Method and related device for intelligent verification of form image corresponding to power industry expansion

Abstract

The application provides a kind of intelligent verification method of power industry expansion corresponding form image and related equipment, the method includes in response to receiving form image, image enhancement processing is carried out to form image;Using power character recognition model, the form image after image enhancement is identified, and field data is obtained;Using power knowledge graph, the semantic rectification is carried out to field data, and corrected field data is obtained;At least one first suspicious data is obtained by dynamic rule verification to corrected field data, and at least one second suspicious data is obtained by static rule verification to corrected field data;For each target suspicious data, the similarity of target suspicious data and its corresponding predetermined field data is determined, in response to determining that the similarity is less than predetermined similarity, target suspicious data is determined as abnormal field data, the technical problem that the verification accuracy of form image is poor in the prior art is solved, and the efficiency of verifying form image is improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to an intelligent verification method and related equipment for form images corresponding to power industry expansion applications. Background Technology

[0002] With the accelerating pace of digitalization and informatization, users need to fill out various necessary forms during both online and offline electricity service processing, as well as in multiple stages of on-site operations. The information in these forms must be properly stored in the system. Electricity expansion application, as a core business process in the power industry, involves processing diverse forms such as low-voltage residential electricity, high-voltage customer electricity, on-site survey forms, and power supply plan responses, as well as verifying and entering supporting materials. The quality of the form data directly affects the smoothness of business processing and customer service experience. Currently, the data entry and verification of electricity expansion application forms mainly rely on manual work, which is prone to errors. Furthermore, during the subsequent verification of the entered form images, it is difficult for humans to detect all errors, resulting in poor accuracy in manual verification of form images. Summary of the Invention

[0003] In view of this, the purpose of this application is to propose an intelligent verification method and related equipment for the form images corresponding to power industry expansion application, so as to overcome all or part of the shortcomings of the prior art.

[0004] To achieve the above objectives, this application provides an intelligent verification method for form images corresponding to power industry expansion applications, comprising: in response to receiving a form image, performing image enhancement processing on the form image; using a pre-constructed power character recognition model to recognize the image-enhanced form image to obtain field data; using a pre-constructed power knowledge graph to perform semantic correction on the field data to obtain corrected field data; performing dynamic rule verification on the corrected field data to obtain at least one first suspicious data, and performing static rule verification on the corrected field data to obtain at least one second suspicious data, and taking both the first suspicious data and the second suspicious data as target suspicious data; for each target suspicious data, determining the similarity between the target suspicious data and its corresponding predetermined field data, and in response to determining that the similarity is less than the predetermined similarity, identifying the target suspicious data as abnormal field data.

[0005] Optionally, the step of performing dynamic rule verification on the corrected field data to obtain at least one first suspicious data includes: obtaining at least one currently stored derivative rule, wherein each derivative rule is determined by mining multiple historical abnormal field data; and verifying the corrected field data using each derivative rule to obtain the at least one first suspicious data.

[0006] Optionally, the at least one second suspicious data includes at least one first sub-suspicious data, at least one second sub-suspicious data, and at least one third sub-suspicious data; the step of performing static rule validation on the corrected field data to obtain at least one second suspicious data includes: obtaining historical stored data corresponding to the form image; comparing the corrected field data with the historical stored data to obtain the at least one first sub-suspicious data; checking the corrected field data according to predetermined form filling rules to obtain the at least one second sub-suspicious data; and checking the corrected field data according to predetermined business filling rules to obtain the at least one third sub-suspicious data.

[0007] Optionally, determining the similarity between the target suspicious data and its corresponding predetermined field data includes: determining the character similarity, pinyin similarity, abbreviation similarity, and semantic segmentation similarity between the target suspicious data and its corresponding predetermined field data; and performing a weighted calculation on the character similarity, the pinyin similarity, the abbreviation similarity, and the semantic segmentation similarity to obtain the similarity.

[0008] Optionally, the step of using a pre-constructed power knowledge graph to perform semantic correction on the field data to obtain corrected field data includes: extracting key fields from the field data to obtain multiple key field data; for each key field data, using the power knowledge graph to perform unit standardization, address normalization, industry classification mapping, and semantic error correction on the key field data to obtain at least one sub-corrected field data; and determining at least one sub-corrected field data corresponding to all key field data as the corrected field data.

[0009] Optionally, the image enhancement processing of the form image includes: sequentially performing multi-level filtering processing, tilt correction processing, crease repair processing, background interference removal processing, illumination interference removal processing, and character recognition enhancement processing on the form image to obtain the form image after image enhancement processing.

[0010] Optionally, after identifying the target suspicious data as abnormal field data, the method includes: generating and sending a verification report corresponding to the form image based on at least one abnormal field data and the corrected field data.

[0011] Based on the same inventive concept, this application also provides an intelligent verification device for form images corresponding to power industry expansion applications, comprising: a processing module configured to perform image enhancement processing on the form image in response to receiving a form image; a recognition module configured to recognize the image-enhanced form image using a pre-built power character recognition model to obtain field data; a semantic correction module configured to perform semantic correction on the field data using a pre-built power knowledge graph to obtain corrected field data; a verification module configured to perform dynamic rule verification on the corrected field data to obtain at least one first suspicious data, and perform static rule verification on the corrected field data to obtain at least one second suspicious data, and regard both the first suspicious data and the second suspicious data as target suspicious data; and a determination module configured to determine the similarity between the target suspicious data and its corresponding predetermined field data for each target suspicious data, and determine the target suspicious data as abnormal field data in response to determining that the similarity is less than the predetermined similarity.

[0012] Based on the same inventive concept, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the processor implements the method described above when executing the computer program.

[0013] Based on the same inventive concept, this application also provides a non-transitory computer-readable storage medium that stores computer instructions for causing a computer to perform the method described above.

[0014] As can be seen from the above, the intelligent verification method and related equipment for the form image corresponding to the power industry expansion application provided in this application include: in response to receiving a form image, performing image enhancement processing on the form image; using a pre-built power character recognition model to recognize the image-enhanced form image to obtain field data; using a pre-built power knowledge graph to perform semantic correction on the field data to obtain corrected field data; performing dynamic rule verification on the corrected field data to obtain at least one first suspicious data, and performing static rule verification on the corrected field data to obtain at least one second suspicious data, and taking both the first suspicious data and the second suspicious data as target suspicious data; for each target suspicious data, determining the similarity between the target suspicious data and its corresponding predetermined field data, and in response to determining that the similarity is less than the predetermined similarity, determining the target suspicious data as abnormal field data. Attached Figure Description

[0015] 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.

