CN113255498A - Financial reimbursement invoice management method based on block chain technology - Google Patents

Financial reimbursement invoice management method based on block chain technology Download PDF

Info

Publication number
CN113255498A
CN113255498A CN202110535995.1A CN202110535995A CN113255498A CN 113255498 A CN113255498 A CN 113255498A CN 202110535995 A CN202110535995 A CN 202110535995A CN 113255498 A CN113255498 A CN 113255498A
Authority
CN
China
Prior art keywords
feature
invoice
reimbursement
semantic
classification
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Withdrawn
Application number
CN202110535995.1A
Other languages
Chinese (zh)
Inventor
吴雨淅
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hangzhou Xiyu Technology Co ltd
Original Assignee
Hangzhou Xiyu Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Hangzhou Xiyu Technology Co ltd filed Critical Hangzhou Xiyu Technology Co ltd
Priority to CN202110535995.1A priority Critical patent/CN113255498A/en
Publication of CN113255498A publication Critical patent/CN113255498A/en
Withdrawn legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/40—Document-oriented image-based pattern recognition
    • G06V30/41—Analysis of document content
    • G06V30/412—Layout analysis of documents structured with printed lines or input boxes, e.g. business forms or tables
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/12—Accounting
    • G06Q40/125—Finance or payroll
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L9/00—Cryptographic mechanisms or cryptographic arrangements for secret or secure communications; Network security protocols
    • H04L9/50—Cryptographic mechanisms or cryptographic arrangements for secret or secure communications; Network security protocols using hash chains, e.g. blockchains or hash trees

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Biophysics (AREA)
  • Computing Systems (AREA)
  • Business, Economics & Management (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Multimedia (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • Technology Law (AREA)
  • General Business, Economics & Management (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The method adopts a deep neural network model to perform feature extraction and recognition on an invoice image aiming at the storage of distributed data of a block chain and the non-tamper property and traceability of the data stored by the block chain so as to determine whether errors exist in submitted invoices. Like this, through adopting the block chain to store and manage the invoice image, can utilize the tamper-proof nature and the traceability of block chain, detect the error condition of invoice to avoid invoice itself to have the management confusion of financial reimbursement invoice that the mistake leads to.

