure plate images;

[0032] A clustering processing module, used for performing clustering processing on the pressure plate detection frame image and the reference detection frame image respectively, to obtain first pressure plate position information corresponding to the pressure plate detection frame image and second pressure plate position information corresponding to the reference detection frame image;

[0033] The state recognition module is used to perform state comparison processing based on the first pressure plate position information and the second pressure plate position information to identify the pressure plate with abnormal state in the real-time pressure plate image.

[0034] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0035] Acquire a real-time pressure plate image collected from a target substation, and search for a reference pressure plate image corresponding to the real-time pressure plate image;

[0036] The real-time pressure plate image is subjected to pressure plate detection processing by a preset pressure plate detection model to obtain a pressure plate detection frame image, and the reference pressure plate image is subjected to pressure plate detection processing by a preset reference detection model to obtain a reference detection frame image, wherein the preset image detection model is obtained by training an initial YOLOv5 model with historical pressure plate images collected from the target substation, wherein the initial YOLOv5 model includes a small-scale detection head and uses EIoU as a coordinate loss function, and the preset reference detection model is obtained by training the initial YOLOv5 model with various types of reference pressure plate images;

[0037] Performing clustering processing on the pressure plate detection frame image and the reference detection frame image respectively to obtain first pressure plate position information corresponding to the pressure plate detection frame image and second pressure plate position information corresponding to the reference detection frame image;

[0038] A state comparison process is performed based on the first pressure plate position information and the second pressure plate position information to identify a pressure plate with an abnormal state in the real-time pressure plate image.

[0039] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0040] Acquire a real-time pressure plate image collected from a target substation, and search for a reference pressure plate image corresponding to the real-time pressure plate image;

[0041] The real-time pressure plate image is subjected to pressure plate detection processing by a preset pressure plate detection model to obtain a pressure plate detection frame image, and the reference pressure plate image is subjected to pressure plate detection processing by a preset reference detection model to obtain a reference detection frame image, wherein the preset image detection model is obtained by training an initial YOLOv5 model with historical pressure plate images collected from the target substation, wherein the initial YOLOv5 model includes a small-scale detection head and uses EIoU as a coordinate loss function, and the preset reference detection model is obtained by training the initial YOLOv5 model with various types of reference pressure plate images;

[0042] Performing clustering processing on the pressure plate detection frame image and the reference detection frame image respectively to obtain first pressure plate position information corresponding to the pressure plate detection frame image and second pressure plate position information corresponding to the reference detection frame image;

[0043] A state comparison process is performed based on the first pressure plate position information and the second pressure plate position information to identify a pressure plate with an abnormal state in the real-time pressure plate image.

[0044] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0045] Acquire a real-time pressure plate image collected from a target substation, and search for a reference pressure plate image corresponding to the real-time pressure plate image;

[0046] The real-time pressure plate image is subjected to pressure plate detection processing by a preset pressure plate detection model to obtain a pressure plate detection frame image, and the reference pressure plate image is subjected to pressure plate detection processing by a preset reference detection model to obtain a reference detection frame image, wherein the preset image detection model is obtained by training an initial YOLOv5 model with historical pressure plate images collected from the target substation, wherein the initial YOLOv5 model includes a small-scale detection head and uses EIoU as a coordinate loss function, and the preset reference detection model is obtained by training the initial YOLOv5 model with various types of reference pressure plate images;

[0047] Performing clustering processing on the pressure plate detection frame image and the reference detection frame image respectively to obtain first pressure plate position information corresponding to the pressure plate detection frame image and second pressure plate position information corresponding to the reference detection frame image;

[0048] A state comparison process is performed based on the first pressure plate position information and the second pressure plate position information to identify a pressure plate with an abnormal state in the real-time pressure plate image.

[0049] The above-mentioned substation pressure plate identification method, device, computer equipment, computer-readable storage medium and computer program product first obtain the real-time pressure plate image collected from the target substation, and find the reference pressure plate image corresponding to the real-time pressure plate image, so as to judge whether the real-time pressure plate state is abnormal through the reference pressure plate image, and then perform pressure plate detection processing on the real-time pressure plate image through a preset pressure plate detection model to obtain a pressure plate detection frame image, and perform pressure plate detection processing on the reference pressure plate image through a preset reference detection model to obtain a reference detection frame image, and in the detection process, the detection model is obtained by training the initial YOLOv5 model containing a small-scale detection head and using EIoU as the coordinate loss function, and the wide loss plus the high loss of EioU effectively measure the important information of the target detection process, and the small-scale detection head is used to accurately detect the pressure plate as a small target, and finally the detected pressure plate detection frame image and the reference detection frame image are clustered respectively to determine the pressure plate position information of the two. Finally, based on the position information, the real-time pressure plate state on the real-time pressure plate image and the reference pressure plate state on the reference pressure plate image are compared to identify the pressure plate with abnormal state. By improving the YOLOv5 model, the accuracy and robustness of the pressure plate recognition and detection process can be effectively improved, thereby effectively improving the accuracy of pressure plate state detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0051] Figure 1 A diagram showing an application environment of a method for identifying a pressure plate in a substation in an embodiment;

[0052] Figure 2 A schematic diagram of a process of identifying a pressure plate in a substation according to an embodiment;

