Document analysis method and device based on computing power resources of intelligent computing center

Abstract

The invention relates to the technical field of computing power infrastructure, and provides a document analysis method and device based on computing power resources of an intelligent computing center, and the method comprises the following steps: S1, obtaining a to-be-processed document in a first format, and converting the document in the first format into a document in a second format, S2, inputting the document in the second format into a layout recognition model, obtaining a layout recognition result of the document, and inputting the layout recognition result into a content recognition model to obtain content information of the document; and S3, combining the layout recognition result and the content information to obtain an analysis result of the document. Therefore, the layout recognition model and the content recognition model which are operated based on the computing power resources of the intelligent computing center are used for analyzing the document, the document analysis speed can be remarkably increased, and the text semantic understanding accuracy and the text information extraction accuracy are greatly improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of computing infrastructure, and in particular, to a document parsing method and device based on computing resources of an intelligent computing center. Background Art

[0002] There are some obvious limitations and challenges in the current PDF (Portable Document Format) document parsing technology. First, traditional document parsing methods have weak parsing capabilities for scanned PDFs, because scanned PDFs usually store document content in the form of images, and traditional text extraction methods cannot directly read and parse the text information therein. Secondly, for non-scanned PDF documents, traditional methods usually rely on fixed-length splitting or regular expressions for content parsing, which cannot effectively understand the semantic structure and contextual associations of the text, which may affect the subsequent semantic understanding and information extraction based on large models, and lack flexibility in processing text. In addition, there is currently a lack of a document parsing solution specifically for paragraph structures, which cannot fully utilize the semantic connections and logical relationships between paragraphs to improve the parsing effect.

[0003] In summary, traditional PDF document parsing methods have low parsing precision and poor parsing accuracy. Summary of the invention

[0004] The embodiments of the present application provide a document parsing method and device based on the computing power resources of an intelligent computing center to solve the technical problems of low parsing precision and poor parsing accuracy of traditional PDF document parsing methods.

[0005] In order to solve the above technical problems, this application is implemented as follows:

[0006] In a first aspect, an embodiment of the present application provides a document parsing method based on computing resources of an intelligent computing center, the method comprising:

[0007] Step S1: obtaining a document in a first format to be processed, and converting the document in the first format into a document in a second format, wherein the first format is a portable document format PDF, and the second format is a picture;

[0008] Step S2: inputting the document in the second format into a layout recognition model to obtain a layout recognition result of the document, and inputting the layout recognition result into a content recognition model to obtain content information of the document; the computing power resource includes a GPU, and the layout recognition model and the content recognition model run in the GPU;

[0009] Step S3: combining the layout recognition result and the content information to obtain the parsing result of the document.

[0010] Optionally, the computing power resource is a distributed computing power resource, and the distributed computing power resource includes: multiple GPUs located on the same node, each GPU runs a layout recognition model and a content recognition model, and the step S2 includes:

[0011] Step S21: splitting the document in the second format into multiple sub-document groups according to the document page numbers;

[0012] Step S22: inputting each of the plurality of sub-document groups into the layout recognition models on different GPUs to obtain the layout recognition results corresponding to each sub-document group, and inputting the layout recognition results corresponding to each sub-document group into the content recognition models on different GPUs to obtain the content information corresponding to each sub-document group;

[0013] Step S23: combining the layout recognition results corresponding to each of the sub-document groups to obtain the layout recognition result of the document, and combining the content information corresponding to each of the sub-document groups to obtain the content information of the document.

[0014] Optionally, the content recognition model includes: a first content recognition model and a second content recognition model, and step S2 further includes:

[0015] Step S24: Determine whether the computing power resources of the intelligent computing center are greater than a preset threshold, if yes, execute step S25; if no, execute step S26;

[0016] Step S25: inputting the document in the second format into the layout recognition model to obtain a layout recognition result of the document, and inputting the layout recognition result into the first content recognition model to obtain complete content information of the document, wherein the complete content information includes at least one of the following: text information, table information, and picture information of the document;

[0017] Step S26: Input the document in the second format into the layout recognition model to obtain the layout recognition result of the document, and input the layout recognition result into the second content recognition model to obtain the basic content information of the document, wherein the basic content information includes: text information of the document.

