Automatic reimbursement auditing method and system based on large language model

Abstract

The invention discloses an automatic reimbursement auditing method and system based on a large language model, and the method comprises the steps: enabling a system to receive a reimbursement bill picture uploaded by a user, and carrying out the preprocessing of the to-be-audited reimbursement bill picture; the method automatically confirms the identity of the reimbursement person and the department where the reimbursement person is located, analyzes the name information of the purchased article, judges whether the purchased article is the necessary expense of the company or not, if yes, reimbursement is directly carried out, and if not, for example, in real life, each department of the company needs to have a meal together frequently, it is judged that the expenditure is the department team expenditure. The method comprises the following steps: a company can compete and give a certain reimbursement amount to each department team every month, summing the reimbursement amount of this time and the reimbursement amount of the department team expenditure of this month, judging whether the sum reaches a reimbursement upper limit or not, if so, not reimbursement, and transferring the reimbursement sheet of this time to an automatic reimbursement table of the team; therefore, the basic operation cost of an enterprise can be well managed, and the requirements of workers are met.

Description

Technical Field

[0001] The present invention relates to the technical field of reimbursement auditing, and in particular to an automated reimbursement auditing method and system based on a large language model. Background Art

[0002] Accounting vouchers can record the occurrence and completion of economic activities and provide original basis for accounting. They can check the authenticity, legality and rationality of economic activities and provide important basis for accounting supervision. They can clarify economic responsibilities and provide important documents for implementing the post responsibility system. They can reflect relevant economic interest relations and provide legal evidence for safeguarding legitimate rights and interests. They can supervise economic activities and control economic operations.

[0003] When existing enterprises make reimbursements, users usually log in to the enterprise system and upload a picture of the reimbursement form on their own. After multiple levels of approval, it is determined whether to make reimbursement. It is impossible to automatically determine the identity of the person making the reimbursement and the department they are in based on the reimbursement form picture. In real life, the company needs to reimburse a certain amount of expenses for each department and each team in the company within its capabilities. The expenses can be used for team dinners, etc. It is impossible to automatically review whether the reimbursement form has reached the reimbursement limit. The reimbursement review and statistics are slow, which makes it difficult to meet the management needs of the enterprise. Summary of the invention

[0004] In order to solve the above technical problems, an automated reimbursement review method and system based on a large language model are provided. This technical solution solves the problem mentioned in the above background technology that the existing reimbursement system cannot automatically determine the identity and department of the reimbursement recipient based on the reimbursement form picture, while in real life, the company needs to reimburse a certain amount of expenses to each department and each team of the company within its capabilities. The expenses can be used for team dinners, etc., and it is impossible to automatically review whether the reimbursement form has reached the reimbursement limit, and the reimbursement review statistics are slow.

[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0006] The automated reimbursement review method based on the large language model includes:

[0007] The system accepts the reimbursement form picture uploaded by the user, and pre-processes the reimbursement form picture to be reviewed to obtain a pre-processed picture, wherein the pre-processing includes image denoising, image enhancement and image segmentation;

[0008] Analyze and extract the text in the pre-processed reimbursement form image, wherein the extracted text includes the name information of the purchased item, the amount information of the goods, and the name information of the purchaser;

[0009] Compare the extracted purchaser name information with the user name information pre-stored in the system to determine whether there are multiple users in the system with the same name as the purchaser;

[0010] Use information to determine the identities of multiple users with the same name, and confirm the identity of the person making the reimbursement and the department to which the person belongs;

[0011] Determine the expense attribute of the name information of the purchased item, where the expense attribute is necessary expenses of the company and expenses of the department team. If it is necessary expenses of the company, reimbursement is made directly;

[0012] Conduct statistical analysis on the department team expenditures of the department where the reimbursement recipient is located in that month to determine the reimbursement results.

[0013] Preferably, the image denoising specifically comprises the following steps:

[0014] Adaptive wavelet threshold denoising is performed on the expense reimbursement form image;

[0015] By constructing a probability model of wavelet coefficients, the optimal wavelet threshold is determined using Bayesian estimation.

[0016] The wavelet coefficients are processed with adaptive threshold and then reconstructed by wavelet to obtain a denoised image.

[0017] Preferably, the image enhancement specifically comprises the following steps:

[0018] Performing adaptive histogram equalization processing on the denoised image;

[0019] Determine the cumulative distribution function of the local area by calculating the grayscale histogram of the denoised image;

[0020] Based on the cumulative distribution function, the grayscale mapping relationship of the local area is adaptively adjusted to obtain an enhanced image.