[0053] Figure 3 is a schematic diagram of a real-time platen image sample in one embodiment;

[0054] Figure 4 is a schematic diagram of a reference platen image in one embodiment;

[0055] Figure 5 A schematic diagram of the original network structure of YOLOv5 in one embodiment;

[0056] Figure 6 A schematic diagram of a YOLOv5 network structure after adding a small target detection head in one embodiment;

[0057] Figure 7 It is a schematic diagram of an interface for realizing pressure plate detection on a mobile terminal in one embodiment;

[0058] Figure 8 A schematic diagram of a complete detection process in one embodiment;

[0059] Fig. 9 A schematic diagram of a loss situation in a training process of a preset pressure plate detection model in an embodiment;

[0060] Fig.10 A schematic diagram of verifying the accuracy of a preset pressure plate detection model training process in one embodiment;

[0061] Fig.11 A schematic diagram showing a comparison of the accuracy and recall rate of platen recognition between the models before and after improvement in one embodiment;

[0062] Fig.12 A schematic diagram of a loss situation in a training process of a preset benchmark detection model in one embodiment;

[0063] Fig.13 A schematic diagram of verifying accuracy of a preset benchmark detection model training process in one embodiment;

[0064] Fig.14 A schematic diagram of comparing reasoning results in one embodiment;

[0065] Fig.15 is a structural block diagram of a substation pressure plate identification device in one embodiment;

[0066] Fig.16 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0068] The substation pressure plate identification method provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. When the employees of the substation need to accurately implement the pressure plate status identification processing, the substation pressure plate identification method of the present application can be used for identification. At this time, the employees can train the preset pressure plate detection model and the preset benchmark detection model through the server 104, and then deploy these two models to the terminal 102. The employees can use the terminal 102 to identify the substation pressure plate. The terminal 102 can obtain the real-time pressure plate image collected from the target substation, and search for the reference pressure plate image corresponding to the real-time pressure plate image; perform pressure plate detection processing on the real-time pressure plate image through the preset pressure plate detection model to obtain the pressure plate detection frame image; perform pressure plate detection processing on the reference pressure plate image through the preset reference detection model to obtain the reference detection frame image; the preset image detection model is obtained by training the initial YOLOv5 model through the historical pressure plate images collected from the target substation, the initial YOLOv5 model includes a small-scale detection head and uses EIoU as the coordinate loss function; the preset reference detection model is obtained by training the initial YOLOv5 model through various types of reference pressure plate images; cluster the pressure plate detection frame image and the reference detection frame image respectively to obtain the first pressure plate position information corresponding to the pressure plate detection frame image, and the second pressure plate position information corresponding to the reference detection frame image; perform state comparison processing based on the first pressure plate position information and the second pressure plate position information to identify the abnormal pressure plate in the real-time pressure plate image. Employees can re-inspect the detected abnormal pressure plate and correct its state to ensure the normal operation of the substation. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car devices, projection devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. Head-mounted devices can be virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. The server 104 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services.

[0069] In an exemplary embodiment, Figure 2 As shown, a substation pressure plate identification method is provided, and the method is applied to Figure 1 The terminal 102 in the example is used for explanation. Of course, the method can also be applied to the server. When the method is applied to the terminal 102, the following steps 201 to 207 are included. Among them:

[0070] Step 201, obtain a real-time pressure plate image collected from a target substation, and search for a reference pressure plate image corresponding to the real-time pressure plate image.

[0071] The target substation is the substation where the pressure plate detected by the substation pressure plate identification method of the present application is located. The target substation contains power equipment such as substation cabinets and substation boxes, and the real-time pressure plate image is the image captured from the power equipment such as substation cabinets and substation boxes. Figure 3 The reference platen image corresponding to the real-time platen image refers to a template image corresponding to each real-time platen image. The reference platen image is marked with the status of each platen switch. The reference platen image can be referred to Figure 4 As shown. In one embodiment, the reference pressure plate image can select the image in the "Four Meters and One Guide" of the substation. The "Four Meters and One Guide" is a general term for voltmeter, ammeter, active power meter, reactive power meter and indicator, also known as the four meters and one indicator of the power transformer. The guide can be used for substation equipment installation and commissioning, operation monitoring and management. Generally, when the pressure plate state needs to be adjusted, the substation staff can manually adjust the pressure plate by comparing with the reference pressure plate image. The present application is suitable for state judgment of the adjusted pressure plate state to prevent errors in pressure plate adjustment.

[0072] For example, after the substation staff changes the state of the pressure plate, in order to ensure that there is no error in the adjustment of the pressure plate state, the state of the pressure plate can be identified by the substation pressure plate identification method of the present application. First, the substation staff can collect the real-time pressure plate image after the state change through the terminal 102, such as by taking a photo, and at the same time specify the reference pressure plate image corresponding to the real-time pressure plate image in the identification system. Then the terminal 102 can search for the reference pressure plate image to identify the pressure plate of the substation.