[0016] Figure 1 This is a flowchart illustrating the intelligent verification method for the form image corresponding to the power industry expansion application in this embodiment of the application.

[0017] Figure 2 This is a flowchart illustrating an intelligent verification method for form images corresponding to power industry expansion application, as shown in another embodiment of this application. Figure 3 This is a schematic diagram of the intelligent verification device for the form image corresponding to the power industry expansion application in this embodiment of the application. Figure 4 This is a schematic diagram of the hardware structure of an electronic device 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] As described in the background section, with the continuous acceleration of digital informatization and the ongoing improvement of the level of lean management in power services, more stringent and detailed requirements have been placed on document management, document flow efficiency, and rapid sharing of document information. Whether in online or offline electricity business processing, or in various stages of on-site operations, users need to fill out various necessary forms, and the information in these forms must be properly stored in the system. Business expansion and connection application, as a core business link in power marketing, involves the processing of diverse forms such as low-voltage residential electricity, high-voltage customer electricity, on-site survey forms, and power supply plan responses, as well as the verification and entry of supporting materials. Its form processing efficiency and data quality directly affect the smoothness of business processing and customer service experience. Specifically, business expansion and connection application refers to the general business process of power supply companies accepting customer electricity applications, handling power supply and electricity-related procedures based on customer needs and power supply conditions, and completing the entire process of grid connection.

[0021] Currently, data entry and verification of business expansion application forms mainly rely on manual work. This manual entry process is prone to errors, resulting in low efficiency and inconsistent accuracy. Furthermore, during the subsequent verification of the resulting form images, it is difficult for humans to detect all errors, leading to poor accuracy in manual verification. This process is not only cumbersome and time-consuming, but also lacks verification precision.

[0022] Existing technologies attempt to use software to replace manual form image validation; however, the following problems exist: (i) Data quality is difficult to guarantee: On the one hand, existing technologies mostly use general OCR (Optical Character Recognition) models, which are not optimized for the characteristics of power scenarios such as professional terminology, multiple fields in forms, and handwritten annotations. They also do not integrate multimodal features such as text, layout, and visual elements, resulting in low recognition accuracy for key fields such as customer number and electricity capacity. On the other hand, the verification rules are limited to basic levels such as field non-empty and fixed format, lacking business logic verification such as capacity-voltage matching and industry-electricity characteristic correlation, resulting in insufficient anomaly detection rate. Low-precision recognition and verification results not only cannot directly support the automatic filling requirements of the system's business processing interface, but also easily lead to subsequent business data chaos, affecting the standardization of file management and increasing the cost of subsequent data traceability and correction.

[0023] (ii) Low standardization and lack of closed-loop optimization capabilities: Existing technologies have not established a unified verification rule system, and the verification standards of different power supply stations vary, resulting in the recurrence of similar errors in different areas, which seriously affects the consistency of business processing. At the same time, the correct data after manual correction is not fed back to the recognition model or rule base for iterative optimization, resulting in weak model generalization ability, difficulty in adapting to the dynamic changes in business scenarios, and inability to form a closed-loop process of "recognition-verification-optimization".

[0024] (III) Imbalance between processing efficiency and cost control: The existing technology relies on manual completion of the entire process of form entry and review. At the same time, the form recognition and tamper detection functions are separate and lack an integrated processing architecture, further lengthening the process. This model not only cannot adapt to the daily business processing needs of 100,000+ under the large-scale development of power marketing, but also requires a large investment of manpower to maintain operation, while increasing the burden on front-line staff, becoming the core bottleneck restricting the improvement of back-end service response speed.

[0025] The aforementioned problems not only lead to low efficiency in validating form images, but also result in poor accuracy in validating form images.

[0026] In view of this, embodiments of this application propose an intelligent verification method for form images corresponding to power industry expansion applications, referring to... Figure 1 This includes the following steps: Step 101: In response to receiving the form image, perform image enhancement processing on the form image.

[0027] In this step, this application focuses on the automated and high-precision verification of business expansion application forms. It proposes an innovative intelligent verification method for business expansion application forms. Relying on multimodal OCR recognition technology, it achieves accurate extraction of key fields in the form. Combining power knowledge graph, it completes semantic-level standardization and correction. Through a streaming comparison engine and a multi-level verification rule system, it achieves real-time intelligent verification of extracted information and original system data. Finally, it continuously optimizes the model and rules with dynamic error analysis and feedback closed-loop mechanism, forming a full-process intelligent verification logic of "collection and preprocessing - identification and correction - real-time verification - optimization and iteration".

[0028] First, upon receiving the form image, to ensure accurate verification later, image enhancement processing is performed on the form image to improve its image quality. The form image is a power-related form; for example, it is a business expansion application form. This original image of the business expansion application form is acquired through image acquisition equipment. The business expansion application form includes forms such as: Low-voltage residential electricity registration form, Low-voltage non-residential electricity registration form, Low-voltage bulk electricity registration form, High-voltage customer electricity registration form, High-voltage site survey form, Low-voltage site survey form, High-voltage power supply scheme response form, Customer power receiving project design document submission form, Customer power receiving project design document review opinion form, Customer power receiving project design change application contact form, Customer power receiving project intermediate inspection report form, Customer power receiving project intermediate inspection opinion form, Customer power receiving project new equipment addition to system application form, and New installation (capacity increase) power supply form, etc. To ensure the validity of form images, the form image acquisition device should be properly configured. For example, a CCD (Charge-Coupled Device) sensor scanner with a resolution of 300-600 DPI should be selected, along with an adaptive light source adjustment of 200-1000 lux, which can effectively suppress reflections and shadow interference.

[0029] Step 102: Using a pre-built power character recognition model, the image of the enhanced form is recognized to obtain field data.

[0030] In this step, the form image needs to be recognized in order to identify the data in the form image. In existing technologies, a general character recognition model is used to recognize form images, but this model has poor accuracy in recognizing form images from the power industry. This application utilizes a pre-built power character recognition model to recognize the field data from the image-enhanced form image. The power character recognition model is trained based on historical form images from the power industry. For example, a dedicated OCR recognition model is constructed based on the pre-trained LayoutLMv3 multimodal model (a pre-trained multimodal document understanding model based on the Transformer architecture) for business expansion application forms (including low-voltage / high-voltage electricity registration forms, site survey forms, power supply scheme response forms, etc.). LayoutLMv3 achieves accurate recognition of form images by jointly modeling text, layout, and vision modalities. The power character recognition model in this application is a multimodal feature-optimized OCR recognition model.

