An audit method and system for intelligently screening abnormal financial data

Abstract

The application belongs to the field of production and the field of artificial intelligence system research and development, and discloses an audit abnormal financial data intelligent screening method and system; audit client collects uploaded audit project financial data, and obtains financial bill image features through processing of scanned document images, constructs an audit abnormal financial data intelligent screening model by using the audit project financial data, the financial bill image features and corresponding audit abnormal financial data screening methods, and uses the model to generate a new audit abnormal financial data screening method to determine the financial data abnormal screening, the application uses audit financial document image extraction technology to obtain accurate financial bill image features, and uses a cross-modal neural network model to accurately generate the screening method for the scene of audit abnormal financial data intelligent screening, thereby assisting in reducing the artificial screening time of auditors, reducing the audit time cost, and improving the accuracy of audit abnormal financial data screening.

Description

Technical Field

[0001] This invention belongs to the field of intelligent audit system research and development technology, and in particular relates to an intelligent screening method and system for abnormal financial data in audits. Background Technology

[0002] With the deepening of enterprise digital transformation, financial data is experiencing explosive growth, posing a severe challenge to traditional auditing methods. Traditional auditing mainly relies on manual data verification and sampling methods, which are inefficient, time-consuming, and dependent on human experience, making it difficult to cope with the complex and ever-changing financial environment of modern enterprises. In particular, traditional methods can no longer meet the actual needs of enterprise financial management in areas such as massive data processing, abnormal transaction identification, and real-time risk warning.

[0003] Against this backdrop, the rapid development of artificial intelligence and big data technologies has brought about a revolutionary change in financial auditing. AI-powered financial anomaly detection technology, through algorithms such as machine learning, deep learning, and natural language processing, can intelligently analyze and recognize patterns in massive amounts of financial data, enabling automatic detection and risk warning of abnormal transactions. This intelligent auditing approach not only significantly improves audit efficiency and accuracy but also achieves full data coverage, breaking through the limitations of traditional sampling audits and providing enterprises with more comprehensive and precise risk control measures for financial management.

[0004] Therefore, existing technologies have the following problems: how to select common fields for audit projects, how to process regular financial document images to obtain features closely related to audit financial anomalies, how to construct an intelligent screening model for audit anomaly financial data, how to use the model to generate audit anomaly financial data screening methods in real time to determine financial data anomalies, and how to use the generated classification result data to determine audit anomaly financial data screening methods, thereby helping to reduce the manual screening time of auditors, reduce audit time costs, and improve the accuracy of anomaly financial data screening during audits. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes an intelligent screening method and system for auditing abnormal financial data.

[0006] In a first aspect of the present invention, a method for intelligent screening of audit anomaly financial data is provided, the method comprising: S1. Collect and upload financial data of audit projects through the audit client, obtain financial document image features by processing the financial document images collected by document scanning, and obtain the corresponding audit abnormal financial data screening method by the auditors through tagging. S2. Cross-modal data processing is performed using the financial data of the audit project and the image features of the financial documents to obtain the audit abnormal financial features; S3. Receive and construct an intelligent screening model for audit abnormal financial data based on the aforementioned audit abnormal financial characteristics and the aforementioned audit abnormal financial data screening method; S4. Using the intelligent screening model for abnormal financial data in audits, a real-time screening method for abnormal financial data in audits is obtained by screening abnormal financial data for different audit projects. The real-time screening method for abnormal financial data in audits is then used to intelligently screen abnormal financial data for audit projects.

[0007] Furthermore, the financial data of the audit project includes audit project field characteristics and amount characteristics.

[0008] Furthermore, the process of obtaining financial document image features by processing the financial document images collected through document scanning specifically employs an audit financial image selection method to process the financial document images to obtain the financial document image features. The financial document images are then processed to separate the financial document image features from the background features, and the inter-class variance C between the financial document image features and the background is defined.

[0009] Furthermore, the audit financial image selection method processes the data to obtain a feature vector for the project name field in the financial invoice image features. The financial invoice image features also include the financial invoice reimbursement time and the financial invoice reimbursement amount. The financial invoice reimbursement time is set to 1 if it occurs during the project's time period and 0 if it does not occur during the project's time period. The financial invoice reimbursement amount is obtained by image recognition using ORC technology.

[0010] Furthermore, the step of obtaining audit anomaly financial features by performing cross-modal data processing using the audit project financial data and the financial document image features is specifically achieved by concatenating feature vectors of the audit project financial data and the financial document image features.