[0018] Optionally, the layout recognition result includes the position of each element frame in the document, the size of each element frame and the type of each element frame, and the type includes at least one of the following: title frame, paragraph frame, header frame, footer frame.

[0019] Optionally, step S2 includes:

[0020] Step S27: inputting the document in the second format into the layout recognition model to obtain a layout recognition result of the document;

[0021] Step S28: adjusting the size of each element frame to a preset size to obtain an adjusted layout recognition result;

[0022] Step S29: input the adjusted layout recognition result into the content recognition model to obtain the content information of the document.

[0023] Optionally, step S3 includes:

[0024] Step S31: sorting each element frame according to the position of each element frame;

[0025] Step S32: adjusting the format of the content information corresponding to each element frame to obtain adjusted content information;

[0026] Step S33: Combine each element frame with the corresponding adjusted content information to obtain the parsing result of the document.

[0027] Optionally, the content recognition model is an optical character recognition (OCR) content recognition model.

[0028] In a second aspect, an embodiment of the present application provides a document parsing device based on computing resources of an intelligent computing center, the device comprising:

[0029] An acquisition module, used for acquiring a document in a first format to be processed, and converting the document in the first format into a document in a second format, wherein the first format is PDF and the second format is a picture;

[0030] An execution module is used to input the document in the second format into a layout recognition model to obtain a layout recognition result of the document, and input the layout recognition result into a content recognition model to obtain content information of the document; the computing power resources include a GPU, and the layout recognition model and the content recognition model run in the GPU; the layout recognition result and the content information are combined to obtain a parsed result of the document.

[0031] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the steps of a document parsing method based on computing power resources of an intelligent computing center as described in the first aspect are implemented.

[0032] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of a document parsing method based on computing power resources of an intelligent computing center as described in the first aspect are implemented.

[0033] In a fifth aspect, an embodiment of the present application provides a computer program product, comprising computer instructions, which, when executed by a processor, implement the steps of a document parsing method based on computing power resources of an intelligent computing center as described in the first aspect.

[0034] In the embodiment of the present application, the document is parsed using a layout recognition model and a content recognition model that are run based on the computing resources of an intelligent computing center. The computing resources of the intelligent computing center can provide powerful computing power, so that in document parsing, complex layout and content recognition tasks can be completed quickly and efficiently, which can significantly improve the speed of document parsing; and layout recognition is performed first, then content recognition, and finally the layout recognition result is combined with the content recognition information to reconstruct the semantics and structure of the document, ensuring that the final parsing result can not only display the accurate extracted text information, but also maintain the logical and semantic structural integrity of the original document, which can greatly improve the accuracy of text semantic understanding and the accuracy of text information extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0036] Figure 1 A flowchart of a document parsing method based on computing resources of an intelligent computing center provided in an embodiment of the present application;

[0037] Figure 2 A flowchart of a document parsing method based on computing resources of an intelligent computing center provided in an embodiment of the present application;

[0038] Figure 3 A flowchart of a document parsing method based on computing resources of an intelligent computing center is provided for an embodiment of the present application;

[0039] Figure 4 A structural block diagram of a document parsing device based on computing resources of an intelligent computing center provided in an embodiment of the present application;

[0040] Figure 5 A schematic diagram of the structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0042] The following is a brief description of the technical terms involved in this application.

[0043] The "computing power" mentioned in this application is the ability of computer equipment or computing / data center to process information. It is the ability of computer hardware and software to work together to perform certain computing needs. It is the computing power to achieve target result output by processing information data. It is a new type of productivity that integrates information computing power, network carrying capacity, and data storage capacity. It mainly provides services to society through computing power infrastructure.

