TWI716761B - Intelligent accounting system and identification method for accounting documents - Google Patents

Abstract

An intelligent accounting system and an identification method for accounting documents, the intelligent accounting system is electrically connected to a client-end electronic device, the client-end electronic device stores the document image data, and the intelligent accounting system includes a document memory module and a document management module. The document memory module is configured to store the document image data transmitted by the client-end electronic device, and the document management module captures the document image data in the document memory module. The document management module includes an image processing unit, a text recognition unit, and a semantic analysis unit. The image processing unit performs image processing on the document image data, and the text recognition unit converts the document image data into the document text data, and the semantic analysis unit receives the document text data transmitted from the text recognition unit and analyzes the document text data and classifies its accounting items to generate a billing data.

Description

Intelligent accounting system and the method of identification and entry of accounting documents

The present invention provides an intelligent accounting system and an identification method of accounting vouchers, in particular to an intelligent accounting system with automatic processing of accounting data and an identification method of accounting vouchers.

Today's small and medium-sized enterprises usually do not set up a separate accounting department, so every month at the end of the month, the piles of invoices and the densely packed income and expenditure details on the account are definitely a headache for the managers of small and medium-sized enterprises. In addition, the accounting process is cumbersome and error-proof. Most managers of small and medium-sized enterprises have limited accounting professional resources, so they appoint an external bookkeeper to help manage the accounts. However, just recording and categorizing various vouchers (such as invoices, electronic invoices, import declarations, receipts, accounting vouchers, etc.) will consume a lot of labor and time, resulting in poor timeliness. And manual processing and classification are also prone to errors or omissions in accounting.

Therefore, how to manage the accounting affairs of enterprises more efficiently and accurately, and reduce the errors of human error, is worth considering for those with ordinary knowledge in this field.

The purpose of the present invention is to provide an intelligent accounting system and a method for identifying accounting vouchers. The intelligent accounting system and the method for identifying accounting vouchers can efficiently manage the enterprise’s intelligent accounting and reduce human errors. .

The invention provides an intelligent accounting system. The intelligent accounting system is communicatively connected to a client electronic device. The client electronic device stores at least one voucher image data, and the voucher image data corresponds to an accounting voucher. The intelligent accounting system includes a certificate memory module and a certificate management module. The certificate memory module is used to store the certificate image data transmitted from the client electronic device, and the certificate management module retrieves the certificate image data in the certificate memory module. The voucher management module includes: an image processing unit, a text recognition unit, a vocabulary, a word vector corpus, a semantic analysis unit, and an automatic correction unit. The vocabulary includes a departmental vocabulary, a product noun vocabulary, and a transaction object vocabulary. The text recognition unit receives the voucher image data, and converts the voucher image data into voucher text data. The lexicon in the lexicon records the correspondence between various product names and accounting items. The word vector corpus includes multiple word vectors, and these word vectors correspond to various product names classified in the lexicon. The semantic analysis unit receives the voucher text data from the text recognition unit. The voucher text data includes a product name and a transaction object data. The semantic analysis unit performs the following steps:

(a1) Search for the product name in the voucher text data in the vocabulary of the product name and the transaction object, and determine whether there is the product name contained in the voucher text data in the vocabulary, and if so, perform the following steps (a2), if it does not exist, proceed to the following step (a3).

(a2) Find out the accounting item corresponding to the product name.

(a3) Find the word vector closest to the product name in the word vector corpus, compare and classify the word vector with the word vector, and find the accounting item with the vector distance closest to the word vector, and Add the product name to the dictionary.

(a4) According to step (a2) or step (a3), an accounting data is generated, the accounting data includes the item name contained in the voucher text data and the accounting item corresponding to the item name.

The present invention provides an accounting voucher identification method, including the following steps: (a) reading a voucher image data, the voucher image data is corresponding to an accounting voucher; (b) Preprocess the voucher image data to improve the clarity of the voucher image data, and measure the appearance of the voucher image data to determine the size of the accounting voucher; (c) Convert the voucher image data into a voucher Text data, the voucher text data includes a product name and a transaction object data; (d) Search for the product name in the voucher text data in a vocabulary database, and determine whether the voucher text exists in the vocabulary The product name contained in the data, if it exists, proceed to the following step (e), if it does not exist, proceed to the following step (f), where the thesaurus records the correspondence between various product names and accounting items; (e) find out the corresponding product name Accounting items; (f) Find the word vector closest to the product name in a word vector corpus, compare the word vector with the word vector, and find the accounting item with the vector distance closest to the word vector, The word vector corpus includes a plurality of word vectors, and these word vectors correspond to various product names in the word database; and (g) according to step (e) or step (f), an accounting data is generated, and the accounting data includes the The first product name and the accounting item corresponding to the first product name.

