Document information processing method and system

Abstract

The invention provides a receipt information processing method and system, and the method comprises the following steps: collecting receipt information based on machine vision, and carrying out the standardization processing of the receipt information, and obtaining the standard receipt information; classifying the standard information to obtain document categories; based on a deep learning algorithm and the document category, the document standard information is subjected to classified auditing, and an auditing result is obtained; and based on an auditing result, classifying and storing the approved document standard information, and establishing a document information processing file. According to the technical scheme, advanced technologies such as machine vision and deep learning are introduced, automatic and intelligent processing of receipt information is achieved, processing efficiency and data accuracy are remarkably improved, meanwhile, storage and management modes are optimized, powerful decision support is provided for enterprises, and expandability and customer satisfaction of the system are enhanced.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent information processing, and in particular relates to a document information processing method and system. Background Art

[0002] With the continuous expansion and complexity of corporate business, document processing has become an indispensable part of daily operations. Traditional manual processing methods have problems such as low efficiency and prone to errors, which cannot meet the needs of modern enterprise management. Therefore, an automated and intelligent document information processing method is needed to improve processing efficiency and accuracy.

[0003] The rapid development of machine vision and deep learning technologies has provided new solutions for document information processing. Machine vision technology can automatically capture document images and extract key information from them; deep learning technology can automatically review and classify document information through training and learning of large amounts of data. The application of these technologies has greatly improved the efficiency and accuracy of document information processing.

[0004] However, the prior art still lacks a complete document information processing method and system to achieve efficient collection and processing of document information. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides a document information processing method and system to improve the efficiency of document information processing and realize automatic and intelligent processing of document information.

[0006] To achieve the above object, the present invention provides the following solutions: A document information processing method comprises the following steps: Based on machine vision, document information is collected and standardized to obtain standard document information; Classifying the standard information to obtain a document category; Based on the deep learning algorithm and the document category, the document standard information is reviewed by category to obtain the review result; Based on the audit results, the audited document standard information is classified and stored, and a document information processing file is established.

[0007] Preferably, the method for collecting document information based on machine vision is: Collecting an initial document image, and binarizing the initial document image using an Otsu algorithm to obtain a binary document image; Using a Radon operator, performing tilt correction on the binary document image to obtain a corrected document image; Using multi-edge operator detection, the document and background areas in the corrected document image are segmented to obtain a document information image; The OCR technology is used to identify the information in the document information image to complete the collection of the document information.

[0008] Preferably, the document information includes the title, date, number, content description and signature or seal of the document.

[0009] Preferably, the method for classifying the standard information to obtain the document category is: Segmenting the document information, extracting the BERT vector representation of the document information after segmentation, and obtaining initial information features; Using the BiLSTM model to learn the context representation of the initial information features to obtain information text features; Based on the initial information features and the information text features, construct a text graph structure; Based on the text graph structure, obtaining a feature representation of the document information; The feature representation is input into the fully connected layer, and the feature category probability value is obtained through the Softmax function; Based on the feature category probability value, the document information is classified to obtain the document category.

[0010] Preferably, the document categories include financial documents, commercial documents, funding documents, shipping documents, insurance documents, official documents, ancillary documents and other documents.

[0011] Preferably, the document standard information is reviewed by category, including reviewing whether the document standard information meets the business requirements and the authenticity of the signature; wherein the method for reviewing whether the document standard information meets the business requirements is: Based on the document category, collect the corresponding business rules and regulatory requirements; Use parsers to convert business rules and regulatory requirements of various documents into internal data structures; Matching and evaluating the document standard information with the internal data structure to obtain triggered rules; Execute the action part of the triggered rule, generate a document review exception log, and complete the review of whether the document standard information meets business requirements.

[0012] Preferably, the method for verifying the authenticity of the signature is: Based on the document source, obtain the real signature image; De-noising the binarized document image using a Gaussian filter to obtain a de-noised image; Locate the signature part in the denoised image and extract the signature part to obtain a document signature image; Extracting the geometric, texture and edge features of the document signature image to obtain a comprehensive feature vector of the document signature; Matching the document signature comprehensive feature vector with the real signature image to calculate the similarity; Based on the similarity, the authenticity of the document signature is evaluated.

