Intelligent identification method for nameplates and related devices

Abstract

This application provides a method and related equipment for intelligent nameplate recognition. The method includes: acquiring image data of a nameplate area; extracting character space features from the image data of the nameplate area; capturing character sequence dependencies based on the character space features to obtain time series features; decoding the time series features to map them into a readable character sequence; performing semantic parsing on the character sequence to obtain structured nameplate data; and repairing the structured nameplate data based on preset standard fields to obtain the final nameplate information. This application's embodiments optimize the image, locate the nameplate area, extract character features and serialize them to generate a readable character sequence. Combined with semantic parsing to extract parameters and repair data, it ensures the completeness and accuracy of the information. It adapts to complex scenarios, improves efficiency by 80%, meets dynamic recognition needs, and provides support for business operations.

Description

Technical Field

[0001] This application relates to the field of nameplate recognition technology, and in particular to a method and related equipment for intelligent nameplate recognition. Background Technology

[0002] With the acceleration of the global clean energy transition, intelligent recognition technology for distributed photovoltaic inverter nameplates has gradually become an important requirement in the new energy industry. Existing technologies mainly fall into two categories: recognition schemes based on traditional OCR and template matching, and intelligent recognition schemes based on deep learning. Traditional OCR technology extracts nameplate information through fixed template matching, suitable for nameplates of highly standardized power equipment, but its adaptability to complex scenarios is poor. Deep learning schemes utilize object detection and sequence recognition technologies to automatically extract nameplate image features and recognize text content, improving recognition capabilities in certain scenarios. However, existing technologies still have significant shortcomings in dealing with the complexity and dynamic changes of distributed photovoltaic inverter nameplates.

[0003] Existing technologies have limitations in scene adaptability. Traditional OCR solutions cannot handle complex shooting scenarios such as backlighting, blurring, and occlusion, while deep learning solutions lack technical support for uneven lighting and repairing worn areas, leading to decreased recognition accuracy. Regarding format compatibility, traditional template matching only works with preset formats. While deep learning solutions offer some flexibility, they cannot automatically parse the semantic relationships of parameters on nameplates from multiple manufacturers, making it difficult to cover non-standardized formats. High model update costs are also a major issue. Adding new samples requires full retraining, which is time-consuming and resource-intensive, and may lead to a decline in existing recognition performance. Furthermore, existing technologies are highly dependent on samples; insufficient sample library coverage and a lack of sample augmentation mechanisms result in insufficient generalization ability of the model in scenarios with incomplete information. Summary of the Invention

[0004] In view of this, the purpose of this application is to propose a method and related equipment for intelligent identification of nameplates.

[0005] To achieve the above objectives, this application provides a method for intelligent identification of nameplates, comprising: Acquire image data of the nameplate area; Extract the character space features from the image data of the nameplate area; Based on the character space features, character sequence dependencies are captured, and time series features are obtained; The time series features are decoded and mapped to obtain a readable character sequence; The character sequence is semantically parsed to obtain structured nameplate data; Based on preset standard fields, the structured nameplate data is repaired to obtain the final nameplate information.

[0006] In one possible implementation, acquiring the nameplate area image data includes: Obtain the original nameplate image data; The original nameplate image data is intelligently optimized to obtain optimized nameplate image data; the intelligent optimization process includes processing ambient light and processing the shooting angle. The optimized nameplate image data is located and cropped using the YOLOv8 model to obtain the nameplate area image data.

[0007] In one possible implementation, extracting the character space features of the nameplate area image data includes: Multiple convolution kernels are used to perform sliding convolution processing on the nameplate area image data to extract the local edge features of the characters and the spatial features of the characters in the nameplate image, thereby obtaining the character spatial features.

[0008] In one possible implementation, the step of capturing character sequence dependencies based on the character space features to obtain time series features includes: The character space features are compressed into a one-dimensional feature sequence; Capture the forward and backward contextual relationships of the one-dimensional feature sequence; The time series features are obtained by combining the forward and backward contextual relationships.

[0009] In one possible implementation, decoding the time series features to map them into a readable character sequence includes: At each time step, the probability distribution of characters in the time series features is predicted, and the optimal character sequence is obtained based on the maximum path probability. In response to the presence of duplicate characters and whitespace characters in the character sequence, the duplicate characters and whitespace characters are merged, and finally mapped to obtain the readable character sequence.

[0010] In one possible implementation, the semantic parsing of the character sequence to obtain structured nameplate data includes: The semantic parsing rule base is invoked to identify and match the parameter names in the character sequence to obtain the parameter names; Use regular expressions to match the numerical values ​​and units in the character sequence. The parameter name, the value, and the unit are recombined into key-value pairs to obtain the structured nameplate data.

[0011] In one possible implementation, the structured nameplate data is repaired based on preset standard fields to obtain the final nameplate information, including: Verify whether the matching relationship between the value and the unit is correct, and verify whether the association relationship between the range of the value and the parameter type of the parameter name is correct; In response to the presence of low-confidence data or missing data in the structured nameplate data, the missing data is restored, and the low-confidence data is repaired based on contextual semantic matching to obtain the final nameplate information.

[0012] Based on the same inventive concept, embodiments of this application also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the nameplate intelligent recognition method as described in any of the above.

[0013] Based on the same inventive concept, embodiments of this application also provide a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute any of the above-described nameplate intelligent recognition methods.

[0014] Based on the same inventive concept, this application also provides a computer program product, which includes computer program instructions, the computer instructions being used to cause the computer program product to execute any of the above-described nameplate intelligent recognition methods.

