PDF image screening method driven by intelligent analysis

Abstract

The invention discloses a PDF (Portable Document Format) image screening method driven by intelligent analysis, which belongs to the technical field of document processing and comprises the following steps of: intelligently analyzing a plurality of PDF files to obtain a plurality of pieces of image information for format conversion; traversing the plurality of PDF files based on the plurality of bitmap formats for image capture; performing image content identification on the multiple pieces of image information according to the multiple pieces of image proportion data; identifying the plurality of pieces of image information based on a plurality of text data sets, and traversing a plurality of PDF files according to a plurality of image indexes; simulation preselection is carried out based on the target PDF file, full-text search is carried out based on multiple preselected images, preselected feedback information is sent to the remote control terminal, and an image screening instruction is generated to carry out intelligent screening response on the PDF images. According to the method and the device, the problems of low PDF file image processing efficiency and inaccurate information extraction in the prior art are solved, so that a user faces tedious manual screening and processing flows in document management and data analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of document processing, and in particular to a PDF image screening method driven by intelligent analysis. Background Art

[0002] In the digital age, PDF files, as a widely used document format, carry a large amount of text and image information. With the surge in the amount of information, the traditional way of manually screening and processing PDF documents can no longer meet the needs of efficiency, security and accuracy. Therefore, PDF processing solutions based on artificial intelligence and automation technologies have emerged to improve the efficiency and accuracy of data processing. These technologies can not only speed up the process of information extraction, but also help users quickly locate key information in massive documents, promoting the digital transformation of various industries.

[0003] Currently, there are many PDF processing tools on the market, but most of them still rely on manual intervention or rule-based parsing methods, and cannot achieve comprehensive intelligent and automated coverage. These tools have limitations in processing complex formats, image quality optimization, and information retrieval, resulting in low efficiency and high error rates. In addition, existing image screening technologies usually lack the support of deep learning and data analysis, and cannot fully utilize the information in the document. Therefore, developing a PDF image screening method based on intelligent parsing to achieve automated and efficient document processing has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] This application provides a PDF image screening method driven by intelligent parsing, aiming to solve the problems of low efficiency in PDF file image processing and inaccurate information extraction in the prior art, which results in users facing cumbersome manual screening and processing processes in document management and data analysis.

[0005] In view of the above problems, the present application provides a PDF image screening method driven by intelligent parsing.

[0006] The present application provides a PDF image screening method driven by intelligent parsing, the method comprising: performing intelligent parsing on multiple PDF files to obtain multiple image information, performing format conversion based on the multiple image information to obtain multiple bitmap formats; traversing multiple PDF files based on the multiple bitmap formats to capture images and determine multiple image ratio data; performing image content recognition on the multiple image information according to the multiple image ratio data to generate multiple text data sets; marking the multiple image information based on the multiple text data sets to generate multiple image indexes, traversing multiple PDF files according to the multiple image indexes to determine a target PDF file; performing simulated pre-selection based on the target PDF file to generate multiple pre-selected images, performing full-text search based on the multiple pre-selected images to obtain pre-selected feedback information, sending the pre-selected feedback information to a remote control terminal, and generating an image screening instruction to perform intelligent screening response on the PDF image.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] Due to the adoption of a PDF image screening method driven by intelligent analysis, this technical solution solves the problems of low image screening efficiency and inaccurate information extraction in the prior art by intelligently parsing multiple PDF files, extracting image information and performing format conversion. By traversing PDF files for image capture and proportional data analysis, the recognition of image content and the generation of text data sets are realized, and then an image index is generated to facilitate the rapid location of the target PDF file. Finally, based on pre-selected feedback information and semantic analysis, intelligent screening instructions are generated to effectively screen PDF images, thereby greatly improving the efficiency and accuracy of image processing, reducing the need for manual intervention, and optimizing the document management and data analysis processes.

[0009] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A flowchart of a PDF image screening method driven by intelligent analysis is provided for an embodiment of the present application.

[0011] Figure 2 A schematic diagram of a process for obtaining multiple image information in a PDF image screening method driven by intelligent analysis is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0012] The overall idea of ​​the technical solution provided by this application is as follows:

[0013] The embodiment of the present application provides a PDF image screening method driven by intelligent analysis. By performing intelligent analysis on multiple PDF files, image information is extracted and format conversion is performed to generate a bitmap format. Then, based on the image ratio data, the image content is identified and a text data set is generated. Finally, an image index is generated to quickly locate the target PDF file, thereby realizing an automated and intelligent image screening process.

[0014] After introducing the basic principles of the present application, various non-limiting implementation methods of the present application will be specifically described below in conjunction with the drawings in the specification.

