AI-based archive content analysis and intelligent labeling system

Abstract

The invention belongs to the technical field of intelligent identification, and provides an AI-based archive content analysis and intelligent labeling system, which is characterized in that based on analysis of an abnormal archive, an area where non-coincident content is located is marked in the analysis of the abnormal archive to obtain a non-coincident area, and meanwhile, a damaged area is marked in the analysis of the abnormal archive to obtain a damaged area; the non-overlapping area and the damaged area are compared and analyzed to obtain an overlapping influence value, the overlapping influence value is compared with an overlapping influence threshold value, if the overlapping influence value is larger than the overlapping influence threshold value, archive analysis abnormity is related to archive damage, and therefore the overlapping degree of the damaged area and the non-overlapping area is high, and archive analysis abnormity is related to archive damage. The damaged content of the paper archive has great influence on the precision of image recognition and extraction.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent recognition technology, and specifically is an archive content parsing and intelligent tagging system based on AI. Background Art

[0002] OCR technology is an image recognition technology that can convert text information in images into editable text data. The core of this technology is to use computer vision and artificial intelligence algorithms to automatically identify and extract text from printed materials, handwritten text or screenshots, making it part of the electronic text for easy storage, retrieval and processing.

[0003] In the existing OCR technology, it is possible to automatically identify and extract text from printed materials, handwritten texts, or screenshots, parse and extract the contents of paper archives, and annotate them through electronic text based on the parsed and extracted contents. However, after parsing and extracting paper archives using OCR technology, the identified extracted images are different from the contents on the paper archives. Therefore, this will affect the annotation of the paper archives in the later stages. Now, through an AI-based archive content parsing and intelligent annotation system, it is verified that the abnormality of the paper archive image is not caused by damage to the paper archive itself.

[0004] To this end, the present invention provides an archive content parsing and intelligent tagging system based on AI. Summary of the invention

[0005] In order to make up for the deficiencies of the prior art, at least one technical problem raised in the background technology is solved.

[0006] The technical solution adopted by the present invention to solve the technical problem is: an AI-based archive content analysis and intelligent annotation system, comprising:

[0007] Data analysis module: extract images from paper archives to obtain archive extraction images, compare and analyze the archive extraction images, mark paper archives according to the comparison and analysis results, and obtain analysis of abnormal archives and analysis of normal archives;

[0008] File evaluation module: Based on parsing abnormal files, mark the area where non-overlapping content is located in the parsing abnormal files to obtain non-overlapping areas, and mark the damaged areas in the parsing abnormal files to obtain damaged areas. Compare and analyze the non-overlapping areas and damaged areas to obtain the overlap impact value, and compare the overlap impact value with the overlap impact threshold. If the overlap impact value is greater than the overlap impact threshold, it means that the file parsing abnormality is related to file damage, and an impact signal is generated;

[0009] Abnormal analysis module: Based on the impact signal, the paper archives are optimized and processed, and the images of the optimized paper archives are extracted again to obtain the secondary extracted images of the archives. The secondary extracted images of the archives are compared and analyzed, and according to the comparison and analysis results, it is judged whether the analysis is abnormal.

[0010] As a further solution of the present invention, the method for obtaining the abnormal image is as follows:

[0011] If the content in the recognized image completely overlaps with the page content in the corresponding paper file, the recognized image is marked as a normal image;

[0012] If the content in the recognized image does not completely overlap with the page content in the corresponding paper file, the recognized image will be marked as a parsing abnormality image, and the content in the parsing abnormality image that does not completely overlap with the page content in the corresponding paper file will be marked as non-overlapping content.

[0013] As a further solution of the present invention, the method for obtaining the average value of non-overlapping length is:

[0014] The number of abnormal images is analyzed statistically, and the ratio is processed with the total number of recognition images contained in the recognition image set to obtain the abnormal number ratio.

[0015] As a further solution of the present invention, the non-overlapping values ​​are obtained by:

[0016] Obtain characters of non-overlapping content corresponding to the abnormal parsing images, add and average the characters of non-overlapping content corresponding to all abnormal parsing images, and obtain the average value of non-overlapping length;

[0017] The non-overlapping value is obtained by calculating the ratio of the average non-overlapping length to the total length of the paper archive content.

[0018] As a further solution of the present invention, the method for obtaining the parsed abnormal value is:

[0019] The non-coincidence values ​​and the anomaly number ratio are summed to obtain the analytical anomaly value.