[0015] As can be seen from the above, the nameplate intelligent recognition method and related equipment provided in this application acquire nameplate area image data; extract character space features from the nameplate area image data; capture character sequence dependencies based on the character space features to obtain time series features; decode the time series features to map them into readable character sequences; perform semantic parsing on the character sequences to obtain structured nameplate data; and perform repair processing on the structured nameplate data based on preset standard fields to obtain the final nameplate information. This application embodiment collects and optimizes nameplate image data using a smart terminal device. This includes intelligently adjusting ambient lighting and shooting angle, and using the optimized image data to accurately locate and crop the nameplate area, providing high-quality input for subsequent recognition. Based on the cropped nameplate image data, local edge features and overall spatial features of the characters are extracted to form character spatial features. These character spatial features are then serialized to capture the forward and backward contextual relationships between characters, generating time-series features. By decoding the time-series features, a readable character sequence is mapped, and semantic parsing is performed to extract parameter names, values, and unit information, generating structured nameplate data. The structured data is repaired and validated based on preset standard fields to ensure the correct association between parameter types and units, and that the numerical range conforms to business rules. Low-confidence data and missing data are restored and repaired, ultimately outputting complete, accurate, and business-compliant nameplate information. This method achieves end-to-end optimization from image acquisition to information output through innovative data optimization, feature extraction, semantic parsing, and data repair mechanisms. It significantly improves recognition efficiency and accuracy, adapts to complex scenarios such as low light, backlight, and occlusion, and reduces processing time by more than 80% compared to traditional manual input methods. It provides reliable technical support for the recognition of dynamically changing nameplate information from multiple manufacturers and in multiple formats. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the nameplate intelligent recognition method according to an embodiment of this application; Figure 2 This is a schematic diagram of the electronic device structure according to an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0019] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0020] It is understood that before using the technical solutions of the various embodiments in this disclosure, users will be informed of the type, scope of use, and usage scenarios of the personal information involved in an appropriate manner, and user authorization will be obtained.

[0021] For example, upon receiving a user's active request, a prompt message is sent to the user to explicitly inform them that the requested operation will require the acquisition and use of the user's personal information. This allows the user to independently choose, based on the prompt message, whether to provide personal information to the software or hardware such as electronic devices, applications, servers, or storage media performing the operations of this disclosed technical solution.

[0022] As an optional but not limited implementation, in response to a user's active request, sending a prompt message to the user can be done via a pop-up window, where the prompt message can be presented in text format. Furthermore, the pop-up window can also include a selection control allowing the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0023] It is understood that the above notification and user authorization process are merely illustrative and do not constitute a limitation on the implementation of this disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of this disclosure.

[0024] As described in the background section, existing technologies mainly include traditional OCR and template matching schemes, as well as deep learning schemes, but none of them can meet the complex requirements of nameplates for distributed photovoltaic inverters. Traditional OCR cannot handle scenarios such as backlighting, blurring, and occlusion. While deep learning improves recognition capabilities, it lacks technical support such as illumination compensation and wear repair, resulting in insufficient adaptability. Fixed template matching is difficult to cover non-standardized formats from multiple manufacturers, and deep learning cannot automatically parse the semantic relationships between parameters, leading to poor format compatibility. Model updates require full retraining, which is time-consuming and resource-intensive. High sample dependence and a lack of enhancement mechanisms result in insufficient generalization ability.

[0025] Based on the above considerations, this application proposes an intelligent nameplate recognition method. The method involves acquiring image data of the nameplate area; extracting character space features from the image data; capturing character sequence dependencies based on the character space features to obtain time-series features; decoding the time-series features to map them into readable character sequences; performing semantic parsing on the character sequences to obtain structured nameplate data; and repairing the structured nameplate data based on preset standard fields to obtain the final nameplate information. This application intelligently optimizes image data, accurately locates the nameplate area, extracts and serializes character features, generates time-series features, and decodes them into readable character sequences. It combines a semantic parsing rule base to extract parameter names, values, and units, generating structured data and repairing low-confidence and missing data to ensure complete and accurate information. This method is adaptable to complex scenarios, improves processing efficiency by more than 80% compared to manual input, meets the dynamic recognition needs of nameplates from multiple manufacturers and in multiple formats, and provides reliable support for business systems.

[0026] The technical solutions of the embodiments of this application will be described in detail below through specific examples.

[0027] refer to Figure 1 The nameplate intelligent recognition method of this application includes the following steps: Step S101: Obtain image data of the nameplate area; Step S102: Extract the character space features of the nameplate area image data; Step S103: Based on the character space features, the character sequence dependency is captured to obtain the time series features; Step S104: Decode the time series features to obtain a readable character sequence; Step S105: Perform semantic parsing on the character sequence to obtain structured nameplate data; Step S106: Based on preset standard fields, the structured nameplate data is repaired to obtain the final nameplate information.

[0028] For step S101, obtain the image data of the nameplate area.

[0029] In some embodiments, acquiring the nameplate area image data includes: acquiring original nameplate image data; performing intelligent optimization processing on the original nameplate image data to obtain optimized nameplate image data; the intelligent optimization processing includes processing ambient light and processing the shooting angle; and using the YOLOv8 model to locate and crop the optimized nameplate image data to obtain the nameplate area image data.

[0030] In this embodiment, the process of acquiring nameplate area image data includes acquiring the original nameplate image data and generating the final nameplate area image data. The entire process aims to ensure the clarity and integrity of the nameplate area image data through a series of intelligent processing and high-precision cropping. First, acquiring the original nameplate image data is the starting point of the entire process, typically completed through a smart terminal device. The smart terminal device is equipped with a high-resolution camera to capture high-quality nameplate images, ensuring complete coverage of the nameplate area in the original image data. However, due to the complexity of shooting environment conditions (such as excessively strong or dark lighting, shooting angle deviations, etc.), the original nameplate image data may not be of directly usable quality. Therefore, further intelligent optimization processing is needed on the original nameplate image data to improve the overall image quality and provide a reliable data foundation for subsequent processing steps.

[0031] Ambient light processing is an essential part of intelligent optimization. Using the light sensor in the smart terminal device, this application can detect the ambient light intensity in real time. When the ambient light is insufficient (e.g., below 300 lux), this application automatically triggers a supplementary light to increase image brightness and optimize contrast, ensuring that the character information on the nameplate is effectively captured. Conversely, when the ambient light is too strong (e.g., above 1500 lux), this application automatically adjusts the camera's exposure parameters to avoid overexposure, while enhancing the contrast between the nameplate characters and the background, thereby better separating the nameplate information from interference information. After completing ambient light processing, the shooting angle also needs to be processed. Using the angle sensor in the smart terminal device, this application can detect whether the shooting angle meets the requirements. If the angle between the shooting angle and the nameplate plane exceeds 15°, this application will issue a prompt, reminding the staff to adjust the shooting posture to obtain more standardized nameplate image data. Furthermore, for slight angle deviations, this application can also correct them through subsequent image processing algorithms, thereby further improving the usability of the image.

