Abnormal certificate detection method and device and electronic equipment

Abstract

The invention discloses an abnormal certificate detection method and apparatus, and an electronic device. The method comprises the following steps: acquiring a voucher image of a target voucher according to a preset acquisition frequency, and identifying the voucher image of the target voucher from the voucher image; identifying text information in the voucher image, and determining a content identification result of the target voucher according to the text information; inputting the voucher image into an appearance recognition model to obtain an appearance recognition result of the target voucher; and determining an abnormal state of the target voucher according to the appearance recognition result and the content recognition result. According to the method and the device, the problem that the efficiency and the timeliness of performing abnormity check on the bank business voucher in the related technology are relatively low is solved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and more specifically, to a method, apparatus, and electronic device for detecting abnormal credentials. Background Technology

[0002] Currently, when checking whether bank branch business documents (such as business licenses) are displayed correctly according to bank regulations, the following two methods are usually used: one is manual on-site inspection, where staff regularly or irregularly visit various branches to visually inspect whether there are any abnormalities in the display status of business documents and manually record them; the other is random photo upload inspection, where branch staff take photos daily and upload them to the management system for manual review by back-office staff.

[0003] However, both of the above methods have the following problems: 1. Manual inspection is less efficient and accurate, and incurs labor costs; 2. Manual inspection has a gap period, making it impossible to detect business vouchers in real time.

[0004] There is currently no effective solution to the problem of low efficiency and timeliness in the anomaly detection of bank business documents in related technologies. Summary of the Invention

[0005] The main objective of this application is to provide a method, apparatus, and electronic device for detecting abnormal documents, so as to solve the problem of low efficiency and timeliness in the related technologies for abnormal inspection of bank business documents.

[0006] To achieve the above objectives, according to one aspect of this application, a method for detecting abnormal vouchers is provided. The method includes: acquiring voucher images of a target voucher at a preset acquisition frequency, and identifying voucher images of the target voucher from the voucher images; identifying text information in the voucher images, and determining the content recognition result of the target voucher based on the text information; inputting the voucher images into an appearance recognition model to obtain the appearance recognition result of the target voucher; and determining the abnormal state of the target voucher based on the appearance recognition result and the content recognition result.

[0007] Optionally, identifying the target voucher image from the voucher image includes: inputting the voucher image into a coordinate recognition model to obtain the coordinate information of the target voucher in the voucher image, wherein the coordinate information includes the vertex coordinates of the target voucher, the coordinate recognition model is obtained by training an initial neural network model with a first training set, the first training set consists of multiple first sample data, each first sample data includes the voucher image of a sample voucher and the coordinate information of the sample voucher; determining the coverage area of ​​the target voucher in the voucher image based on the coordinate information, and cropping the coverage area to obtain the voucher image.

[0008] Optionally, the appearance recognition model is trained as follows: M historical voucher images are acquired, and the M historical voucher images are occluded or damaged to obtain N sample voucher images. The appearance recognition label of the historical voucher images is "no anomaly", and the appearance recognition label of the sample voucher images is "anomaly". M and N are both positive integers, and N is greater than or equal to M. The M historical voucher images and N sample voucher images, as well as the appearance recognition labels of each historical voucher image and each sample voucher image, are determined as the second training set. The initial neural network model is trained using the second training set to obtain the appearance recognition model.

[0009] Optionally, inputting the voucher image into the appearance recognition model to obtain the appearance recognition result of the target voucher includes: inputting the voucher image into the appearance recognition model to obtain a first recognition result; determining a reference edge in the border of the voucher image and determining a contrast edge of the voucher image corresponding to the reference edge; determining the angle value between the reference edge and the contrast edge, and determining a second recognition result of the voucher image based on the angle value; determining the appearance recognition result as normal if both the first recognition result and the second recognition result indicate no abnormality; and determining the appearance recognition result as abnormal if the first recognition result and / or the second recognition result indicate an abnormality.

[0010] Optionally, determining the content recognition result of the target credential based on the text information includes: obtaining preset information of the target credential, comparing the preset information with the text information to obtain text similarity, and determining a third recognition result based on the text similarity; recognizing the expiration date data in the text information, judging whether the expiration date data is valid, and determining a fourth recognition result based on the judgment result; if both the third and fourth recognition results indicate no abnormality, determining the content recognition result as normal; if the third and / or fourth recognition results indicate abnormality, determining the content recognition result as abnormal.

[0011] Optionally, identifying text information in a voucher image includes: scanning the voucher image to obtain initial text data of the voucher image; determining the attribute information of each initial text data, and filtering the initial text data according to preset filtering rules and attribute information to obtain filtered initial text data; and determining the filtered initial text data as text information.

[0012] Optionally, determining the abnormal state of the target credential based on the appearance recognition result and the content recognition result includes: if both the appearance recognition result and the content recognition result indicate no abnormality, determining the abnormal state of the target credential as no abnormality; if the appearance recognition result and / or the content recognition result indicate an abnormality, determining the abnormal state of the target credential as abnormal.

[0013] To achieve the above objectives, according to another aspect of this application, an abnormal document detection device is provided. The device includes: a first identification unit, configured to acquire document images of a target document at a preset acquisition frequency and identify a document image of the target document from the document images; a first determination unit, configured to identify text information in the document image and determine the content recognition result of the target document based on the text information; a second identification unit, configured to input the document image into an appearance recognition model to obtain the appearance recognition result of the target document; and a second determination unit, configured to determine the abnormal state of the target document based on the appearance recognition result and the content recognition result.

[0014] Optionally, the first recognition unit includes: a first recognition module, used to input the voucher image into a coordinate recognition model to obtain the coordinate information of the target voucher in the voucher image, wherein the coordinate information includes the vertex coordinates of the target voucher, and the coordinate recognition model is obtained by training an initial neural network model with a first training set, the first training set consisting of multiple first sample data, each first sample data including a voucher image of a sample voucher and the coordinate information of the sample voucher; and a first determination module, used to determine the coverage area of ​​the target voucher in the voucher image based on the coordinate information, and to crop the coverage area to obtain a voucher image.

[0015] Optionally, the appearance recognition model is trained using the following apparatus: an acquisition unit, used to acquire M historical voucher images and perform occlusion or damage operations on the M historical voucher images to obtain N sample voucher images, wherein the appearance recognition label of the historical voucher images is "no anomaly" and the appearance recognition label of the sample voucher images is "anomaly", M and N are both positive integers, and N is greater than or equal to M; a training unit, used to determine the M historical voucher images and N sample voucher images, as well as the appearance recognition label of each historical voucher image and the appearance recognition label of each sample voucher image, as a second training set, and to train the initial neural network model using the second training set to obtain the appearance recognition model.

[0016] Optionally, the second recognition unit includes: a second recognition module for inputting the voucher image into the appearance recognition model to obtain a first recognition result; a second determination module for determining a reference edge in the border of the voucher image and determining a comparison edge of the voucher image corresponding to the reference edge; a third determination module for determining the angle value between the reference edge and the comparison edge, and determining a second recognition result of the voucher image based on the angle value; a fourth determination module for determining that the appearance recognition result is normal if both the first recognition result and the second recognition result indicate no abnormality; and a fifth determination module for determining that the appearance recognition result is abnormal if the first recognition result and / or the second recognition result indicate an abnormality.