[0020] As a further solution of the present invention, the analytical abnormality value is compared with the analytical abnormality threshold, and the comparison process is as follows:

[0021] If the parsing anomaly value is greater than the parsing anomaly threshold, the paper file is marked as a parsing anomaly file;

[0022] If the parsing anomaly is less than or equal to the parsing anomaly threshold, the paper file is marked as a parsing normal file.

[0023] As a further solution of the present invention, the method of distinguishing overlapping damaged pages from non-overlapping damaged pages is as follows:

[0024] Marking the area where the non-overlapping content is located in the parsed abnormal file to obtain the non-overlapping area, marking the file page with the non-overlapping area as an abnormal file page, and integrating the page numbers of all abnormal file pages into an abnormal page number group;

[0025] Marking the damaged area in the parsed abnormal file to obtain the damaged area, and marking the file page with the damaged area as a damaged page;

[0026] Get the damaged page number. If the damaged page number exists in the abnormal page number group, mark the damaged page as a coincident damaged page.

[0027] If the damaged page number does not exist in the abnormal page number group, the damaged page is marked as a non-overlapping damaged page.

[0028] As a further solution of the present invention, the overlap influence value is obtained by:

[0029] The obtained overlap damage number ratio CH, double overlap damage number ratio SH and double overlap area ratio SM are processed by the formula: The overlap influence value CY is calculated.

[0030] As a further solution of the present invention, the method of obtaining the overlap damage quantity ratio is as follows:

[0031] Count the number of overlapping damaged pages and compare it with the number of abnormal archive pages to get the overlapping damage ratio, which is marked as CH;

[0032] The double combined damage ratio is obtained as follows:

[0033] Based on the overlapped damaged page, if the non-overlapped area and the damaged area overlap in the overlapped damaged page, the overlapped damaged page will be marked as a double overlapped damaged page;

[0034] If the non-overlapping area and the damaged area do not overlap in the overlapping damaged page, no operation is performed;

[0035] Count the number of double-damaged pages and compare it with the number of overlapped damaged pages to get the double-damaged number ratio, which is marked as SH.

[0036] The double composite area ratio is obtained as follows:

[0037] Based on the double-overlapping damaged pages, the areas of the overlapping areas between the non-overlapping areas and the damaged areas in all double-overlapping damaged pages are summed and averaged to obtain the mean overlapping area. The mean overlapping area is ratioed with the total area of ​​the non-overlapping areas to obtain the double-overlapping area ratio, which is marked as SM.

[0038] As a further solution of the present invention, a comparison analysis is performed on the secondary extracted images of the archives, and according to the comparison analysis results, whether the analysis is abnormal is determined, and the process is as follows:

[0039] The optimized paper archives are image extracted by OCR technology to obtain the archive secondary extraction image, wherein the archive secondary extraction image includes a secondary recognition image set, and the secondary recognition image set is analyzed and processed to obtain the analytical abnormal value again;

[0040] The retrieved parsing anomaly value is compared with the parsing anomaly threshold as follows:

[0041] If the re-obtained parsing anomaly value is greater than the parsing anomaly threshold, the file parsing is abnormal;

[0042] If the retrieved parsing anomaly value is less than the parsing anomaly threshold, the file parsing is normal.

[0043] The beneficial effects of the present invention are as follows:

[0044] (1) The present invention extracts images from paper archives to obtain archive extraction images, compares and analyzes the archive extraction images, and marks paper archives according to the comparison and analysis results to obtain abnormal analysis archives and normal analysis archives. The overall recognition accuracy and completeness of paper archives after image recognition are reflected by the ratio of the number of abnormal analysis images and the relative length of the erroneous contents in these abnormal analysis images.

[0045] (2) The present invention is based on parsing abnormal files. In the parsing abnormal files, the area where the non-overlapping content is located is marked to obtain the non-overlapping area. At the same time, the damaged area is marked in the parsing abnormal files to obtain the damaged area. The non-overlapping area and the damaged area are compared and analyzed to obtain the overlap influence value. The overlap influence value is compared with the overlap influence threshold. If the overlap influence value is greater than the overlap influence threshold, it means that the file parsing abnormality is related to the file damage, which means that the damage area and the non-overlapping area overlap to a high degree, and the damaged content of the paper file has a great influence on the accuracy of image recognition extraction;

[0046] (3) The present invention optimizes the paper archive based on the influencing signal, and extracts the image of the optimized paper archive again to obtain a secondary extracted image of the archive, and compares and analyzes the secondary extracted image of the archive. According to the comparison and analysis result, if the re-obtained analytical anomaly value is greater than the analytical anomaly threshold, it is verified that the abnormality of the paper archive image is not caused by damage to the paper archive itself. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The present invention will be further described below in conjunction with the accompanying drawings.

