A method, device, and electronic equipment for reading optical character meters in substations.

Abstract

This invention belongs to the field of image detection technology and provides a method, device, and electronic device for reading optical character meters in substations. The method includes: acquiring an image of an optical character meter; processing the optical character meter image using a target detection model to obtain the meter type corresponding to the optical character meter image and the dial area image in the optical character meter image; extracting character features of the dial area image using a feature extraction network of a reading recognition model; obtaining historical character features corresponding to the meter type; fusing the character features and historical character features using a historical information fusion convolution module of the reading recognition model to obtain fused character features; and parsing the fused character features to obtain the reading corresponding to the optical character meter image. This invention also provides a device for reading optical character meters in substations, a computer program product, an electronic device, and a substation inspection system. This invention improves reading reliability and can be adapted to different types of meters.

Description

Technical Field

[0001] This invention relates to the field of image detection technology, and in particular to a method, apparatus, and electronic device for reading optical character meters in substations. Background Technology

[0002] As a crucial node in the power system, substations undertake the core tasks of power transmission and distribution, making real-time monitoring and management of their internal equipment essential. To record and analyze power operation data, substations are widely equipped with metering devices, including electricity meters, power factor meters, and voltmeters. These devices primarily use optical character meters to display key operating parameters. However, traditional meter reading methods rely mainly on manual recording, which is not only time-consuming and labor-intensive but also susceptible to human error, leading to inaccurate or incomplete data recordings. Furthermore, the working environment of substations is typically complex, potentially facing harsh conditions such as high temperature, high humidity, and electromagnetic interference, further increasing the difficulty and risk of manual reading.

[0003] In recent years, with the rapid development of image processing technology and artificial intelligence, automatic meter reading methods based on optical character recognition (OCR) have gradually become a research and application hotspot. This technology, which uses cameras to capture meter images and employs computer vision algorithms to extract and recognize character information from the images, greatly reduces manual intervention and improves the accuracy and efficiency of data recording. However, existing optical character recognition (OCR) technologies still face multiple technical bottlenecks in practical applications. For example, under adverse conditions such as complex background light interference, blurred character displays, tilted dials, or reflections, the accuracy of reading recognition will significantly decrease. Furthermore, meter characters may not be fully rotated due to mechanical limitations, resulting in blurred character shapes, which further increases the difficulty of recognition. Besides the difficulty in recognition accuracy, substations also contain various types of optical character meters, such as leakage current meters with action counts, electronic meters, action count meters, electronic meters in pressure plate cabinets, and oil temperature gauges. How to achieve compatibility with various types of optical character meters and ensure the universality of the reading method under different types of optical character meter structures is also a problem that existing technologies urgently need to solve. Summary of the Invention

[0004] This application aims to at least solve the technical problems existing in the prior art, and to provide a method, device and electronic device for reading optical character meters in substations.

[0005] In a first aspect, this application provides a method for reading optical character meters in a substation. The method includes: acquiring an optical character meter image; processing the optical character meter image using a target detection model to obtain the meter type corresponding to the optical character meter image and the dial area image in the optical character meter image; extracting character features of the dial area image using a feature extraction network of a reading recognition model; acquiring historical character features corresponding to the meter type, wherein the historical character features are obtained by processing historical dial area images corresponding to the meter type using a pre-trained image encoder; fusing the character features and historical character features using a historical information fusion convolution module of the reading recognition model to obtain fused character features; and parsing the fused character features to obtain the reading corresponding to the optical character meter image.

[0006] Secondly, this application provides a substation optical character meter reading device for implementing the substation optical character meter reading method provided in the first aspect of this application. The device includes: an image acquisition module for acquiring an optical character meter image; a target detection module for processing the optical character meter image using a target detection model to obtain the meter type corresponding to the optical character meter image and the dial area image in the optical character meter image; a character feature extraction module for extracting character features of the dial area image using a feature extraction network of a reading recognition model; a historical character feature acquisition module for acquiring historical character features corresponding to the meter type, wherein the historical character features are obtained by processing the historical dial area image corresponding to the meter type using a pre-trained image encoder; a fusion processing module for fusing character features and historical character features using a fusion convolution module based on historical information from the reading recognition model to obtain fused character features; and a parsing module for parsing the fused character features to obtain the reading corresponding to the optical character meter image.

[0007] Thirdly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of a substation optical character meter reading method provided in the first aspect of this application.

