Meter monitoring identification model training method, identification method, device and medium

Abstract

The application provides a meter monitoring identification model training method, an identification method, equipment and a medium, and belongs to the technical field of intelligent power equipment monitoring. The method obtains a plurality of training samples. For each training sample, an image feature map of a sample meter image is extracted through a backbone network module of an initial identification model, the image feature map is subjected to feature enhancement through a neck network module of the initial identification model, the image enhanced feature map is processed through a detection head module of the initial identification model, and prediction labeling information is obtained. Finally, based on sample labeling information of the sample meter image, the prediction labeling information and a preset loss function, the initial identification model is trained to obtain a meter monitoring identification model. The technical scheme trains the initial identification model, so that the meter monitoring identification model has high identification accuracy, the obtained meter monitoring identification model can be used for meter identification, and meter monitoring and state identification can be efficiently performed.

Description

Technical Field

[0001] This application relates to the field of intelligent power equipment monitoring technology, and in particular to a training method, identification method, equipment and medium for a meter monitoring and identification model. Background Technology

[0002] With the digital and intelligent transformation of the power industry and the widespread application of advanced technologies such as smart grids, the Internet of Things, big data, and cloud computing, the monitoring and management of power equipment has entered a new stage of development. Therefore, real-time monitoring and identification of the meter status of power equipment has become an important part of the power system.

[0003] In existing technologies, meter monitoring and identification methods for power equipment suffer from low meter identification accuracy.

[0004] Therefore, improving the accuracy of meter monitoring and identification methods has become an urgent technical problem to be solved. Summary of the Invention

[0005] This application provides a training method, identification method, device, and medium for a meter monitoring and identification model, so as to improve the identification accuracy of the meter monitoring and identification method.

[0006] In a first aspect, embodiments of this application provide a training method for a meter monitoring and identification model, comprising:

[0007] Obtain multiple training samples; each training sample includes a sample meter image and sample annotation information corresponding to the sample meter image;

[0008] For each training sample, the image feature map corresponding to the sample meter image is extracted through the backbone network module of the initial recognition model;

[0009] The image feature map is enhanced by using the neck network module of the initial recognition model to obtain an enhanced image feature map.

[0010] The image enhancement feature map is processed by the detection head module of the initial recognition model to obtain the prediction annotation information;

[0011] Based on the sample annotation information, prediction annotation information, and preset loss function of the sample meter images, the initial recognition model is trained to obtain the meter monitoring recognition model.

[0012] In one possible implementation, the neck network module includes a pooling layer, a self-attention mechanism layer, and a splicing layer;

[0013] The image feature map is enhanced by using the neck network module of the initial recognition model, resulting in an enhanced image feature map, including:

[0014] The image feature map is input into the pooling layer to obtain a multi-scale feature map corresponding to the image feature map;

[0015] The image feature map and the corresponding multi-scale feature map are respectively input into the self-attention mechanism layer to adjust the channel weights of the image feature map and the multi-scale feature map respectively;

[0016] The image feature map with adjusted channel weights and the corresponding multi-scale feature map are input into the stitching layer for fusion processing to obtain the image enhancement feature map.

[0017] In one possible implementation, an initial identification model is trained based on sample annotation information, prediction annotation information, and a preset loss function to obtain a meter monitoring identification model, including:

[0018] Based on a preset loss function, calculate the loss value between sample annotation information and predicted annotation information;

[0019] Adjust the parameters of the initial identification model based on the loss value until the preset convergence condition is met to obtain the meter monitoring identification model.

[0020] In one possible implementation, multiple training samples are obtained, including:

[0021] Obtain the raw meter image;

[0022] The original meter images are augmented to obtain sample meter images;

[0023] The visual detection algorithm is used to obtain the sample annotation information corresponding to the sample meter image.

[0024] In one possible implementation, a visual detection algorithm is used to obtain sample annotation information corresponding to the sample meter image, including:

[0025] The sample meter images are processed using a visual detection algorithm to obtain a binarized image corresponding to the sample meter images;

[0026] Pointer separation processing is performed on the binarized image to obtain a pointer-separated image;

[0027] The pointer separation image is identified to obtain the pointer recognition result;

[0028] The pointer recognition result is used to determine the sample annotation information corresponding to the sample meter image.

[0029] Secondly, embodiments of this application provide a meter monitoring and identification method, wherein the executing entity is a meter identification terminal, including:

[0030] Acquire the image of the meter to be monitored for identification;

[0031] The image to be identified is input into the meter monitoring and identification model for processing to obtain the meter identification result. The meter monitoring and identification model is pre-trained according to the training method of the meter monitoring and identification model of the first aspect or any other method.

[0032] The meter readings are sent to the server.

[0033] In one possible implementation, acquiring the image to be identified of the meter to be monitored includes:

[0034] Set the camera parameters of the image acquisition device according to real-time environmental data;

[0035] Based on camera parameters, capture images of the meters to be monitored for identification.

[0036] In one possible implementation, sending the meter reading to the server includes:

[0037] Based on the WAPI protocol, a wireless LAN authentication and confidentiality infrastructure, the meter reading results are encrypted and then sent to the server.

[0038] Thirdly, embodiments of this application provide a meter monitoring and identification method, wherein the execution entity is a server, including:

[0039] The meter identification terminal receives the meter identification result of the meter to be monitored; wherein, the meter identification result is the meter identification terminal acquiring the image of the meter to be monitored, identifying the image of the meter to be monitored through the meter monitoring and identification model, and encrypting the meter identification result based on the Wireless Local Area Network Authentication and Confidentiality Infrastructure (WAPI) protocol before sending it; the meter monitoring and identification model is pre-trained according to the training method of the meter monitoring and identification model in the first aspect or the second aspect.

[0040] The meter readings are decrypted and displayed based on the WAPI protocol.

[0041] Fourthly, embodiments of this application provide a training device for a meter monitoring and identification model, comprising:

[0042] The first acquisition module is used to acquire multiple training samples; wherein each training sample includes a sample meter image and sample annotation information corresponding to the sample meter image;

[0043] The first processing module is used to extract the image feature map corresponding to the sample meter image through the backbone network module of the initial recognition model for each training sample.

