Picture auditing method and device, electronic equipment and program product

Abstract

The invention provides a picture auditing method and device, electronic equipment and a program product, and belongs to the technical field of image processing. The method comprises the steps of obtaining a to-be-audited picture set; the to-be-audited picture set comprises a plurality of to-be-audited pictures and certificate labels corresponding to the to-be-audited pictures; performing scene classification prediction according to the to-be-audited picture set through the audit prediction model to obtain a scene classification result; and selecting an audit judgment model corresponding to the scene classification result to audit the plurality of to-be-audited pictures to obtain a target audit result. According to the embodiment of the invention, manual intervention and repetitive configuration work can be greatly reduced through automatic scene recognition, process dynamic configuration and accurate marking, and meanwhile, the dynamic process and rule can respond to novel risk or strategy changes more quickly, so that the efficiency and accuracy of business auditing based on pictures can be improved.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to an image auditing method, apparatus, electronic device, and program product. Background Technology

[0002] Business auditing refers to a method used by financial auditing departments to supervise and inspect the authenticity, legality, security, and effectiveness of various business operations conducted by financial institutions. It is primarily used to check whether the financial institution's business operations comply with relevant laws, regulations, regulatory requirements, and internal systems, and to identify potential risks in business processes, such as credit risk, market risk, and operational risk, thereby assessing the effectiveness of its risk control measures. In traditional real-name business auditing practices, manual verification and review of various supporting documents submitted by the applicant is mainly relied upon after the business processing is completed. However, this operating model inherently contains a series of unavoidable drawbacks. Summary of the Invention

[0003] The main objective of this application is to provide an image auditing method, apparatus, electronic device, storage medium, and program product that can improve the efficiency and accuracy of image-based business auditing.

[0004] To achieve the above objectives, one aspect of this application proposes an image verification method, the method comprising: Obtain the image set to be audited; the image set to be audited contains multiple images to be audited and the corresponding document tags for each image; The scene classification result is obtained by using the audit prediction model to perform scene classification prediction based on the set of images to be audited; The audit judgment model corresponding to the scene classification result is selected to audit the multiple images to be audited, and the target audit result is obtained.

[0005] In some embodiments, before obtaining the set of images to be audited, the method further includes: Obtain business processing data; The business processing data is integrated according to the preset audit conditions to obtain the audit list data; An external robot is used to capture the audit list data to obtain the multiple images to be audited and the corresponding document tags for each image.

[0006] In some embodiments, before the external robot is used to retrieve the audit list data to obtain the multiple images to be audited and the corresponding document tags for each image, the method further includes: Obtain the scenario information corresponding to each of the multiple audit checklist data; Based on the scenario information corresponding to each audit checklist data, priority data corresponding to each audit checklist data is generated according to the priority generation conditions. The process of using an external robot to crawl the audit list data to obtain the multiple images to be audited and the corresponding document tags for each image includes: The audit list data with the highest priority is selected as the target audit list data, and an external robot is used to crawl the target audit list data to obtain the multiple images to be audited and the corresponding document tags for each image.

[0007] In some embodiments, the step of using an external robot to crawl the audit list data to obtain the multiple images to be audited and the corresponding document tags for each image includes: The external robot identifies multiple initial business images in the target audit checklist data to obtain the identification fields corresponding to the initial business images. Select the document tag that matches the recognition field corresponding to the initial business image as the document tag corresponding to the initial business image; Data cleaning is performed on the multiple initial business images to obtain the multiple images to be audited and the corresponding document tags for each image to be audited.

[0008] In some embodiments, selecting an audit judgment model corresponding to the scene classification result to audit the multiple images to be audited, and obtaining a target audit result, includes: The completeness of the image set to be audited and the audit list data is analyzed using the audit judgment model. If it is determined that the set of images to be audited and the audit list data are incomplete, a prompt message will be output and new images to be audited will be obtained again. If it is determined that the set of images to be audited and the audit list data are complete, then the audit judgment model is used to audit each image to be audited according to the audit list data to obtain the target audit result.

[0009] In some embodiments, the step of auditing each image to be audited according to the audit checklist data using the audit judgment model to obtain the target audit result includes: The audit judgment model is used to perform image recognition on the image to be audited to obtain key image information corresponding to the image to be audited. The audit judgment model matches the audit list data with the key information of the images to be audited according to the preset audit configuration rules, thereby obtaining the matching degree data for each image to be audited. The target audit result is determined based on the matching degree data corresponding to each image to be audited.

[0010] In some embodiments, after selecting an audit judgment model corresponding to the scene classification result to audit the multiple images to be audited and obtaining the target audit result, the method further includes: Obtain the manual judgment result corresponding to the first set of images to be audited; the first set of images to be audited can be any set of images to be audited; If the manual judgment result indicates that the target audit result corresponding to the first set of images to be audited is correct, then the first set of images to be audited and the target audit result corresponding to the first set of images to be audited are input into the audit judgment model to feed back into the audit judgment model for training.

[0011] To achieve the above objectives, another aspect of this application provides an image verification device, the device comprising: The image acquisition module is used to acquire a set of images to be audited; the set of images to be audited includes multiple images to be audited and the corresponding document tags for each image. The scene prediction module is used to perform scene classification prediction based on the set of images to be audited using the audit prediction model, and obtain the scene classification result. The image audit module is used to select an audit judgment model corresponding to the scene classification result to audit the multiple images to be audited, and obtain the target audit result.

[0012] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method.