[0017] Optionally, the first determining unit includes: an acquisition module, used to acquire preset information of the target voucher, compare the preset information with text information to obtain text similarity, and determine a third identification result based on the text similarity; a judgment module, used to identify the expiration date data in the text information, judge the expiration date data to determine whether the target voucher is valid, and determine a fourth identification result based on the judgment result; a sixth determining module, used to determine that the content identification result is normal if both the third and fourth identification results indicate no abnormality; and a seventh determining module, used to determine that the content identification result is abnormal if the third and / or fourth identification results indicate an abnormality.

[0018] Optionally, the first determining unit includes: a scanning module for scanning the voucher image to obtain initial text data of the voucher image; a filtering module for determining the attribute information of each initial text data respectively, and filtering the initial text data according to preset filtering rules and attribute information to obtain filtered initial text data; and an eighth determining module for determining the filtered initial text data as text information.

[0019] Optionally, the second determining unit includes: a ninth determining module, used to determine the abnormal state of the target credential as no abnormality when both the appearance recognition result and the content recognition result indicate no abnormality; and a tenth determining module, used to determine the abnormal state of the target credential as abnormal when the appearance recognition result and / or the content recognition result indicate an abnormality.

[0020] To achieve the above objectives, according to another aspect of this application, an electronic device is provided, the electronic device including a memory storing an executable program; and a processor for running the program, wherein the program executes the above-described method for detecting abnormal credentials during runtime.

[0021] To achieve the above objectives, according to another aspect of this application, a computer program product is provided, including computer instructions that, when executed by a processor, implement the steps of the above-described method for detecting abnormal credentials.

[0022] In this embodiment, the method involves acquiring images of target vouchers at a preset acquisition frequency, identifying the target voucher image from the voucher image, identifying text information in the voucher image, determining the content recognition result of the target voucher based on the text information, inputting the voucher image into an appearance recognition model to obtain the appearance recognition result of the target voucher, and determining the abnormal state of the target voucher based on the appearance recognition result and the content recognition result. By acquiring images of target vouchers and identifying text information based on the voucher image, and then performing anomaly detection on the target voucher from both content and appearance dimensions, the method achieves the goal of automatically and accurately detecting target vouchers whenever an inspection is required. This improves the efficiency and timeliness of voucher inspection and solves the technical problem of low efficiency and timeliness in anomaly inspection of bank business vouchers in related technologies. Attached Figure Description

[0023] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0024] Figure 1 A hardware structure block diagram of a computer terminal for implementing a method for detecting abnormal credentials is shown.

[0025] Figure 2 This is a flowchart of the abnormal credential detection method according to Embodiment 1 of this application;

[0026] Figure 3 This is a schematic diagram of an optional abnormal certificate detection system provided according to Embodiment 1 of this application;

[0027] Figure 4 This is an execution flowchart of the abnormal credential detection system provided in Embodiment 1 of this application;

[0028] Figure 5 This is an execution flowchart of the business license area detection module provided in Embodiment 1 of this application;

[0029] Figure 6 This is a schematic diagram of the suspension state analysis module provided according to Embodiment 1 of this application;

[0030] Figure 7 This is an execution flowchart of the suspension state analysis module provided in Embodiment 1 of this application;

[0031] Figure 8 This is an execution flowchart of the integrity analysis submodule provided in Embodiment 1 of this application;

[0032] Figure 9 This is an execution flowchart of the content compliance analysis submodule provided in Embodiment 1 of this application;

[0033] Figure 10 This is an execution flowchart of the compliance judgment subunit provided in Embodiment 1 of this application;

[0034] Figure 11 This is a schematic diagram of the detection platform of the detection system for the deployment of abnormal credentials according to Embodiment 1 of this application;

[0035] Figure 12 This is a schematic diagram of an abnormal certificate detection device according to Embodiment 2 of this application;

[0036] Figure 13 This is a structural block diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0037] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0038] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0039] It should be noted that the terms "first," "second," etc., in the specification, claims, 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.

[0040] It should be noted that the abnormal credential detection methods, devices, and electronic devices defined in this disclosure can be used in the field of artificial intelligence, or in any field other than artificial intelligence. The application fields of the abnormal credential detection methods, devices, and electronic devices defined in this disclosure are not limited.

[0041] It should be noted that all information, user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, and displayed data) used in this application are information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with the relevant laws, regulations, and standards of the relevant regions, have taken necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse use. If the user chooses to refuse, the process will proceed to the expert decision-making process. For example, this system has an interface with relevant users or organizations. Before obtaining relevant information, a request to obtain the information needs to be sent to the aforementioned user or organization through the interface. After receiving consent from the aforementioned user or organization, the relevant information is obtained. Users can view the purpose of data use in real time through the authorization interface and have the right to withdraw authorization or delete data at any time. After the authorization is withdrawn, the system will terminate the relevant data processing within 24 hours.

[0042] The embodiments or examples disclosed herein are not exhaustive, but merely illustrative of some embodiments or examples, and are not intended to limit the scope of protection of this disclosure. Unless otherwise specified, each step in a particular embodiment or example can be implemented as an independent embodiment, and the steps can be arbitrarily combined. For example, a solution after removing some steps in a particular embodiment or example can also be implemented as an independent embodiment, and the order of the steps in a particular embodiment or example can be arbitrarily interchanged. Furthermore, optional methods or examples in a particular embodiment or example can be arbitrarily combined; moreover, embodiments or examples can be arbitrarily combined. For example, some or all steps of different embodiments or examples can be arbitrarily combined, and a particular embodiment or example can be arbitrarily combined with optional methods or examples of other embodiments or examples.

[0043] Example 1

[0044] According to an embodiment of this application, an embodiment of a method for detecting abnormal credentials is also provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0045] The method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1 A hardware block diagram of a computer terminal for implementing a method for detecting abnormal credentials is shown. Figure 1 As shown, the computer terminal 10 (or mobile device) may include one or more processors 102 (shown as 102a, 102b, ..., 102n in the figure) 102 (processor 102 may include, but is not limited to, processing devices such as microprocessors or programmable logic devices), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface, a universal serial bus port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0046] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0047] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the abnormal credential detection method in this embodiment. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby implementing the above-mentioned abnormal credential detection method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0048] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0049] The display may be, for example, a touchscreen LCD display that allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0050] Under the aforementioned operating environment, this application provides the following: Figure 2 The method for detecting abnormal credentials is shown. Figure 2 This is a flowchart of the abnormal credential detection method according to Embodiment 1 of this application, as follows: Figure 2 As shown, the method includes:

[0051] Step S201: Collect the image of the target voucher according to the preset collection frequency, and identify the voucher image of the target voucher from the voucher image.

[0052] It should be noted that the executing entity in this embodiment can be a document detection system installed in a financial institution. This system can be connected to a camera installed near the hanging position of the target document in the financial institution, so as to collect the image of the target document through the camera, and then determine whether there is any abnormality in the hanging state of the target document based on the image.