[0015] Examples, such as Figure 1 As shown, the embodiment of the present application provides a PDF image screening method driven by intelligent analysis, the method comprising:

[0016] Step S100: intelligently parse multiple PDF files to obtain multiple image information, and perform format conversion based on the multiple image information to obtain multiple bitmap formats.

[0017] Specifically, image information refers to image data extracted from PDF files, including visual content such as pictures, charts, and illustrations. Format conversion is the conversion of image information from one format (such as PDF) to another format (such as bitmap) for subsequent processing and analysis. Bitmap format is a format that stores images in pixels, such as BMP, JPEG, PNG, etc., which is suitable for image display and editing.

[0018] First, you need to use a PDF parsing library (such as Apache PDFBox or PyMuPDF) to read the document. This process usually uses streaming reading technology to ensure that large PDF files can be loaded efficiently and their contents can be processed dynamically in memory. The parsing library loads each page of the PDF file one by one and extracts the image information.

[0019] Through this step, important image information can be efficiently extracted from multiple PDF files and converted into bitmap format. This improves the efficiency of data processing, allows for quick acquisition of required visual information, reduces the need for manual intervention, improves the accuracy and usability of information extraction, and provides a reliable basis for subsequent data analysis and decision-making.

[0020] Step S200: traversing multiple PDF files to capture images based on the multiple bitmap formats, and determining multiple image ratio data.

[0021] Specifically, image capture refers to the process of extracting image content from a PDF file. Image ratio data describes the relative size and position relationship of an image at different sizes, which can be used for image comparison and positioning.

[0022] Based on multiple bitmap formats, multiple PDF files are traversed. PDF files are opened one by one to extract the image data. In this process, the implementation of image capture relies on parsing the graphic content in the PDF and identifying the image part that conforms to the bitmap format.

[0023] After extraction, the images are processed to determine their image ratio data. Specifically, the system calculates the width and height of the image, derives the aspect ratio, and also obtains scaling information. With this data, the geometric characteristics of the image can be established. For example, when processing a PDF containing multiple charts, the charts can be captured in bitmap format and the size ratio of each chart can be recorded.

[0024] For example, suppose there is a PDF file of a research report, which contains multiple charts. Use PDFBox to open the file, traverse all pages, and extract the image information of each chart. Each chart is converted to PNG bitmap format, and then the width and height of the chart are calculated, and the image ratio data such as 4:3 or 16:9 is recorded.

[0025] Through this step, various types of image information can be extracted from PDF files efficiently and accurately, and their proportion data can be obtained. This not only improves the efficiency of image capture, but also lays the foundation for subsequent image analysis and processing, allowing end users to easily obtain and use image resources in PDF files.

[0026] Step S300: performing image content recognition on the plurality of image information according to the plurality of image ratio data to generate a plurality of text data sets.

[0027] Specifically, a text dataset is a structured data set generated through the recognition process, which usually includes the extracted text information and its related attributes.

[0028] First, we rely on multiple image ratio data to perform content recognition on each image. Load the extracted images and perform preprocessing to ensure that the image size is consistent with the model requirements. For example, assuming that the image ratio data is 16:9, adjust the image to this ratio and input it into the image content recognition model. Specifically, the steps of the image content recognition model include data preparation, model selection, training, evaluation, and deployment. First, extract images from a variety of PDF files and annotate key information to generate a training dataset. Next, select a suitable deep learning architecture, such as a convolutional neural network (CNN), and design the model's hierarchy. Then, train the model using the annotated dataset and adjust the hyperparameters to optimize performance. During the training process, use cross-validation to evaluate the accuracy of the model and continuously improve it based on feedback. Finally, after sufficient testing, deploy the model to a production environment to ensure that it can process input images in real time, generate text datasets, and perform effective content recognition.

[0029] Next, the image content is analyzed through the image content recognition model to identify elements in the image, such as identifying columns, lines or text content in a chart. This process uses open source tools such as TensorFlow or PyTorch to extract and classify images using pre-trained image content recognition models.

[0030] After the recognition is completed, the results are organized into multiple text data sets, including the extracted text information and its location, category, etc. For example, when processing a chart containing statistical data, the value and corresponding label of each data point can be extracted.

[0031] Through this step, valuable text information can be efficiently extracted from images and a structured data set can be constructed. This not only improves the accuracy of data analysis, but also provides users with convenient data retrieval and utilization methods, making the extraction and processing of information from PDF documents more systematic and efficient.

[0032] Step S400: Identifying the multiple image information based on the multiple text data sets, generating multiple image indexes, traversing multiple PDF files according to the multiple image indexes, and determining a target PDF file.

[0033] Specifically, the image index refers to an index generated based on a text dataset, which can quickly locate the location and association of image information in multiple PDF files. The target PDF file refers to a specific PDF file obtained by traversing the image index, which is usually a file related to the user's query or demand.