[0032] After the above optimization process, optimized nameplate image data is obtained. The optimized image data meets the recognition requirements in terms of brightness, contrast, and angle. However, even the optimized nameplate image data may still contain a large amount of background information (such as the device casing, environmental background, etc.), which can interfere with subsequent character recognition. Therefore, in order to extract the nameplate area image data, it is necessary to use the YOLOv8 model to locate and crop the optimized nameplate image data. As a highly efficient object detection algorithm, the YOLOv8 model can quickly identify and locate the nameplate area in the image by performing convolution operations on the optimized image through its pre-trained feature extractor. To adapt to the characteristics of the inverter nameplate scenario, targeted optimizations were made to the general YOLOv8 model to further improve the positioning accuracy.

[0033] Specifically, the improved YOLOv8 model incorporates unique features of the nameplate, including the metallic background color, rectangular border, and the layout pattern of the nameplate parameters. The metallic background color, as a typical background material for nameplates, possesses relatively stable texture characteristics; therefore, the improved model incorporates the ability to recognize metallic textures in its feature extractor. The rectangular border is a common shape for nameplates; by enhancing the model's sensitivity to rectangular boundaries, it can more accurately detect the nameplate area. The layout pattern of the nameplate parameters provides additional contextual information to the model by analyzing the arrangement of parameters on the nameplate (such as uniform arrangement from left to right or vertical distribution), thereby improving localization accuracy. Furthermore, to further enhance the model's localization capabilities, the improved YOLOv8 model also adds an "edge gradient enhancement" module. This module identifies the features of edge regions in the image by calculating the grayscale change rate of image pixels, thereby strengthening the nameplate contour information. In practice, the edge gradient enhancement module performs pixel-by-pixel analysis of the nameplate boundaries in the image, identifying areas with significant grayscale value changes. This processing method highlights the boundary between the nameplate and the background, allowing the model to more clearly distinguish the nameplate area while effectively filtering out interfering information (such as the device casing or background environment). Through these improvements, the YOLOv8 model significantly enhances the accuracy of nameplate positioning, achieving a positioning accuracy of ≥98.5%, while reducing the false detection rate to <1%.

[0034] After localization, the cropping function of the YOLOv8 model was used to crop the optimized nameplate image data. Bounding boxes were used to define the effective area of ​​the nameplate; the cropping operation extracted this area from the image while removing redundant background. Through the cropping operation, image data containing only the nameplate content, i.e., the nameplate region image data, was finally obtained. This nameplate region image data has high resolution and high integrity, ensuring that the nameplate information is fully preserved. Through a series of optimization operations and efficient object detection algorithms, high-quality nameplate region image data was generated from the original nameplate image data. This data provides a high-quality input foundation for subsequent character extraction, semantic parsing, and structured information generation, and also lays a solid data foundation for the reliable operation of this application.

[0035] Furthermore, in step S102, the character space features of the nameplate area image data are extracted.

[0036] In some embodiments, extracting the character spatial features of the nameplate area image data includes: performing sliding convolution processing on the nameplate area image data using multiple convolution kernels to extract the local edge features of the characters and the spatial features of the characters in the nameplate image, thereby obtaining the character spatial features.

[0037] In this embodiment, the process of extracting the character spatial features from the nameplate area image data is a key step in extracting character morphological features from the nameplate area image data, providing necessary foundational support for subsequent character sequence modeling and semantic parsing. In some embodiments, this process uses multiple convolution kernels to perform sliding convolution processing on the nameplate area image data, gradually extracting the local edge features and spatial features of the characters, ultimately obtaining the character spatial features. First, the input data is the nameplate area image data obtained and cropped in step S101. This data has been optimized and already possesses high-quality image clarity and integrity, ensuring that the character information is fully presented in the image, providing a good input foundation for subsequent processing stages. However, the nameplate area image data is essentially still in the form of a two-dimensional image, with character information embedded in the pixel composition of the image. It is necessary to use feature extraction methods to transform the visual features of the characters into feature representations that can be used for subsequent processing.

[0038] To extract spatial features of the characters, convolutional neural networks (CNNs) were used. The convolution operation involves sliding convolution on the nameplate area image data using multiple convolutional kernels, scanning each region of the image pixel-by-pixel to capture the local edge features and spatial distribution features of the characters. Specifically, the convolutional kernel is a weighted matrix used to detect features at different directions and scales in the image. Through the sliding operation of the convolutional kernels, local features such as character edges, stroke thickness, and curve shapes can be identified, while the spatial arrangement patterns of the characters can also be extracted. The size and number of convolutional kernels directly affect the feature extraction effect. For example, smaller convolutional kernels (such as 3×3) are suitable for extracting detailed character features, while larger convolutional kernels (such as 7×7) can capture the global shape of the characters. In practical applications, different sizes of convolutional kernels are often used in combination to ensure comprehensive feature extraction.

[0039] Besides convolution operations, feature extraction also includes pooling. Pooling downsamples the feature map output by convolution, further simplifying the data volume while preserving key feature information. For example, max pooling selects the local maximum value from the convolution output feature map, highlighting significant edge features of characters; average pooling calculates the average value of local regions, preserving the overall distribution information of characters. The purpose of pooling is to reduce subsequent computational complexity and enhance the model's robustness to changes in character morphology in the image, enabling the model to accurately extract spatial features of characters even when faced with variations in font size or position.

[0040] During feature extraction, the unique scene features of the nameplate are also incorporated into the processing logic. For example, nameplate characters typically have high-contrast edge characteristics (such as black characters on a metallic background), which can be enhanced through edge detection in convolution operations. Furthermore, the arrangement of nameplate characters usually follows certain layout rules (such as uniform horizontal or vertical arrangement), which is gradually captured through multiple convolution operations and reflected in the feature map as the spatial distribution information of the characters. Through layer-by-layer convolution and pooling processing, the final character spatial features are generated. These features are output in the form of multi-dimensional feature tensors, where each dimension of the tensor corresponds to the feature description of different regions in the image, including character edge information, spatial arrangement rules, and local visual features.

[0041] The entire feature extraction process not only extracts character space features but also enhances the adaptability to nameplate scene features through convolution operations and pooling, ensuring that the model maintains high-precision feature extraction performance even in complex scenes (such as low light and backlight). The final generated character space features provide high-quality input data for subsequent steps (such as character sequence dependency capture and decoding), ensuring the accuracy and efficiency of the entire intelligent recognition process.