[0053] It should be noted that the preset acquisition frequency is the pre-set acquisition frequency of the target document, which can be a business license of a bank branch or other hanging document. The document image is a video frame image obtained by the camera capturing the hanging position of the target document. This image may contain the target document and other objects nearby. The document image can be used to determine whether a document is hanging, and whether the hanging method and content of the document are accurate. The document image is an image extracted from the document image that contains only the target document.

[0054] For example, the credential verification system can connect to the surveillance cameras at a bank branch via a real-time streaming protocol. The system selectively captures video frames from the cameras based on a set acquisition frequency τ (e.g., τ = 1 second), rather than processing every single frame, to balance computational resources with the real-time monitoring requirements.

[0055] Furthermore, after obtaining the image of the voucher at its location, the features in each frame of the image can be analyzed using a deep neural network model to determine whether the target voucher exists in that frame. If the target voucher does not exist, it indicates that the financial institution has not properly displayed the voucher, and the detection result is determined to be abnormal.

[0056] In the presence of a target voucher, it can be determined that the financial institution has suspended the voucher. At this point, it is necessary to identify the target voucher in the voucher image and extract the target voucher from the voucher image to obtain the voucher image. This image is then further inspected to determine whether the target voucher is abnormal.

[0057] For example, the system can connect to existing surveillance cameras in bank branches via a preset protocol to receive and decode video streams in H.264 / H.265 formats in real time. Considering that a business license is a static object, the system does not need to analyze it frame by frame. Instead, by setting a video frame extraction time interval, it extracts frames from the input video stream to obtain keyframes for analysis, represented as follows:

[0058] ;

[0059] in, This is a video keyframe extraction function, where S is the video stream address. In one embodiment, the preset video frame extraction time interval is used. =1, where V is the extracted video frame image. Video keyframe extraction effectively balances computational overhead and real-time performance.

[0060] Step S202: Identify the text information in the voucher image and determine the content recognition result of the target voucher based on the text information.

[0061] It should be noted that the text information can be the text content extracted from the business license image, including but not limited to the company name, unified social credit code, validity period, and other information.

[0062] For example, after acquiring the image of the target document, the recognized image can be submitted to an optical character recognition engine. A deep neural network model is then used to locate and recognize the text, converting it into readable text information. Subsequently, the system compares this text information with preset business license standard information to verify the authenticity and accuracy of the content, ensuring that the information on the target document has not been tampered with or contains errors.

[0063] Step S203: Input the image of the voucher into the appearance recognition model to obtain the appearance recognition result of the target voucher.

[0064] For example, the system can also directly input the image of the credential into the appearance recognition model. The model analyzes the image based on a deep neural network to determine whether the target credential has physical damage, whether it is obscured by external factors, and whether its hanging is tilted. Each analysis dimension produces a result, including a normal or abnormal state judgment and the confidence level of that judgment, based on which the system comprehensively assesses the appearance health of the target credential.

[0065] Step S204: Determine the abnormal status of the target credential based on the appearance recognition result and the content recognition result.

[0066] For example, after obtaining the appearance recognition results and text recognition results, the system will comprehensively analyze the content recognition results and appearance recognition results to determine whether the target document is abnormal. If the text information completely matches the standard information, and the appearance inspection shows that the target document has not suffered physical damage, is not obstructed, and is hung in the correct posture, the system determines that the target document is in a normal display state; otherwise, if any one of the checks fails, it is considered an abnormal state, and the system will immediately trigger an early warning mechanism, generate an alarm, record the abnormal details in detail, and notify relevant personnel to handle it. This allows for real-time detection of the status of target documents hung by financial institutions, and by comprehensively analyzing the content recognition results and appearance recognition results, the accuracy of the detection results is improved.

[0067] This embodiment enables financial institutions to monitor the display status of business licenses at their branches around the clock through an automated monitoring process. By using cameras to capture video frames and analyzing the content and appearance of the documents in the video frame images, the detection results of the document display status are obtained, thereby improving the efficiency and timeliness of document display status detection.

[0068] The abnormal document detection method provided in this application collects the document image of the target document and identifies text information based on the document image. Then, it performs abnormal detection on the target document from two dimensions: content and appearance, and obtains the detection result of the target document. This achieves the goal of automatically and accurately detecting the target document whenever it needs to be checked, thereby improving the efficiency and timeliness of document detection and solving the technical problem of low efficiency and timeliness of abnormal checks on bank business documents in related technologies.

[0069] To accurately obtain a voucher image from a voucher image, optionally, in the abnormal voucher detection method provided in this application embodiment, identifying the voucher image of a target voucher from the voucher image includes: inputting the voucher image into a coordinate recognition model to obtain the coordinate information of the target voucher in the voucher image, wherein the coordinate information includes the vertex coordinates of the target voucher, the coordinate recognition model is obtained by training an initial neural network model with a first training set, the first training set consists of multiple first sample data, each first sample data includes a voucher image of a sample voucher and the coordinate information of the sample voucher; determining the coverage area of ​​the target voucher in the voucher image based on the coordinate information, and cropping the coverage area to obtain the voucher image.

[0070] It should be noted that the coordinate recognition model can be a deep learning-based model used to identify the location information of the target document in the document image, especially to determine its vertex coordinates. The vertex coordinates can be the coordinates of the four corners of the target document in the image.

[0071] For example, when determining the voucher image, the initial neural network model needs to be trained using a first training set to obtain a coordinate recognition model. In the initial stage, the initial neural network model is trained using deep learning with the first training set to adapt to the specific features and location patterns of bank branch vouchers. The first training set may contain multiple image samples, each of which is labeled with the coordinates of the voucher in the image, i.e., the vertex coordinates, so that the model can determine the vertex coordinates of the voucher based on the image samples.

[0072] Furthermore, in practical applications, the system inputs the image of the voucher into a pre-trained coordinate recognition model. The model analyzes the image features through deep learning algorithms and accurately predicts the top corner coordinates of the target voucher (business license) in the image. This can include the coordinates of the top left and bottom right corners, or generate the coordinates of the four top corners.

[0073] Furthermore, based on the vertex coordinate information, the system can calculate the coverage area of ​​the target voucher in the image, i.e., the bounding box of the voucher. This calculation process involves determining the distance and angle between coordinate points to accurately delineate the image area of ​​the target voucher. At this point, the system can use the determined bounding box information to crop out an independent image of the target voucher from the original voucher image, obtaining a complete voucher image.

[0074] For example, the coordinates of a business license identified by a coordinate recognition model could be:

[0075] ;

[0076] Where Z represents the identified business license target area, expressed as the x-coordinate of the top left corner of the area. y-axis and the x-coordinate of the bottom right corner y-axis Record it. This represents a target recognition deep neural network with a parameter set of N. The input of this deep neural network is a surveillance video frame image, and the output is the result of recognizing a business license in the surveillance video frame image. It can be any deep neural network with target recognition capabilities. A confidence threshold is set for identifying targets related to business licenses, used to filter out low-confidence identification results. In one embodiment, the value is set to... =0.5.

[0077] This embodiment identifies the top corner coordinates of the target voucher using a coordinate recognition model, and then crops the voucher image based on the top corner coordinates, achieving the technical effect of accurately obtaining a complete voucher image.