[0044] The "computing power (CP)" described in this application is the ability of a data center server to process data and output results. It is a comprehensive indicator to measure the computing power of a data center, including general computing power, supercomputing power and intelligent computing power. The commonly used unit of measurement is the number of floating point operations per second (Floating Point Operations Per Second, FLOPS, where 1EFLOPS=10^18FLOPS). The larger the value, the stronger the comprehensive computing power. According to calculations, 1EFLOPS is approximately the computing power output of 5 Tianhe-2A or 500,000 mainstream server CPUs (Central Processing Unit) or 2 million mainstream notebooks. The calculation formula is: CP=CP 通用 +CP 智能 +CP 超级 In the embodiment of the present application, the computing power adopts the half-precision floating point computing power (FP16) of the GPU (Graphics Processing Unit).

[0045] The "carrying capacity" (Network Power, NP) described in this application is a manifestation of the data transmission capability of computing facilities, including comprehensive capabilities such as network architecture, network bandwidth, transmission latency, intelligent management and scheduling. Carrying capacity involves network transmission within and between data centers, and is a comprehensive indicator for measuring network transmission scheduling capabilities. In the embodiment of this application, carrying capacity uses video memory bandwidth. In the embodiment of this application, storage capacity uses video memory bandwidth.

[0046] The "Storage Power" (SP) described in this application refers to the comprehensive capabilities of a data center in terms of data storage capacity, performance, safety and reliability, and green and low-carbon. It is a comprehensive indicator for measuring the data storage capacity of a data center, including external storage devices such as storage arrays and server built-in storage devices. The commonly used unit of measurement for storage capacity is exabyte (EB, 1EB = 2^60bytes), and the commonly used unit of measurement for performance is the number of reads and writes per second per unit capacity (IOPS / TB, Input / Output Operations Per Second / TB). The disaster recovery ratio is an important manifestation of safety and reliability. In the embodiment of this application, storage power uses video memory capacity.

[0047] The "computing power infrastructure" described in this application is a new type of information infrastructure that integrates information computing power, network carrying capacity, and data storage capacity. It can realize centralized computing, storage, transmission and application of information, and presents characteristics such as diversity and ubiquity, intelligence and agility, security and reliability, green and low-carbon. It is of great significance to promote industrial transformation and upgrading, enable my country's scientific and technological innovation, meet people's better life and achieve high-efficiency social governance.

[0048] The “computing power” mentioned in this application includes general computing power CP 通用 、Intelligent computing power CP 智能 And super computing power CP 超级 .

[0049] The "general computing power" mentioned in this application refers to the computing power provided by servers based on CPU chips, which is used to support basic general computing such as cloud computing and edge computing.

[0050] The "intelligent computing power" mentioned in this application refers to the large-scale deployment of intelligent computing centers based on special chips such as GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), ASIC (Application Specific Integrated Circuit) for various innovative artificial intelligence applications, such as natural language processing, machine vision, etc.

[0051] The "supercomputing power" mentioned in this application is mainly the computing power provided by high-performance computing clusters such as supercomputers. It utilizes the centralized computing resources of multiple computer systems working in parallel and uses a dedicated operating system to handle extremely complex or data-intensive problems. It is mainly used for calculations in cutting-edge scientific fields, such as planetary simulation, drug molecule design, and gene analysis.

[0052] The "intelligent computing center" mentioned in this application refers to a facility that provides the required computing power, data and algorithms for artificial intelligence applications (such as artificial intelligence deep learning model development, model training and model reasoning) by using large-scale heterogeneous computing resources, including general computing power (CPU) and intelligent computing power (GPU, FPGA, ASIC, etc.). The intelligent computing center covers facilities, hardware, and software, and can provide full-stack capabilities from bottom-level computing power to top-level application enablement.

[0053] The "computing resources" mentioned in this application refer to technologies and facilities with information computing, transmission, storage and application capabilities required for the development of a digital society, including but not limited to computing resources such as CPUs and GPUs, network resources such as switches and routers, storage resources such as storage arrays and distributed storage, security resources such as firewalls and intrusion detection systems, and supporting and guarantee resources such as wind, fire, water and electricity.