[0042] Furthermore, in step S103, character sequence dependencies are captured based on the character space features to obtain time series features.

[0043] In some embodiments, the step of capturing character sequence dependencies based on the character space features to obtain time series features includes: compressing the character space features into a one-dimensional feature sequence; capturing the forward and backward contextual relationships of the one-dimensional feature sequence; and combining the forward and backward contextual relationships to obtain the time series features.

[0044] In this embodiment, the process of capturing character sequence dependencies based on the character spatial features and ultimately generating time-series features is a crucial step in the transformation from spatial features to sequence features, providing necessary serialized input for subsequent character decoding and semantic parsing. In some embodiments, this process includes compressing the character spatial features into a one-dimensional feature sequence, capturing the forward and backward contextual relationships of the one-dimensional feature sequence, and combining these contextual relationships to generate time-series features. The input data is the character spatial features extracted in step S102. These features are represented in the form of multi-dimensional feature tensors, containing the local edge features of the characters, spatial distribution patterns, and overall visual information of the nameplate area image. However, character spatial features are essentially two-dimensional spatial distribution forms and cannot directly express the time-series dependencies of characters. Therefore, to meet the needs of character sequence modeling, it is necessary to serialize the multi-dimensional feature tensors and convert them into one-dimensional feature sequences.

[0045] During serialization, the character space features are first compressed along the horizontal direction of the image (i.e., the character arrangement direction). Specifically, the model aggregates the features of each column, compressing the two-dimensional feature tensor into a one-dimensional feature sequence, where each time step corresponds to a column of features in the image. This compression method preserves the contextual information of the characters while significantly reducing the complexity of the features, making subsequent sequence modeling more efficient. After compression, the generated one-dimensional feature sequence is used as input to further capture the contextual relationships between characters.

[0046] To capture the dependencies in a character sequence, it is necessary to analyze the forward and backward contextual relationships between characters. For this purpose, a bidirectional Long Short-Term Memory (BiLSTM) neural network is used to process the one-dimensional feature sequence. BiLSTM processes the character sequence through two LSTM networks operating in opposite directions. The forward LSTM processes the time steps of the one-dimensional feature sequence from left to right, capturing the forward dependencies between characters. For example, when processing the character sequence "220V", the forward LSTM can identify the continuity between the digits "220" and its association with the subsequent character "V". Simultaneously, the backward LSTM processes the one-dimensional feature sequence from right to left, capturing the backward dependencies between characters. For example, the backward LSTM can identify the association between the character "V" and the preceding digit "220". Through bidirectional processing, BiLSTM can generate contextual information for each time step, ensuring that the model has a comprehensive understanding of the relationships between characters.

[0047] By combining forward and backward contextual relationships, BiLSTM ultimately generates time-series features. These features are represented as a feature matrix, where each row corresponds to a time step in the one-dimensional feature sequence, and each column represents the contextual information for that time step. This representation not only preserves the serialization information of the characters but also includes the contextual dependencies between them, providing high-quality input data for subsequent decoding. Furthermore, the generation process of time-series features is robust and can adapt to the diversity of nameplate character arrangements. For example, regardless of whether the characters are arranged horizontally or slightly tilted, BiLSTM can capture global contextual information, identify the relationships between characters, and generate accurate time-series features.

[0048] The entire process successfully transformed spatial features into temporal series features through character spatial feature serialization, contextual relationship capture, and feature integration. This step ensures that the model can fully express the sequence information of the nameplate characters, laying a solid foundation for subsequent steps (such as decoding temporal series features and solving the character adhesion problem). Simultaneously, this process incorporates the specific needs of the nameplate scenario, further optimizing the adaptability to nameplate characters through bidirectional modeling and serialization processing. Even in complex scenarios (such as character adhesion or uneven character spacing), it can generate accurate temporal series features.

[0049] Furthermore, in step S104, the time series features are decoded to obtain a readable character sequence.

[0050] In some embodiments, decoding the time series features to map them into a readable character sequence includes: predicting the probability distribution of characters in the time series features at each time step, and obtaining the optimal character sequence based on the maximum path probability; in response to the presence of duplicate characters and whitespace characters in the character sequence, merging the duplicate characters and whitespace characters, and finally mapping them into the readable character sequence.

[0051] In this embodiment, the process of decoding the time series features to map them into a readable character sequence is a crucial step in converting the time series features into actual character representations, providing direct input for subsequent semantic parsing. In some embodiments, the decoding process generates an optimal character sequence by predicting the probability distribution of characters step-by-step based on the time series features, combined with a maximum path probability algorithm. During this process, the system also performs post-processing on the generated character sequence to resolve issues of repeated characters and whitespace characters, ultimately ensuring a clear and accurate readable character sequence output. The input data is the time series features generated in step S103, which are represented in the form of a feature matrix. Each row corresponds to a time step in the time series, and each column contains the context information of that time step and the feature representation of the character. However, the time series features themselves do not directly correspond to actual characters; they only contain character-related feature information and need to be mapped into a readable character sequence through the decoding process.

[0052] The first step in the decoding process is to predict the character probability distribution for each time step based on the time series features. In this process, the system progressively decodes the time series features using the Connected Temporal Classification (CTC) algorithm. Specifically, CTC generates a probability distribution for each time step in the time series, covering the probabilities of all possible characters, including the probability of a special "blank character." The "blank character" represents the case where no actual character corresponds to a time step, which is particularly important in scenarios with uneven character spacing or characters overlapping. The core of CTC decoding is to calculate the probabilities of all possible paths mapping from the time series features to the character sequence using dynamic programming, and select the character sequence with the highest path probability as the output. In this way, CTC can accurately decode time series features into character sequences even without explicit alignment annotations.

[0053] After generating the optimal character sequence, post-processing is required to address the issues of duplicate and whitespace characters. Since the CTC decoding process may generate identical characters (e.g., "AAABB") in adjacent time steps and insert additional whitespace characters (e.g., "AB_blank_C"), a merging process is necessary. Specifically, the system scans the decoded character sequence, removes all whitespace characters, and merges consecutive duplicate characters. For example, the decoded sequence "AAABB_blank_C" will be processed into "ABC". Through this post-processing step, a clear and readable character sequence without redundancy is finally generated.

