Substation meter identification method, device and equipment

Abstract

The embodiment of the invention provides a substation meter identification method, device and equipment. The method comprises the following steps: acquiring a to-be-recognized image; inputting the to-be-recognized image into a pre-acquired target detection model to obtain a target type of a target meter and position information of the target meter in the to-be-recognized image; wherein the target detection model is a model which is obtained by training a lightweight YOLOv11 model based on an image in which the type and the position of the meter are calibrated in advance and is used for classifying the meter in the image and determining the position of the meter; wherein the lightweight YOLOv11 model is obtained by setting a backbone network of the YOLOv11 to be a StarNet lightweight network and setting a detection head to be BNLSCD; cutting out an area where the target meter is located from the to-be-identified image according to the position information to obtain a dial plate image; and reading the meter based on the target type and the dial plate image to obtain the reading of the target meter. The method is used for achieving the effect of reducing computing resources consumed for identifying the meter reading.

Description

Technical Field

[0001] This application relates to the field of data recognition, and in particular, to a method, device, and equipment for recognizing substation meters. Background Art

[0002] Due to advantages such as high reliability and low cost, pointer meters are widely used in various industrial equipment and facilities, and usually rely on manual observation for reading. In many industrial scenarios, since meters are often installed in places with long distances or poor environmental conditions, there are problems of low efficiency of manual meter reading and a high incidence of meter reading errors.

[0003] Therefore, with the continuous advancement of industrial automation, the automatic recognition of meters has become one of the key technologies for realizing intelligent operation and maintenance and improving production efficiency.

[0004] However, in the existing methods for automatic recognition of meters, when automatically recognizing meter readings, there are often problems of high consumption of computing resources and low recognition efficiency. Summary of the Invention

[0005] Embodiments of this application provide a method, device, and equipment for recognizing substation meters, so as to achieve the effects of reducing the computing resources required for recognizing meter readings and improving the recognition efficiency.

[0006] In a first aspect, embodiments of this application provide a method for recognizing substation meters, including:

[0007] Obtain an image to be recognized, where the image to be recognized includes a target meter for which data is to be read;

[0008] Input the image to be recognized into a pre-obtained target detection model to obtain the target type of the target meter and the position information of the target meter in the image to be recognized; where the target detection model is a model for classifying and determining the position of the meter in the image, which is trained based on an image with pre-calibrated meter types and positions, and the lightweight YOLOv11 model; the lightweight YOLOv11 model is obtained by setting the backbone network of YOLOv11 to the StarNet lightweight network and setting the detection head to BN_LSCD;

[0009] Crop out the area where the target meter is located from the image to be recognized according to the position information to obtain a dial image;

[0010] Read the meter based on the target type and the dial image to obtain the reading of the target meter.

[0011] In a possible implementation, reading the meter based on the target type and the dial image to obtain the reading of the target meter includes:

[0012] Obtain the scale lines and the pointer of the target meter dial from the dial image;

[0013] According to the target type, obtain the corresponding dial model of the target meter, where the dial model includes parameters of the circular outer frame of the meter of the target type, scale line information, the variation range of the reading, and the angular range of the pointer;

[0014] Align the scale lines of the target meter dial with the dial model, and after alignment, determine the pointer angle based on the pointer of the target meter dial and the dial model;

[0015] Obtain the reading of the target meter according to the pointer angle.

[0016] In a possible implementation, the obtaining the reading of the target meter according to the pointer angle includes:

[0017] Calculate the reading of the target meter according to the pointer angle and the reading calculation formula of the meter of the target type;

[0018] Or,

[0019] Determine the reading corresponding to the pointer angle according to the mapping relationship between the angle and the reading of the pointer of the meter of the target type included in the dial model, as the reading of the target meter.

[0020] In a possible implementation, the obtaining the scale lines and the pointer of the target meter dial from the dial image includes:

[0021] Perform perspective transformation on the dial image to obtain the dial image in the front view;

[0022] Adopt the Hough transform method for the dial image in the front view to extract the scale lines and the pointer of the target meter dial.

[0023] In a possible implementation, before inputting the image to be recognized into the pre-obtained target detection model, the method further includes:

[0024] Annotate the images of different types of meters collected to obtain an image data set, where the image data set includes multiple images of different types of meters, and the annotation information of each image includes the meter type and the position of the meter;

[0025] Set the backbone network of the YOLOv11 model to the StarNet lightweight network, and set the detection head to BN_LSCD to obtain a lightweight YOLOv11 model;

[0026] Based on the lightweight YOLOv11 model, construct an initial object detection model;

[0027] Train the initial object detection model according to the image dataset to obtain the object detection model.

[0028] In a possible implementation manner, the method further includes:

[0029] According to the reading of the target meter at the current moment and the readings of the target meter within a preset duration before the current moment, use a reading prediction model to predict the reading of the target meter, and obtain the predicted reading of the target meter at the next time step; wherein, the reading prediction model is a prediction model trained based on historical meter reading data for an LSTM model;

[0030] Diagnose the predicted reading according to the fluctuation of historical readings and the reading range of the meter of the preset target type, and obtain a first diagnosis result, where the first diagnosis result is used to indicate whether the predicted reading is abnormal.

[0031] In a possible implementation manner, the method further includes:

[0032] Diagnose the reading of the target meter according to the prior knowledge obtained in advance, and obtain a second diagnosis result, where the second diagnosis result is used to indicate whether the reading of the target meter is abnormal; wherein, the prior knowledge includes: a key reading range and judgment rules for at least one abnormal situation;

[0033] If the first diagnosis result indicates that the predicted reading is abnormal and / or the second diagnosis result indicates that the reading is abnormal, push an alarm message; wherein, the alarm message includes: meter type, current reading, prediction trend, abnormal reason, and treatment measures.

[0034] In a possible implementation manner, the method further includes:

[0035] Update the prior knowledge and / or the reading prediction model according to the data in the alarm message.

[0036] In a second aspect, an identification device for a substation meter provided by an embodiment of the present application includes:

[0037] An acquisition unit, configured to acquire an image to be recognized, where the image to be recognized includes a target meter for which data is to be read;

[0038] A detection unit, configured to input the image to be recognized into a pre-acquired target detection model, and obtain the target type of the target meter and the position information of the target meter in the image to be recognized; wherein, the target detection model is a model trained based on images with pre-calibrated meter types and positions, for classifying and determining the position of meters in the image; the lightweight YOLOv11 model is obtained by setting the backbone network of YOLOv11 as the StarNet lightweight network and setting the detection head as BN_LSCD;

[0039] A processing unit, configured to crop the area where the target meter is located from the image to be recognized according to the position information, to obtain a dial image;

[0040] A determination unit, configured to read the meter based on the target type and the dial image, to obtain the reading of the target meter.