[0073] Step 203, perform pressure plate detection processing on the real-time pressure plate image through a preset pressure plate detection model to obtain a pressure plate detection frame image, perform pressure plate detection processing on the benchmark pressure plate image through a preset benchmark detection model to obtain a benchmark detection frame image, the preset image detection model is obtained by training the initial YOLOv5 model with historical pressure plate images collected from the target substation, the initial YOLOv5 model includes a small-scale detection head and uses EIoU as the coordinate loss function, the preset benchmark detection model is obtained by training the initial YOLOv5 model with various types of benchmark pressure plate images.

[0074] Among them, the preset pressure plate detection model and the preset benchmark detection model are both detection models that have been pre-trained and can be deployed on the mobile terminal 102. The server 104 can train the preset pressure plate detection model from the historical pressure plate images collected from the target substation. And the preset benchmark detection model is trained by using various types of benchmark pressure plate images. Both are trained with the improved YOLOv5 model as the basic model. The YOLOv5 model network belongs to an end-to-end single-stage target detection model, which adopts a method based on single-stage detection to achieve target detection by dividing the image into smaller grids and predicting the object category, bounding box and confidence in each grid. It adopts a design based on the combination of feature pyramid network (FPN) and PAN network, which can effectively process targets of different scales, thereby improving the accuracy and stability of target detection. The native YOLOv5 network structure includes three parts: backbone, neck and head. Among them, the backbone includes multiple Conv modules, C3 modules and an SPPF module, which is mainly responsible for feature extraction; the neck is composed of Conv modules, C3 modules, and upsampling modules, which is mainly responsible for multi-scale feature fusion; the head is responsible for predicting output. The YOLOv5 network structure is as follows Figure 5 As shown. The YOLOv5 loss function consists of three parts: classification loss (L cls ), confidence loss L conf and coordinate loss L bbox :

[0075] L=L cls +L conf +L bbox

[0076] Among them, the coordinate loss adopts the CIoU calculation method by default. CIoU is developed based on DIoU. It comprehensively considers the overlap and distance scale between the real box and the predicted box, making the target regression more stable. The calculation formula is as follows:

[0077]

[0078] In the formula, IoU represents the intersection-over-union ratio between the real box and the predicted box, b represents the predicted box, bgt represents the real box, ρ represents the Euclidean distance between the center point of the real box and the predicted box, c represents the diagonal distance of the minimum closure area between the predicted box and the real box, α is the weight coefficient, and the expression is:

[0079]

[0080] Where ν is used to measure the consistency of aspect ratio, and the expression is:

[0081]

[0082] In the formula, w gt Represents the width of the real box, h gt Represents the width of the prediction box, w represents the width of the prediction box, and h represents the width of the prediction box.

[0083] The classification loss and confidence loss are calculated by binary cross entropy function (BCE) by default. The following formula is the calculation formula of binary cross entropy function:

[0084]

[0085] Among them, y is the label corresponding to the input sample (positive sample is 1, negative sample is 0), and p is the probability that the model predicts that the input sample is a positive sample. The confidence represents the reliability of the current prediction box. The larger the value, the closer the prediction box is to the real box. The confidence label is calculated based on the IoU score between the prediction box corresponding to the grid and the real box, and a smoothing coefficient α is added. When α=1, the confidence label is equal to the CIoU score:

[0086] label = 1-α+α*score_CIoU

[0087] This application improves the native YOLOv5 model. The native YOLOv5 structure contains three detection heads. The feature map sizes output by the three detection heads are 20×20, 40×40, and 80×80, which is equivalent to 8x, 16x, and 32x downsampling on the original image (640×640). The higher the sampling multiple, the more abstract the feature and the larger the receptive field. Most of the pressure plate targets in this application are small and medium-sized targets. Therefore, the YOLOv5 network is improved for the targets of pressure plate detection. Including one, adding a detection head. Through cluster analysis of the label boxes, it is found that the average width of the largest target box is 126.24, the minimum width is 24.35, the average height is 134.55, and the minimum height is 36.16. Generally, the model can detect targets of size 32×32. In order to detect pressure plate targets more accurately, the original YOLOv5 network is improved and a small-scale detection head is added. Compared with the original YOLOv5 model, this improved network structure introduces an upsampling operation, which enables the model to process small targets more attentively. The network structure after adding the detection head is as follows: Figure 6As shown, the light-colored line represents the newly added part. The second is the improvement in the model loss function. Although the coordinate loss function CIoU of the YOLOv5 model takes into account the overlapping area, center point distance and aspect ratio of the bounding box regression, when the aspect ratio and center point of the two boxes are consistent, but the sizes are inconsistent, the loss calculated by CIoU will be the same. This is inconsistent with the actual situation. Therefore, in order to better adapt to the actual scenario of pressure plate detection, this application replaces the original CIoU loss function with EIoU. The improved YOLOv5 model adopts the EIoU loss function, which more accurately measures the important information in target detection and improves the performance and accuracy of the model. Its calculation formula is as follows:

[0088]

[0089] Among them, C w and C h They represent the width and height of the minimum bounding box covering the two Boxes respectively. Compared with the CIoU loss function, the EIoU loss function modifies the aspect ratio loss into a width loss plus a height loss.