[0034] First, the extracted image information is identified using multiple text datasets. Specifically, the text information of each image is matched to generate a corresponding image index. This process uses natural language processing tools such as NLTK or spaCy to process text data in order to extract keywords and features.

[0035] Next, the generated image index is applied to traverse multiple PDF files. In specific implementation, the PDF parsing library is used to quickly retrieve and match the information in the image index. The index information is used to quickly find PDF files related to the image and determine the target PDF file.

[0036] For example, suppose the text dataset contains "sales data report", find the image index matching the keyword, and then locate the document containing the image in multiple PDF files.

[0037] Through this step, relevant documents can be efficiently extracted from numerous PDF files, significantly improving the speed and accuracy of information retrieval. It not only optimizes the data processing process, but also enables users to quickly obtain the required information, enhancing the convenience and flexibility of data utilization.

[0038] Step S500: Perform simulated pre-selection based on the target PDF file to generate multiple pre-selected images, perform full-text search based on the multiple pre-selected images, obtain pre-selection feedback information, send the pre-selection feedback information to the remote control terminal, and generate image filtering instructions to perform intelligent filtering response on the PDF images.

[0039] Specifically, the pre-selected images are images extracted from the target PDF file as candidates for subsequent analysis. Full-text search refers to a comprehensive search of the extracted text to find information related to specific keywords or topics. Pre-selected feedback information refers to feedback on the relevance and applicability of the pre-selected images to user needs based on the results of the full-text search. The remote control terminal refers to a device used to receive pre-selected feedback information and send instructions, which can be a mobile phone, tablet or computer.

[0040] First, the target PDF file is simulated and pre-selected, and multiple pre-selected images are generated by parsing its content. Image processing tools such as OpenCV or PIL (Python Imaging Library) are usually used to extract image data from the file. Then, these pre-selected images are processed to ensure image quality and usability, such as image enhancement and denoising.

[0041] Subsequently, a full-text search is performed based on the generated pre-selected images to retrieve the text content related to these images. To this end, a natural language processing library (such as NLTK or spaCy) is used to segment and tokenize the text and extract keywords for efficient search.

[0042] During the search process, pre-selected feedback information is collected and generated, such as the matching degree between the image and the text, the relevance score, etc. This feedback information will be sent to the remote control terminal, such as the user's mobile phone or computer, using network transmission protocols (such as HTTP or WebSocket) to achieve real-time communication.

[0043] Finally, based on the pre-selected feedback information, image filtering instructions are generated to intelligently respond to the user's filtering needs. For example, if the user's goal is to "find images of customer feedback", images related to customer feedback will be recommended first, and corresponding instructions will be generated to facilitate users to quickly filter the required content.

[0044] For example, to find images related to "product quality complaints". First, pre-selected images related to this topic are extracted from the target PDF file, and text searches are performed on these images to find paragraphs related to "product quality". Based on the search results, pre-selected feedback information is generated to indicate which images best meet the user's needs. Finally, this information is sent to the user's mobile phone, and the user selects the required image through a simple interface, and the corresponding filtering instructions are generated based on the selection.

[0045] This step significantly improves the efficiency of image screening, allowing users to quickly find images related to a specific topic. Through simulated pre-selection and intelligent feedback, it can more accurately meet user needs, reduce the workload of manual screening, and improve work efficiency and user experience.

[0046] Further, such as Figure 2 As shown, multiple PDF files are intelligently parsed to obtain multiple image information, and the method includes: streaming reading based on multiple PDF files, determining multiple file formats, parsing the multiple PDF files according to the multiple file formats, and obtaining multiple file object types; integrating the multiple file formats with the multiple file object types to build an object tree structure; traversing the multiple PDF files according to the object tree structure to perform content stream parsing and extract multiple content data; mapping the multiple file object types with the multiple file formats based on the multiple content data to build a PDF parsing library; traversing multiple PDF files through the PDF parsing library to match, and batch parsing the multiple PDF files according to the matching results to obtain the multiple image information.

[0047] Specifically, streaming reading is a method of reading data that saves memory and improves efficiency by loading data gradually instead of all at once. File object types refer to different types of data structures in PDF files, such as text, images, tables, etc., which represent different contents in the document. The object tree structure is a data structure that organizes the various objects in the document in a hierarchical manner for fast access and processing. Content stream parsing is the process of analyzing the content stored in the file and extracting useful information. PDF parsing library is a tool or framework specifically used to parse and process PDF files and extract content and structure information.

[0048] First, multiple PDF files are read one by one using streaming reading technology (such as Apache PDFBox's Streaming API) to determine the file format. In this process, the parser identifies the file format type and creates corresponding file object types, such as image objects, text objects, etc.

