Paper archive digital quality self-checking system based on AI detection

Abstract

The invention discloses a paper archive digital quality self-inspection system based on AI detection, which relates to the technical field of archive management and comprises an image acquisition module, a content extraction module, a threshold adjustment module, a quality detection module, a result evaluation module, a feedback correction module and a data management module. The image acquisition module is responsible for acquiring images of paper archives by using a scanner or a digital camera and performing preliminary image preprocessing. According to the AI detection paper archive digital quality self-inspection system, unified execution of quality inspection standards is ensured through a rule engine, subjective differences of quality standard understanding of different quality inspection personnel are eliminated, and a detection threshold value is dynamically adjusted according to the age, the material and the storage state of the archive so as to adapt to quality inspection requirements of different archives; and particularly, the difficulty in identifying handwritings, faded handwritings, special symbols and the like in historical archives is optimized, so that the comprehensiveness and the efficiency of detection are greatly improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of file management, and in particular to an AI detection-based paper file digitization quality self-checking system. BACKGROUND

[0002] Paper file digitization refers to a processing process in which a scanner or a digital camera and the like are used to systemically process paper files, so that the paper files are converted into digital images or digital texts stored on a carrier such as a magnetic tape, a magnetic disk or an optical disk and can be recognized by a computer. It is a profound reform in the field of file management in the information age, and realizes long-term preservation and efficient utilization of file resources by converting file information on a traditional carrier into a digital format through a technical means.

[0003] Traditional paper file digitization quality detection mainly relies on a combination of manual inspection and simple computer program verification, forming a relatively fixed but low-efficiency working mode. With the expansion of the digitization scale and the improvement of user quality requirements, the manual detection speed cannot match the digitization production speed, forming a serious process blockage, and different quality inspection personnel have subjective differences in understanding the quality standards, resulting in inconsistent scale grasping. Therefore, the traditional manual detection is gradually replaced by a self-checking system with higher efficiency and stronger quality control capability. Compared with manual detection (10-30% sampling inspection), the self-checking system not only realizes automatic detection of all samples, but also improves the detection speed by hundreds of times, and ensures uniform execution of standards through a rule engine, eliminating manual subjective bias.

[0004] However, in actual operation, the existing digitization quality self-checking system has insufficient recognition accuracy for handwritten characters, faded characters and special symbols (such as engineering drawing marks) in historical files, which is far lower than the recognition level of more than 95% of printed characters, and still needs manual intervention for supplementary inspection; and a fixed detection threshold is usually used, which is difficult to dynamically adjust parameters to adapt to files of different ages, materials or storage states, such as using standard contrast detection for yellowed paper, resulting in a 3-5 times increase in false positive rate of historical files. SUMMARY

[0005] The application aims to provide an AI detection-based paper file digitization quality self-checking system to solve the problems in the background art.

[0006] To achieve the above-mentioned purpose, the application provides the following technical scheme: an AI detection-based paper file digitization quality self-checking system, comprising: Image acquisition module: responsible for using devices such as scanners or digital cameras to collect images of paper archives and performing preliminary image preprocessing to improve image quality, providing clear and contrast images for subsequent recognition; Content extraction module: responsible for text recognition on preprocessed images to extract text information and special symbol features in archives, providing basic data for subsequent quality detection; Threshold adjustment module: responsible for dynamically adjusting detection thresholds according to the age, material and storage state of the archives to adapt to the quality detection needs of different archives and reduce false positive rates; Quality detection module: responsible for quality detection of digitized archives based on extracted text information and features combined with dynamically adjusted detection thresholds; Result evaluation module: responsible for comprehensive detection results, evaluating overall quality levels, generating pass rate reports, and identifying and reporting potential quality problems; Feedback correction module: responsible for feeding back quality detection results to users and providing correction suggestions or automatic correction functions to correct recognition errors and improve the quality of digitized archives; Data management module: responsible for managing and storing digitized archive data and its detection records by building a database.

[0007] Further, the image acquisition module integrates image acquisition hardware interfaces, calls image processing libraries (such as OpenCV) for preprocessing, and the preprocessing steps are as follows: Denoising: applying Gaussian filtering, median filtering and other algorithms to remove random noise in the image; Binarization: for text archives, binarization is used to convert images to black and white; Contrast enhancement: histogram equalization, contrast stretching and other methods are used to enhance the contrast of the image, making the text and background more distinct; Tilt correction: detect the tilt angle in the image and perform rotation correction to ensure that the text lines are horizontal; Size normalization: adjust the image to a uniform size; Crop and scale: according to pre-set rules or user specifications, crop and scale the image to remove unnecessary edge areas and adjust the image size.