[0008] Fourthly, this application provides an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to execute a method for reading optical character meters in a substation provided in the first aspect of this application.

[0009] Fifthly, this application provides a substation inspection system, comprising: one or more optical character meters deployed in the substation; one or more cameras for acquiring images of the optical character meters; and a processing center for acquiring the optical character meter images from the cameras and executing a substation optical character meter reading method provided in the first aspect of this application to obtain readings corresponding to the optical character meter images.

[0010] The beneficial technical effects of this application are as follows: In the process of acquiring the reading corresponding to the optical character meter image, the target detection model is used to obtain the meter type and dial area image. The character features of the dial area image are extracted through the feature extraction network of the pre-trained reading recognition model. The corresponding historical character features are obtained according to the meter type. The historical information fusion convolution module of the reading recognition model is used to fuse the character features and historical character features to obtain fused character features. The fused character features are analyzed to obtain the reading corresponding to the optical character meter image. This reduces the impact of background light interference, character display blurring, dial tilt or reflection, etc., that may exist in the optical character meter image due to the complex working environment of the substation, on the reading accuracy and improves the reliability of the reading. In addition, the corresponding historical character features are obtained according to the meter type corresponding to the optical character meter image, and the information can be automatically filled in according to the meter type. This can adapt to different types of meters. Different types of meters can share the target detection model, image encoder and reading recognition model, which has universality and greatly improves the operating efficiency of the substation. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating a preferred embodiment of the method for reading optical character meters in a substation according to the present invention.

[0012] Figure 2 This is a schematic diagram of the process of obtaining readings in a preferred embodiment of the present invention;

[0013] Figure 3 This is a schematic diagram of the training process of an image encoder in a preferred embodiment of the present invention;

[0014] Figure 4 This is a flowchart of reading recognition in one embodiment of the present invention;

[0015] Figure 5 This is a schematic diagram of the structure of an electronic device in a preferred embodiment of the present invention. Detailed Implementation

[0016] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0017] In the description of this invention, it should be understood that the terms "longitudinal", "lateral", "up", "down", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0018] In the description of this invention, unless otherwise specified and limited, it should be noted that the terms "installation", "connection" and "linking" should be interpreted broadly. For example, they can refer to mechanical or electrical connections, or internal connections between two components. They can be direct connections or indirect connections through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms according to the specific circumstances.

[0019] The substation optical character meter reading method provided by this invention can be executed by at least one of the following electronic devices: a server, a terminal, or any other electronic device that can be configured to execute the substation optical character meter reading method provided in this application. In other words, the substation optical character meter reading method can be executed by software or hardware installed on a terminal device or a server device. The software can be a blockchain platform. Servers include, but are not limited to: a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0020] This invention provides a method for reading optical character meters in substations. In a preferred embodiment, please see... Figure 1 and Figure 2 The method includes:

[0021] Step S1: Obtain the image of the optical character meter.

[0022] In this embodiment, the optical character meter image is an image of an optical character meter in a substation, captured by a camera. Optical character meters are used to measure operating parameters of the substation, such as power, current, and voltage. Optical character meters can be broadly categorized into two types: those that cannot be read using OCR and those that can be read using OCR. OCR-readable meters are further divided into subtypes such as leakage current meters with actuation counts, electronic meters, actuation count meters, electronic meters for pressure plate cabinets, and oil temperature meters. The corresponding readings can be obtained by processing these meter images using optical character recognition (OCR) technology.

[0023] Step S2: Process the optical character meter image using the target detection model to obtain the meter type corresponding to the optical character meter image and the dial area image in the optical character meter image.

[0024] Step S3 involves using the feature extraction network of the reading recognition model to extract character features from the dial area image. Steps S2 and S3 can be executed in parallel or sequentially.

[0025] In this embodiment, the network structure of the object detection model is preferably, but not limited to, existing YOLO series networks, Dino networks, FCOS networks (fully convolutional one-stage object detection networks), Faster RCNN networks (faster region-based convolutional neural networks), or CenterNet networks (center-point-based object detection networks). YOLO is an abbreviation for You Only LookOnce, a real-time object detection network. First, the object detection model network is constructed. Dino is an abbreviation for DIstillation with NO labels. The object detection model network is trained using a pre-constructed object detection sample set to obtain the object detection model. The object detection sample set includes multiple optical character meter images of different meter types from a substation, and object detection labels for each optical character meter image. The object detection labels include the actual meter type of the optical character meter image and the actual detection box location information of the dial area in the image.