[0078] To construct a more accurate appearance recognition model, optionally, in the abnormal certificate detection method provided in this application embodiment, the appearance recognition model is trained in the following manner: M historical certificate images are acquired, and the M historical certificate images are occluded or damaged to obtain N sample certificate images, wherein the appearance recognition label of the historical certificate images is no abnormality, and the appearance recognition label of the sample certificate images is abnormal, M and N are both positive integers, and N is greater than or equal to M; the M historical certificate images and N sample certificate images, as well as the appearance recognition label of each historical certificate image and the appearance recognition label of each sample certificate image are determined as a second training set, and the initial neural network model is trained through the second training set to obtain the appearance recognition model.

[0079] It should be noted that historical document images refer to genuine images of bank branch business licenses that have not been manually modified. These images are initially marked as having no visual abnormalities. Obscuring or damaging operations refer to simulating visual abnormalities such as obscuring, tearing, or smearing on historical document images to create sample document images with abnormality labels. Visual identification labels are used to mark the appearance of document images and are divided into two categories: no abnormalities and abnormalities.

[0080] For example, when training the appearance recognition model, training samples need to be obtained. First, the system filters and collects M business license images that have not been modified by anyone from the historical data of bank branch surveillance videos. These images all show that the business licenses are intact and unobstructed, and are marked as having no abnormalities in appearance recognition.

[0081] Furthermore, in order to simulate various abnormal situations that may occur during the hanging of vouchers, the system performs a series of occlusion or damage operations on these M historical voucher images, such as deliberately adding occlusions to the images, simulating damage, wear, etc., to create N sample voucher images, and the appearance identification label of these images is abnormal.

[0082] Furthermore, the M historical certificate images and N sample certificate images, along with their respective appearance recognition labels (no abnormality or abnormality), are integrated into a second training set. The training set contains both positive and negative samples, enabling the model to learn and distinguish between normal and abnormal business license appearances in diverse contexts.

[0083] After determining the training set, an initial deep neural network model can be selected as the starting point for the appearance recognition model. This model can be a convolutional neural network architecture, consisting of a series of convolutional layers, residual network blocks, fully connected layers, and an output layer. The convolutional layers are used to extract basic features of the image, the residual network blocks help the model learn complex features in depth, the fully connected layers are used for feature integration, and the output layer is responsible for making classification decisions based on the learned features. Thus, by training this model, it can automatically extract image features and perform target recognition or classification.

[0084] It should be noted that the appearance recognition model trained in this embodiment can also be composed of multiple sub-models. Each sub-model can correspond to a different appearance anomaly. For example, sub-model 1 corresponds to damage, sub-model 2 corresponds to occlusion, etc. The various sub-models in the appearance recognition model can be plugged in or removed according to the detection requirements, so that the appearance recognition model has different recognition functions.

[0085] This embodiment achieves the technical effect of improving the accuracy of model recognition results by constructing sample data in the training set and training the model.

[0086] Optionally, in the abnormal document detection method provided in this application embodiment, inputting the document image into the appearance recognition model to obtain the appearance recognition result of the target document includes: inputting the document image into the appearance recognition model to obtain a first recognition result; determining a reference edge in the border of the document image and determining a comparison edge of the document image corresponding to the reference edge; determining the angle value between the reference edge and the comparison edge, and determining a second recognition result of the document image based on the angle value; determining the appearance recognition result as normal if both the first recognition result and the second recognition result indicate no abnormality; and determining the appearance recognition result as abnormal if the first recognition result and / or the second recognition result indicate an abnormality.

[0087] It should be noted that the first recognition result is a preliminary result of the appearance recognition model analyzing the document image to identify whether the business license has physical damage. The reference edge is a selected edge in the document image, used as a reference for subsequent angle measurements. The comparison edge is another edge corresponding to the reference edge, used to measure the tilt angle. The angle value is the angle formed between the reference edge and the comparison edge, used to assess the degree of tilt of the business license. The second recognition result is the result of judging whether the business license is tilted based on the angle value.

[0088] For example, when inspecting the appearance of a target document, it is necessary to determine whether the target document has defects such as damage or obstruction that affect the recognition of the document content, and whether the target document has placement defects such as tilting. The document image is first transmitted to the appearance recognition model, which uses a deep neural network to analyze image features to identify whether there is physical damage on the surface of the business license, such as tearing, abrasion, or breakage.

[0089] Simultaneously, the system also needs to assess the hanging tilt of the business license. First, a reference edge is determined within the frame of the document image, usually selected as the top or bottom of the business license. Then, the system determines the corresponding comparison edge, i.e., the other side of the business license, to calculate the angle between the reference edge and the reference edge. Further, the system measures the angle value between the reference edge and the comparison edge, and then calculates the angle between the two straight lines. Finally, it can determine whether the angle value exceeds the range of normal hanging state based on a preset tilt threshold. If the angle value is within the threshold, the second recognition result is no abnormality; otherwise, the result is abnormal, indicating that the business license is hanging tilted.

[0090] Finally, after obtaining the first and second identification results, the system performs a comprehensive analysis of the first identification result (physical damage state) and the second identification result (suspension tilt state). If both identification results indicate no abnormality, the appearance identification result is deemed normal; conversely, if either identification result is abnormal, the comprehensive appearance identification result is determined to be abnormal.

[0091] This embodiment detects the appearance of the target document from two dimensions, thereby improving the accuracy and comprehensiveness of the detection results.

[0092] To accurately obtain the content recognition result, optionally, in the abnormal credential detection method provided in this application embodiment, determining the content recognition result of the target credential based on the text information includes: obtaining preset information of the target credential, comparing the preset information with the text information to obtain text similarity, and determining a third recognition result based on the text similarity; recognizing the expiration date data in the text information, judging whether the expiration date data is valid, and determining a fourth recognition result based on the judgment result; if both the third and fourth recognition results indicate no abnormality, determining the content recognition result as no abnormality; if the third and / or fourth recognition results indicate abnormality, determining the content recognition result as abnormal.

[0093] It should be noted that the preset information can refer to standard business document information recorded in the banking system, including but not limited to the company name and unified social credit code. The expiration date data can be the validity period information on the business license, used to determine whether the license has expired.

[0094] For example, when determining the content recognition result of the target voucher, it is necessary to determine it from two dimensions. First, the preset information of the target voucher (business license) is retrieved from the internal database. This information is pre-entered and verified to be correct. Then, a deep learning model is used to analyze the text information recognized from the voucher image and extract key fields, including but not limited to the company name and unified social credit code.

[0095] Furthermore, the system compares the identified text information with the preset information word by word, calculates the text similarity to assess the degree of matching between the two texts, and determines the third recognition result based on the text similarity. If the similarity is higher than the preset threshold, the third recognition result is no abnormality; otherwise, if the similarity is low, it indicates that there is a significant difference between the text information and the preset information, and the third recognition result is judged as abnormal.

[0096] For example, it is also necessary to identify and extract the validity period data of the target certificate from the text information, and compare the validity period data with the current system time to determine whether the business license is within its validity period. If the validity period data indicates that the target certificate is valid, the fourth identification result is no anomaly; otherwise, if the target certificate has expired, or is about to expire and has not been updated in advance, the fourth identification result is marked as anomaly.

