An archive image data management system and method based on deep learning
By using deep learning-based OCR and CNN models, combined with dynamic integrity assessment, the inefficiency and lack of intelligence in traditional archival image management systems have been solved, enabling intelligent and refined management of archival images.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHENZHEN SHIXINDA TECH CO LTD
- Filing Date
- 2025-07-18
- Publication Date
- 2026-05-15
AI Technical Summary
Traditional archival image management systems suffer from inefficiency, high costs of manual intervention, and insufficient intelligence when processing multimodal data, verifying complex elements, and assessing dynamic integrity. As a result, they fail to meet the intelligent and refined needs of enterprises for archival management.
This paper adopts a deep learning-based approach, utilizes OCR technology to extract features from archival images, combines a CNN model for image denoising and correction, and uses a dynamic integrity assessment model to calculate the probability of text features appearing based on historical data to achieve intelligent management of archival images.
It improves the efficiency of archival image preprocessing, reduces the cost of manual intervention, enhances the intelligence of text feature extraction, dynamically adjusts the integrity assessment, and meets the enterprise's needs for intelligent and refined archival management.
Smart Images
Figure CN120833613B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of archival image data management technology, specifically an archival image data management system and method based on deep learning. Background Technology
[0002] With the deepening of enterprise digital transformation, the management of archival image data, as a core link in information asset management, affects the efficiency of enterprise operation and decision-making quality in terms of its integrity, accuracy and security. The demand for intelligent management of massive archival images is becoming increasingly urgent.
[0003] Traditional management methods exhibit significant limitations when dealing with multimodal data processing, complex element verification, and dynamic integrity assessment: First, archival image preprocessing is inefficient and costly due to manual intervention. Traditional methods rely on manual labor for scanning, noise reduction, and correction of archival images. Faced with tens of thousands of new archival documents daily, this not only results in long processing cycles but also increases the risk of missing crucial information due to human error. Second, text feature extraction and element verification lack intelligent methods. Traditional OCR technology can only perform basic text recognition and cannot perform semantic understanding of archival text features based on domain knowledge, leading to a high rate of missed detections for essential elements. Traditional manual, line-by-line verification remains a significant challenge. The current verification method is insufficient to detect hidden risks, resulting in a high rate of missed judgments and failing to meet the auditing needs of complex business scenarios. In addition, the archive integrity assessment system lacks dynamic adjustment capabilities. Existing systems typically use fixed rules and cannot quantitatively assess the probability of the presence of a corresponding image when a certain feature appears in similar archives in historical data. This leads to a disconnect between the integrity score and actual business needs. To address these issues, although existing archive image management systems have introduced some automated tools, they still have shortcomings in archive image data management, especially in processing archive images. They lack the ability to manage archive image data and cannot meet the needs of enterprises for intelligent and refined archive management. Summary of the Invention
[0004] The purpose of this invention is to provide a deep learning-based archival image data management system and method to solve the problems raised in the prior art.
[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for managing archival image data based on deep learning, the method comprising the following steps:
[0006] Using a collection of archival images as input, OCR technology is used to recognize each image and extract its feature set;
[0007] Keyword monitoring is conducted based on the characteristics of archival texts to verify the first and second elements in the archives;
[0008] When the archival text feature tki appears in the archival image, it is determined whether the first element exists. If it does not exist, the image completeness is adjusted according to the preset rules. If it exists, the probability of its occurrence is calculated based on the frequency of the corresponding required image when the text feature appears in the historical data. Here, tki represents the i-th archival text feature in the k-th archival image data in the archival image dataset.
[0009] If the calculated probability is compared with the set probability threshold, the completeness of the archive image is increased by a set percentage, and the adjusted completeness is used as the initial value for calculating the probability of the next required image for that text feature.
[0010] Using a collection of archival images as input, OCR technology is used to recognize each image and extract its feature set. The specific steps include:
[0011] Based on OCR technology, the archival image data set I={I1,I2,...,Im} is used as input. OCR recognition is performed on any archival image Ik to extract archival text features, denoted as Tk={tk1,tk2,...,ktn}. Here, I1,I2,...,Im represent the 1st, 2nd,...,mth archival images in the archival image data set, respectively, m represents the total number of archival images, Ik represents the kth archival image in the archival image data set, tk1,tk2,...,ktn represent the 1st, 2nd,...,nth archival text features in the kth archival image in the archival image data set, respectively, and n represents the number of archival text features in each archival image.