[0054] The entire decoding process combines the unaligned decoding capability of the CTC algorithm with the character merging capability of post-processing to achieve efficient mapping from time-series features to readable character sequences. During decoding, CTC not only solves alignment problems caused by uneven character spacing or character adhesion but also ensures the global optimality of the decoding result through dynamic programming. Simultaneously, character merging further improves the readability of the decoding result, eliminating interference from repeated characters and "whitespace characters," ensuring that the output is consistent with the actual characters on the nameplate. This decoding method has significant advantages in adapting to nameplate scenarios; even when faced with complex nameplate formats or blurry or damaged characters, it can still generate accurate and readable character sequences, providing standardized input for subsequent semantic parsing and structured data generation. Ultimately, this step outputs a complete and accurate readable character sequence, laying the core foundation for intelligent recognition of nameplate information.

[0055] Furthermore, in step S105, the character sequence is subjected to semantic parsing processing to obtain structured nameplate data.

[0056] In some embodiments, the step of performing semantic parsing processing on the character sequence to obtain structured nameplate data includes: calling a semantic parsing rule library to identify and match parameter names in the character sequence to obtain parameter names; using regular expressions to match numerical values ​​and units in the character sequence; and recombining the parameter names, numerical values, and units into key-value pairs to obtain the structured nameplate data.

[0057] In this embodiment, the process of semantically parsing the character sequence and ultimately generating structured nameplate data is a crucial step in transforming a readable character sequence into structured data with clear logical relationships, providing standardized input for subsequent field repair and business system integration. In some embodiments, this process identifies and matches parameter names in the character sequence by calling a semantic parsing rule base, extracts numerical values ​​and units from the character sequence using regular expressions, and reassembles the parameter names, numerical values, and units into key-value pairs to generate structured nameplate data. The input data is the readable character sequence decoded in step S104, which has removed redundant information (such as repeated characters and whitespace characters) and is presented in a clear character arrangement. However, the character sequence itself only contains a linear representation of the characters and does not yet express the logical relationships and structured information between parameters. Therefore, further semantic parsing is required to extract key information from the nameplate and transform it into an easily processed structured form.

[0058] The first step in semantic parsing is to identify and match parameter names within a character sequence. By calling the semantic parsing rule base, the system can accurately identify parameter names within the character sequence. The semantic parsing rule base contains a rich dictionary of nameplate parameters, covering common parameter types found on inverter nameplates (such as "model," "rated power," and "voltage") and their corresponding semantic rules. For example, for the character sequence "Pn" or "Power," the rule base can match it as "rated power"; for "Model" or "Type," it can match it as "model." This process achieves automatic parameter name recognition through rule matching, ensuring that the parameter name correctly corresponds to the actual content on the nameplate. The rule matching design also considers the diversity of parameter names; for example, the same parameter may have different expressions. The rule base can associate multiple expressions through keyword vectors to improve the accuracy of semantic parsing.

[0059] After recognizing the parameter names, the system needs to extract the corresponding numerical values ​​and units from the character sequence. To do this, regular expressions are used to match the character sequence. Regular expressions are powerful pattern matching tools capable of extracting numerical values ​​and their corresponding units according to preset format rules. For example, for the character sequence "Pn: 30kW", regular expressions can identify the numerical value "30" and the unit "kW". Through this process, the system can accurately extract the parameter values ​​and unit information from the nameplate, avoiding information loss due to irregular character arrangement or format changes. Furthermore, the flexibility of regular expressions allows them to adapt to different nameplate formats, achieving accurate extraction regardless of whether the numerical values ​​and units are closely arranged or separated.

[0060] After extracting the parameter names, values, and units, the system reassembles this information into key-value pairs to generate structured nameplate data. The reassembly process uses the parameter name as the key and the value and unit as the value, combining them to form a standardized key-value pair structure. For example, the character sequence "Pn: 30kW" will generate the following key-value pair after semantic parsing: "Rated Power: 30kW"; the character sequence "Model: INV123" will generate "Model Number: INV123". This structured representation clearly expresses the logical relationships between the parameters on the nameplate, making the data easier to store, transmit, and process.

[0061] The entire semantic parsing process successfully transformed character sequences into structured nameplate data by invoking a semantic parsing rule base, regular expression matching, and key-value pair recombination. This process not only extracted key parameter information from the nameplates but also standardized the data to ensure its integrity and accuracy. The semantic parsing process also possesses strong adaptability, capable of handling diverse nameplate content (e.g., nameplates from different manufacturers and in different formats) and complex scenarios (e.g., incomplete or blurred characters). Ultimately, this step outputs a complete and standardized set of structured nameplate data, providing a reliable input foundation for subsequent steps (such as field repair and business system integration) and laying a solid technical foundation for the intelligent application of nameplate information.

[0062] Furthermore, in step S106, the structured nameplate data is repaired based on preset standard fields to obtain the final nameplate information.

[0063] In some embodiments, the step of repairing the structured nameplate data based on preset standard fields to obtain the final nameplate information includes: verifying whether the matching relationship between the value and the unit is correct, and verifying whether the association relationship between the range of the value and the parameter type of the parameter name is correct; in response to the presence of low-confidence data or missing data in the structured nameplate data, restoring the missing data, and repairing the low-confidence data based on contextual semantic matching to obtain the final nameplate information.

[0064] In this embodiment, the process of repairing the structured nameplate data based on preset standard fields and finally generating nameplate information is a key step in ensuring the accuracy, completeness, and standardization of the output data. Through the verification and semantic repair of standard fields, this process can effectively solve the problem of low-confidence data or missing data in the structured nameplate data, ensuring that the final nameplate information meets actual needs and business rules. In some embodiments, this process includes verifying the matching relationship between the numerical values ​​and units of the structured nameplate data, verifying the association relationship between the numerical range and the parameter type of the parameter name, restoring missing data, and repairing low-confidence data based on contextual semantic matching.