[0054] Please refer to Figure 1 , Figure 1 A flowchart of a document parsing method based on computing resources of an intelligent computing center according to an embodiment of the present application is shown. Figure 1 As shown, the method includes:

[0055] Step S1: obtaining a document in a first format to be processed, and converting the document in the first format into a document in a second format;

[0056] Among them, the first format is PDF and the second format is image;

[0057] Step S2: inputting the document in the second format into the layout recognition model to obtain the layout recognition result of the document, and inputting the layout recognition result into the content recognition model to obtain the content information of the document;

[0058] Computing resources include GPUs, where layout recognition models and content recognition models run;

[0059] Step S3: Combine the layout recognition result and the content information to obtain the document analysis result.

[0060] In step S1, first obtain a first format document to be processed, that is, a PDF document. Subsequently, the PDF document is converted into a second format document, usually a picture format (such as JPEG, PNG, etc.) through a conversion tool. The purpose of the conversion is to use image processing technology to perform subsequent layout recognition and content recognition. PDF documents usually contain text, images, and complex typesetting structures. Converting them into pictures can provide clear and processable input for subsequent deep learning models, avoiding the parsing difficulties that may be caused by the PDF format.

[0061] In step S2, the converted image document is input into a layout recognition model, which is pre-trained based on rich training data. The image information of the document can be analyzed by a deep learning algorithm to identify the structural layout of the document, such as elements such as titles, paragraphs, pictures and tables. The result of layout recognition can show the position and type of each layout element of the document. In a possible implementation, the layout recognition result includes the position of each element box in the document, the size of each element box and the type of each element box, and the type includes at least one of the following: title box, paragraph box, header box, footer box. Next, the recognized layout result is passed to the content recognition model, which can be an OCR (Optical Character Recognition) content recognition model. The content recognition model further analyzes the specific content of the document and extracts text information, table content, and image-related data.

[0062] It should also be noted that the computing resources (such as GPU) of the intelligent computing center are used to run layout recognition models and content recognition models to achieve efficient and fast calculations and ensure the feasibility of large-scale document processing.

[0063] Now we will briefly explain the training of the layout recognition model. Relying on the powerful computing power of the intelligent computing center, we can perform fine-grained labeling and data optimization on open source data sets and enterprise proprietary data sets through manual annotation. Based on this, we can build a detailed fine-grained layout recognition data set. Using this annotated data set, we can train multiple high-performance layout recognition models and test the recognition effect of the layout recognition models to achieve accuracy and robustness.

[0064] Through the trained layout recognition model, it is possible to accurately extract multiple content elements such as paragraphs, titles, pictures, tables, etc. from the document, thereby achieving efficient document processing and accurate information extraction. This method fully utilizes the computing power of the intelligent computing center and provides reliable technical support for increasingly complex document processing needs. It not only improves the accuracy and efficiency of document parsing, but also provides powerful flexibility to handle various types and complex documents.

[0065] In a possible implementation, the computing power resources are distributed computing power resources, and the distributed computing power resources include: multiple GPUs located on the same node, each GPU running a layout recognition model and a content recognition model, such as Figure 2 As shown, step S2 includes:

[0066] Step S21: splitting the document in the second format into a plurality of sub-document groups according to the document page numbers;

[0067] Step S22: inputting each of the multiple sub-document groups into the layout recognition models on different GPUs to obtain the layout recognition results corresponding to each sub-document group, and inputting the layout recognition results corresponding to each sub-document group into the content recognition models on different GPUs to obtain the content information corresponding to each sub-document group;

[0068] Step S23: combining the layout recognition results corresponding to each sub-document group to obtain the layout recognition result of the document, and combining the content information corresponding to each sub-document group to obtain the content information of the document.