[0021] Preferably, the image segmentation specifically comprises the following steps:

[0022] Segmenting the enhanced image into multiple superpixels using a superpixel segmentation algorithm;

[0023] Determine the similarity matrix between superpixels as the edge weight of the undirected graph, and construct a superpixel graph cut undirected graph;

[0024] In combination with the pre-set probability distribution of the text area, the global optimal solution of the superpixel graph cut undirected graph is solved by the maximum flow minimum cut algorithm;

[0025] The segmentation result of the text area is obtained to obtain the preprocessed image.

[0026] Preferably, the analysis of the Chinese characters in the pre-processed expense claim form image specifically comprises the following steps:

[0027] Mark the area of ​​the expense report image in lines;

[0028] The labeled images are combined into a training set and trained through the object detection model until convergence;

[0029] The segmented image blocks are put into the trained object detection model for object detection of text line areas, and the coordinates of the text line areas in each image block are obtained and output.

[0030] Preferably, the extraction of Chinese characters from the pre-processed expense report image specifically comprises the following steps:

[0031] Based on the coordinates of the complete text line area, obtain each complete text line area in the expense report image;

[0032] Based on the text line content in the expense report image, the regional coordinates of the expense report image are marked and formed into a training set;

[0033] Train a hybrid model consisting of CNN and LSTM based on deep learning attention mechanism until convergence;

[0034] The text line area of ​​the segmented image block is put into the trained hybrid model for text line recognition to obtain the specific text in the text line.

[0035] Preferably, the step of comparing the extracted purchaser name information with the user name information pre-stored in the system specifically comprises the following steps:

[0036] Retrieve the user name information stored in the system;

[0037] The extracted purchaser name information is traversed through the entire pre-stored user name information;

[0038] If the number of pre-stored user names corresponding to the purchaser's name information is one, the identity of the reimbursement recipient is directly determined, and the department to which the reimbursement recipient belongs is directly obtained;

[0039] If there are multiple pre-stored user names corresponding to the purchaser's name information, information judgment is required.

[0040] Preferably, the steps of determining the identities of multiple users with the same name and confirming the identity of the reimbursement recipient and the department to which the reimbursement recipient belongs specifically include the following steps:

[0041] Retrieve the mobile phone number information of multiple users with the same name from the database;

[0042] Automatically send verification information to the mobile phones of multiple users with the same name, wherein the verification information includes the name of the purchased item to be reimbursed and the amount of the item;

[0043] The user confirms the verification information and, based on the confirmation result, determines the identity of the person making the reimbursement and the department to which the person belongs.

[0044] Preferably, the statistical analysis of the department team expenditures of the department where the reimbursement recipient is located in that month to determine the reimbursement results specifically includes the following steps:

[0045] Collect statistics on the expenditures of the department team of the department where the reimbursement requester is located in that month and obtain the statistical value;

[0046] The statistical value is summed with the department team expenditure to be reimbursed to obtain the sum value;

[0047] The summed value is compared with a threshold value preset in the system;

[0048] If the sum is greater than or equal to the preset threshold, the department team's expenses will not be reimbursed and will be transferred to the team's reimbursement list for the next month;

[0049] If the sum is less than the preset threshold, reimbursement will be made directly.

[0050] In addition, the present invention also proposes an automated reimbursement review system based on a large language model, comprising:

[0051] An image preprocessing module, wherein the image preprocessing module internally integrates a first processing unit, a second processing unit and a third processing unit, wherein the first processing unit is used to denoise the image, the second processing unit is used to enhance the image, and the third processing unit is used to segment the image;

[0052] An extraction module, the extraction module is used to analyze and extract the characters in the pre-processed expense report image;

[0053] A first judgment module, which is used to compare the extracted purchaser name information with the user name information pre-stored in the system to determine whether there are multiple users in the system with the same name as the purchaser;

[0054] An identity confirmation module, which is used to determine the identities of multiple users with the same name, and confirm the identity of the reimbursement recipient and the department to which the reimbursement recipient belongs;

[0055] A second judgment module, the second judgment module is used to judge the cost attribute of the name information of the purchased item;

[0056] The reimbursement module is used to directly reimburse the necessary expenses of the company and to conduct statistical analysis on the department team expenses of the department where the reimburser is located in that month to determine the reimbursement results.