[0026] In this embodiment, in step S2, the optical character meter image is input into the target detection model to obtain the target detection result. The target detection result includes the meter type corresponding to the optical character meter image and the detection box position information of the dial area in the optical character meter image. The dial area image is extracted from the optical character meter image based on the detection box position information of the dial area.

[0027] In this embodiment, preferably, in step S2, after obtaining the meter type, it is necessary to exclude types that cannot be read by OCR, as the images of these types of meters cannot be read using OCR technology. Figure 4In the example shown, after the target detection model outputs the meter type, it first determines whether the meter type is one that cannot be read via OCR. If so, the reading process ends; otherwise, it continues with subsequent steps. This filters out meters that do not have OCR reading capabilities.

[0028] Step S4: Obtain the historical character features corresponding to the meter type. The historical character features are obtained by processing the historical dial area image corresponding to the meter type through a pre-trained image encoder.

[0029] Step S5: Use the historical information of the reading recognition model to fuse character features and historical character features using the fusion convolution module to obtain fused character features.

[0030] Step S6: Analyze the fused character features to obtain the reading corresponding to the optical character meter image.

[0031] In this embodiment, to address the problem of reduced text recognition accuracy caused by background light interference and character blurring in optical character meter images, which hinders the utilization of historical character features, this application improves existing text recognition networks and proposes a new reading recognition model network structure. Specifically, improvements are made to existing lightweight OCR models (not limited to PPocrv2, PPocrv3, and PPocrv4) text recognition networks. PPocrv2, PPocrv3, and PPocrv4 are the first, second, and third versions of the practical ultra-lightweight OCR model launched by the Baidu team, respectively.

[0032] All network structures before the last layer in the text recognition network are used as the feature extraction network of the reading recognition model in this application. The last layer of the text recognition network is replaced with the historical information fusion convolutional module of this application. The last layer of the text recognition network may be a fully connected layer or a convolutional layer. Therefore, please see Figure 2 The reading recognition model in this application includes a cascaded feature extraction network and a historical information fusion convolution module.

[0033] In one example, step S4, obtaining the historical character features corresponding to the meter type, includes: pre-storing the historical character features corresponding to each meter type in the database, and querying the corresponding historical character features from the database according to the obtained meter type.

[0034] In another example, step S4, obtaining the historical character features corresponding to the meter type, includes: pre-storing historical dial area images corresponding to each meter type in the database, querying the corresponding historical dial area images from the database according to the obtained meter type, and processing the historical dial area images using a pre-trained image encoder to obtain the corresponding historical character features.

[0035] In another example, to better match the acquired historical character features with the meters in the optical character meter images and improve reading accuracy, preferably, each optical character meter in the substation is assigned a patrol point ID. While the camera is acquiring the optical character meter image, it simultaneously acquires the patrol point ID of that optical character meter. Based on the patrol point ID, the database is queried to find one or more historical meter area images or one or more historical character features corresponding to that patrol point ID. Then, the required historical meter area image is determined from one or more historical meter area images using the acquired meter type, or the historical character feature is determined from one or more historical character features using the acquired meter type.

[0036] In this embodiment, step S6 is not limited to using the parsing method in existing OCR models to convert the feature map containing fused character features into reading text, such as... Figure 2 Example of the reading text 00668 in the text.

[0037] For example, suppose the dimensions of the fused character features are (b, w, char_num), where b is the training batch size, w is the width of the fused character features, and char_num is the character type. Taking the existing parsing module CTCLabelDecode (Connectivist Temporal Classification Label Decode) as an example, it will obtain the index of the character type with the highest confidence. For example, if the character sequence is [a, b, c, d, e], assuming that d has the highest confidence, then its index is taken as 3. This forms a result with a dimension of (b, w, 1). Then, this result is filtered to remove spaces, thereby obtaining the reading text.

[0038] In this embodiment, preferably, please see Figure 2 As shown, the historical information fusion convolutional module includes:

[0039] A mapping network maps historical character features to historical information modulation coefficients.

[0040] The fusion convolutional layer performs convolution processing on character features based on the injected modulated convolutional weights to obtain fused character features;

[0041] The weight modulation unit modulates the original convolution weights of the fused convolutional layer using historical information modulation coefficients to obtain modulated convolution weights, and then injects the modulated convolution weights into the fused convolutional layer.