[0097] Finally, the system comprehensively analyzes the third and fourth identification results. Only when both are determined to be without anomalies is the content identification result confirmed as without anomalies, indicating that the business license information is accurate and valid. If either the third or fourth identification result is determined to be abnormal, the content identification result will also be determined to be abnormal, indicating that there may be information errors or an expired license.

[0098] This embodiment verifies the accuracy and timeliness of the voucher content, achieving the technical effect of accurately determining the text recognition result.

[0099] To accurately obtain text information, optionally, in the abnormal voucher detection method provided in this application embodiment, identifying text information in the voucher image includes: scanning the voucher image to obtain initial text data of the voucher image; determining the attribute information of each initial text data, and filtering the initial text data according to preset filtering rules and attribute information to obtain filtered initial text data; and determining the filtered initial text data as text information.

[0100] It should be noted that the initial text data can be a set of text information initially extracted from the voucher image using optical character recognition technology, which may contain irrelevant or redundant content. Attribute information can be additional descriptive information of the initial text data, such as position, size, font, and shape, used to assist in filtering and identifying key fields. Pre-defined filtering rules can be a set of predefined rules used to determine whether the initial text data meets the characteristics of the target voucher's key fields, including but not limited to field length, position distribution, and information type.

[0101] For example, when recognizing text information in a voucher image, the voucher image is first scanned using optical character recognition through a deep learning model. The model can extract image features through a convolutional neural network, then locate the text region through a text prediction layer, and recognize the text content to form a series of initial text data.

[0102] Furthermore, for the initially extracted text data, the system will further analyze the attribute information of each data point, including the position coordinates of the text, font size, shape features, and information type (numbers, letters, or special symbols), in order to determine whether it matches the expected attributes of the key fields.

[0103] Furthermore, after obtaining the attribute information, the attribute information of each initial text data can be compared with the preset filtering rules to eliminate non-critical, redundant, or rule-incompatible data, retaining the initial text data that best matches the key field attributes. This data not only lies within the preset area but also conforms to the field's format and information type requirements.

[0104] Ultimately, the system determines the filtered initial text data set as text information, which is a clean dataset containing core fields such as company name, unified social credit code, and validity period, and obtains the text information of the target voucher.

[0105] This embodiment achieves the technical effect of improving the accuracy and efficiency of text information acquisition by automating and accurately identifying and filtering key information of vouchers, thereby improving the accuracy of subsequent identification of abnormal status of target vouchers.

[0106] To accurately determine the abnormal state of the target credential, optionally, in the abnormal credential detection method provided in this application embodiment, determining the abnormal state of the target credential based on the appearance recognition result and the content recognition result includes: when both the appearance recognition result and the content recognition result indicate no abnormality, determining the abnormal state of the target credential as no abnormality; when the appearance recognition result and / or the content recognition result indicate an abnormality, determining the abnormal state of the target credential as abnormal.

[0107] For example, after obtaining the appearance recognition result and the content recognition result, the system first checks the appearance recognition result output by the appearance recognition model to confirm whether the target credential is properly hung, i.e., without physical damage, obstruction, or tilting. Subsequently, based on the content recognition result output by the content recognition module, the system confirms the accuracy of key information on the target credential, including the authenticity of the information and whether it is within its validity period.

[0108] Furthermore, the system performs a comprehensive analysis of the appearance recognition results and content recognition results, determining the final abnormal status of the target document based on whether both indicate no abnormalities. If the appearance recognition results show that the target document has no physical damage, obstruction, or tilting, and the content recognition results confirm that the document information is authentic and within its validity period, then the system determines the abnormal status of the target document to be no abnormality, indicating that the document complies with regulatory requirements. Conversely, if either the appearance recognition result or the content recognition result indicates an abnormality, i.e., the document has physical damage, obstruction, tilting, or information inconsistencies, or is expired, then the system determines the abnormal status of the target document to be abnormal, prompting the need for further manual verification or immediate corrective measures.

[0109] This embodiment comprehensively analyzes the appearance recognition results and content recognition results to fully evaluate and detect the abnormal status of the target voucher, thereby improving the accuracy of the voucher detection results.

[0110] Figure 3 This is a schematic diagram of an optional abnormal certificate detection system provided in Embodiment 1 of this application, as shown below. Figure 3 As shown, the system includes a video data interface module, a business license area detection module, a hanging status analysis module, and a configuration module. The system can detect the display status of business licenses hanging in bank branches in real time, thereby improving the efficiency and timeliness of business license display status detection.

[0111] Figure 4This is an execution flowchart of the abnormal credential detection system provided in Embodiment 1 of this application, as follows: Figure 4 As shown:

[0112] The configuration module is used to configure the parameters required to execute the above-mentioned abnormal certificate detection method. The video data docking module is used to dock with the bank branch surveillance camera, receive and decode the video stream. The business license area detection module is used to perform business license detection on the video frames read by the video data docking module. The hanging status analysis module is used to determine whether the business license is correctly hung and within the validity period.

[0113] The video data interface module is used to interface with video data from surveillance cameras in bank branches. It connects to existing surveillance cameras in bank branches via a preset protocol, receiving and decoding video streams in formats such as H.264 / H.265 in real time. Considering that business licenses are static objects, the system does not need to analyze them frame by frame. Instead, by setting a video frame extraction time interval, it extracts frames from the input video stream to obtain keyframes for analysis, represented as follows:

[0114] ;

[0115] in, This is a video keyframe extraction function, where S is the video stream address. In one embodiment, the preset video frame extraction time interval is used. =1, where V is the extracted video frame image. Video keyframe extraction effectively balances computational overhead and real-time performance.

[0116] The business license area detection module is used to detect business licenses in video frames read by the video data integration module. The business license area detection module mainly includes two sub-modules: target recognition and area cropping.

[0117] In the target recognition submodule, for each video frame read by the video data interface module, a deep neural network is used to identify the business license in the video frame image, and the recognition result is obtained, represented as:

[0118] ;

[0119] Where Z represents the identified business license target area, expressed as the x-coordinate of the top left corner of the area. y-axis and the x-coordinate of the bottom right corner y-axis Record it. This represents a target recognition deep neural network with a parameter set of N. The input of this deep neural network is a surveillance video frame image, and the output is the result of recognizing a business license in the surveillance video frame image. It can be any deep neural network with target recognition capabilities. A confidence threshold is set for identifying targets related to business licenses, used to filter out low-confidence identification results. In one embodiment, the value is set to... =0.5.

[0120] The region cropping submodule primarily crops the business license region from the video frame image based on the business license region identified in the target recognition step, obtaining a business license image, represented as:

[0121] P={w,h}=G p (V, Z);

[0122] in, This represents an image cropping function, whose inputs are a video frame image V and the target region. The output is the cropped business license image P, where w and h are the width and height of the business license image, respectively, satisfying: w=|x2-x1|, h=|y2-y1|.

[0123] Figure 5 The flowchart below shows the execution process of the business license area detection module according to Embodiment 1 of this application. Figure 5 As shown, when performing business license area detection using the business license area detection module, the target recognition sub-model first uses a deep neural network to identify the business license in each video frame image read by the video data interface module, obtaining the recognition result. Further, the area cropping sub-module crops the business license area from the video frame image based on the business license area identified by the target recognition sub-model, obtaining the business license image.