[0031] The image-enhanced form image is input into the power character recognition model. The CNN (Convolutional Neural Network) feature extraction network with spatial attention mechanism in the power character recognition model extracts features from the image-enhanced form image, obtaining the corresponding visual features, text layout features, and preliminary text features. Multi-scale grid fusion is applied to the visual features, text layout features, and preliminary text features to enhance the edge detection capability for blurred text. For example, multi-scale fusion using 16-pixel × 16-pixel, 32-pixel × 32-pixel, and 64-pixel × 64-pixel grids enhances the edge detection capability for blurred text at 200-300 DPI. Simultaneously, for handwritten annotations in the form (such as surveyor signatures and modification opinions), a bidirectional LSTM-CTC (Long Short-Term Memory Network-Connected Time Classifier) ​​temporal modeling framework is added, introducing features such as pen pressure change rate to improve recognition accuracy. Finally, the power character recognition model outputs structured field information. This application utilizes the power character recognition model to recognize image-enhanced form images, ensuring the accuracy of character recognition in form images.

[0032] Step 103: Use the pre-built power knowledge graph to perform semantic correction on the field data to obtain corrected field data.

[0033] In this step, to ensure consistency in subsequent image verification and to ensure standardized management of the verified form images, a pre-built power knowledge graph is used to perform semantic correction on the field data, resulting in corrected field data. Based on the semantic correction of the power knowledge graph, the extracted field data is matched and corrected with the pre-built power knowledge graph. The power knowledge graph is dynamically updated, ensuring that accurate semantic correction of the field data is always possible. For example, the power knowledge graph contains 32,000 power administrative terms (covering organization names, voltage levels, coding rules, address standard formats, etc.) and 18,000 power industry-specific terms (covering electricity consumption types, equipment models, industry classifications, etc.).

[0034] By integrating multimodal OCR with a power knowledge graph, high-precision recognition and semantic correction of key fields in business expansion application forms were achieved, solving the problems of insufficient recognition accuracy and numerous non-standard expressions in traditional OCR. Experiments have shown that the recognition accuracy of key fields in the data of this application all reached over 97%.

[0035] Step 104: Perform dynamic rule verification on the corrected field data to obtain at least one first suspicious data, and perform static rule verification on the corrected field data to obtain at least one second suspicious data, and take both the first suspicious data and the second suspicious data as target suspicious data.

[0036] In this step, the existing verification rules are limited to basic levels such as field non-empty and fixed format, resulting in poor verification accuracy. This application, however, verifies the corrected field data using both dynamic and static rules, obtaining at least one first suspicious data point after dynamic rule verification, and at least one second suspicious data point after dynamic rule verification. Both the first and second suspicious data points are considered target suspicious data, which are suspected abnormal field data that, after static or dynamic rule verification, are determined to require similarity assessment. Static rules include multiple pre-set verification rules, while dynamic rules are dynamically changing verification rules obtained by mining all historical abnormal field data. By performing dynamic rule verification on the corrected field data, the recurrence of similar errors in different areas is avoided, ensuring consistency in subsequent business processing.

[0037] By applying both dynamic and static rule validation to the corrected field data, not only are validation standards for different business processes unified, but the goal of comprehensive validation of the corrected field data is also achieved. The multi-layered validation mechanism in this application ensures both the efficiency of basic validation and improves validation accuracy in complex scenarios. The combination of the multi-layered validation mechanism and the streaming comparison engine enables real-time and comprehensive form verification. Experiments have shown that the anomaly marking accuracy reaches over 95%, and the verification efficiency is more than 50 times higher than manual verification, significantly reducing business processing time.

[0038] Step 105: For each target suspicious data, determine the similarity between the target suspicious data and its corresponding predetermined field data; in response to determining that the similarity is less than the predetermined similarity, determine the target suspicious data as abnormal field data.

[0039] In this step, for each target suspicious data, the similarity between the target suspicious data and its corresponding predetermined field data is determined. The predetermined field data corresponding to the target suspicious data are standard data pre-set and stored in the system, determined based on historical experience. Since the target suspicious data is only data deemed potentially abnormal after preliminary static or dynamic rule verification, false positives are possible. The predetermined field data corresponding to the target suspicious data, however, are standard data that conforms to business specifications and is considered correct. By calculating the similarity between the target suspicious data and its corresponding predetermined field data, the degree of closeness between the two can be quantified. It should be noted that the algorithm used to determine the similarity between the target suspicious data and its corresponding predetermined field data is a dedicated power similarity algorithm, ensuring the reliability of the obtained similarity scores.

[0040] If the similarity is less than a predetermined similarity, the target suspicious data is identified as anomalous field data. The predetermined similarity is determined based on historical experience. When the similarity is less than the predetermined similarity, it indicates that the target suspicious data differs significantly from the standard data, exceeding the normal range. This data can be determined as not meeting business requirements and thus identified as anomalous field data. This effectively improves the accuracy and reliability of anomalous data detection, reduces false positives and false negatives, and provides more accurate data support for subsequent data processing and business decisions.

[0041] This application proposes an intelligent verification method and system based on multimodal OCR and power knowledge graph to optimize business processing efficiency. It optimizes the OCR recognition model through multimodal features, achieves semantic standardization and correction by combining a power-specific knowledge graph, and builds a multi-level intelligent verification rule engine. This achieves end-to-end consistency and business semantics in form processing, from automatic filling of interface information to business logic verification and data access and storage. Simultaneously, a closed-loop optimization mechanism improves automation and reduces manual intervention costs, while standardized solutions enhance collaborative efficiency, ensuring that technology and business needs are aligned. Ultimately, this improves business processing efficiency and data quality. This application relates to the field of digital technology in power marketing, specifically to business expansion application form processing technology, involving deep correlation image processing, optical character recognition, knowledge graph construction, and rule engine development. It constructs a complete end-to-end intelligent form processing solution capable of automated collection, high-precision recognition, intelligent verification, and secure storage of customer forms, comprehensively improving the efficiency and data quality of business expansion application processing, thereby providing strong technical support and assurance for the business development and customer service of power companies.