[0069] It should be noted that multiple sub-document groups will be input into the layout recognition model running on each GPU respectively. Distributed computing resources mean that multiple GPUs can process multiple sub-document groups in parallel. The layout recognition model on each GPU will analyze its corresponding sub-document group and output the layout recognition results of the group. These results describe the position and type of each element in each sub-document group. Afterwards, these layout recognition results will be input into the content recognition model on the corresponding GPU for further text and content extraction. The content recognition model will analyze the text, images, tables and other information in each sub-document group, and output the specific content information of each sub-document group. Finally, the layout recognition results and specific content information of the sub-document groups from all GPUs are combined to form a complete document recognition result.

[0070] It should be noted that each sub-document group can also be processed separately by different computing nodes. Different computing nodes can run the layout recognition model and the content recognition model respectively. This parallel processing uses multiple nodes to work simultaneously, which can improve processing efficiency. The results after processing by each node will be summarized and combined into a complete layout recognition result and content information to obtain a complete document recognition result.

[0071] In summary, by splitting the document into multiple sub-document groups according to page numbers and using distributed computing resources to concurrently execute layout and content recognition, this implementation method not only improves the efficiency of document parsing, but also enhances the intelligent computing center's processing capabilities for large documents. It can effectively meet the needs of large-scale document parsing, shorten processing time, and improve the accuracy of document parsing.

[0072] In one possible implementation, the content recognition model includes: a first content recognition model and a second content recognition model, and step S2 also includes: step S24: determine whether the computing power resources of the intelligent computing center are greater than a preset threshold, if so, execute step S25; if not, execute step S26; step S25: input the document in the second format into the layout recognition model to obtain the layout recognition result of the document, and input the layout recognition result into the first content recognition model to obtain the complete content information of the document, wherein the complete content information includes at least one of the following: text information, table information and picture information of the document; step S26: input the document in the second format into the layout recognition model to obtain the layout recognition result of the document, and input the layout recognition result into the second content recognition model to obtain the basic content information of the document, wherein the basic content information includes: text information of the document.

[0073] In other words, the granularity and complexity of document parsing can be adjusted according to the computing power resources. For example, when computing power resources are limited, priority is given to key information (such as titles and paragraph information), and more complex tasks (such as image recognition and table recognition) are processed when computing power resources are sufficient. In this way, the appropriate processing method can be flexibly selected according to the current computing power resources. The dynamic adaptive adjustment capability can maintain a certain performance for the intelligent computing center when resources are limited, and provide effective information extraction capabilities. At the same time, it can achieve a more comprehensive analysis effect when resources are sufficient, ensuring that users can obtain as much and comprehensive information as possible in the document.

[0074] In a possible implementation, step S2 includes:

[0075] Step S27: inputting the document in the second format into the layout recognition model to obtain a layout recognition result of the document;

[0076] Step S28: adjusting the size of each element frame to a preset size to obtain an adjusted layout recognition result;

[0077] Step S29: input the adjusted layout recognition result into the content recognition model to obtain the content information of the document.

[0078] That is to say, in order to ensure that the content can be filled into each element box accordingly, the size of the element box can be adaptively adjusted to avoid missing content, and the size of the element box can also be standardized, which is conducive to subsequent information extraction and layout of the final presentation results.

[0079] In a possible implementation, step S3 includes:

[0080] Step S31: sorting each element frame according to the position of each element frame;

[0081] Step S32: adjusting the format of the content information corresponding to each element frame to obtain adjusted content information;

[0082] Step S33: Combine each element frame with the corresponding adjusted content information to obtain the document parsing result.

[0083] Thus, by gradually sorting, formatting and combining elements, the results of document parsing are not only accurate, but also more in line with the user's reading habits in terms of structure. The final output information can be made more readable and practical, making it easier for users to understand and apply the parsing results. Through these steps, not only is information extraction completed, but the presentation quality of the final document is also improved, enhancing the user experience.

[0084] It should be noted that when data extraction is based solely on the layout recognition model, since information is processed in units of "pages", there may be missing information in the data across pages, which in turn affects the complete semantics of the document. In order to solve this problem, a cross-page data merging processing logic can be designed. In this way, not only the completeness and accuracy of document parsing are improved, but also a more consistent semantic understanding can be provided, ensuring that users can obtain global and meaningful information when using the parsing results. At the same time, the introduction of cross-page processing logic has greatly enhanced the ability of the document parsing system to process complex, multi-page documents, expanding its application scenarios and practical value.