[0041] In a possible implementation manner, the determination unit includes:

[0042] A first acquisition module, configured to acquire the scale lines and the pointer of the target meter dial from the dial image;

[0043] A second acquisition module, configured to acquire a dial model corresponding to the target meter according to the target type, where the dial model includes parameters of the circular outer frame of the meter of the target type, scale line information, the change range of the reading, and the angular range of the pointer;

[0044] An alignment module, configured to align the scale lines of the target meter dial with the dial model, and determine the pointer angle based on the pointer of the target meter dial and the dial model after alignment;

[0045] A determination module, configured to obtain the reading of the target meter according to the pointer angle.

[0046] In a possible implementation manner, the determination module is specifically configured to:

[0047] Calculate the reading of the target meter according to the pointer angle and the reading calculation formula of the meter of the target type;

[0048] Or,

[0049] Determine the reading corresponding to the pointer angle according to the mapping relationship between the angle and the reading of the pointer of the meter of the target type included in the dial model, as the reading of the target meter.

[0050] In a possible implementation manner, the first acquisition module is specifically configured to:

[0051] Perform a perspective transformation on the dial image to obtain a dial image in the front view.

[0052] Use the Hough transform method on the dial image in the front view to extract the scale lines and pointers of the target meter dial.

[0053] In a possible implementation manner, the identification device for substation meters further includes:

[0054] A labeling unit for labeling the images of different types of meters collected to obtain an image data set, where the image data set includes multiple images of different types of meters, and the labeling information of each image includes the meter type and the position of the meter.

[0055] A setting unit for setting the backbone network of the YOLOv11 model to the StarNet lightweight network and setting the detection head to BN_LSCD to obtain a lightweight YOLOv11 model.

[0056] A construction unit for constructing an initial target detection model based on the lightweight YOLOv11 model.

[0057] A training unit for training the initial target detection model according to the image data set to obtain the target detection model.

[0058] In a possible implementation manner, the identification device for substation meters further includes:

[0059] A prediction unit for predicting the reading of the target meter using a reading prediction model according to the reading of the target meter at the current moment and the readings of the target meter within a preset time period before the current moment, to obtain a predicted reading of the target meter at the next time step; wherein, the reading prediction model is a prediction model trained on the LSTM model based on historical meter reading data.

[0060] A first diagnosis unit for diagnosing the predicted reading according to the fluctuation of historical readings and the reading range of the meter of the preset target type to obtain a first diagnosis result, and the first diagnosis result is used to indicate whether the predicted reading is abnormal.

[0061] In a possible implementation manner, the identification device for substation meters further includes:

[0062] A second diagnostic unit, configured to diagnose the reading of the target meter according to the pre-acquired prior knowledge, and obtain a second diagnostic result, where the second diagnostic result is used to indicate whether the reading of the target meter is abnormal; wherein, the prior knowledge includes: a key reading range and a judgment rule for at least one abnormal condition;

[0063] An alarm unit, configured to push an alarm message if the first diagnostic result indicates that the predicted reading is abnormal and / or the second diagnostic result indicates that the reading is abnormal; wherein, the alarm message includes: meter type, current reading, predicted trend, cause of abnormality, and treatment measures.

[0064] In a possible implementation manner, the identification device for substation meters further includes:

[0065] An update unit, configured to update the prior knowledge and / or the reading prediction model according to the data in the alarm message.

[0066] In a third aspect, an embodiment of the present application provides a computer device, including: a memory, a processor;

[0067] The memory stores computer execution instructions;

[0068] The processor executes the computer execution instructions stored in the memory, so that the processor executes the above first aspect and / or various possible implementation manners of the first aspect.

[0069] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer execution instructions are stored, and when the computer execution instructions are executed by a processor, they are used to implement the above first aspect and / or various possible implementation manners of the first aspect.

[0070] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the above first aspect and / or various possible implementation manners of the first aspect.

[0071] The method, device, and equipment for identifying substation meters provided by the embodiments of the present application use a lightweight object detection model to detect and classify the image to be recognized of the target meter including the data to be read, obtain the target type of the target meter and the position information of the target meter in the image to be recognized, and then further process the image to be recognized based on the target type of the target meter and the position information of the target meter in the image to be recognized to obtain the reading of the target meter. Among them, the object detection model is a model trained based on the images with pre-calibrated meter types and positions to classify and determine the positions of the meters in the images. The lightweight YOLOv11 model is obtained by setting the backbone network of YOLOv11 as the StarNet lightweight network and setting the detection head as BN_LSCD. This method processes the image to be recognized by using a lightweight object detection model to achieve the technical effects of reducing the computing resources consumed for identifying the meter reading and improving the recognition efficiency. Description of the Drawings

[0072] The drawings here are incorporated into the specification and form a part of this specification, showing the embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application.

[0073] Figure 1 It is a schematic flowchart of the method for identifying substation meters provided in Embodiment 1 of the present application;

[0074] Figure 2 It is a schematic flowchart of the method for identifying substation meters provided in Embodiment 2 of the present application;

[0075] Figure 3 It is a schematic flowchart of the method for identifying substation meters provided in Embodiment 3 of the present application;

[0076] Figure 4 It is a schematic flowchart of the method for identifying substation meters provided in Embodiment 4 of the present application;

[0077] Figure 5 It is a schematic structural diagram of the device for identifying substation meters provided in Embodiment 5 of the present application;

[0078] Figure 6 It is a schematic structural diagram of the device for identifying substation meters provided in Embodiment 6 of the present application;

[0079] Figure 7 It is a schematic structural diagram of the computer equipment provided by the present application.

[0080] Through the above-mentioned accompanying drawings, specific embodiments of the present application have been shown, and will be described in more detail hereinafter. These drawings and the textual description are not intended to limit the scope of the concept of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by reference to specific embodiments. Detailed Description of the Embodiments

[0081] Exemplary embodiments will be described in detail herein, and examples thereof are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0082] To clearly understand the technical solution of the present application, the solutions of the prior art will be introduced in detail below.

[0083] With the continuous expansion of the scale of the power system, as the core component of the power system, the safe and stable operation state of the substation is particularly important. With the continuous advancement of industrial automation, especially in the power system and substation, the automatic recognition of meters has become one of the key technologies for realizing intelligent operation and maintenance and improving production efficiency. Traditional pointer meters are widely used in various industrial equipment and facilities due to their advantages such as reliability and low cost. However, the reading of traditional pointer meters usually relies on manual operation. Especially in the case of a complex on-site environment, a long or restricted installation position of the meter, manual meter reading has problems of low efficiency and high error rate. These factors make the automation demand for meter readings increasingly urgent, and how to achieve intelligent recognition of meters through image processing and deep learning technologies has become an important research direction in current technology.

[0084] In the prior art, there is a method for recognizing meters based on deep learning. This method can extract deep features in images and perform classification and localization through deep learning models such as Convolutional Neural Network (CNN), so as to achieve automatic reading. However, such methods often require high computing resources, which in turn leads to problems of poor practicability and low recognition efficiency of the recognition method.