[0042] The above scheme involves the following steps: Upon receiving a form image, image enhancement processing is performed on the form image; a pre-built power character recognition model is used to recognize the enhanced form image to obtain field data; a pre-built power knowledge graph is used to perform semantic correction on the field data to obtain corrected field data; dynamic rule verification is performed on the corrected field data to obtain at least one first suspicious data, and static rule verification is performed on the corrected field data to obtain at least one second suspicious data; both the first and second suspicious data are designated as target suspicious data; for each target suspicious data, the similarity between the target suspicious data and its corresponding predetermined field data is determined; and upon determining that the similarity is less than the predetermined similarity, the target suspicious data is identified as abnormal field data.

[0043] In some embodiments, performing dynamic rule verification on the corrected field data to obtain at least one first suspicious data includes: obtaining at least one currently stored derivative rule, wherein each derivative rule is determined by mining multiple historical abnormal field data; and verifying the corrected field data using each derivative rule to obtain the at least one first suspicious data.

[0044] In this embodiment, the system continuously analyzes all historical abnormal field data, uses association rule mining algorithms to automatically discover frequently occurring error patterns, and generates and stores at least one derived rule. For example, if historical abnormal field data frequently shows "address format mismatch" and is also accompanied by "invalid GIS coordinates," a combined verification rule for the aforementioned anomalies is automatically generated. Furthermore, external data fusion verification is performed; for example, the GIS geographic information system interface is called in real time to verify the validity and coordinates of the "electricity address"; the marketing system interface is called to obtain the customer's historical electricity consumption data to verify the rationality of the electricity capacity filled in the current form compared to historical capacity.

[0045] When performing dynamic rule validation on the corrected field data, at least one derived rule is retrieved from the current storage. Each derived rule is determined by mining multiple historical abnormal field data, and the currently stored at least one derived rule is updated in real time. The corrected field data is then validated using each derived rule to obtain at least one first suspicious data point. Through dynamic error analysis and a feedback closed-loop optimization mechanism, continuous iterative optimization of the model and rules is achieved, enabling the system to adapt to the dynamic needs of form type updates and business rule adjustments, thus improving the system's robustness and applicability.

[0046] In some embodiments, the at least one second suspicious data includes at least one first sub-suspicious data, at least one second sub-suspicious data, and at least one third sub-suspicious data; the step of performing static rule validation on the corrected field data to obtain at least one second suspicious data includes: obtaining historical stored data corresponding to the form image; comparing the corrected field data with the historical stored data to obtain the at least one first sub-suspicious data; checking the corrected field data according to predetermined form filling rules to obtain the at least one second sub-suspicious data; and checking the corrected field data according to predetermined business filling rules to obtain the at least one third sub-suspicious data.

[0047] In this embodiment, the corrected field data is compared in real time with its corresponding historical stored data to obtain at least one first sub-suspicious data. For example, an Apache Flink (an open-source stream processing framework for stateful computation of bounded and unbounded data streams) is used to build a streaming comparison engine, which compares the corrected field data with its corresponding original customer files, historical work orders, and other data in the system in real time. The corrected field data is checked according to predetermined form filling rules to obtain at least one second sub-suspicious data. These predetermined form filling rules are basic rules used to verify the completeness (e.g., customer number is not empty) and format compliance (e.g., whether the voltage level is an enumerated value such as "10kV" or "380V"). The corrected field data is then checked according to predetermined business filling rules to obtain at least one third sub-suspicious data. The rules for filling in pre-ordered business data are used to validate the business logic between fields. For example, they validate whether a customer with a "voltage level" of "10kV" has a "power capacity" greater than or equal to 100kVA; whether a customer with an "industry classification" of "catering" applies for electricity during peak dining hours; and whether the associated "power supply plan response form" for a "new installation and delivery order" is signed and the "inspection opinion form" is "qualified". By performing static rule validation on the corrected field data, on the one hand, the basic quality of the data can be guaranteed, ensuring that key fields are not missing and that the data format conforms to the specifications, avoiding the impact of data problems on subsequent business processes and decision-making accuracy; on the other hand, it can improve data processing efficiency, quickly filter invalid data, and reduce the amount of data to be processed later.

[0048] Static and dynamic rules constitute a multi-level verification rule system, including a historical rule layer, a basic rule layer, and a business rule layer corresponding to static rules, as well as a dynamic rule layer corresponding to dynamic rules, completing the multi-level rule engine verification. This application constructs a real-time verification engine based on the Apache Flink streaming computing framework (a distributed streaming data platform for building real-time data pipelines and streaming applications) and innovatively integrates a multi-level rule verification system. This architecture can receive OCR recognition results in milliseconds and simultaneously perform real-time data comparison with marketing business systems and GIS systems (Geographic Information Systems). The multi-level rule engine covers historical rule verification, basic format verification, cross-field business logic verification (such as capacity-voltage matching), and dynamic risk warning based on historical error pattern mining, realizing the leap from "data comparison" to "business compliance review," transforming the verification work from lagging manual spot checks to real-time online automated risk control.

[0049] In some embodiments, determining the similarity between the target suspicious data and its corresponding predetermined field data includes: determining the character similarity, pinyin similarity, abbreviation similarity, and semantic segmentation similarity between the target suspicious data and its corresponding predetermined field data; and performing a weighted calculation on the character similarity, the pinyin similarity, the abbreviation similarity, and the semantic segmentation similarity to obtain the similarity.

[0050] In this embodiment, key text fields such as electricity address and customer name are neither enumerated fields nor involved in logical checks, yet their consistency needs to be compared in actual business operations. In such cases, simply using a simple string exact match method is often insufficient to meet practical requirements. This application proposes a multi-dimensional feature-integrated power-specific text similarity algorithm. The core idea of ​​this algorithm is: for two text strings T to be compared... form (Target suspicious data) and T system (Pre-defined field data), its overall similarity S total It is a weighted harmonic sum of the similarities of multiple features. The comprehensive similarity can be defined using the following formula: Stotal(T form T system )=ω1×S char +ω2×S pinyin +ω3×S abbr +ω4×S semantic Formula 1 Among them, Total(T) form T system The similarity (ω) represents the similarity between the target suspicious data and its corresponding predetermined field data, where ω1 + ω2 + ω3 + ω4 = 1. ω1 is the weight corresponding to character similarity, ω2 is the weight corresponding to pinyin similarity, ω3 is the weight corresponding to abbreviation similarity, and ω4 is the weight corresponding to semantic segmentation similarity. These weights can be adjusted based on field characteristics and historical data training. char S represents character similarity. pinyin S represents the similarity of pinyin. abbr S represents the similarity of abbreviations. semantic This represents the semantic segment similarity.