[0057] Compared with the prior art, the present invention provides an automated reimbursement review method and system based on a large language model, which has the following beneficial effects:

[0058] The present invention improves the accuracy of subsequent text recognition by removing noise from the reimbursement form image. In order to enhance the contrast of the image, improve the image clarity and improve the effect of text recognition, the denoised image is subjected to image enhancement processing, and the image is divided into different areas to better identify and locate the text area. Secondly, text line area detection and image text conversion to computer text are performed through deep learning technology to extract the name information of the purchased item, the amount information of the goods and the name information of the purchaser. The extracted name information of the purchaser is first compared with the user name information pre-stored in the system. If the number of pre-stored user names corresponding to the name information of the purchaser is one, the identity of the person making the reimbursement and the department to which the person making the reimbursement belongs are directly confirmed. Of course, in real life, there are also companies with multiple people with the same name and surname. The amount information of the purchased items is sent to the user with the same name, so that the identity of the person making the reimbursement and the department to which the person making the reimbursement belongs can be accurately locked. The name information of the purchased items is then analyzed to determine whether it is the company's necessary expenses. The company's necessary expenses include many types, such as the purchase of chairs, computers, printing, and other necessary expenses by the department, which are directly reimbursed. If not, for example, in real life, each department of the company often needs to have dinner together, which is determined to be department team expenditure. The company will do its best to give each department team a certain reimbursement amount each month, and sum the amount reimbursed with the amount of department team expenditure reimbursed for that month to determine whether it has reached the reimbursement limit. If so, it will not be reimbursed, and the reimbursement form will be transferred to the team's automatic reimbursement form, thereby managing the company's basic operating costs well and meeting the needs of staff. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a schematic diagram of the automated reimbursement review method of the present invention;

[0060] Figure 2 Schematic diagram of the method for image denoising in the present invention;

[0061] Figure 3 Schematic diagram of the image enhancement method in the present invention;

[0062] Figure 4 It is a schematic diagram of the method for image segmentation in the present invention;

[0063] Figure 5 It is a schematic diagram of analyzing the Chinese characters in the pre-processed expense claim form image in the present invention;

[0064] Figure 6 It is a schematic diagram of extracting characters from a pre-processed reimbursement form image in the present invention. DETAILED DESCRIPTION

[0065] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art may think of other obvious variations.

[0066] Please refer to Figure 1-Figure 6 As shown, the automated reimbursement review method based on the large language model includes:

[0067] The system accepts the reimbursement form images uploaded by the user, and preprocesses the reimbursement form images to be reviewed to obtain preprocessed images. The preprocessing includes image denoising, image enhancement and image segmentation.

[0068] Analyze and extract the text in the pre-processed expense report image, the extracted text includes the name of the purchased item, the amount of the goods, and the name of the purchaser;

[0069] Compare the extracted purchaser name information with the user name information pre-stored in the system to determine whether there are multiple users in the system with the same name as the purchaser;

[0070] Use information to determine the identities of multiple users with the same name, and confirm the identity of the person making the reimbursement and the department to which the person belongs;

[0071] Determine the expense attribute of the name information of the purchased item. The expense attribute is the necessary expense of the company and the expense of the department team. If it is the necessary expense of the company, it will be reimbursed directly.

[0072] Conduct statistical analysis on the department team expenditures of the department where the reimbursement recipient is located in that month to determine the reimbursement results.

[0073] It is understandable to those skilled in the art that the present invention improves the accuracy of subsequent text recognition by removing noise from the reimbursement form image. In order to enhance the contrast of the image, improve the image clarity, and improve the effect of text recognition, the denoised image is subjected to image enhancement processing, and the image is divided into different regions to better identify and locate the text region. Secondly, text line region detection and image text conversion to computer text are performed through deep learning technology to extract the name information of the purchased item, the amount of the goods, and the name information of the purchaser. The extracted purchaser's name information is first compared with the user name information pre-stored in the system. If the number of pre-stored user names corresponding to the purchaser's name information is one, the identity of the person making the reimbursement and the department to which the person making the reimbursement belongs are directly confirmed. Of course, in real life, there are also companies with multiple people with the same name and surname, and the purchases being reimbursed this time are compared with each other. The name information and amount information of the goods are sent to users with the same name, so that the identity of the person making the reimbursement and the department to which the person making the reimbursement belongs can be accurately locked. The name information of the purchased items is then analyzed to determine whether it is a necessary expense for the company. Necessary expenses for the company include many types, such as the purchase of chairs, computers, printing, and other necessary expenses for the department, which are directly reimbursed. If not, for example, in real life, each department of the company often needs to have dinner together, which is judged as department team expenditure. The company will do its best to give each department team a certain reimbursement amount each month, and sum the amount reimbursed with the amount of department team expenditure reimbursed for that month to determine whether it has reached the reimbursement limit. If so, it will not be reimbursed, and the reimbursement form will be transferred to the team's automatic reimbursement form, thereby managing the company's basic operating costs well and meeting the needs of staff.