[0048] Figure 1It is a module diagram of an archive content analysis and intelligent annotation system based on AI of the present invention;

[0049] Figure 2 It is a flowchart of the steps of analyzing archive content in an archive content analysis and intelligent tagging system based on AI of the present invention. DETAILED DESCRIPTION

[0050] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0051] Example 1

[0052] like Figure 1 - Figure 2 As shown, an AI-based archive content parsing and intelligent annotation system according to an embodiment of the present invention includes:

[0053] Data analysis module: extract images from paper archives to obtain archive extraction images, compare and analyze the archive extraction images, mark paper archives according to the comparison and analysis results, and obtain analysis of abnormal archives and analysis of normal archives;

[0054] In some embodiments, an image is extracted from a paper archive by using OCR technology to obtain an archive extracted image, wherein the archive extracted image includes a recognition image set, and the recognition image set is analyzed and processed, and the process is as follows:

[0055] It should be noted that the identification image set includes multiple identification images, wherein one identification image corresponds to one page of the paper file;

[0056] Based on any one recognition image in the recognition image set;

[0057] Compare the content in the recognized image with the corresponding paper file content;

[0058] If the content in the recognized image completely overlaps with the page content in the corresponding paper file, the recognized image is marked as a normal image;

[0059] If the content in the recognized image does not completely overlap with the page content in the corresponding paper file, the recognized image is marked as a parsing abnormal image, and the content in the parsing abnormal image that does not completely overlap with the page content in the corresponding paper file is marked as non-overlapping content;

[0060] The number of abnormal images is analyzed statistically, and the ratio is processed with the total number of recognition images contained in the recognition image set to obtain the abnormal number ratio;

[0061] It can be understood that the meaning of the abnormal number ratio is: by analyzing the proportion of the number of abnormal images in the recognition image set, it reflects the consistency between the recognition image set and the paper archive content. Specifically, the larger the value, the greater the difference between the recognition image and the paper archive content. On the contrary, the smaller the value, the smaller the difference between the recognition image and the paper archive content.

[0062] Obtain characters of non-overlapping content corresponding to the abnormal parsing images, add and average the characters of non-overlapping content corresponding to all abnormal parsing images, and obtain the average value of non-overlapping length;

[0063] The non-overlapping value is obtained by calculating the ratio of the average non-overlapping length to the total length of the paper archive content;

[0064] The non-coincidence value and the anomaly number ratio are summed to obtain the analytical anomaly value;

[0065] It can be understood that: the meaning of the analysis outlier value is: through the ratio of the number of analysis outlier images and the relative length of the erroneous content in these analysis outlier images, it reflects the overall recognition accuracy and completeness of the paper archives after image recognition. Specifically, the larger this value is, the larger the proportion of the analysis outlier images in the recognition image set is, and the content of these analysis outlier images deviates greatly from the content in the paper archives. On the contrary, the smaller this value is, the smaller the proportion of the analysis outlier images in the recognition image set is, and the content of these analysis outlier images deviates less from the content in the paper archives.

[0066] The analysis anomaly value is compared with the analysis anomaly threshold. The comparison process is as follows:

[0067] If the parsing anomaly value is greater than the parsing anomaly threshold, it means that in the recognition image set, the number of parsing anomaly images accounts for a large proportion, and the content of these parsing anomaly images deviates greatly from the content of the paper archives, then the paper archives are marked as parsing anomaly archives;

[0068] If the parsing anomaly is less than or equal to the parsing anomaly threshold, it means that in the recognition image set, the number of parsing anomaly images accounts for a small proportion, and the content of these parsing anomaly images is less deviated from the content in the paper archives, then the paper archives are marked as parsing normal archives;

[0069] It should be noted that the analytical anomaly threshold is obtained by personnel in this technical field through a large amount of experimental data;

[0070] The specific implementation scheme of the embodiment of the present invention is as follows: image extraction is performed on the content of paper archives to obtain archive extraction images, comparison and analysis are performed on the archive extraction images, and the paper archives are marked according to the comparison and analysis results to obtain parsed abnormal archives and parsed normal archives, and the overall recognition accuracy and completeness of the paper archives after image recognition are reflected by the ratio of the number of parsed abnormal images and the relative length of the erroneous contents in these parsed abnormal images.