[0011] It also provides an intelligent screening system for audit anomaly financial data, which includes a financial data collection module for audit projects, a financial document image scanning and acquisition module, an audit anomaly financial feature processing module, an intelligent screening model construction module for audit anomaly financial data, and an audit anomaly financial data filtering module. The audit project financial data collection module collects uploaded audit project financial data through the audit client. The financial document image scanning and acquisition module: obtains financial document images by processing project audit documents and scanning them. The audit abnormal financial feature processing module: receives the financial document image and processes it using the audit financial image selection method to obtain financial document image features; receives and processes the audit project financial data and the financial document image features to perform cross-modal data processing to obtain audit abnormal financial features; The intelligent screening model construction module for audit abnormal financial data: obtains the audit abnormal financial data screening method for the audit project corresponding to the financial data of the audit project obtained by the auditors through tagging the uploaded audit project and the financial document image; receives and constructs an intelligent screening model for audit abnormal financial data based on the audit abnormal financial characteristics and the audit abnormal financial data screening method. The audit abnormal financial data screening module: uses the intelligent screening model for audit abnormal financial data to screen audit abnormal financial data for different audit projects to obtain a real-time screening method for audit abnormal financial data, and uses the real-time screening method for audit abnormal financial data to intelligently screen audit abnormal financial data for audit projects.

[0012] Furthermore, the financial data of the audit project includes audit project field characteristics and amount characteristics.

[0013] Furthermore, the process of obtaining financial document image features by processing the financial document images collected through document scanning specifically employs an audit financial image selection method to process the financial document images to obtain the financial document image features. The financial document images are then processed to separate the financial document image features from the background features, and the inter-class variance C between the financial document image features and the background is defined.

[0014] Therefore, the beneficial effects of this invention are as follows: It collects and uploads audit project financial data, including project field values ​​(such as consulting fees, travel expenses, service fees, etc.) and amount values, and obtains financial document image features by processing scanned financial document images. It also obtains the time of all reimbursement documents (0 for other time periods if the project exists, 1 for non-existent periods), the project expenditure name, and all corresponding amounts. Using the audit project financial data, the financial document image features, and the corresponding audit anomaly financial data screening method, it constructs an intelligent audit anomaly financial data screening model. The model generates a new audit anomaly financial data screening method to determine financial data anomalies. This invention utilizes audit financial document image extraction technology to obtain accurate financial document image features. Furthermore, for the scenario of intelligent audit anomaly financial data screening, this invention uses a neural network model that processes cross-modal data to accurately generate audit anomaly financial data screening methods, thereby helping to reduce the manual screening time for auditors, reduce audit time costs, and improve the accuracy of anomaly financial data screening during audits. Attached Figure Description

[0015] Figure 1 This is a flowchart of the intelligent screening method for audit abnormal financial data of the present invention; Figure 2 This is a schematic diagram of the intelligent screening system for audit anomaly financial data of the present invention; Figure 3 This is a schematic diagram of the activation function model in this invention; Figure 4 This is a schematic diagram of the original image of the financial document in an embodiment of the present invention; Figure 5 This is a schematic diagram of an electronic device structure for implementing the method of the present invention in an embodiment of the present invention. Detailed Implementation

[0016] The invention will now be further described in conjunction with the accompanying drawings and specific embodiments. The neural network model used in this invention is a specific model suitable for the intelligent screening method of audit abnormal financial data, and the activation function is a preferred activation function for processing the selected audit project financial data and cross-modal data of financial invoice images.

[0017] like Figure 2 As shown, this invention belongs to the field of intelligent audit system research and development technology, and is for the research and development of information system integration service technology such as artificial intelligence system in the audit report production field. Therefore, it belongs to the information system integration service such as artificial intelligence system in the production field.

[0018] In a first aspect of the present invention, a method for intelligent screening of audit anomaly financial data is provided, the method comprising: S1. Collect and upload financial data of audit projects through the audit client, obtain financial document image features by processing the financial document images collected by document scanning, and obtain the corresponding audit abnormal financial data screening method by the auditors through tagging. S2. Cross-modal data processing is performed using the financial data of the audit project and the image features of the financial documents to obtain the audit abnormal financial features; S3. Receive and construct an intelligent screening model for audit abnormal financial data based on the aforementioned audit abnormal financial characteristics and the aforementioned audit abnormal financial data screening method; S4. Using the intelligent screening model for abnormal financial data in audits, a real-time screening method for abnormal financial data in audits is obtained by screening abnormal financial data for different audit projects. The real-time screening method for abnormal financial data in audits is then used to intelligently screen abnormal financial data for audit projects.