[0085] In view of the technical problems in the above background art, when researching the identification method of substation meters, the inventor considered improving the backbone network and detection head in the advanced YOLOv11 model to obtain a lightweight YOLOv11 model, and then using this lightweight YOLOv11 model as the object detection model for identifying the readings of the meters. Therefore, when using this object detection model to identify the readings of the meters, not only can the accuracy of the identification be guaranteed, but also the resources consumed by the calculation can be greatly reduced, thereby improving the practicability of the identification method and the identification efficiency.

[0086] Based on the above technical concept of the inventor, this application obtains an image to be recognized; inputs the image to be recognized into a pre-obtained object detection model to obtain the target type of the target meter and the position information of the target meter in the image to be recognized; wherein, the object detection model is a model trained based on an image with pre-calibrated meter types and positions for classifying and determining the positions of meters in the image; wherein, the lightweight YOLOv11 model is obtained by setting the backbone network of YOLOv11 as the StarNet lightweight network and setting the detection head as BN_LSCD (Batch Normalization_Lightweight Shared Convolutional Separamter); crops the area where the target meter is located from the image to be recognized according to the position information to obtain a dial image; and reads the meter reading based on the target type and the dial image to obtain the reading of the target meter, achieving the technical effects of reducing the calculation resources required for identifying the meter reading and improving the identification efficiency.

[0087] The following uses specific embodiments to elaborate in detail on the technical solutions of this application and how the technical solutions of this application solve the above technical problems. These several specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0088] Figure 1 It is a schematic flowchart of the identification method of substation meters provided in the first embodiment of this application. As Figure 1 shown, this method includes:

[0089] S101. Obtain an image to be recognized, where the image to be recognized includes a target meter for which data is to be read.

[0090] In this step, in order to read the reading in the target meter, it is necessary to obtain an image to be recognized including the target meter.

[0091] In a specific implementation manner, the image to be recognized can be obtained based on a fixed camera or the image acquisition device of an inspection robot.

[0092] S102. Input the image to be recognized into a pre-obtained target detection model to obtain the target type of the target meter and the position information of the target meter in the image to be recognized, where the target detection model is a model trained based on images with pre-calibrated meter types and positions for classifying and determining the positions of meters in the image, by using a lightweight YOLOv11 model.

[0093] In this step, it is necessary to input the obtained image to be recognized into the pre-obtained target detection model for detection and classification to obtain the target type of the target meter and the position information of the target meter in the image to be recognized. The target detection model is trained by using an image dataset on a lightweight YOLOv11 model. The image dataset includes multiple images with pre-calibrated meter types and positions. The lightweight YOLOv11 model is obtained by setting the backbone network of YOLOv11 to the StarNet lightweight network and setting the detection head to BN_LSCD.

[0094] Among them, the meter types at least include: barometer, pressure gauge, oil level gauge, oil pressure gauge.

[0095] It should be understood that the YOLOv11 model mainly consists of an input, a backbone network, a neck, and a detection head. In this solution, the StarNet lightweight network is selected as the backbone network of the model, and BN_LSCD is selected as the detection head of the model to complete the improvement of the YOLOv11 model and obtain the lightweight YOLOv11 model. Further, by using the obtained image dataset to train the YOLOv11 model, a target detection model can be obtained.

[0096] In the specific application of this solution, first, the StarNet lightweight network of the target detection model extracts features from the input image to be recognized and outputs a high-dimensional feature map with a low resolution. Then, the neck of the target detection model performs multi-scale feature fusion on the high-dimensional feature map and outputs a multi-scale feature map. Next, the BN_LSCD detection head of the target detection model performs bounding box regression and class prediction processing based on the multi-scale feature map and outputs the target type of the target meter and the position information of the target meter in the image to be recognized.

[0097] It should be understood that the core of the StarNet lightweight network is the star operation. Through this star operation, the StarNet lightweight network can significantly increase the feature dimension in a recursively stacked form. By using the StarNet lightweight network as the backbone network of the object detection model, it is possible to obtain a richer and more expressive feature representation while maintaining computational efficiency, enabling efficient and rapid recognition of meters under different lighting and background conditions.

[0098] Furthermore, in a specific implementation, the specific structure of the StarNet lightweight network is as follows:

[0099] The StarNet lightweight network consists of a Convolutional-Batch-Normalization-ReLU (CBR) module and 4 Star Layer modules. Among them, each CBR module includes a Convolutional (Conv) module with a convolutional kernel size of 3×3, a stride of 2, and an output channel of 24, a Batch Normalization (BN) module, and a ReLu6 activation module; each Star Layer module consists of a Space-to-Depth Convolution (SPD Conv) module and several Star Block modules.

[0100] In the above structure of the StarNet lightweight network, the SPD Conv module in the Star Layer module performs convolutional processing on each pixel in the feature map corresponding to the image to be recognized to complete downsampling and modifying the number of embedded channels, obtaining an intermediate feature map with reduced channels.

[0101] Specifically, the SPD Conv module first divides the feature map into multiple sub-feature maps in the spatial dimension; then splices the above sub-feature maps in the channel dimension to generate an intermediate feature map; then, the point convolutional layer in the SPD Conv module reduces the channels of the intermediate feature map to obtain an intermediate feature map with reduced channels.

[0102] Exemplarily, if the image to be recognized is input into the object detection model and after being processed by the CBR module, the feature map corresponding to the image to be recognized output has a size of , then the SPD Conv module will process the feature map as follows:

[0103] ① Divide the feature map into sub-feature maps from the spatial dimension, where the sub-feature maps are represented by the following formula:

[0104]

[0105]

[0106]

[0107]

[0108] Among them, the size of each sub-feature map is .

[0109] ② Concatenate the above sub-feature maps in the channel dimension to generate an intermediate feature map .

[0110] Among them, the size of the first intermediate feature map is .

[0111] ③ Use convolution kernels with a size of and a stride of 1 to perform convolution operations to achieve channel dimensionality reduction of the intermediate feature map, obtaining the intermediate feature map after channel dimensionality reduction.

[0112] Among them, the size of the intermediate feature map after channel dimensionality reduction is .

[0113] In the above formula, is the spatial size of the feature map , is 2, is the channel dimension of the feature map , is the embedding channel number of the intermediate feature map after channel dimensionality reduction.

[0114] It should be understood that this structure using the SPD Conv module can effectively avoid the problem of loss of fine-grained information caused by using strided convolution for downsampling in traditional convolutional neural networks. While ensuring the feature representation ability during the downsampling process, it can more effectively retain the fine-grained feature information of the image, achieving the effect of improving the object detection ability for different meters under different scenarios and different lighting conditions.