[0013] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0014] The embodiments of this application include at least the following beneficial effects: This application provides an image auditing method, apparatus, electronic device, and program product. The solution includes obtaining a set of images to be audited; the set of images to be audited includes multiple images to be audited and document tags corresponding to each image; performing scene classification prediction based on the set of images to be audited using an auditing prediction model to obtain scene classification results; selecting an auditing judgment model corresponding to the scene classification results to audit the multiple images to be audited to obtain a target auditing result. Implementing the embodiments of this application, the audit prediction model performs scene classification prediction based on the acquired set of images to be audited, and obtains scene classification results. Based on the data characteristics of the input set of images to be audited, it automatically identifies and classifies them into a preset audit algorithm, and selects the audit judgment model corresponding to the scene classification results to audit multiple audit images. It can dynamically configure the audit process according to the set of images to be audited, and realize the audit of multiple audit images based on the scene. Through automated scene recognition, dynamic process configuration and accurate labeling, it greatly reduces manual intervention and repetitive configuration work. At the same time, the dynamic process and rules can respond more agilely to new risks or strategy changes, and can improve the efficiency and accuracy of image-based business audits. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of an implementation environment provided in an embodiment of this application; Figure 2 This is a flowchart of the image verification method provided in the embodiments of this application; Figure 3 This is a sequence diagram illustrating the classification of audit images in one embodiment; Figure 4 This is a flowchart of another image verification method provided in the embodiments of this application; Figure 5 This is a schematic diagram of a multi-dimensional fraud factor scoring system and a multi-level risk warning system in one embodiment; Figure 6 This is a schematic diagram of an AI circuit breaker mechanism in one embodiment; Figure 7 This is a timing diagram illustrating the generation of an electronic receipt in one embodiment; Figure 8 yes Figure 2 The flowchart of step S203 in the process; Figure 9 This is a flowchart illustrating the image audit process in one embodiment; Figure 10 This is a schematic diagram of the structure of the image auditing device provided in the embodiments of this application; Figure 11 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0017] It is understood that the terms "first," "second," etc., used in this application may be used to describe various concepts herein, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of this application, a first set of images to be audited may also be referred to as a second set of images to be audited, and similarly, a second set of images to be audited may also be referred to as a first set of images to be audited. Depending on the context, the words "if" or "when" as used herein may be interpreted as "when," "in response to determination," or "when," "in the event of a determination."

[0018] As used in this application, the terms "at least one", "multiple", "each", "any", etc., "at least one" includes one, two or more, "multiple" includes two or more, "each" refers to each of the corresponding multiples, and "any" refers to any one of the multiples.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0020] Before providing a detailed description of the embodiments of this application, some of the nouns and terms involved in the embodiments of this application will be explained first. The nouns and terms involved in the embodiments of this application are subject to the following interpretations.

[0021] 1) Optical Character Recognition (OCR) technology can be used to extract text information from images. It is often used for text recognition on identity documents, such as ID numbers and names. 2) A CRM system is a tool used by businesses to manage customer relationships and improve the quality of customer service.

[0022] In traditional real-name business audit practices, manual review and verification of all supporting documents submitted by applicants are primarily conducted after the business process is completed. However, this operational model inherently suffers from a series of unavoidable drawbacks: First, its post-audit nature means it cannot identify and intercept potential business risks in a timely manner at critical stages of business acceptance, leading to risks only surfacing later in the process and potentially persisting. Second, due to the lag in auditing, problematic data discovered during manual verification has often already been generated and circulated, making tracing and correction extremely difficult and timely rectification challenging, resulting in the accumulation of erroneous data. Furthermore, the entire audit process is highly dependent on manual operation, consuming significant human resources and significantly increasing audit operating costs. Moreover, constrained by the speed and effort required for manual processing, overall audit efficiency is generally low, resulting in poor overall effectiveness in risk control and business compliance assurance.

[0023] In view of this, this application provides an image auditing method, apparatus, electronic device, storage medium, and program product. This solution uses an auditing prediction model to perform scene classification prediction based on the acquired set of images to be audited, and obtains scene classification results. Based on the data characteristics of the input set of images to be audited, it automatically identifies and classifies them into a preset auditing algorithm, and selects an auditing judgment model corresponding to the scene classification results to audit multiple auditing images. It can dynamically configure the auditing process according to the set of images to be audited, and realize the auditing of multiple auditing images based on scenes. Through automated scene recognition, dynamic process configuration, and accurate labeling, it significantly reduces manual intervention and repetitive configuration work. At the same time, the dynamic process and rules can respond more agilely to new risks or strategy changes, and can improve the efficiency and accuracy of image-based business auditing.

[0024] The image verification method provided in this application relates to the field of image processing technology. The image verification method provided in this application can be applied to a terminal, a server, or software running on a terminal or server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or in-vehicle terminal, but is not limited to these. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing the image verification method, but is not limited to the above forms.

[0025] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0026] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data required for the proper functioning of these embodiments acquired.

[0027] like Figure 1 The diagram shown is a schematic representation of an implementation environment provided in an embodiment of this application. (Refer to...) Figure 1 The implementation environment includes at least one terminal 102 and a server 101. The terminal 102 and the server 101 can be connected via a network, either wirelessly or via a wired connection, to complete data transmission and exchange.

[0028] Server 101 can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0029] Additionally, server 101 can also be a node server in a blockchain network. Blockchain is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms.

[0030] Terminal 102 can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. Terminal 102 and server 101 can be directly or indirectly connected via wired or wireless communication, and this embodiment of the application does not impose any limitations.

[0031] For example, based on Figure 1 The implementation environment shown in this application embodiment provides an image auditing method. The following description uses the application of this image auditing method in server 101 as an example. It can be understood that the image auditing method can also be applied in terminal 102.

[0032] Figure 2 This is an optional flowchart of the image auditing method provided in the embodiments of this application. The executing entity of the image auditing method can be any of the aforementioned electronic devices (including servers or terminals). Figure 2 The method may include, but is not limited to, steps S201 to S203.

[0033] Step S201: Obtain the image set to be audited; the image set to be audited contains multiple images to be audited and the corresponding document labels for each image.