[0124] The hanging status analysis module is used to evaluate the hanging quality of business licenses, determine whether the business licenses are hung correctly, and whether the business licenses are within their validity period. Figure 6 This is a schematic diagram of the suspension state analysis module provided in Embodiment 1 of this application, as shown below. Figure 6 As shown, the suspension status analysis module mainly includes an integrity analysis submodule and a content compliance analysis submodule. The integrity analysis submodule includes a damage detection subunit, an occlusion detection subunit, and a tilt detection subunit. The content compliance analysis submodule includes a content recognition subunit and a compliance judgment subunit. The compliance judgment subunit includes a content authenticity judgment subbranch and a validity judgment subbranch.

[0125] Figure 7 This is an execution flowchart of the suspension state analysis module provided in Embodiment 1 of this application, as follows: Figure 7As shown, the integrity analysis submodule is used to identify whether the business license has problems such as damage, obstruction by foreign objects, or tilting. The content compliance analysis submodule is used to detect whether the business license is genuine and whether it is within the validity period. The integrity analysis submodule and the content compliance analysis submodule are executed in parallel.

[0126] The integrity analysis submodule is used to identify whether a business license has issues such as damage, obstruction by foreign objects, or tilting. The integrity analysis submodule mainly consists of three parallel subunits: a damage detection subunit, an obstruction detection subunit, and a tilt detection subunit. These subunits are designed with a plug-and-play approach, allowing for flexible addition or removal. The integrity analysis submodule is represented as follows:

[0127] C c =G c ({F p F z F q},P);

[0128] Among them, C c (·) refers to the integrity analysis submodule, whose inputs mainly include pluggable sub-unit combinations and the business license image P obtained from the business license area detection module. p (·) represents the damage detection subunit, F z (·) represents the occlusion detection subunit, F q (·) represents the oblique detection subunit. C c =(c c ) represents the detection results of the integrity analysis submodule, c c ∈{0,1} represents the result category, mainly including two categories: 0 and 1, where 0 indicates failure and 1 indicates success. The integrity analysis submodule obtains the final result by integrating the results of all its sub-units. When the damage detection sub-unit result is no damage, the occlusion detection sub-unit result is no occlusion, and the tilt detection sub-unit result is no tilt, then the integrity analysis submodule's result is success, i.e., c c =1, otherwise the result is failure, i.e., c c =0, which means:

[0129] ;

[0130] Among them, c p c z and c q These are the result categories for the damage detection subunit, the occlusion detection subunit, and the tilt detection subunit, respectively.

[0131] In the damage detection subunit, the business license image obtained by the business license area detection module is used to identify whether there is damage in the business license image through a deep neural network classification model, represented as follows:

[0132] ;

[0133] in, The results of the damage detection subunit mainly include result category c. p ∈{0,1} and result confidence Result category c p It mainly includes two categories: 0 and 1, where 0 indicates damage and 1 indicates no damage. F p (·) represents the damage detection subunit, which can be any deep neural network with image classification capabilities. In one embodiment, F p (·) can be implemented using an image classification network. F p The input to (·) includes the business license image P and the confidence threshold ρ. When F p If the image classification model identifies the image as damaged (·) and the confidence level is greater than ρ, then the final result category is damaged, i.e., c. p =0. In one embodiment, the value ρ is 0.5.

[0134] In the occlusion detection subunit, the business license image obtained by the business license area detection module is used to identify whether there is an occlusion problem in the business license image through a deep neural network classification model, represented as follows:

[0135] ;

[0136] Among them, C z The results of the occlusion detection subunit mainly include result category c. z ∈{0,1} and result confidence Result category c z It mainly includes two categories: 0 and 1, where 0 represents occlusion and 1 represents no occlusion. F z (·) represents the occlusion detection subunit, which can be any deep neural network with image classification capabilities. In one embodiment, F z (·) This is implemented using an image classification network. F z The inputs for (·) include the business license image P and the confidence threshold. When F z The image classification model (·) identifies it as occlusion, and the confidence level is greater than 0.5%. When the result is occlusion, the final result category is occlusion, i.e., c. z =0. In one embodiment, the value is... =0.5.

[0137] In the tilt detection subunit, the tilt detection algorithm is used to detect whether the business license image obtained by the business license area detection module has a tilt problem, as shown below:

[0138] ;

[0139] Among them, C q The results for the tilt detection subunit mainly include result category c. q ∈{0,1} and result skewness Result category c q It mainly includes two categories: 0 and 1, where 0 indicates tilt and 1 indicates no tilt. F z (·) represents the tilt detection sub-unit, implemented by the tilt detection algorithm. The tilt angle of the business license edge is calculated using the inverse tangent function, expressed as:

[0140] Fq = arctan(|h| / |w|) × 180 / π;

[0141] Where arctan(·) is the inverse tangent function, w and h are the width and height of the business license image P, respectively. θ is the tilt threshold, that is, when the tilt of the business license is greater than θ degrees, it is judged as tilted. In one embodiment, the value of θ = 15 is taken.

[0142] Figure 8 This is an execution flowchart of the integrity analysis submodule provided in Embodiment 1 of this application, as follows: Figure 8 As shown, the damage detection subunit uses a deep neural network classification model to identify whether the business license image is damaged, the occlusion detection subunit uses a deep neural network classification model to identify whether the business license image is occluded, and the tilt detection subunit uses a tilt detection algorithm to detect whether the business license image is tilted.

[0143] The content compliance analysis submodule is used to check whether the business license is genuine and valid. Figure 9 This is an execution flowchart of the content compliance analysis submodule provided in Embodiment 1 of this application, as follows: Figure 9 As shown, the content compliance analysis submodule mainly includes two sub-units: the content recognition sub-unit and the compliance judgment sub-unit. The content recognition sub-unit is used to perform text recognition on the business license image using optical character recognition technology to extract key fields. The compliance judgment sub-unit is used to judge the compliance of the business license, including the content authenticity judgment sub-branch and the date validity judgment sub-branch.

[0144] In the content recognition subunit, the business license image obtained by the business license area detection module is processed using optical character recognition (OCR) technology to extract key fields, which may include the company name, unified social credit code, and validity period, as shown below:

[0145] C o ={t1, ...,t}i}=F o (P, ο);

[0146] Among them, C o ={t1, ...,t} i} represents the collection of text content extracted from the business license image, t i That is, the i-th segment of text. F o (·) represents a deep neural network for optical character recognition. The input to this deep neural network is a business license image, and the output is a set of extracted text content from the business license image. F o (·) can be any deep neural network with optical character recognition capabilities. ο is the confidence threshold for the optical character recognition deep neural network, and in one embodiment, the value ο = 0.4.