Description

Financial reimbursement invoice management method based on block chain technology
Technical Field
The present invention relates to financial data management in the field of blockchain, and more particularly, to a method for managing financial reimbursement invoices based on blockchain technology, a system for managing financial reimbursement invoices based on blockchain technology, and an electronic device.
Background
The block chain is a distributed shared account book and a database, and has the characteristics of decentralization, no tampering, trace retaining in the whole process, traceability, collective maintenance, openness and transparency and the like. The characteristics ensure the honesty and the transparency of the block chain and lay a foundation for creating trust for the block chain. In recent years, as the technology of blockchain matures and develops, various data management technologies based on blockchain technology and applications thereof are developed due to the unique non-alterable characteristic of blockchain.
Expense reimbursement belongs to the internal financial management items of a company, and the expense reimbursement is required to be carried out to an employee bank card by the company according to reimbursement documents personally submitted by the employees. Here, the reimbursement document personally submitted by the employee may contain a large number of invoices. On the one hand, due to the trend of centralized management of finance, the corporate financial department desires to uniformly manage and review the invoice for reimbursement of expenses of departments, subsidiaries and the like, which involves the distributed storage function of data of the block chain technology, and furthermore, because of the importance of financial data, it is also desired to ensure the non-tampering property and traceability of data using the block chain technology.
However, even for a certain reimbursement item, there may be a case where a plurality of invoices are submitted by employees, and if these invoices themselves have errors, such as duplicate reimbursement, date error, or number-added tax invoice, due to non-tamper-resistance of data stored in the block, there is a problem in managing the financial reimbursement invoice.
Therefore, an optimized solution for financial reimbursement invoice management is desired.
Disclosure of Invention
The present application is proposed to solve the above-mentioned technical problems. The embodiment of the application provides a block chain technology-based financial reimbursement invoice management method, a block chain technology-based financial reimbursement invoice management system and an electronic device, which adopt a deep neural network model to perform feature extraction and identification on an invoice image aiming at the storage of distributed data of a block chain and the non-tamper-property and traceability of the data stored by the block chain so as to determine whether a submitted invoice has errors. Like this, through adopting the block chain to store and manage the invoice image, can utilize the tamper-proof nature and the traceability of block chain, detect the error condition of invoice to avoid invoice itself to have the management confusion of financial reimbursement invoice that the mistake leads to.
According to one aspect of the present application, there is provided a financial reimbursement invoice management method based on blockchain technology, comprising:
acquiring invoice images of a plurality of invoices submitted by reimbursement applicants;
inputting invoice images of the plurality of invoices into a convolutional neural network to obtain an initial feature map, wherein the initial feature map has four dimensions: an image width W, an image height H, a channel dimension C and a sample dimension S, wherein the channel dimension C corresponds to the number of convolution kernels of each layer in the convolutional neural network;
pooling a feature matrix consisting of image width W and image height H in the initial feature map to obtain a feature matrix with the dimension C S;
acquiring reimbursement item descriptions associated with the plurality of invoices and extracting text information in the reimbursement item descriptions;
enabling the text information in the reimbursement item description to pass through a semantic understanding model based on deep learning to obtain a semantic feature vector;
calculating Softmax-like function values of feature values of all positions of the semantic feature vector, and weighting all feature values of the same sample in the feature matrix with the scale of C S to obtain an attention feature matrix, wherein the Softmax-like function values are quotient of weighted sum of natural constant exponent function values taking the negative number of the feature values of all positions in the semantic feature vector as power and natural constant exponent function values taking the negative number of the feature values of all positions in the semantic feature vector as power;
passing the attention feature matrix through a plurality of convolutional layers to obtain a classification feature map;
passing the classification feature map through a classifier to obtain a classification result, wherein the classification result is used for indicating whether errors exist in the submitted invoice or not; and
in response to the classification result being that there is no error in the submitted invoices, storing invoice images of the plurality of invoices in a storage block of a block chain architecture.
According to another aspect of the present application, there is provided a financial reimbursement invoice management system based on blockchain technology, comprising:
the invoice image acquisition unit is used for acquiring invoice images of a plurality of invoices submitted by the reimbursement applicant;
an initial feature map generation unit, configured to input the invoice images of the multiple invoices obtained by the invoice image acquisition unit into a convolutional neural network to obtain an initial feature map, where the initial feature map has four dimensions: an image width W, an image height H, a channel dimension C and a sample dimension S, wherein the channel dimension C corresponds to the number of convolution kernels of each layer in the convolutional neural network;
a feature matrix generating unit, configured to pool feature matrices, which are composed of image width W and image height H, in the initial feature map obtained by the initial feature map generating unit, so as to obtain a feature matrix with a scale of C × S;
a text information acquisition unit for acquiring reimbursement item descriptions associated with the plurality of invoices and extracting text information in the reimbursement item descriptions;
a semantic feature vector generating unit, configured to pass the text information in the reimbursement item description obtained by the text information obtaining unit through a semantic understanding model based on deep learning to obtain a semantic feature vector;
an attention feature matrix generating unit, configured to calculate a Softmax-like function value of the feature value at each position of the semantic feature vector obtained by the semantic feature vector generating unit, and then weight each feature value of the same sample in the feature matrix obtained by the feature matrix generating unit with a scale of C × S to obtain an attention feature matrix, where the Softmax-like function value is a quotient of a weighted sum of a natural constant exponent function value raised to a negative number of the feature value at each position in the semantic feature vector and a natural constant exponent function value raised to a negative number of the feature value at each position in the semantic feature vector;
a classification feature map generation unit configured to obtain a classification feature map by passing the attention feature matrix obtained by the attention feature matrix generation unit through a plurality of convolution layers;
a classification result generating unit, configured to pass the classification feature map obtained by the classification feature map generating unit through a classifier to obtain a classification result, where the classification result is used to indicate whether an error exists in a submitted invoice; and
and the storage unit is used for storing the invoice images of the invoices in a storage block of a block chain architecture in response to the fact that the classification result obtained by the classification result generation unit is that the submitted invoice has no error.
According to still another aspect of the present application, there is provided an electronic apparatus including: a processor; and a memory having stored therein computer program instructions which, when executed by the processor, cause the processor to perform the method of financial reimbursement invoice management based on blockchain techniques as described above.
According to yet another aspect of the present application, there is provided a computer readable medium having stored thereon computer program instructions which, when executed by a processor, cause the processor to perform the method of financial reimbursement invoice management based on blockchain techniques as described above.
Compared with the prior art, the block chain technology-based financial reimbursement invoice management method, the block chain technology-based financial reimbursement invoice management system and the electronic device provided by the application are used for extracting and identifying the characteristics of the invoice image by adopting a deep neural network model aiming at the storage of distributed data of the block chain and the non-tamper-property and traceability of the data stored in the block chain so as to determine whether the submitted invoice has errors. Like this, through adopting the block chain to store and manage the invoice image, can utilize the tamper-proof nature and the traceability of block chain, detect the error condition of invoice to avoid invoice itself to have the management confusion of financial reimbursement invoice that the mistake leads to.
Drawings
The above and other objects, features and advantages of the present application will become more apparent by describing in more detail embodiments of the present application with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings, like reference numbers generally represent like parts or steps.
FIG. 1 is a block chain-based invoice image database architecture according to an embodiment of the present application;
FIG. 2 is a diagram illustrating an application scenario of a block chain technology-based financial reimbursement invoice management method according to an embodiment of the present application;
FIG. 3 is a flow chart of a method for financial reimbursement invoice management based on blockchain techniques according to an embodiment of the present application;