[0019] In this embodiment, given the revolutionary changes brought about by the rapid development of artificial intelligence and big data technologies to financial auditing, the financial anomaly detection technology, through algorithms such as machine learning, deep learning, and natural language processing, can intelligently analyze and recognize patterns in massive amounts of financial data, achieving automatic detection and risk warning of abnormal transactions. This intelligent auditing method not only significantly improves audit efficiency and accuracy but also achieves full data coverage, breaking through the limitations of traditional sampling audits and providing enterprises with more comprehensive and precise risk control measures for financial management. In this embodiment, the intelligent screening model for audit anomaly financial data is used as a dynamic training model for real-time auditing of financial data from different projects. The more training data, the more accurate the audit anomaly financial data screening method obtained by the model classification. The specific process of the intelligent screening method for audit anomaly financial data of this invention is as follows: Figure 1 As shown.

[0020] Furthermore, the financial data of the audit project includes audit project field characteristics and amount characteristics.

[0021] In this embodiment, each audit project has a different expenditure item name. The project name of the payable financial item is represented by a field value to facilitate subsequent model classification and processing to obtain the audit abnormal financial data filtering method. In addition, the amount feature is the amount feature value of the corresponding expenditure item.

[0022] In this embodiment, the preferred audit item fields are travel expenses (value 1), service fees (value 2), and consulting fees (value 3). These are partial audit item field values ​​set by those skilled in the art. Other fields include general audit item expenditures such as logistics costs, testing and processing fees, and material costs, all set according to the needs of model training. The monetary feature represents the reimbursable amount contained in the corresponding audit item field. For example, if travel expenses are 50,000, the monetary feature would be 50,000. The final audit item financial data is a feature vector composed of the audit item field features and the monetary feature.

[0023] Furthermore, the process of obtaining financial document image features by processing the scanned financial document images specifically employs an audit financial image selection method to process the financial document images and obtain the financial document image features. Image processing is then performed on the financial document images to separate the financial document image features from the background features. The inter-class variance C between the financial document image features and the background is defined as follows: In the formula, Let g be the number of pixels that belong to the image features of financial documents, and g be the total average gray level of the financial document image data. This represents the number of pixels that belong to the background features of the financial document image. The average gray level of the financial document image features. The average gray level represents the background features of the financial document image.

[0024] To implement the audit financial image selection method, this embodiment also prioritizes thickening the financial document image features in the financial document image to make it more suitable for accurately reading the financial document image features. At this time, the pixel value greater than the inter-class variance C can be used to determine the location of the financial document image.

[0025] In this embodiment, one of the feature vectors in the financial document image features is the pixel value of the project name field in the electronic invoice. The important feature vector of the expenditure project name field pixel value in the financial document image features is determined and selected using the audit financial image selection method.

[0026] Furthermore, the audit financial image selection method processes the data to obtain a feature vector for the project name field in the financial invoice image features. The financial invoice image features also include the financial invoice reimbursement time and the financial invoice reimbursement amount. The financial invoice reimbursement time is set to 1 if it occurs during the project's time period and 0 if it does not occur during the project's time period. The financial invoice reimbursement amount is obtained by image recognition using ORC technology.

[0027] OCR (Optical Character Recognition) is a technology that converts text in an image into editable text, widely used in scenarios such as digit recognition, document digitization, and invoice processing. Its core process includes image preprocessing, text detection, character segmentation, feature extraction, and character recognition. In this embodiment, the reimbursement amount for financial invoices is the total amount including tax.

[0028] In this embodiment, a feature of a financial invoice image can be expressed as a feature vector of (project name field feature vector, financial invoice reimbursement time, financial invoice reimbursement amount).

[0029] Furthermore, the step of obtaining audit anomaly financial features by performing cross-modal data processing using the audit project financial data and the financial document image features is specifically achieved by concatenating feature vectors of the audit project financial data and the financial document image features.

[0030] Feature concatenation is the most basic and commonly used feature fusion method in deep learning. Its core idea is to directly concatenate feature vectors from different sources or levels along a specified dimension to form a new, higher-dimensional feature vector. This method preserves the original information of all input features, providing richer feature representations for subsequent models. In this embodiment, the feature vector concatenation method can be horizontal or vertical, selected based on the model's processing capabilities. Based on the training and processing speed of the subsequent model, this invention preferably uses horizontal concatenation.

