Asset information identification method, system and device for returned electric energy meter, and storage medium
By combining a pre-trained target detection model and a deep learning OCR model with a convolutional neural network and a connection-time classification mechanism, the problem of low efficiency and poor accuracy in identifying asset information of decommissioned energy meters has been solved. This has enabled fast and accurate automatic identification of nameplate information and generation of structured data, thereby improving the efficiency and accuracy of energy meter management.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HAINAN POWER GRID CO LTD
- Filing Date
- 2025-12-09
- Publication Date
- 2026-05-01
AI Technical Summary
Existing technologies rely on manual or general OCR recognition for asset information identification of decommissioned energy meters, resulting in low efficiency and poor fault tolerance, making it difficult to quickly and accurately identify nameplate information under complex on-site conditions.
By employing a pre-trained object detection model and a deep learning OCR model, combined with a convolutional neural network and a connection-based temporal classification mechanism, key asset information regions are automatically identified and bounded, and structured data is generated through rule verification or historical database comparison and error correction.
It enables rapid and accurate identification of decommissioned electricity meter asset information under complex field conditions, with the identification accuracy rate increased to 99.7%, the processing speed reduced from 6.5 seconds to 4 seconds, and the processing capacity increased from 1,000 meters/day to 1,600 meters/day, significantly reducing labor intensity and error rate.
Smart Images
Figure CN121963162A_ABST
Abstract
Description
A method, system, device, and storage medium for asset information identification of decommissioned electricity meters. Technical Field
[0001] This invention relates to the field of automated identification technology for power metering equipment, and in particular to a method, system, device, and storage medium for identifying asset information of decommissioned power meters. Background Technology
[0002] For metering assets centered on smart meters, accurate management throughout their entire lifecycle is crucial. Efficient sorting, testing, and reuse of returned meters are key business processes for power grid companies to achieve resource recycling, cost reduction, and efficiency improvement. However, in practice, discrepancies exist between the physical nameplate information (such as asset code, specifications, and manufacturer) of many returned meters and the records in the marketing system, necessitating verification and correction on a meter-by-meter basis.
[0003] Currently, asset verification is primarily done manually. Operators must use handheld barcode scanners to scan asset barcodes, visually read various characters on the nameplates, and then manually enter the information into spreadsheets for comparison. For older meters without barcodes or with damaged barcodes (such as the 2011 version of electricity meters), manual identification and data entry are entirely required. Existing automation attempts, such as using general optical character recognition technology, suffer from low accuracy when faced with complex field conditions (e.g., diverse nameplate formats, fonts, complex backgrounds, and potential damage, tilting, or uneven lighting on electricity meter nameplates), failing to meet the stringent accuracy requirements of the business. In summary, existing technologies generally suffer from low efficiency, high manual labor intensity, high error rates, and a lack of adaptive recognition capabilities for complex, non-standard nameplates. Summary of the Invention
[0004] In view of the above-mentioned problems, the present invention provides a method, system, device and storage medium for asset information identification of decommissioned electricity meters.
[0005] Therefore, the technical problem solved by this invention is: how to overcome the problems of low efficiency and poor fault tolerance of existing technologies that rely on manual or general OCR (Optical Character Recognition), and to provide a method that can adapt to different nameplate styles and still quickly and accurately identify the asset information of decommissioned energy meters under complex field conditions.
[0006] To address the aforementioned technical problems, the present invention provides the following technical solution: Firstly, the present invention provides a method for identifying asset information of decommissioned energy meters, comprising: placing a single decommissioned energy meter in a fixed position on the device's workbench, triggering a start command, and automatically focusing and capturing an image containing complete nameplate information using a smart camera integrated above the device; calling a pre-trained target detection model to process the image containing complete nameplate information, identifying and selecting sub-image regions containing key asset information; sequentially inputting each sub-image region into a deep learning OCR model for character recognition, converting pixel information in the image into an original string sequence; assigning semantic labels to the corresponding text results based on the region categories detected by the target detection model, correcting errors and determining confidence levels in the recognition results through rule verification or comparison with a historical database, automatically filling them into corresponding fields, and generating a structured asset information record.