In order to make the above-mentioned features and advantages of the present invention more obvious and understandable, a detailed description is given below of preferred embodiments in conjunction with the accompanying drawings.

10: Client electronic device

11: lens

12: Client operating program

81: Accounting documents

100: Intelligent accounting system

110: certificate memory module

13: Certificate image data

13’: Voucher text data

120: Certificate Management Module

122: Voucher Classification Unit

122a: Credential feature style classification library

124: image processing unit

126: text recognition unit

127: Accounting Processing Unit

128a: Thesaurus

128b: word vector corpus

129: Correction machine learning unit

23: Accounting information

S110~S195: Step flow chart

131a~131f: field

20: Terminal control unit

2331: subpoena

2332: Journal

2333: report

2334: business tax

83: Electronic invoice

830: Two-dimensional barcode

Figure 1 shows the relationship between the smart accounting system and the client electronic device of this embodiment.

FIG. 2 shows a schematic diagram of taking accounting vouchers with the lens of the client electronic device.

Figure 3 shows a flowchart of the method for identifying accounting documents in this embodiment.

Figure 4 shows a schematic diagram of accounting documents.

Figure 5 shows a more detailed classification of accounting information.

FIG. 6 is a schematic diagram of photographing an electronic invoice with the lens of the client electronic device.

Please refer to FIG. 1. FIG. 1 illustrates the relationship between the smart accounting system of this embodiment and the client electronic device. The client electronic device 10 includes a lens 11 and a client operating program 12, and the client operating program 12 can communicate with the intelligent accounting system 100 for data exchange. Please refer to FIG. 2. FIG. 2 shows a schematic diagram of using the lens 11 of the client electronic device to photograph the accounting voucher 81. The accounting voucher 81 is photographed by the lens 11, and the image of the accounting voucher 81 is stored as voucher image data 13. In this embodiment, the client electronic device 10 is a smart phone, but it is not limited to this. The client electronic device 10 can also be a desktop computer or a notebook computer, and the accounting certificate 81 can be used by a scanner And converted into credential image data 13. In this creation, accounting vouchers refer to written certificates that record economic activities and clarify economic responsibilities. Accounting vouchers are an important basis for registering account books and conducting accounting supervision, such as unified invoices, tax vouchers, electronic invoices, and various receipts.

Please continue to refer to FIG. 1, the intelligent accounting system 100 includes a certificate memory module 110 and a certificate management module 120. The smart accounting system 100 is communicatively connected to a client electronic device 10, so the smart accounting system 100 can store the credential image data 13 transmitted from the client electronic device 10 in the credential memory module 110. The certificate management module 120 retrieves the certificate image data in the certificate memory module 110. The certificate management module 120 includes a certificate classification unit 122, a certificate characteristic pattern classification database 122a, an image processing unit 124, a text recognition unit 126, An accounting processing unit 127, a semantic analysis unit 128, a word database 128a, a word vector corpus 128b, and a correction machine learning unit 129. Among them, the vocabulary 128a includes department vocabulary, product noun vocabulary, and transaction object vocabulary. The item vocabulary records the correspondence between various item names and accounting items. The transaction object vocabulary in the vocabulary 128a also records the transaction object (ie : Customer and supplier), and the department dictionary in the thesaurus 128a records the company's department name. Moreover, the word vector corpus 128b includes a plurality of word vectors, and these word vectors correspond to various product names in the word database 128a. Word vector (also known as word embedding or representation) is a technology that has been widely used in recent years. It uses a vector to represent each word (vector representation).

Please refer to FIG. 1 and FIG. 3 at the same time. FIG. 3 shows a flowchart of the method for identifying accounting documents in this embodiment. First, in step S110, the intelligent accounting system 100 reads the voucher image data 13, which in this embodiment is transmitted from the client electronic device 10, for example. After that, step S120 is implemented, and the image processing unit 124 preprocesses the voucher image data 13 to improve the clarity of the voucher image data 13. In this embodiment, preprocessing refers to image noise reduction or binarization processing on the voucher image data 13 to improve the clarity of the voucher image data 13. In addition, the image processing unit 124 also measures the appearance of the voucher image data 13 to determine the size of the accounting voucher 81. Moreover, the image processing unit 124 can also perform feature point detection on the credential image data 13 to find multiple feature points (not shown).