[0012] Image denoising, correction, and contrast enhancement based on CNN models achieve multi-dimensional technical optimization: Through convolutional neural network models such as U-Net, noise such as stains and scratches in scanned documents are intelligently identified and filtered out, reducing image noise interference rate; by using CNN combined with perspective transformation algorithm, text areas tilted by more than 5° are automatically corrected, shortening the correction time for a single image; for special media such as yellowed paper and thermal printing, CNN adaptively adjusts the grayscale mapping curve to improve the contrast between text and background, thus reducing the OCR recognition error rate.
[0013] Keyword monitoring based on archival text features is used to verify the first and second elements in the archives. Specific steps include:
[0014] Keyword monitoring is performed based on the extracted archival text features to verify the first and second elements in the archive. The first element represents the necessary image requirement when the archival text feature tki appears, and the second element represents the optional image requirement when the archival text feature tki appears. The set of image requirements required by the archival text feature tki is P={p1,p2,...,pq}, where p1,p2,...,pq represent the 1st, 2nd,...,qth required images when the archival text feature tki appears.
[0015] When the archival text feature tki appears in the archival image, it is determined whether the first element exists. If it does not exist, the image completeness is adjusted according to a preset rule; if it does exist, the probability of its occurrence is calculated based on the frequency of the corresponding required image when the text feature appears in historical data. The specific steps include:
[0016] When the archive image Ik contains the archive text feature tki, determine whether the archive image Ik has the first element. If the archive image Ik does not have the first element, then adjust the integrity of the archive image Ik to 0.
[0017] When the first element exists in the archival image Ik, the completeness of the archival image Ik is adjusted to 50%. Then, the probability of the archival text feature tki appearing in the required image pj is determined. The probability of the archival text feature tki appearing in the required image pj is defined as follows: P(pj|tki) = the number of times image pj exists when tki appears in the historical data / the total number of times tki appears in the historical data; where pj represents the j-th required image when the archival text feature tki appears.
[0018] The calculated probability is compared with a set probability threshold. If the probability exceeds the threshold, the image completeness is increased by a set percentage. The adjusted completeness is then used as the initial value for calculating the probability of the next required image for that text feature. The specific steps include:
[0019] Based on historical data, a probability threshold is set for the occurrence of the required image pj using the calculated archival text feature tki. The calculated probability of the required image pj using the archival text feature tki is compared with the set probability threshold. When the calculated probability of the required image pj using the archival text feature tki exceeds the set threshold, the completeness of the archival image Ik is increased by a percentage point each time. The adjusted completeness of the archival image Ik is used as the initial completeness of the archival image Ik adjusted according to the above content, where a represents the percentage increase when the completeness of the archival image is increased, and pj+1 represents the (j+1)th required image when the archival text feature tki appears.
[0020] A deep learning-based archival image data management system includes a feature extraction module, an element verification module, a completeness evaluation module, and a completeness adjustment module. The feature extraction module takes an archival image set as input and uses OCR technology to recognize each image and extract its feature set. The element verification module performs keyword monitoring based on archival text features to verify the first and second elements in the archive. The completeness evaluation module determines whether the first element exists when an archival text feature tki appears in an archival image. If not, the image completeness is adjusted according to a preset rule; if it exists, the probability of its occurrence is calculated based on the frequency of occurrence of the corresponding required image in historical data, where tki represents the i-th archival text feature in the k-th archival image data set. The completeness adjustment module compares the calculated probability with a set probability threshold. If the threshold is exceeded, the archival image completeness is increased by a set percentage, and the adjusted completeness is used as the initial value for calculating the probability of the next required image for that text feature.
[0021] The feature extraction module includes an OCR recognition unit and a feature structuring unit. The OCR recognition unit is used to extract archival text features from any archival image by taking the archival image dataset as input based on OCR technology. The feature structuring unit is used to parse the OCR-recognized text data into a structured feature set. The output of the OCR recognition unit is connected to the input of the feature structuring unit, and the output of the feature structuring unit is connected to the input of the element verification module.
[0022] The element verification module includes a keyword monitoring unit, a first element verification unit, and a second element verification unit. The keyword monitoring unit is used to perform keyword retrieval on the extracted feature set and match it with a preset business keyword library. The first element verification unit is used to determine the necessary requirements for the image when archival text features are present. The second element verification unit is used to determine the optional requirements for the image when archival text features are present. The output of the keyword monitoring unit is connected to the input of the first element verification unit and the input of the second element verification unit. The output of the element verification module is connected to the input of the integrity evaluation module.