[0013] The present invention also provides a document information processing system for implementing the document information processing method, comprising: a document information collection module, a document classification module, a document information review module and a storage module; The document information collection module is used to collect document information based on machine vision, and perform standard processing on the document information to obtain document standard information; The document classification module is used to classify the standard information to obtain the document category; The document information review module is used to classify and review the document standard information based on the deep learning algorithm and the document category to obtain the review result; The storage module is used to classify and store the document standard information that has passed the review based on the review result, and to establish a document information processing file.

[0014] Preferably, the document information collection module includes: An initial document image acquisition unit, used to acquire an initial document image, and perform binarization processing on the initial document image using an Otsu algorithm to obtain a binarized document image; An image correction unit, used to perform tilt correction on the binary document image using a Radon operator to obtain a corrected document image; A segmentation unit, used to segment the document and the background area in the corrected document image by using multi-edge operator detection to obtain a document information image; An information collection unit, used to recognize information in the document information image using OCR technology to complete the collection of the document information; The information standardization unit is used to standardize the document information to obtain document standard information.

[0015] Preferably, the document classification module includes: An initial information feature extraction unit, used to segment the document information, extract the BERT vector representation of the document information after segmentation, and obtain initial information features; An information text feature extraction unit, used to learn the context representation of the initial information feature using a BiLSTM model to obtain information text features; A text graph structure construction unit, used to construct a text graph structure based on the initial information features and the information text features; A feature representation unit, used for obtaining a feature representation of the document information based on the text graph structure; The classification unit is used to input the feature representation into the fully connected layer, obtain the feature category probability value through the Softmax function; based on the feature category probability value, classify the document information to obtain the document category.

[0016] Compared with the prior art, the present invention has the following beneficial effects: Improve processing efficiency: Automatically collect document information through machine vision, reducing the time and error rate of manual input, and significantly improving the efficiency of document processing. Standardized processing ensures the uniformity and comparability of information, laying a good foundation for subsequent steps.

[0017] Enhanced data accuracy: The application of machine vision and deep learning algorithms can accurately identify and extract key information from documents, reducing the possibility of human error. Classification technology helps to group documents according to predetermined rules for subsequent management and analysis.

[0018] Realize intelligent auditing: Deep learning algorithms can automatically audit according to document categories and preset rules, reducing the workload of manual auditing while improving the accuracy and consistency of auditing. Audit results can be fed back instantly, making it easier to find and correct problems in a timely manner.

[0019] Optimize storage and management: Organize the approved document information in an orderly manner through classified storage, which is convenient for subsequent query and retrieval. Establish document information processing archives to achieve long-term preservation and traceability of document information, which is helpful for corporate compliance management and auditing.

[0020] Improve decision support: Accurate and comprehensive document information provides valuable data resources for enterprises, helping management to make more scientific and reasonable decisions. Through data analysis, problems and improvement points in business operations can be discovered, promoting continuous optimization and improvement of enterprises.

[0021] Enhance system scalability: This method adopts modular design, each step is relatively independent, and easy to maintain and upgrade. As the business of the enterprise develops and the needs change, the processing flow can be flexibly adjusted and optimized to meet new needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0023] Figure 1 The present invention is a flowchart of a document information processing method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments. Example

[0026] like Figure 1 As shown, a document information processing method includes the following steps: S1: Based on machine vision, collect document information, standardize the document information, and obtain document standard information; A further implementation method is that, based on machine vision, the method for collecting document information is: Collect the initial document image, use the Otsu algorithm to binarize the initial document image, and obtain the binary document image; use the Otsu algorithm to automatically perform threshold segmentation on the image and convert the image into a binary image. The Otsu algorithm can automatically calculate a threshold value so that the intra-class variance between the foreground and background of the image is minimized.

[0027] Using Radon operator, the binarized document image is tilt-corrected to obtain a corrected document image. Multi-edge operator detection is used to segment the document and the background area in the corrected document image to obtain a document information image. It should be noted that for the collection of paper documents, there may be a background image (blank edges of paper, desktop, etc.) in the initial document image. Therefore, in this embodiment, the Canny edge detector is used to detect the edges of the document, and morphological operations (such as dilation and erosion) are used to enhance and clean the edges, and finally the document is segmented from the background.