Then, in step S130, the voucher classification unit 122 classifies the voucher image data 13 by using the voucher feature pattern classification database 122a. Since the voucher feature pattern classification database 122a stores multiple voucher patterns, the voucher classification unit 122 can classify the voucher image data 13, for example, the voucher image data 13 can be classified into water, electricity, and gas charges by the size of the accounting voucher 81 Wait. In this embodiment, the credential classification unit 122 uses Convolutional Neural Networks to classify the credential image data 13, which can further improve the efficiency of classifying the credential image data 13.

Again, please refer to Figure 4 at the same time, which is a schematic diagram of an accounting document. Step S132 is implemented to divide the voucher image data 13 into a plurality of fields 131a-131f based on the feature points detected by the image processing unit 124. After that, step S134 is implemented to label each field of the voucher image data 13, for example, field 131a is marked as the unit of issuing receipt 131a, field 131b is marked as the transaction object, field 131c is marked as product name, field 131d is marked Is the amount, the field 131e is marked as the issuance time, and the field 131f is marked as the receipt number. The reason why the fields 131a to 131f are labeled is that the semantic analysis unit 128 is less likely to make errors when analyzing the data, which will be described in more detail below.

Then, in step S140, the text recognition unit 126 converts the voucher image data 13 that has undergone image processing into a voucher text data 13'. In this embodiment, the text recognition unit 126 uses OCR text recognition technology to convert the text image in the voucher image data 13 into real text data, that is, the voucher text The word data 13' is the data that can be used for text editing by text editing software. The voucher text data 13' includes at least one product name, that is, the product name. In addition, since the image processing unit 124 has already labeled each field of the voucher image data 13, the text recognition unit 126 can also correspond the voucher text data 13' with the tags of each field. For example, the text "○ ○ ○ Sushi Famous Restaurant Co., Ltd." corresponds to the label "Transaction Object" in the field 131b, and the text "NT$63,000" corresponds to the label "Amount" in the field 131d.

In addition, it is mentioned in the previous paragraph that the reason for labeling the fields 131a to 131f is that the semantic analysis unit 128 is less likely to make errors when analyzing the data. As shown in Fig. 4, since the column 131b corresponding to the "transaction target" may have a similar product name (sushi in this embodiment), if the column 131b is labeled in advance, the semantic analysis unit 128 It will not make a mistake and classify the voucher into incorrect accounting items.

Next, step S150 is implemented. The semantic analysis unit 128 searches the vocabulary text data 13' in the vocabulary 128a (for example, "Accounting and Financial Statement Preparation" shown in the field 131c in FIG. 4), and performs a search in step S150. In S160, it is determined whether there is a product name corresponding to the voucher text data 13' in the thesaurus 128a, if it exists, the following step S170 is performed, and if it does not exist, the following step S180 is performed. In step S170, the semantic analysis unit 128 finds the accounting item corresponding to the product name in the voucher text data 13'. For example, "accounting and financial statement preparation" corresponds to the accounting item "labor expenses" (No. 2071). It is worth noting that when the semantic analysis unit 128 finds out the accounting items corresponding to the product name in the voucher text data 13', it will not only refer to the product noun database, but also the department vocabulary and the transaction object vocabulary; , The department reporting the accounting voucher 8 and the transaction object on the accounting voucher 8 will also affect the classification of accounting items.

In step S180, the semantic analysis unit 128 finds the word vector closest to the product name in the voucher text data 13' in the word vector corpus 128b, compares the word vector with the word database 128a, and finds The product name corresponding to the word vector closest to the vector distance of the word vector and the accounting item corresponding to the product name.

After that, step S190 is executed, and the semantic analysis unit 128 generates an accounting data 23 (as shown in FIG. 5). The accounting data 23 is the item name in the voucher text data 13' and the accounting item corresponding to the item name. Please also refer to Figure 5, Figure 5 shows a more detailed classification of accounting information. In other embodiments, after the multiple accounting data 23 are sorted by the accounting processing unit 127, further accounting data such as a voucher 2331, one-day accounting 2332, a statement 2333, and a business tax 2334 are generated. Therefore, compared to using manual to process the internal accounting affairs of the enterprise, the intelligent accounting system 100 of the present invention can calculate the business tax to be paid more efficiently, greatly reducing the time cost.