[0023] The completeness assessment module includes a stage division strategy unit and a demand image probability calculation unit. The stage division strategy unit is used to set an initial completeness based on the first element verification result. The demand image probability calculation unit is used to divide the number of times image pj appears when the i-th archival text feature appears in the k-th archival image data in the historical data by the total number of times the i-th archival text feature appears in the k-th archival image data in the historical data, where pj represents the j-th demand image when the i-th archival text feature appears in the k-th archival image data. The output of the stage division strategy unit is connected to the input of the demand image probability calculation unit, and the output of the demand image probability calculation unit is connected to the input of the completeness adjustment module.
[0024] The integrity adjustment module includes a threshold comparison unit, an integrity increase unit, and an initial value update unit. The threshold comparison unit compares the probability value output by the demand image probability calculation unit with a preset probability threshold. The integrity increase unit increases the integrity of the archive image when the calculated probability that the i-th archive text feature in the k-th archive image data exceeds the set threshold. The initial value update unit uses the increased integrity value as the initial integrity for the next demand image probability calculation of that text feature. The output of the threshold comparison unit is connected to the input of the integrity increase unit, and the output of the integrity increase unit is connected to the input of the initial value update unit.
[0025] Compared with the prior art, the beneficial effects of the present invention are:
[0026] 1. A dynamic completeness assessment model based on historical data is introduced. The model calculates the probability of the appearance of the required image corresponding to the text feature and compares it with a threshold to achieve adjustment. Unlike the fixed rule assessment in the existing technology, this invention combines the probability of the appearance of the corresponding image in the historical data of similar archives to dynamically quantify the completeness.
[0027] 2. By using a deep learning-driven automated preprocessing workflow, combined with a CNN model, image noise reduction, correction, and contrast enhancement are achieved. Unlike traditional manual preprocessing methods, this invention reduces the preprocessing time for a single archive image. Attached Figure Description
[0028] Figure 1 This is a flowchart illustrating a deep learning-based method for managing archival image data according to the present invention.
[0029] Figure 2 This is a schematic diagram of the structure of a deep learning-based archival image data management system according to the present invention. Detailed Implementation
[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0031] In the embodiment: such as Figures 1-2 As shown, the present invention provides a technical solution, a method for managing archival image data based on deep learning, the method comprising the following steps:
[0032] Using a collection of archival images as input, OCR technology is used to recognize each image and extract its feature set;
[0033] Keyword monitoring is conducted based on the characteristics of archival texts to verify the first and second elements in the archives;
[0034] When the archival text feature tki appears in the archival image, it is determined whether the first element exists. If it does not exist, the image completeness is adjusted according to the preset rules. If it exists, the probability of its occurrence is calculated based on the frequency of the corresponding required image when the text feature appears in the historical data. Here, tki represents the i-th archival text feature in the k-th archival image data in the archival image dataset.
[0035] If the calculated probability is compared with the set probability threshold, the completeness of the archive image is increased by a set percentage, and the adjusted completeness is used as the initial value for calculating the probability of the next required image for that text feature.
[0036] Using a collection of archival images as input, OCR technology is used to recognize each image and extract its feature set. The specific steps include:
[0037] Based on OCR technology, the archival image data set I={I1,I2,...,Im} is used as input. OCR recognition is performed on any archival image Ik to extract archival text features, denoted as Tk={tk1,tk2,...,ktn}. Here, I1,I2,...,Im represent the 1st, 2nd,...,mth archival images in the archival image data set, respectively, m represents the total number of archival images, Ik represents the kth archival image in the archival image data set, tk1,tk2,...,ktn represent the 1st, 2nd,...,nth archival text features in the kth archival image in the archival image data set, respectively, and n represents the number of archival text features in each archival image.
[0038] Specifically, a company digitized 1,000 contract files. The contract images included scanned copies (PDF format) and handwritten annotation pages (JPG format). The average size of a single page image was 2.5MB. Among them, 30% were scanned copies of yellowed paper, and 20% had blurry seals.
[0039] Of the 1000 archival images, 300 are contract type 1, 300 are contract type 2, and 400 are contract type 3. OCR recognition is performed on any archival image to extract the archival text features.