[0007] As a preferred embodiment of an asset information identification method for decommissioned energy meters, the step of calling a pre-trained target detection model to process an image containing complete nameplate information and identify and select the sub-image region where the key asset information region is located includes: extracting multi-scale visual features of the image containing complete nameplate information through the feature extraction network in the target detection model; and based on the multi-scale visual features, predicting the bounding boxes and categories of multiple different types of key asset information regions in the nameplate image in parallel during the same forward inference process.
[0008] As a preferred solution for asset information identification of decommissioned energy meters, the method of calling a pre-trained target detection model to process the image containing complete nameplate information and identify and select the sub-image region where the key asset information region is located further includes: the key asset information region includes the asset identification code region and the text information region; and the sub-image corresponding to each key information region is segmented from the nameplate image according to the prediction results.
[0009] As a preferred embodiment of an asset information identification method for decommissioned energy meters, the step of sequentially inputting each sub-image region into a deep learning OCR model for character recognition and converting pixel information in the image into an original string sequence includes: extracting visual features from each input sub-image region and forming a feature sequence in the width dimension through a convolutional feature extraction network in an end-to-end character recognition model; inputting the feature sequence into a sequence modeling network, and generating an encoded sequence containing contextual information based on the contextual dependencies between features at each position in the sequence.
[0010] The beneficial effects of this preferred technical solution are as follows: by extracting robust visual features and serializing them through a convolutional network, and then using a sequence modeling network to capture contextual dependencies, the model's ability to recognize blurred, deformed or partially occluded characters is enhanced, thereby improving the accuracy of text recognition in complex scenes.
[0011] As a preferred scheme for asset information identification of decommissioned energy meters, the step of sequentially inputting each sub-image region into a deep learning OCR model for character recognition and converting pixel information in the image into an original string sequence further includes: decoding the encoded sequence based on a connection-time classification mechanism to output a character sequence corresponding to the content of the sub-image region; wherein, the end-to-end character recognition model is optimized through a training process that includes a connection-time classification loss function.
[0012] The beneficial effects of this preferred technical solution are as follows: by using a connection-based temporal classification mechanism for decoding and training, the model can directly output character sequences without relying on precise character segmentation, which reduces the requirements for image preprocessing and improves the robustness of recognition of irregular layouts or low-quality text.
[0013] As a preferred embodiment of an asset information identification method for decommissioned energy meters, the method involves: assigning semantic tags to the corresponding text results identified based on the region categories detected by the target detection model; correcting errors and assessing confidence levels by verifying the identification results through rule validation or comparison with a historical database; and automatically filling the corresponding fields to generate structured asset information records. This includes: assigning corresponding semantic tags to the corresponding character sequences based on the region category information output by the target detection model; comparing the identified text results with preset validation rules or a historical identification database; performing error correction processing; and evaluating the confidence level of each identification result.
[0014] As a preferred embodiment of an asset information identification method for decommissioned energy meters, the method further includes: assigning semantic tags to the corresponding text results identified based on the region category detected by the target detection model; correcting errors and determining confidence levels in the identification results through rule verification or comparison with historical databases; and automatically filling the corresponding fields to generate structured asset information records. This also includes: automatically filling the corresponding data fields of a preset template based on the evaluated text identification results and their semantic tags through field mapping relationships; and generating asset information records with standard field structures based on the filling results.
[0015] Secondly, this invention provides an asset information identification system for decommissioned energy meters, comprising: an image acquisition and positioning module, used to place a single decommissioned energy meter in a fixed position on the device's workbench, trigger a start command, and automatically focus and capture an image containing complete nameplate information using a smart camera integrated on top of the device; an information region detection module, used to call a pre-trained target detection model to process the image containing complete nameplate information, identify and select the sub-image region where the key asset information region is located; a character sequence recognition module, used to input each sub-image region sequentially into a deep learning OCR model for character recognition, converting the pixel information in the image into the original string sequence; and a structured processing and output module, used to assign semantic labels to the corresponding text results based on the region categories detected by the target detection model, perform error correction and confidence judgment on the recognition results through rule verification or comparison with historical databases, automatically fill in the corresponding fields, and generate a structured asset information record.
[0016] Thirdly, the present invention provides a computer device, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, wherein when the computer-executable instructions are executed by the processor, the steps of an asset information identification method for decommissioned energy meters are implemented.
[0017] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of an asset information identification method for decommissioned energy meters.