[0074] Image denoising specifically includes the following steps:

[0075] Adaptive wavelet threshold denoising is performed on the expense reimbursement form image;

[0076] By constructing a probability model of wavelet coefficients, the optimal wavelet threshold is determined using Bayesian estimation.

[0077] The wavelet coefficients are processed with adaptive threshold and then reconstructed by wavelet to obtain a denoised image.

[0078] Image enhancement specifically includes the following steps:

[0079] Perform adaptive histogram equalization on the denoised image;

[0080] By calculating the grayscale histogram of the denoised image, the cumulative distribution function of the local area is determined;

[0081] Based on the cumulative distribution function, the grayscale mapping relationship of the local area is adaptively adjusted to obtain an enhanced image.

[0082] Image segmentation specifically includes the following steps:

[0083] The enhanced image is segmented into multiple superpixels by a superpixel segmentation algorithm;

[0084] Determine the similarity matrix between superpixels as the edge weight of the undirected graph, and construct a superpixel graph cut undirected graph;

[0085] Combined with the pre-set probability distribution of the text area, the global optimal solution of the superpixel graph cut undirected graph is solved by the maximum flow minimum cut algorithm;

[0086] The segmentation result of the text area is obtained and the preprocessed image is obtained.

[0087] The analysis of the Chinese characters in the pre-processed expense report image specifically includes the following steps:

[0088] Mark the area of ​​the expense report image in lines;

[0089] The labeled images are combined into a training set and trained through the object detection model until convergence;

[0090] The segmented image blocks are put into the trained object detection model for object detection of text line areas, and the coordinates of the text line areas in each image block are obtained and output.

[0091] Extracting Chinese characters from the pre-processed expense report image specifically includes the following steps:

[0092] Based on the coordinates of the complete text line area, obtain each complete text line area in the expense report image;

[0093] Based on the text line content in the expense report image, the regional coordinates of the expense report image are marked and formed into a training set;

[0094] Train a hybrid model consisting of CNN and LSTM based on deep learning attention mechanism until convergence;

[0095] The text line area of ​​the segmented image block is put into the trained hybrid model for text line recognition to obtain the specific text in the text line.

[0096] Comparing the extracted purchaser name information with the user name information pre-stored in the system specifically includes the following steps:

[0097] Retrieve the user name information stored in the system;

[0098] The extracted purchaser name information is traversed through the entire pre-stored user name information;

[0099] If the number of pre-stored user names corresponding to the purchaser's name information is one, the identity of the reimbursement recipient is directly determined, and the department to which the reimbursement recipient belongs is directly obtained;

[0100] If there are multiple pre-stored user names corresponding to the purchaser's name information, information judgment is required.

[0101] The steps for determining the identities of multiple users with the same name and confirming the identity of the reimbursement recipient and the department to which the reimbursement recipient belongs are as follows:

[0102] Retrieve the mobile phone number information of multiple users with the same name from the database;

[0103] Automatically send verification information to the mobile phones of multiple users with the same name. The verification information includes the name of the purchased item to be reimbursed and the amount of the goods;

[0104] The user confirms the verification information and, based on the confirmation result, determines the identity of the person making the reimbursement and the department to which the person belongs.

[0105] Conduct statistical analysis on the department team expenditures of the department where the reimbursement is made in that month, and determine the reimbursement results, including the following steps:

[0106] Collect statistics on the expenditures of the department team of the department where the reimbursement requester is located in that month and obtain the statistical value;

[0107] The statistical value is summed with the department team expenditure to be reimbursed to obtain the sum value;

[0108] The summed value is compared with a threshold value preset in the system;

[0109] If the sum is greater than or equal to the preset threshold, the department team's expenses will not be reimbursed and will be transferred to the team's reimbursement list for the next month;

[0110] If the sum is less than the preset threshold, reimbursement will be made directly.