[0042] In this embodiment, the mapping network is not limited to including one linear layer or two or more cascaded linear layers. For example, the mapping network includes three cascaded fully connected layers. Let the historical character features be... Historical character features are processed through a mapping network. The historical information modulation coefficients were then obtained. , , This represents a mapping network.

[0043] In this embodiment, the original convolution weights of the fused convolutional layer are the convolution weights obtained after the reading recognition model has been trained. .

[0044] In this embodiment, more preferably, the weight modulation unit obtains the modulated convolution weights of the fused convolutional layer by multiplying the historical information modulation coefficients with the original convolution weights of the fused convolutional layer. After obtaining the modulated convolution weights, the weight modulation unit will... Updated to The modulation and convolution weight injection is completed.

[0045] This invention also discloses an image encoder training method. The training framework for the image encoder is as follows: Figure 3 As shown. In a preferred embodiment, the training process of the image encoder includes:

[0046] Step A: Construct a sample image set using multiple historical dial area images.

[0047] Multiple historical optical character meter images corresponding to various meter types in the substation are collected. A pre-trained object detection model is used to process these images, obtaining a historical dial area image for each image. Each historical dial area image is used as a sample image, and the readings in each image are represented as text labels, thus each sample image has a corresponding label.

[0048] Step B involves constructing an image encoder training framework, which includes an autoencoder and a large-scale contrastive learning model. The autoencoder includes an image encoder to be trained and an image decoder to reconstruct the input image from the output features of the image encoder. The image encoder is not limited to multiple stacked convolutional and pooling layers. The image decoder is primarily used for upsampling and may include multiple layers of transposed convolutions. The large-scale contrastive learning model is not limited to existing CLIP or ALIGN models (natural language supervised scaling models). Preferably, the large-scale contrastive learning model is a Contrastive Language-Image Pretraining model, abbreviated as CLIP model.

[0049] Step C involves iteratively training the autoencoder using the training set until the training stopping condition is met. To prevent the image encoder from only learning the absolute values ​​of historical dial area images, which could lead to performance degradation, this application uses a large contrastive learning model as a supervisory agent. The training stopping condition is that the number of training iterations reaches a preset maximum, or the loss value is less than or equal to a loss threshold. Each training iteration includes the following steps:

[0050] Step C1, Input sample image The autoencoder obtains the restored image corresponding to the sample image. Among them, sample images For any meter type, the historical dial area image is represented by an autoencoder, which includes an image encoder and an image decoder connected in sequence.

[0051] Step C2: Use the contrastive learning image encoder in the contrastive learning large model to acquire sample images respectively. Image features and restored image Image features For example, the CLIP image encoder of the CLIP model is used to acquire sample images respectively. Image features and restored image Image features .

[0052] Step C3: Extract sample images using the contrastive learning text encoder in the contrastive learning large model. The text features of the corresponding character tags. For example, such as... Figure 3 As shown, the CLIP text encoder of the CLIP model is used to extract sample images. The corresponding character label 00667 yields text features. .

[0053] Step C4, based on sample images Image features of sample images Image restoration Image features of the restored image Text features of character tags Calculate the loss value.

[0054] In this embodiment, the image features of the sample image are utilized. Image features of the restored image Calculate the loss value so that the sample image and restored image Aligning the large model feature space in contrastive learning helps improve the accuracy of loss value calculation and enhances the image encoder's extraction accuracy of character features from historical dial area images.

[0055] Step C5: Adjust the network parameters of the autoencoder based on the loss value. Specifically, gradient descent is used to adjust the network parameters of the image encoder and image decoder in the autoencoder based on the loss value.

[0056] In this embodiment, preferably, in step C4, based on the sample image Image features of sample images Image restoration Image features of the restored image Text features of character tags Calculate the loss value, including:

[0057] Based on sample images and restored image Calculate the first loss term .

[0058] Image features based on sample images Image features of the restored image Calculate the second loss term .

[0059] Image features based on restored images Text features of character tags Calculate the third loss term .

[0060] Combined with the first loss item Second loss item and the third loss item Obtain loss value . , , , These represent the first weight, second weight, and third weight, respectively, and their value ranges are: , .