[0034] In some embodiments, an electronic device can receive and load a structured set of images to be audited via its image acquisition interface. This set of images can be a collection of images included in a business audit. The core components of this set of images to be audited include two key elements: first, multiple images submitted by the user, which can be photographs or scans from various physical documents; and second, "document tag" information precisely bound to each image. The document tag corresponding to the image to be audited can be used to describe the specific category of document represented by the corresponding image, such as the front and back of an ID card, the information page of a passport, a driver's license, etc.

[0035] Optionally, when processing user transactions, the electronic device can collect various transaction processing data required for the transactions. This data may include images uploaded by the user and information from various forms filled out, etc., and is not limited thereto. After collecting the transaction processing data, the electronic device can extract the data to generate a set of images to be audited.

[0036] Step S202: The audit prediction model performs scene classification prediction based on the set of images to be audited, and obtains the scene classification result.

[0037] In some embodiments, the scene classification results can be used to describe the business scenario corresponding to the business to be processed for the image to be audited. Electronic devices can input a structured set of images to be audited into a pre-trained audit prediction model for intelligent analysis. This audit prediction model can be an advanced deep neural network, such as a Convolutional Neural Network (CNN) or a Visual Transformer, whose core processing method lies in automatically learning and extracting the multi-dimensional visual features and contextual information contained in each image to be audited. This audit prediction model can automatically determine the specific business scenario category to which the current business processing request belongs by comprehensively analyzing the entire set of images to be audited, such as "new personal account opening," "document information change," "enterprise document verification," or "multi-document combination authentication," etc.

[0038] This scenario classification is not simply image recognition, but rather based on complex pattern recognition and semantic understanding to infer the user's intent and the essential type of business transaction. After obtaining the scenario classification result, the electronic device can intelligently call the most suitable refined audit rule chain and verification process based on the scenario classification result. For example, for the "enterprise document verification" scenario, it automatically strengthens the cross-verification of the consistency and validity period of the business license and legal representative's ID card information.

[0039] Electronic devices use an audit prediction model to perform scene classification prediction based on the set of images to be audited, and obtain scene classification results. This achieves automated and intelligent identification of business scenarios, getting rid of the limitations of manual pre-setting or judgment of scenarios. It not only provides a crucial decision basis for subsequent differentiated and precise risk control, but also significantly improves the adaptability and processing efficiency of the entire audit system, ensuring that business requests of different complexities can obtain the most appropriate and rigorous risk assessment.

[0040] Step S203: Select the audit judgment model corresponding to the scene classification result to audit multiple images to be audited, and obtain the target audit result.

[0041] In some embodiments, electronic devices can dynamically select an audit judgment model that precisely matches the aforementioned scenario classification results through a built-in intelligent routing mechanism, thereby performing in-depth analysis on the image set to be audited. These audit judgment models are typically lightweight deep learning networks trained independently for specific business scenarios. Specifically, the audit judgment model can be built based on MobileNet or knowledge distillation models, without limitation. The audit judgment model usually includes audit rules that need to be considered when handling the corresponding business. For example, for the "enterprise document verification" scenario, the corresponding audit judgment model can automatically strengthen the cross-validation of the consistency and validity period of the business license and legal representative's ID card information.

[0042] Furthermore, electronic devices can analyze the visual information of multiple images to be audited, such as document textures, printing features, and facial biometric information. Combined with the document tags corresponding to the images, they can perform joint semantic understanding and logical correlation verification on multiple images. The audit judgment model can autonomously focus on key risk dimensions, such as document authenticity verification, information consistency verification, and validity period compliance judgment, to execute end-to-end automated audit decisions and ultimately output the target audit result. This target audit result can include a binary conclusion of pass / reject for the images to be audited, or it can include fine-grained risk markers for the entire set of images to be audited, such as "ID card photo blurriness exceeds the standard" or "business license and legal representative's name are inconsistent."

[0043] Electronic devices can select an audit judgment model corresponding to the scene classification result to audit multiple images to be audited, and obtain the target audit result. This realizes the scenario-based on-demand calling of the audit model. By replacing the general algorithm with a highly customized special model, the accuracy of identifying complex business risks and the ability to resist interference are significantly improved.

[0044] As an optional implementation, after obtaining the target audit result, the electronic device can acquire the manual judgment result corresponding to the first set of images to be audited; the first set of images to be audited can be any set of images to be audited; if the manual judgment result indicates that the target audit result corresponding to the first set of images to be audited is correct, then the first set of images to be audited and the target audit result corresponding to the first set of images to be audited are input into the audit judgment model to feed back into the audit judgment model for training.

[0045] Electronic devices can include a CRM processing system. After the business processing data uploaded by users through the electronic device is completed in the CRM processing system, the registration information is automatically collected into CDAP (Cast Data Application Platform, a unified platform for big data applications). Audit data is automatically generated based on a preset data model and sent to the audit system to form a list of data to be audited. Within the audit system, an audit queue is generated according to a preset priority. Subsequently, an external robot automatically captures and saves one or more of the relevant images, fields, and documents. An external classification and recognition model automatically identifies and classifies one or more of the captured and saved images, fields, and documents, extracting key information. Within the audit system, a scenario logic judgment algorithm is automatically run to determine and obtain the target audit result.