[0085] The cross-page data merging processing logic is a process designed to reconstruct and maintain the global semantic structure of the document. This logic ensures that this information is correctly restored and integrated across multiple pages by identifying and tracking cross-page elements, such as paragraphs and tables. During the processing, the algorithm detects the continuity of information between pages, identifies content belonging to the same logical unit, and merges it into a whole. This merging is particularly suitable for tables, because the information in the table usually extends to multiple pages, and the integrity of this information is crucial for data analysis and decision support.

[0086] Now let's explain the logic of cross-page data merging in a specific application scenario. When processing a cross-page PDF document, you can design a logical merging strategy based on the page number sequence after PDF is converted to an image to ensure the integrity and order of text and tables. For text paragraphs, the merging strategy can be based on the punctuation marks of the paragraphs. For example, if the last paragraph of page 1 ends with a period, it is considered a complete paragraph and does not need to be merged with the first paragraph of page 2; if it does not end with a period, it may be an unfinished paragraph, and it can be merged with the first paragraph of page 2 to maintain the continuity of the paragraph.

[0087] When dealing with tables, the situation is relatively complicated. Generally, tables have titles. If a table spans across pages, the table on the second page will lack a title. To solve this problem, you can first remove interfering elements such as headers and footers, and then sort the elements of the current page in reading order. For tables that span across pages, when the table on the first page is the last one on the page, and the table on the second page is the first one on the second page, you can try to merge the two tables.

[0088] In addition, if you encounter some non-standard PDF documents, the last element of the first page and the first element of the second page may be a table, and the table on the second page has no title (non-standard). In this case, if the row and column structures of the two tables are different, the merge fails and the two tables will be retained separately. If the structures are the same, they may be merged, but the probability of this happening is small, and a check mechanism can be introduced. For example, when two tables with the same structure are detected, a data validity check mechanism is introduced. By checking the key data values ​​in the table, such as the title and important fields in the table, it can be ensured that the data will not cause ambiguity or errors after the merge. In this way, the integrity and accuracy of the data can be maintained, while providing users with more reliable document parsing results.

[0089] In summary, relying on the powerful computing power of the intelligent computing center, the intelligent document parsing method provided in the embodiment of the present application can efficiently process and parse various types of PDF documents, ensuring that the parsing process remains accurate and efficient regardless of the complexity of the document. Utilizing the powerful computing power resources of the intelligent computing center, firstly, the various elements such as titles, paragraphs, and tables in the document are accurately identified through the pre-trained layout recognition model. Next, the paragraphs or tables are effectively associated and spliced ​​with their corresponding titles (see Figure 3 ). This processing method not only retains the complete semantic information of the document, but also ensures the logical connection between the title and the content, thus providing users with a detailed and easy-to-understand document parsing result. This process fully utilizes the computing power advantage of the intelligent computing center to achieve efficient processing of complex document structures and accurate information extraction.

[0090] In the embodiment of the present application, the document is parsed using a layout recognition model and a content recognition model that are run based on the computing resources of an intelligent computing center. The computing resources of the intelligent computing center can provide powerful computing power, so that in document parsing, complex layout and content recognition tasks can be completed quickly and efficiently, which can significantly improve the speed of document parsing; and layout recognition is performed first, then content recognition, and finally the layout recognition result is combined with the content recognition information to reconstruct the semantics and structure of the document, ensuring that the final parsing result can not only display the accurate extracted text information, but also maintain the logical and semantic structural integrity of the original document, which can greatly improve the accuracy of text semantic understanding and the accuracy of text information extraction.