After step S190 is completed, step S195 is executed, and the correction machine learning unit 129 receives the accounting data 23 from the semantic analysis unit 128, and can generate an audit data page (not shown) on the screen of a terminal control unit 20, For the operator of the terminal control unit 20 to check the accounting data 23, if the correction machine learning unit 129 determines that the correspondence between the product name and the accounting item in the voucher text data 13' is incorrect, then correct the relationship between the product name and the accounting item in the accounting data 23 Correspondence, and add the product name and the corresponding relationship between the revised product name and the accounting item to the thesaurus 128a. That is to say, if the corresponding product name cannot be found in the thesaurus 128a by using the word vector in step S180, and then the classified accounting item cannot be found, the data page is reviewed and machine learning is used to correct the increase in the thesaurus 128a. The product name of and the corresponding word vector in the word vector corpus 128b. In this way, if there is an accounting voucher containing the product name after processing, the accounting item to which it belongs can be immediately determined and classified. Therefore, the semantic analysis unit 128 can also increase the accuracy of subsequent classification of accounting items through the feedback of machine learning as shown in step S195.

In this embodiment, the calibration machine learning unit 129 is communicatively connected to the terminal control unit 20. The operator of the terminal control unit 20 is, for example, a financial staff in the organization. The financial staff uses the terminal control unit 20 to check accounts. Whether the service data 23 is correct.

Here, the thesaurus 128a is further explained. Since the thesaurus 128a stores the correspondence between multiple product names and various accounting items, the semantic analysis unit 128 can sort the voucher text data 13' into accounting data. For example, when the product name in the voucher text data 13' includes the keyword "pencil", if the word database 128a records "pencil" and its corresponding accounting item, the semantic analysis unit 128 can find the accountant in the word database 128a The corresponding name of the item is "Stationery", and the item number is "5153". Or, when the product name in the voucher text data 13' includes keywords such as "water fee" or "electricity fee", if the word database 128a records "water fee" or "electricity fee" and its corresponding accounting item, the semantic analysis unit 128 Find the corresponding name in the thesaurus 128a as "Water and Electricity Gas Charge", and the item number is "5161". In addition, the thesaurus 128a may also record the name, uniform number, or other relevant information of the transaction object, so that the accounting data 23 generated by the semantic analysis unit 128 can include the transaction object data.

Please refer to FIGS. 1 and 6 at the same time. In addition to allowing the client electronic device 10 to read general accounting vouchers, the client operating program 12 also allows the client electronic device 10 to read an electronic invoice 83 to generate voucher text data. 13'. In detail, the electronic invoice 83 itself has a QR code 830, and the basic information of the voucher text 13' is stored in the two-dimensional barcode 830, so when the lens 11 is aligned with the two-dimensional barcode 830 of the electronic invoice 83 At this time, the client electronic device 10 can also read the voucher text data 13' from the two-dimensional barcode 830 and transmit it to the accounting processing unit 127 of the intelligent accounting system 100. In this way, the accounting processing unit 127 can perform subsequent processing (generating a voucher 2331, a journal 2332, a report 2333, and a business tax 2334).

To sum up, the intelligent accounting system of the present invention can convert the image data of accounting documents into text data, and automatically classify accounting items to make various accounting data. In addition, the intelligent accounting system integrates reinforcement learning to improve its recognition rate.

Although the present invention has been disclosed as above in preferred embodiments, it is not intended to limit the present invention. Anyone with ordinary knowledge in the relevant technical field can make some changes and modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be subject to those defined by the attached patent scope.

10: Client electronic device

11: lens

12: Client operating program

13: Certificate image data

13’: Voucher text data

20: Terminal control unit

100: Intelligent accounting system

110: certificate memory module

120: Certificate Management Module

122: Voucher Classification Unit

122a: Credential feature style classification library

124: image processing unit

126: text recognition unit

127: Accounting Processing Unit

128: Semantic Analysis Unit

128a: Thesaurus

128b: word vector corpus

129: Correction machine learning unit

Claims (11)