[0018] The beneficial effects of this invention are as follows: Addressing the practical problems of low efficiency and high error rate in manual verification of decommissioned electricity meter assets, this invention applies intelligent image recognition technology to the sorting scenario of electricity metering assets. It achieves automatic collection and structured input of nameplate information from decommissioned electricity meters. It can adaptively handle nameplate formats from different manufacturers and versions, effectively overcoming complex working conditions such as uneven lighting, nameplate damage, and character tilt. The verification time for a single meter asset information is reduced from 6.5 seconds manually to less than 4 seconds, and the recognition accuracy rate is improved to over 99.7%. The original scanning, reading, and input process, which required two people, is simplified to a single-person placement operation, increasing the daily processing capacity from 1000 meters to 1600 meters. This significantly reduces labor intensity and avoids errors such as misalignment and omissions that are prone to occur during manual input. It provides efficient and reliable data support for the accurate recycling, repair, and scrapping of decommissioned electricity meters, thus improving the quality and efficiency of metering asset verification work. Attached Figure Description
[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0020] Figure 1 is an overall flowchart of an asset information identification method for decommissioned electricity meters provided by the present invention.
[0021] Figure 2 is a schematic diagram of the hardware device for an asset information identification method for decommissioned electricity meters provided by the present invention. Detailed Implementation
[0022] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.
[0023] Example 1, referring to Figure 1, is the first embodiment of the present invention. This embodiment provides a method for identifying asset information of decommissioned energy meters, including: S1: After placing a single decommissioned energy meter in a fixed position on the device workbench, a start command is triggered, and an image containing complete nameplate information is automatically focused and captured by a smart camera integrated on top of the device; S2: A pre-trained target detection model is invoked to process the image containing complete nameplate information, identify and select the sub-image region where the key asset information area is located; S3: Each sub-image region is sequentially input into a deep learning OCR model for character recognition, converting the pixel information in the image into the original string sequence; S4: According to the region category detected by the target detection model, the corresponding text result is assigned a semantic label, and the recognition result is corrected and confidence judged through rule verification or comparison with a historical database, and automatically filled into the corresponding field to generate a structured asset information record.
[0024] It should be noted that steps S1-S4 achieve an end-to-end automated identification process from physical meters to structured data, transforming manual visual judgment and data entry into machine vision-driven intelligent processing. This effectively solves the problems of low efficiency and high error rate in manual identification of returned energy meters due to their diverse models and formats. The entire process is completed in a closed loop within a single device, significantly improving the accuracy and speed of asset information verification. Furthermore, the structured data output provides a directly usable data foundation for the subsequent asset management system, realizing a crucial shift from manual operation to intelligent operation in the metered asset sorting process.
[0025] Example 2, referring to Figures 1-2, is an embodiment of the present invention. Based on the previous embodiment, it provides a method for identifying asset information of decommissioned energy meters, including: In this embodiment, after placing a single decommissioned energy meter at a fixed position on the device workbench in step S1 above, triggering a start command, and automatically focusing and capturing an image containing complete nameplate information using a smart camera integrated on top of the device, the method includes: manually picking up the decommissioned energy meter and placing it at a fixed position on the device workbench, triggering a start command. The smart camera on top of the device automatically focuses and captures an image of the energy meter nameplate based on optical technology, obtaining a digital image containing complete nameplate information including asset code, equipment name, manufacturer, and specifications.
[0026] In another possible implementation, when the energy meter is detected to be placed in a designated position on the workbench, an electrical signal can be automatically generated by a pressure sensor or a photoelectric sensor to trigger the start command, thus achieving contactless automatic triggering.
[0027] In another possible implementation, the start command can be triggered by the operator manually pressing a physical button on the device or a virtual button on the touchscreen, so as to provide a clear point of human control.
[0028] It should be noted that the device refers to the hardware device that applies the method of the present invention, as shown in Figure 2.
[0029] In this embodiment, step S2 above, which calls a pre-trained object detection model to process the image containing complete nameplate information and identify and select the sub-image region where the key asset information region is located, includes: In this embodiment, it is preferable to call a pre-trained YOLO object detection model to process the acquired nameplate image. The YOLO model processes the image through its convolutional neural network (CNN), generates multi-scale feature maps, and detects all target regions in the image at once based on these feature maps.