[0090] Exemplarily, after obtaining the real-time pressure plate image and the reference pressure plate image, in order to detect the pressure plate switch contained in the two images, the pressure plate detection can be implemented by a pre-trained machine learning model. For the real-time pressure plate image and the reference pressure plate image, a preset pressure plate detection model and a preset reference detection model are respectively set. The real-time pressure plate image is subjected to pressure plate detection processing by the preset pressure plate detection model, so as to frame the detection frame position in the real-time pressure plate image by the detection frame, and obtain a pressure plate detection frame image. At the same time, for the reference pressure plate image, the corresponding pressure plate position can also be framed.

[0091] Step 205 , clustering the platen detection frame image and the reference detection frame image respectively to obtain first platen position information corresponding to the platen detection frame image and second platen position information corresponding to the reference detection frame image.

[0092] Step 207 , performing status comparison processing based on the first pressing plate position information and the second pressing plate position information, and identifying the pressing plate with abnormal status in the real-time pressing plate image.

[0093] Among them, clustering processing refers to clustering the detection frames in the pressure plate detection frame image and the reference detection frame image respectively. The clustering here is row and column clustering, that is, clustering the detection frames in the same row and the detection frames in the same column together. Through clustering, the relative position information of each detection frame in the image can be determined, that is, the first pressure plate position information and the second pressure plate position information.

[0094] Exemplarily, after the pressure plate position is framed by the detection frame, and the pressure plate detection frame image and the reference detection frame image are obtained, each pressure plate detection frame can be positioned by clustering to determine the relative position of the detection frame. Clustering here can specifically use a density-based clustering method. Through clustering, the detection frames in the same row and the detection frames in the same column are clustered together to determine the row position and column position of each detection frame in the image. For the pressure plate detection frame image, the first pressure plate position information is obtained, and for the reference detection frame image, the second pressure plate position information is obtained. For example, for the detection frame with the second row position and the third column position in the real-time pressure plate image, the detection frame with the second row position and the third column position can be found from the reference detection frame image, and then the pressure plate states of the two are compared. If the pressure plate states are the same, it means that the switch state of the pressure plate at the current position is normal. If the pressure plate states are different, it means that the pressure plate state at the current position is abnormal.

[0095] The above-mentioned substation pressure plate recognition method first obtains the real-time pressure plate image collected from the target substation, and searches for the reference pressure plate image corresponding to the real-time pressure plate image, so as to judge whether the real-time pressure plate state is abnormal through the reference pressure plate image, and then performs pressure plate detection processing on the real-time pressure plate image through the preset pressure plate detection model to obtain the pressure plate detection frame image, and performs pressure plate detection processing on the reference pressure plate image through the preset reference detection model to obtain the reference detection frame image, and in the detection process, the detection model is trained with the initial YOLOv5 model containing a small-scale detection head and using EIoU as the coordinate loss function, and the wide loss plus the high loss of EioU effectively measures the important information of the target detection process, and the small-scale detection head is used to accurately detect the pressure plate as a small target, and finally the detected pressure plate detection frame image and the reference detection frame image are clustered respectively to determine the pressure plate position information of the two. Finally, based on the position information, the real-time pressure plate state on the real-time pressure plate image and the reference pressure plate state on the reference pressure plate image are compared to identify the pressure plate with abnormal state. By improving the YOLOv5 model, the accuracy and robustness of the pressure plate recognition and detection process can be effectively improved, thereby effectively improving the accuracy of pressure plate state detection.

[0096] In an exemplary embodiment, the step 203 of performing pressure plate detection processing on the real-time pressure plate image through a preset pressure plate detection model to obtain a pressure plate detection frame image includes: performing feature extraction processing on the real-time pressure plate image through the backbone network of the preset pressure plate detection model to obtain an original pressure plate image feature map; performing multi-scale feature fusion processing on the original pressure plate image feature map through the neck network of the preset pressure plate detection model to obtain a pressure plate image fusion feature map; performing output prediction processing on the pressure plate image fusion feature map through the detection head network of the preset pressure plate detection model to obtain a pressure plate detection frame image.

[0097] Exemplarily, the present application constructs a preset pressure plate detection model based on the network structure of YOLOv5, and the YOLOv5 network structure includes three parts: backbone network backbone, neck network neck and detection head head. Among them, the backbone network backbone includes multiple Conv modules, C3 modules and an SPPF module, which is mainly responsible for feature extraction; the neck network neck is composed of Conv modules, C3 modules, and upsampling modules, which is mainly responsible for multi-scale feature fusion; the detection head head is responsible for predicting output. Therefore, in the detection process of the detection frame, the input real-time pressure plate image is first subjected to feature extraction processing by the backbone network backbone to obtain the original pressure plate image feature map. Then, the neck network neck performs multi-scale feature fusion processing on the extracted original pressure plate image feature map to obtain the pressure plate image fusion feature map. Finally, the detection head network head performs output prediction processing on the pressure plate image fusion feature map to obtain the pressure plate detection frame image. Similarly, for the preset reference detection model, the reference detection frame image can also be extracted through the above-mentioned processing process. In this embodiment, the backbone network backbone, the neck network neck and the detection head head are used to complete the recognition of the pressure plate detection frame in the pressure plate image, which can effectively improve the detection accuracy.