[0051] Determine character similarity S char The edit distance (Levenshtein Distance) is normalized using dynamic programming and calculated using the following formula: Formula 2 in, This indicates that the similarity metric between the target suspicious data and its corresponding predetermined field data is measured using edit distance based on dynamic programming, and max means to maximize.

[0052] Determining character similarity is used to handle character-level differences such as typos, omissions, and extra characters. Lower replacement costs are set for key power characters (such as "phase" and "line", "KW" and "kW") to improve fault tolerance.

[0053] Determine the similarity S between pinyin pinyin The Chinese string needs to be converted into a full Pinyin sequence P first. form and P system Then, their longest common subsequence (LCS) similarity is calculated using the following formula: Formula 3 Among them, P form P represents the full sequence of the target suspicious data. system This represents the full sequence of the predefined field data.

[0054] Determining pinyin similarity helps address homophones and writing differences caused by dialect accents. An integrated library of commonly used pinyin abbreviations for power grid place names (e.g., "SL" for "Victory" and "People") enhances robustness.

[0055] Determine the abbreviation similarity S abbr Based on a pre-built "Mapping Table of Commonly Used Abbreviations in the Power Industry", the abbreviations and full names in the strings are bidirectionally standardized before the character similarity S´ is calculated. char The calculation is performed using the following formula: Formula 4 in, This indicates bidirectional standardization of abbreviations and full names in a string.

[0056] Determining abbreviation similarity is used to handle abbreviations that are commonly used in addresses (such as “Limited Company” abbreviated as “Ltd.” or “Shares” abbreviated as “Co.”).

[0057] Determine the semantic segment similarity S semantic Addresses are segmented into ordered semantic units using an address resolution model (or rules) (e.g., province, city, district, road / street, number, building, room). Then, unit matching is performed on each semantic unit, using the aforementioned S... char S pinyin Calculate the similarity between units, allowing for a certain degree of order tolerance (e.g., "18th Floor of XX Building" vs. "18th Floor of XX Building"). Finally, assign different weights to different units based on their importance (e.g., "street name" has a higher weight than "room number"), and perform a weighted average, calculated using the following formula: Formula 5 Where n refers to the total number of semantic units, representing the number of segments in the standard address structure, typically 7 segments (province, city, district, street, number, building, room), β i The weight of the i-th semantic unit is given by β. i ∈[0,1]; For the i-th semantic unit corresponding to the target suspicious data, For the i-th semantic unit corresponding to the predefined field data, matchScore is the unit matching score function that calculates the similarity between two semantic units.

[0058] Based on preset thresholds and rules, anomalies are automatically marked. A similarity threshold θ is set for each key field. field If S total <θ field If the difference is not found, it is considered to be a discrepancy. The power-specific text similarity algorithm not only outputs a comprehensive similarity score but also sub-scores for each dimension, providing accurate attribution for anomaly labeling. For example, if S... char Very low and S pinyin High, can be marked as "suspected homonym misspelling"; if S abbr If it's very low, it can be marked as "abbreviation does not match the full name"; if S semantic If a display unit is missing or severely misaligned, it can be marked as "incomplete or disordered address structure".

[0059] It should be noted that after identifying the suspicious target data as anomalous field data, the recognition confidence level C corresponding to the power character recognition model can also be obtained. ocr Based on the identification confidence level C ocr And similarity score S total It determines whether abnormal field data requires manual review. Upon determining that abnormal field data requires manual review, it generates and sends the corresponding review information for the abnormal field data. For example, even if S... total Slightly above the threshold, but C ocr Even with extremely low confidence levels, an anomaly flag of "low confidence identification result, manual review recommended" may still be triggered. All verification and comparison results (including pass, anomaly flags, and details) are output to the downstream data stream in real time. This data stream is pushed to the front-end business interface in real time to remind staff to handle the issue; on the other hand, as a high-quality structured data source, it is persistently stored in the database of the dynamic error analysis platform to support subsequent analysis and optimization.

[0060] To address the challenge of fuzzy text matching, this application proposes a composite similarity algorithm specifically designed for the power industry. This algorithm does not rely on a single string comparison but comprehensively calculates features across four dimensions: character edit distance, pinyin similarity, common abbreviation matching, and semantic segmentation structure similarity. The final similarity score is obtained through weighted summation. This algorithm is specifically optimized for scenarios involving homophones in power addresses and customer names, simplified and traditional characters, and the mixing of abbreviations and full names, significantly improving the accuracy and interpretability of difference judgment in real-world business environments. This application introduces a multi-dimensional feature fusion verification logic and innovatively constructs a dedicated similarity algorithm adapted to power industry expansion and installation scenarios. Combined with a multi-level verification rule system and the Apache Flink streaming engine, it achieves real-time, accurate comparison and anomaly marking between form information and marketing system data.

[0061] In some embodiments, the step of using a pre-constructed power knowledge graph to perform semantic correction on the field data to obtain corrected field data includes: extracting key fields from the field data to obtain multiple key field data; for each key field data, using the power knowledge graph to perform unit standardization, address normalization, industry classification mapping, and semantic error correction on the key field data to obtain at least one sub-corrected field data; and determining at least one sub-corrected field data corresponding to all key field data as the corrected field data.