[0049] Next, based on the identified file format and file object type, an object tree structure is constructed to clearly represent the relationship between each object, which is convenient for subsequent content stream parsing. Then, multiple PDF files are traversed according to the object tree structure, content stream parsing is performed, and content data such as text and images are extracted.

[0050] Using content data, different object types are mapped to file formats, and finally a PDF parsing library is built for subsequent quick access. Finally, using this parsing library, multiple PDF files are traversed for matching, and batch parsing is performed through the matching results to obtain multiple image information.

[0051] Through this step, the intelligent parsing method can efficiently and systematically process multiple PDF files, automatically identifying and extracting various types of information. This processing method not only improves the speed and efficiency of file processing, but also reduces the need for manual intervention. Ultimately, users can quickly obtain all the required image information, greatly improving the quality of document analysis and information extraction, and providing a solid foundation for subsequent data analysis and decision-making.

[0052] Furthermore, the PDF parsing library is used to traverse multiple PDF files for matching, and the multiple PDF files are batch parsed according to the matching results to obtain the multiple image information. The method includes: loading multiple PDF files based on the file usage path, using the PDF parsing library to parse and allocate the multiple PDF files, and determining multiple parsing threads; matching multiple PDF files according to the multiple parsing threads, and processing the multiple PDF files in parallel according to the parsing matching results to generate parsing instructions; activating an image processing module through the parsing instructions, and performing image enhancement on multiple PDF files through the image processing module to extract graphic contours and generate a geometric data set; performing deep learning based on the geometric data set to determine the multiple image information.

[0053] Specifically, the file usage path refers to the specific location of the file in the storage system, which usually includes the file name and directory structure. Parsing thread refers to the computing unit that independently executes tasks during parallel processing, which can improve processing efficiency. The geometric data set contains the data set of information such as image shape, edge and contour, which is used for subsequent analysis and processing.

[0054] First, multiple PDF files are loaded based on the file usage path, and the PDF parsing library is used to parse and distribute the files. This process involves creating multiple parsing threads so that multiple PDF files can be processed in parallel, thereby improving efficiency.

[0055] Next, multiple PDF files are matched according to the parsing thread, and the corresponding parsing instructions are generated by analyzing the similarity or other features of the file content. These instructions will then activate the image processing module (such as OpenCV) to enhance the image in each PDF file to extract the shape contour and generate a geometric data set.

[0056] Finally, based on the geometric dataset, deep learning technology is applied to identify and extract the required multiple image information through training models. Specifically, a deep learning model is built using the generated geometric dataset, and a convolutional neural network (CNN) is usually used for image recognition and feature extraction. The input of the model is the enhanced image and its geometric data, which includes contours, edges, and shape features. Next, the model is trained, using the annotated image information as a training set, and the model parameters are adjusted to optimize the recognition accuracy. After training, the model is able to identify specific features and patterns in the image, such as changes in edges or the shape of graphics, and then extract multiple image information. Finally, by reasoning on the newly input image, the recognition result will be output, and detailed information about the image content, such as the image category, size, and its position in the document, will be provided.

[0057] This method can efficiently perform batch parsing and image processing on multiple PDF files, automatically identifying the required image information. This parallel processing method significantly improves processing speed and accuracy, while reducing the need for manual intervention. Ultimately, users can quickly obtain all relevant image information, support data analysis and decision-making in the fields of medicine and scientific research, and provide a solid foundation for subsequent work.

[0058] Furthermore, based on the multiple bitmap formats, multiple PDF files are traversed to perform image capture and multiple image ratio data are determined. The method includes: reading images from multiple PDF files based on the multiple bitmap formats to generate a bitmap image set, capturing positions according to the bitmap image set, and generating image matrix data; binarizing the image matrix data through mean filtering, performing edge detection according to the image processing results, and determining multiple straight line data; performing equidistance analysis according to the multiple straight line data to generate a detection ruler; and performing geometric deduction on multiple PDF files according to the geometric data set using the detection ruler to determine the multiple image ratio data.

[0059] Specifically, a bitmap image set is a collection of multiple extracted images, usually stored in a bitmap format for subsequent processing. Site capture is the process of identifying and recording specific locations or feature points from an image, usually used to analyze image structure. Image matrix data is a numerical representation of an image, usually a two-dimensional array containing the color or grayscale value of each pixel. Mean filtering is an image processing technique that smoothes an image and reduces noise by calculating the average value of a pixel neighborhood. Edge detection is the process of identifying the boundaries of objects in an image. Common algorithms include Canny and Sobel. Detection ruler is a reference tool for determining specific distances or proportions in an image, helping to deduce other dimensions. Geometric derivation is the process of obtaining unknown geometric features by inferring based on known geometric information.