[0115] Meanwhile, in the structure of the above StarNet lightweight network, in the Star Block module of the Star Layer module, the intermediate feature map after channel dimensionality reduction sequentially undergoes processing steps such as the first depthwise separable convolution, channel number expansion, non-linear activation, element-wise multiplication, channel number adjustment, the second depthwise separable convolution, and residual connection, and finally generates a high-dimensional feature map.

[0116] Specifically, in each Star Block module, the intermediate feature map after the input channel dimension reduction is first processed by the 7×7 first depth separable convolution module to extract the basic feature map; the obtained basic feature map is then passed to two 1×1 convolution layers in parallel to expand the number of channels to four times the original number. The output of the first 1×1 convolution layer is processed by the ReLU6 activation function and then multiplied element-by-element with the output of the second 1×1 convolution layer to generate a higher-dimensional feature map; then, the generated higher-dimensional feature map is passed through a 1×1 convolution layer to obtain a higher-dimensional feature map with the number of channels restored to the original number of channels; then, the higher-dimensional feature map with the number of channels restored to the original number of channels is further extracted by the 7×7 second depth separable convolution module to obtain a high-order feature map; finally, the obtained high-order feature map is regularized by the drop_path layer to prevent overfitting, and a regularized high-order feature map is obtained, and the obtained regularized high-order feature map is residually connected with the feature map originally input to the StarNet lightweight network to generate the final high-dimensional feature map.

[0117] Among them, the ReLU6 activation function is a linear activation function truncated between 0 and 6, which can be expressed by the following formula: , Represents the output of the first 1×1 convolutional layer.

[0118] Optionally, the path discarding probability of the drop_path layer in each Star Block module can be set to random. Based on this, the diversity and generalization ability of the target detection model can be enhanced, and the model's anti-interference ability to overfitting can be further improved.

[0119] In addition, in a specific implementation, the specific structure of BN_LSCD is obtained by replacing the group normalization (GN) in the lightweight LSCD (Lightweight Shared Convolutional Separamter) structure with BN, and setting the parameters of the BN layer not to be shared at different positions.

[0120] Specifically, when the BN_LSCD detection head receives the multi-scale feature map output after the neck of the target detection model performs multi-scale feature fusion on the high-dimensional feature map, first, the BN layer of the BN_LSCD detection head normalizes the feature map according to the mean and variance of the input image to obtain the normalized feature map. Among them, the parameters of the BN layer are all calculated independently and are not shared at different positions; then, the BN_Conv 1×1 module is used to expand the number of channels of the normalized feature map through 1×1 convolution, and deep feature extraction is performed on the feature map in combination with BN to generate the first intermediate feature map; then, the BN_Conv 3×3 module performs convolution operations on the first intermediate feature map to extract the spatial and channel information of the image, and the second intermediate feature map output by each of the two convolutional layers is obtained. Among them, each BN_Conv 3×3 module contains two convolutional layers with a convolution kernel of 3×3 that share weights; then, first, the element-wise multiplication module combines the second intermediate feature maps output by the two convolutional layers through element-wise multiplication, and then enhances the non-linear expression of the features through the ReLU6 activation function in the non-linear activation module to obtain a higher-dimensional feature map. Further, first, the number of channels of the higher-dimensional feature map is reduced to one-fourth of the original through the convolutional layer with a convolution kernel of 1×1 in the channel number adjustment module to obtain the feature map after channel dimensionality reduction, and then the second depthwise separable convolution module performs further feature extraction on the feature map after channel dimensionality reduction to obtain the high-order feature map; then, the obtained high-order feature map is connected with the multi-scale feature map originally input to the BN_LSCD detection head through a residual connection to obtain the final high-level feature map; finally, the prediction box convolution module (Convolution_Box, Conv_Box) and the classification convolution module (Convolution_Classification, Conv_Cls) perform object detection and classification prediction on the final high-level feature map, and output the target type of the target meter and the position information of the target meter in the image to be recognized.

[0121] It should be understood that the parameters of the BN layer of the BN_LSCD detection head provided in this application are not shared at different positions, which ensures that each BN layer can independently normalize the input data, making the normalization process at each position more flexible and adaptable to the feature distributions in different regions. It can be seen that by making the above improvements to the lightweight LSCD and applying it to the target detection model, while ensuring the lightweight and high efficiency of the model, the adaptability of the target detection model to different scales and different regions can be effectively improved, the stability of the model in large-scale data training can be enhanced, and the performance and generalization ability of the model can be improved.

[0122] Further, in a specific implementation manner, the target detection model can be obtained in the following way:

[0123] Step 1: Annotate the images of different types of meters collected to obtain an image dataset. The image dataset includes multiple images of different types of meters, and the annotation information of each image includes the meter type and the position of the meter.

[0124] In this step, it is necessary to annotate the meters in the image in the form of manual annotation, using the minimum bounding rectangle to obtain the annotation box of each image and the corresponding type and position of the annotation box. The type corresponding to the annotation box is used to indicate the meter type of the meter contained in the annotation box, and the position corresponding to the annotation box includes the center point coordinates, width, and height of the annotation box; according to the type and position corresponding to the annotation box in each image, the meter type and the position of the meter of each image are obtained, and then an image dataset including multiple images and the annotation information of each image is obtained.

[0125] Among them, the types of the annotation boxes at least include: barometer, pressure gauge, oil level gauge, oil pressure gauge.

[0126] In a specific implementation, among the images of multiple meters collected, different types of meters should be included, and the images of each type of meter under different lighting conditions and environments should be covered. Therefore, training the target detection model with the image dataset obtained based on the images of the meters obtained according to this implementation can improve the generalization ability and stability of the model.

[0127] Step 2: Set the backbone network of the YOLOv11 model to the StarNet lightweight network, and set the detection head to BN_LSCD to obtain a lightweight YOLOv11 model.

[0128] Step 3: Based on the lightweight YOLOv11 model, construct an initial target detection model;

[0129] Step 4: Train the initial target detection model according to the image dataset to obtain a target detection model.

[0130] In this implementation, by training the initial target detection model constructed based on the lightweight YOLOv11 model with an image dataset with annotation information, a lightweight target detection model capable of detecting and classifying the images to be recognized is obtained. Based on this target detection model, the recognition of meters can reduce the computing resources consumed by the recognition and improve the recognition efficiency.

[0131] S103: Crop the area where the target meter is located from the image to be recognized according to the position information to obtain a dial image.

[0132] In this step, in order to remove the interference information in the image to be recognized, it is necessary to crop the area where the target meter is located from the image to be recognized according to the position information of the target meter obtained by detection in the corresponding image to be recognized, so as to obtain an image containing the dial.

[0133] S104. Read the meter based on the target type and the dial image to obtain the reading of the target meter.

[0134] In this step, it is necessary to read the target meter in the dial image according to the target type of the target meter, so as to obtain the reading of the target meter.

[0135] Exemplarily, the range of the target meter can be determined according to the target type of the target meter, and geometric information such as the scale lines, pointers, and dial shape of the target meter can be extracted by using image processing technology on the dial image; based on the range of the target meter and the geometric information of the target meter, the reading of the target meter can be completed.