[0044] The second processing module is used to enhance the image feature map through the neck network module of the initial recognition model to obtain the image enhanced feature map.

[0045] The third processing module is used to process the image enhancement feature map through the detection head module of the initial recognition model to obtain the prediction annotation information;

[0046] The determination module is used to train the initial recognition model based on the sample annotation information, prediction annotation information and preset loss function of the sample meter images to obtain the meter monitoring recognition model.

[0047] In one possible implementation, the neck network module includes a pooling layer, a self-attention mechanism layer, and a splicing layer;

[0048] The second processing module is specifically used for:

[0049] The image feature map is input into the pooling layer to obtain a multi-scale feature map corresponding to the image feature map;

[0050] The image feature map and the corresponding multi-scale feature map are respectively input into the self-attention mechanism layer to adjust the channel weights of the image feature map and the multi-scale feature map respectively;

[0051] The image feature map with adjusted channel weights and the corresponding multi-scale feature map are input into the stitching layer for fusion processing to obtain the image enhancement feature map.

[0052] In one possible implementation, the determining module is specifically used for:

[0053] Based on a preset loss function, calculate the loss value between sample annotation information and predicted annotation information;

[0054] Adjust the parameters of the initial identification model based on the loss value until the preset convergence condition is met to obtain the meter monitoring identification model.

[0055] In one possible implementation, the first acquisition module is specifically used for:

[0056] Obtain the raw meter image;

[0057] The original meter images are augmented to obtain sample meter images;

[0058] The visual detection algorithm is used to obtain the sample annotation information corresponding to the sample meter image.

[0059] In one possible implementation, the first acquisition module acquires sample annotation information corresponding to the sample meter image through a visual detection algorithm, specifically for:

[0060] The sample meter images are processed using a visual detection algorithm to obtain a binarized image corresponding to the sample meter images;

[0061] Pointer separation processing is performed on the binarized image to obtain a pointer-separated image;

[0062] The pointer separation image is identified to obtain the pointer recognition result;

[0063] The pointer recognition result is used to determine the sample annotation information corresponding to the sample meter image.

[0064] Fifthly, embodiments of this application provide a meter monitoring and identification device, the executing entity of which is a meter identification terminal, including:

[0065] The second acquisition module is used to acquire the image to be identified of the meter to be monitored;

[0066] The fourth processing module is used to input the image to be identified into the meter monitoring and identification model for processing to obtain the meter identification result. The meter monitoring and identification model is pre-trained according to or any of the meter monitoring and identification model training methods.

[0067] The sending module is used to send the meter reading results to the server.

[0068] In one possible implementation, the second acquisition module is specifically used for:

[0069] Set the camera parameters of the image acquisition device according to real-time environmental data;

[0070] Based on camera parameters, capture images of the meters to be monitored for identification.

[0071] In one possible implementation, the sending module is specifically used for:

[0072] Based on the WAPI protocol, a wireless LAN authentication and confidentiality infrastructure, the meter reading results are encrypted and then sent to the server.

[0073] Sixthly, embodiments of this application provide a meter monitoring and identification device, the execution entity of which is a server, including:

[0074] The receiving module is used to receive the meter identification result of the meter to be monitored sent by the meter identification terminal; wherein, the meter identification result is the image of the meter to be monitored that the meter identification terminal obtains, identifies the image of the meter to be monitored through the meter monitoring and identification model, and encrypts the meter identification result based on the Wireless Local Area Network Authentication and Confidentiality Infrastructure (WAPI) protocol before sending it; the meter monitoring and identification model is pre-trained according to the training method of the meter monitoring and identification model in the first aspect or the second aspect.

[0075] The fifth processing module is used to decrypt and display the meter identification results based on the WAPI protocol.

[0076] In a seventh aspect, embodiments of this application provide an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0077] The memory stores the instructions that the computer executes;

[0078] The processor executes computer-executable instructions stored in memory to implement the method as described in the first aspect or any of the above.

[0079] Eighthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the method described in the first aspect or any of the above methods.

[0080] Ninthly, embodiments of this application provide a computer program, the computer program product including the computer program, the computer program being stored in a computer-readable storage medium, at least one processor being able to read the computer program from the computer-readable storage medium, and at least one processor executing the computer program being able to implement the methods of the first aspect or any of the above-described methods.

[0081] This application provides a training method, recognition method, device, and medium for a meter monitoring and identification model. The training method first acquires multiple training samples. Each training sample includes a sample meter image and corresponding sample annotation information. For each training sample, the backbone network module of the initial identification model extracts the image feature map corresponding to the sample meter image. Then, the neck network module of the initial identification model enhances the image feature map to obtain an enhanced image feature map. The detection head module of the initial identification model processes the enhanced image feature map to obtain predicted annotation information. Finally, based on the sample annotation information, predicted annotation information, and a preset loss function, the initial identification model is trained to obtain the meter monitoring and identification model. This technical solution, through training the initial identification model, enables the meter monitoring and identification model to have high recognition accuracy. The obtained meter monitoring and identification model can be used for meter identification, enabling efficient meter monitoring and status recognition. Attached Figure Description

[0082] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0083] Figure 1 A schematic diagram of the architecture of the meter monitoring and identification model provided in the embodiments of this application;

[0084] Figure 2 A flowchart illustrating the training method for the meter monitoring and identification model provided in this application embodiment;

[0085] Figure 3 Flowchart of the meter monitoring and identification method provided in the embodiments of this application Figure 1 ;

[0086] Figure 4 Flowchart of the meter monitoring and identification method provided in the embodiments of this application Figure 2 ;

[0087] Figure 5 System architecture diagram of the training method for the meter monitoring and identification model provided in the embodiments of this application;

[0088] Figure 6 Flowchart of the meter monitoring and identification method provided in the embodiments of this application Figure 3 ;

[0089] Figure 7 A schematic diagram of the structure of the training device for the meter monitoring and identification model provided in the embodiments of this application;

[0090] Figure 8 Schematic diagram of the structure of the meter monitoring and identification device provided in the embodiments of this application Figure 1 ;

[0091] Figure 9 Schematic diagram of the structure of the meter monitoring and identification device provided in the embodiments of this application Figure 2 ;

[0092] Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0093] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0094] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0095] Before introducing the embodiments of this application, the application background of the embodiments of this application will be explained first:

[0096] With the digital and intelligent transformation of the power industry and the widespread application of advanced technologies such as smart grids, the Internet of Things, big data, and cloud computing, the monitoring and management of power equipment has entered a new stage of development. Therefore, real-time monitoring and identification of the meter status of power equipment has become an important part of the power system.