[0046] Figure 3 This is a timeline diagram illustrating the classification of audit images in one embodiment, such as... Figure 3As shown, when a user conducts a business transaction, they can register the transaction through the CRM system, which allows the electronic device to obtain the transaction data. The electronic device then captures the transaction data to obtain a list of items to be audited and adds the list to the audit queue so that the electronic device can process the audits of the list in order. Electronic devices acquire a list of images to be audited and capture and save one or more of the relevant images, fields, and documents. On one hand, they can generate order management data based on these images, fields, and documents and store it in the CRM system. On the other hand, they can train a classification and recognition model on these images, fields, and documents—that is, train an audit prediction model. The identified scene classification results are saved, and the audit judgment model corresponding to the scene classification results is selected to audit multiple images to be audited, obtaining the target audit result. The target audit result is then manually verified. If the manual judgment indicates that the target audit result corresponding to the first set of images to be audited is correct, the first set of images to be audited and its corresponding target audit result are input into the audit judgment model to provide feedback training. If the manual judgment indicates that the target audit result corresponding to the first set of images to be audited is incorrect, a new target audit result input by the user is obtained and used as the target audit result corresponding to the first set of images to be audited. This new target audit result is then input into the audit judgment model to provide feedback training.

[0047] The electronic device uses the set of images to be audited to train the audit judgment model. Users can manually check and correct the identification results based on the target audit results. The number of corrections feeds back into the audit prediction model and the audit judgment model, which can improve the accuracy of the target audit results, thus forming an intelligent closed loop of identification-judgment-feedback-optimization.

[0048] In steps S201 to S203 of this embodiment, the electronic device performs scene classification prediction based on the acquired set of images to be audited using an audit prediction model, obtains scene classification results, automatically identifies and classifies the data features of the input set of images to be audited into a preset audit algorithm, and selects an audit judgment model corresponding to the scene classification results to audit multiple audit images. It can dynamically configure the audit process according to the set of images to be audited, realize the audit of multiple audit images based on the scene. Through automated scene recognition, dynamic process configuration and accurate labeling, manual intervention and repetitive configuration work are greatly reduced. At the same time, the dynamic process and rules can respond more agilely to new risks or strategy changes, and can improve the efficiency and accuracy of image-based business audits.

[0049] Figure 4 This is a flowchart of another image verification method provided in the embodiments of this application, such as... Figure 4 As shown, the image auditing method may also include steps S401 to S406.

[0050] Step S401: Obtain business processing data.

[0051] In some embodiments, when an electronic device performs user business transactions, it captures or receives full transaction data from the front-end business system in real time. This dataset is a fusion of structured and unstructured data, and may include original form information submitted by the user, such as account type and service package, as well as system logs during the business operation process, such as operation timestamps, device fingerprints, and behavior trajectories.

[0052] To ensure the computability of business processing data, the electronic device's built-in data cleaning and feature encoding module standardizes the raw streaming data: key business entities are extracted using semantic parsing technology, discrete operations are transformed into risk quantification indicators using feature engineering methods, and finally, a tensor format adapted to the input of a deep learning model is generated, thus obtaining the processed business processing data. The electronic device can then integrate the business processing data according to preset audit conditions to obtain audit checklist data.

[0053] Step S402: Integrate the business processing data according to the preset audit conditions to obtain the audit list data.

[0054] In some embodiments, audit conditions can be used to describe the conditions that business processing data needs to meet, enabling electronic devices to generate audit checklist data. Electronic devices can first determine whether the business processing data is complete based on preset audit conditions. Specifically, they can identify whether the business processing data contains fields, images, etc., that are required for auditing. More specifically, electronic devices can identify whether an image contains the required fields to determine if the image is complete. If the required fields are identified, the image is considered complete. For example, the required fields for an image of the "front of an ID card" include "name|birth", "gender", "ethnicity|travel|flag", "identity", and "address|preferred address|address|province". To improve the error tolerance of image recognition, similar characters can be set as required fields. For example, it is allowed to identify either the "address" or "preferred address" field as a required field. The required fields for an image of the "back of an ID card" include "issuing authority", "validity period", "resident ID card", "People's Republic of China", and "public security bureau". Electronic devices can also determine whether the matching degree between the fields contained in an image and the fields required for that image in the audit conditions is greater than a preset matching threshold. If it is determined that the matching degree between the fields contained in the image and the fields required for that image in the audit conditions is greater than the preset matching threshold, then the image is considered complete. If the business processing data is determined to be complete, the business processing data can be integrated to obtain the audit list data.

[0055] Furthermore, the electronic device can also determine the accuracy of business processing data based on preset audit conditions. Specifically, it can identify whether the fields contained in the business processing data are consistent with the corresponding fields entered by the user. If they are consistent, the business processing data is considered accurate; if they are inconsistent, the data is considered inaccurate. For example, if the business processing data includes a photo of an ID card and the user-entered ID card number, the electronic device can perform optical character recognition (OCR) on the ID card photo to obtain a recognition field describing the ID card number. This field is then compared with the field of the user-entered ID card number. If the field matches, the identity information is considered accurate. If all the information required for the audit conditions is accurate, the business processing data is considered accurate. If the data is determined to be accurate, it can be integrated to obtain an audit list.

[0056] As an alternative implementation, electronic devices can use RPA (Robotic Process Automation) technology to retrieve order information and related image queries from the CRM processing system using the order number. Based on the order codes required by the business itself, the data is imported into the query module to obtain and download thousands of image addresses provided by the order codes. In this way, using RPA technology to replace repetitive manual CRM information collection can free up manpower and reduce errors caused by human subjectivity.

[0057] As an optional implementation, the electronic device can acquire scene information corresponding to multiple audit checklists; based on the scene information corresponding to each audit checklist, it generates priority data for each audit checklist according to priority generation conditions; it selects the audit checklist with the highest priority as the target audit checklist, and uses an external robot to capture the target audit checklist to obtain multiple images to be audited and the corresponding document tags for each image. The priority data describes the priority of capturing the corresponding audit checklist data. Audit checklist data can be divided into two categories: manually imported and system imported. Users can set the priority of manually imported data to be higher or lower than that of system imported data, which is not limited here. System-imported audit checklist data can also be divided into categories such as fraud-related, key scenarios, and others. Users can also set the priority of these categories as needed, allowing the electronic device to automatically generate the corresponding priority data for the audit checklist. For audit checklist data in the same scenario, the priority of the audit checklist data can be determined according to the order of import time. Specifically, the priority of the audit checklist data imported earlier can be set to be higher. For example, if audit checklist data A is imported at 15:00:10 and audit checklist data B is imported at 15:00:30, and audit checklist data A and audit checklist data B are audit checklist data in the same scenario, then the priority of the priority data corresponding to audit checklist data A is higher than the priority of the priority data corresponding to audit checklist data B.