[0098] In an exemplary embodiment, step 205 includes: performing row and column clustering processing on the pressure plate detection frame image through a density-based clustering algorithm to obtain the first row clustering result and the first column clustering result of the pressure plate detection frame image, and performing row and column clustering processing on the reference detection frame image through a density-based clustering algorithm to obtain the second row clustering result and the second column clustering result of the reference detection frame image; based on the first row clustering result and the first column clustering result, determining the row position and column position of each pressure plate detection frame image in the real-time pressure plate image to obtain the first pressure plate position information, and based on the second row clustering result and the second column clustering result, determining the row position and column position of each reference detection frame image in the reference pressure plate image to obtain the second pressure plate position information.

[0099] Among them, the density-based clustering algorithm can specifically use DBSCAN (Density-Based Spatial Clustering of Applications with Noise) clustering. DBSCAN clustering divides data points into core points, boundary points and noise points. Core points are points with more points than MinPts within a certain radius, boundary points are points with less points than MinPts within the radius but connected to the core points, and noise points are points that do not belong to any cluster. It defines two important parameters: ∈, which represents the maximum distance between two samples in the same cluster. If the distance between two samples is less than ∈, they are considered to belong to the same cluster. MinPts, represents the minimum number of samples in the neighborhood of ∈. When the number of samples in the neighborhood of ∈ of a sample is greater than or equal to MinPts, the sample is considered to be a core point. The DBSCAN algorithm has the following characteristics:

[0100] 1) Core points: Points that contain at least MinPts samples in their neighborhood are considered core points.

[0101] 2) Boundary point: a point whose number of samples in the neighborhood of ∈ is less than MinPts, but is in the neighborhood of the core point.

[0102] 3) Noise points: Points that are neither core points nor boundary points are noise points.

[0103] The steps of the DBSCAN algorithm are as follows:

[0104] 1) Set the domain radius eps and the minimum number of points min_samples.

[0105] 2) Find core points to form temporary clusters. For each unvisited point, calculate all points in its eps neighborhood. If the number of points in this neighborhood is greater than or equal to min_samples, then create a new cluster and mark this point as visited, otherwise mark it as noise.

[0106] 3) If a point is part of a cluster, then all points in its eps neighborhood are also part of the cluster. Therefore, all unvisited points in the eps neighborhood are added to the cluster and marked as visited.

[0107] 4) Repeat steps 2) and 3) until all points have been visited.

[0108] Exemplarily, in order to determine the state of the pressure plate at the corresponding position in the real-time pressure plate image and the reference pressure plate image, the present application adopts the improved YOLOv5 model for target detection, and then performs row and column clustering on the detected target to determine the relative position of each pressure plate. For the row and column clustering process, the present application adopts the DBSCAN clustering algorithm to cluster the detected target frame. DBSCAN clustering is a commonly used density clustering algorithm, which can identify areas with high density and can distinguish areas with different densities. This algorithm does not require the number of clusters to be set in advance, and can handle noise and outliers better. Through row and column clustering, for the pressure plate detection frame image, the first row clustering result and the first column clustering result obtained by clustering can be used to determine which cluster each detection frame belongs to, determine the row position and column position of each pressure plate detection frame image in the real-time pressure plate image, and obtain the first pressure plate position information. Similarly, for the reference detection frame images, after clustering, the row position and column position of each reference detection frame image in the reference pressure plate image can be determined based on the second row clustering result and the second column clustering result obtained by clustering, thereby obtaining the second pressure plate position information. In this embodiment, the detection frame is clustered in rows and columns by a density-based clustering algorithm to determine the relative position of the detection frame in the image, thereby providing a basis for the subsequent detection frame discrimination processing, and the pressure plates in the same position are identified by clustering, thereby improving the accuracy of pressure plate identification.

[0109] In an exemplary embodiment, step 207 includes: determining a platen detection frame image and a reference detection frame image having the same row position and column position based on the first platen position information and the second platen position information; performing state comparison processing on the platen detection frame image and the reference detection frame image having the same row position and column position, determining a platen detection frame image having a different state from the reference detection frame image, and obtaining an abnormal state platen in the real-time platen image.

[0110] Exemplarily, for the recognition process of the pressure plate state, the pressure plate detection frame image and the reference detection frame image with the same row position and column position can be determined based on the first pressure plate position information and the second pressure plate position information. The pressure plate state in the pressure plate detection frame image at the same position should be the same as the pressure plate state marked in the reference detection frame image. Therefore, the pressure plate detection frame image and the reference detection frame image with the same row position and column position can be directly subjected to state comparison processing to determine the pressure plate detection frame image with a different state from the reference detection frame image. If the pressure plate states represented by the reference detection frame image and the pressure plate detection frame image are different, it means that the switch state corresponding to the pressure plate is abnormal. By comparing the significance of each pressure plate, all the pressure plates with abnormal states in the real-time pressure plate image can be determined. In this embodiment, by comparing the pressure plate detection frame image and the reference detection frame image at the same position, the pressure plate with abnormal state can be effectively identified to ensure the accuracy of detection.

[0111] In one of the exemplary embodiments, the method also includes: acquiring historical pressure plate images collected from the target substation; resizing the historical pressure plate images to obtain original sample images of the same specifications; annotating the switch status of each pressure plate in the original sample image in the form of a rectangular frame to obtain a training sample image; and training an initial YOLOv5 model based on the training sample image to obtain a preset image detection model.