[0062] In this embodiment, key fields are extracted from the field data to obtain multiple key field data. Key business fields are extracted from the field data, including but not limited to: customer number, electricity address, electricity capacity, voltage level, industry classification, and equipment model. Simultaneously, auxiliary fields such as business type and application date are extracted to form a structured raw field set, providing a data foundation for subsequent semantic correction. The key field data is matched and corrected with a pre-constructed power knowledge graph, which is dynamically updated. A graph neural network is used to implement terminology association reasoning, and the power knowledge graph is used to perform semantic correction on the extracted key fields. For each key field data, the power knowledge graph is used to standardize units, normalize addresses, map industry classifications, and correct semantic errors, resulting in at least one sub-corrected field data. The at least one sub-corrected field data corresponding to all key field data is determined as the corrected field data, specifically including: Standardize key field data by unit: Using the capacity unit standard attribute in the power knowledge graph, correct non-standard units in the identification results to standard units, such as correcting "kva" to "kVA" and "kw" to "kW". Standardize key field data by address: Based on the address standard format in the knowledge graph, standardize the identified electricity addresses, such as supplementing "No. 1, XX Road" to "No. 001, XX Road", and restoring abbreviations in addresses. Map key field data to industry categories: Based on the industry entity standard names in the power knowledge graph, map the identified non-standard industry names to standard categories, such as mapping "catering services" to "catering industry". Correct key field data for semantic errors: Using entity relationships in the power knowledge graph, correct semantically contradictory identification results. For example, when identifying "voltage level = 220V" and "electricity capacity = 200kVA", based on the association rule in the power knowledge graph that "220V voltage level corresponds to capacity ≤ 100kVA", the identification result is determined to have a semantic contradiction, triggering a re-identification process.

[0063] This application creatively combines LayoutLMv3-based multimodal OCR technology with a power industry knowledge graph, forming a unique semantic correction mechanism. After the OCR recognizes the field data in the form, the system automatically calls the standard entities, attributes, and association rules in the power knowledge graph to perform unit standardization (e.g., "kva" → "kVA"), address normalization (e.g., "XX Road No. 1" → "XX Road No. 001"), and professional terminology mapping (e.g., "catering" → "catering industry"). This collaboration not only improves the accuracy of character recognition but also realizes intelligent "meaning recognition," upgrading the traditional OCR literal recognition to semantic understanding oriented towards power business.

[0064] In some embodiments, the image enhancement processing of the form image includes: sequentially performing multi-level filtering processing, tilt correction processing, crease repair processing, background interference removal processing, illumination interference removal processing, and character recognition enhancement processing on the form image to obtain a form image after image enhancement processing.

[0065] In this embodiment, the form image undergoes a step-by-step preprocessing process, forming a complete workflow of "noise removal - tilt correction - crease repair - background separation - character enhancement - accurate recognition," specifically including: (1) Multi-level filtering: First, the salt-and-pepper noise generated during scanning or shooting is eliminated by median filtering of a predetermined value kernel. For example, the predetermined value is 5×5. Then, Gaussian noise is processed by dynamic parameter Gaussian filtering (σ=0.5~1.2). The filtering kernel parameters are adaptively adjusted according to the image noise density to improve the signal-to-noise ratio by 8.7 dB, eliminate noise in the form image, and provide a clear image basis for subsequent processing.

[0066] (2) Tilt correction processing: Automatic image tilt correction within a predetermined angle range is achieved by Radon transform (an integral transform method). For example, the predetermined range is ±45°. Radon transform maps the image to polar coordinate space through line integral projection. Its mathematical expression is R(ρ, θ) = ∫ f(x, y) ds, Formula 6, where ρ is the distance from the line to the origin and θ is the projection angle. The tilt error of the corrected image is ≤0.5°, ensuring the horizontality of the form text lines and avoiding character cutting deviation during subsequent recognition.

[0067] (3) Crease repair processing: For creased paper forms, a directional diffusion algorithm is introduced to enhance the effective signal in the crease area. By analyzing the texture direction of the crease, diffusion filtering is performed along the direction perpendicular to the crease to weaken the crease shadow on the characters. Experiments have shown that the character recognition rate in the crease area is increased to 91.2%.

[0068] (4) Background interference removal: The foreground extraction module constructed using Generative Adversarial Network (GAN) separates the form foreground (text, table lines) from the complex background (desktop, paper texture). At the same time, a background suppression module is added to eliminate gradient background interference. This module extracts the brightness channel through adaptive color space conversion to perform background color uniformization processing, and then suppresses the grayscale fluctuation of the gradient background by constraining the pixel grayscale value range. Experiments have shown that this improves the contrast between the foreground characters and the background by more than 30%.

[0069] (5) Illumination Interference Processing: The Niblack adaptive thresholding algorithm is used to perform binarization segmentation on the preprocessed image. The Niblack adaptive thresholding algorithm is a local adaptive image binarization algorithm that dynamically calculates the threshold based on the mean and standard deviation of the local neighborhood of each pixel, and it has good effects on images with uneven illumination and low contrast. The local threshold is dynamically calculated through a sliding window. The core formula for threshold calculation is: T(x,y)=m(x,y) Formula 7, k·s(x,y), where m(x,y) is the mean grayscale value within the local neighborhood window, s(x,y) is the standard deviation of grayscale values ​​within the neighborhood window, and k is an adjustment coefficient ranging from 0.1 to 0.3, adaptively optimized according to the form paper type. This algorithm can accurately distinguish characters from the background under different lighting conditions, avoiding the problems of character loss in dark areas or misjudgment of background in bright areas caused by fixed threshold segmentation.

[0070] (6) Character recognition enhancement processing: A hybrid recognition model of Niblack adaptive threshold and ResNet-50 is constructed. The binarized character region image is input into the ResNet-50 (50-layer residual network) deep convolutional neural network. The residual connection structure of this network solves the problem of gradient vanishing during deep network training and enhances the feature extraction capability of blurred and slightly deformed characters. The output layer of the model uses the Softmax function to achieve character classification. It has efficient recognition capability for printed characters and handwritten annotation characters (such as survey signatures) in forms. Experiments have shown that the character-level recognition accuracy reaches 99.1%.

[0071] To address the diverse challenges of form images in real-world business scenarios, this application integrates multiple image preprocessing techniques to form a highly efficient preprocessing pipeline. It employs nonlocal mean denoising to suppress Gaussian noise, utilizes Radon transform for tilt correction, and innovatively applies a GAN (Generative Adversarial Network)-based background separation network to effectively solve interference problems such as blurred scanned documents, misaligned placement, complex background textures, or stamp coverage. This combined technology significantly improves the input image quality for subsequent OCR recognition modules, laying a solid foundation for high-precision recognition. After the above-described full-process preprocessing, a form image with a clear foreground and complete characters is obtained, providing high-quality input data for subsequent multimodal OCR recognition. Experiments have shown that the preprocessing time per page is ≤120 ms, meeting the high-efficiency processing requirements for batch forms.

[0072] In some embodiments, after identifying the target suspicious data as abnormal field data, the method includes: generating and sending a verification report corresponding to the form image based on at least one abnormal field data and the corrected field data.