[0058] Furthermore, the electronic device can adjust the order in which audit checklist data is retrieved by allowing users to directly input priority data. The device supports manual adjustment of the execution order, enabling emergency queue-jumping functionality. This allows for flexible responses to order audit needs from higher-level authorities and other special scenarios, quickly addressing temporary compliance inspection requirements and effectively reducing the risk of liability determination for enterprises due to surprise inspections.

[0059] Electronic devices generate priority data for each audit checklist based on priority conditions and the scene information corresponding to each audit checklist data. The audit checklist data with the highest priority is selected as the target audit checklist data for capture to obtain multiple images to be audited and the corresponding document tags for each image. Based on a scene-aware elastic scheduling mechanism, this completely reconstructs the traditional first-come-first-served mechanical work mode, enabling high-risk, high-value businesses to obtain second-level response privileges and effectively improving the efficiency of auditing key businesses.

[0060] Step S403: Use an external robot to capture the audit list data to obtain multiple images to be audited and the corresponding document labels for each image.

[0061] In some embodiments, the electronic device can crawl audit checklist data using an external robot, which can be an algorithm for crawling images to be audited from the audit checklist data. Optionally, the electronic device can use the external robot to identify multiple initial business images in the target audit checklist data to obtain the recognition fields corresponding to the initial business images; select the document tags that match the recognition fields corresponding to the initial business images as the document tags corresponding to the initial business images; and perform data cleaning on the multiple initial business images to obtain multiple images to be audited and the document tags corresponding to each image to be audited. The electronic device can use image deep learning frameworks, including PaddleClas, to identify the initial business images in the audit checklist data, and supports model compression and optimization to adapt to different hardware and performance requirements. It has powerful feature extraction and classification capabilities and is widely used in object recognition, scene classification, image retrieval, and other fields. Its key features include model training, multi-label classification, and incremental training, enabling the rapid construction of customized image recognition systems.

[0062] Furthermore, electronic devices can first utilize OCR technology combined with deep learning layout analysis, such as document region detection based on Mask R-CNN, to extract structured recognition fields from the initial business images in the audit checklist data. Subsequently, semantic association matching algorithms can be used to intelligently map the extracted fields to a predefined document label system. For example, when the words "Resident Identity Card" and the national emblem are recognized, the "Front of Identity Card" label is automatically bound, and the corresponding anti-counterfeiting point verification rules are activated. Electronic devices can also adaptively clean the data pipeline, using generative adversarial networks for image denoising and perspective correction, while using a rule engine to remove blurry, incomplete, or other invalid images, outputting a standardized set of images to be audited.

[0063] Since Chinese paths are inconvenient, electronic devices can convert Chinese fields to English fields, and for simplification, use document labels such as ABCD instead, as shown in Table 1: Table 1

[0064] Electronic devices can use external robots to identify multiple initial business images in the target audit list data, obtain the recognition fields corresponding to the initial business images, and select the document tags that match the recognition fields of the initial business images as the document tags corresponding to the initial business images. After data cleaning of multiple initial business images, multiple images to be audited and the document tags corresponding to each image to be audited are obtained. This solves the subjective bias and efficiency bottleneck of manual annotation in the traditional mode, improves the accuracy of document recognition, and increases the processing efficiency of single batch processing.

[0065] As an optional implementation, the electronic device can automatically analyze the target audit results to filter out the set of images to be audited corresponding to the target audit results that show abnormal results. When the electronic device is in audit mode, if it detects that the external robot is not offline and the processing time exceeds a preset time threshold, it can automatically trigger a re-audit and re-fetch the current audit list data. The audit status can be that the audit list data is being fetched by an external robot, and when auditing the images to be audited, the audit status can be: processing, pending crawling, pending OCR technology execution, pending audit, or auditing. The time threshold can be set to 10 seconds. If the electronic device detects that the processing time exceeds the preset time twice, it will automatically trigger an alert and request manual intervention for verification. By fixing the anomalies, the execution efficiency and accuracy of the model can be continuously optimized.

[0066] Based on the aforementioned audit scenarios and rules, this audit method needs further expansion to enhance the actual audit effectiveness of these scenarios during the actual audit process. Electronic devices can construct multi-dimensional fraud factor scoring levels and multi-level risk warning systems. This can involve establishing multi-dimensional high-risk factor scoring levels, subdividing factors such as network access compliance, customer attributes, call characteristics, traffic patterns, and user characteristics to form a multi-dimensional application foundation for high-risk factors. Electronic devices can also establish number risk levels and multi-level warning systems, encapsulating high-risk factor modules according to different business characteristics, and forming a dynamic module and handling model integration application based on business scenarios. Figure 5 This is a schematic diagram of a multi-dimensional fraud factor scoring system and a multi-level risk warning system in one embodiment, such as... Figure 5As shown, the electronic device can construct a number database containing a database of fraudulent number cases and numbers at specific risk levels. By aggregating the underlying data of the number database, a feature wide table is extracted. This feature wide table can contain historical feature data of fraudulent numbers and feature data of numbers at specific risk levels. Based on the feature wide table, a risk grading model and multiple rule models are trained to obtain the trained risk grading model and multiple rule models. The electronic device can use these models to assess the risk of numbers involved in business transactions and obtain the corresponding risk result. If the risk result of a number is detected to be higher than the high-risk warning line, a customer service callback signal can be issued to notify relevant customer service personnel to respond and handle the situation. If the risk result is detected to be lower than the low-risk warning line, no action is required. If customer service personnel determine that the number is involved in fraud during the callback, the number can be suspended or blacklisted.