[0008] Further, the content extraction module uses OCR (Optical Character Recognition) technology combined with deep learning models CNN and RNN to improve the recognition accuracy of handwritten, faded text and special symbols. Specifically, it trains deep learning models to recognize different fonts and text clarity, and uses feature extraction algorithms to capture special symbol features.

[0009] Further, the content extraction module includes the following technical architecture: 1) OCR base layer: using traditional OCR engines (e.g. Tesseract, ABBYY) for preliminary text localization and segmentation, quickly extracting printed text; 2) Deep learning enhancement layer: combining convolutional neural networks (CNN) and recurrent neural networks (RNN) to form a hybrid model architecture for fine recognition and correction; CNN (Convolutional Neural Network): used for recognition of handwritten and faded text, extracting local features (such as strokes and connected structures) through convolutional layers, and classifying characters through fully connected layers; RNN (Recurrent Neural Network): combining LSTM or GRU units to process sequential text (such as line text and special symbol arrangements), capturing context dependencies, and suitable for scenarios such as engineering drawing symbols and mathematical formulas that require context understanding.

[0010] Further, the training data of the CNN includes handwritten samples of different eras and writing styles (such as historical archives of brush and pen writing), as well as faded text simulation data generated by image degradation.

[0011] Further, the CNN and RNN use a collaborative training method, including a two-stage training strategy: Stage 1, independent pre-training: The CNN converges on the character image classification task until the accuracy is >92%; The RNN converges on the text line recognition task until the CER is <8%; Stage 2, joint fine-tuning: The feature vectors extracted by the CNN and RNN are weighted and fused in a 6:4 ratio, and CTCLoss is used to optimize the sequence recognition effect; The CNN contributes 60% weight: mainly bearing local visual features, including handwriting form (pen features of brush / pen), faded traces (texture patterns of ink diffusion), and symbol geometric features (topological structure of engineering drawings); The RNN contributes 40% weight: mainly bearing sequence semantic features, including inter-character syntax relationships (word order of historical documents), professional symbol logic (operation priority of mathematical formulas), and cross-line context association (row-column correspondence of table data).

[0012] Further, the threshold adjustment module uses machine learning algorithms to analyze historical archive data, establishes a mapping relationship between different archive types and the best detection threshold, and automatically adjusts detection parameters such as contrast and clarity according to the archive situation during actual detection, including the following steps: Step 1, data collection and preprocessing: Collect a large amount of historical archive data, including archive samples of different ages, materials and preservation states, and corresponding quality detection results (such as qualified, unqualified and specific problem types), and preprocess the data, such as feature extraction and data cleaning, for subsequent analysis; Step 2, feature extraction and selection: Extract key features from the archive data, such as the age of the archive, the type of material, the preservation environment, etc., and use feature selection techniques to select the features that have the greatest impact on the detection threshold, in order to reduce computational complexity and improve model accuracy; Step 3, model training and verification: Select appropriate machine learning algorithms (such as decision trees, random forests, support vector machines, etc.) to train the model, establish the mapping relationship between the type of archive and the optimal detection threshold, and use cross-validation to verify the model; Step 4, threshold mapping relationship establishment: Through the trained model, the optimal detection threshold is predicted for each type of archive, and a threshold mapping table or function is established for quick query and application in the actual detection process.

[0013] Further, the quality detection module uses a rule engine to automatically detect all samples based on the recognition results, and evaluates the digitalization quality, that is, all digitalized archives are detected according to the unified standard through the pre-set standardized rule library, achieving 100% full-sample detection, and the rule content includes: Basic rules: whether the parameters such as image resolution, contrast, color mode meet the requirements of digital storage; Content rules: text integrity (such as missing characters, garbled characters), recognition accuracy of special symbols (such as engineering drawing marks); Dynamic rules: combined with the dynamic parameters output by the threshold adjustment module (such as reducing the contrast detection threshold for yellowing paper), flexible adaptation to different archive characteristics.