[0061] In this embodiment, through the first loss term The restored image and sample image obtained by the supervised autoencoder are not limited to those obtained by computing the sample image. and restored image The mean absolute error loss is used as the first loss term. ,Right now , This represents the function for calculating the mean absolute error.

[0062] In this embodiment, through the second loss term In contrastive learning of large model feature spaces (such as CLIP feature spaces), image features of sample images are brought closer together. Image features of the restored image The distance is used to align the restored image with the sample image. This is not limited to calculating image features of the sample image. Image features of the restored image The average absolute error loss is used as the second loss term. ,Right now .

[0063] In this embodiment, the third loss item The preferred method is cosine similarity loss. This is achieved through a third loss term. By comparing and learning the feature space of a large model, the image features of the restored image are brought closer together. Text features of character tags The distance between them.

[0064] In this implementation, the contrastive learning large model (such as the CLIP model) is pre-trained on a large-scale dataset, making it more robust to features extracted from the main subject in the image. In the comparison training of the contrastive learning large model (such as the CLIP model), the character label text information can be regarded as a cluster center. By narrowing the distance between text features and image features, the image encoder can be guided to retain the most core character features and ignore the influence of lighting, thereby improving the accuracy of readings.

[0065] This invention also discloses a training method for a reading recognition model. In a preferred embodiment, the training process of the reading recognition model includes:

[0066] Step 1: Construct a reading recognition sample set. Each reading recognition sample includes a dial area image and the corresponding real reading of the dial area image.

[0067] Step 2, construct the network structure of the reading recognition model, such as Figure 2 As shown, the reading recognition model includes a feature extraction network and a historical information fusion convolutional module connected in sequence.

[0068] Step 3: Divide the reading recognition sample set into a training set, a test set, and a validation set in a ratio of 8:1:1.

[0069] Step 4: Iteratively train the network structure of the reading recognition model using the training set until the training stopping condition is met. The training stopping condition is that the number of training iterations has reached a preset maximum, or the cross-entropy loss calculated during training is less than or equal to the loss threshold.

[0070] In each iteration of training: the input sample's dial area image is fed into the feature extraction network of the reading recognition model to obtain the character features of the sample; historical character features corresponding to the meter type to which the sample's dial area image belongs (which can be preset) are obtained. These historical character features can be directly extracted from the database, or historical dial area images corresponding to the meter type to which the sample's dial area image belongs can be obtained, and then processed by the trained image encoder to obtain historical character features; the historical information fusion convolution module of the reading recognition model is used to fuse the character features and historical character features to obtain fused character features; the fused character features are parsed to obtain the reading corresponding to the optical character meter image, which is recorded as the recognized reading; the cross-entropy loss is calculated based on the recognized reading and the actual reading of the sample, and the network parameters of the reading recognition model are optimized using the gradient descent method based on the cross-entropy loss.

[0071] Step 5: Test and validate the trained reading recognition model using the test set and validation set respectively. When the test and validation are passed, output the trained reading recognition model, and use the weights of the fusion convolution in the historical information fusion convolution module of the reading recognition model at this time as the original convolution weights of the fusion convolution.

[0072] The present invention also discloses a substation optical character meter reading device for implementing the above-mentioned substation optical character meter reading method. In a preferred embodiment, the device includes:

[0073] The image acquisition module acquires images of optical character meters;

[0074] The target detection module uses a target detection model to process optical character meter images, and obtains the meter type corresponding to the optical character meter image and the dial area image in the optical character meter image;

[0075] The character feature extraction module uses the feature extraction network of the reading recognition model to extract character features from the dial area image;

[0076] The historical character feature acquisition module acquires the historical character features corresponding to the meter type. The historical character features are obtained by processing the historical dial area image corresponding to the meter type through a pre-trained image encoder.

[0077] The fusion processing module utilizes historical information from the reading recognition model to fuse character features and historical character features into a convolutional module, thereby obtaining fused character features.

[0078] The parsing module analyzes and fuses character features to obtain the readings corresponding to the optical character meter images.

[0079] In this embodiment, the image acquisition module, target detection module, character feature extraction module, historical character feature acquisition module, fusion processing module, and parsing module correspond one-to-one with steps S1, S2, S3, S4, S5, and S6 of the above-described substation optical character meter reading method, and will not be described in detail here.