[0073] In this embodiment, a verification report corresponding to the form image is generated and sent based on at least one abnormal field data and corrected field data. For example, the verification report includes at least the abnormal field data corresponding to the form image, the position of the abnormal field data in the corrected field data, and the reason for the abnormality corresponding to the abnormal field data. It should be noted that after the form image is verified and no abnormal field data is found, a verification pass report corresponding to the form image is generated and sent. For forms that pass verification, they are automatically entered into the system to generate electronic files; for forms that fail verification, they are pushed to the business personnel's processing interface, displaying the abnormality type, abnormal field, and related evidence (original image, recognition result, original data). After the business personnel complete the manual correction, they feed the corrected data back to the feedback loop mechanism of the previous step. At the same time, a form verification quality report is generated, including indicators such as the verification pass rate, error rate, and abnormality type distribution of each power supply station, providing decision support for power marketing management. This application realizes full-process automation of business expansion application form verification, reduces manual intervention, lowers the human error rate, and the generated quality report provides data support for power marketing management, improving the scientific nature of management decisions.

[0074] The system in this application automatically diagnoses the root causes of errors (such as image quality, model defects, or rule loopholes) by collecting end-to-end data from image input to final verification and utilizing clustering analysis and association rule mining techniques. The analysis results provide management insights in the form of quality heatmaps, and automatically form a feedback loop: converting manually verified correct results into training samples to drive incremental learning of the OCR model, optimize verification rule thresholds, and enrich the knowledge graph content, enabling the system to self-iterate and become increasingly intelligent with use.

[0075] In another embodiment provided in this application, after generating and sending the verification report corresponding to the form image, (1) all verification reports and all verification pass reports are subjected to error data aggregation analysis, and the identification and verification results of all forms are collected and aggregated and statistically analyzed according to power supply station, form type, abnormal type, and time dimension. (2) Quality heat map generation and insight: based on the error analysis results, a visual quality heat map is generated to intuitively display the form filling and identification quality distribution in each area and locate high-frequency error points. (3) A closed-loop feedback optimization mechanism is established: based on the quality heat map and error classification results, the OCR recognition model and verification rules are dynamically optimized: for form types and fields with high error rates, additional labeled samples are added, and incremental training is performed on the LayoutLMv3 dedicated recognition model; for semantic correction errors, the entity attributes and association rules of the power industry knowledge graph are updated; for the weight ω of the intelligent verification similarity algorithm, the weight ω is adjusted. i and the threshold θ fieldBased on historical correct / incorrect samples, Bayesian optimization is used to automatically adjust the optimal parameters, forming a closed loop of "identification-verification-analysis-optimization" to drive continuous iteration and upgrading of the model and rules.

[0076] In another embodiment of this application, such as Figure 2 As shown, the process includes: Form Acquisition Preprocessing: Acquiring the original image of the business expansion application form, performing noise reduction, tilt correction, and background separation preprocessing on the original image, and outputting a clear form image. Multimodal OCR Recognition and Semantic Correction: Implementing multimodal recognition and semantic correction of form information, including LayoutLMv3 multimodal recognition and knowledge graph semantic correction. Streaming Intelligent Verification: Real-time comparison and intelligent verification of form information and original data, including three-level verification rules, Apache Flink streaming comparison, and similarity calculation. Dynamic Error Analysis and Optimization: Implementing dynamic analysis of verification errors and optimization of models and rules, including error analysis units and feedback loops. Result Output and Anomaly Handling: Outputting verification results, processing abnormal forms, and generating quality reports.

[0077] 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.

[0078] 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.

[0079] Based on the same inventive concept, and corresponding to any of the above embodiments, this application also provides an intelligent verification device for form images corresponding to power industry expansion applications.

[0080] refer to Figure 3 The intelligent verification device for the form image corresponding to the power industry expansion application includes: The processing module 10 is configured to perform image enhancement processing on the form image in response to receiving the form image.

[0081] The recognition module 20 is configured to use a pre-built power character recognition model to recognize the image-enhanced form image and obtain field data.

[0082] The semantic correction module 30 is configured to perform semantic correction on the field data using a pre-built power knowledge graph to obtain corrected field data.

[0083] The verification module 40 is configured to perform dynamic rule verification on the modified field data to obtain at least one first suspicious data, and to perform static rule verification on the modified field data to obtain at least one second suspicious data, and to regard both the first suspicious data and the second suspicious data as target suspicious data.

[0084] The determination module 50 is configured to determine the similarity between each target suspicious data and its corresponding predetermined field data, and in response to determining that the similarity is less than the predetermined similarity, determine the target suspicious data as abnormal field data.

[0085] The aforementioned apparatus, in response to receiving a form image, performs image enhancement processing on the form image; uses a pre-built power character recognition model to recognize the enhanced form image and obtain field data; uses a pre-built power knowledge graph to perform semantic correction on the field data and obtain corrected field data; performs dynamic rule verification on the corrected field data to obtain at least one first suspicious data, and performs static rule verification on the corrected field data to obtain at least one second suspicious data, and designates both the first suspicious data and the second suspicious data as target suspicious data; for each target suspicious data, determines the similarity between the target suspicious data and its corresponding predetermined field data, and in response to determining that the similarity is less than the predetermined similarity, identifies the target suspicious data as abnormal field data.

[0086] In some embodiments, the verification module 40 is further configured to obtain at least one derived rule currently stored, wherein each derived rule is determined by mining multiple historical abnormal field data; and the corrected field data is verified using each derived rule to obtain the at least one first suspicious data.

[0087] In some embodiments, the at least one second suspicious data includes at least one first sub-suspicious data, at least one second sub-suspicious data, and at least one third sub-suspicious data; the verification module 40 is further configured to acquire historical stored data corresponding to the form image, compare the corrected field data with the historical stored data to obtain the at least one first sub-suspicious data; check the corrected field data according to predetermined form filling rules to obtain the at least one second sub-suspicious data; and check the corrected field data according to predetermined business filling rules to obtain the at least one third sub-suspicious data.

[0088] In some embodiments, the determining module 50 is further configured to determine the character similarity, pinyin similarity, abbreviation similarity, and semantic segmentation similarity between the target suspicious data and its corresponding predetermined field data; and to perform a weighted calculation on the character similarity, the pinyin similarity, the abbreviation similarity, and the semantic segmentation similarity to obtain the similarity.