FIG. 4 is a system architecture diagram illustrating a block-chain based financial reimbursement invoice management method according to an embodiment of the present application;
FIG. 5 is a flowchart of passing text information in the reimbursement item description through a deep learning-based semantic understanding model to obtain semantic feature vectors in a block chain technology-based financial reimbursement invoice management method according to an embodiment of the present application;
fig. 6 is a flowchart illustrating weighting, after calculating Softmax-like function values of feature values at positions of the semantic feature vector, feature values of the same sample in the feature matrix with a scale C × S to obtain an attention feature matrix in the financial reimbursement invoice management method based on the blockchain technique according to the embodiment of the present application;
FIG. 7 is a flowchart of passing the classification feature map through a classifier to obtain classification results in a financial reimbursement invoice management method based on blockchain technology according to an embodiment of the present application;
FIG. 8 is a block diagram of a financial reimbursement invoice management system based on blockchain techniques according to an embodiment of the present application;
FIG. 9 is a block diagram of a semantic feature vector generation unit in a blockchain technology-based financial reimbursement invoice management system according to an embodiment of the present application;
FIG. 10 is a block diagram of an attention feature matrix generation unit in a block-chain technology based financial reimbursement invoice management system according to an embodiment of the present application;
FIG. 11 is a block diagram of a classification result generation unit in a block chain technology-based financial reimbursement invoice management system according to an embodiment of the present application;
FIG. 12 is a block diagram of an electronic device according to an embodiment of the application.
Detailed Description
Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be understood that the described embodiments are only some embodiments of the present application and not all embodiments of the present application, and that the present application is not limited by the example embodiments described herein.
Block chain architecture overview
Fig. 1 illustrates an architectural diagram of a block chain-based invoice image database according to an embodiment of the present application. As shown in fig. 1, the invoice image database based on the block chain according to the embodiment of the present application adopts a typical block chain architecture, and invoice images, such as invoice images P1, P2, …, Pn of departments, subsidiaries, and the like, are stored in respective storage blocks B1, B2, …, Bn constructed in a block chain. Of course, it will be understood by those skilled in the art that the invoice images for different departments may also be stored in separate blocks, for example, one block is dedicated to storing invoice images for department a, and another block is dedicated to storing invoice images for department B.
According to a typical blockchain storage architecture, each block B1, B2, …, Bn includes pointers H1, H2, …, Hn and data portions D1, D2, …, Dn. The pointers H1, H2, …, Hn may be various types of hash pointers, such as SHA-256 hash functions commonly used in blockchain storage architectures, that point to the last chunk.
In the embodiment of the present application, the value of the hash pointer of the next chunk is based on the value of the hash pointer of the previous chunk and the hash function value of the data portion, for example, H2 ═ H1 × H (D1), and H (D1) represents the hash function value of the data portion D1. The value of the hash pointer for the first chunk may be a random value. In this way, any modification to the portion of data within a block will react on the value of the hash pointer of the next block and further change the values of the hash pointers of all subsequent blocks, making modifications to the portion of data virtually impossible.
Also, in each data portion D1, D2, …, Dn, the hash function value for that data portion may be based on a hash function value generated separately for each invoice image in that data portion. For example, all invoice images in the data portion may be stored in a hash pointer-based data structure of the mekerr tree, thereby facilitating the backtracking of specific invoice images through hash pointers and establishing appropriate membership between the respective invoice images.
Here, those skilled in the art can understand that the block chain-based invoice image database according to the embodiment of the present application may adopt any general block chain architecture, and the embodiment of the present application is not intended to limit the specific implementation of the block chain architecture.
Moreover, in the embodiment of the present application, the block chain preferably adopts a private chain or a federation chain, so as to facilitate distributed storage management of the invoice image database in a financial department inside a company or an enterprise, and accordingly, each storage block for storing an invoice image may be configured in advance without being generated based on a consensus algorithm, so that consumption of computing resources caused by the consensus algorithm may be avoided.
That is to say, the block chain architecture of the block chain-based invoice image database according to the embodiment of the present application focuses on storage management of invoice images, and does not relate to a value transfer function based on a block chain similar to electronic money, so that the block chain architecture can be configured in advance at a cloud end by a management department in a company or an enterprise, and is accessed from a terminal by each technical department to upload the invoice images, and is uniformly stored and managed at the cloud end. Therefore, the application of the blockchain architecture can conveniently realize the distributed storage of the invoice image, since each technical department is likely to be distributed in different geographic positions.
On the other hand, each block in the block chain architecture according to the embodiment of the present application may also be associated with a block of the public chain, so that each block has time stamp information corresponding to the associated block of the public chain. Thus, when information requiring a time attribute, such as the uploading time of an invoice image, needs to be recorded to determine whether the invoice image is an early version, the time sequence attribute of each block in the block chain can be utilized.
Overview of a scene
As previously mentioned, expense reimbursement pertains to a company's internal financial management issues that require the company to reimburse expenses to the employee's bank card based on reimbursement documentation personally submitted by the employee. Here, the reimbursement document personally submitted by the employee may contain a large number of invoices. The applicant of the present application considers that the block chain technology is used to solve the problem, but on the one hand, due to the great trend of centralized management of finance, a group finance department desires to uniformly manage and audit the expense reimbursement invoices of departments, subsidiaries and the like, which relates to the distributed storage function of data of the block chain technology, and in addition, due to the importance of financial data, desires to ensure the non-tamper property and traceability of data by using the block chain technology.
Here, even with respect to a certain reimbursement item, there may be a case where a plurality of invoices are submitted by employees, and if there are errors in the invoices themselves, such as duplicate reimbursements, date errors, or even numbered value-added tax invoices, due to the non-tamper-ability of the data stored in the block, there is a problem in managing the financial reimbursement invoices.
Therefore, the applicant of the present application expects that the error condition of the invoice is first detected before the reimbursement invoice is stored in the blockchain architecture, so as to avoid causing subsequent invoice management confusion. Considering that computer vision technology based on deep learning is basically mature in image recognition at present, the applicant of the present application adopts a deep neural network model to perform feature extraction and recognition on an invoice image so as to determine whether errors exist in a submitted invoice.
Based on this, in the technical scheme of this application, obtain the invoice image of a plurality of invoices that reimburse the applicant and submit first, and input it into convolutional neural network in order to obtain initial characteristic map. Here, the initial feature map includes four dimensions, namely an image width W, an image height H, a channel dimension C and a sample dimension S, wherein the channel dimension C corresponds to the number of convolution kernels, i.e., the number of filters, of each layer in the convolutional neural network, and thus the feature matrix of W × H on the channel dimension C can be regarded as a high-dimensional image feature extracted by each filter of the convolutional neural network, such as corresponding to an invoice number, an invoice date, an invoice amount, and the like. Of course, those skilled in the art will appreciate that in the case of deep convolutional neural networks, the high-dimensional image features have not simply corresponded to the physical features described above, but rather a highly abstract representation of the object features in a high-dimensional feature space.
Thus, an average or maximum pooling of a feature matrix of image width W and image height H is performed on the initial feature map to obtain a feature matrix of C × S, wherein the feature matrix corresponds to each feature value of C, e.g. each column in a row represents a different high-dimensional feature extracted by the convolutional neural network for a single sample, and S, i.e. each row, corresponds to a different sample. Here, in order to more fully utilize other reimbursement information associated with invoice reimbursement, text information of reimbursement item description is extracted, converted into a text vector, and passed through a semantic understanding model to obtain a semantic feature vector.