[0080] The present invention also discloses a substation inspection system, in a preferred embodiment of which the system includes:

[0081] One or more optical character meters are deployed in a substation, and the optical character meters include one or more meter types, for monitoring various operating parameters of the substation.

[0082] One or more cameras are used to capture images from optical character meters. These cameras can be fixedly installed at various inspection points of the optical character meters in the substation using support mechanisms to capture their images. The support mechanisms are not limited to support rods or support frames.

[0083] The processing center acquires optical character meter images from the cameras and executes the aforementioned substation optical character meter reading method to obtain the corresponding readings from the optical character meter images. The processing center is a server, with a network connection device installed at each camera. The server communicates with the cameras via the network connection device, which is not limited to wireless or wired communication devices. Alternatively, the processing center may include multiple edge processors, each located at one camera and communicating with the camera.

[0084] In this embodiment, preferably, it also includes a database that stores historical dial area images or historical character features corresponding to different meter types. The different historical dial area images or historical character features in the database can also correspond to the inspection point IDs of optical character meters.

[0085] Figure 4 This document presents a flowchart illustrating the reading recognition process in one embodiment of the present invention. Specifically, it includes: Upon startup, a substation field camera captures an image of an optical character meter; a target detection model detects the captured optical character meter image to obtain the meter type corresponding to the optical character meter image, determining whether the meter type is a type that cannot be read using OCR. If so, the program exits; otherwise, the dial area image is cropped from the optical character meter image. The dial area image is input into the reading recognition model to obtain character features. The historical character features corresponding to the meter type and the character features are input together into the prediction head (i.e., the historical information fusion convolution module and the parsing module) to obtain the detection result.

[0086] The present invention also discloses a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method for reading optical character meters in substations provided by the present invention. The computer program product should be understood as a software product that primarily implements its solution through a computer program, such as a program product integrated in the cloud or a software library.

[0087] The present invention also discloses an electronic device, in one embodiment of which the electronic device includes at least one processor; and a memory communicatively connected to the at least one processor; wherein,

[0088] The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the substation optical character meter reading method provided by the present invention.

[0089] like Figure 5 The diagram shown is a schematic representation of an electronic device for a substation optical character meter reading method according to an embodiment of the present invention. The electronic device may include a processor 10, a memory 11, a communication bus 12, and a communication interface 13. It may also include a computer program, such as a substation optical character meter reading method program, stored in the memory 11 and executable on the processor 10.

[0090] In some embodiments, the processor 10 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 10 is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the memory 11 (e.g., executing a method for reading optical character meters in a substation), and calls data stored in the memory 11 to perform various functions of the electronic device and process data.

[0091] The memory 11 includes at least one type of readable storage medium, including flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of an electronic device, such as a portable hard drive. In other embodiments, the memory 11 can be an external storage device of the electronic device, such as a plug-in portable hard drive, SmartMediaCard (SMC), SecureDigital (SD) card, FlashCard, etc. Furthermore, the memory 11 can include both internal and external storage units of the electronic device. The memory 11 can be used not only to store application software and various types of data installed on the electronic device, such as the code for a substation optical character meter reading method program, but also to temporarily store data that has been output or will be output.

[0092] The communication bus 12 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. The bus is configured to enable communication between the memory 11 and at least one processor 10, etc.

[0093] Communication interface 13 is used for communication between the aforementioned electronic device and other devices, including a network interface and a user interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a Wi-Fi interface, Bluetooth interface, etc.), typically used to establish communication connections between the electronic device and other electronic devices. The user interface may be a display, an input unit (such as a keyboard), and optionally, a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the electronic device and to display a visual user interface.

[0094] Figure 5 Only electronic devices with components are shown; it will be understood by those skilled in the art that... Figure 5The structure shown does not constitute a limitation on the electronic device and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0095] For example, although not shown, the electronic device may also include a power supply (such as a battery) to power various components. Preferably, the power supply can be logically connected to at least one processor 10 via a power management device, thereby enabling functions such as charging management, discharging management, and power consumption management. The power supply may also include one or more DC or AC power supplies, recharging devices, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0096] It should be understood that the embodiments are for illustrative purposes only and are not limited to this structure in the scope of the patent application.