[0065] The input data is structured nameplate data generated in step S105. This data has been converted into key-value pairs through semantic parsing, containing information such as parameter names, values, and units. However, due to the complexity of the nameplate information sources (such as ambiguous characters and inconsistent formats), the structured data may still contain low-confidence data (such as abnormal numerical ranges or incorrect unit matching) or missing key fields (such as some parameters not being extracted). Therefore, in order to further improve data quality and ensure that it meets the standard requirements of the business system, the structured nameplate data needs to be repaired.

[0066] The first step in the repair process is to validate the data based on preset standard fields. These preset standard fields contain business rules for nameplate parameters and their corresponding matching relationships. For example, "rated power" must be paired with the unit "kW" or "MW," and "voltage" must be within the range of "100V-1000V." These rules are defined by the nameplate parameter semantic library and the business system field mapping table, enabling accurate validation of the matching relationship between numerical values ​​and units in structured data. For example, when "rated power: 500W" is detected, the system will determine that the unit is mismatched because "rated power" should correspond to "kW" or "MW." Simultaneously, the system also validates the range of values ​​to ensure they conform to the business rules for the parameter type. For example, the numerical range of "rated power" is typically "10kW-500kW." If a value outside this range (such as "600kW") is detected, the system will mark the data as abnormal. Through this validation process, the system can quickly identify potential errors in structured data and provide a basis for subsequent repairs.

[0067] After data validation, the system repairs missing and low-confidence data. For missing data, the system attempts to complete the information through restoration processing. Restoration processing uses contextual information and the relationship between the nameplate parameter rule base to infer and complete missing fields. For example, when the "rated power" field is missing, the system can infer it based on other known parameters (such as equipment model or rated voltage) and combine this with the nameplate parameter rule base to generate a high-confidence completion result. For low-confidence data, the system uses contextual semantic matching for repair. Contextual semantic matching re-infers low-confidence data by analyzing other fields in the structured data and their relationships. For example, when there is doubt about the matching relationship between the value and unit of a field, the system can correct the data by referring to the contextual information of other parameters (such as the numerical range relationship between "rated power" and "voltage"). Contextual semantic matching can also combine the rule base of the data source to generate repair results with high confidence and conforming to business rules.

[0068] The entire repair process involves three stages: verification, restoration, and semantic matching. This progressively improves the quality of structured nameplate data, ultimately generating nameplate information that meets business requirements. After repair, the nameplate information is converted into a standardized format, corresponding one-to-one with the field requirements of the business system. This ensures the data can be directly uploaded to the system and supports subsequent device management and data analysis functions. Simultaneously, this process addresses the complexity of the nameplate scenario, using contextual information and a rule base to efficiently repair missing and low-confidence data. Even when faced with blurry, damaged, or non-standardized nameplates, it can generate complete and accurate nameplate information. Finally, this step outputs a set of standardized nameplate information, providing reliable results for the entire intelligent nameplate recognition process and laying a solid data foundation for the efficient operation of the business system.

[0069] Furthermore, in some feasible embodiments, this application also designs an incremental learning dynamic iteration mechanism, which is an important innovative design to ensure that the model can adapt to dynamic changes in nameplate format. Its core lies in enabling the model to quickly adapt to new samples through automated sample quality screening, semi-automatic annotation, and local model fine-tuning, while avoiding the forgetting of existing knowledge. The purpose of this mechanism is not only to solve the problems of high update costs and reliance on large-scale samples in existing models, but also to shorten the model update cycle, reduce computational resource consumption, and improve the model's continuous adaptability. In practical applications, the incremental learning dynamic iteration mechanism first responds to new sample events. When the system detects new nameplate image data uploaded to the sample library, it automatically initiates the quality screening process. New samples may originate from new manufacturers or changes in nameplate format, and their identification confidence may be lower than a preset threshold (e.g., confidence < 80%). Such samples are marked as "samples to be labeled" by the system and pushed to the semi-automatic annotation platform.

[0070] In the semi-automatic annotation platform, sample annotation is effectively handled through a combination of "model pre-annotation + manual verification." The model pre-annotation stage utilizes the existing model to initially identify new samples, automatically annotating parameter names such as "model" and "rated power," while simultaneously predicting corresponding parameter values. This process significantly reduces the workload of manual annotation and improves overall annotation efficiency. Subsequently, the manual verification stage checks and corrects the model pre-annotation results. Staff only need to confirm the correctness of the parameter values, without needing to re-annotate parameter types, thereby further improving annotation efficiency. Using this annotation method, the pre-annotation accuracy can reach over 85%, while the overall annotation efficiency is improved by approximately 60%. Annotated samples are added to the incremental sample library, providing high-quality training data for subsequent model updates.

[0071] After acquiring incremental samples, the system performs local fine-tuning on the existing model to adapt to the new samples while avoiding negative impacts on the recognition ability of existing samples. The core strategy of local model fine-tuning is "parameter freezing + knowledge distillation". First, the low-level feature extraction layers of the model (such as the CNN part) are frozen to retain the ability to extract general visual features, and only the upper-level semantic parsing layers (such as the BiLSTM and CTC parts) are updated to reduce the overall computational burden of the model. Second, knowledge distillation technology is used to transfer the knowledge of the original model to the updated model, ensuring that the knowledge of the new samples can be integrated with the knowledge of the existing samples, avoiding catastrophic forgetting. During training, mini-batch training (batch size=8) and a low learning rate (1e-5) are used for optimization, and the model update can be completed in only 10-20 iterations. Through this fine-tuning strategy, the consumption of computational resources is reduced by about 70%, and the model update cycle is shortened from the traditional 24 hours to less than 1 hour.

[0072] After model fine-tuning, the system verifies the performance of the updated model to ensure it functions correctly and adapts to new sample formats. During testing, the recognition accuracy of new samples must meet a preset standard (e.g., ≥92%), while ensuring the model's ability to recognize existing samples remains stable. The validated model is then deployed to smart terminal devices or cloud servers for real-time nameplate information recognition tasks. Thanks to the incremental learning dynamic iteration mechanism, the adaptation cycle for new manufacturers is shortened from the traditional 7 days to 1 day, significantly improving the system's adaptability and response speed.

[0073] Overall, the incremental learning dynamic iteration mechanism forms an efficient dynamic update closed loop through sample quality screening, semi-automatic annotation, local model fine-tuning, and knowledge distillation. This mechanism not only solves the adaptation problem caused by nameplate format changes, but also significantly reduces model update costs and resource consumption, enabling the model to maintain continuous adaptability in complex scenarios, while providing technical support for the long-term operation of business systems.