Further, in order to help the high-dimensional image features extracted by the convolutional neural network focus on features meaningful for reimbursement matters by means of the semantic feature vectors, Softmax-like functions of feature values at positions of the semantic feature vectors are calculated and then the feature values of the same sample in the feature matrix of C × S are weighted, i.e., the feature values in each row are weighted as described above, and the Softmax-like functions are expressed as exp (-xi)/∑ eiexp (-xi). Thus, it is equivalent to applying an attention mechanism to the feature matrix by sample based on the text semantic information to obtain an attention feature matrix.
Then, in order to further mine the correlation between the eigenvalues corresponding to the samples in the attention feature matrix, the attention feature matrix is further passed through a plurality of convolutional layers to obtain a classification feature map, and the classification feature map is input to a classifier to obtain a classification result. The classification result indicates whether the submitted invoice has errors, and accordingly, if the submitted invoice has no errors, the invoice image can be stored in the storage block of the block chain architecture.
Fig. 2 illustrates an application scenario diagram of a financial reimbursement invoice management method based on a blockchain technique according to an embodiment of the application. As shown in fig. 2, in this application scenario, first, invoice images of a plurality of invoices submitted by a reimbursement applicant and their associated reimbursement items descriptions are acquired from a terminal device (e.g., D as illustrated in fig. 2); the invoice images and their associated reimbursement statement for the plurality of invoices are then input into a server (e.g., cloud server S as illustrated in fig. 2) deployed with a financial reimbursement invoice management algorithm based on blockchain techniques, wherein the server is capable of processing the invoice images and their associated reimbursement statement for the plurality of invoices based on the financial reimbursement invoice management algorithm of blockchain techniques to generate a classification result indicating whether the submitted invoice is erroneous. Then, in response to the classification result being that there is no error for the submitted invoice, storing the invoice images of the plurality of invoices in a storage block of a block chain architecture (e.g., block T as illustrated in fig. 2).
Having described the general principles of the present application, various non-limiting embodiments of the present application will now be described with reference to the accompanying drawings.
Exemplary method
FIG. 3 illustrates a flow chart of a method of financial reimbursement invoice management based on blockchain techniques. As shown in fig. 3, a block chain technology-based financial reimbursement invoice management method according to an embodiment of the present application includes: s110, acquiring invoice images of a plurality of invoices submitted by the reimbursement applicant; s120, inputting the invoice images of the invoices into a convolutional neural network to obtain an initial feature map, wherein the initial feature map has four dimensions: an image width W, an image height H, a channel dimension C and a sample dimension S, wherein the channel dimension C corresponds to the number of convolution kernels of each layer in the convolutional neural network; s130, pooling a feature matrix consisting of the image width W and the image height H in the initial feature map to obtain a feature matrix with the dimension C S; s140, acquiring reimbursement item descriptions associated with the plurality of invoices and extracting text information in the reimbursement item descriptions; s150, enabling the text information in the reimbursement item description to pass through a semantic understanding model based on deep learning to obtain a semantic feature vector; s160, calculating Softmax-like function values of feature values at respective positions of the semantic feature vector, and then weighting each feature value of the same sample in the feature matrix with a scale of C × S to obtain an attention feature matrix, where the Softmax-like function value is a quotient of a weighted sum of a natural constant exponent function value raised by a negative number of the feature value at each position in the semantic feature vector divided by a natural constant exponent function value raised by a negative number of the feature value at each position in the semantic feature vector; s170, passing the attention feature matrix through a plurality of convolution layers to obtain a classification feature map; s180, enabling the classification characteristic graph to pass through a classifier to obtain a classification result, wherein the classification result is used for indicating whether errors exist in the submitted invoices or not; and S190, in response to the classification result being that the submitted invoices have no errors, storing the invoice images of the invoices in a storage block of a block chain architecture.
Fig. 4 is a schematic diagram illustrating an architecture of a block chain technology-based financial reimbursement invoice management method according to an embodiment of the present application. As shown IN fig. 4, IN the network architecture of the block chain technology-based financial reimbursement invoice management method, first, invoice images (e.g., IN1 as illustrated IN fig. 4) of a plurality of invoices submitted by reimbursement applicants are acquired; then, inputting the invoice images of the plurality of invoices into a convolutional neural network (e.g., CNN as illustrated in fig. 4) to obtain an initial feature map (e.g., F1 as illustrated in fig. 4); then, pooling a feature matrix consisting of an image width W and an image height H in the initial feature map to obtain a feature matrix with a dimension C × S (e.g., M1 as illustrated in fig. 4); next, obtaining reimbursement item descriptions associated with the plurality of invoices and extracting textual information IN the reimbursement item descriptions (e.g., as illustrated IN2 IN fig. 4); then, passing the text information in the reimbursement item description through a deep learning-based semantic understanding model (e.g., SUM as illustrated in fig. 4) to obtain a semantic feature vector (e.g., V1 as illustrated in fig. 4); then, after calculating a Softmax-like function value of the feature value at each position of the semantic feature vector, weighting each feature value of the same sample in the feature matrix with the scale C × S to obtain an attention feature matrix (e.g., as illustrated in fig. 4, M2); then, passing the attention feature matrix through a plurality of convolutional layers (e.g., CL as illustrated in fig. 4) to obtain a classification feature map (e.g., Fc as illustrated in fig. 4); then, the classification feature map is passed through a classifier (e.g., a classifier as illustrated in fig. 4) to obtain a classification result, which is used to indicate whether there is an error in the submitted invoice; then, in response to the classification result being that there is no error for the submitted invoice, storing the invoice images of the plurality of invoices in a storage block of a blockchain architecture (e.g., T as illustrated in fig. 4).
In step S110, invoice images of a plurality of invoices submitted by the reimbursement applicant are obtained. As mentioned above, the inventors of the present application expect that the error condition of the invoice is first detected before the reimbursement invoice is stored in the blockchain architecture, so as to avoid causing subsequent invoice management confusion. The method and the device adopt a deep neural network model to perform feature extraction and recognition on the invoice image so as to determine whether errors exist in the submitted invoice. Therefore, in the embodiment of the present application, the invoice images of a plurality of invoices submitted by the reimbursement applicant are obtained first. Here, in the embodiment of the present application, the invoice image may be an image generated by shooting a paper invoice, or an electronic image formed by scanning, which is not limited in the present application.
In step S120, the invoice images of the plurality of invoices are input to a convolutional neural network to obtain an initial feature map, wherein the initial feature map has four dimensions: an image width W, an image height H, a channel dimension C, and a sample dimension S, wherein the channel dimension C corresponds to the number of convolution kernels for each layer in the convolutional neural network. That is, the convolutional neural network extracts high-dimensional features in the invoice images of the plurality of invoices.
Specifically, in the present embodiment, in the initial feature map, a feature matrix composed of an image width W and an image height H represents high-dimensional image features extracted by the respective filters of the convolutional neural network. It should be understood that the channel dimension C corresponds to the number of convolution kernels, i.e., the number of filters, of each layer in the convolutional neural network, and thus the feature matrix of W × H on the channel dimension C can be regarded as a high-dimensional image feature extracted by each filter of the convolutional neural network, such as a feature corresponding to an invoice number, an invoice date, an invoice amount, and the like. Of course, those skilled in the art will appreciate that the high-dimensional image features may not simply correspond to the above-described physical features under the influence of a deep convolutional neural network, but rather a highly abstract representation of the object features in a high-dimensional feature space.
In particular, in the embodiment of the present application, the convolutional neural network is a deep residual network. Those skilled in the art will appreciate that deep networks are difficult to train because the gradient disappears, because the gradient propagates back to the previous layer, and repeated multiplication may make the gradient infinitesimally small, with the result that the performance tends to saturate, or even drop off rapidly, as the number of layers in the network is deeper. The residual error network is characterized by easy optimization and can improve the accuracy by increasing the equivalent depth. An identical shortcut key (also called jump connecting line) is introduced into an internal residual block, one or more layers are directly skipped, and the structure is stacked on the network, so that even if the gradient disappears, the original output is at least mapped onto the past in an identical manner, namely a 'copy layer' is stacked on a shallow network, and the gradient disappearance problem caused by depth increase in a deep neural network is relieved.