[0097] Furthermore, if the modules / units integrated into the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, a computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0098] In the description of this specification, the references to terms such as "an embodiment," "some embodiments," "example," "specific example," "a implementation," "a preferred implementation," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0099] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for reading optical character meters in a substation, characterized in that, The method includes: Acquire images of optical character meters; The optical character meter image is processed using a target detection model to obtain the meter type corresponding to the optical character meter image and the dial area image in the optical character meter image; the meter type includes non-OCR readable type and OCR readable type. When the meter type is OCR readable type, the following steps are performed: The feature extraction network of the reading recognition model is used to extract character features from the dial area image; The historical character features corresponding to the meter type are obtained by processing the historical dial area image corresponding to the meter type through a pre-trained image encoder. By utilizing the historical information of the reading recognition model, a convolutional module is used to fuse character features and historical character features to obtain fused character features; Analyze and fuse character features to obtain the readings corresponding to the optical character meter images; The historical information fusion convolution module includes: A mapping network maps historical character features to historical information modulation coefficients. The fusion convolutional layer performs convolution processing on character features based on the injected modulated convolutional weights to obtain fused character features; The weight modulation unit modulates the original convolution weights of the fused convolutional layer using historical information modulation coefficients to obtain modulated convolution weights, and then injects the modulated convolution weights into the fused convolutional layer.

2. The method as described in claim 1, characterized in that, The weight modulation unit obtains the modulated convolution weights of the fused convolutional layer by multiplying the historical information modulation coefficients with the original convolution weights of the fused convolutional layer.

3. The method as described in claim 1 or 2, characterized in that, The training process of the image encoder includes: Input a sample image to the autoencoder to obtain the restored image corresponding to the sample image. The sample image is a historical dial area image corresponding to any meter type. The autoencoder includes an image encoder and an image decoder connected in sequence. The image features of the sample image and the image features of the restored image are obtained by using the contrastive learning image encoder in the contrastive learning large model; The contrastive learning text encoder in the contrastive learning large model is used to extract the text features of the character labels corresponding to the sample images; The loss value is calculated based on the sample image, the image features of the sample image, the restored image, the image features of the restored image, and the text features of the character labels; Adjust the network parameters of the autoencoder based on the loss value.

4. The method as described in claim 3, characterized in that, The calculation of the loss value based on the sample image, the image features of the sample image, the restored image, the image features of the restored image, and the text features of the character labels includes: The first loss term is calculated based on the sample image and the restored image; The second loss term is calculated based on the image features of the sample image and the image features of the restored image; The third loss term is calculated based on the image features of the restored image and the text features of the character labels; The loss value is obtained by combining the first loss term, the second loss term, and the third loss term.

5. The method as described in claim 4, characterized in that, The first and / or second loss terms are the average absolute error loss; And / or, the third loss term is the cosine similarity loss.

6. A reading device for optical character meters in a substation, characterized in that, The apparatus for implementing the method according to any one of claims 1-5 comprises: The image acquisition module acquires images of optical character meters; The target detection module processes optical character meter images using a target detection model to obtain the meter type corresponding to the optical character meter image and the dial area image within the optical character meter image. Meter types include those that cannot be read via OCR and those that can be read via OCR. When the meter type is an OCR-readable type, it also includes: The character feature extraction module uses the feature extraction network of the reading recognition model to extract character features from the dial area image; The historical character feature acquisition module acquires the historical character features corresponding to the meter type. The historical character features are obtained by processing the historical dial area image corresponding to the meter type through a pre-trained image encoder. The fusion processing module utilizes historical information from the reading recognition model to fuse character features and historical character features into a convolutional module, thereby obtaining fused character features. The analysis module analyzes and fuses character features to obtain the readings corresponding to the optical character meter images; The historical information fusion convolution module includes: A mapping network maps historical character features to historical information modulation coefficients. The fusion convolutional layer performs convolution processing on character features based on the injected modulated convolutional weights to obtain fused character features; The weight modulation unit modulates the original convolution weights of the fused convolutional layer using historical information modulation coefficients to obtain modulated convolution weights, and then injects the modulated convolution weights into the fused convolutional layer.

7. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method described in any one of claims 1-5.

8. An electronic device, characterized in that, The electronic device includes: At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the method as described in any one of claims 1 to 5.

9. A substation inspection system, characterized in that, include: One or more optical character meters deployed in a substation; One or more cameras are used to capture optical character meter images; a processing center acquires optical character meter images from the cameras and executes the method of any one of claims 1 to 5 to obtain the readings corresponding to the optical character meter images.

Images