[0111] In addition, the present invention also proposes an automated reimbursement review system based on a large language model, comprising:

[0112] An image preprocessing module, wherein the image preprocessing module internally integrates a first processing unit, a second processing unit and a third processing unit, wherein the first processing unit is used to denoise the image, the second processing unit is used to enhance the image, and the third processing unit is used to segment the image;

[0113] An extraction module, which is used to analyze and extract the text in the pre-processed expense report image;

[0114] A first judgment module, the first judgment module is used to compare the extracted purchaser name information with the user name information pre-stored in the system to determine whether there are multiple users in the system with the same name as the purchaser;

[0115] The identity confirmation module is used to determine the identities of multiple users with the same name, confirm the identity of the reimbursement recipient and the department to which the reimbursement recipient belongs;

[0116] A second judgment module, the second judgment module is used to judge the cost attribute of the name information of the purchased item;

[0117] The reimbursement module is used to directly reimburse the necessary expenses of the company and to conduct statistical analysis on the department team expenses of the department where the reimburser is located in that month to determine the reimbursement results.

[0118] In summary, the present invention improves the accuracy of subsequent text recognition by removing noise from the reimbursement form image. In order to enhance the contrast of the image, improve the image clarity, and improve the effect of text recognition, the denoised image is subjected to image enhancement processing, and the image is divided into different areas to better identify and locate the text area. Secondly, text line area detection and image text conversion to computer text are performed through deep learning technology to extract the name information of the purchased item, the amount information of the goods, and the name information of the purchaser. The extracted purchaser's name information is first compared with the user name information pre-stored in the system. If the number of pre-stored user names corresponding to the purchaser's name information is one, the identity of the person making the reimbursement and the department to which the person making the reimbursement belongs are directly confirmed. Of course, in real life, there are also companies with multiple people with the same name and surname, and the name information of the purchased item being reimbursed is compared with the name information of the purchased item. The information of the amount of the goods is sent to the user with the same name, so as to accurately lock the identity of the person making the reimbursement and the department to which the person making the reimbursement belongs, and then analyze the name information of the purchased items to determine whether it is the necessary expenses of the company. The necessary expenses of the company include many kinds, such as the purchase of chairs, computers, printing and other necessary expenses by the department, which are directly reimbursed. If not, for example, in real life, each department of the company often needs to have dinner together, which is judged as department team expenditure. The company will do its best to give each department team a certain reimbursement amount every month, and sum the amount of this reimbursement with the amount of reimbursement of the department team expenditure for that month to determine whether it has reached the reimbursement limit. If reached, it will not be reimbursed, and the reimbursement form will be transferred to the team's automatic reimbursement form, so as to manage the basic operating costs of the enterprise well and meet the needs of the staff.

[0119] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. An automated reimbursement review method based on a large language model, characterized in that: include: The system accepts the reimbursement form picture uploaded by the user, and pre-processes the reimbursement form picture to be reviewed to obtain a pre-processed picture, wherein the pre-processing includes image denoising, image enhancement and image segmentation; Analyze and extract the text in the pre-processed reimbursement form image, wherein the extracted text includes the name information of the purchased item, the amount information of the goods, and the name information of the purchaser; Compare the extracted purchaser name information with the user name information pre-stored in the system to determine whether there are multiple users in the system with the same name as the purchaser; Use information to determine the identities of multiple users with the same name, and confirm the identity of the person making the reimbursement and the department to which the person belongs; Determine the expense attribute of the name information of the purchased item, where the expense attribute is necessary expenses of the company and expenses of the department team. If it is necessary expenses of the company, reimbursement is made directly; Conduct statistical analysis on the department team expenditures of the department where the reimbursement recipient is located in that month to determine the reimbursement results.

2. The automated reimbursement review method based on a large language model according to claim 1, characterized in that: The image denoising specifically comprises the following steps: Adaptive wavelet threshold denoising is performed on the expense reimbursement form image; By constructing a probability model of wavelet coefficients, the optimal wavelet threshold is determined using Bayesian estimation. The wavelet coefficients are processed with adaptive threshold and then reconstructed by wavelet to obtain a denoised image.

3. The automated reimbursement review method based on a large language model according to claim 1, characterized in that: The image enhancement specifically comprises the following steps: Performing adaptive histogram equalization processing on the denoised image; Determine the cumulative distribution function of the local area by calculating the grayscale histogram of the denoised image; Based on the cumulative distribution function, the grayscale mapping relationship of the local area is adaptively adjusted to obtain an enhanced image.