[0060] Read images from multiple PDF files based on multiple bitmap formats, extract images and generate bitmap image sets. Use tools such as OpenCV or Pillow to convert images in PDF to bitmap format. Then, use site capture technology to identify feature points in the bitmap and generate image matrix data.

[0061] Next, a mean filter is applied to binarize the image matrix, which helps remove noise and enhance the clarity of the image. Subsequently, an edge detection algorithm (such as Canny edge detection) is used to identify straight line data in the image, which provides the basis for subsequent analysis.

[0062] Using these straight line data, we conduct equidistance analysis, generate detection scales, and determine the reference scale in the image. Finally, we combine the geometric data set and use the detection scale to perform geometric derivation to determine the scale data of multiple images.

[0063] Through this process, image data can be extracted and analyzed from PDF files efficiently and accurately, and the scale information of each image can be determined. This not only improves the efficiency of image processing, but also ensures the accuracy of data, laying a solid foundation for subsequent image recognition and data analysis, allowing users to better utilize the graphic resources in PDF documents.

[0064] Furthermore, the multiple image information is identified based on the multiple text data sets to generate multiple image indexes, and multiple PDF files are traversed according to the multiple image indexes to determine the target PDF file. The method includes: performing character matching on the multiple image information based on the multiple text data sets to generate character matching information; performing text processing on the multiple image information according to the character matching information to generate multiple text vectors; performing hashing based on the multiple image information according to the multiple text vectors to generate multiple image indexes; traversing and calculating multiple PDF files according to the multiple image indexes to obtain multiple image feature values; serializing the multiple image feature values ​​in descending order to generate a feature value sequence, and quickly matching the feature value sequence with multiple PDF files to determine the target PDF file.

[0065] Specifically, character matching refers to identifying and comparing characters or strings in a text dataset to determine the correspondence between image information and text. Text vector refers to the representation of text information converted into a numerical vector for easy computer processing and analysis, usually using word embedding technology such as Word2Vec or TF-IDF. Hashing refers to converting data into a fixed-length value through a hash function for fast search and comparison. Image feature values ​​refer to key features extracted from an image, which are used to describe the numerical representation of the image content and are usually obtained through image processing algorithms. Feature value sequence refers to a list of image feature values ​​arranged in descending order, which is used to quickly match and locate the target PDF file.

[0066] First, characters in multiple text datasets are compared with image information through character matching to generate character matching information. Natural language processing tools such as NLTK or spaCy are used to improve matching accuracy. Then, text processing is performed on the matched image information to generate multiple text vectors. Common methods are to use TF-IDF to vectorize text, or to use models such as BERT for deep embedding.

[0067] Next, these text vectors are hashed to generate multiple image indexes. Hash algorithms such as SHA-256 can ensure the uniqueness and fast search of different inputs. Subsequently, according to the generated image index, multiple PDF files are traversed and the feature values ​​of each image are calculated. Feature value extraction usually uses OpenCV or other image processing libraries to obtain image features such as shape and color.

[0068] Finally, all image feature values ​​are serialized in descending order to form a feature value sequence, so as to facilitate rapid matching with multiple PDF files according to the feature value sequence and determine the target PDF file.

[0069] For example, if the image information of "Customer Satisfaction Survey" is identified in the text dataset, the information is matched with characters and a text vector is generated. Assuming the hash value of the text vector is "abc123", the hash value is used to traverse the PDF file to find the document with the same feature value, and finally confirm the target PDF file.

[0070] This step significantly improves the efficiency of information retrieval. Through accurate identification and fast matching of image information, users can quickly find PDF files related to specific needs. Through hashing and feature value serialization, it remains efficient even in the face of large amounts of data, improving overall retrieval performance and user experience.

[0071] Furthermore, a simulated pre-selection is performed based on the target PDF file to generate a plurality of pre-selected images, and a full-text search is performed based on the plurality of pre-selected images to obtain pre-selection feedback information. The method comprises: setting a pre-selection condition for the target PDF file according to the feature value sequence; performing a semantic analysis on the target PDF file based on the pre-selection condition to generate semantic understanding information, and constructing a pre-selection decision according to the semantic understanding information; executing the pre-selection decision to perform a simulated pre-selection on the target PDF file to generate the plurality of pre-selected images; performing text extraction based on the plurality of pre-selected images to generate a plurality of pre-selected texts, and performing a full-text search based on the plurality of pre-selected texts to generate a plurality of search results; performing a context feedback evaluation on the plurality of pre-selected images based on the plurality of search results to generate a search relevance score list; and adding the search relevance score list to the pre-selection feedback information.