[0028] Use OCR technology to identify information in document images and complete the collection of document information.

[0029] A further implementation manner is that the document information includes the title, date, number, content description and signature or seal of the document.

[0030] S2: Classify the standard information to obtain the document category; A further implementation method is to classify the standard information and obtain the document category as follows: The document information is segmented, and the BERT vector representation of the document information after segmentation is extracted to obtain the initial information features; in this embodiment, first, the document information is segmented and split into independent vocabulary units. Then, the segmented text is encoded using the pre-trained BERT (Bidirectional Encoder Representations from Transformers) model. The BERT model can capture rich semantic information and convert the text into a high-dimensional vector representation, that is, each word or phrase is mapped to a vector space of fixed dimension, and these vectors are the initial information features.

[0031] The BiLSTM model is used to learn the contextual representation of the initial information features to obtain the information text features; in this embodiment, the initial information features generated by BERT are used as input and sent to the bidirectional long short-term memory network (BiLSTM). BiLSTM can consider both forward and backward contextual information at the same time, thereby capturing the dependency relationship between each word and other words in the text and generating richer information text features.

[0032] Based on the initial information features and information text features, a text graph structure is constructed; based on the initial information features and information text features, a text graph structure can be further constructed. Here, the nodes in the graph structure can represent words or phrases in the text, and the edges represent the semantic relationship or co-occurrence relationship between words. The construction of the graph structure can be based on the similarity, co-occurrence frequency or other semantic relevance indicators between words.

[0033] Based on the text graph structure, the feature representation of the document information is obtained; based on the text graph structure, the nodes in the graph structure can be converted into low-dimensional dense vectors through graph embedding technology, and these vectors are the feature representation of the document information. Graph embedding can capture the complex relationship between words and generate feature vectors with rich semantic information.

[0034] The feature representation is input into the fully connected layer, and the feature category probability value is obtained through the Softmax function. Specifically, the feature representation of the document information is input into the fully connected layer, and the fully connected layer will perform further nonlinear transformation on the features to extract the most useful information for the classification task.

[0035] Finally, the output of the fully connected layer is converted into a probability distribution through the Softmax function. The probability value of each category represents the possibility that the document belongs to that category.

[0036] Based on the feature category probability value, the document information is classified to obtain the document category. Specifically, according to the feature category probability value output by the Softmax function, the category with the highest probability is selected as the final classification result of the document. In actual deployment, the above operation steps are optimized to reduce computing resource consumption and model training time.

[0037] S3: Based on the deep learning algorithm and document category, classify and review the document standard information to obtain the review results; In a further embodiment, the document categories include financial documents, commercial documents, funding documents, shipping documents, insurance documents, official documents, ancillary documents and other documents.

[0038] Specifically, financial documents include bills of exchange, promissory notes, checks, etc.; Commercial documents include invoices, transport documents, insurance documents, packing documents, etc.; Fund documents include documents used for the receipt and payment of funds; Shipping documents include ocean bills of lading, air waybills, road, rail and inland waterway transport documents, courier and postal receipts, etc.; Insurance documents include insurance policy, advance insurance policy, insurance certificate, etc.; Official documents include customs invoice, certificate of origin, commodity inspection certificate, etc.; The supplementary documents include shipping notice, bill of lading certificate, beneficiary certificate, sample delivery certificate, etc. Other documents include sales orders, purchase orders, delivery notes, goods receipts, sales contracts, purchase contracts, receipts, payment vouchers, employee payroll, etc.

[0039] A further implementation method is to classify and review the standard information of the document, including reviewing whether the standard information of the document meets the business requirements and the authenticity of the signature; wherein the method for reviewing whether the standard information of the document meets the business requirements is: Based on the document type, collect the corresponding business rules and regulatory requirements; Use parsers to convert business rules and regulatory requirements of various documents into internal data structures; Match and evaluate the document standard information with the internal data structure to obtain the triggered rules; Execute the action part of the triggered rule, generate a document review exception log, and complete the review of whether the document standard information meets business requirements.