[0031] Furthermore, the activation function of the intelligent screening model for audit abnormal financial data is: In the formula, For the activation function value, Bias-weighted sum of the audited financial features of the input.

[0032] A neural network is a computational model composed of a large number of interconnected neurons. Inspired by the human nervous system, it achieves a highly adaptive and non-linear mapping from input data to output results through the combination and training of multiple layers of neurons. For example... Figure 3 The diagram shown is a schematic of the activation function model.

[0033] In this embodiment, the final output value of the neural network is used to determine the corresponding audit abnormal financial data screening method. If the final output value of the neural network is less than 0.12 and greater than 0.05, the audit abnormal financial data screening method is determined to be that the travel expense field of the audit project has financial data abnormalities. If the final output value of the neural network is greater than or equal to 0.12 and less than 0.25, the audit abnormal financial data screening method is determined to be that the test and analysis processing fee field of the audit project has financial data abnormalities. The correspondence between the final output value of the neural network and the audit abnormal financial data screening method is set by the auditors using the training data of the neural network model. The above two situations are the corresponding results obtained after the model is trained and used in this application. This application is not limited to these two results in determining the final audit abnormal financial data screening method.

[0034] Neural networks typically consist of multiple layers, including an input layer, hidden layers, and an output layer. The input layer receives input data, while the hidden and output layers are responsible for calculating the output. Each neuron receives multiple inputs from the previous layer, which are weighted and calculated, then a bias is applied, followed by a nonlinear transformation through an activation function to produce the final output. Connections between neurons are usually represented by weights, where the weight values ​​represent the strength of the connection. These weights can be updated during training to adjust the connection strength between neurons. Neural networks possess strong adaptability and nonlinear mapping capabilities, allowing them to adapt to diverse input data and complex problems. In practical applications, neural networks have been widely used in image recognition, speech recognition, natural language processing, and intelligent control, achieving considerable success.

[0035] Tanh Activation Function: The Tanh (hyperbolic tangent) activation function is a commonly used non-linear activation function in deep learning, with a shape similar to the Sigmoid activation function. This function maps input values ​​to the range of -1 to 1. This zero-centered characteristic makes Tanh more powerful than the Sigmoid function in terms of representation.

[0036] It also provides an intelligent screening system for audit anomaly financial data, which includes a financial data collection module for audit projects, a financial document image scanning and acquisition module, an audit anomaly financial feature processing module, an intelligent screening model construction module for audit anomaly financial data, and an audit anomaly financial data filtering module. The audit project financial data collection module collects uploaded audit project financial data through the audit client. The financial document image scanning and acquisition module: obtains financial document images by processing project audit documents and scanning them. The audit abnormal financial feature processing module: receives the financial document image and processes it using the audit financial image selection method to obtain financial document image features; receives and processes the audit project financial data and the financial document image features to perform cross-modal data processing to obtain audit abnormal financial features; The intelligent screening model construction module for audit abnormal financial data: obtains the audit abnormal financial data screening method for the audit project corresponding to the financial data of the audit project obtained by the auditors through tagging the uploaded audit project and the financial document image; receives and constructs an intelligent screening model for audit abnormal financial data based on the audit abnormal financial characteristics and the audit abnormal financial data screening method. The audit abnormal financial data screening module: uses the intelligent screening model for audit abnormal financial data to screen audit abnormal financial data for different audit projects to obtain a real-time screening method for audit abnormal financial data, and uses the real-time screening method for audit abnormal financial data to intelligently screen audit abnormal financial data for audit projects.

[0037] Furthermore, the financial data of the audit project includes audit project field characteristics and amount characteristics.

[0038] In this embodiment, each audit project has a different expenditure item name. The project name of the payable financial item is represented by a field value to facilitate subsequent model classification and processing to obtain the audit abnormal financial data filtering method. In addition, the amount feature is the amount feature value of the corresponding expenditure item.

[0039] In this embodiment, the preferred audit item fields are travel expenses (value 1), service fees (value 2), and consulting fees (value 3). These are some of the audit item field values ​​set by those skilled in the art. Other fields include general audit item expenditures such as testing and processing fees and material costs, all set according to the needs of model training. The monetary feature is the reimbursable amount contained in the corresponding audit item field. For example, if travel expenses are 50,000, the monetary feature would be 50,000. The final audit item financial data is a feature vector composed of the audit item field features and the monetary feature.