[0112] Exemplarily, the resizing process refers to resizing the historical platen images of different specifications to a certain extent, so as to ensure the consistency of the image format. Since the captured platen images are of different sizes and have relatively high resolutions, the image size can be adjusted before being sent to the network. For example, the long side of the platen image is adjusted to 1280, and the short side is adaptively processed to obtain the original sample image of the same specification. After that, these images are only annotated. Specifically, the switch state of each platen in the original sample image can be annotated in the form of a rectangular frame to obtain a training sample image. The data annotation work is implemented using the open source annotation software labelImg, and there are three annotated categories: open, closed, and standby. Finally, the initial YOLOv5 model is trained with these training sample images to obtain a preset image detection model. Similarly, for the preset benchmark detection model, it is also necessary to use the adjusted benchmark platen images of various types as training samples to complete the relevant processing of model training. In this embodiment, historical pressure plate images are collected and resized to construct model training data, thereby completing the training of the preset image detection model, thereby effectively ensuring the accuracy and efficiency of the pressure plate detection frame detection.

[0113] In an exemplary embodiment, the method also includes: obtaining a trained preset pressure plate detection model, converting the preset pressure plate detection model into an open neural network exchange format; converting the preset pressure plate detection model in the open neural network exchange format into a binary model file through a model conversion tool; and deploying the binary model file on a mobile terminal for substation pressure plate identification.

[0114] Exemplarily, the present application can also be applied on a server. In this case, after completing the training process of the model, the trained model can be deployed to a mobile terminal so that grassroots substation operation and maintenance staff can complete the detection through the mobile terminal. During deployment, it is necessary to first convert the preset pressure plate detection model into an open neural network exchange format onnx, and then use a model conversion tool, such as the NCNN tool to convert the preset pressure plate detection model in the open neural network exchange format into a binary model file, and finally successfully deploy the preset pressure plate detection model to the mobile terminal. Similarly, the preset benchmark detection model can also be deployed in the same way. For the corresponding application process, from uploading the real-time pressure plate image to displaying the output results, the average response time is about 2 seconds. The test results of a single image are as follows. Figure 7 In this embodiment, by compressing the detection model, packaging it into a program, and installing it in a mobile terminal, the user only needs to take two photos to achieve a quick comparison, and the entire operation takes less than 5 seconds, greatly improving work efficiency.

[0115] In one embodiment, the flowchart of the pressure plate detection of the present application can refer to Figure 8 As shown in the figure, firstly, historical pressure plate images are collected from the target substation, and these images are resized to construct model training data. Then, the improved YOLOv5 network is trained based on these training data to obtain the pressure plate detection model. After the training is completed, the network is trained with the "four meters and one guide" image to obtain the "four meters and one guide" detection model. Finally, the DBSCAN clustering method is used to cluster the pressure plate detection model and the "four meters and one guide" detection model respectively to obtain the relative position of each pressure plate, and finally the state comparison is performed to find the pressure plates with different states.

[0116] In one embodiment, for the solution of this application, pictures taken on-site at the substation and screenshot data of the "Four Tables and One Guide" are selected. Among them, a total of 738 pictures and 21,037 pressure plate switches are annotated. The training set and the test set are divided according to the ratio of 8:2, and then 590 pictures of the training set and 148 pictures of the verification set are divided. The division of data training and verification sets is shown in Table 1.

[0117] Table 1. Platen data division

[0118]

[0119] The “Four Tables and One Guide” data contains a total of 675 images. The division of data training and validation sets is shown in Table 2 below.

[0120] Table 2. Data division of "Four Tables and One Guide"

[0121]

[0122] The training platform of this application uses NVIDIA GTX3090 GPU, 12th generation Intel(R) Core(TM) i7-12700KF CPU, and Ubuntu 18.04 operating system. The main parameters of the training process are as follows:

[0123] Input image size (imgsz): 1280; number of training epochs: 300 rounds; batch size (batch_size): 10; initial learning rate (lr): 0.01; cosine annealing algorithm is used as the learning rate decay strategy; other parameters use the default parameters; data enhancement hyperparameters use the default parameters in hyp.scratch-high.yaml, but set the mosaic, mixup, and copy_pastec parameters to 0.

[0124] The loss of the platen data training process and the verification accuracy are as follows Fig. 9 , Fig.10 As shown in the figure. From the training process curve, it can be seen that the improved YOLOv5 model is more stable during the training process, and the loss decreases more smoothly. From the training results and verification results, it can be seen that after 200 rounds, both the original model and the improved model can converge well. From the verification loss, the loss of the two models when they converge is basically the same. However, when we compare the accuracy and recall rate of each category, the improved YOLOv5 model is higher than the original model in all indicators (see the results of each indicator). Fig.11 ).

[0125] The improved model is used to train the "Four Tables and One Guide" data. Since the "Four Tables and One Guide" images are small in size, the input size is unified to 640×640, and training is performed for 200 rounds. The training process is as follows: Fig.12 As shown, the verification results after training are as follows Fig.13 shown.