[0097] Among them, the meters of power equipment can be sulfur hexafluoride (SF6) meters. SF6 is a key insulation and arc-extinguishing medium in high-voltage power equipment, and its real-time monitoring is crucial to ensuring the safe and stable operation of the power system.

[0098] Existing meter monitoring methods rely on manual inspections, which are inefficient and inaccurate, and cannot achieve real-time monitoring and identification. Other methods use traditional wireless networks to transmit images of the meters to be monitored to a server, which then performs meter identification. However, traditional wireless networks have security vulnerabilities.

[0099] Therefore, improving the accuracy of meter monitoring and identification methods while enhancing the security of data transmission has become an urgent technical problem to be solved.

[0100] To address the technical problems existing in the prior art, the inventors of this application propose the following concept: To address the low accuracy of existing meter monitoring and identification methods, an initial identification model is adopted, comprising a neck network module including a pooling layer, a self-attention mechanism layer, and a stitching layer. The neck network module can obtain image enhancement feature maps, thereby improving the meter identification accuracy of the initial identification model. Specifically, training samples consisting of sample meter images and their corresponding annotation information are input into the initial identification model for training. During training, model parameters are adjusted until the model converges, resulting in the meter monitoring and identification model. Then, the image of the meter to be monitored is input into the meter monitoring and identification model to obtain the meter identification result. The meter identification result is transmitted to a server for display using the Wireless Local Area Network Authentication and Privacy Infrastructure (WAPI) protocol. The WAPI protocol has an encryption function, and using WAPI for data transmission can improve the security of the meter monitoring and identification method.

[0101] Specifically, Figure 1 This is a schematic diagram of the architecture of the meter monitoring and identification model provided in the embodiments of this application, as shown below. Figure 1 The meter monitoring and identification model provided in the embodiments of this application is described below:

[0102] The meter monitoring and identification model includes an input module, a backbone network module, a neck network module, and a prediction module.

[0103] The input module is responsible for inputting the meter image; the backbone network module is responsible for extracting image features from the input meter image to help identify key information such as the meter's numbers and symbols; the neck network module is responsible for further processing and optimizing the features extracted by the backbone network so that they can be passed to the detection head module; the detection head module is responsible for performing target detection and recognition tasks based on the features processed by the neck network, obtaining the values ​​displayed on the meter, and generating the final meter recognition result.

[0104] The parts not described in detail are disclosed in the following embodiments.

[0105] The technical solution of this application will now be described in detail through specific embodiments. It should be noted that the following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0106] Figure 2 This is a flowchart illustrating the training method for the meter monitoring and identification model provided in the embodiments of this application, as shown below. Figure 2 As shown, the method may include the following steps:

[0107] Step 21: Obtain multiple training samples.

[0108] Each training sample includes a sample meter image and sample annotation information corresponding to the sample meter image.

[0109] In this step, multiple training samples containing sample meter images and corresponding sample annotation information are obtained to provide training data for the subsequent initial recognition model training. This ensures that the model has enough samples to train and learn image features, thereby improving the robustness and accuracy of the model.

[0110] Furthermore, step 21 can be implemented as follows:

[0111] Step 1: Obtain the original meter image.

[0112] This implementation acquires original meter images under different conditions, thereby improving the model's generalization ability.

[0113] Among them, the original meter images under different conditions can be meter images under different shooting angles, different lighting conditions, and different resolution conditions.

[0114] In one possible implementation, raw meter images can be obtained from a historical database, or raw meter image data can be acquired using a camera or other visual acquisition devices.

[0115] Step 2: Perform data augmentation processing on the original meter images to obtain sample meter images.

[0116] In this implementation, due to limitations in data acquisition conditions, the original meter images may have certain limitations or uniformity in image data. Therefore, it is necessary to augment the original meter images to simulate changes in image features under different environmental conditions and obtain sample meter images. This allows the model to see richer sample images and corresponding image features during training, enhancing the robustness of the initial recognition model and enabling it to make more accurate predictions in different real-world environments.

[0117] Data augmentation is a technique to enhance sample diversity. Common augmentation methods include image rotation, horizontal flipping, random cropping, and color adjustment (e.g., adjusting brightness and contrast). This embodiment does not specifically limit the augmentation method used.

[0118] Step 3: Obtain sample annotation information corresponding to the sample meter images through visual detection algorithms.

[0119] In this implementation, a visual detection algorithm is used to obtain the pointer and dial scale features of the sample meter image, and sample annotation information corresponding to the sample meter image is generated based on the features of the pointer and dial scale.

[0120] Optionally, step 3 can be implemented in the following ways:

[0121] S1. The sample meter images are processed using a visual detection algorithm to obtain a binarized image corresponding to the sample meter images.

[0122] In this implementation, the color information in the sample meter image is converted into black and white using a visual detection algorithm to eliminate unnecessary background information and obtain a binarized image corresponding to the sample meter image, making subsequent processing (i.e. pointer separation and recognition) simpler and more efficient.

[0123] Among them, visual detection algorithms for binarized image processing can be adaptive thresholding, global thresholding, local thresholding, etc.

[0124] S2. Perform pointer separation processing on the binarized image to obtain the pointer-separated image.

[0125] In this implementation, a visual detection algorithm is used to perform pointer separation processing on the obtained binarized image, separating the pointer on the meter from the image and generating a pointer-separated image. This separates the pointer from the background, scale, and other information in the meter image, facilitating accurate identification in the future.