[0040] Furthermore, the process of obtaining financial document image features by processing the scanned financial document images specifically employs an audit financial image selection method to process the financial document images and obtain the financial document image features. Image processing is then performed on the financial document images to separate the financial document image features from the background features. The inter-class variance C between the financial document image features and the background is defined as follows: In the formula, Let g be the number of pixels that belong to the image features of financial documents, and g be the total average gray level of the financial document image data. This represents the number of pixels that belong to the background features of the financial document image. The average gray level of the financial document image features. The average gray level represents the background features of the financial document image.

[0041] To implement the image selection method for auditing financial documents, this embodiment prioritizes thickening the image features of financial invoices within the financial document images to better facilitate accurate reading of these features. In this case, pixel values ​​greater than the inter-class variance C can be used to identify locations within the financial invoice image. For example... Figure 4 This is a schematic diagram of the original image of the financial document in this invention.

[0042] In this embodiment, one of the feature vectors in the financial document image features is the pixel value of the project name field in the electronic invoice. The important feature vector of the expenditure project name field pixel value in the financial document image features is determined and selected using the audit financial image selection method.

[0043] Furthermore, the activation function of the intelligent screening model for audit abnormal financial data is: In the formula, For the activation function value, Bias-weighted sum of the audited financial features of the input.

[0044] A neural network is a computational model composed of a large number of interconnected neurons. Inspired by the human nervous system, it can achieve a highly adaptive and nonlinear mapping from input data to output results through the combination and training of multiple layers of neurons. Figure 5 This is a schematic diagram of the electronic device structure required for the implementation of the present invention. The electronic device includes a processor, a memory, a communication interface, and a bus.

[0045] Neural networks typically consist of multiple layers, including an input layer, hidden layers, and an output layer. The input layer receives input data, while the hidden and output layers are responsible for calculating the output. Each neuron receives multiple inputs from the previous layer, which are weighted and calculated, then a bias is applied, followed by a nonlinear transformation through an activation function to produce the final output. Connections between neurons are usually represented by weights, where the weight values ​​represent the strength of the connection. These weights can be updated during training to adjust the connection strength between neurons. Neural networks possess strong adaptability and nonlinear mapping capabilities, allowing them to adapt to diverse input data and complex problems. In practical applications, neural networks have been widely used in image recognition, speech recognition, natural language processing, and intelligent control, achieving considerable success.

[0046] Tanh Activation Function: The Tanh (hyperbolic tangent) activation function is a commonly used non-linear activation function in deep learning, with a shape similar to the Sigmoid activation function. This function maps input values ​​to the range of -1 to 1. This zero-centered characteristic makes Tanh more powerful than the Sigmoid function in terms of representation.

[0047] Therefore, the beneficial effects of this invention are as follows: It collects and uploads audit project financial data, including project field values ​​(such as consulting fees, travel expenses, service fees, etc.) and amount values, and obtains financial document image features by processing scanned financial document images. It also obtains the time of all reimbursement documents (0 for other time periods if the project exists, 1 for non-existent periods), the project expenditure name, and all corresponding amounts. Using the audit project financial data, the financial document image features, and the corresponding audit anomaly financial data screening method, it constructs an intelligent audit anomaly financial data screening model. The model generates a new audit anomaly financial data screening method to determine financial data anomalies. This invention utilizes audit financial document image extraction technology to obtain accurate financial document image features. Furthermore, for the scenario of intelligent audit anomaly financial data screening, this invention uses a neural network model that processes cross-modal data to accurately generate audit anomaly financial data screening methods, thereby helping to reduce the manual screening time for auditors, reduce audit time costs, and improve the accuracy of anomaly financial data screening during audits.

[0048] The combination of multiple embodiments of the present invention can achieve all the above effects, but it is not required that each embodiment of the present invention achieve all the above advantages and effects, because each embodiment of the present invention can constitute a separate technical solution and make one or more contributions to the prior art.

[0049] For any module structures not specifically defined in this invention, the existing technical specifications shall prevail. The existing technical specifications mentioned in the foregoing background and specific embodiments sections are considered part of this invention and are used to understand the meaning of certain technical features or parameters. The scope of protection of this invention is determined by the actual contents of the claims.