[0040] U-Net convolutional neural network is used for noise reduction. For yellowed paper images, the model automatically adjusts the grayscale mapping parameters to improve the contrast between text and background.
[0041] Images with a tilt of more than 5° (280 images in total) were corrected using a perspective transformation algorithm.
[0042] Image denoising, correction, and contrast enhancement based on CNN models achieve multi-dimensional technical optimization: Through convolutional neural network models such as U-Net, noise such as stains and scratches in scanned documents are intelligently identified and filtered out, reducing image noise interference rate; by using CNN combined with perspective transformation algorithm, text areas tilted by more than 5° are automatically corrected, shortening the correction time for a single image; for special media such as yellowed paper and thermal printing, CNN adaptively adjusts the grayscale mapping curve to improve the contrast between text and background, thus reducing the OCR recognition error rate.
[0043] Keyword monitoring based on archival text features is used to verify the first and second elements in the archives. Specific steps include:
[0044] Keyword monitoring is performed based on the extracted archival text features to verify the first and second elements in the archive. The first element represents the necessary image requirement when the archival text feature tki appears, and the second element represents the optional image requirement when the archival text feature tki appears. The set of image requirements required by the archival text feature tki is P={p1,p2,...,pq}, where p1,p2,...,pq represent the 1st, 2nd,...,qth required images when the archival text feature tki appears.
[0045] Specifically, for Contract I5, the specific text features include necessary requirement feature 1, necessary requirement feature 2, and necessary requirement feature 3, as well as optional requirement feature 4.
[0046] When the archival text feature tki appears in the archival image, it is determined whether the first element exists. If it does not exist, the image completeness is adjusted according to a preset rule; if it does exist, the probability of its occurrence is calculated based on the frequency of the corresponding required image when the text feature appears in historical data. The specific steps include:
[0047] When the archive image Ik contains the archive text feature tki, determine whether the archive image Ik has the first element. If the archive image Ik does not have the first element, then adjust the integrity of the archive image Ik to 0.
[0048] When the first element exists in the archival image Ik, the completeness of the archival image Ik is adjusted to 50%. Then, the probability of the archival text feature tki appearing in the required image pj is determined. The probability of the archival text feature tki appearing in the required image pj is defined as follows: P(pj|tki) = the number of times image pj exists when tki appears in the historical data / the total number of times tki appears in the historical data; where pj represents the j-th required image when the archival text feature tki appears.
[0049] Specifically, due to the existence of essential elements, the initial completeness of I5 is adjusted to 50% according to the rules; for essential element feature 1, the probability of its corresponding required image p1 is calculated to be 84% based on historical data; the preset probability threshold is 70%, and 84% exceeds the threshold, so the completeness is increased by the set range a=10%.
[0050] The calculated probability is compared with a set probability threshold. If the probability exceeds the threshold, the image completeness is increased by a set percentage. The adjusted completeness is then used as the initial value for calculating the probability of the next required image for that text feature. The specific steps include:
[0051] Based on historical data, a probability threshold is set for the occurrence of the required image pj using the calculated archival text feature tki. The calculated probability of the required image pj using the archival text feature tki is compared with the set probability threshold. When the calculated probability of the required image pj using the archival text feature tki exceeds the set threshold, the completeness of the archival image Ik is increased by a percentage point each time. The adjusted completeness of the archival image Ik is used as the initial completeness of the archival image Ik adjusted according to the above content, where a represents the percentage increase when the completeness of the archival image is increased, and pj+1 represents the (j+1)th required image when the archival text feature tki appears.
[0052] A deep learning-based archival image data management system includes a feature extraction module, an element verification module, a completeness evaluation module, and a completeness adjustment module. The feature extraction module takes an archival image set as input and uses OCR technology to recognize each image and extract its feature set. The element verification module performs keyword monitoring based on archival text features to verify the first and second elements in the archive. The completeness evaluation module determines whether the first element exists when an archival text feature tki appears in an archival image. If not, the image completeness is adjusted according to a preset rule; if it exists, the probability of its occurrence is calculated based on the frequency of occurrence of the corresponding required image in historical data, where tki represents the i-th archival text feature in the k-th archival image data set. The completeness adjustment module compares the calculated probability with a set probability threshold. If the threshold is exceeded, the archival image completeness is increased by a set percentage, and the adjusted completeness is used as the initial value for calculating the probability of the next required image for that text feature.