[0071] Example 2

[0072] like Figure 1 - Figure 2 As shown, based on Example 1, an AI-based archive content parsing and intelligent annotation system according to an embodiment of the present invention includes:

[0073] File evaluation module: Based on parsing abnormal files, mark the area where non-overlapping content is located in the parsing abnormal files to obtain non-overlapping areas, and mark the damaged areas in the parsing abnormal files to obtain damaged areas. Compare and analyze the non-overlapping areas and damaged areas to obtain the overlap impact value, and compare the overlap impact value with the overlap impact threshold. If the overlap impact value is greater than the overlap impact threshold, it means that the file parsing abnormality is related to file damage, and an impact signal is generated;

[0074] Specifically, the area where the non-overlapping content is located is marked in the parsed abnormal file to obtain the non-overlapping area, the file page with the non-overlapping area is marked as an abnormal file page, and the page numbers of all abnormal file pages are integrated into an abnormal page number group;

[0075] Marking the damaged area in the parsed abnormal file to obtain the damaged area, and marking the file page with the damaged area as a damaged page;

[0076] Get the damaged page number. If the damaged page number exists in the abnormal page number group, mark the damaged page as a coincident damaged page.

[0077] If the damaged page number does not exist in the abnormal page number group, the damaged page is marked as a non-overlapping damaged page;

[0078] Count the number of overlapping damaged pages and compare it with the number of abnormal archive pages to get the overlapping damage ratio, which is marked as CH;

[0079] It should be noted that the meaning of the overlapping damage number ratio is: the proportion of pages in the abnormal file pages that are marked as damaged pages and whose numbers also exist in the abnormal page number group (i.e., overlapping damaged pages), which reflects the overlap between damage and parsing anomalies in the parsing abnormal file. Specifically, the larger this value is, the more overlaps there are between the abnormal file page and the damaged page. Conversely, the smaller this value is, the less overlaps there are between the abnormal file page and the damaged page.

[0080] Based on the overlapped damaged page, if the non-overlapped area and the damaged area overlap in the overlapped damaged page, the overlapped damaged page will be marked as a double overlapped damaged page;

[0081] If the non-overlapping area and the damaged area do not overlap in the overlapping damaged page, no operation is performed;

[0082] Count the number of double-damaged pages and compare it with the number of overlapped damaged pages to get the double-damaged number ratio, which is marked as SH.

[0083] It should be noted that the meaning of the double overlap damage ratio is: it reflects the proportion of pages where the non-overlap area overlaps with the damaged area in the overlap damaged pages. Specifically, the larger the value, the higher the overlap between the non-overlap area and the damaged area. On the contrary, the smaller the value, the lower the overlap between the non-overlap area and the damaged area.

[0084] Based on the double overlap damaged pages, the areas of the overlapped areas between the non-overlapped areas and the damaged areas in all double overlap damaged pages are summed and averaged to obtain the mean overlapped area, and the mean overlapped area is ratioed to the total area of ​​the non-overlapped areas to obtain the double overlapped area ratio, which is marked as SM.

[0085] Among them, the total area of ​​the non-overlapping area is the sum of the non-overlapping areas in all abnormal file pages;

[0086] The obtained overlap damage number ratio CH, double overlap damage number ratio SH and double overlap area ratio SM are processed by the formula: The overlap influence value CY is calculated;

[0087] It can be understood that the meaning of the overlap impact value CY is: the influence degree of the overlap between the damaged area and the non-overlapping area (i.e., the specific area marked when analyzing the anomaly) in the anomaly file. Specifically, the larger the value, the higher the overlap degree between the damaged area and the non-overlapping area. On the contrary, the smaller the value, the lower the overlap degree between the damaged area and the non-overlapping area.