Claims

1. A method for intelligent screening of audit anomaly financial data, characterized in that, The method includes: S1. Collect and upload financial data of audit projects through the audit client, obtain financial document image features by processing the financial document images collected by document scanning, and obtain the corresponding audit abnormal financial data screening method by the auditors through tagging. S2. Cross-modal data processing is performed using the financial data of the audit project and the image features of the financial documents to obtain the audit abnormal financial features; S3. Receive and construct an intelligent screening model for audit abnormal financial data based on the aforementioned audit abnormal financial characteristics and the aforementioned audit abnormal financial data screening method; S4. Using the intelligent screening model for abnormal financial data in audits, a real-time screening method for abnormal financial data in audits is obtained by screening abnormal financial data for different audit projects. The real-time screening method for abnormal financial data in audits is then used to intelligently screen abnormal financial data for audit projects.

2. The intelligent screening method for audit anomaly financial data as described in claim 1, characterized in that: The financial data of the audit project includes audit project field characteristics and amount characteristics.

3. The intelligent screening method for audit anomaly financial data as described in claim 2, characterized in that: The process involves obtaining financial document image features by processing scanned financial document images. Specifically, the audit financial image selection method is used to process the financial document images to obtain the financial document image features. Image processing is performed on the financial document images to separate the financial document image features from the background features. The inter-class variance C between the financial document image features and the background is defined.

4. The intelligent screening method for audit anomaly financial data as described in claim 3, characterized in that: The audit financial image selection method processes the data to obtain a feature vector for the project name field in the financial invoice image features. The financial invoice image features also include the financial invoice reimbursement time and the financial invoice reimbursement amount. The financial invoice reimbursement time is set to 1 if it occurs during the project's time period and 0 if it does not occur during the project's time period. The financial invoice reimbursement amount is obtained by image recognition using ORC technology.

5. The intelligent screening method for audit anomaly financial data as described in claim 1 or 3, characterized in that: The method of obtaining audit anomaly financial features by performing cross-modal data processing using the audit project's financial data and the financial document image features is specifically achieved by concatenating feature vectors of the audit project's financial data and the financial document image features.

6. An intelligent screening system for audit anomaly financial data, comprising an audit project financial data collection module, a financial document image scanning and acquisition module, an audit anomaly financial feature processing module, an audit anomaly financial data intelligent screening model construction module, and an audit anomaly financial data filtering module, characterized in that: The audit project financial data collection module collects uploaded audit project financial data through the audit client. The financial document image scanning and acquisition module: obtains financial document images by processing project audit documents and scanning them. The audit abnormal financial feature processing module: receives the financial document image and processes it using the audit financial image selection method to obtain financial document image features; receives and processes the audit project financial data and the financial document image features to perform cross-modal data processing to obtain audit abnormal financial features; The intelligent screening model construction module for audit abnormal financial data: obtains the audit abnormal financial data screening method for the audit project corresponding to the financial data of the audit project obtained by the auditors through tagging the uploaded audit project and the financial document image; receives and constructs an intelligent screening model for audit abnormal financial data based on the audit abnormal financial characteristics and the audit abnormal financial data screening method. The audit abnormal financial data screening module: uses the intelligent screening model for audit abnormal financial data to screen audit abnormal financial data for different audit projects to obtain a real-time screening method for audit abnormal financial data, and uses the real-time screening method for audit abnormal financial data to intelligently screen audit abnormal financial data for audit projects.

7. The intelligent screening system for audit anomaly financial data as described in claim 6, characterized in that: The financial data of the audit project includes audit project field characteristics and amount characteristics.

8. The intelligent screening system for audit anomaly financial data as described in claim 7, characterized in that: The process involves obtaining financial document image features by processing scanned financial document images. Specifically, the audit financial image selection method is used to process the financial document images to obtain the financial document image features. Image processing is performed on the financial document images to separate the financial document image features from the background features. The inter-class variance C between the financial document image features and the background is defined.

9. The intelligent screening system for audit anomaly financial data as described in claim 6 or 7, characterized in that: The intelligent screening model for audit anomaly financial data has an activation function that includes a biased weighted sum of the input audit anomaly financial features.

10. The intelligent screening system for audit anomaly financial data as described in claim 8, characterized in that: The audit financial image selection method processes the data to obtain a feature vector for the project name field in the financial invoice image features. The financial invoice image features also include the financial invoice reimbursement time and the financial invoice reimbursement amount. The financial invoice reimbursement time is set to 1 if it occurs during the project's time period and 0 if it does not occur during the project's time period. The financial invoice reimbursement amount is obtained by image recognition using ORC technology.

Images