[0091] Please refer to Figure 4 , Figure 4 A document parsing device based on computing resources of an intelligent computing center provided according to an embodiment of the present application is shown, and the device 40 includes:

[0092] An acquisition module 41 is used to acquire a document in a first format to be processed, and convert the document in the first format into a document in a second format, wherein the first format is PDF and the second format is a picture;

[0093] The execution module 42 is used to input the document in the second format into the layout recognition model to obtain the layout recognition result of the document, and input the layout recognition result into the content recognition model to obtain the content information of the document; the computing resources include GPU, and the layout recognition model and the content recognition model run in the GPU; the layout recognition result and the content information are combined to obtain the parsing result of the document.

[0094] In a possible implementation, the computing power resources are distributed computing power resources, and the distributed computing power resources include: multiple GPUs located on the same node, each GPU running a layout recognition model and a content recognition model,

[0095] The execution module 42 is further used to split the document in the second format into a plurality of sub-document groups according to the document page numbers;

[0096] Input each of the multiple sub-document groups into the layout recognition models on different GPUs to obtain the layout recognition results corresponding to each sub-document group, and input the layout recognition results corresponding to each sub-document group into the content recognition models on different GPUs to obtain the content information corresponding to each sub-document group;

[0097] The layout recognition results corresponding to each sub-document group are combined to obtain the layout recognition result of the document, and the content information corresponding to each sub-document group is combined to obtain the content information of the document.

[0098] In a possible implementation, the content recognition model includes: a first content recognition model and a second content recognition model, and the execution module 42 is further used to determine whether the computing power resources of the intelligent computing center are greater than a preset threshold, and if so, input the document in the second format into the layout recognition model to obtain the layout recognition result of the document, and input the layout recognition result into the first content recognition model to obtain the complete content information of the document, wherein the complete content information includes at least one of the following: text information, table information, and picture information of the document;

[0099] If not, the document in the second format is input into the layout recognition model to obtain the layout recognition result of the document, and the layout recognition result is input into the second content recognition model to obtain the basic content information of the document, wherein the basic content information includes: text information of the document.

[0100] In a possible implementation, the layout recognition result includes the position of each element frame in the document, the size of each element frame and the type of each element frame, and the type includes at least one of the following: title frame, paragraph frame, header frame, footer frame.

[0101] In a possible implementation, the execution module 42 is further configured to input the document in the second format into the layout recognition model to obtain a layout recognition result of the document;

[0102] Adjust the size of each element frame to a preset size to obtain an adjusted layout recognition result;

[0103] The adjusted layout recognition result is input into the content recognition model to obtain the content information of the document.

[0104] In a possible implementation, the execution module 42 is further configured to sort each element box according to the position of each element box;

[0105] Adjust the format of the content information corresponding to each element frame to obtain adjusted content information;

[0106] Each element frame is combined with the corresponding adjusted content information to obtain the parsing result of the document.

[0107] In a possible implementation, the content recognition model is an OCR content recognition model.

[0108] In the embodiment of the present application, the document is parsed using a layout recognition model and a content recognition model that are run based on the computing resources of an intelligent computing center. The computing resources of the intelligent computing center can provide powerful computing power, so that in document parsing, complex layout and content recognition tasks can be completed quickly and efficiently, which can significantly improve the speed of document parsing; and layout recognition is performed first, then content recognition, and finally the layout recognition result is combined with the content recognition information to reconstruct the semantics and structure of the document, ensuring that the final parsing result can not only display the accurate extracted text information, but also maintain the logical and semantic structural integrity of the original document, which can greatly improve the accuracy of text semantic understanding and the accuracy of text information extraction.

[0109] Please refer to Figure 5 The embodiment of the present application also provides an electronic device 50, including a processor 51, a memory 52, and a computer program stored in the memory 52 and executable on the processor 51. When the computer program is executed by the processor 51, the various processes of the document parsing method embodiment based on the computing power resources of the intelligent computing center are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0110] The embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, each process of the document parsing method embodiment based on the computing power resources of the intelligent computing center is implemented, and the same technical effect can be achieved. To avoid repetition, it is not repeated here. The computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0111] The embodiment of the present application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the various processes of the above-mentioned document parsing method embodiment based on the computing power resources of an intelligent computing center, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0112] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0113] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0114] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A document parsing method based on computing resources of an intelligent computing center, characterized in that: The method comprises: Step S1: obtaining a document in a first format to be processed, and converting the document in the first format into a document in a second format, wherein the first format is a portable document format PDF, and the second format is a picture; Step S2: inputting the document in the second format into a layout recognition model to obtain a layout recognition result of the document, and inputting the layout recognition result into a content recognition model to obtain content information of the document; the computing power resource includes a graphics processing unit GPU, and the layout recognition model and the content recognition model run in the GPU; Step S3: combining the layout recognition result and the content information to obtain the parsing result of the document.