[0072] Specifically, the pre-selection condition refers to the criteria or rules set according to the feature value sequence, which are used to filter the relevant content in the target PDF file. The pre-selection decision refers to the judgment formed based on the semantic understanding information to determine which content needs to be pre-selected and extracted. The context feedback evaluation is to evaluate the relevance and effectiveness of the pre-selected images based on the search results to form a score list.

[0073] First, pre-selection conditions are set based on the feature value sequence, such as "find content containing the keywords 'quality' and 'complaint'". Then, a natural language processing tool (such as spaCy or NLTK) is used to perform semantic analysis on the target PDF file to extract semantic understanding information related to the pre-selection conditions, such as determining the main topic and related keywords of the file.

[0074] Based on this understanding information, a pre-selection decision is constructed to determine which images need to be pre-selected. On this basis, a simulated pre-selection is performed on the target PDF file, and multiple pre-selected images that meet the conditions are extracted through an image processing library (such as OpenCV).

[0075] Next, text extraction is performed on the generated pre-selected images, usually using OCR (optical character recognition) technology, such as Tesseract, to extract the text contained in the image. The extracted multiple pre-selected texts will be used for full-text search to find content related to user needs and generate multiple search results.

[0076] Subsequently, the pre-selected images are evaluated for contextual feedback based on these search results, and the matching degree of each image with the search results is analyzed to generate a search relevance score list. This score list will be added to the pre-selected feedback information to provide a basis for subsequent image screening.

[0077] For example, a user wants to find images related to "customer satisfaction". First, set the pre-selection conditions, including the two keywords "customer" and "satisfaction". Through semantic analysis, extract the topics related to these keywords and build a pre-selection decision. Then, perform image extraction on the target PDF file to find multiple relevant images. Next, use OCR technology to extract the text in the image, and perform a full-text search based on these texts to find paragraphs related to customer satisfaction. Finally, evaluate each pre-selected image and the search results, generate a search relevance score list, and integrate it into the pre-selection feedback information.

[0078] This step improves the accuracy and efficiency of image screening. By combining pre-selection conditions, semantic analysis and contextual feedback, it can more intelligently identify and extract images that are highly relevant to user needs, reducing the need for manual intervention, enhancing the user experience and ensuring that users can quickly find the information they need.

[0079] Furthermore, the pre-selection decision is executed to simulate the pre-selection of the target PDF file to generate the multiple pre-selected images. The method includes: setting a decision rule, based on multiple dimensions, executing the pre-selection decision according to the decision rule to generate multiple initial pre-selected images, and the decision rule includes multiple dimensions; judging whether the multiple initial pre-selected images meet the preset standard threshold, if the multiple pre-selected images do not meet the preset standard threshold, performing distortion analysis on the multiple pre-selected images to generate distortion analysis results; performing image ratio correction on the multiple initial pre-selected images according to the distortion analysis results to obtain the multiple pre-selected images.

[0080] Specifically, dimensions refer to different aspects or features that need to be considered in the decision-making process, such as the clarity, contrast, color distribution, etc. of an image. Initial pre-selected images refer to the first batch of images generated by the pre-selection decision, which have been initially screened but have not yet met all quality criteria. Preset standard thresholds are used to evaluate specific criteria for image quality, usually including image size, contrast, etc. Distortion analysis refers to analyzing whether an image has distortion or deformation problems to ensure that the image's proportions and structure are correct. Image ratio correction refers to improving image quality by adjusting the size and proportion of an image to meet preset standards.

[0081] First, a series of criteria are set to perform pre-selection decisions based on multi-dimensional decision rules. These rules include image size requirements (such as minimum width and height) and contrast (such as the distribution of gray levels). Subsequently, multiple initial pre-selected images are generated.

[0082] Next, it is determined whether these initial pre-selected images meet the preset standard threshold. For example, image processing tools (such as OpenCV) are used to check whether the size of each image is within the set range and whether the contrast reaches the expected level. If some images do not meet the standards, distortion analysis will be performed to detect whether the image has distortion problems such as stretching or compression.

[0083] After the distortion analysis is completed, analysis results are generated to identify which images need to be scaled. Using image processing techniques (such as image resampling and interpolation), scale correction is performed on images that do not meet the standards to ensure that the size and quality of the images meet the preset standards.

[0084] This step not only effectively screens out images that meet the standards, but also automatically adjusts the image size and quality when necessary, ensuring that the final pre-selected images are clear and have good visual effects.

[0085] Furthermore, the pre-selected feedback information is sent to a remote control terminal, and an image screening instruction is generated to perform intelligent screening response on the PDF image. The method includes: weighting and distributing the pre-selected feedback information according to the semantic understanding information by the remote control terminal to generate a plurality of weight coefficients; performing reinforcement learning according to the plurality of weight coefficients and the search relevance score list to generate an image decision tree; verifying the target PDF file based on the image decision tree to determine the target PDF image; performing image retrieval according to the target PDF image, and generating the image screening instruction according to the retrieval result.