[0136] The method for identifying a substation meter provided in this embodiment obtains an image to be recognized; inputs the image to be recognized into a pre-acquired target detection model to obtain the target type of the target meter and the position information of the target meter in the image to be recognized; wherein, the target detection model is a model trained based on an image pre-calibrated with meter types and positions for classifying and determining the position of meters in the image; wherein, the lightweight YOLOv11 model is obtained by setting the backbone network of YOLOv11 as the StarNet lightweight network and setting the detection head as BN_LSCD; crop the area where the target meter is located from the image to be recognized according to the position information to obtain a dial image; read the meter based on the target type and the dial image to obtain the reading of the target meter, achieving the technical effects of reducing the computing resources consumed for identifying the meter reading and improving the recognition efficiency.

[0137] Furthermore, in the prior art, only the geometric features of the dial are extracted from the image to be recognized through image processing technology first, and then the meter is directly read based on the extracted geometric features. However, because the geometric features of the meter extracted based on the image are affected by fluctuations in image quality, the diversity of dial shapes, and the variability of the environment, the reading of the meter obtained only based on the meter image has a problem of poor accuracy.

[0138] Based on this, Embodiment 2 of this application provides a method for identifying a substation meter, which only extracts the scale lines and pointers of the target meter dial based on image processing technology, and obtains the reading of the target meter based on the scale lines and pointers of the dial and a pre-constructed dial model corresponding to the meter type of the target meter.

[0139] Specifically, Figure 2 is a schematic flowchart of the method for identifying substation meters provided in the second embodiment of the present application. As Figure 2 shown, based on the above embodiment, this embodiment provides a specific method for reading the meter based on the target type and the dial image to obtain the reading of the target meter, including:

[0140] S201. Obtain the scale lines and pointers of the target meter dial from the dial image.

[0141] In this step, the geometric features of the dial image should be extracted through image processing technology first, and the scale lines and pointers of the target meter dial are extracted.

[0142] Exemplarily, the image processing technology can be the Hough transform method, the edge detection method, etc.

[0143] In a specific implementation manner, obtaining the scale lines and pointers of the target meter dial from the dial image includes:

[0144] Step 1. Perform perspective transformation on the dial image to obtain the dial image in the front view.

[0145] In this step, due to factors such as the camera shooting angle or illumination, there will be a problem of perspective distortion in the dial image. Therefore, it is necessary to perform perspective transformation on the dial image to obtain the dial image in the front view for subsequent extraction of the scale lines and pointers of the target meter dial.

[0146] Among them, the mathematical expression of the perspective transformation can be represented by the homography matrix H:

[0147]

[0148] Among them, is the coordinate after perspective transformation, is the coordinate in the original image, and H is the homography matrix.

[0149] Step 2. Use the Hough transform method for the dial image in the front view to extract the scale lines and pointers of the target meter dial.

[0150] In this step, it is necessary to use the Hough transform method to detect straight lines in the dial image in the front view and extract the scale lines and pointers of the target meter dial.

[0151] S202. According to the target type, obtain the dial model corresponding to the target meter. The dial model includes the parameters of the circular outer frame of the meter of the target type, the scale line information, the change range of the reading, and the angular range of the pointer.

[0152] In this step, according to the target type, it is necessary to select the dial model corresponding to the target type from multiple pre-constructed dial models corresponding to different meter types as the dial model corresponding to the target meter. Among them, the dial model includes the parameters of the circular outer frame of the meter corresponding to its meter type, scale line information, the variation range of readings, and the angular range of the pointer.

[0153] Specifically, the parameters of the circular outer frame of the meter include the radius and center of the circular outer frame; the scale line information includes the position of the scale line, the angle and length of each scale line. Among them, the angle of each scale line is determined based on the included angle between the line connecting the center of the circular frame to the scale line and the line connecting the center of the circular frame to the zero scale line; the variation range of readings refers to the range composed of the maximum reading and the minimum reading of the meter (i.e., the range of the meter); the angular range of the pointer refers to the range composed of the maximum value and the minimum value that the pointer can rotate.

[0154] In a specific implementation manner, the following method can be used to construct the dial models corresponding to different meter types:

[0155] For each meter of each meter type, select a clear image including the meter, and preprocess the image to obtain the preprocessed image. Among them, the preprocessing includes denoising processing, cropping processing based on the minimum bounding rectangle, and scaling processing to ensure that the image is of a unified size; extract the parameters of the circular frame of the meter through circle detection by the Hough transform and extract the position of the scale line of the meter and the length of each scale line through line detection by the Hough transform; determine the variation range of the meter readings and the angular range of the pointer according to prior knowledge, and construct the dial model corresponding to the meter type based on the parameters of the circular frame of the meter, the position of the scale line, the length of each scale line, the variation range of the meter readings, and the angular range of the pointer.

[0156] S203. Align the scale lines of the target meter dial with the scale lines of the dial model, and determine the pointer angle based on the pointer of the target meter dial and the dial model after alignment.

[0157] In this step, it is necessary to align the scale lines of the target meter dial with the scale lines of the dial model according to the position and length of the scale lines in the scale line information of the dial model, and determine the pointer angle based on the pointer of the target meter dial and the scale line information in the dial model after alignment. Among them, the pointer angle refers to the angle between the pointer and the reference line with the line connecting the center of the circular frame to the zero scale line as the reference line.

[0158] It should be understood that this step determines the pointer angle based on the scale line information in the dial model, which can effectively avoid the problem of inaccurate readings of the target meter caused by incorrect recognition of the scale lines of the dial extracted from the dial image corresponding to the image to be recognized (for example, there are scale lines that are not extracted or there are mis-extracted scale lines), and improves the robustness of the meter recognition method.

[0159] Optionally, it can first be determined whether the pointer angle of the target meter falls within the angle range of the pointer in the dial model. If it falls within the angle range of the pointer in the dial model, the processing in step S204 can be performed to obtain the degree of the target meter; if it does not fall within the angle range of the pointer in the dial model, new scale lines and a new pointer of the target meter dial are obtained from the dial image again, and a new pointer angle is determined based on the new pointer to complete the correction of the pointer angle, effectively avoiding problems such as reading drift caused by incorrect extraction of the pointer.

[0160] S204. Obtain the reading of the target meter according to the pointer angle.

[0161] In this step, according to the obtained pointer angle and the change range of the readings in the dial model, the reading of the target meter is obtained.

[0162] In a specific implementation manner, the following two methods can be used to obtain the reading of the target meter according to the pointer angle:

[0163] Method 1. Calculate the reading of the target meter according to the pointer angle and the reading calculation formula of the meter of the target type.