[0074] Furthermore, in some feasible embodiments, this application also designs a low-sample-dependency training mechanism, which is a significant innovation designed to address the problems of large data requirements and high costs of sample generation and annotation during model construction and iteration. The core of this mechanism lies in significantly reducing the dependence on large-scale samples through a combination of feature transfer learning, sample augmentation generation, and semi-automatic annotation tools, while simultaneously improving the model's generalization and adaptability. In practical applications, low-sample-dependency training first utilizes general text features for feature transfer learning to reduce the model's need for specific samples. Characters in nameplates have certain universality, such as stroke structure and layout rules, and these features are not limited by specific manufacturers or specific nameplate formats. Therefore, by extracting and transferring these general features, the system enables the model to converge quickly with fewer specific samples. Feature transfer learning not only improves the model's adaptability to general scenarios but also significantly accelerates the model's training speed, increasing the convergence speed by approximately two times and greatly shortening the model's training cycle.

[0075] Building upon feature transfer learning, low-sample-dependency training further expands the training data volume through sample augmentation generation technology to cover more scenarios and situations. The sample augmentation generator generates a large number of simulated samples by randomly abrading, occluding, and mutilating existing samples. These augmented samples can simulate nameplate images in complex scenes, such as information loss due to character wear and content blurring caused by dust occlusion. The design goal of sample augmentation generation is to significantly improve the coverage of edge scenes. After sample augmentation, the proportion of "information-mutilated samples" increases from 5% to 30%, effectively improving the model's adaptability to complex scenes. Simultaneously, the number of samples is expanded to 10 times the size of the original sample set, significantly alleviating the problem of insufficient data. Through sample augmentation generation technology, model training can be optimized in a wider range of scenarios, ensuring its adaptability to diverse nameplate formats and environmental conditions in practical applications.

[0076] To further reduce the cost of manual annotation, low-sample-dependency training incorporates a semi-automatic annotation tool. This tool has a built-in "automatic parameter type recognition" function, automatically annotating parameter names in new samples, such as fields like "rated power" and "model." Staff only need to verify the parameter values ​​in the automatic annotation results, significantly reducing repetitive manual operations and improving annotation efficiency. Compared to traditional manual annotation methods, this semi-automatic method is 3 times more efficient and reduces annotation costs by approximately 60%. Furthermore, the semi-automatic annotation tool can perform secondary verification by combining model pre-annotation results with a semantic rule base, ensuring the accuracy and consistency of the annotation results. By combining sample augmentation generation and the semi-automatic annotation tool, low-sample-dependency training can generate high-quality training datasets with minimal manual intervention, providing reliable data support for model optimization.

[0077] In low-sample-dependency training, the model training process fully leverages the characteristics of augmented samples and transfer learning, enabling the model to maintain high accuracy and strong generalization ability even with only 1 / 5 the sample size of existing techniques. During training, the system optimizes the model's parameter settings by combining transfer learning techniques and the characteristics of augmented sample generation, allowing it to quickly adapt to the nameplate recognition needs of different scenarios. Ultimately, the optimized model achieves a recognition accuracy of ≥92% in new manufacturer nameplate scenarios, with a 75% improvement in generalization ability. This not only solves the problem of insufficient sample data but also significantly improves the model's stability and adaptability in complex environments.

[0078] Overall, the low-sample-dependency training mechanism, through the synergy of feature transfer learning, sample augmentation generation, and semi-automatic annotation tools, forms an efficient and resource-low-dependency training system. This mechanism not only significantly reduces the model's need for large-scale samples but also improves training efficiency and model adaptability, providing strong support for the long-term development of nameplate intelligent recognition technology. Through this mechanism, the model can achieve high accuracy and strong generalization ability with a small sample size, while significantly reducing data generation and annotation costs, laying a solid foundation for continuous model optimization and the efficient operation of business systems.

[0079] As can be seen from the above embodiments, the intelligent nameplate recognition method described in this application acquires nameplate area image data; extracts character space features from the nameplate area image data; captures character sequence dependencies based on the character space features to obtain time series features; decodes the time series features to map a readable character sequence; performs semantic parsing on the character sequence to obtain structured nameplate data; and repairs the structured nameplate data based on preset standard fields to obtain the final nameplate information. This application embodiment acquires nameplate area image data and uses a smart terminal device to collect the nameplate area. During the collection process, the original nameplate image data is intelligently optimized, including processing ambient light and correcting the shooting angle to improve image quality. The optimized image data is then precisely located and cropped using the YOLOv8 model to obtain nameplate area image data for subsequent processing. The process ensures the clarity and integrity of the acquired images through real-time illumination detection and angle sensors, while using the edge gradient enhancement module of the YOLOv8 model to strengthen the boundary of the nameplate area, resulting in a positioning accuracy of over 98.5%, providing high-quality input data for subsequent character recognition.

[0080] Based on the obtained nameplate area image data, a convolutional neural network (CNN) in a deep learning framework is used to extract features from the image. Multiple convolutional kernels are used to perform sliding convolution operations on the nameplate image data to extract local edge features and overall spatial features of the characters. The convolutional kernels capture key visual features such as character strokes, font size, and color contrast to generate a high-dimensional character spatial feature tensor, ensuring that character information in the nameplate image can be completely extracted, providing a foundation for subsequent modeling of character sequence dependencies.

[0081] Based on the extracted character spatial features, a bidirectional long short-term memory (BiLSTM) network is used to model the sequence dependencies of characters, compressing the character spatial features into a one-dimensional feature sequence. By capturing the forward and backward contextual relationships of the one-dimensional feature sequence and combining the forward and backward semantic information, time-series features containing time-series dependencies are generated. The bidirectional modeling process ensures that the model can capture the semantic associations between characters, such as the dependency between "220" and "V" in "220V". This process effectively addresses the diversity of character layout and order, providing global contextual support for subsequent character sequence decoding.

[0082] The connection-time classification (CTC) algorithm is used to decode time series features. First, the probability distribution of characters is predicted at each time step. Dynamic programming is then used to calculate the optimal path probability, thus decoding the most probable character sequence. Within the character sequence, the CTC algorithm further merges and eliminates duplicate and whitespace characters to ensure the coherence and accuracy of the sequence. This decoding process effectively addresses the problems of character concatenation, repetition, or missing characters, ultimately generating a readable character sequence that provides data support for subsequent semantic parsing.