[0030] The YOLO object detection model uses an end-to-end training approach and is optimized for key asset information areas on the nameplate of electricity meters, including barcodes, asset codes, manufacturers, and specifications, making it suitable for real-time object detection.
[0031] It should be noted that, through the processing of the YOLO object detection model, precise sub-image regions containing key information can be efficiently identified and selected under conditions of varying lighting, occlusion, and complex backgrounds. This solves the problem of inconsistent placement of key information on nameplates caused by the aging and diverse models and specifications of decommissioned energy meters.
[0032] In another possible implementation, multi-scale visual features can be extracted by employing a Feature Pyramid Network (FPN) structure. By fusing feature maps from different depths of the backbone network through top-down and lateral connections, a multi-scale feature representation with rich semantic information and accurate spatial location can be generated.
[0033] In another possible implementation, multi-scale visual features can be extracted by replacing ordinary convolutional layers with dilated convolutions, thereby expanding the receptive field without increasing parameters and computational cost, enabling the network to capture contextual information at different scales on a single-layer feature map.
[0034] In this embodiment, step S3 above, where each sub-image region is sequentially input into a deep learning OCR model for character recognition, converting pixel information in the image into the original string sequence, includes: sequentially inputting each key asset information sub-image region into a deep learning OCR model based on a CRNN (Convolutional Recurrent Neural Network) + CTC (Connectionist Temporal Classification) architecture for character recognition. The deep learning OCR model primarily utilizes intelligent AI algorithms and machine vision technology to convert image information into editable text or numbers.
[0035] Specifically, deep learning OCR models based on the CRNN+CTC architecture include: Convolutional Neural Network (CNN) feature extraction: Let the input single-line text image be I, with dimensions H x W. After passing through a CNN backbone (such as VGG / ResNet) for feature mapping: X = CNN(I), where X is the extracted feature map with dimensions H' x W' x C (C is the number of channels). The feature map can be viewed as a sequence in the width dimension, with a sequence length of W', and the feature at each time step is a C-dimensional vector. This yields the feature sequence x = (x1, x2, ..., x...). W').
[0036] Recurrent Neural Network (RNN) sequence modeling: The feature sequence x extracted by the CNN is input into a bidirectional LSTM (Bi-LSTM) to model the contextual information of the sequence. The output of each time step contains information about the current character and its context.
[0037] h = BiLSTM(x) where h = (h1, h2, ..., h T (T = W') is the sequence of hidden states output by the RNN, where each h t It is a vector representing the feature at position t.
[0038] In another possible implementation, the sequence modeling network can also be performed using a gated recurrent unit (GRU) network. GRU captures long-term dependencies in the sequence with a simpler structure through update gate and reset gate mechanisms, thereby ensuring the sequence modeling effect while potentially having a faster training and inference speed.
[0039] In another possible implementation, a sequence modeling network can be constructed using a Transformer encoder, which utilizes a self-attention mechanism to compute global dependencies between all positional features in the sequence, processes the entire sequence in parallel, and is particularly adept at handling long sequences and capturing complex contextual information.
[0040] CTC Loss Function: The CTC (Connectionist Temporal Classification) loss function is used to train the model to solve the problem of character and label alignment. CTC allows the system to handle irregular character sequences without requiring precise label alignment. The loss function is expressed as: L CTC =-log Where β is the set of all possible label sequences, and p(A / X) is the probability of the model predicting the label sequence.
[0041] The model uses the CTC loss function to address the character-label alignment issue during training. In the recognition (prediction) phase, the system processes the RNN output sequence. The decoding mechanism converts the sequence information into the final character sequence, thereby converting the image pixel information into the original string sequence (such as "XX manufacturer" or "DDZY102-Z").
[0042] It should be noted that the deep learning OCR model based on the CRNN+CTC architecture has been specifically optimized for the recognition of Chinese characters, Arabic numerals, and letters. Even for characters with poor print quality, slight angular deviations, or partial obscuration by dust or foreign objects, it can maintain a high recognition rate through contextual analysis. Furthermore, each successfully recognized character is recorded and stored in the deep learning database. By accumulating image information, incremental learning and adaptive optimization are achieved, continuously improving the model's adaptability to different styles, backgrounds, and fonts, ensuring consistently accurate character recognition.