[0126] The pointer separation process is performed using techniques such as morphological operations or edge detection. The visual detection algorithms used for pointer separation can be the Hough circle detection algorithm and the contour detection algorithm.

[0127] In one possible implementation, a contour detection algorithm is used for pointer separation. Specifically, the binarized image is first denoised to reduce unnecessary details and noise. Then, edge information is extracted from the binarized image. Based on the edge information, OpenCV is used to extract contours from the image. The extracted contours are filtered by conditions such as area and perimeter to remove contours that do not meet the requirements. Finally, the detected contours are drawn onto the original image, which usually displays the boundaries of the contours, thereby achieving the purpose of pointer separation.

[0128] S3. Recognize the pointer separation image to obtain the pointer recognition result.

[0129] In this implementation, the scale lines on the edge of the dashboard in the binarized image are unfolded into a rectangular strip (that is, a straight line containing all the scale lines). Then, the image recognition algorithm detects the current position of the pointer, calculates the positional relationship between the pointer and the scale lines on the dial, and identifies the specific value or state that the pointer is pointing to, thereby obtaining the pointer recognition result.

[0130] In one possible implementation, the scale lines of the circular dashboard in the binarized image are unfolded into a rectangular band using polar coordinate expansion.

[0131] S4. Determine the pointer recognition result as the sample annotation information corresponding to the sample meter image.

[0132] Under this implementation, the pointer recognition result is analyzed to determine the specific value or state displayed in the meter image. The pointer recognition result is then identified as the sample annotation information corresponding to the sample meter image, providing annotation data support for subsequent initial recognition model training.

[0133] Step 22: For each training sample, extract the image feature map corresponding to the sample meter image through the backbone network module of the initial recognition model.

[0134] In this step, the sample meter images from each training sample are first input into the backbone network module of the initial recognition model for processing, such as... Figure 1As shown, the image features containing both low-level and high-level feature information of the sample meter image are extracted through the Convolutional Layer + Batch Normalization + Mish Activation Function (CBM) module and five Cross Stage Partial Network with X repeated modules (CSPX) modules, resulting in the image feature map corresponding to the sample meter image.

[0135] The aforementioned low-level feature information includes basic visual information (such as edges and corners), while high-level feature information includes more complex and abstract visual information (such as the shape and details of objects).

[0136] The CBM module includes a convolutional layer (Conv), a batch normalization (BN) layer, and a Mish layer. The CSPX module includes five CBM layers, X residual components, and a concatenation layer (Concat).

[0137] Step 23: Enhance the image feature map by using the neck network module of the initial recognition model to obtain the image enhanced feature map.

[0138] In this step, the image feature map corresponding to the sample meter image extracted from the backbone network is input into the neck network module of the initial recognition model for feature depth processing and optimization. By fusing multi-level feature maps and combining the processing of modules such as nonlinear activation functions, feature fusion, and attention mechanisms, the key features (such as pointers) and detailed information (such as dial markings) in the sample meter image are enhanced, thus obtaining an image-enhanced feature map. The image-enhanced feature map is more recognizable than the image feature map corresponding to the sample meter image extracted from the backbone network, contains more accurate visual information, and provides richer feature input for subsequent prediction and analysis tasks.

[0139] Furthermore, the neck network module includes pooling layers, self-attention mechanism layers, and concatenation layers. The neck network module of the initial recognition model enhances the image feature map to obtain an enhanced image feature map, which can be implemented in the following ways:

[0140] Among them, such as Figure 1As shown, the pooling layer, the self-attention mechanism layer, and the splicing layer together form the improved Spatial Pyramid Pooling (SPP) module. The embodiments of this application adopt the SPP* module, which has a similar structure to the SPP module.

[0141] Step 1: Input the image feature map into the pooling layer to obtain a multi-scale feature map corresponding to the image feature map.

[0142] In this implementation, the image feature map corresponding to the sample meter image extracted from the backbone network is input into the pooling layer in the neck network module to perform spatial dimensionality reduction on the image feature map and obtain the multi-scale feature map corresponding to the image feature map.

[0143] Among them, multi-scale feature maps can contain receptive fields of different sizes, which can better capture detailed information in the image, especially when the target size, position and shape vary greatly, and can provide more layers of feature information for subsequent feature enhancement.

[0144] In one possible implementation, the multi-scale feature maps include 5×5 scale feature maps, 9×9 scale feature maps, and 13×13 scale feature maps.

[0145] Step 2: Input the image feature map and the corresponding multi-scale feature map into the self-attention mechanism layer respectively, so as to adjust the channel weights of the image feature map and the multi-scale feature map respectively.

[0146] In this implementation, the image feature map and the aforementioned multi-scale feature map are respectively input into the self-attention mechanism layer. The correlation between the various parts of the input feature map is calculated, and the weight of each channel in the multiple input feature maps is automatically adjusted. The initial recognition model can assign greater weight to more important regions and channels in different spatial regions and channels, thereby improving the initial recognition model's attention to important features. This allows key regions in the sample meter image (such as pointers, dial edges, etc.) to be represented more prominently, while reducing the influence of irrelevant background information.

[0147] In one possible implementation, such as Figure 1 As shown, the image feature map, 5×5 scale feature map, 9×9 scale feature map and 13×13 scale feature map are input into the self-attention mechanism layer.

[0148] Step 3: Input the image feature map after adjusting the channel weights and the corresponding multi-scale feature map into the stitching layer for fusion processing to obtain the image enhancement feature map.

[0149] In this implementation, after adjusting the channel weights of the image feature map and the multi-scale feature map using the self-attention mechanism, the adjusted feature map is passed to the stitching layer for processing to obtain the image enhancement feature map.

[0150] The specific processing of the stitching layer is to fuse the image feature map and the multi-scale feature map together to form an image enhancement feature map that integrates feature maps of multiple scales and key features. While ensuring that features of different scales are effectively fused, it retains the detailed features and semantic information at different levels, and finally obtains an image enhancement feature map with rich feature information, thereby improving the accuracy of the initial recognition model.