In step S130, a feature matrix composed of the image width W and the image height H in the initial feature map is pooled to obtain a feature matrix with a scale C × S. That is, the average value pooling or the maximum value pooling of the feature matrix composed of the image width W and the image height H is performed on the initial feature map to obtain a feature matrix of C × S, wherein the feature matrix corresponds to each feature value of C, for example, each column in one row represents a different high-dimensional feature extracted by the convolutional neural network for a single sample, and S, i.e., each row, corresponds to a different sample. It will be appreciated by those skilled in the art that pooling, also known as down-sampling, acts to reduce the size of the image by averaging or maximizing the values in the feature matrix of W H by global mean pooling or global maximum pooling, assigning the corresponding position of the output (i.e., representing all the feature representations in the feature matrix by global mean or global maximum values), and thus reducing the initial feature map with four dimensions to a feature matrix with a scale of cs. Although the feature matrix with the scale C x S has relatively less information than the initial feature map, it reduces the amount of computation.
In step S140, reimbursement item descriptions associated with the plurality of invoices are acquired and text information in the reimbursement item descriptions is extracted. That is, in order to more fully utilize other reimbursement information associated with invoice reimbursement, reimbursement item descriptions associated with the plurality of invoices are obtained and textual information in the reimbursement item descriptions is extracted.
In step S150, the text information in the reimbursement item description is passed through a semantic understanding model based on deep learning to obtain a semantic feature vector. That is, the text information is processed by the semantic understanding model.
Specifically, in this embodiment of the present application, the process of passing the text information in the reimbursement item description through a deep learning-based semantic understanding model to obtain a semantic feature vector includes: firstly, embedding a text information input word in the reimbursement item description into a model to obtain a text vector. It should be understood that text is a very important type of unstructured data, and text can be converted into structured data, i.e., text vectors, through a word embedding model. Then, the text vector is input into a bidirectional long-short term memory model to obtain the semantic feature vector.
It should be known to those skilled in the art that a Long Short-Term Memory network (LSTM) is a time-cycle neural network, which enables weights of the neural network to be updated by adding an input gate, an output gate and a forgetting gate, and the scale of the weights at different times can be dynamically changed with fixed parameters of the network model, so as to avoid the problem of gradient disappearance or gradient expansion. The bidirectional long and short term memory model is formed by combining a forward LSTM and a backward LSTM, the forward LSTM can learn the information of the front text of the current word, and the backward LSTM can learn the information of the subsequent text of the current word, so the semantic feature vector obtained by the bidirectional long and short term memory model learns the information of the text vector context.
Fig. 5 illustrates a flowchart of passing text information in the reimbursement item description through a deep learning-based semantic understanding model to obtain semantic feature vectors in a block chain technology-based financial reimbursement invoice management method according to an embodiment of the present application. As shown in fig. 5, in the embodiment of the present application, passing the text information in the reimbursement item description through a deep learning-based semantic understanding model to obtain a semantic feature vector includes: s210, embedding a text information input word in the reimbursement item description into a model to obtain a text vector; and S220, inputting the text vector into a bidirectional long-short term memory model to obtain the semantic feature vector.
In step S160, a Softmax-like function value of the feature value at each position of the semantic feature vector is calculated, and then each feature value of the same sample in the feature matrix with the scale C × S is weighted to obtain an attention feature matrix, where the Softmax-like function value is a quotient of a weighted sum of a natural constant exponent function value raised to the power of the negative number of the feature value at each position in the semantic feature vector and a natural constant exponent function value raised to the power of the negative number of the feature value at each position in the semantic feature vector. In order to help the high-dimensional image features extracted by the convolutional neural network to focus on the features meaningful to the reimbursement matters by means of the semantic feature vectors, Softmax-like functions of feature values of positions of the semantic feature vectors are calculated and then the Softmax-like functions are carried out on all the features of the same sample in a feature matrix of C & ltS & gtThe eigenvalues are weighted, i.e. as described above for each eigenvalue in each row, the Softmax-like function is expressed as exp (-xi)/∑iexp (-xi). Thus, it is equivalent to applying an attention mechanism to the feature matrix by sample based on the text semantic information to obtain an attention feature matrix.
Specifically, in this embodiment of the present application, after calculating the Softmax-like function values of the eigenvalues at each position of the semantic eigenvector, weighting each eigenvalue of the same sample in the eigenvalue matrix with the scale C × S to obtain the attention eigen matrix, includes: firstly, calculating a Softmax-like function value of the feature value of each position of the semantic feature vector by using a first formula, wherein the first formula is as follows: yi is exp (-xi)/[ sigma ]iexp (-xi), Yi represents the Softmax-like function value, and xi represents the feature value of each position in the semantic feature vector. Then, weighting each eigenvalue of the same sample in the feature matrix with the scale C S by the Softmax-like function value to obtain the attention feature matrix. That is, the characteristic matrix is weighted by using the Softmax-like function values of the characteristic values of the positions of the semantic characteristic vector as the weighting coefficients of each column in the characteristic matrix, so that different weights are given to the characteristic values of the same sample in the characteristic matrix to obtain the attention matrix. In this way, the semantic feature vector can be used to help the high-dimensional image features extracted by the convolutional neural network focus on the features meaningful for the reimbursement.
Fig. 6 is a flowchart illustrating that, in the block chain technology-based financial reimbursement invoice management method according to the embodiment of the present application, after Softmax-like function values of feature values at positions of the semantic feature vectors are calculated, feature values of the same sample in the feature matrix with the scale C × S are weighted to obtain an attention feature matrix. As shown in fig. 6, in the embodiment of the present application, after calculating the Softmax-like function values of the eigenvalues at each position of the semantic eigenvector, weighting each eigenvalue of the same sample in the eigenvalue matrix with the scale C × S to obtain the attention eigen matrix, includes: s310, calculating the first formulaA Softmax-like function value of the eigenvalues at each position of the semantic eigenvector, wherein the first formula is: yi is exp (-xi)/[ sigma ]iexp (-xi), Yi represents the Softmax-like function value, xi represents the feature value of each position in the semantic feature vector; and S320, weighting each characteristic value of the same sample in the characteristic matrix with the scale C S by the Softmax-like function value to obtain the attention characteristic matrix.
In step S170, the attention feature matrix is passed through a plurality of convolutional layers to obtain a classification feature map. That is, the correlation between the eigenvalues corresponding to the respective samples in the attention feature matrix is further mined by a plurality of convolutional layers. As will be appreciated by those skilled in the art, the convolutional layer may perform feature extraction on the input attention feature matrix through a convolution operation, where features learned by one layer of convolution are often local, and the higher the number of layers, the more global the learned features are.
In step S180, the classification feature map is passed through a classifier to obtain a classification result, which is used to indicate whether there is an error in the submitted invoice.
Specifically, in the embodiment of the present application, the process of passing the classification feature map through a classifier to obtain a classification result includes: first, the classification feature map is passed through one or more fully connected layers to obtain a classification feature vector. That is, the classification feature map is encoded using one or more fully-connected layers as an encoder to fully utilize information at various locations in the classification feature map to generate a classification feature vector. The classification feature vector is then input into a Softmax classification function to obtain a first probability that the submitted invoice is erroneous and a second probability that the submitted invoice is non-erroneous. Then, based on the first probability and the second probability, the classification result is determined. That is, the larger of the first probability and the second probability is determined as a classification result.
Fig. 7 illustrates a flowchart of passing the classification feature map through a classifier to obtain a classification result in a financial reimbursement invoice management method based on a blockchain technique according to an embodiment of the present application. As shown in fig. 7, in the embodiment of the present application, passing the classification feature map through a classifier to obtain a classification result includes: s410, enabling the classification feature map to pass through one or more full-connection layers to obtain a classification feature vector; s420, inputting the classification feature vector into a Softmax classification function to obtain a first probability that errors exist in the submitted invoice and a second probability that errors do not exist in the submitted invoice; and S430, determining the classification result based on the first probability and the second probability.