[0067] At the same time, electronic devices establish an AI circuit breaker mechanism to promptly cut off risks and stop losses. Specifically, this can be achieved by opening up the CMR permission configuration port to enable circuit breakers for sales acceptance, login, order rejection, outlets, and settlement for employee IDs, outlets, and packages, so as to promptly cut off risky behaviors and subject permissions and prevent risks from spreading. Figure 6 This is a schematic diagram of an AI circuit breaker mechanism in one embodiment, such as... Figure 6 As shown, if an electronic device identifies a yellow-risk behavior in an employee ID performing business, it can stop the current business being performed by that employee ID; if it identifies an orange-risk behavior, it can stop the automated control process for that employee ID; if it identifies a red-risk behavior, it can stop the use and settlement of that employee ID. Yellow-risk behaviors can be any one or more of the following: failure to generate an electronic business document, lack of a valid signature on the electronic business document, and low matching rate of the employee ID development contract serial number. Orange-risk behaviors can be any one or more of the following: large number of queries, high-frequency queries, large number of ID card queries, employee ID downgrade renewals reaching 10, employee ID downgrade renewal rate reaching 15%, abnormal registration of individuals under 18 years old, and employee ID activation by individuals on the negative list. Red-risk behaviors can be any one or more of the following: employee IDs at ineffective branches not frozen in time, low matching rate of the branch development contract serial number, and activation of branches by partners on the negative list.

[0068] Furthermore, electronic devices leverage core capabilities such as image recognition and document signature recognition to promote cross-industry applications. Specifically, these devices can provide embedded pages that upload images and return recognition results. Uploading an image triggers an API call, using the uploaded image as input to output the image's classification result and accompanying text. Figure 7 This is a timing diagram illustrating the generation of an electronic receipt in one embodiment, as shown below. Figure 7As shown, when a user completes a business transaction, they register the transaction in the CRM system, and then the data is processed through CDAP. Next, the robot retrieves the electronic receipt / field information from the electronic receipt. This information is submitted to the signature recognition system, which uses an image analysis algorithm to predict the electronic signature result and saves the result. Simultaneously, the information is transmitted to the audit system. The audit system first generates a list of items to be audited, then creates an audit queue, and uses a scenario-based logic judgment algorithm to evaluate the signature recognition results, etc. During this process, humans can verify the audit results, ultimately arriving at the final audit conclusion. The entire process encompasses multiple stages, including initialization, information acquisition, content recognition, and audit results, involving the collaborative operation of multiple participants and system modules, including users, CRM, CDAP, the robot, the signature recognition system, and the audit system.

[0069] Furthermore, the electronic device integrates PaddleClase's image processing technology, allowing users to edit configuration files and set model parameters before training begins. Training parameters are defined as 100 training epochs, a cosine learning rate, a decay strategy, and specified training / validation dataset paths (train.txt / val.txt). Model performance is evaluated using Top-1 accuracy, Top-5 accuracy, confusion matrix, and category-level metrics. Top-1 accuracy, the proportion of correctly predicted classes, is the core metric; Top-5 accuracy, the proportion of correct predictions of the top 5 classes, is a metric for multi-class tasks; the confusion matrix analyzes misclassification among classes; and category-level metrics include precision and recall for each class. The criteria for retraining are: if Top-1 accuracy is <85%, retraining is required; or if the recall for any class is <80% or shows a significant tendency to misclassify, retraining is also required.

[0070] Electronic devices can undergo 100 rounds of training using the constructed image depth training framework, and the performance of the trained model can be evaluated. 200 samples are randomly selected from the training set for manual verification, focusing on analyzing low-confidence correct samples and high-confidence incorrect samples combining frequently misclassified categories. File parameters in the script are replaced, and the performance of the trained model is evaluated using a confusion matrix. Based on the decision criteria, it is determined whether secondary training optimization is necessary. Secondary training optimization can involve retaining high-confidence correct samples (confidence > 0.85), focusing on collecting samples with blurred boundaries (confidence in the 0.4-0.6 range), and removing low-quality samples (consistently inconsistent predictions or confidence < 0.2). Occlusion processing is applied to frequently misclassified samples to create hybrid samples, i.e., samples that fuse features from different categories, to adjust the data distribution, i.e., increasing the proportion of weaker category samples to balance the amount of training data for each category.

[0071] At this point, a progressive dynamic learning rate is adopted: a constant learning rate of (5e-4) for the first 10 rounds; cosine annealing decay for the middle rounds; and exponential decay for the last 5 rounds (γ=0.9). It should be noted that during training, strong enhancements are applied to high-frequency misclassified samples: MixUp and CutMix are used to generate adversarial examples that are added to the training set. Based on the optimized model after secondary training, steps 2-3 are repeated until the judgment criteria determine that secondary training optimization is no longer needed. Finally, the multi-dimensional acceptance criteria are validated and derived: Top-1 accuracy ≥ 92%, recall for each category > 85%, and maximum class preference < 30% on the test set. Finally, the image classification provides an API interface, as shown in Table 2. Table 2

[0072] The specific interface of the Result object is shown in Table 3: Table 3

[0073] The specific interface for electronic receipt signature verification is shown in Table 4: Table 4

[0074] Step S404: Obtain the image set to be audited; the image set to be audited contains multiple images to be audited and the corresponding document labels for each image.

[0075] Step S405: The audit prediction model performs scene classification prediction based on the set of images to be audited, and obtains the scene classification result.

[0076] Step S406: Select the audit judgment model corresponding to the scene classification result to audit multiple images to be audited, and obtain the target audit result.

[0077] The descriptions of steps S404 to S406 can be found in the descriptions of steps S201 to S203 in the above embodiments, and will not be repeated here.