[0014] Further, the result evaluation module locates the root cause of systematic defects (such as the performance bias of a specific scanner) through correlation analysis, and provides repair suggestions, including adjusting the scanning device settings, rescan specific archives, post-processing of already digitized archives, etc., and marking difficult cases that need manual review.

[0015] Further, the repair suggestion generation method: combined with the knowledge base and expert system, according to the type and severity of the detection results, generate targeted repair guidance to help users quickly solve problems.

[0016] The present application provides an AI detection paper archive digitalization quality self-checking system, which has the following beneficial effects: The application ensures the unified execution of quality detection standards through the rule engine, eliminates subjective differences in the understanding of quality standards by different quality inspectors, and dynamically adjusts the detection threshold according to the age, material and preservation state of the archives to adapt to the quality detection needs of different archives, especially optimizes the recognition difficulties of handwritten body, faded handwriting and special symbols in historical archives, realizes the full-sample automatic detection of digital archives compared with the traditional artificial detection sampling, greatly improves the comprehensiveness and efficiency of detection, and the detection speed is hundreds of times higher than that of artificial detection, which can quickly process large-scale digital archives, avoids process blockage, and has high detection precision, significantly reduces the false positive rate. In addition, the system can also evaluate the overall quality level based on the comprehensive detection results, generate a qualified rate report, and provide correction suggestions or automatic correction functions, which can help further improve the quality of digital archives. BRIEF DESCRIPTION OF DRAWINGS

[0017] Fig. 1 A logic block diagram of an AI-detected paper archive digitalization quality self-checking system of the application; Fig. 2 A threshold adjustment module implementation step flowchart of an AI-detected paper archive digitalization quality self-checking system of the application. DETAILED DESCRIPTION

[0018] The embodiments of the application will be further described in detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate the application, but cannot be used to limit the scope of the application.

[0019] As shown in the drawings, Figs. 1-2 An AI-detected paper archive digitalization quality self-checking system includes: An image acquisition module: uses a scanner or a digital camera and other devices to acquire the image of the paper archive and performs preliminary image preprocessing to improve image quality and provide clear and contrast images for subsequent recognition.

[0020] In this embodiment, the image acquisition module integrates an image acquisition hardware interface, calls an image processing library (such as OpenCV) for preprocessing, and the preprocessing steps are as follows: Denoising: applying Gaussian filtering, median filtering and other algorithms to remove random noise in the image; Binarization: for text archives, the image is converted to black and white through binarization processing; Contrast enhancement: enhance the contrast of the image through histogram equalization, contrast stretching and other methods to make the text and background more distinct; Skew correction: detect the skew angle in the image and perform rotation correction to ensure that the text lines are horizontal; Size normalization: adjust the image to a uniform size; Crop and resize: Crop and resize the image according to preset rules or user specifications, remove unnecessary edge areas, and adjust the image size.

[0021] Content extraction module: Perform text recognition on the preprocessed image, extract text information and special symbol features in the archive, and provide basic data for subsequent quality detection.

[0022] In this embodiment, the content extraction module uses OCR (Optical Character Recognition) technology combined with deep learning models CNN and RNN to improve the recognition accuracy of handwritten, faded text and special symbols. Specifically, it trains deep learning models to recognize different fonts and text clarity, and uses feature extraction algorithms to capture special symbol features.

[0023] In this embodiment, the content extraction module includes the following technical architecture: 1) OCR base layer: Use traditional OCR engines (such as Tesseract, ABBYY) for preliminary text positioning and segmentation, and quickly extract printed text.

[0024] 2) Deep learning enhancement layer: Combine convolutional neural networks (CNN) and recurrent neural networks (RNN) to form a hybrid model architecture for fine recognition and correction.

[0025] CNN (Convolutional Neural Network): Used for recognition of handwritten and faded text, extracts local features (such as strokes and connected structures) through convolutional layers, and classifies characters through fully connected layers; CNN training data includes handwritten samples of different ages and writing styles (such as historical archives of brush calligraphy and penmanship), as well as faded text simulation data generated by image degradation.

[0026] RNN (Recurrent Neural Network): Combines LSTM or GRU units to process sequential text (such as line text and special symbol arrangements), capture context dependencies, and is suitable for scenarios such as engineering drawing symbols and mathematical formulas that require context understanding.