4. The automated reimbursement review method based on a large language model according to claim 1, characterized in that: The image segmentation specifically comprises the following steps: Segmenting the enhanced image into multiple superpixels using a superpixel segmentation algorithm; Determine the similarity matrix between superpixels as the edge weight of the undirected graph, and construct a superpixel graph cut undirected graph; In combination with the pre-set probability distribution of the text area, the global optimal solution of the superpixel graph cut undirected graph is solved by the maximum flow minimum cut algorithm; The segmentation result of the text area is obtained to obtain the preprocessed image.

5. The automated reimbursement review method based on a large language model according to claim 1, characterized in that: The analysis of the Chinese characters in the pre-processed expense report image specifically includes the following steps: Mark the area of ​​the expense report image in lines; The labeled images are combined into a training set and trained through the object detection model until convergence; The segmented image blocks are put into the trained object detection model for object detection of text line areas, and the coordinates of the text line areas in each image block are obtained and output.

6. The automated reimbursement review method based on a large language model according to claim 1, characterized in that: The extraction of Chinese characters from the pre-processed expense report image specifically comprises the following steps: Based on the coordinates of the complete text line area, obtain each complete text line area in the expense report image; Based on the text line content in the expense report image, the regional coordinates of the expense report image are marked and formed into a training set; Train a hybrid model consisting of CNN and LSTM based on deep learning attention mechanism until convergence; The text line area of ​​the segmented image block is put into the trained hybrid model for text line recognition to obtain the specific text in the text line.

7. The automated reimbursement review method based on a large language model according to claim 1, characterized in that: The step of comparing the extracted purchaser name information with the user name information pre-stored in the system specifically includes the following steps: Retrieve the user name information stored in the system; The extracted purchaser name information is traversed through the entire pre-stored user name information; If the number of pre-stored user names corresponding to the purchaser's name information is one, the identity of the reimbursement recipient is directly determined, and the department to which the reimbursement recipient belongs is directly obtained; If there are multiple pre-stored user names corresponding to the purchaser's name information, information judgment is required.

8. The automated reimbursement review method based on a large language model according to claim 1, characterized in that: The steps of determining the identities of multiple users with the same name and confirming the identity of the reimbursement recipient and the department to which the reimbursement recipient belongs specifically include the following steps: Retrieve the mobile phone number information of multiple users with the same name from the database; Automatically send verification information to the mobile phones of multiple users with the same name, wherein the verification information includes the name of the purchased item to be reimbursed and the amount of the item; The user confirms the verification information and, based on the confirmation result, determines the identity of the person making the reimbursement and the department to which the person belongs.

9. The automated reimbursement review method based on a large language model according to claim 1, characterized in that: The statistical analysis of the department team expenditures of the department where the reimbursement is made in that month and the determination of the reimbursement results specifically include the following steps: Collect statistics on the expenditures of the department team of the department where the reimbursement requester is located in that month and obtain the statistical value; The statistical value is summed with the department team expenditure to be reimbursed to obtain the sum value; The summed value is compared with a threshold value preset in the system; If the sum is greater than or equal to the preset threshold, the department team's expenses will not be reimbursed and will be transferred to the team's reimbursement list for the next month; If the sum is less than the preset threshold, reimbursement will be made directly.

10. An automated reimbursement review system based on a large language model, used to implement the automated reimbursement review method based on a large language model as described in any one of claims 1 to 9, characterized in that: include: An image preprocessing module, wherein the image preprocessing module internally integrates a first processing unit, a second processing unit and a third processing unit, wherein the first processing unit is used to denoise the image, the second processing unit is used to enhance the image, and the third processing unit is used to segment the image; An extraction module, the extraction module is used to analyze and extract the characters in the pre-processed expense report image; A first judgment module, which is used to compare the extracted purchaser name information with the user name information pre-stored in the system to determine whether there are multiple users in the system with the same name as the purchaser; An identity confirmation module, which is used to determine the identities of multiple users with the same name, and confirm the identity of the reimbursement recipient and the department to which the reimbursement recipient belongs; A second judgment module, the second judgment module is used to judge the cost attribute of the name information of the purchased item; The reimbursement module is used to directly reimburse the necessary expenses of the company and to conduct statistical analysis on the department team expenses of the department where the reimburser is located in that month to determine the reimbursement results.

Images