[0151] Step 24: Process the image enhancement feature map through the detection head module of the initial recognition model to obtain the prediction annotation information;

[0152] In this step, the image enhancement feature map is input into the detection head module of the initial recognition model for processing. The numerical display area of ​​the sample meter image is identified, the relative position of the pointer and scale line on the meter in the sample meter image is accurately located, the meter pointer reading in the sample meter image is identified, and finally the predicted annotation information of each meter in the sample meter image is generated. This provides important reference data for the subsequent initial recognition model training, optimization and evaluation.

[0153] The predicted labeling information includes meter readings and meter type (e.g., electricity meter, SF6 meter, etc.).

[0154] The detection head module includes a CBL module and a Conv layer for identifying large meter targets in the marker recognition image, a CBL module and a Conv layer for identifying medium meter targets in the marker recognition image, and a CBL module and a Conv layer for identifying small meter targets in the marker recognition image.

[0155] Step 25: Based on the sample annotation information, prediction annotation information and preset loss function of the sample meter images, train the initial recognition model to obtain the meter monitoring recognition model.

[0156] In this step, the initial recognition model uses a preset loss function to calculate the difference between the predicted annotation information corresponding to the generated sample meter image and the sample annotation information corresponding to the sample meter image. The preset loss function calculates a loss value for each prediction result, reflecting the performance of the initial recognition model. Then, the initial recognition model transmits the loss value back to the initial recognition model through the backpropagation algorithm, and uses an optimization algorithm to adjust the parameters of the initial recognition model. After multiple iterations, the parameters of the model are gradually adjusted and optimized until the model converges to obtain the meter monitoring recognition model.

[0157] Furthermore, step 25 can be implemented as follows:

[0158] Step 1: Calculate the loss value between sample annotation information and predicted annotation information based on the preset loss function.

[0159] In this implementation, the initial recognition model compares the predicted annotation information with the sample annotation information, and uses a preset loss function to calculate the loss value between the predicted annotation information and the sample annotation information for each sample meter image, thereby quantifying the error of the initial recognition model's prediction.

[0160] In one possible implementation, the default loss function is the Generalized Intersection over Union (GIoU) ​​loss function. By introducing a minimum closed region, the failure of the Intersection over Union (IoU) in non-overlapping regions is solved, providing a more robust gradient signal and position sensitivity, thereby improving the accuracy of the initial recognition model.

[0161] Step 2: Adjust the parameters of the initial identification model according to the loss value until the preset convergence condition is met to obtain the meter monitoring identification model.

[0162] In this implementation, the initial identification model adjusts the model parameters based on the calculated loss value, and then the loss value is propagated back to the initial identification model through the backpropagation algorithm, thereby updating the network weights and biases. After multiple backpropagation processes, the preset convergence condition is reached, and the initial identification model corresponding to the preset convergence condition is used as the meter monitoring identification model.

[0163] During model training, the ReduceLearningRateOnPlateau (ReduceLROnPlateau) method in TensorFlow is used to control the learning rate, which gradually decreases the learning rate during training and helps the initial recognition model converge.

[0164] The preset convergence condition can be that the change in the loss value is small enough or that the maximum number of iterations is reached.

[0165] In one possible implementation, the loss value is 0.1 and the maximum number of iterations is 100.

[0166] This application provides a training method for a meter monitoring and identification model. The method first acquires multiple training samples; each training sample includes a sample meter image and corresponding sample annotation information. For each training sample, the backbone network module of the initial identification model extracts the image feature map corresponding to the sample meter image. Then, the neck network module of the initial identification model enhances the image feature map to obtain an enhanced image feature map. The detection head module of the initial identification model processes the enhanced image feature map to obtain predicted annotation information. Finally, based on the sample annotation information, predicted annotation information, and a preset loss function, the initial identification model is trained to obtain the meter monitoring and identification model. This technical solution, by training the initial identification model, enables the meter monitoring and identification model to have high recognition accuracy. The obtained meter monitoring and identification model can be used for meter identification, enabling efficient meter monitoring and status recognition.

[0167] Based on the above embodiments, Figure 3 Flowchart of the meter monitoring and identification method provided in the embodiments of this application Figure 1 ,like Figure 3 As shown, the method may include the following steps:

[0168] Step 31: Obtain the image of the meter to be monitored.

[0169] In this step, an image acquisition device is used to acquire the video stream of the meter to be monitored, and then the image frames in the video stream are extracted to obtain the image of the meter to be monitored.

[0170] Furthermore, step 31 can be implemented as follows:

[0171] Step 1: Set the camera parameters of the image acquisition device according to the real-time environmental data.

[0172] Under this implementation, the parameters of the image acquisition device (such as a camera) are optimized according to different factors in the real-time environment (such as lighting, background noise, meter angle, etc.).

[0173] Among them, the camera parameters of the image acquisition device can be adjusted as follows: if the current ambient light is dark, the exposure time and gain of the camera can be adjusted to ensure that the brightness of the image of the meter to be monitored is moderate and the details are clear. If there is reflection or dirt on the surface of the meter to be monitored, the focus and contrast of the camera can be adjusted to reduce the influence of these interference factors. Through precise camera parameter settings, it can be ensured that the quality of the acquired image meets the needs of the subsequent meter monitoring and identification model.

[0174] Step 2: Based on the camera parameters, capture the image of the meter to be monitored for identification.

[0175] In this implementation, based on the camera parameters of the image acquisition device, the image acquisition device captures images of the meter to be monitored, while ensuring that all important information of the meter (such as digital display, pointer position, etc.) can be clearly displayed in the image of the meter to be monitored.

[0176] The image acquisition device can monitor and acquire images of only one meter, or it can monitor images of multiple meters simultaneously. That is, the image to be identified of the meter to be monitored must include at least one meter.

[0177] In addition, the camera parameters of the image acquisition device can be adjusted on the meter recognition terminal. The image acquisition device stores the acquired image to be recognized in the meter recognition terminal, and then the meter recognition terminal sends the image to be recognized to the meter monitoring and recognition model configured in the meter recognition terminal for processing.