[0147] In the compliance judgment subunit, the compliance of the business license is judged based on the recognition results obtained in the content recognition subunit. This mainly includes two sub-branches: content authenticity judgment and date validity judgment. These two sub-branches are designed as pluggable components, allowing for flexible addition or removal. The content recognition subunit is represented as follows:

[0148] C h =G h ({F a F e}, C o );

[0149] Among them, C h (·) refers to the compliance judgment subunit, whose input mainly includes pluggable sub-branch combinations and the set of text content C obtained from the business license image in the content recognition subunit. o F a (·) represents the sub-branch for judging the authenticity of the content, F e ( The first branch is for date validity checks. h =(c h ) represents the detection result of this sub-unit, c h ∈{0,1} represents the result category, mainly including two categories: 0 and 1, where 0 indicates failure and 1 indicates success. The compliance judgment subunit obtains the final result by combining the results of all its sub-branches. When the content authenticity judgment sub-branch result is "content is authentic" and the date validity judgment sub-branch result is "within the validity period," then the result of the compliance judgment subunit is "pass," i.e., c. h =1, otherwise the result is failure, i.e., c h =0, which means:

[0150] ;

[0151] Among them, ca and c e The result categories are the content authenticity judgment sub-branch and the date validity judgment sub-branch, respectively.

[0152] Figure 10 This is an execution flowchart of the compliance judgment subunit provided in Embodiment 1 of this application, as follows: Figure 10 As shown, the content authenticity judgment sub-branch compares the recognition result obtained by the content recognition sub-unit with the configured business license standard information to verify the authenticity of the license. The date validity judgment sub-branch extracts the business license validity period from the recognition result obtained by the content recognition sub-unit to determine whether the business license is within the validity period.

[0153] In the content authenticity judgment sub-branch, the recognition result obtained from the content recognition sub-unit is compared with the configured business license standard information to verify the authenticity of the license, as shown below:

[0154] C a =(c a s a )=F a (C o C config , α);

[0155] Among them, C a The results of the content authenticity judgment sub-branch mainly include result category c. a ∈{0,1} and result matching degree Result category c a It mainly includes two categories: 0 and 1, where 0 indicates that the content is untrue and 1 indicates that the content is true. config This refers to the configured business license standard information. α is the matching threshold, when c a and C config When the matching degree is higher than α, the content is determined to be true; otherwise, the content is determined to be untrue. In one embodiment, the value α is 0.8.

[0156] In the date validity judgment sub-branch, the business license validity period is extracted from the recognition results obtained in the content recognition sub-unit. It is then compared with the current date to determine whether the business license is still valid, as shown below:

[0157] C e =(c e )=F e (C o T now );

[0158] Among them, C e The results of the date validity judgment sub-branch mainly include result category c. e ∈{0,1}, result category ce It mainly includes two categories: 0 and 1, where 0 indicates that it is not within the validity period, and 1 indicates that it is within the validity period. now Indicates the current date.

[0159] It should be noted that this system also requires a configuration module to configure the parameters necessary for executing the above process, including the video stream address S, and various thresholds, including the video frame extraction time interval τ, the confidence threshold ε for business license target recognition, the confidence threshold ρ for the damage detection subunit, and the confidence threshold for the occlusion detection subunit. The tilt detection subunit tilt threshold θ, the content recognition subunit confidence threshold ο, the content authenticity judgment sub-branch matching threshold α, and the configured business license standard information C config wait.

[0160] It should be noted that, Figure 11 This is a schematic diagram of the detection platform of the detection system for the deployment of abnormal credentials provided in Embodiment 1 of this application, as shown below. Figure 11 As shown, the abnormal credential detection system can be connected to the data management module, model management module, and configuration management module. The data management module can record and preprocess the images sent by the surveillance camera and send the processed video stream to the abnormal credential detection system. The model management module can manage and provide the models configured in the abnormal credential detection system. The configuration management module can manage and provide the configuration data configured in the abnormal credential detection system.

[0161] It should be noted that the data management module and the model management module can also be connected to the model training module. The data management module can provide training set data to the model training module, and the model training module can store the model in the model management module for management after completing the model training.

[0162] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0163] Example 2

[0164] This application also provides an abnormal credential detection device. It should be noted that the abnormal credential detection device of this application can be used to execute the abnormal credential detection method provided in the above embodiments. The abnormal credential detection device provided in this application is described below.

[0165] According to an embodiment of this application, an apparatus for implementing the above-described method for detecting abnormal credentials is also provided. Figure 12This is a schematic diagram of an abnormal certificate detection device according to Embodiment 2 of this application, as shown below. Figure 12 As shown, the device includes:

[0166] The first identification unit 121 is used to acquire the image of the target voucher according to a preset acquisition frequency, and to identify the voucher image of the target voucher from the voucher image.

[0167] The first determining unit 122 is used to identify text information in the voucher image and determine the content recognition result of the target voucher based on the text information.

[0168] The second recognition unit 123 is used to input the voucher image into the appearance recognition model to obtain the appearance recognition result of the target voucher.

[0169] The second determining unit 124 is used to determine the abnormal state of the target credential based on the appearance recognition result and the content recognition result.

[0170] The abnormal document detection device provided in this application collects the document image of the target document and identifies text information based on the document image. Then, it performs abnormal detection on the target document from two dimensions: content and appearance, and obtains the detection result of the target document. This achieves the goal of automatically and accurately detecting the target document whenever it needs to be checked, thereby improving the efficiency and timeliness of document detection and solving the technical problem of low efficiency and timeliness of abnormal checks on bank business documents in related technologies.

[0171] Optionally, in the abnormal voucher detection device provided in this application embodiment, the first identification unit 121 includes: a first identification module, used to input the voucher image into a coordinate recognition model to obtain the coordinate information of the target voucher in the voucher image, wherein the coordinate information includes the vertex coordinates of the target voucher, and the coordinate recognition model is obtained by training an initial neural network model with a first training set, the first training set consisting of multiple first sample data, each first sample data including the voucher image of a sample voucher and the coordinate information of the sample voucher; and a first determination module, used to determine the coverage area of ​​the target voucher in the voucher image according to the coordinate information, and to crop the coverage area to obtain a voucher image.

[0172] Optionally, in the abnormal certificate detection device provided in this application embodiment, the appearance recognition model is trained by the following device: an acquisition unit, used to acquire M historical certificate images and perform occlusion or damage operations on the M historical certificate images to obtain N sample certificate images, wherein the appearance recognition label of the historical certificate images is no abnormality, and the appearance recognition label of the sample certificate images is abnormal, M and N are both positive integers, and N is greater than or equal to M; a training unit, used to determine the M historical certificate images and N sample certificate images, as well as the appearance recognition label of each historical certificate image and the appearance recognition label of each sample certificate image as a second training set, and train the initial neural network model through the second training set to obtain the appearance recognition model.

[0173] Optionally, in the abnormal document detection device provided in this application embodiment, the second identification unit 123 includes: a second identification module, used to input the document image into the appearance recognition model to obtain a first identification result; a second determination module, used to determine a reference edge in the border of the document image and determine a comparison edge of the document image corresponding to the reference edge; a third determination module, used to determine the angle value between the reference edge and the comparison edge, and determine the second identification result of the document image based on the angle value; a fourth determination module, used to determine that the appearance recognition result is normal when both the first identification result and the second identification result indicate no abnormality; and a fifth determination module, used to determine that the appearance recognition result is abnormal when the first identification result and / or the second identification result indicate an abnormality.