[0027] CNN and RNN use a collaborative training approach, including a two-stage training strategy: Stage 1, independent pre-training: CNN converges on the character image classification task until the accuracy is >92%; RNN converges on the text line recognition task until CER <8%.

[0028] Stage 2, joint fine-tuning: Weighted fusion of CNN and RNN feature vectors in a 6:4 ratio, and use CTCLoss to optimize sequence recognition results; CNN contributes 60% weight: mainly carries local visual features, including handwriting form (brush / pen writing characteristics), fading traces (ink diffusion texture patterns), symbol geometric features (topological structure of engineering drawings); RNN contributes 40% weight: mainly carries sequence semantic features, including inter-character syntax relationships (historical document classical Chinese word order), professional symbol logic (mathematical formula operation priority), cross-row context association (table data row-column correspondence).

[0029] Threshold adjustment module: dynamically adjust the detection threshold according to the age, material and preservation state of the archives, to adapt to the quality detection needs of different archives and reduce the false positive rate.

[0030] In this embodiment, the threshold adjustment module uses machine learning algorithms to analyze historical archive data, establishes a mapping relationship between different archive types and the best detection threshold, and automatically adjusts the detection parameters such as contrast and clarity according to the archive conditions in the actual detection process, and the implementation includes the following steps: Step 1, data collection and preprocessing: Collect a large amount of historical archive data, including archive samples of different ages, materials and preservation states, and corresponding quality detection results (such as qualified, unqualified and specific problem types), and preprocess the data, such as feature extraction, data cleaning, etc., in order to facilitate subsequent analysis.

[0031] Step 2, feature extraction and selection: Extract key features from archive data, such as archive age, material type, storage environment, etc., and use feature selection techniques to filter out the features that have the greatest impact on detection threshold, in order to reduce computational complexity and improve model accuracy.

[0032] Step 3, model training and verification: Select appropriate machine learning algorithms (such as decision tree, random forest, support vector machine, etc.) to train the model, establish the mapping relationship between archive type and best detection threshold, and use cross-validation to verify the model.

[0033] Step 4, threshold mapping relationship establishment: Through the trained model, predict the best detection threshold for each archive type, establish a threshold mapping table or function, and use it to quickly query and apply in the actual detection process.

[0034] In practical applications, when a new file enters the self-inspection system, the threshold adjustment module extracts the characteristics of the file's age, material, and preservation state, and inputs these characteristics into the trained machine learning model. The model predicts the optimal detection threshold suitable for the file. The threshold adjustment module dynamically adjusts the detection threshold based on the prediction results to ensure the accuracy and effectiveness of quality detection.

[0035] Case 1: Processing of brush manuscript in earlier period Initial state: Fixed threshold setting: contrast 0.65, clarity 0.7; Problem manifestation: 38% of valid handwriting is misjudged as stains due to ink bleeding.

[0036] Dynamic adjustment process: Recognize the feature combination of "vertical layout", "brush calligraphy", and "rice paper texture", and match the best parameters of similar files in historical data; Lower the contrast threshold to 0.52 (adapt to light ink writing); Increase the clarity threshold to 0.75 (strengthen the detection of stroke continuity); Enable local adaptive strategy: use 0.45 contrast for seal area separately.

[0037] Effectiveness improvement: The misjudgment rate decreases from 38% to 9%, and the key information recall rate increases by 62%.

[0038] Case 2: Optimization of 1960s oil printing files Initial state: General threshold: contrast 0.7, clarity 0.6; Main problem: wax paper printing causes backside text interference, with a misjudgment rate of 43%.

[0039] Dynamic adjustment process: Detect "wax paper texture", "typewriter font", and "double-sided light transmission" features, and match the best parameters of similar files in historical data; Increase the contrast threshold to 0.78 (suppress interference from printing); Add a color detection channel (filter light blue printing marks); Relax the clarity requirement for oil printing-specific handwriting breakpoints to 0.55.

[0040] Effectiveness improvement: The printing misjudgment elimination rate reaches 92%, and the effective information capture rate remains above 95%.

[0041] Case 3: Parameter adaptation of engineering blueprints Initial scenario Default settings: contrast 0.6, sharpness 0.8; Actual performance: the fine line detection rate is as high as 51%.