[0089] In some embodiments, the semantic correction module 30 is further configured to extract key fields from the field data to obtain multiple key field data; for each key field data, the power knowledge graph is used to perform unit standardization, address normalization, industry classification mapping, and semantic error correction on the key field data to obtain at least one sub-corrected field data; and at least one sub-corrected field data corresponding to all key field data is determined as the corrected field data.

[0090] In some embodiments, the processing module 10 is further configured to sequentially perform multi-level filtering, tilt correction, crease repair, background interference removal, illumination interference removal, and character recognition enhancement processing on the form image to obtain a form image after image enhancement processing.

[0091] In some embodiments, a sending module is further included, which is configured to generate and send a verification report corresponding to the form image based on at least one abnormal field data and the corrected field data after the target suspicious data is determined to be abnormal field data.

[0092] For ease of description, the above devices are described in terms of function, divided into various modules. Of course, in implementing this application, the functions of each module can be implemented in one or more software and / or hardware.

[0093] The apparatus described above is used to implement the intelligent verification method for the form image corresponding to the power industry expansion application in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0094] Based on the same inventive concept, corresponding to the methods of 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. When the processor executes the program, it implements the intelligent verification method for the form image corresponding to the power industry expansion application as described in any of the above embodiments.

[0095] Figure 4 This embodiment illustrates a more specific hardware structure of an electronic device. The device 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.

[0096] 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.

[0097] 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.

[0098] The input / output interface 1030 is used to connect input / output modules to realize information input and output. The input / output modules can be configured as components in the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touch screens, microphones, various sensors, etc., and output devices may include displays, speakers, vibrators, indicator lights, etc.

[0099] 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.).

[0100] 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.

[0101] 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.

[0102] The electronic device described above is used to implement the intelligent verification method for the form image corresponding to the power industry expansion application in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0103] 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 intelligent verification method for form images corresponding to power industry expansion applications as described in any of the above embodiments.

[0104] 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 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.

[0105] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the intelligent verification method for the form image corresponding to the power industry expansion application as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0106] Based on the same concept, corresponding to the methods of any of the above embodiments, this application also provides a computer program product, including computer program instructions. When the computer program instructions are run on a computer, the computer executes the intelligent verification method for the form image corresponding to the power industry expansion application as described in any of the above embodiments, which has the beneficial effects of the corresponding method embodiments, and will not be repeated here.

[0107] It should be noted that the embodiments of this application can also be further described in the following ways: 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.

[0108] 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.

[0109] 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.

[0110] 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.

[0111] 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 is limited to these examples; under the concept 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 detail for the sake of brevity.

[0112] 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.

[0113] 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.

[0114] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. 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 intelligently verifying form images corresponding to power industry expansion applications, characterized in that, include: In response to receiving a form image, perform image enhancement processing on the form image; Using a pre-built electricity character recognition model, the enhanced form image is recognized to obtain field data; The field data is semantically corrected using a pre-built power knowledge graph to obtain corrected field data; Dynamic rule validation is performed on the corrected field data to obtain at least one first suspicious data, and static rule validation is performed on the corrected field data to obtain at least one second suspicious data, and both the first suspicious data and the second suspicious data are taken as target suspicious data; For each target suspicious data, the similarity between the target suspicious data and its corresponding predetermined field data is determined. In response to determining that the similarity is less than the predetermined similarity, the target suspicious data is identified as abnormal field data.

2. The method according to claim 1, characterized in that, The dynamic rule validation of the corrected field data, resulting in at least one first suspicious data, includes: Obtain at least one derived rule currently stored, wherein each derived rule is determined by mining multiple historical anomaly field data; Each derived rule is used to verify the data in the corrected field to obtain at least one first suspicious data.

3. The method according to claim 1, characterized in that, The at least one second suspicious data includes at least one first sub-suspicious data, at least one second sub-suspicious data, and at least one third sub-suspicious data; The static rule validation of the corrected field data, resulting in at least one second suspicious data, includes: Obtain the historical stored data corresponding to the form image, compare the corrected field data with the historical stored data, and obtain the at least one first sub-suspicious data; According to the predetermined form filling rules, the data in the correction field is checked to obtain at least one second sub-suspicious data; According to the predetermined business filling rules, the data in the correction field is checked to obtain at least one third sub-suspicious data.

4. The method according to claim 1, characterized in that, Determining the similarity between the target suspicious data and its corresponding predetermined field data includes: Determine the character similarity, pinyin similarity, abbreviation similarity, and semantic segmentation similarity between the target suspicious data and its corresponding predetermined field data; The similarity is obtained by weighting the character similarity, the pinyin similarity, the abbreviation similarity, and the semantic segmentation similarity.

5. The method according to claim 1, characterized in that, The step of using a pre-built power knowledge graph to perform semantic correction on the field data to obtain corrected field data includes: The key fields of the field data are extracted to obtain multiple key field data; For each key field data, the power knowledge graph is used to standardize the unit, normalize the address, map the industry classification, and correct semantic errors to obtain at least one sub-corrected field data. The correction field data is determined by identifying at least one sub-correction field data corresponding to all key field data.

6. The method according to claim 1, characterized in that, The image enhancement processing of the form image includes: The form image is subjected to multi-level filtering, tilt correction, crease repair, background interference removal, lighting interference removal, and character recognition enhancement processing in sequence to obtain the form image after image enhancement processing.

7. The method according to claim 1, characterized in that, After identifying the target suspicious data as anomalous field data, the method includes: Based on at least one abnormal field data and the corrected field data, a verification report corresponding to the form image is generated and sent.

8. An intelligent verification device for form images corresponding to power industry expansion applications, characterized in that, include: The processing module is configured to perform image enhancement processing on the form image in response to receiving the form image; The recognition module is configured to use a pre-built power character recognition model to recognize the image after image enhancement to obtain field data; The semantic correction module is configured to perform semantic correction on the field data using a pre-built power knowledge graph to obtain corrected field data; The verification module is configured to perform dynamic rule verification on the modified field data to obtain at least one first suspicious data, and to perform static rule verification on the modified field data to obtain at least one second suspicious data, and to take both the first suspicious data and the second suspicious data as target suspicious data. The determination module is configured to, for each target suspicious data, determine the similarity between the target suspicious data and its corresponding predetermined field data, and in response to determining that the similarity is less than the predetermined similarity, determine the target suspicious data as abnormal field data.

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

10. 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 described in any one of claims 1 to 7.

Images