[0086] Specifically, first, the pre-selected feedback information is sent to the remote control terminal, and these feedbacks are weighted and distributed through semantic understanding information to generate multiple weight coefficients. This process can use natural language processing tools (such as NLTK or spaCy) to analyze the feedback content and extract key information to determine the importance of each feedback.

[0087] Next, reinforcement learning is performed based on the generated weight coefficients and the search relevance score list, which usually requires a machine learning framework (such as TensorFlow or PyTorch) for model training. At this stage, the image screening decision is optimized by continuously adjusting the weight coefficients to generate an image decision tree. Among them, reinforcement learning specifically combines the weight coefficients with the search relevance score through a reinforcement learning algorithm (such as Q-learning) to build a training environment. Using Python and deep learning frameworks (such as TensorFlow or PyTorch), a neural network structure is designed for the model. The input includes image features and relevance scores, and the output is a decision action (such as screening or excluding images). Next, by continuously simulating different selection processes, the selected images, the feedback obtained, and the corresponding reward values ​​are recorded. For example, if the selection of an image leads to an increase in the relevance score, a positive reward is given, otherwise a negative reward is given. As the training proceeds, the model will gradually learn effective decision paths and generate an image decision tree to help quickly make the best choice in future screening. In this way, the image decision tree can not only efficiently screen out the images most relevant to user needs, but also continuously optimize the decision strategy, significantly improving the accuracy and efficiency of image screening. For example, when processing large amounts of documents, key information images can be identified within seconds, greatly improving work efficiency.

[0088] Finally, based on the constructed image decision tree, the target PDF file is verified to determine which images are the most relevant. Finally, image retrieval will be performed based on the target PDF images, which can be achieved with the help of database query tools (such as Elasticsearch) to ensure that the most suitable images are found quickly.

[0089] Through this step, intelligent screening of images is achieved, which can respond to user needs more effectively. The combination of weighted feedback and reinforcement learning ensures the accuracy and relevance of image selection, greatly improving the user experience.

[0090] In summary, the analysis-driven PDF image screening method provided in the embodiments of the present application has the following technical effects:

[0091] 1. Through intelligent analysis of image information in PDF files, efficient image extraction and format conversion are achieved. Through automated processing, the need for manual intervention is reduced, the speed and accuracy of information processing are improved, allowing users to quickly obtain the required image data and enhancing the efficiency of file management.

[0092] 2. Based on the streaming reading and parsing of multiple PDF files, the object tree structure is constructed, which helps to systematically manage and efficiently extract content data. Through this structured processing, file objects of different formats can be effectively identified and integrated, improving the parsing ability of complex PDF files and ensuring the integrity and consistency of information.

[0093] 3. The combination of decision rules and distortion analysis makes the image screening process more scientific and accurate. By performing standard threshold judgment and ratio correction on the image, it is ensured that the final selected pre-selected image meets the specific standards, which greatly improves the quality of image screening.

[0094] Any step of the method described above can be stored as a computer instruction or program in an unlimited computer memory, and can be called and recognized by an unlimited computer processor to implement any method in the embodiments of the present application, without any unnecessary restrictions.

[0095] Furthermore, the first or second mentioned above not only represents an order relationship, but also represents a specific concept, and / or refers to the selection of multiple elements individually or in whole. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technologies, the present application intends to include these modifications and variations.

Claims

1. The PDF image screening method driven by intelligent analysis is characterized by: The method comprises: Intelligently analyze multiple PDF files to obtain multiple image information, and perform format conversion based on the multiple image information to obtain multiple bitmap formats; Traversing a plurality of PDF files to capture images based on the plurality of bitmap formats, and determining a plurality of image ratio data; Performing image content recognition on the plurality of image information according to the plurality of image ratio data to generate a plurality of text data sets; Identify the multiple image information based on the multiple text data sets, generate multiple image indexes, traverse multiple PDF files according to the multiple image indexes, and determine a target PDF file; A simulated pre-selection is performed based on the target PDF file to generate multiple pre-selected images, a full-text search is performed based on the multiple pre-selected images, pre-selected feedback information is obtained, the pre-selected feedback information is sent to a remote control terminal, and an image screening instruction is generated to perform intelligent screening response on the PDF image.

2. The intelligent analysis driven PDF image screening method according to claim 1, characterized in that: Intelligently parse multiple PDF files to obtain multiple image information, the method includes: Performing streaming reading based on multiple PDF files, determining multiple file formats, parsing the multiple PDF files according to the multiple file formats, and obtaining multiple file object types; Integrate the multiple file formats and the multiple file object types to construct an object tree structure; Traversing the plurality of PDF files according to the object tree structure to perform content stream parsing and extract a plurality of content data; Mapping the multiple content data according to the multiple file object types and the multiple file formats, building a PDF parsing library; The PDF parsing library traverses multiple PDF files for matching, and batch parses the multiple PDF files according to the matching results to obtain the multiple image information.