[0083] The decoded character sequence undergoes semantic parsing. A pre-defined semantic parsing rule base is invoked to accurately identify and match parameter names within the character sequence, ensuring that the parameter names conform to industry standards and format requirements for nameplates. Regular expressions are used to match numerical values ​​and unit information within the character sequence, completing a comprehensive parsing of the nameplate parameters. After parsing, the parameter names, numerical values, and unit information are recombine into key-value pairs, generating structured nameplate data. This process ensures the accuracy of the parsing results through predefined semantic and format rules in the rule base, particularly verifying the logical relationships between parameter types, units, and numerical values.

[0084] Based on the generated structured nameplate data, the data is repaired using preset standard fields. First, the matching relationship between parameter values ​​and units is validated to ensure the correct association between parameter types and units; simultaneously, the parameter value range is checked to ensure it conforms to business rules, such as whether the rated power value exceeds a reasonable range. For low-confidence or missing data in the structured nameplate data, contextual semantic matching and completion algorithms are used for repair. For missing data, it is restored by combining parameter types and rule bases; for low-confidence data, confidence is increased by referring to contextual information. Finally, repaired and validated nameplate information is generated, ensuring data integrity, accuracy, and business adaptability.

[0085] The above methods optimize the entire process from nameplate image data acquisition to high-quality information output, significantly improving recognition efficiency and accuracy. Furthermore, intelligent optimization and repair mechanisms effectively adapt to nameplate data from different manufacturers and formats. The entire process is more than 80% faster than traditional manual data entry, significantly improving information processing efficiency and the automation level of business integration.

[0086] It should be noted that the method in this embodiment can be executed by a single device, such as a computer or server. The method can also be applied in a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method in this embodiment, and the multiple devices will interact with each other to complete the method described.

[0087] It should be noted that the above description describes some embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0088] Based on the same inventive concept, corresponding to any of the above embodiments, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the nameplate intelligent recognition method described in any of the above embodiments.

[0089] Figure 2 This embodiment illustrates a more specific hardware structure of an electronic device, which may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected internally via the bus 1050.

[0090] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0091] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.

[0092] The input / output interface 1030 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.

[0093] The communication interface 1040 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0094] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.

[0095] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.

[0096] The electronic devices described above are used to implement the corresponding nameplate intelligent recognition method in any of the foregoing embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0097] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides a non-transitory computer-readable storage medium that stores computer instructions for causing the computer to execute the nameplate intelligent recognition method as described in any of the above embodiments.

[0098] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0099] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the nameplate intelligent recognition method as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0100] Based on the same inventive concept, corresponding to the nameplate intelligent recognition method described in any of the above embodiments, this disclosure also provides a computer program product, which includes computer program instructions. In some embodiments, the computer program instructions can be executed by one or more processors of a computer to cause the computer and / or the processor to perform the nameplate intelligent recognition method. Corresponding to the execution entity for each step in each embodiment of the nameplate intelligent recognition method, the processor executing the corresponding step can belong to the corresponding execution entity.

[0101] The computer program product of the above embodiments is used to cause the computer and / or the processor to execute the nameplate intelligent identification method as described in any of the above embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0102] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application (including the claims) is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this application as described above, which are not provided in the details for the sake of brevity.

[0103] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this application, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this application, and this also takes into account the fact that the details of the implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this application will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of this application, it will be apparent to those skilled in the art that the embodiments of this application can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0104] Although this application has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0105] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of this application.

Claims

1. A method for intelligent identification of nameplates, characterized in that, include: Acquire image data of the nameplate area; Extract the character space features from the image data of the nameplate area; Based on the character space features, character sequence dependencies are captured, and time series features are obtained; The time series features are decoded and mapped to obtain a readable character sequence; The character sequence is semantically parsed to obtain structured nameplate data; Based on preset standard fields, the structured nameplate data is repaired to obtain the final nameplate information.

2. The method according to claim 1, characterized in that, The acquisition of the nameplate area image data includes: Obtain the original nameplate image data; The original nameplate image data is intelligently optimized to obtain optimized nameplate image data; the intelligent optimization process includes processing ambient light and processing the shooting angle. The optimized nameplate image data is located and cropped using the YOLOv8 model to obtain the nameplate area image data.

3. The method according to claim 1, characterized in that, The extraction of character space features from the nameplate area image data includes: Multiple convolution kernels are used to perform sliding convolution processing on the nameplate area image data to extract the local edge features of the characters and the spatial features of the characters in the nameplate image, thereby obtaining the character spatial features.

4. The method according to claim 1, characterized in that, The process of capturing character sequence dependencies based on the character space features to obtain time series features includes: The character space features are compressed into a one-dimensional feature sequence; Capture the forward and backward contextual relationships of the one-dimensional feature sequence; The time series features are obtained by combining the forward and backward contextual relationships.

5. The method according to claim 1, characterized in that, The step of decoding the time series features to map them into a readable character sequence includes: At each time step, the probability distribution of characters in the time series features is predicted, and the optimal character sequence is obtained based on the maximum path probability. In response to the presence of duplicate characters and whitespace characters in the character sequence, the duplicate characters and whitespace characters are merged, and finally mapped to obtain the readable character sequence.

6. The method according to claim 1, characterized in that, The semantic parsing process of the character sequence to obtain structured nameplate data includes: The semantic parsing rule base is invoked to identify and match the parameter names in the character sequence to obtain the parameter names; Use regular expressions to match the numerical values ​​and units in the character sequence; The parameter name, the value, and the unit are recombined into key-value pairs to obtain the structured nameplate data.

7. The method according to claim 6, characterized in that, The structured nameplate data is repaired based on preset standard fields to obtain the final nameplate information, including: Verify whether the matching relationship between the value and the unit is correct, and verify whether the association relationship between the range of the value and the parameter type of the parameter name is correct; In response to the presence of low-confidence data or missing data in the structured nameplate data, the missing data is restored, and the low-confidence data is repaired based on contextual semantic matching to obtain the final nameplate information.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 7.

9. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 7.

10. A computer program product comprising computer program instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1 to 7.

Images