In step S190, in response to the classification result being that there is no error in the submitted invoices, storing the invoice images of the plurality of invoices in a storage block of a block chain architecture. That is, the invoice image is managed using a blockchain technique to facilitate distributed storage and to ensure the non-tamper and traceability of the data.
In summary, the financial reimbursement invoice management method based on the block chain technology is stated, and the method adopts a deep neural network model to perform feature extraction and identification on an invoice image according to the storage of distributed data of the block chain and the non-tamper-ability and traceability of the data stored by the block chain so as to determine whether an error exists in a submitted invoice. Like this, through adopting the block chain to store and manage the invoice image, can utilize the tamper-proof nature and the traceability of block chain, detect the error condition of invoice to avoid invoice itself to have the management confusion of financial reimbursement invoice that the mistake leads to.
Exemplary System
FIG. 8 illustrates a block diagram of a financial reimbursement invoice management system based on blockchain techniques according to an embodiment of the present application. As shown in fig. 8, a block chain technology-based financial reimbursement invoice management system 800 according to an embodiment of the present application includes: an invoice image obtaining unit 810, configured to obtain invoice images of a plurality of invoices submitted by the reimbursement applicant; an initial feature map generating unit 820, configured to input the invoice images of the multiple invoices obtained by the invoice image obtaining unit 810 into a convolutional neural network to obtain an initial feature map, where the initial feature map has four dimensions: an image width W, an image height H, a channel dimension C and a sample dimension S, wherein the channel dimension C corresponds to the number of convolution kernels of each layer in the convolutional neural network; a feature matrix generating unit 830, configured to pool feature matrices, which are composed of the image width W and the image height H, in the initial feature map obtained by the initial feature map generating unit 820, so as to obtain a feature matrix with a scale C × S; a text information obtaining unit 840 configured to obtain reimbursement item descriptions associated with the plurality of invoices and extract text information in the reimbursement item descriptions; a semantic feature vector generation unit 850, configured to pass the text information in the reimbursement item description obtained by the text information obtaining unit 840 through a semantic understanding model based on deep learning to obtain a semantic feature vector; an attention feature matrix generation unit 860 configured to calculate a Softmax-like function value of the feature value at each position of the semantic feature vector obtained by the semantic feature vector generation unit 850, and then weight each feature value of the same sample in the feature matrix obtained by the feature matrix generation unit with a scale of C × S to obtain an attention feature matrix, where the Softmax-like function value is a quotient of a weighted sum of a natural constant exponent function value raised to a negative number of the feature value at each position of the semantic feature vector and a natural constant exponent function value raised to a negative number of the feature value at each position of the semantic feature vector; a classification feature map generation unit 870 for passing the attention feature matrix obtained by the attention feature matrix generation unit 860 through a plurality of convolutional layers to obtain a classification feature map; a classification result generating unit 880, configured to pass the classification feature map obtained by the classification feature map generating unit 870 through a classifier to obtain a classification result, where the classification result is used to indicate whether an error exists in the submitted invoice; and a storage unit 890, configured to store the invoice images of the multiple invoices in a storage block of a block chain architecture in response to that the classification result obtained by the classification result generation unit 880 is that there is no error in the submitted invoice.
In one example, in the above-described financial reimbursement invoice management system 800, in the initial feature map, a feature matrix composed of an image width W and an image height H represents high-dimensional image features extracted by the respective filters of the convolutional neural network.
In an example, in the above-mentioned financial reimbursement invoice management system 800, as shown in fig. 9, the semantic feature vector generation unit 850 includes: a text vector generation subunit 851, configured to embed the text information input word in the reimbursement item description into a model to obtain a text vector; and a feature extraction subunit 852, configured to input the text vector obtained by the text vector generation subunit 851 into a bidirectional long-short term memory model to obtain the semantic feature vector.
In one example, in the above-mentioned financial reimbursement invoice management system 800, as shown in fig. 10, the attention feature matrix generation unit 860 includes: a function value generating subunit 861, configured to calculate a Softmax-like function value of the feature value at each position of the semantic feature vector according to a first formula, where the first formula is: yi is exp (-xi)/∑ i exp (-xi), Yi represents the Softmax-like function value, xi represents the eigenvalue of each position in the semantic eigenvector; and a weighting subunit 862 configured to weight, with the Softmax-like function value obtained by the function value generation subunit 861, each feature value of the same sample in the feature matrix with the scale C × S to obtain the attention feature matrix.
In an example, in the above-mentioned financial reimbursement invoice management system 800, as shown in fig. 11, the classification result generating unit 880 includes: a classification feature vector generation subunit 881, configured to pass the classification feature map through one or more fully connected layers to obtain a classification feature vector; a probability generating subunit 882, configured to input the classification feature vector obtained by the classification feature vector generating subunit 881 into a Softmax classification function to obtain a first probability that the submitted invoice has an error and a second probability that the submitted invoice has no error; and a determining subunit 883, configured to determine the classification result based on the first probability and the second probability obtained by the probability generating subunit 882.
In one example, in the financial reimbursement invoice management system 800 described above, the convolutional neural network is a deep residual network.
Here, it can be understood by those skilled in the art that the detailed functions and operations of the respective units and modules in the above-described financial reimbursement invoice management system 800 have been described in detail in the above description of the block chain technology-based financial reimbursement invoice management method with reference to fig. 1 to 7, and thus, a repetitive description thereof will be omitted.
As described above, the financial reimbursement invoice management system 800 according to the embodiment of the present application may be implemented in various terminal devices, such as a server for financial reimbursement invoice management, and the like. In one example, the financial reimbursement invoice management system 800 according to embodiments of the present application may be integrated into the terminal device as a software module and/or a hardware module. For example, the financial reimbursement invoice management system 800 may be a software module in the operating system of the terminal device, or may be an application developed for the terminal device; of course, the financial reimbursement invoice management system 800 may also be one of many hardware modules of the terminal device.
Alternatively, in another example, the financial reimbursement invoice management system 800 and the terminal device may be separate devices, and the financial reimbursement invoice management system 800 may be connected to the terminal device through a wired and/or wireless network and transmit the interactive information in an agreed data format.
Exemplary electronic device
Next, an electronic apparatus according to an embodiment of the present application is described with reference to fig. 12. As shown in fig. 12, the electronic device 10 includes one or more processors 11 and a memory 12. The processor 11 may be a Central Processing Unit (CPU) or other form of processing unit having data processing capabilities and/or instruction execution capabilities, and may control other components in the electronic device 10 to perform desired functions.
Memory 12 may include one or more computer program products that may include various forms of computer-readable storage media, such as volatile memory and/or non-volatile memory. The volatile memory may include, for example, Random Access Memory (RAM), cache memory (cache), and/or the like. The non-volatile memory may include, for example, Read Only Memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium and executed by processor 11 to implement the functions of the blockchain technology-based financial reimbursement invoice management method of the various embodiments of the present application described above and/or other desired functions. Various contents such as a semantic feature vector, an attention feature matrix, etc. may also be stored in the computer-readable storage medium.
In one example, the electronic device 10 may further include: an input system 13 and an output system 14, which are interconnected by a bus system and/or other form of connection mechanism (not shown).
The input system 13 may comprise, for example, a keyboard, a mouse, etc.
The output system 14 may output various information including classification results and the like to the outside. The output system 14 may include, for example, a display, speakers, a printer, and a communication network and its connected remote output devices, among others.
Of course, for simplicity, only some of the components of the electronic device 10 relevant to the present application are shown in fig. 12, and components such as buses, input/output interfaces, and the like are omitted. In addition, the electronic device 10 may include any other suitable components depending on the particular application.