[0042] Dynamic adjustment process: Identify "blue background", "black line", "CAD font" features, and match the best parameters of similar archives in historical data; Trigger professional drawing processing mode: 1) replace spatial domain detection with frequency domain analysis method; 2) contrast threshold is converted to dynamic range [0.55-0.85]; 3) enable sub-pixel detection for <0.3mm fine lines; Implement hierarchical threshold strategy for different line widths.

[0043] Effectiveness improvement: Fine line detection rate increased from 49% to 97%, and size annotation accuracy reached 99.2%.

[0044] Quality detection module: based on extracted text information and features, combined with dynamically adjusted detection threshold, quality detection is performed on digitized archives.

[0045] In this embodiment, the quality detection module uses a rule engine to automatically detect all samples and evaluate digitization quality. By using a pre-set standardized rule library, all digitized archives are detected according to a unified standard, achieving 100% full-sample detection. Compared with traditional manual detection, which only detects 10-30% of samples, the detection speed is improved by hundreds of times, and the standard is executed with zero deviation. Rule content: Basic rules: whether parameters such as image resolution, contrast, color mode meet the requirements of digitization storage; Content rules: text integrity (such as missing characters, garbled characters), recognition accuracy of special symbols (such as engineering drawing marks); Dynamic rules: combined with dynamic parameters output by the threshold adjustment module (such as reducing contrast detection threshold for yellowed paper), flexibly adapt to different archive characteristics.

[0046] Result evaluation module: comprehensive detection results, evaluate the overall quality level, generate qualified rate report, and identify and report potential quality problems.

[0047] In this embodiment, the result evaluation module locates the root cause of systematic defects (such as performance deviation of a specific scanner) through correlation analysis, and provides repair suggestions, including adjusting scanning device settings, rescanning specific archives, post-processing of digitized archives, etc. Mark difficult cases that need manual review.

[0048] The generation method of repair suggestions: combined with knowledge base and expert system, according to the type and severity of detection results, generate targeted repair guidance to help users quickly solve problems.

[0049] Feedback correction module: feedback the quality detection result to the user, and provide correction suggestions or automatic correction function to correct the recognition error and improve the quality of the digital archives; in the embodiment, the feedback correction module develops a user interface to display the detection result, integrates an automatic correction algorithm or provides a manual correction tool for correcting the recognition error.

[0050] Data management module: manage and store the digital archives data and its detection records by constructing a database.

[0051] The embodiments of the present application are given for illustration and description only, and are not intended to be exhaustive or to limit the application to the forms disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art. Embodiments were chosen and described in order to best explain the principles of the application and its practical application, and to enable others skilled in the art to understand the application for various embodiments with various modifications as are suited to the particular use contemplated.

Claims

1. An AI-based self-inspection system for the quality of digitized paper archives, characterized in that, It includes an image acquisition module, a content extraction module, a threshold adjustment module, a quality detection module, a result evaluation module, a feedback correction module, and a data management module; The image acquisition module is responsible for acquiring images of paper documents using a scanner or digital camera and performing preliminary image preprocessing; The content extraction module is responsible for performing text recognition on the preprocessed image and extracting text information and special symbol features from the file; The threshold adjustment module is responsible for dynamically adjusting the detection threshold according to the age, material and preservation status of the archives to adapt to the quality inspection needs of different archives. The quality inspection module is responsible for performing quality inspection on digital archives based on extracted text information and features, combined with dynamically adjusted detection thresholds. The result evaluation module is responsible for integrating the test results, evaluating the overall quality level, generating a pass rate report, and identifying and reporting potential quality problems. The feedback correction module is responsible for feeding back the quality inspection results to the user and providing correction suggestions or automatic correction functions to correct identification errors. The data management module is responsible for managing and storing digitized archival data and its inspection records by building a database.

2. The AI-based self-inspection system for the digitization quality of paper archives according to claim 1, characterized in that, The image acquisition module integrates an image acquisition hardware interface and calls an image processing library for preprocessing. The preprocessing steps are as follows: Denoising: Apply Gaussian filtering and median filtering algorithms to remove random noise from the image; Binarization: For text-based archives, binarization converts the image into black and white. Contrast enhancement: Enhances image contrast through methods such as histogram equalization and contrast stretching, making text and background more distinct; Tilt correction: Detect the tilt angle in the image and perform rotation correction to ensure that the text lines are horizontal; Size normalization: Adjusts the image to a uniform size; Cropping and scaling: Crops and scales images according to preset rules or user specifications, removing unnecessary edge areas and adjusting image size.