[0178] Step 32: Input the image to be identified into the meter monitoring and identification model for processing to obtain the meter identification result.

[0179] Among them, the meter monitoring and identification model is based on Figure 1 The meter monitoring and identification model in the corresponding embodiment is pre-trained.

[0180] In this step, the image to be identified is input into the meter monitoring and recognition model for image preprocessing (such as noise reduction, contrast enhancement, etc.) to ensure that the meter monitoring and recognition model can extract clear features from the image to be identified. Then, the meter monitoring and recognition model will automatically obtain the relative position of the meter pointer and the dial scale lines according to the features of the image to be identified, thereby obtaining the meter pointer reading and finally obtaining the meter recognition result.

[0181] The meter monitoring and identification model is configured on the meter identification terminal, which can be a computer or a microcomputer control chip.

[0182] In one possible implementation, the meter identification result includes the current value of the meter, the confidence score of the identification, and the meter type.

[0183] Step 33: Send the meter reading results to the server.

[0184] In this step, after the meter monitoring and identification model configured in the meter identification terminal obtains the meter identification result, the meter identification terminal sends the meter identification result to the server. After receiving the identification result, the server can further perform data analysis, storage, alarm, or report generation operations.

[0185] If the server determines that the meter identification result indicates that a certain meter has an abnormal value, the server can immediately issue an alarm and notify relevant personnel to handle it. If the identification results of multiple meters need to be summarized, the server can also perform trend analysis based on these meter data, generate charts or statistical information, and assist in subsequent decision-making and optimization work.

[0186] In one possible implementation, the server includes a remote management platform that can display meter readings.

[0187] Furthermore, step 33 can be implemented as follows:

[0188] Based on the WAPI protocol, the meter reading results are encrypted and then sent to the server.

[0189] In this implementation, the meter identification terminal encrypts the meter identification results based on the WAPI protocol, then encapsulates the encrypted data using the WAPI protocol to ensure data integrity and security, and finally sends the encrypted meter identification results to the server using the WAPI protocol.

[0190] After receiving the encrypted meter identification result, the server parses the data using the WAPI protocol and decrypts it using the corresponding key to restore the meter identification result before encryption. The server then performs subsequent analysis and processing based on the decrypted meter identification result. Using the WAPI protocol to encrypt and transmit the meter identification result can prevent the meter identification result from being stolen or tampered with during transmission, ensuring the security and integrity of the meter identification result during transmission.

[0191] This application provides a meter monitoring and identification method. The method first acquires an image of the meter to be monitored for identification, then inputs the image into a meter monitoring and identification model for processing to obtain the meter identification result, and finally sends the meter identification result to a server. This technical solution improves the accuracy of meter monitoring and identification by acquiring an image of the meter to be monitored, inputting the image into a meter monitoring and identification model for precise analysis, obtaining accurate meter identification results, and sending the identification results to a server.

[0192] Based on the above embodiments, Figure 4 Flowchart of the meter monitoring and identification method provided in the embodiments of this application Figure 2 ,like Figure 4 As shown, the method may include the following steps:

[0193] Step 41: Receive the meter identification results of the meter to be monitored sent by the meter identification terminal.

[0194] The meter identification result is obtained by the meter identification terminal from the image of the meter to be monitored, and the image is identified by the meter monitoring and identification model. The meter identification result is then encrypted and sent based on the WAPI protocol. The meter monitoring and identification model is pre-trained according to the training method of the above-mentioned meter monitoring and identification model.

[0195] In this step, the server receives the meter identification results of the meter to be monitored, which are encrypted using the WAPI protocol, from the meter identification terminal. Then, it verifies the integrity of the meter identification result data to ensure that the meter identification result data has not been tampered with during transmission.

[0196] Integrity verification includes checking whether the meter identification result data packet is complete, conforms to the WAPI protocol format, and contains a valid check code or digital signature (i.e., the authentication of the meter identification terminal).

[0197] In one possible implementation, if integrity verification fails, the server will refuse to process the meter identification result data packet, ensuring that invalid or potentially tampered meter identification result data is not accepted, thereby ensuring the security of meter identification result data transmission.

[0198] Step 42: Decrypt and display the meter identification results based on the WAPI protocol.

[0199] In this step, the server uses the shared key in the WAPI protocol to decrypt the encrypted meter identification result to obtain the unencrypted meter identification result. Then, the unencrypted meter identification result is parsed and stored in the server's database for subsequent querying, analysis, and report generation.

[0200] The server can be a high-efficiency device management platform that can perform real-time monitoring and precise data analysis.

[0201] This application provides a meter monitoring and identification method. The method first receives the meter identification result of the meter to be monitored from a meter identification terminal. The meter identification result is obtained by the meter identification terminal from the image of the meter to be monitored, and then identified using a pre-trained meter monitoring and identification model. The identification result is encrypted using the WAPI protocol before being sent, and then decrypted and displayed using the same WAPI protocol. This technical solution obtains the image of the meter to be monitored by the meter identification terminal and processes it using a pre-trained identification model to obtain accurate meter identification results. The identification results are encrypted using the WAPI protocol before being sent, and the receiver decrypts the data using the same protocol, ensuring the security of the meter identification result data transmission.

[0202] In one possible implementation, Figure 5 This is a system architecture diagram of the training method for the meter monitoring and identification model provided in the embodiments of this application. The architecture includes a server, a meter identification terminal, an image acquisition device, meters, and a WAPI access point.

[0203] The server receives data from the meter identification terminal, analyzes, processes, and displays it, and is responsible for the centralized management and access of the data.

[0204] The meter identification terminal is used to control the image acquisition equipment to acquire meter images, and to process the meter images through the meter monitoring model to obtain the meter identification results.

[0205] A meter is a hardware device used to actually measure and record data.

[0206] The WAPI access point is responsible for providing wireless network connectivity for meter reading terminals and servers, ensuring real-time encrypted data transmission and guaranteeing data security. Figure 5 The area marked by the dashed line represents the coverage area of ​​the WAPI wireless network.