Claims (10)

1. A financial reimbursement invoice management method based on a block chain technology is characterized by comprising the following steps:
acquiring invoice images of a plurality of invoices submitted by reimbursement applicants;
inputting invoice images of the plurality of invoices into a convolutional neural network to obtain an initial feature map, wherein the initial feature map has four dimensions: an image width W, an image height H, a channel dimension C and a sample dimension S, wherein the channel dimension C corresponds to the number of convolution kernels of each layer in the convolutional neural network;
pooling a feature matrix consisting of image width W and image height H in the initial feature map to obtain a feature matrix with the dimension C S;
acquiring reimbursement item descriptions associated with the plurality of invoices and extracting text information in the reimbursement item descriptions;
enabling the text information in the reimbursement item description to pass through a semantic understanding model based on deep learning to obtain a semantic feature vector;
calculating Softmax-like function values of feature values of all positions of the semantic feature vector, and weighting all feature values of the same sample in the feature matrix with the scale of C S to obtain an attention feature matrix, wherein the Softmax-like function values are quotient of weighted sum of natural constant exponent function values taking the negative number of the feature values of all positions in the semantic feature vector as power and natural constant exponent function values taking the negative number of the feature values of all positions in the semantic feature vector as power;
passing the attention feature matrix through a plurality of convolutional layers to obtain a classification feature map;
passing the classification feature map through a classifier to obtain a classification result, wherein the classification result is used for indicating whether errors exist in the submitted invoice or not; and
in response to the classification result being that there is no error in the submitted invoices, storing invoice images of the plurality of invoices in a storage block of a block chain architecture.
2. The financial reimbursement invoice management method based on blockchain technique of claim 1, wherein in said initial feature map, a feature matrix consisting of image width W and image height H represents high-dimensional image features extracted by respective filters of said convolutional neural network.
3. The method for financial reimbursement invoice management based on blockchain technology as claimed in claim 1, wherein passing the text information in the reimbursement item description through a deep learning based semantic understanding model to obtain semantic feature vectors comprises:
embedding text information input words in the reimbursement item description into a model to obtain a text vector;
and inputting the text vector into a bidirectional long-short term memory model to obtain the semantic feature vector.
4. The method for managing financial reimbursement invoices according to claim 1, wherein the step of weighting each eigenvalue of the same sample in the eigenvalue matrix with the scale of C S after calculating the Softmax-like function value of the eigenvalue of each position of the semantic eigenvector to obtain the attention eigenvalue matrix comprises the steps of:
calculating Softmax-like function values of the feature values of the semantic feature vectors at the positions by using a first formula, wherein the first formula is as follows: yi is exp (-xi)/[ sigma ]iexp (-xi), Yi represents the Softmax-like function value, xi represents the feature value of each position in the semantic feature vector; and
weighting each characteristic value of the same sample in the characteristic matrix with the scale C S by the Softmax-like function value to obtain the attention characteristic matrix.
5. The method for financial reimbursement invoice management based on blockchain technology as claimed in claim 1, wherein passing said classification feature map through a classifier to obtain classification results comprises:
passing the classification feature map through one or more fully connected layers to obtain a classification feature vector;
inputting the classification feature vector into a Softmax classification function to obtain a first probability that the submitted invoice is erroneous and a second probability that the submitted invoice is not erroneous; and
determining the classification result based on the first probability and the second probability.
6. The method for financial reimbursement invoice management based on block-chain techniques of claim 1, wherein said convolutional neural network is a deep residual network.
7. A financial reimbursement invoice management system based on blockchain technology, comprising:
the invoice image acquisition unit is used for acquiring invoice images of a plurality of invoices submitted by the reimbursement applicant;
an initial feature map generation unit, configured to input the invoice images of the multiple invoices obtained by the invoice image acquisition unit into a convolutional neural network to obtain an initial feature map, where the initial feature map has four dimensions: an image width W, an image height H, a channel dimension C and a sample dimension S, wherein the channel dimension C corresponds to the number of convolution kernels of each layer in the convolutional neural network;
a feature matrix generating unit, configured to pool feature matrices, which are composed of image width W and image height H, in the initial feature map obtained by the initial feature map generating unit, so as to obtain a feature matrix with a scale of C × S;
a text information acquisition unit for acquiring reimbursement item descriptions associated with the plurality of invoices and extracting text information in the reimbursement item descriptions;
a semantic feature vector generating unit, configured to pass the text information in the reimbursement item description obtained by the text information obtaining unit through a semantic understanding model based on deep learning to obtain a semantic feature vector;
an attention feature matrix generating unit, configured to calculate a Softmax-like function value of the feature value at each position of the semantic feature vector obtained by the semantic feature vector generating unit, and then weight each feature value of the same sample in the feature matrix obtained by the feature matrix generating unit with a scale of C × S to obtain an attention feature matrix, where the Softmax-like function value is a quotient of a weighted sum of a natural constant exponent function value raised to a negative number of the feature value at each position in the semantic feature vector and a natural constant exponent function value raised to a negative number of the feature value at each position in the semantic feature vector;
a classification feature map generation unit configured to obtain a classification feature map by passing the attention feature matrix obtained by the attention feature matrix generation unit through a plurality of convolution layers;
a classification result generating unit, configured to pass the classification feature map obtained by the classification feature map generating unit through a classifier to obtain a classification result, where the classification result is used to indicate whether an error exists in a submitted invoice; and
and the storage unit is used for storing the invoice images of the invoices in a storage block of a block chain architecture in response to the fact that the classification result obtained by the classification result generation unit is that the submitted invoice has no error.
8. The financial reimbursement invoice management system for blockchain-based technologies of claim 7, wherein said semantic feature vector generation unit comprises:
the text vector generation subunit is used for embedding the text information input words in the reimbursement item description into a model to obtain a text vector;
and the feature extraction subunit is used for inputting the text vector obtained by the text vector generation subunit into a bidirectional long-short term memory model so as to obtain the semantic feature vector.
9. The financial reimbursement invoice management system for blockchain-based techniques of claim 7, wherein said attention feature matrix generation unit comprises:
a function value generating subunit, configured to calculate a Softmax-like function value of the feature value at each position of the semantic feature vector by using a first formula, where the first formula is: yi is exp (-xi)/[ sigma ]iexp (-xi), Yi represents the Softmax-like function value, xi represents the feature value of each position in the semantic feature vector; and
a weighting subunit, configured to weight, with the Softmax-like function value obtained by the function value generating subunit, each feature value of the same sample in the feature matrix with the scale C × S to obtain the attention feature matrix.
10. An electronic device, comprising:
a processor; and
memory having stored therein computer program instructions which, when executed by the processor, cause the processor to perform the method of block-chain technology based financial reimbursement invoice management according to any of claims 1-6.
CN202110535995.1A 2021-05-17 2021-05-17 Financial reimbursement invoice management method based on block chain technology Withdrawn CN113255498A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110535995.1A CN113255498A (en) 2021-05-17 2021-05-17 Financial reimbursement invoice management method based on block chain technology