[0040] This embodiment also includes: handling priorities and conflicts between rules to ensure the correct execution of rules. The rule engine is integrated into the system and fully tested. The test should cover various possible scenarios and boundary conditions to ensure the correctness and stability of the rule engine. At the same time, the performance of the rule engine needs to be evaluated and optimized to meet the needs of the business system. After the rule engine is deployed, it needs to be continuously monitored and maintained. Monitoring can help to promptly discover and solve problems such as rule execution errors and performance bottlenecks. Maintenance includes updating rules, optimizing rule engine performance, and fixing vulnerabilities. The rule engine is iteratively optimized according to business changes and demand changes. This includes adding new rules, modifying existing rules, adjusting rule priorities, etc. By continuously optimizing the rule engine, its accuracy and efficiency can be improved to better meet business needs.

[0041] A further implementation method is that the method for verifying the authenticity of the signature is: Based on the document source, obtain the real signature image; Use Gaussian filter to denoise the binary document image to obtain a denoised image; Locate the signature part in the denoised image and extract it to obtain the document signature image; Extract the geometric, texture and edge features of the document signature image to obtain the comprehensive feature vector of the document signature; Using cosine similarity, the document signature comprehensive feature vector is matched with the real signature image to calculate the similarity; Based on the similarity, the authenticity of the document signature is evaluated. Specifically, when the similarity meets the preset threshold, the signature is determined to be authentic, otherwise it is forged.

[0042] In this embodiment, the detected abnormal situation is processed, such as correcting the error, returning the document or taking other remedial measures. A feedback mechanism is established to feed back the processing results to the relevant departments or personnel in a timely manner for timely follow-up and processing.

[0043] S4: Based on the audit results, the audited document standard information is classified and stored, and a document information processing archive is established. In this embodiment, the processed document information is stored in a safe and reliable database to ensure the long-term accessibility and recoverability of the data. The documents are archived and managed, and an archiving strategy is formulated according to business needs, such as archiving by time, type or importance. Example

[0044] The present invention also provides a document information processing system for implementing a document information processing method, comprising: a document information collection module, a document classification module, a document information review module and a storage module; The document information collection module is used to collect document information based on machine vision, and to standardize the document information to obtain document standard information; The document classification module is used to classify standard information and obtain document categories; The document information review module is used to classify and review the standard document information based on the deep learning algorithm and document category to obtain the review results; The storage module is used to classify and store the audited document standard information based on the audit results, and to establish a document information processing file.

[0045] In a further embodiment, the document information collection module includes: An initial document image acquisition unit is used to acquire an initial document image, and perform binarization processing on the initial document image using an Otsu algorithm to obtain a binarized document image; An image correction unit, used to perform tilt correction on the binary document image using a Radon operator to obtain a corrected document image; A segmentation unit, used for segmenting the document and the background area in the corrected document image by using multi-edge operator detection to obtain a document information image; An information collection unit is used to recognize information in the document information image using OCR technology to complete the collection of document information; The information standardization unit is used to standardize document information and obtain document standard information.

[0046] In a further embodiment, the document classification module comprises: An initial information feature extraction unit is used to segment the document information, extract the BERT vector representation of the segmented document information, and obtain the initial information features; An information text feature extraction unit, used to learn the context representation of the initial information feature using the BiLSTM model to obtain the information text feature; A text graph structure building unit, used to build a text graph structure based on initial information features and information text features; A feature representation unit, used to obtain feature representation of document information based on the text graph structure; The classification unit is used to input the feature representation into the fully connected layer and obtain the feature category probability value through the Softmax function; based on the feature category probability value, the document information is classified to obtain the document category.

[0047] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A document information processing method, characterized in that: The following steps are involved: Based on machine vision, document information is collected and standardized to obtain standard document information; Classifying the standard information to obtain a document category; Based on the deep learning algorithm and the document category, classify and review the document standard information to obtain the review result; Based on the audit results, the audited document standard information is classified and stored, and a document information processing file is established.