2. The method according to claim 1, characterized in that The computing power resources are distributed computing power resources, and the distributed computing power resources include: multiple GPUs located on the same node, each GPU runs a layout recognition model and a content recognition model, and the step S2 includes: Step S21: splitting the document in the second format into multiple sub-document groups according to the document page numbers; Step S22: inputting each of the plurality of sub-document groups into the layout recognition models on different GPUs to obtain the layout recognition results corresponding to each sub-document group, and inputting the layout recognition results corresponding to each sub-document group into the content recognition models on different GPUs to obtain the content information corresponding to each sub-document group; Step S23: combining the layout recognition results corresponding to each of the sub-document groups to obtain the layout recognition result of the document, and combining the content information corresponding to each of the sub-document groups to obtain the content information of the document.

3. The method according to claim 1, characterized in that The content recognition model includes: a first content recognition model and a second content recognition model, and the step S2 further includes: Step S24: Determine whether the computing power resources of the intelligent computing center are greater than a preset threshold, if yes, execute step S25; if no, execute step S26; Step S25: inputting the document in the second format into the layout recognition model to obtain a layout recognition result of the document, and inputting the layout recognition result into the first content recognition model to obtain complete content information of the document, wherein the complete content information includes at least one of the following: text information, table information, and picture information of the document; Step S26: Input the document in the second format into the layout recognition model to obtain the layout recognition result of the document, and input the layout recognition result into the second content recognition model to obtain the basic content information of the document, wherein the basic content information includes: text information of the document.

4. The method according to claim 1, characterized in that: The layout recognition result includes the position of each element frame in the document, the size of each element frame and the type of each element frame, and the type includes at least one of the following: title frame, paragraph frame, header frame, and footer frame.

5. The method according to claim 4, characterized in that The step S2 comprises: Step S27: inputting the document in the second format into the layout recognition model to obtain a layout recognition result of the document; Step S28: adjusting the size of each element frame to a preset size to obtain an adjusted layout recognition result; Step S29: input the adjusted layout recognition result into the content recognition model to obtain the content information of the document.

6. The method according to claim 4, characterized in that The step S3 comprises: Step S31: sorting each element frame according to the position of each element frame; Step S32: adjusting the format of the content information corresponding to each element frame to obtain adjusted content information; Step S33: Combine each element frame with the corresponding adjusted content information to obtain the parsing result of the document.

7. The method according to any one of claims 1 to 6, characterized in that The content recognition model is an optical character recognition (OCR) content recognition model.

8. A document parsing device based on computing power resources of an intelligent computing center, characterized in that: The device comprises: An acquisition module, used for acquiring a document in a first format to be processed, and converting the document in the first format into a document in a second format, wherein the first format is PDF and the second format is a picture; An execution module is used to input the document in the second format into a layout recognition model to obtain a layout recognition result of the document, and input the layout recognition result into a content recognition model to obtain content information of the document; the computing power resources include a GPU, and the layout recognition model and the content recognition model run in the GPU; the layout recognition result and the content information are combined to obtain a parsed result of the document.

9. An electronic device, characterized in that: include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the steps of a document parsing method based on computing power resources of an intelligent computing center as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the steps of a document parsing method based on computing power resources of an intelligent computing center as described in any one of claims 1 to 7 are implemented.

11. A computer program product, characterized in that It includes computer instructions, which, when executed by a processor, implement the steps of a document parsing method based on computing power resources of an intelligent computing center as described in any one of claims 1 to 7.

Images