[0164] Among them, the reading calculation formula of the meter of the target type is as follows:

[0165]

[0166] In the above formula, is the minimum value of the change range of the readings included in the dial model;

[0167] is the maximum value of the change range of the readings included in the dial model; is the angle of the full scale line included in the dial model; is the angle of the zero scale line included in the dial model; represents the pointer angle.

[0168] Method 2. Determine the reading corresponding to the pointer angle according to the mapping relationship between the angle and the reading of the pointer of the meter of the target type included in the dial model, and use it as the reading of the target meter.

[0169] Specifically, based on the angle of each scale line included in the dial model and the range of readings, the mapping relationship between the angle of the scale line in the dial model and the reading can be obtained. Since the pointer angle has a corresponding relationship with the scale line angle in the dial model, the mapping relationship between the angle of the pointer of the meter in the dial model and the reading can be obtained. Based on this, the reading corresponding to the pointer angle can be determined as the reading of the target meter according to the mapping relationship between the angle of the pointer of the target type of meter included in the dial model and the reading.

[0170] The method for identifying substation meters provided in an embodiment of the present application first obtains the scale lines and pointer of the target meter dial from the dial image; then obtains the dial model corresponding to the target meter according to the target type; then, aligns the scale lines of the target meter dial with the dial model, and after the alignment, determines the pointer angle based on the pointer of the target meter dial and the dial model; finally, obtains the means for obtaining the reading of the target meter according to the pointer angle, thereby improving the accuracy of the reading of the identified target meter and the robustness of the method for identifying substation meters.

[0171] Figure 3 A flow chart of a method for identifying a substation meter provided in Embodiment 3 of the present application is shown as follows: Figure 3 Based on the above embodiment, the identification method of the substation meter provided in this embodiment further includes:

[0172] S301. According to the reading of the target meter at the current moment and the reading of the target meter within a preset time period before the current moment, a reading prediction model is used to predict the reading of the target meter to obtain the predicted reading of the target meter at the next time step; wherein the reading prediction model is a prediction model obtained by training the LSTM model based on historical meter reading data.

[0173] In this solution, it is necessary to collect the image to be identified of the target meter at a preset time interval, and for each image to be identified, the method in the above embodiment is used to obtain the reading of the target meter corresponding to the image.

[0174] In this step, it is necessary to use a reading prediction model to predict the reading of the target meter based on the reading of the target meter obtained based on the image to be identified including the target meter at the current moment, and the readings of the target meter at multiple time steps obtained based on multiple images to be identified within a preset time period before the current moment, so as to obtain the predicted reading of the target meter at the next time step, wherein the reading prediction model is a prediction model obtained by training a Long Short-Term Memory (LSTM) model based on historical meter reading data.

[0175] In the actual application of this solution, first, readings of each type of meter are taken at preset time intervals to obtain time-series reading data corresponding to different types of meters (i.e., historical meter reading data). Among them, the preset time interval is determined according to the meter change speed and the data storage capacity of the computer device. Exemplarily, the preset time interval can be ten minutes, and the present application does not limit the setting of the preset time interval; the obtained historical meter reading data is subjected to data cleaning, and the cleaned data is normalized to obtain a historical meter reading data set for training the LSTM model. Optionally, the maximum-minimum (Minimum-Maximum, Min-Max) normalization method can be used to normalize the cleaned data.

[0176] In a specific implementation manner, during the training process of the LSTM model, the mean squared error (MSE) can be used as the loss function for optimizing the parameters of the LSTM model; gradient descent can be performed according to the Adam optimizer for optimizing the parameters of the LSTM model. Among them, the parameters of the model include the number of hidden layer units, the number of layers, the learning rate, etc. set by the LSTM model.

[0177] Optionally, the training period of the LSTM model can be set to 150.

[0178] S302. According to the fluctuation condition of the historical readings and the reading range of the meter of the preset target type, diagnose the predicted readings to obtain a first diagnosis result.

[0179] In this step, the normal fluctuation range, change trend, and periodic characteristics of the historical readings can be determined according to the fluctuation condition of the historical readings; the reading range of the meter of the preset target type refers to the reasonable reading interval corresponding to the target type of meter; based on the fluctuation condition of the historical readings and the reading range of the meter of the preset target type, diagnose the predicted readings to obtain a first diagnosis result, where the first diagnosis result is used to indicate whether the predicted readings are abnormal.

[0180] The method for identifying substation meters provided by the embodiments of the present application adopts a reading prediction model to predict the readings of a target meter based on the readings of the target meter at the current moment and the readings of the target meter within a preset duration before the current moment, so as to obtain the predicted readings of the target meter at the next time step. Then, according to the fluctuation of historical readings and the reading range of meters of a preset target type, the predicted readings are diagnosed to obtain a first diagnosis result. This realizes the prediction of the reading change trend of the meter. Based on the predicted readings of the meter at the next time step, early warning of abnormal readings of the meter can be achieved, effectively preventing the occurrence of sudden failures. In addition, by combining the fluctuation of historical readings to diagnose the predicted readings, the error of the diagnosis result caused by short-term fluctuations can be effectively avoided.

[0181] Figure 4 FIG. 4 is a schematic flowchart of the method for identifying substation meters provided by Embodiment 4 of the present application. As Figure 4 described, on the basis of the above embodiments, the method for identifying substation meters provided in this embodiment further includes:

[0182] S401. Diagnose the readings of the target meter according to the prior knowledge obtained in advance to obtain a second diagnosis result, where the prior knowledge includes: the key reading range and the judgment rules for at least one abnormal situation.

[0183] In this step, according to the prior knowledge obtained in advance, the readings of the target meter obtained from the to-be-identified image including the target meter at the current moment are diagnosed to obtain a second diagnosis result, where the second diagnosis result is used to indicate whether the readings of the target meter are abnormal.

[0184] Specifically, the prior knowledge includes: the key reading range and the judgment rules for at least one abnormal situation.

[0185] Among them, the key reading range refers to the range in which the equipment monitored by the target meter operates normally. Exemplarily, if the reading change range of a certain pressure gauge is [0, 100], and when the reading of the target meter is within the interval [20, 80], it indicates that the pressure of the equipment monitored by the target meter is within the normal range, then the key reading range of the target meter is [20, 80]. If the reading of the target meter is not within the key reading range, it is diagnosed that the reading of the target meter is abnormal.

[0186] Exemplarily, the judgment rules for abnormal situations may include:

[0187] Rule 1. Reading increase rule: If the target meter type is an oil level meter and the change between two consecutive readings exceeds a certain threshold, it is diagnosed that the equipment detected by the meter has a failure, and at this time, it is considered that the reading of the target meter is abnormal.

[0188] Rule 2, Pressure Fluctuation Rule: If the target meter type is an oil and gas pressure gauge and the reading of the meter fluctuates continuously by more than ±5 Pa within a short period of time, it is considered that there may be a problem with the pressure control system monitored by the meter. At this time, the second diagnostic result needs to indicate that the reading of the target meter is abnormal.