[0053] The feature extraction module includes an OCR recognition unit and a feature structuring unit. The OCR recognition unit is used to extract archival text features from any archival image by taking the archival image dataset as input based on OCR technology. The feature structuring unit is used to parse the OCR-recognized text data into a structured feature set. The output of the OCR recognition unit is connected to the input of the feature structuring unit, and the output of the feature structuring unit is connected to the input of the element verification module.
[0054] The element verification module includes a keyword monitoring unit, a first element verification unit, and a second element verification unit. The keyword monitoring unit is used to perform keyword retrieval on the extracted feature set and match it with a preset business keyword library. The first element verification unit is used to determine the necessary requirements for the image when archival text features are present. The second element verification unit is used to determine the optional requirements for the image when archival text features are present. The output of the keyword monitoring unit is connected to the input of the first element verification unit and the input of the second element verification unit. The output of the element verification module is connected to the input of the integrity evaluation module.
[0055] The completeness assessment module includes a stage division strategy unit and a demand image probability calculation unit. The stage division strategy unit is used to set an initial completeness based on the first element verification result. The demand image probability calculation unit is used to divide the number of times image pj appears when the i-th archival text feature appears in the k-th archival image data in the historical data by the total number of times the i-th archival text feature appears in the k-th archival image data in the historical data, where pj represents the j-th demand image when the i-th archival text feature appears in the k-th archival image data. The output of the stage division strategy unit is connected to the input of the demand image probability calculation unit, and the output of the demand image probability calculation unit is connected to the input of the completeness adjustment module.
[0056] The integrity adjustment module includes a threshold comparison unit, an integrity increase unit, and an initial value update unit. The threshold comparison unit compares the probability value output by the demand image probability calculation unit with a preset probability threshold. The integrity increase unit increases the integrity of the archive image when the calculated probability that the i-th archive text feature in the k-th archive image data exceeds the set threshold. The initial value update unit uses the increased integrity value as the initial integrity for the next demand image probability calculation of that text feature. The output of the threshold comparison unit is connected to the input of the integrity increase unit, and the output of the integrity increase unit is connected to the input of the initial value update unit.
[0057] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.
Claims
1. A method for managing archival image data based on deep learning, characterized in that: The method includes the following steps: Using a collection of archival images as input, OCR technology is used to recognize each image and extract its feature set; Keyword monitoring based on archival text features is used to verify the first and second elements in the archives. Specific steps include: Keyword monitoring is performed based on the extracted archival text features to verify the first and second elements in the archive. The first element represents the necessary image requirement when the archival text feature tki appears, and the second element represents the optional image requirement when the archival text feature tki appears. The set of image requirements required by the archival text feature tki is P={p1,p2,...,pq}, where p1,p2,...,pq represent the 1st, 2nd,...,qth required images when the archival text feature tki appears. When the archival text feature tki appears in the archival image, it is determined whether the first element exists. If it does not exist, the image completeness is adjusted according to the preset rules. If it exists, the probability of its occurrence is calculated based on the frequency of the corresponding required image when the text feature appears in historical data. Here, tki represents the i-th archival text feature in the k-th archival image data set. The specific steps include: When the archive image Ik contains the archive text feature tki, determine whether the archive image Ik has the first element. If the archive image Ik does not have the first element, adjust the integrity of the archive image Ik to 0. When the first element exists in the archival image Ik, the completeness of the archival image Ik is adjusted to 50%. Then, the probability of the archival text feature tki appearing in the required image pj is determined. The probability of the archival text feature tki appearing in the required image pj is defined as follows: P(pj|tki) = the number of times image pj exists when tki appears in the historical data / the total number of times tki appears in the historical data; where pj represents the j-th required image when the archival text feature tki appears. The calculated probability is compared with a set probability threshold. If the probability exceeds the threshold, the image completeness is increased by a set percentage. The adjusted completeness is then used as the initial completeness value for calculating the probability of the next required image for that text feature. The specific steps include: Based on historical data, a probability threshold is set for the occurrence of the required image pj by the calculated archival text feature tki. The calculated probability of the required image pj by the archival text feature tki is compared with the set probability threshold. When the calculated probability of the required image pj by the archival text feature tki exceeds the set threshold, the completeness of the archival image Ik is increased by a percentage point each time. The adjusted completeness of the archival image Ik is used as the initial completeness value for calculating the probability of the next required image pj+1 of the archival text feature tki. Here, a represents the percentage point increase when the completeness of the archival image is increased, and pj+1 represents the (j+1)th required image when the archival text feature tki occurs.