[0207] Figure 6 Flowchart of the meter monitoring and identification method provided in the embodiments of this application Figure 3 ,like Figure 6 As shown, the method may include the following steps:

[0208] Step 1, Begin.

[0209] Step 2: Use a meter reading terminal to adjust the camera parameters.

[0210] Step 3: Use a camera to capture an image of the meter to be identified and send it to the meter identification terminal.

[0211] Step 4: The meter identification terminal sends the image of the meter to be identified to the meter monitoring and identification model for processing to obtain the meter identification result.

[0212] Step 5: The meter identification terminal uses the WAPI protocol to encrypt the meter identification results to obtain the encrypted meter identification results.

[0213] Step 6: The meter identification terminal sends the encrypted meter identification result to the remote management platform (i.e., the server) using the WAPI protocol.

[0214] Step 7: The remote management platform decrypts the encrypted meter identification results to obtain the unencrypted meter identification results.

[0215] Step 8: The remote management platform analyzes the meter identification results and displays them on the page, while also storing the meter identification results.

[0216] Step 9, End.

[0217] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0218] Figure 7 This is a schematic diagram of the structure of the training device for the meter monitoring and identification model provided in an embodiment of this application. Figure 7 As shown, the device includes:

[0219] The first acquisition module 71 is used to acquire multiple training samples; wherein each training sample includes a sample meter image and sample annotation information corresponding to the sample meter image;

[0220] The first processing module 72 is used to extract the image feature map corresponding to the sample meter image through the backbone network module of the initial recognition model for each training sample.

[0221] The second processing module 73 is used to enhance the image feature map through the neck network module of the initial recognition model to obtain an enhanced image feature map.

[0222] The third processing module 74 is used to process the image enhancement feature map through the detection head module of the initial recognition model to obtain prediction annotation information;

[0223] The determination module 75 is used to train the initial recognition model based on the sample annotation information, prediction annotation information and preset loss function of the sample meter image to obtain the meter monitoring recognition model.

[0224] In one possible implementation, the neck network module includes a pooling layer, a self-attention mechanism layer, and a splicing layer;

[0225] The second processing module 73 is specifically used for:

[0226] The image feature map is input into the pooling layer to obtain a multi-scale feature map corresponding to the image feature map;

[0227] The image feature map and the corresponding multi-scale feature map are respectively input into the self-attention mechanism layer to adjust the channel weights of the image feature map and the multi-scale feature map respectively;

[0228] The image feature map with adjusted channel weights and the corresponding multi-scale feature map are input into the stitching layer for fusion processing to obtain the image enhancement feature map.

[0229] In one possible implementation, the determining module 75 is specifically used for:

[0230] Based on a preset loss function, calculate the loss value between sample annotation information and predicted annotation information;

[0231] Adjust the parameters of the initial identification model based on the loss value until the preset convergence condition is met to obtain the meter monitoring identification model.

[0232] In one possible implementation, the first acquisition module 71 is specifically used for:

[0233] Obtain the raw meter image;

[0234] The original meter images are augmented to obtain sample meter images;

[0235] The visual detection algorithm is used to obtain the sample annotation information corresponding to the sample meter image.

[0236] In one possible implementation, the first acquisition module 71 acquires sample annotation information corresponding to the sample meter image through a visual detection algorithm, specifically for:

[0237] The sample meter images are processed using a visual detection algorithm to obtain a binarized image corresponding to the sample meter images;

[0238] Pointer separation processing is performed on the binarized image to obtain a pointer-separated image;

[0239] The pointer separation image is identified to obtain the pointer recognition result;

[0240] The pointer recognition result is used to determine the sample annotation information corresponding to the sample meter image.

[0241] Figure 8 Schematic diagram of the structure of the meter monitoring and identification device provided in the embodiments of this application Figure 1 .like Figure 8 As shown, the main executing component of this device is a meter identification terminal, which includes:

[0242] The second acquisition module 81 is used to acquire the image to be identified of the meter to be monitored;

[0243] The fourth processing module 82 is used to input the image to be identified into the meter monitoring and identification model for processing to obtain the meter identification result. The meter monitoring and identification model is pre-trained according to or any of the meter monitoring and identification model training methods.

[0244] The sending module 83 is used to send the meter reading results to the server.

[0245] In one possible implementation, the second acquisition module 81 is specifically used for:

[0246] Set the camera parameters of the image acquisition device according to real-time environmental data;

[0247] Based on camera parameters, capture images of the meters to be monitored for identification.

[0248] In one possible implementation, the sending module 83 is specifically used for:

[0249] Based on the WAPI protocol, a wireless LAN authentication and confidentiality infrastructure, the meter reading results are encrypted and then sent to the server.

[0250] Figure 9 Schematic diagram of the structure of the meter monitoring and identification device provided in the embodiments of this application Figure 2 .like Figure 9 As shown, the main body of this device is a server, including:

[0251] The receiving module 91 is used to receive the meter identification result of the meter to be monitored sent by the meter identification terminal; wherein, the meter identification result is that the meter identification terminal obtains the image of the meter to be monitored, identifies the image of the meter to be monitored through the meter monitoring and identification model, and encrypts the meter identification result based on the Wireless Local Area Network Authentication and Confidentiality Infrastructure (WAPI) protocol before sending it; the meter monitoring and identification model is pre-trained according to the training method of the meter monitoring and identification model in any of the above embodiments;

[0252] The fifth processing module 92 is used to decrypt and display the meter identification results based on the WAPI protocol.

[0253] The apparatus provided in this application embodiment can be used to execute the determination method in any of the above embodiments. Its implementation principle and technical effect are similar, and will not be described again here.

[0254] It should be noted that the division of the various modules in the above device is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, these modules can be implemented entirely in software via processing element calls; they can be fully implemented in hardware; or some modules can be implemented in software via processing element calls, while others are implemented in hardware. Additionally, these modules can be fully or partially integrated together, or implemented independently. The processing element here can be an integrated circuit with signal processing capabilities. During implementation, each step of the above method or each of the above modules can be completed through the integrated logic circuits in the hardware of the processor element or through software instructions.