[0174] Optionally, in the abnormal credential detection device provided in this application embodiment, the first determining unit 122 includes: an acquisition module, used to acquire preset information of the target credential, compare the preset information with text information to obtain text similarity, and determine a third identification result based on the text similarity; a judgment module, used to identify the expiration date data in the text information, judge the expiration date data to determine whether the target credential is valid, and determine a fourth identification result based on the judgment result; a sixth determining module, used to determine that the content identification result is not abnormal when both the third and fourth identification results indicate no abnormality; and a seventh determining module, used to determine that the content identification result is abnormal when the third and / or fourth identification results indicate abnormality.

[0175] Optionally, in the abnormal voucher detection device provided in this application embodiment, the first determining unit 122 includes: a scanning module, used to scan the voucher image to obtain the initial text data of the voucher image; a filtering module, used to determine the attribute information of each initial text data respectively, and filter the initial text data according to the preset filtering rules and attribute information to obtain the filtered initial text data; and an eighth determining module, used to determine the filtered initial text data as text information.

[0176] Optionally, in the abnormal credential detection device provided in this application embodiment, the second determining unit 124 includes: a ninth determining module, used to determine the abnormal state of the target credential as no abnormality when both the appearance recognition result and the content recognition result indicate no abnormality; and a tenth determining module, used to determine the abnormal state of the target credential as abnormal when the appearance recognition result and / or the content recognition result indicate abnormality.

[0177] It should be noted that the first identification unit 121, the first determination unit 122, the second identification unit 123, and the second determination unit 124 mentioned above correspond to steps S201 to S204 in Embodiment 1. The instances and application scenarios implemented by each of the above units and the corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1. It should be noted that the above modules or units can be hardware components or software components stored in memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n). The above modules can also be part of a device and can run in the computer terminal 10 provided in Embodiment 1.

[0178] Example 3

[0179] Embodiments of this application may provide an electronic device. Figure 13 This is a structural block diagram of an electronic device according to an embodiment of this application. Figure 13 As shown, the electronic device may include: one or more ( Figure 13 (Only one is shown) processor 1002, memory 1004, memory controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0180] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the methods and apparatus in the embodiments of this application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby implementing the above-described methods. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0181] Those skilled in the art will understand that Figure 13The structure shown is for illustrative purposes only. Electronic devices can also be smartphones, tablets, handheld computers, mobile internet devices (MIDs), PADs, and other terminal devices. Figure 13 This does not limit the structure of the aforementioned electronic device. For example, electronic devices may also include components that are more... Figure 13 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 13 The different configurations shown.

[0182] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0183] Example 4

[0184] Embodiments of this application also provide a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the abnormal credential detection method provided in Embodiment 1.

[0185] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0186] Embodiments of this application also provide a computer program product, which, when executed on a data processing device, is adapted to perform the steps of a method for detecting abnormal credentials.

[0187] Embodiments of this application also provide a computer-readable storage medium, which includes a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to execute the above-described method for detecting abnormal credentials.

[0188] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0189] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0190] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units 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 displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0191] The units described 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.

[0192] 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.

[0193] 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 several 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 described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0194] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for detecting abnormal vouchers, characterized in that, include: Acquire the image of the target voucher at a preset acquisition frequency, and identify the voucher image of the target voucher from the voucher image; Identify the text information in the voucher image, and determine the content recognition result of the target voucher based on the text information; The image of the voucher is input into the appearance recognition model to obtain the appearance recognition result of the target voucher; The abnormal state of the target credential is determined based on the appearance recognition result and the content recognition result.

2. The method according to claim 1, characterized in that, The document image for identifying the target document from the document image includes: The image of the voucher is input into the coordinate recognition model to obtain the coordinate information of the target voucher in the voucher image. The coordinate information includes the vertex coordinate of the target voucher. The coordinate recognition model is obtained by training an initial neural network model with a first training set. The first training set consists of multiple first sample data, and each first sample data includes the voucher image of a sample voucher and the coordinate information of the sample voucher. The target voucher is determined to cover an area in the voucher image based on the coordinate information, and the covered area is cropped to obtain the voucher image.

3. The method according to claim 1, characterized in that, The appearance recognition model is trained in the following manner: Get M historical voucher images, and perform occlusion or damage operations on the M historical voucher images to obtain N sample voucher images. The appearance identification label of the historical voucher images is no abnormality, and the appearance identification label of the sample voucher images is abnormal. M and N are both positive integers, and N is greater than or equal to M. The M historical voucher images and the N sample voucher images, along with the appearance recognition labels for each historical voucher image and each sample voucher image, are determined as the second training set. The initial neural network model is then trained using the second training set to obtain the appearance recognition model.

4. The method according to claim 1, characterized in that, Inputting the image of the voucher into the appearance recognition model to obtain the appearance recognition result of the target voucher includes: The image of the credential is input into the appearance recognition model to obtain a first recognition result; A reference edge is determined within the border of the voucher image, and a comparison edge of the voucher image corresponding to the reference edge is determined. Determine the angle value between the reference edge and the comparison edge, and determine the second recognition result of the voucher image based on the angle value; If both the first identification result and the second identification result indicate no abnormality, the appearance identification result is determined to be without abnormality. If the first recognition result and / or the second recognition result indicate an anomaly, the appearance recognition result is determined to be abnormal.

5. The method according to claim 1, characterized in that, The content recognition result of the target credential determined based on the text information includes: Obtain preset information of the target credential, compare the preset information with the text information to obtain text similarity, and determine a third recognition result based on the text similarity; Identify the expiration date data in the text information, determine whether the target voucher is valid based on the expiration date data, and determine the fourth identification result based on the determination result; If both the third and fourth identification results indicate no abnormality, the content identification result is determined to be without abnormality. If the third identification result and / or the fourth identification result indicate an anomaly, the content identification result is determined to be anomaly.

6. The method according to claim 1, characterized in that, Identifying the text information in the voucher image includes: The voucher image is scanned to obtain the initial text data of the voucher image; The attribute information of each initial text data is determined, and the initial text data is filtered according to the preset filtering rules and the attribute information to obtain the filtered initial text data. The filtered initial text data is determined as the text information.

7. The method according to claim 1, characterized in that, Determining the abnormal state of the target credential based on the appearance recognition result and the content recognition result includes: If both the appearance recognition result and the content recognition result indicate no abnormality, the abnormality status of the target credential is determined to be no abnormality. If the appearance recognition result and / or the content recognition result indicate an anomaly, the abnormal state of the target credential is determined to be abnormal.

8. A device for detecting abnormal vouchers, characterized in that, include: The first identification unit is used to acquire the image of the target voucher at a preset acquisition frequency, and to identify the voucher image of the target voucher from the voucher image; The first determining unit is used to identify text information in the voucher image and determine the content recognition result of the target voucher based on the text information; The second recognition unit is used to input the voucher image into the appearance recognition model to obtain the appearance recognition result of the target voucher. The second determining unit is used to determine the abnormal state of the target credential based on the appearance recognition result and the content recognition result.

9. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the abnormal credential detection method according to any one of claims 1 to 7.

10. An electronic device, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program, when running, performs the method for detecting abnormal credentials as described in any one of claims 1 to 7.

Images