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110535995.1A CN113255498A (en) 2021-05-17 2021-05-17 Financial reimbursement invoice management method based on block chain technology

Publications (1)

Publication Number Publication Date
CN113255498A true CN113255498A (en) 2021-08-13

Family

ID=77182403

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110535995.1A Withdrawn CN113255498A (en) 2021-05-17 2021-05-17 Financial reimbursement invoice management method based on block chain technology

Country Status (1)

Country Link
CN (1) CN113255498A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114238765A (en) * 2021-12-16 2022-03-25 吉林大学 Block chain-based position attention recommendation method
CN115375980A (en) * 2022-06-30 2022-11-22 杭州电子科技大学 Block chain-based digital image evidence storing system and method
CN115775116A (en) * 2023-02-13 2023-03-10 华设设计集团浙江工程设计有限公司 BIM-based road and bridge engineering management method and system

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114238765A (en) * 2021-12-16 2022-03-25 吉林大学 Block chain-based position attention recommendation method
CN115375980A (en) * 2022-06-30 2022-11-22 杭州电子科技大学 Block chain-based digital image evidence storing system and method
CN115375980B (en) * 2022-06-30 2023-05-09 杭州电子科技大学 Digital image certification system and certification method based on blockchain
CN115775116A (en) * 2023-02-13 2023-03-10 华设设计集团浙江工程设计有限公司 BIM-based road and bridge engineering management method and system

Similar Documents

Publication Publication Date Title
US11776244B2 (en) Systems and methods for generating and using semantic images in deep learning for classification and data extraction
US11481605B2 (en) 2D document extractor
US20210150338A1 (en) Identification of fields in documents with neural networks without templates
US11074442B2 (en) Identification of table partitions in documents with neural networks using global document context
EP3640847A1 (en) Systems and methods for identifying form fields
US11157816B2 (en) Systems and methods for selecting and generating log parsers using neural networks
US11816138B2 (en) Systems and methods for parsing log files using classification and a plurality of neural networks
CN113255498A (en) Financial reimbursement invoice management method based on block chain technology
US11830269B2 (en) System for information extraction from form-like documents
CN113255496A (en) Financial expense reimbursement management method based on block chain technology
CN114139522B (en) Key information identification method based on hierarchical attention and label guided learning
US11232299B2 (en) Identification of blocks of associated words in documents with complex structures
US20230138491A1 (en) Continuous learning for document processing and analysis
RU2768233C1 (en) Fuzzy search using word forms for working with big data
CN115860271A (en) System and method for managing art design scheme
Ciaburro et al. Python Machine Learning Cookbook: Over 100 recipes to progress from smart data analytics to deep learning using real-world datasets
EP4089568A1 (en) Cascade pooling for natural language document processing
CN111008624A (en) Optical character recognition method and method for generating training sample for optical character recognition
CN112434889A (en) Expert industry analysis method, device, equipment and storage medium
CA3060293A1 (en) 2d document extractor
CN113297849B (en) Financial pre-charge management method based on block chain technology
CN112347738B (en) Bidirectional encoder characterization quantity model optimization method and device based on referee document
US20240143906A1 (en) Character level embeddings for spreadsheet data extraction using machine learning
CN118506391A (en) Bill information extraction method, device, computer equipment and storage medium
CN113158689A (en) Product detection report management method based on block chain technology

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
WW01 Invention patent application withdrawn after publication

Application publication date: 20210813

WW01 Invention patent application withdrawn after publication