[0189] Rule 3, Long-term Trend Abnormality Rule: If the target meter type is an oil level gauge and the reading of the oil level gauge suddenly shows a gradually rising or falling trend after a stable change over a period of time, it is considered that there may be a potential leakage problem with the equipment monitored by the meter. At this time, it is considered that the reading of the target meter is abnormal.

[0190] S402. When the first diagnostic result indicates that the predicted reading is abnormal and / or the second diagnostic result indicates that the reading is abnormal, push an alarm message.

[0191] In this step, according to the content indicated by the first diagnostic result obtained in step S302 and the content indicated by the second diagnostic result obtained in step S401, when the first diagnostic result indicates that the predicted reading is abnormal and / or the second diagnostic result indicates that the reading is abnormal, an alarm message is pushed to the staff to facilitate the staff to take timely handling measures. Among them, the alarm message includes: meter type, current reading, predicted trend, abnormal reason, and handling measures.

[0192] It should be understood that the current reading refers to the reading of the target meter obtained based on the to-be-identified image including the target meter at the current moment; the predicted trend is the change trend of the reading of the target meter determined according to the predicted reading; the abnormal reason refers to the reason for the abnormal reading determined according to the diagnostic conditions, where the diagnostic conditions include the prior knowledge for diagnosing the current reading, the fluctuation of the historical readings for diagnosing the predicted reading, and the reading range of the meters of the preset target type; the prior rules may also include the handling measures corresponding to the judgment rules for each abnormal situation.

[0193] Exemplarily, the predicted trend may refer to the change trend of the predicted reading obtained in step S301 compared with the current reading. By comparing the two, it is possible to further determine whether the current reading and the predicted reading of the meter are abnormal.

[0194] In a possible implementation manner, the prior knowledge and / or the reading prediction model can be updated according to the data in the alarm message.

[0195] In this implementation manner, a judgment rule for a new abnormal situation can be formed based on the meter type, current reading, predicted trend, and abnormal reason given in the alarm information to update the prior knowledge and further improve the reliability of the second diagnosis result. In addition, the alarm information may further include new historical meter reading data including the readings with abnormal reading situations. Correspondingly, according to the new historical meter reading data, the reading prediction model can be updated by means of incremental learning, so that the reading prediction model adapts to the state change of the device monitored by the meter and improves the accuracy of the predicted reading of the meter predicted by the model.

[0196] The method for identifying a substation meter provided in this embodiment analyzes the meter reading from two aspects, the current perspective and the future trend, by comprehensively analyzing the second diagnosis result obtained according to the current reading and the first diagnosis result obtained according to the predicted reading, so as to improve the accuracy of abnormal detection. By promptly alarming the staff about the abnormal situation of the meter, reliable monitoring of the device monitored by the meter is realized.

[0197] Figure 5 It is a schematic structural diagram of the substation meter identification device provided in the fifth embodiment of the present application, as Figure 5 shown, the substation meter identification device 50 provided in this embodiment includes:

[0198] An acquisition unit 501, configured to acquire an image to be recognized, where the image to be recognized includes a target meter for which data is to be read;

[0199] A detection unit 502, configured to input the image to be recognized into a pre-acquired target detection model to obtain the target type of the target meter and the position information of the target meter in the image to be recognized; wherein, the target detection model is a model for classifying and determining the position of the meter in the image obtained by training the lightweight YOLOv11 model based on the image with the meter type and position pre-calibrated; the lightweight YOLOv11 model is obtained by setting the backbone network of YOLOv11 to the StarNet lightweight network and setting the detection head to BN_LSCD;

[0200] A processing unit 503, configured to crop out the area where the target meter is located from the image to be recognized according to the position information to obtain a dial image;

[0201] A determination unit 504, configured to read the meter based on the target type and the dial image to obtain the reading of the target meter.

[0202] The recognition device 50 of the substation meter provided in this embodiment can execute the method provided in the above method embodiment, and its implementation principle and technical effects are similar, which will not be elaborated here in this embodiment.

[0203] Figure 6 It is a schematic structural diagram of the recognition device of the substation meter provided in the sixth embodiment of the present application. As Figure 6 shown, on the basis of the above embodiment, the recognition device 50 of the substation meter provided in this embodiment includes:

[0204] A labeling unit 505, configured to label the images of different types of meters collected to obtain an image data set, where the image data set includes multiple images of different types of meters, and the labeling information of each image includes the meter type and the position of the meter;

[0205] A setting unit 506, configured to set the backbone network of the YOLOv11 model to the StarNet lightweight network and set the detection head to BN_LSCD to obtain a lightweight YOLOv11 model;

[0206] A construction unit 507, configured to construct an initial object detection model based on the lightweight YOLOv11 model;

[0207] A training unit 508, configured to train the initial object detection model according to the image data set to obtain the object detection model.

[0208] In a possible implementation manner, the determination unit 504 includes:

[0209] A first acquisition module 5041, configured to acquire the scale lines and pointers of the target meter dial from the dial image;

[0210] A second acquisition module 5042, configured to acquire the corresponding dial model of the target meter according to the target type, where the dial model includes the parameters of the circular outer frame of the meter of the target type, scale line information, the change range of the reading, and the angle range of the pointer;

[0211] An alignment module 5043, configured to align the scale lines of the target meter dial with the dial model, and determine the pointer angle based on the pointer of the target meter dial and the dial model after alignment;

[0212] A determination module 5044, configured to obtain the reading of the target meter according to the pointer angle.

[0213] In a possible implementation manner, the determination module 5044 is specifically configured to:

[0214] Calculate the reading of the target meter according to the pointer angle and the reading calculation formula of the meter of the target type.

[0215] Or,

[0216] Determine the reading corresponding to the pointer angle according to the mapping relationship between the angle and the reading of the pointer of the meter of the target type included in the dial model, and use it as the reading of the target meter.

[0217] In a possible implementation manner, the first acquisition module 5041 is specifically configured to:

[0218] Perform perspective transformation on the dial image to obtain a dial image in a front view angle;

[0219] Use the Hough transform method on the dial image in the front view angle to extract the scale lines and pointers of the dial of the target meter.

[0220] In a possible implementation manner, the substation meter identification device 50 further includes:

[0221] A prediction unit 509, configured to perform reading prediction on the target meter according to the reading of the target meter at the current moment and the readings of the target meter within a preset time period before the current moment, and obtain a predicted reading of the target meter at the next time step; wherein, the reading prediction model is a prediction model trained based on historical meter reading data for an LSTM model;

[0222] A first diagnosis unit 510, configured to diagnose the predicted reading according to the fluctuation condition of historical readings and the preset reading range of the meter of the target type, and obtain a first diagnosis result, where the first diagnosis result is used to indicate whether the predicted reading is abnormal.