2. The method for managing archival image data based on deep learning according to claim 1, characterized in that: Using a collection of archival images as input, OCR technology is used to recognize each image and extract its feature set. The specific steps include: Based on OCR technology, the archival image data set I={I1,I2,...,Im} is used as input. OCR recognition is performed on any archival image Ik to extract archival text features, denoted as Tk={tk1,tk2,...,ktn}. Here, I1,I2,...,Im represent the 1st, 2nd,...,mth archival images in the archival image data set, respectively, m represents the total number of archival images, Ik represents the kth archival image in the archival image data set, tk1,tk2,...,ktn represent the 1st, 2nd,...,nth archival text features in the kth archival image in the archival image data set, respectively, and n represents the number of archival text features in each archival image.
3. A deep learning-based archival image data management system, applied to the deep learning-based archival image data management method described in any one of claims 1-2, characterized in that: The system includes a feature extraction module, an element verification module, a completeness assessment module, and a completeness adjustment module. The feature extraction module takes a set of archival images as input and uses OCR technology to recognize each image and extract its feature set. The element verification module performs keyword monitoring based on archival text features to verify the first and second elements in the archives. The completeness assessment module determines whether the first element exists when the archival text feature tki appears in an archival image. If not, the image completeness is adjusted according to a preset rule. If it exists, the probability of its occurrence is calculated based on the frequency of occurrence of the corresponding required image in historical data, where tki represents the i-th archival text feature in the k-th archival image data set. The completeness adjustment module compares the calculated probability with a set probability threshold. If the threshold is exceeded, the archival image completeness is increased by a set percentage, and the adjusted completeness is used as the initial completeness value for calculating the probability of the next required image for that text feature.
4. The archival image data management system based on deep learning according to claim 3, characterized in that: The feature extraction module includes an OCR recognition unit and a feature structuring unit. The OCR recognition unit is used to extract archival text features from any archival image by taking the archival image dataset as input based on OCR technology. The feature structuring unit is used to parse the OCR-recognized text data into a structured feature set. The output of the OCR recognition unit is connected to the input of the feature structuring unit, and the output of the feature structuring unit is connected to the input of the element verification module.
5. The archival image data management system based on deep learning according to claim 4, characterized in that: The element verification module includes a keyword monitoring unit, a first element verification unit, and a second element verification unit. The keyword monitoring unit is used to perform keyword retrieval on the extracted feature set and match it with a preset business keyword library. The first element verification unit is used to determine the necessary requirements for the image when archival text features are present. The second element verification unit is used to determine the optional requirements for the image when archival text features are present. The output of the keyword monitoring unit is connected to the input of the first element verification unit and the input of the second element verification unit. The output of the element verification module is connected to the input of the integrity evaluation module.
6. The archival image data management system based on deep learning according to claim 5, characterized in that: The completeness assessment module includes a stage division strategy unit and a demand image probability calculation unit. The stage division strategy unit is used to set an initial completeness value based on the first element verification result. The demand image probability calculation unit is used to divide the number of times image pj appears when the i-th archival text feature appears in the k-th archival image data in the historical data by the total number of times the i-th archival text feature appears in the k-th archival image data in the historical data, where pj represents the j-th demand image when the i-th archival text feature appears in the k-th archival image data. The output of the stage division strategy unit is connected to the input of the demand image probability calculation unit, and the output of the demand image probability calculation unit is connected to the input of the completeness adjustment module.
7. The archival image data management system based on deep learning according to claim 6, characterized in that: The integrity adjustment module includes a threshold comparison unit, an integrity adjustment unit, and an initial value update unit. The threshold comparison unit is used to compare the probability value output by the required image probability calculation unit with a preset probability threshold. The integrity adjustment unit is used to adjust the integrity of the archive image when the probability of the i-th archive text feature appearing in the k-th archive image data in the calculated archive image data set exceeds a set threshold. The initial value update unit is used to use the adjusted integrity value as the initial integrity value for calculating the probability of the next required image for the text feature. The output of the threshold comparison unit is connected to the input of the integrity adjustment unit, and the output of the integrity adjustment unit is connected to the input of the initial value update unit.