[0255] Figure 10 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application, such as... Figure 10 As shown, the electronic device may include: a processor 101, a memory 102, and computer program instructions stored in the memory 102 and executable on the processor 101. When the processor 101 executes the computer program instructions, it implements the method provided in any of the foregoing embodiments.

[0256] Optionally, the various components of the electronic device can be connected via a system bus.

[0257] The memory 102 can be a separate storage unit or a storage unit integrated into the processor 101. The number of processors 101 can be one or more.

[0258] It should be understood that processor 101 can be a Central Processing Unit (CPU), or other general-purpose processor 101, digital signal processor 101 (DSP), application-specific integrated circuit (ASIC), etc. General-purpose processor 101 can be a microprocessor 101, or any conventional processor 101. The steps of the method disclosed in this application can be directly manifested as execution by the hardware processor 101, or execution by a combination of hardware and software modules in processor 101.

[0259] The system bus can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The system bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the figure, but this does not indicate that there is only one bus or one type of bus. Memory 102 may include random access memory (RAM) 102, and may also include non-volatile memory (NVM) 102, such as at least one disk storage device 102.

[0260] All or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a readable memory 102. When the program is executed, it performs the steps of the above-described method embodiments; and the aforementioned memory 102 (storage medium) includes: read-only memory 102 (ROM), RAM, flash memory 102, hard disk, solid-state hard disk, magnetic tape, floppy disk, optical disk, and any combination thereof.

[0261] The electronic device provided in this application embodiment can be used to execute the method provided in any of the above method embodiments. Its implementation principle and technical effect are similar, and will not be repeated here.

[0262] This application provides a computer-readable storage medium storing computer instructions that, when executed on a computer, cause the computer to perform the above-described method.

[0263] The aforementioned computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory, electrically erasable programmable read-only memory, erasable programmable read-only memory, programmable read-only memory, read-only memory, magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0264] Optionally, a readable storage medium can be coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Alternatively, the readable storage medium can be an integral part of the processor. Both the processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components within the device.

[0265] This application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium, and the at least one processor can implement the above-described method when executing the computer program.

[0266] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A training method for a meter monitoring and identification model, characterized in that, The method includes: Multiple training samples are acquired; wherein each training sample includes a sample meter image and sample annotation information corresponding to the sample meter image; For each training sample, the image feature map corresponding to the sample meter image is extracted through the backbone network module of the initial recognition model; The image feature map is enhanced by using the neck network module of the initial recognition model to obtain an enhanced image feature map; The image enhancement feature map is processed by the detection head module of the initial recognition model to obtain prediction annotation information; Based on the sample annotation information, the predicted annotation information, and the preset loss function of the sample meter images, the initial recognition model is trained to obtain the meter monitoring recognition model.

2. The method according to claim 1, characterized in that, The neck network module includes a pooling layer, a self-attention mechanism layer, and a splicing layer; The image feature map is enhanced by the neck network module of the initial recognition model to obtain an enhanced image feature map, including: The image feature map is input into the pooling layer to obtain a multi-scale feature map corresponding to the image feature map; The image feature map and the corresponding multi-scale feature map are respectively input into the self-attention mechanism layer to adjust the channel weights of the image feature map and the multi-scale feature map respectively; The image feature map with adjusted channel weights and the corresponding multi-scale feature map are input into the stitching layer for fusion processing to obtain the image enhancement feature map.

3. The method according to claim 1, characterized in that, The process of training the initial identification model based on the sample annotation information, the predicted annotation information, and a preset loss function to obtain the meter monitoring identification model includes: Based on the preset loss function, calculate the loss value between the sample annotation information and the predicted annotation information; The parameters of the initial identification model are adjusted according to the loss value until the preset convergence condition is met, thereby obtaining the meter monitoring identification model.

4. The method according to claim 1, characterized in that, The acquisition of multiple training samples includes: Obtain the raw meter image; The original meter image is subjected to data augmentation processing to obtain the sample meter image; The sample labeling information corresponding to the sample meter image is obtained through a visual detection algorithm.

5. The method according to claim 4, characterized in that, The step of obtaining sample annotation information corresponding to the sample meter image through a visual detection algorithm includes: The sample meter image is processed by the visual detection algorithm to obtain a binarized image corresponding to the sample meter image; The pointer-separated image is obtained by performing pointer separation processing on the binarized image; The pointer separation image is then identified to obtain the pointer identification result; The pointer recognition result is determined as the sample annotation information corresponding to the sample meter image.

6. A meter monitoring and identification method, characterized in that, The method includes: Acquire the image of the meter to be monitored for identification; The image to be identified is input into the meter monitoring and identification model for processing to obtain the meter identification result. The meter monitoring and identification model is pre-trained by the training method of the meter monitoring and identification model according to any one of claims 1 to 5. The meter readings are sent to the server.

7. The method according to claim 6, characterized in that, The process of acquiring the image to be identified of the meter to be monitored includes: Set the camera parameters of the image acquisition device according to real-time environmental data; Based on the camera parameters, an image of the meter to be monitored is captured for identification.

8. The method according to claim 6, characterized in that, Sending the meter reading result to the server includes: The meter identification result is encrypted and sent to the server based on the WAPI protocol, a wireless LAN authentication and confidentiality infrastructure.

9. A meter monitoring and identification method, characterized in that, The method includes: The device receives the meter identification result of the meter to be monitored sent by the meter identification terminal; wherein the meter identification result is the meter identification terminal acquiring the image to be identified of the meter to be monitored, identifying the image to be identified through the meter monitoring and identification model, and encrypting the meter identification result based on the Wireless Local Area Network Authentication and Privacy Infrastructure (WAPI) protocol before sending it; the meter monitoring and identification model is pre-trained by the meter monitoring and identification model training method according to any one of claims 1 to 5. The meter identification result is decrypted and displayed based on the WAPI protocol.

10. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 1 to 5.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1 to 5.

Images