[0223] In a possible implementation manner, the substation meter identification device 50 further includes:

[0224] A second diagnosis unit 511, configured to diagnose the reading of the target meter according to the prior knowledge obtained in advance, and obtain a second diagnosis result, where the second diagnosis result is used to indicate whether the reading of the target meter is abnormal; wherein, the prior knowledge includes: a key reading range and judgment rules for at least one abnormal situation;

[0225] An alarm unit 512, configured to push an alarm message if the first diagnosis result indicates that the predicted reading is abnormal and / or the second diagnosis result indicates that the reading is abnormal; wherein, the alarm message includes: meter type, current reading, prediction trend, abnormal reason, and treatment measures.

[0226] In a possible implementation manner, the identification device 50 for substation meters further includes:

[0227] An update unit 513, configured to update the prior knowledge and / or the reading prediction model according to the data in the alarm information.

[0228] The identification device 50 for substation meters provided in this embodiment can execute the method provided in the above method embodiment, and its implementation principle and technical effect are similar, which will not be elaborated here in this embodiment.

[0229] Figure 7 It is a schematic structural diagram of the computer device provided in this application. As Figure 7 shown, the computer device 60 provided in this embodiment includes: at least one processor 601 and a memory 602. Optionally, the device 60 further includes a communication component 603. Among them, the processor 601, the memory 602, and the communication component 603 are connected through a bus 604.

[0230] In a specific implementation process, at least one processor 601 executes the computer execution instructions stored in the memory 602, so that at least one processor 601 executes the above method.

[0231] The specific implementation process of the processor 601 can refer to the above method embodiment, and its implementation principle and technical effect are similar, which will not be elaborated here in this embodiment.

[0232] This application also provides a computer-readable storage medium, in which computer execution instructions are stored. When the processor executes the computer execution instructions, the above method is implemented.

[0233] This application also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0234] Finally, it should be noted that: After those skilled in the art consider the specification and practice the invention disclosed herein, other implementation manners of the invention will be easily thought of. The invention is intended to cover any variations, uses, or adaptations of the invention, which follow the general principles of the invention and include the common general knowledge or conventional technical means in the technical field not disclosed in the invention. It is not limited to the exact structure already described and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is only limited by the appended claims.

Claims

1. A method for identifying a substation meter, characterized in that: include: Acquire an image to be identified, wherein the image to be identified includes a target meter to be read data; Input the image to be identified into a pre-acquired target detection model to obtain the target type of the target meter and the location information of the target meter in the image to be identified; wherein the target detection model is a model for classifying and determining the location of meters in an image obtained by training a lightweight YOLOv11 model based on an image in which the type and location of the meter are pre-calibrated; the lightweight YOLOv11 model is obtained by setting the backbone network of YOLOv11 to the StarNet lightweight network and setting the detection head to BN_LSCD; Cut out the area where the target meter is located from the image to be identified according to the position information to obtain a dial image; A meter reading is performed based on the target type and the dial image to obtain a reading of the target meter.

2. The method according to claim 1, characterized in that: The step of reading a meter based on the target type and the dial image to obtain the target meter reading includes: Acquire the scale lines and pointer of the target meter dial from the dial image; According to the target type, a dial model corresponding to the target meter is obtained, wherein the dial model includes parameters of a circular outer frame of the target type meter, scale line information, a range of change of a reading, and an angle range of a pointer; Aligning the scale lines of the target meter dial with the dial model, and determining the pointer angle based on the pointer of the target meter dial and the dial model after the alignment; According to the pointer angle, the reading of the target meter is obtained.

3. The method according to claim 2, characterized in that The step of obtaining the target meter reading according to the pointer angle includes: Calculate the reading of the target meter according to the pointer angle and the reading calculation formula of the meter of the target type; or, According to the mapping relationship between the angle of the pointer of the target type of meter included in the dial model and the reading, the reading corresponding to the pointer angle is determined as the reading of the target meter.

4. The method according to claim 2, characterized in that: The step of acquiring the scale lines and pointer of the target meter dial from the dial image includes: Performing perspective transformation on the dial image to obtain a dial image at a normal viewing angle; The Hough transform method is used on the dial image under the positive viewing angle to extract the scale lines and pointer of the target meter dial.

5. The method according to any one of claims 1 to 4, characterized in that: Before inputting the image to be recognized into a pre-acquired target detection model, the method further includes: Annotating the collected images of different types of meters to obtain an image data set, wherein the image data set includes multiple images of different types of meters, and the annotation information of each image includes the type of meter and the location of the meter; Set the backbone network of the YOLOv11 model to the StarNet lightweight network, and set the detection head to BN_LSCD to obtain a lightweight YOLOv11 model; Based on the lightweight YOLOv11 model, an initial target detection model is constructed; The initial target detection model is trained according to the image data set to obtain the target detection model.

6. The method according to any one of claims 1 to 4, characterized in that: The method further comprises: According to the reading of the target meter at the current moment and the reading of the target meter within a preset time before the current moment, a reading prediction model is used to predict the reading of the target meter to obtain the predicted reading of the target meter at the next time step; wherein the reading prediction model is a prediction model obtained by training the LSTM model based on historical meter reading data; The predicted reading is diagnosed according to the fluctuation of the historical readings and the preset reading range of the meter of the target type to obtain a first diagnosis result, and the first diagnosis result is used to indicate whether the predicted reading is abnormal.

7. The method according to claim 6, characterized in that The method further comprises: According to the prior knowledge acquired in advance, the reading of the target meter is diagnosed to obtain a second diagnosis result, and the second diagnosis result is used to indicate whether the reading of the target meter is abnormal; wherein the prior knowledge includes: a critical reading range and at least one abnormal situation judgment rule; If the first diagnostic result indicates that the predicted reading is abnormal and / or the second diagnostic result indicates that the reading is abnormal, an alarm message is pushed; wherein the alarm message includes: meter type, current reading, predicted trend, abnormal cause and treatment measures.

8. The method according to claim 7, characterized in that The method further comprises: The prior knowledge and / or the reading prediction model are updated according to the data in the alarm information.

9. A device for identifying a substation meter, characterized in that: include: An acquisition unit, used for acquiring an image to be identified, wherein the image to be identified includes a target meter to be read data; A detection unit, used for inputting the image to be identified into a pre-acquired target detection model to obtain the target type of the target meter and the position information of the target meter in the image to be identified; wherein the target detection model is a model for classifying and determining the position of the meter in the image obtained by training a lightweight YOLOv11 model based on an image in which the meter type and position are pre-calibrated; the lightweight YOLOv11 model is obtained by setting the backbone network of YOLOv11 to the StarNet lightweight network and setting the detection head to BN_LSCD; A processing unit, configured to cut out the area where the target meter is located from the image to be identified according to the position information, to obtain a dial image; A determination unit is used to perform meter reading based on the target type and the dial image to obtain the target meter reading.

10. A computer device, characterized in that: include: Memory, processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory, so that the processor performs the method according to any one of claims 1 to 8.

Images