An intelligent accounting system, which is communicatively connected to a client electronic device. The client electronic device stores at least one voucher image data corresponding to an accounting voucher. The smart accounting system includes: a voucher memory A module for storing the certificate image data transmitted from the client electronic device; a certificate management module, which retrieves the certificate image data in the certificate memory module, the certificate management module including: an image The processing unit preprocesses the voucher image data to improve the clarity of the voucher image data, and measures the appearance of the voucher image data to determine the size of the accounting voucher; a text recognition unit receives the voucher image data , To convert the voucher image data into a voucher text data; a vocabulary to record the correspondence between various product names and accounting items; a word vector corpus, including multiple word vectors, these word vectors correspond to various product names in the vocabulary ; A semantic analysis unit, receiving the voucher text data from the text recognition unit, the voucher text data includes a product name and a transaction partner data, the semantic analysis unit performs the following steps: (a1) in the word database Search for the product name in the voucher text data in the vocabulary, and determine whether the product name contained in the voucher text data exists in the vocabulary. If it exists, proceed to the following step (a2), if not, proceed to the following step (a3 ); (a2) Find out the accounting item corresponding to the product name; (a3) Find the word vector closest to the product name in the word vector corpus, compare the word vector with the word vector, and find the product name that is closest to the vector distance of the word vector and the product name Corresponding accounting item, and add the product name to the word database; and (a4) generate an accounting data according to step (a2) or step (a3), the accounting data includes the product name and the word contained in the voucher text data The accounting item corresponding to the product name; and a calibration machine learning unit, which receives the accounting data from the semantic analysis unit, and checks the accounting data. If the calibration machine learning unit determines what the voucher text data contains If the corresponding relationship between the product name and the accounting item is wrong, correct the corresponding relationship between the product name and the accounting item in the accounting data, and add the product name and the corresponding relationship between the product name and the accounting item to the dictionary. For example, in the intelligent accounting system described in item 1 of the scope of patent application, the step of preprocessing the voucher image data includes: performing image noise reduction on the voucher image data. For the intelligent accounting system described in item 1 of the scope of patent application, the step of preprocessing the voucher image data includes: performing gray-scale binarization processing and correction on the voucher image data. For example, the intelligent accounting system described in item 1 of the scope of patent application further includes: a voucher feature pattern classification library, storing multiple voucher styles; a voucher classification unit connected to the voucher feature pattern classification library, the voucher classification unit The voucher image data is classified according to the voucher characteristic style classification library. For example, in the intelligent accounting system described in item 4 of the scope of patent application, the voucher classification unit will also perform the following steps: (b1) Perform feature point detection on the voucher image data by the image processing unit to find multiple feature points; (b2) Divide the voucher image data into multiple columns by the detected feature points And (b3) label each field of the voucher image data; wherein the text recognition unit corresponds to the voucher text data and the label of each field to further subdivide the voucher text data. A method for identifying an accounting voucher includes the following steps: (a) reading a voucher image data, which corresponds to an accounting voucher; (b) preprocessing the voucher image data to improve the voucher image data And measure the appearance of the voucher image data to determine the size of the accounting voucher; (c) convert the voucher image data into a voucher text data, the voucher text data includes a product name and a transaction object data ; (D) Search for the product name in the voucher text data in a vocabulary, and determine whether there is the product name contained in the voucher text data in the vocabulary. If so, proceed to the following step (e), if If it does not exist, proceed to the following steps (f), where the thesaurus records the correspondence between various product names and accounting items; (e) find out the accounting items corresponding to the product name; (f) find the most relevant item in the word vector corpus A word vector that is close, compare the word vector with the word vector, and find the product name that is closest to the vector distance of the word vector and the accounting item corresponding to the product name, wherein the word vector corpus includes multiple Word vectors, these word vectors correspond to various product names in the vocabulary; and (g) generate an accounting data according to step (d) or step (e), the accounting data includes the product name contained in the voucher text data The accounting item corresponding to the product name; among them, the following steps are included after step (f): Check the accounting data. If the corresponding relationship between the first product name and the accounting item is wrong, correct the corresponding relationship between the product name and the accounting item, and add the product name and the corresponding relationship between the product name and the accounting item to the word database. For example, the method for identifying the accounting voucher described in item 6 of the scope of patent application, wherein the step of preprocessing the voucher image data includes: performing image noise reduction on the voucher image data; or performing gray-scale binary data on the voucher image data Chemical processing and correction. For example, the method for identifying the accounting voucher described in item 6 of the scope of the patent application further includes the following step between step (b) and step (c): classifying the voucher image data by a voucher feature pattern classification database, and The voucher characteristic style classification library stores a variety of voucher styles. As described in item 8 of the scope of patent application, the method for identifying the accounting voucher includes the following steps after classifying the voucher image data: performing feature point detection to find multiple feature points; The obtained characteristic points divide the voucher image data into multiple fields; label each field of the voucher image data; and correspond the voucher text data with the labels of each field to the voucher text data Further breakdown. For the method for identifying the accounting voucher described in item 6 of the scope of the patent application, the voucher classification unit uses a convolutional neural network to classify the voucher image data. For example, in the method for identifying the accounting voucher described in item 6 of the scope of patent application, in step (c), the voucher image data is up-converted into the voucher text data by using OCR text recognition technology.

Images