2. The document information processing method according to claim 1, characterized in that: Based on machine vision, the method of collecting document information is: Collecting an initial document image, and binarizing the initial document image using an Otsu algorithm to obtain a binary document image; Using a Radon operator, performing tilt correction on the binary document image to obtain a corrected document image; Using multi-edge operator detection, the document and background areas in the corrected document image are segmented to obtain a document information image; The OCR technology is used to identify the information in the document information image to complete the collection of the document information.

3. The document information processing method according to claim 1, characterized in that: The document information includes the document title, date, number, content description and signature or seal.

4. The document information processing method according to claim 1, characterized in that: The method of classifying the standard information and obtaining the document category is as follows: Segmenting the document information, extracting the BERT vector representation of the document information after segmentation, and obtaining initial information features; Using the BiLSTM model to learn the context representation of the initial information features to obtain information text features; Based on the initial information features and the information text features, construct a text graph structure; Based on the text graph structure, obtaining a feature representation of the document information; The feature representation is input into the fully connected layer, and the feature category probability value is obtained through the Softmax function; Based on the feature category probability value, the document information is classified to obtain the document category.

5. The document information processing method according to claim 1, characterized in that: The document categories include financial documents, commercial documents, funding documents, shipping documents, insurance documents, official documents, ancillary documents and other documents.

6. The document information processing method according to claim 2, characterized in that: The standard information of the document is reviewed by category, including reviewing whether the standard information of the document meets the business requirements and the authenticity of the signature; wherein the method for reviewing whether the standard information of the document meets the business requirements is: Based on the document category, collect the corresponding business rules and regulatory requirements; Use parsers to convert business rules and regulatory requirements of various documents into internal data structures; Matching and evaluating the document standard information with the internal data structure to obtain triggered rules; Execute the action part of the triggered rule, generate a document review exception log, and complete the review of whether the document standard information meets business requirements.

7. The document information processing method according to claim 6, characterized in that: The method for verifying the authenticity of the signature is: Based on the document source, obtain the real signature image; De-noising the binarized document image using a Gaussian filter to obtain a de-noised image; Locate the signature part in the denoised image and extract the signature part to obtain a document signature image; Extracting the geometric, texture and edge features of the document signature image to obtain a comprehensive feature vector of the document signature; Matching the document signature comprehensive feature vector with the real signature image to calculate the similarity; Based on the similarity, the authenticity of the document signature is evaluated.

8. A document information processing system, used to implement the document information processing method according to any one of claims 1 to 7, characterized in that: include: Document information collection module, document classification module, document information review module and storage module; The document information collection module is used to collect document information based on machine vision, and perform standard processing on the document information to obtain document standard information; The document classification module is used to classify the standard information to obtain the document category; The document information review module is used to classify and review the document standard information based on the deep learning algorithm and the document category to obtain the review result; The storage module is used to classify and store the document standard information that has passed the review based on the review result, and to establish a document information processing file.

9. The document information processing system according to claim 8, characterized in that: The document information collection module includes: An initial document image acquisition unit, used to acquire an initial document image, and perform binarization processing on the initial document image using an Otsu algorithm to obtain a binarized document image; An image correction unit, used to perform tilt correction on the binary document image using a Radon operator to obtain a corrected document image; A segmentation unit, used to segment the document and the background area in the corrected document image by using multi-edge operator detection to obtain a document information image; An information collection unit, used to recognize information in the document information image using OCR technology to complete the collection of the document information; The information standardization unit is used to standardize the document information to obtain document standard information.

10. The document information processing system according to claim 8, characterized in that: The document classification module includes: An initial information feature extraction unit, used to segment the document information, extract the BERT vector representation of the document information after segmentation, and obtain initial information features; An information text feature extraction unit, used to learn the context representation of the initial information feature using a BiLSTM model to obtain information text features; A text graph structure construction unit, used to construct a text graph structure based on the initial information features and the information text features; A feature representation unit, used for obtaining a feature representation of the document information based on the text graph structure; The classification unit is used to input the feature representation into the fully connected layer, obtain the feature category probability value through the Softmax function; based on the feature category probability value, classify the document information to obtain the document category.

Images