[0088] Compare the overlap impact value CY with the overlap impact threshold;

[0089] If the overlap impact value CY is greater than the overlap impact threshold, it means that the damaged area has a high degree of overlap with the non-overlap area, and an impact signal is generated;

[0090] If the overlap impact value CY is less than or equal to the overlap impact threshold, it means that the overlap between the damaged area and the non-overlap area is low, and a non-impact signal is generated;

[0091] It should be noted that the overlap impact threshold is obtained by those skilled in the art through experimental data;

[0092] The specific implementation scheme of the embodiment of the present invention is: based on parsing the abnormal files, marking the area where the non-overlapping content is located in the abnormal files to obtain the non-overlapping area, and marking the damaged area in the abnormal files to obtain the damaged area, comparing and analyzing the non-overlapping area and the damaged area to obtain the overlap impact value, comparing the overlap impact value with the overlap impact threshold, if the overlap impact value is greater than the overlap impact threshold, it means that the file parsing anomaly is related to the file damage, which means that the damaged area has a high degree of overlap with the non-overlapping area, and the damaged content of the paper file has a great influence on the accuracy of image recognition extraction.

[0093] Example 3

[0094] like Figure 1-Figure 2 Based on Example 1 and Example 2, an AI-based archive content parsing and intelligent annotation system described in an embodiment of the present invention includes:

[0095] Abnormal analysis module: Based on the impact signal, the paper archive is optimized and processed again, and the image of the optimized paper archive is extracted again to obtain the secondary extracted image of the archive, and the secondary extracted image of the archive is compared and analyzed. According to the comparison and analysis results, it is judged whether the analysis is abnormal;

[0096] Among them, the method of optimizing the paper archive is: deleting the overlapping and damaged pages in the paper archive to obtain the optimized paper archive;

[0097] Specifically, the optimized paper archive is subjected to image extraction through OCR technology to obtain a secondary extracted image of the archive, wherein the secondary extracted image of the archive includes a secondary recognition image set, and the secondary recognition image set is analyzed and processed to obtain a parsed outlier value;

[0098] The analysis and processing process of the secondary recognition image set is the same as the analysis and processing process of the recognition image set;

[0099] The retrieved parsing anomaly value is compared with the parsing anomaly threshold as follows:

[0100] If the re-obtained parsing anomaly value is greater than the parsing anomaly threshold, the file parsing is abnormal and an abnormal signal is generated;

[0101] Based on abnormal signals, conduct investigation and analysis on file analysis, including but not limited to: OCR recognition accuracy, file content comparison accuracy, etc.;

[0102] If the re-obtained parsing anomaly value is less than the parsing anomaly threshold, the file parsing is normal;

[0103] The specific implementation method of the embodiment of the present invention is as follows: based on the influencing signal, the paper archive is optimized, and the image of the optimized paper archive is extracted again to obtain a secondary extracted image of the archive, and the secondary extracted image of the archive is compared and analyzed. According to the comparison and analysis result, if the re-obtained analysis anomaly value is greater than the analysis anomaly threshold, it is verified that the abnormality of the paper archive image is not caused by damage to the paper archive itself.

[0104] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. An AI-based archive content analysis and intelligent annotation system, characterized by: include: Data analysis module: extract images from paper archives to obtain archive extraction images, compare and analyze the archive extraction images, mark paper archives according to the comparison and analysis results, and obtain analysis of abnormal archives and analysis of normal archives; File evaluation module: Based on parsing abnormal files, mark the area where non-overlapping content is located in the parsing abnormal files to obtain non-overlapping areas, and mark the damaged areas in the parsing abnormal files to obtain damaged areas. Compare and analyze the non-overlapping areas and damaged areas to obtain the overlap impact value, and compare the overlap impact value with the overlap impact threshold. If the overlap impact value is greater than the overlap impact threshold, it means that the file parsing abnormality is related to file damage, and an impact signal is generated; Abnormal analysis module: Based on the impact signal, the paper archives are optimized and processed, and the images of the optimized paper archives are extracted again to obtain the secondary extracted images of the archives. The secondary extracted images of the archives are compared and analyzed, and according to the comparison and analysis results, it is judged whether the analysis is abnormal.

2. The AI-based archive content analysis and intelligent annotation system according to claim 1, characterized in that: The method for obtaining the abnormal image is as follows: If the content in the recognized image completely overlaps with the page content in the corresponding paper file, the recognized image is marked as a normal image; If the content in the recognized image does not completely overlap with the page content in the corresponding paper file, the recognized image will be marked as a parsing abnormality image, and the content in the parsing abnormality image that does not completely overlap with the page content in the corresponding paper file will be marked as non-overlapping content.