[0078] In this embodiment, the electronic device acquires business processing data, integrates the data according to preset audit conditions to obtain audit list data, and uses an external robot to capture the audit list data to obtain multiple images to be audited and the corresponding document tags for each image. This improves the efficiency and accuracy of image-based business audits.

[0079] Figure 8 yes Figure 2 The flowchart of step S203 in the example is as follows: Figure 8 As shown, step S203 may also include steps S801 to S803.

[0080] Step S801: Analyze the completeness of the image set to be audited and the audit list data through the audit judgment model.

[0081] In some embodiments, electronic devices can analyze the integrity of the image content to be audited through a visual feature extraction branch, such as missing national emblem on an ID card or blurred seal on a business license. Simultaneously, a data flow analysis branch can detect the completeness of list elements, such as missing required fields on an account opening application form or missing additional qualification certificates required for cross-border business. The audit judgment model can dynamically calculate the association weight between image semantics and list metadata using an attention mechanism. If an anomaly is identified, corresponding prompts are generated. For example, when it is detected that "the registered capital item in the enterprise account opening list is empty" and "there are traces of smearing in the registered capital area of ​​the business license image," an integrity violation flag is automatically triggered.

[0082] Step S802: If it is determined that the set of images to be audited and the audit list data are incomplete, output a prompt message and re-acquire new images to be audited.

[0083] Step S803: If it is determined that the set of images to be audited and the audit list data are complete, then the audit judgment model is used to audit each image to be audited according to the audit list data to obtain the target audit result.

[0084] In some embodiments, the electronic device can perform graph neural network-driven topological relationship analysis on the image set to be audited and the associated list data through an audit judgment model. If it is determined that the image set to be audited and the audit list data are incomplete, a prompt message is output and new images to be audited are re-acquired. The user can upload new images to be audited through the prompt message.

[0085] If the completeness of the image set to be audited and the audit checklist data is confirmed, the electronic device can perform logical compliance verification by passing the audit checklist data through a lightweight rule tree, such as the matching degree between the validity period of the certificate and the timeliness of the business. At the same time, the image set to be audited can be input into a convolutional attention model for fine-grained risk scanning, thereby obtaining the target audit results.

[0086] Specifically, electronic devices can use an audit judgment model to perform image recognition on the images to be audited, obtaining key image information corresponding to the images to be audited. The audit judgment model then matches the audit list data with the key image information corresponding to the images to be audited according to preset audit configuration rules, obtaining matching degree data for each image to be audited. Based on the matching degree data for each image to be audited, the target audit result is determined. The electronic device performs fine-grained feature extraction on the images to be audited using the audit judgment model, generating a structured image key information matrix, which can include dimensions such as document type, field coordinates, and text confidence. Subsequently, it calls a rule inference engine to apply preset audit configuration rules, such as setting a rule that "ID number must match the account opening form + facial similarity ≥ 0.85". The configuration rules are then converted into computable logical expressions. Semantic alignment technology is used to dynamically map the business fields in the audit list data to the key image information, and a fuzzy matching algorithm is used to calculate the matching degree data for each data item, outputting a confidence score in the 0-1 range. Finally, a risk-weighted aggregation strategy is adopted to perform joint decision analysis on multi-dimensional matching data, and output target audit results including risk classification, anomaly location and handling suggestions.

[0087] For example, auditing personal agency scenarios involves verifying the uploaded ID information and power of attorney in the CRM system to determine if the required materials are met. If the account holder is a minor, family member's household registration information is also required. Specific audit rules include: (a) If the agent for an adult client is different from the client, a power of attorney must be checked, and the name on the power of attorney must match the agent's. (b) If the client or agent is under 18 years old, information about a legal representative must be checked, and the relationship with the legal representative must be verified. The following points prove the legal representative relationship: 1. The client and agent (legal representative) have the same address on their identification documents. 2. The client and agent (legal representative) are on the same household registration book, and the agent (legal representative) is the head of household. 3. If the agent is not the head of household, a medical birth certificate is needed to prove the legal representative's kinship with the head of household; similarly, the kinship with the client can be proven.

[0088] Furthermore, based on the scenario logic judgment algorithm, the audit process can be dynamically set and dynamic set rules can be configured. This enables the integration of dynamic classification audit scenario functions into the model recognition module of the network access audit management platform, realizing dynamic audit processes and scenario labeling based on business flow configuration. This improves audit judgment efficiency and significantly enhances the adaptability and risk control capabilities of different business scenarios.

[0089] Electronic devices can use an audit judgment model to perform image recognition on images to be audited, obtaining key image information corresponding to the images to be audited. The audit judgment model then matches audit list data with the key image information corresponding to the images to be audited according to preset audit configuration rules, obtaining matching degree data for each image to be audited. Based on the matching degree data for each image to be audited, the target audit result is determined. Through deep coupling of machine vision and rule engine, end-to-end automation from image semantic understanding to business compliance judgment is achieved, improving the efficiency and accuracy of image-based business audits.

[0090] Figure 9 This is a flowchart illustrating the image audit process in one embodiment, such as... Figure 9 As shown, government and enterprise clients need to provide the marking scenario and relevant personnel's identity verification, authentic photos, and valid documents. The system will check if the client's name meets the specified character limit. If not, the client needs to supplement the materials as prompted; if it does, they will proceed to the self-service marking process for individual users, where they need to submit their identity verification and authentic photos again. If the client's name does not match the personnel's name, they need to go through the personal user entrusted marking process, submitting the homeowner's and personnel's valid identity verification, the personnel's authentic photos, and proof that the entrusted marking location is valid. Subsequently, the system will initiate an intelligent check to verify whether there are any missing or incorrect documents. If it fails, it will automatically assign a rectification order, and finally, after confirmation that everything is correct, the review will be completed and approved.