3. The intelligent analysis driven PDF image screening method according to claim 2, characterized in that: The PDF parsing library traverses multiple PDF files for matching, and batch parses the multiple PDF files according to the matching results to obtain the multiple image information. The method includes: Load multiple PDF files based on the file usage path, use the PDF parsing library to parse and allocate the multiple PDF files, and determine multiple parsing threads; Matching multiple PDF files according to the multiple parsing threads, processing the multiple PDF files in parallel according to the parsing matching results, and generating parsing instructions; The image processing module is activated by the parsing instruction, and image enhancement is performed on multiple PDF files by the image processing module to extract graphic contours and generate a geometric data set; Deep learning is performed based on the geometric data set to determine the multiple image information.

4. The intelligent analysis driven PDF image screening method according to claim 3, characterized in that: Based on the multiple bitmap formats, multiple PDF files are traversed to capture images and multiple image ratio data are determined. The method includes: Read images from multiple PDF files based on the multiple bitmap formats to generate a bitmap image set, and perform position capture according to the bitmap image set to generate image matrix data; Binarization processing is performed on the image matrix data by using mean filtering, edge detection is performed according to the image processing result, and a plurality of straight line data are determined; Performing equal distance analysis according to the plurality of straight line data to generate a detection scale; According to the geometric data set, the plurality of PDF files are geometrically deduced using the detection ruler to determine the plurality of image ratio data.

5. The intelligent analysis driven PDF image screening method according to claim 1, characterized in that: The plurality of image information are identified based on the plurality of text data sets, a plurality of image indexes are generated, a plurality of PDF files are traversed according to the plurality of image indexes, and a target PDF file is determined, the method comprising: Performing character matching on the plurality of image information based on the plurality of text data sets to generate character matching information; Performing text processing on the plurality of image information according to the character matching information to generate a plurality of text vectors; Based on the multiple image information, hashing is performed according to the multiple text vectors to generate multiple image indexes; Perform traversal calculations on multiple PDF files according to the multiple image indexes to obtain multiple image feature values; The plurality of image feature values ​​are serialized in descending order to generate a feature value sequence, and the feature value sequence is quickly matched with a plurality of PDF files to determine the target PDF file.

6. The intelligent analysis driven PDF image screening method according to claim 5, characterized in that: Performing simulation pre-selection based on the target PDF file to generate multiple pre-selected images, performing full-text search based on the multiple pre-selected images to obtain pre-selected feedback information, the method comprising: Setting a preselection condition for the target PDF file according to the feature value sequence; Performing semantic analysis on the target PDF file based on the pre-selected conditions to generate semantic understanding information, and constructing a pre-selection decision according to the semantic understanding information; Executing the pre-selection decision to simulate pre-selection of the target PDF file to generate the plurality of pre-selected images; Performing text extraction based on the multiple pre-selected images to generate multiple pre-selected texts, and performing full-text search based on the multiple pre-selected texts to generate multiple search results; Performing context feedback evaluation on the plurality of pre-selected images based on the plurality of search results to generate a search relevance score list; The search relevance score list is added to the pre-selected feedback information.

7. The intelligent analysis driven PDF image screening method according to claim 6, characterized in that: Executing the pre-selection decision to simulate pre-selection of the target PDF file to generate the plurality of pre-selected images, the method comprising: Setting a decision rule, based on multiple dimensions, executing the pre-selection decision according to the decision rule to generate multiple initial pre-selected images, wherein the decision rule includes multiple dimensions; Determining whether the multiple initial pre-selected images meet a preset standard threshold, and if the multiple pre-selected images do not meet the preset standard threshold, performing distortion analysis on the multiple pre-selected images to generate a distortion analysis result; The multiple initial pre-selected images are subjected to image ratio correction according to the distortion analysis results to obtain the multiple pre-selected images.

8. The intelligent analysis driven PDF image screening method according to claim 6, characterized in that: The pre-selected feedback information is sent to a remote control terminal, and an image screening instruction is generated to perform intelligent screening response on the PDF image. The method includes: The remote control terminal weights and distributes the pre-selected feedback information according to the semantic understanding information to generate a plurality of weight coefficients; Perform reinforcement learning according to the multiple weight coefficients and the search relevance score list to generate an image decision tree; Verify the target PDF file based on the image decision tree to determine a target PDF image; An image search is performed according to the target PDF image, and the image screening instruction is generated according to the search result.

Images