[0043] In this embodiment, step S4 above assigns semantic tags to the corresponding text results identified by the target detection model based on the region category detected by the target detection model, and corrects and judges the confidence level of the identification results through rule verification or comparison with historical database, and automatically fills them into the corresponding fields to generate structured asset information records. This includes: assigning corresponding semantic tags to the corresponding text results identified by the OCR model based on the region category information (such as "asset code" and "manufacturer") output by the YOLO target detection model.
[0044] By comparing the results with a pre-defined rule base and / or a historical recognition database, logical verification, error correction, and confidence level assessment are performed on the recognition results to improve data accuracy.
[0045] After verification, the structured data (including asset code, equipment name, manufacturer, specifications, year, etc.) is automatically filled into the corresponding field cells of a pre-set spreadsheet template using a field mapping algorithm. The template includes pre-defined fields such as "asset code" and "manufacturer" to ensure accurate matching between the identified content and the template fields, thereby generating a complete structured asset information record that can be directly used for subsequent asset verification and management.
[0046] In another possible implementation, when automatically filling in the corresponding fields, the structured recognition results can also be converted into JSON or XML data format and written directly to the specified table field of the backend database via an application programming interface (API) call.
[0047] In another possible implementation, when automatically filling into the corresponding fields, after establishing a mapping relationship between the recognition results and preset template fields, an automation script (such as a Python script combined with an office software library) can be driven to simulate keyboard input and fill the data item by item into the corresponding cells of an open spreadsheet software (such as Excel).
[0048] Example 3: The above is an illustrative scheme of an asset information identification method for decommissioned energy meters according to this embodiment. It should be noted that the technical solution of an asset information identification system for decommissioned energy meters and the technical solution of the asset information identification method for decommissioned energy meters described above belong to the same concept. Details not described in detail in the technical solution of the asset information identification system for decommissioned energy meters in this embodiment can be found in the description of the technical solution of the asset information identification method for decommissioned energy meters described above.
[0049] This embodiment also provides an asset information identification system for decommissioned energy meters, including: an image acquisition and positioning module, used to place a single decommissioned energy meter in a fixed position on the device's workbench, trigger a start command, and automatically focus and capture an image containing complete nameplate information using a smart camera integrated on top of the device; an information region detection module, used to call a pre-trained target detection model to process the image containing complete nameplate information, identify and select the sub-image region where the key asset information region is located; a character sequence recognition module, used to input each sub-image region sequentially into a deep learning OCR model for character recognition, converting the pixel information in the image into the original string sequence; and a structured processing and output module, used to assign semantic labels to the corresponding text results based on the region categories detected by the target detection model, perform error correction and confidence judgment on the recognition results through rule verification or comparison with historical databases, automatically fill in the corresponding fields, and generate a structured asset information record.
[0050] This embodiment also provides an electronic device applicable to a method for identifying asset information of decommissioned energy meters, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the asset information identification method for decommissioned energy meters as proposed in the above embodiment.
[0051] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements an asset information identification method for decommissioned energy meters as proposed in the above embodiments.
[0052] The storage medium proposed in this embodiment belongs to the same inventive concept as the asset information identification method for decommissioned energy meters proposed in the above embodiment. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.
[0053] Example 4, referring to Table 1, is an embodiment of the present invention, providing a method for identifying asset information of decommissioned electricity meters. To verify the beneficial effects of the present invention, a simulation experiment is conducted for scientific demonstration.
[0054] Through installation and use at a power grid company's metering center, the following results were achieved: 1. The accuracy rate of identifying decommissioned electricity meter asset information reached 99.7%.
[0055] 2. The success rate of identifying returned electricity meter assets reached 99.9%.
[0056] 3. Improve on-site operation efficiency by 60%, increasing the efficiency of verifying returned electricity meter asset information from 1000 meters / day (manual) to 1600 meters / day. This effectively reduces the workload of frontline workers, improves work accuracy, and enhances the quality and efficiency of asset management.