3. The AI-based archive content analysis and intelligent annotation system according to claim 2, characterized in that: The method for obtaining the mean value of non-overlapping length is: The number of abnormal images is analyzed statistically, and the ratio is processed with the total number of recognition images contained in the recognition image set to obtain the abnormal number ratio.

4. The AI-based archive content analysis and intelligent annotation system according to claim 2, characterized in that: The way to obtain non-coincidence values ​​is: Obtain characters of non-overlapping content corresponding to the abnormal parsing images, add and average the characters of non-overlapping content corresponding to all abnormal parsing images, and obtain the average value of non-overlapping length; The non-overlapping value is obtained by calculating the ratio of the average non-overlapping length to the total length of the paper archive content.

5. The AI-based archive content analysis and intelligent annotation system according to claim 2, characterized in that: The way to obtain parsed outliers is as follows: The non-coincidence values ​​and the anomaly number ratio are summed to obtain the analytical anomaly value.

6. The AI-based archive content analysis and intelligent annotation system according to claim 5, characterized in that: The analysis anomaly value is compared with the analysis anomaly threshold. The comparison process is as follows: If the parsing anomaly value is greater than the parsing anomaly threshold, the paper file is marked as a parsing anomaly file; If the parsing anomaly is less than or equal to the parsing anomaly threshold, the paper file is marked as a parsing normal file.

7. The AI-based archive content analysis and intelligent annotation system according to claim 1, characterized in that: The distinction between overlapping damaged pages and non-overlapping damaged pages is as follows: Marking the area where the non-overlapping content is located in the parsed abnormal file to obtain the non-overlapping area, marking the file page with the non-overlapping area as an abnormal file page, and integrating the page numbers of all abnormal file pages into an abnormal page number group; Marking the damaged area in the parsed abnormal file to obtain the damaged area, and marking the file page with the damaged area as a damaged page; Get the damaged page number. If the damaged page number exists in the abnormal page number group, mark the damaged page as a coincident damaged page. If the damaged page number does not exist in the abnormal page number group, the damaged page is marked as a non-overlapping damaged page.

8. The AI-based archive content analysis and intelligent annotation system according to claim 1, characterized in that: The overlap influence value is obtained as follows: The obtained overlap damage number ratio CH, double overlap damage number ratio SH and double overlap area ratio SM are processed by the formula: The overlap influence value CY is calculated.

9. The AI-based archive content analysis and intelligent annotation system according to claim 7, characterized in that: The overlap damage ratio is obtained as follows: Count the number of overlapping damaged pages and compare it with the number of abnormal archive pages to get the overlapping damage ratio, which is marked as CH; The double combined damage ratio is obtained as follows: Based on the overlapped damaged page, if the non-overlapped area and the damaged area overlap in the overlapped damaged page, the overlapped damaged page will be marked as a double overlapped damaged page; If the non-overlapping area and the damaged area do not overlap in the overlapping damaged page, no operation is performed; Count the number of double-damaged pages and compare it with the number of overlapped damaged pages to get the double-damaged number ratio, which is marked as SH. The double composite area ratio is obtained as follows: Based on the double-overlapping damaged pages, the areas of the overlapping areas between the non-overlapping areas and the damaged areas in all double-overlapping damaged pages are summed and averaged to obtain the mean overlapping area. The mean overlapping area is ratioed with the total area of ​​the non-overlapping areas to obtain the double-overlapping area ratio, which is marked as SM.

10. The AI-based archive content analysis and intelligent annotation system according to claim 1, characterized in that: Compare and analyze the images extracted from the archives for the second time, and judge whether the analysis is abnormal based on the comparison and analysis results. The process is as follows: The optimized paper archives are image extracted by OCR technology to obtain the archive secondary extraction image, wherein the archive secondary extraction image includes a secondary recognition image set, and the secondary recognition image set is analyzed and processed to obtain the analytical abnormal value again; The retrieved parsing anomaly value is compared with the parsing anomaly threshold as follows: If the re-obtained parsing anomaly value is greater than the parsing anomaly threshold, the file parsing is abnormal; If the retrieved parsing anomaly value is less than the parsing anomaly threshold, the file parsing is normal.

Images