[0091] In this embodiment, the electronic device analyzes the completeness of the image set to be audited and the audit list data through an audit judgment model. If it is determined that the image set to be audited and the audit list data are incomplete, a prompt message is output and new images to be audited are reacquired. If it is determined that the image set to be audited and the audit list data are complete, the audit judgment model audits each image to be audited according to the audit list data to obtain the target audit result, which can improve the efficiency and accuracy of business auditing based on images.

[0092] Please see Figure 10 This application also provides an image verification device that can implement the above-described method. The device includes: Image acquisition module 1001 is used to acquire the image set to be audited; the image set to be audited contains multiple images to be audited and the corresponding document labels for each image to be audited. Scene prediction module 1002 is used to perform scene classification prediction based on the set of images to be audited using the audit prediction model, and obtain scene classification results; Image audit module 1003 is used to select an audit judgment model corresponding to the scene classification result to audit multiple images to be audited, and obtain the target audit result.

[0093] It is understood that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0094] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0095] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0096] Please see Figure 11 , Figure 11 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 1101 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 application. The memory 1102 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 1102 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1102 and is called and executed by the processor 1101 using the methods described in the embodiments of this application. Input / output interface 1103 is used to implement information input and output; The communication interface 1104 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 1105 transmits information between various components of the device (e.g., processor 1101, memory 1102, input / output interface 1103, and communication interface 1104); The processor 1101, memory 1102, input / output interface 1103 and communication interface 1104 are connected to each other within the device via bus 1105.

[0097] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0098] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0099] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0100] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0101] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0102] The image auditing method, apparatus, electronic device, storage medium, and program products provided in this application significantly reduce manual intervention and repetitive configuration work through automated scene recognition, dynamic process configuration, and precise labeling. At the same time, the dynamic processes and rules can respond more agilely to new risks or strategy changes, thereby improving the efficiency and accuracy of image-based business auditing.

[0103] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0104] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0105] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0106] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0107] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0108] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0109] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0110] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0111] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0112] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0113] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for image verification, characterized in that, The method includes the following steps: Obtain the image set to be audited; the image set to be audited contains multiple images to be audited and the corresponding document tags for each image; The scene classification result is obtained by using the audit prediction model to perform scene classification prediction based on the set of images to be audited; The audit judgment model corresponding to the scene classification result is selected to audit the multiple images to be audited, and the target audit result is obtained.

2. The method according to claim 1, characterized in that, Before obtaining the set of images to be audited, the method further includes: Obtain business processing data; The business processing data is integrated according to the preset audit conditions to obtain the audit list data; An external robot is used to capture the audit list data to obtain the multiple images to be audited and the corresponding document tags for each image.

3. The method according to claim 2, characterized in that, Before using an external robot to retrieve the audit list data to obtain the multiple images to be audited and the corresponding document tags for each image, the method further includes: Obtain the scenario information corresponding to each of the multiple audit checklist data; Based on the scenario information corresponding to each audit checklist data, priority data corresponding to each audit checklist data is generated according to the priority generation conditions. The process of using an external robot to crawl the audit list data to obtain the multiple images to be audited and the corresponding document tags for each image includes: The audit list data with the highest priority is selected as the target audit list data, and an external robot is used to crawl the target audit list data to obtain the multiple images to be audited and the corresponding document tags for each image.

4. The method according to claim 2, characterized in that, The process of using an external robot to crawl the audit list data to obtain the multiple images to be audited and the corresponding document tags for each image includes: The external robot identifies multiple initial business images in the target audit checklist data to obtain the identification fields corresponding to the initial business images. Select the document tag that matches the recognition field corresponding to the initial business image as the document tag corresponding to the initial business image; Data cleaning is performed on the multiple initial business images to obtain the multiple images to be audited and the corresponding document tags for each image to be audited.

5. The method according to claim 1, characterized in that, The audit judgment model corresponding to the scene classification result is selected to audit the multiple images to be audited, and the target audit result is obtained, including: The completeness of the image set to be audited and the audit list data is analyzed using the audit judgment model. If it is determined that the set of images to be audited and the audit list data are incomplete, a prompt message will be output and new images to be audited will be obtained again. If it is determined that the set of images to be audited and the audit list data are complete, then the audit judgment model is used to audit each image to be audited according to the audit list data to obtain the target audit result.

6. The method according to claim 5, characterized in that, The step of auditing each image to be audited based on the audit checklist data using the audit judgment model to obtain the target audit result includes: The audit judgment model is used to perform image recognition on the image to be audited to obtain key image information corresponding to the image to be audited. The audit judgment model matches the audit list data with the key information of the images to be audited according to the preset audit configuration rules, thereby obtaining the matching degree data for each image to be audited. The target audit result is determined based on the matching degree data corresponding to each image to be audited.

7. The method according to any one of claims 1 to 6, characterized in that, After selecting the audit judgment model corresponding to the scene classification result to audit the multiple images to be audited and obtaining the target audit result, the method further includes: Obtain the manual judgment result corresponding to the first set of images to be audited; the first set of images to be audited can be any set of images to be audited; If the manual judgment result indicates that the target audit result corresponding to the first set of images to be audited is correct, then the first set of images to be audited and the target audit result corresponding to the first set of images to be audited are input into the audit judgment model to feed back into the audit judgment model for training.

8. An image verification device, characterized in that, The device includes: The image acquisition module is used to acquire a set of images to be audited; the set of images to be audited includes multiple images to be audited and the corresponding document tags for each image. The scene prediction module is used to perform scene classification prediction based on the set of images to be audited using the audit prediction model, and obtain the scene classification result. The image audit module is used to select an audit judgment model corresponding to the scene classification result to audit the multiple images to be audited, and obtain the target audit result.

9. An electronic device, characterized in that, The electronic device / computer apparatus includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images