[0126] It can be seen from the above verification data that the two models of this application (improved YOLOv5 model) have an average detection accuracy of more than 99% for the pressure plate. Furthermore, we use the DBSCAN clustering method to locate the detected target frames in rows and columns. First, the detection frame of each image is extracted. Then, the horizontal and vertical coordinates of the center points of all its detection frames are clustered respectively to calculate the number of rows and columns in the image, and then the state of each pressure plate is compared with the pressure plate target detected in the "Four Tables and One Guide" image. Finally, the position of the pressure plate in the abnormal state is located. After the model training is completed, reasoning is performed on a single image, and the reasoning output result is as follows. Fig.14 shown. Fig.14 In the figure, the leftmost is the actual pressure plate photographed on site, the middle is the "four tables and one guide", and the rightmost is the pressure plate detected with inconsistent states.

[0127] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0128] Based on the same inventive concept, the embodiment of the present application also provides a substation pressure plate identification device for implementing the substation pressure plate identification method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more substation pressure plate identification device embodiments provided below can refer to the above limitations on the substation pressure plate identification method, and will not be repeated here.

[0129] In an exemplary embodiment, Fig.15 As shown, a substation pressure plate identification device is provided, comprising:

[0130] The image acquisition module 1502 is used to acquire the real-time pressure plate image collected from the target substation and search for the reference pressure plate image corresponding to the real-time pressure plate image.

[0131] The pressure plate detection module 1504 is used to perform pressure plate detection processing on the real-time pressure plate image through a preset pressure plate detection model to obtain a pressure plate detection frame image, perform pressure plate detection processing on the reference pressure plate image through a preset benchmark detection model to obtain a benchmark detection frame image, the preset image detection model is obtained by training the initial YOLOv5 model with historical pressure plate images collected from the target substation, the initial YOLOv5 model includes a small-scale detection head and uses EIoU as the coordinate loss function, and the preset benchmark detection model is obtained by training the initial YOLOv5 model with various types of benchmark pressure plate images.

[0132] The clustering processing module 1506 is used to perform clustering processing on the pressure plate detection frame image and the reference detection frame image respectively to obtain the first pressure plate position information corresponding to the pressure plate detection frame image and the second pressure plate position information corresponding to the reference detection frame image.

[0133] The state identification module 1508 is used to perform state comparison processing based on the first pressure plate position information and the second pressure plate position information to identify the pressure plate with abnormal state in the real-time pressure plate image.

[0134] In one embodiment, the pressure plate detection module 1504 is specifically used to: perform feature extraction processing on the real-time pressure plate image through the backbone network of the preset pressure plate detection model to obtain the original pressure plate image feature map; perform multi-scale feature fusion processing on the original pressure plate image feature map through the neck network of the preset pressure plate detection model to obtain the pressure plate image fusion feature map; perform output prediction processing on the pressure plate image fusion feature map through the detection head network of the preset pressure plate detection model to obtain the pressure plate detection frame image.

[0135] In one embodiment, the clustering processing module 1506 is specifically used to: perform row and column clustering processing on the pressure plate detection frame image through a density-based clustering algorithm to obtain the first row clustering result and the first column clustering result of the pressure plate detection frame image, and perform row and column clustering processing on the reference detection frame image through a density-based clustering algorithm to obtain the second row clustering result and the second column clustering result of the reference detection frame image; based on the first row clustering result and the first column clustering result, determine the row position and column position of each pressure plate detection frame image in the real-time pressure plate image to obtain the first pressure plate position information, and based on the second row clustering result and the second column clustering result, determine the row position and column position of each reference detection frame image in the reference pressure plate image to obtain the second pressure plate position information.

[0136] In one embodiment, the state recognition module 1508 is specifically used to: determine the platen detection frame image and the reference detection frame image with the same row position and column position based on the first platen position information and the second platen position information; perform state comparison processing on the platen detection frame image and the reference detection frame image with the same row position and column position, determine the platen detection frame image with a different state from the reference detection frame image, and obtain the abnormal state platen in the real-time platen image.

[0137] In one embodiment, a model training module is also included, which is used to: obtain historical pressure plate images collected from the target substation; resize the historical pressure plate images to obtain original sample images of the same specifications; annotate the switch status of each pressure plate in the original sample image in the form of a rectangular frame to obtain a training sample image; and train the initial YOLOv5 model based on the training sample image to obtain a preset image detection model.

[0138] In one embodiment, it also includes a model deployment module, which is used to: obtain a trained preset pressure plate detection model, and convert the preset pressure plate detection model into an open neural network exchange format; convert the preset pressure plate detection model in the open neural network exchange format into a binary model file through a model conversion tool; and deploy the binary model file on a mobile terminal for substation pressure plate identification.

[0139] Each module in the above-mentioned substation pressure plate identification device can be implemented in whole or in part by software, hardware and their combination. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the corresponding operations of each of the above modules.

[0140] In an exemplary embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as shown in FIG. Fig.16 As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface, the display unit and the input device are connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, near field communication (NFC) or other technologies. When the computer program is executed by the processor, a substation pressure plate identification method is implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0141] Those skilled in the art will understand that Fig.16 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0142] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0143] In one embodiment, a computer-readable storage medium is provided, storing a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0144] In one embodiment, a computer program product or computer program is provided, the computer program product or computer program includes computer instructions, the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the steps in the above-mentioned method embodiments.