3. The AI-based self-inspection system for the digitization quality of paper archives according to claim 1, characterized in that, The content extraction module uses OCR technology, combined with deep learning models CNN and RNN, to improve the recognition accuracy of handwritten text, faded text and special symbols. Specifically, it trains deep learning models to recognize text with different fonts and handwriting clarity, and uses feature extraction algorithms to capture the features of special symbols.

4. The AI-based self-inspection system for the digitization quality of paper archives according to claim 1, characterized in that, The content extraction module includes the following technical architecture: 1) OCR Basic Layer: Uses a traditional OCR engine for initial text localization and segmentation to quickly extract printed text; 2) Deep learning enhancement layer: Combining convolutional neural networks and recurrent neural networks to form a hybrid model architecture for fine recognition and correction; CNN: Used for recognizing handwritten and faded text, it extracts local features through convolutional layers and classifies characters through fully connected layers; RNN: Combines LSTM or GRU units to process serialized text and capture contextual dependencies.

5. The AI-based self-inspection system for the digitization quality of paper archives according to claim 4, characterized in that, The training data for the CNN includes handwritten samples from different eras and writing styles, as well as simulated data of faded handwriting generated through image degradation.

6. The AI-based self-inspection system for the digitization quality of paper archives according to claim 4, characterized in that, The CNN and RNN are trained collaboratively, including a two-stage training strategy: Phase 1, Independent Pre-training: CNNs converge on character image classification tasks until the accuracy is >92%; RNNs converge on text line recognition tasks until CER < 8%; Phase 2, Joint Fine-tuning: The feature vectors extracted by CNN and RNN are weighted and fused in a 6:4 ratio, and CTCLoss is used to optimize the sequence recognition effect. The CNN contributes 60% of the weights: it mainly carries local visual features, including handwriting shape, fading marks, and geometric features of symbols; The RNN contributes 40% of the weights: it mainly carries the semantic features of the sequence, including the syntactic relationships between characters, the logic of specialized symbols, and the cross-line contextual relationships.

7. The AI-based self-inspection system for the digitization quality of paper archives according to claim 1, characterized in that, The threshold adjustment module uses machine learning algorithms to analyze historical archive data, establishes a mapping relationship between different archive types and the optimal detection threshold, and automatically adjusts the detection parameters accordingly. This is achieved through the following steps: Step 1: Data Collection and Preprocessing Collect historical archive data, including archive samples from different eras, materials, and preservation conditions, as well as corresponding quality inspection results, and preprocess the data; Step 2, Feature Extraction and Selection: Key features are extracted from archival data, and feature selection techniques are used to screen out the features that have the greatest impact on the detection threshold. Step 3, Model Training and Validation: Choose an appropriate machine learning algorithm to train the model, establish a mapping relationship between file type and the optimal detection threshold, and use cross-validation to validate the model; Step 4: Establishing the threshold mapping relationship: The trained model predicts the optimal detection threshold for each file type, and a threshold mapping table or function is established for rapid querying and application during actual detection.

8. The AI-based self-inspection system for the digitization quality of paper archives according to claim 1, characterized in that, The quality inspection module uses a rule engine to perform automated full-sample inspection of the recognition results and evaluate the digitization quality. Specifically, it uses a pre-set standardized rule base to inspect all digitized files according to a unified standard, achieving full-sample inspection. The rule content is as follows: Basic rule: Do the parameters meet the requirements for digital storage? Content rules: text integrity, accuracy of special character recognition; Dynamic rules: Combine the dynamic parameters output by the threshold adjustment module to flexibly adapt to different file characteristics.

9. The AI-based self-inspection system for the digitization quality of paper archives according to claim 1, characterized in that, The results evaluation module locates the root cause of systemic defects through correlation analysis and provides repair suggestions, including adjusting scanning equipment settings, rescanning specific files, post-processing digitized files, and marking difficult cases that require manual review.

10. The AI-based self-inspection system for the digitization quality of paper archives according to claim 1, characterized in that, The repair suggestions are generated by combining a knowledge base and an expert system, based on the type and severity of the detection results, to provide targeted repair guidance and help users quickly resolve the issues.

Images