[0057] The test procedure is shown in Table 1: Table 1 Comparison of operating efficiency between manual verification and automatic identification device test data
[0058] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A method for identifying asset information of decommissioned electricity meters, characterized in that, include: After placing a single decommissioned energy meter in a fixed position on the device's workbench, a start command is triggered. The smart camera integrated on top of the device automatically focuses and captures an image containing complete nameplate information. A pre-trained object detection model is invoked to process the image containing complete nameplate information, identify and select the sub-image region containing the key asset information area. Each sub-image region is then sequentially input into a deep learning OCR model for character recognition, converting the pixel information in the image into the original string sequence. Based on the region categories detected by the target detection model, the corresponding text results are assigned semantic labels. Through rule verification or comparison with historical databases, the recognition results are corrected and confidence levels are judged. The labels are then automatically filled into the corresponding fields to generate structured asset information records.
2. The asset information identification method for decommissioned energy meters as described in claim 1, characterized in that, The process of calling a pre-trained target detection model to process an image containing complete nameplate information and identify and select the sub-image region where the key asset information region is located includes: extracting multi-scale visual features of the image containing complete nameplate information through the feature extraction network in the target detection model; and predicting the bounding boxes and categories of multiple different types of key asset information regions in the nameplate image in parallel during the same forward inference process based on the multi-scale visual features.
3. The asset information identification method for decommissioned energy meters as described in claim 2, characterized in that, The process of calling the pre-trained target detection model to process the image containing complete nameplate information and identify and select the sub-image region where the key asset information region is located also includes: the key asset information region includes the asset identification code region and the text information region; and the sub-image corresponding to each key information region is segmented from the nameplate image according to the prediction results.
4. The asset information identification method for decommissioned energy meters as described in claim 3, characterized in that, The step of sequentially inputting each sub-image region into a deep learning OCR model for character recognition and converting pixel information in the image into an original string sequence includes: extracting visual features from each input sub-image region and forming a feature sequence in the width dimension through a convolutional feature extraction network in an end-to-end character recognition model; inputting the feature sequence into a sequence modeling network and generating an encoded sequence containing contextual information based on the contextual dependencies between features at each position in the sequence.
5. The asset information identification method for decommissioned energy meters as described in claim 4, characterized in that, The step of sequentially inputting each sub-image region into a deep learning OCR model for character recognition and converting pixel information in the image into an original string sequence further includes: decoding the encoded sequence based on a connection-time classification mechanism to output a character sequence corresponding to the content of the sub-image region; wherein, the end-to-end character recognition model is optimized through a training process that includes a connection-time classification loss function.
6. The asset information identification method for decommissioned energy meters as described in claim 5, characterized in that, The process of assigning semantic labels to the corresponding text results based on the region categories detected by the target detection model, correcting errors and judging confidence levels by rule verification or comparison with historical databases, and automatically filling them into the corresponding fields to generate structured asset information records includes: assigning corresponding semantic labels to the corresponding character sequences based on the region category information output by the target detection model; comparing the text recognition results after assigning semantic labels with preset verification rules or historical recognition databases, performing error correction processing, and evaluating the confidence level of each recognition result.
7. The asset information identification method for decommissioned energy meters as described in claim 6, characterized in that, The process of assigning semantic labels to the corresponding text results identified by the target detection model based on the region category, correcting errors and judging confidence levels by rule verification or comparison with historical databases, and automatically filling them into the corresponding fields to generate structured asset information records also includes: automatically filling the corresponding data fields of the preset template based on the evaluated text recognition results and their semantic labels through field mapping relationships; and generating asset information records with standard field structures based on the filling results.
8. An asset information identification system for decommissioned electricity meters, employing the method described in any one of claims 1 to 7, characterized in that, include: The image acquisition and positioning module is used to place a single decommissioned energy meter in a fixed position on the device's workbench, trigger a start command, and automatically focus and capture an image containing complete nameplate information using a smart camera integrated on top of the device; the information area detection module is used to call a pre-trained target detection model to process the image containing complete nameplate information, identify and select the sub-image area where the key asset information area is located. The character sequence recognition module is used to input each sub-image region sequentially into the deep learning OCR model for character recognition, converting the pixel information in the image into the original string sequence; the structured processing and output module is used to assign semantic labels to the corresponding text results based on the region categories detected by the object detection model, perform error correction and confidence judgment on the recognition results through rule verification or comparison with historical databases, and automatically fill in the corresponding fields to generate structured asset information records.
9. An electronic device, characterized in that, include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, It stores computer-executable instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 7.