[0145] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0146] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0147] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0148] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for identifying a pressure plate in a substation, characterized in that: The method comprises: Acquire a real-time pressure plate image collected from a target substation, and search for a reference pressure plate image corresponding to the real-time pressure plate image; The real-time pressure plate image is subjected to pressure plate detection processing by a preset pressure plate detection model to obtain a pressure plate detection frame image, and the reference pressure plate image is subjected to pressure plate detection processing by a preset reference detection model to obtain a reference detection frame image, wherein the preset image detection model is obtained by training an initial YOLOv5 model with historical pressure plate images collected from the target substation, wherein the initial YOLOv5 model includes a small-scale detection head and uses EIoU as a coordinate loss function, and the preset reference detection model is obtained by training the initial YOLOv5 model with various types of reference pressure plate images; Performing clustering processing on the pressure plate detection frame image and the reference detection frame image respectively to obtain first pressure plate position information corresponding to the pressure plate detection frame image and second pressure plate position information corresponding to the reference detection frame image; A state comparison process is performed based on the first pressure plate position information and the second pressure plate position information to identify a pressure plate with an abnormal state in the real-time pressure plate image.

2. The method according to claim 1, characterized in that The performing pressure plate detection processing on the real-time pressure plate image by using a preset pressure plate detection model to obtain a pressure plate detection frame image includes: Performing feature extraction processing on the real-time pressing plate image through a backbone network of a preset pressing plate detection model to obtain an original pressing plate image feature map; Performing multi-scale feature fusion processing on the original pressure plate image feature map through the neck network of the preset pressure plate detection model to obtain a pressure plate image fusion feature map; The pressure plate image fusion feature map is output predicted and processed by the detection head network of the preset pressure plate detection model to obtain a pressure plate detection frame image.

3. The method according to claim 1, characterized in that The clustering process is performed on the pressure plate detection frame image and the reference detection frame image respectively to obtain the first pressure plate position information corresponding to the pressure plate detection frame image and the second pressure plate position information corresponding to the reference detection frame image, including: Performing row-column clustering processing on the platen detection frame image by a density-based clustering algorithm to obtain a first row clustering result and a first column clustering result of the platen detection frame image, and performing row-column clustering processing on the reference detection frame image by a density-based clustering algorithm to obtain a second row clustering result and a second column clustering result of the reference detection frame image; Based on the first row clustering result and the first column clustering result, the row position and column position of each pressure plate detection frame image in the real-time pressure plate image are determined to obtain the first pressure plate position information; based on the second row clustering result and the second column clustering result, the row position and column position of each reference detection frame image in the reference pressure plate image are determined to obtain the second pressure plate position information.

4. The method according to claim 3, characterized in that The performing state comparison processing based on the first pressing plate position information and the second pressing plate position information to identify the pressing plate with abnormal state in the real-time pressing plate image comprises: Determine a platen detection frame image and a reference detection frame image having the same row position and column position based on the first platen position information and the second platen position information; The state comparison processing is performed on the pressure plate detection frame image and the reference detection frame image with the same row position and column position, and the pressure plate detection frame image with a different state from the reference detection frame image is determined to obtain the abnormal state pressure plate in the real-time pressure plate image.

5. The method according to claim 1, characterized in that The method further comprises: Acquire historical pressure plate images collected from the target substation; Resizing the historical platen image to obtain an original sample image with the same specifications; Annotating the switch states of each pressing plate in the original sample image in the form of a rectangular frame to obtain a training sample image; The initial YOLOv5 model is trained based on the training sample images to obtain a preset image detection model.

6. The method according to any one of claims 1 to 5, characterized in that: The method further comprises: Acquire a trained preset pressure plate detection model, and convert the preset pressure plate detection model into an open neural network exchange format; Converting the preset platen detection model in the open neural network exchange format into a binary model file by a model conversion tool; The binary model file is deployed on a mobile terminal for substation pressure plate identification.

7. A substation pressure plate identification device, characterized in that: The device comprises: An image acquisition module is used to acquire a real-time pressure plate image collected from a target substation and search for a reference pressure plate image corresponding to the real-time pressure plate image; A pressure plate detection module, used to perform pressure plate detection processing on the real-time pressure plate image through a preset pressure plate detection model to obtain a pressure plate detection frame image, and perform pressure plate detection processing on the reference pressure plate image through a preset reference detection model to obtain a reference detection frame image, wherein the preset image detection model is obtained by training an initial YOLOv5 model with historical pressure plate images collected from the target substation, wherein the initial YOLOv5 model includes a small-scale detection head and uses EIoU as a coordinate loss function, and the preset reference detection model is obtained by training the initial YOLOv5 model with various types of reference pressure plate images; A clustering processing module, used for performing clustering processing on the pressure plate detection frame image and the reference detection frame image respectively, to obtain first pressure plate position information corresponding to the pressure plate detection frame image and second pressure plate position information corresponding to the reference detection frame image; The state recognition module is used to perform state comparison processing based on the first pressure plate position information and the second pressure plate position information to identify the pressure plate with abnormal state in the real-time pressure plate image.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images