Modular classification and intelligent archiving method of business document information

Abstract

The present application relates to the technical field of archive management, and more particularly to a modular classification and intelligent archiving method for business document information. The method comprises the following steps: obtaining business document information; performing layout analysis, field extraction and content structure identification on the business document information to generate document structure feature data; performing modular classification analysis according to the document structure feature data to determine the classification module to which the document belongs; performing archive structure mapping according to the classification module to which the document belongs to establish the association between the document fields and the archive directory nodes; generating a hierarchical archive directory structure according to the association; generating archiving index data according to the hierarchical archive directory structure; and performing intelligent archiving processing on the business document information according to the archiving index data. The present application realizes intelligent archiving and traceable data organization of the document based on archive management technology, and ensures the archiving integrity and structural consistency.

Description

Technical Field

[0001] This invention relates to the field of document management technology, and in particular to a modular classification and intelligent archiving method for business document information. Background Technology

[0002] In current business document management, staff typically rely on manual classification and archiving based on document name, business type, and content information. This method is susceptible to variations in staff experience, inconsistent archiving standards, and differing operational habits, resulting in low classification efficiency, insufficient archiving accuracy, and difficulties in document retrieval. With the application of electronic document management systems, some solutions have begun to store business documents digitally. However, existing technologies largely rely on document names, keywords, or fixed rules for classification, failing to fully utilize the structured features of document layout, field arrangement, and content hierarchy for analysis, leading to low classification accuracy in complex document scenarios. Furthermore, existing archiving technologies generally employ fixed directory storage, lacking an intelligent mapping mechanism between document fields and archive directory nodes. This hinders the establishment of a unified hierarchical archive directory structure and a traceable archiving system, resulting in insufficient document correlation, low retrieval efficiency, and a low level of intelligence in archive management. Summary of the Invention

[0003] Therefore, it is necessary for the present invention to provide a modular classification and intelligent archiving method for business document information in order to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a modular classification and intelligent archiving method for business document information includes the following steps: Step S1: Obtain business document information; perform layout parsing, field extraction, and content structure recognition on the business document information to generate document structure feature data; Step S2: Perform modular classification analysis based on the document structure feature data to determine the classification module to which the document belongs; Step S3: Perform file structure mapping based on the classification module to which the document belongs, so as to establish the relationship between document fields and file directory nodes; generate a hierarchical file directory structure based on the relationship; generate archive index data based on the hierarchical file directory structure; Step S4: Perform intelligent archiving processing on business document information based on the archived index data to form a traceable archive dataset; perform archive integrity verification based on the traceable archive dataset to generate intelligent archiving result data.

[0005] The beneficial effects of this invention are as follows: (1) By performing layout parsing, field extraction and content structure recognition on business document information, and constructing document structure feature data based on field position relationship and content hierarchy relationship, unified parsing of the layout structure of business documents of different formats is realized, the impact of document format differences on classification processing is reduced, and the accuracy and stability of business document structure recognition are improved.

[0006] (2) By calculating the spatial distribution density of fields based on the structural feature data of the document, and performing spatial clustering analysis and module template matching in combination with structural area units, the automatic identification and modular classification of the structural features of the business document layout are realized, which reduces the subjectivity and inconsistency problems in the manual classification process and improves the classification efficiency and accuracy of business documents.

[0007] (3) By establishing the association between document fields and archive directory nodes, and generating a hierarchical archive directory structure and archive index data, intelligent mapping and automatic archiving of business documents and archive directory are realized, which enhances the standardization and uniformity of archive organization structure and improves archive management efficiency.

[0008] (4) By constructing field-level index associations, data traceability link information and archive version association structure, a traceable archive dataset is formed, realizing full-link tracking and historical version management of business document archiving process, improving archive retrieval efficiency, data traceability and intelligent level of archive management. Attached Figure Description

[0009] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart illustrating the steps of a modular classification and intelligent archiving method for business document information according to the present invention. Figure 2 This is a schematic diagram of the process for identifying field relationships at field positions in this invention; Figure 3 This is a schematic diagram of the process architecture of the modular classification and intelligent archiving method for business document information of the present invention. The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0010] The technical method of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention.

[0011] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0012] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0013] To achieve the above objectives, please refer to Figures 1 to 3 This invention provides a modular classification and intelligent archiving method for business document information, the method comprising the following steps: Step S1: Obtain business document information; perform layout parsing, field extraction, and content structure recognition on the business document information to generate document structure feature data; In one embodiment, after acquiring scanned images or electronic document data of business documents, the documents are first enhanced using a bilateral filtering algorithm and an adaptive threshold segmentation algorithm to eliminate background noise and lighting interference. Then, a layout analysis model is used to identify text areas, table areas, stamp areas, and blank areas, and the boundary contours and spatial coordinates of each area are extracted. Next, character recognition processing is performed on the text areas to obtain field content and field position coordinates. Based on the field position coordinates, the spatial distance, alignment, and adjacency relationships between fields are calculated, and a field topology structure is constructed. Furthermore, based on the field topology structure, the hierarchical relationships between the title layer, content layer, and subordinate layers are identified, and region density features, field connection features, and hierarchical structure features are extracted to generate document structure feature data.

[0014] In another embodiment, assuming the input business document image resolution is 3508×2480 pixels, a total of 500 business documents are obtained; after image enhancement, 9200 text areas, 1350 table areas, and 280 stamp areas are identified; 18600 field nodes are extracted, and 72400 field connection relationships are established; the average adjacency distance of the fields is calculated to be 16.8 pixels, and the average alignment deviation angle is 2.3 degrees; finally, 500 sets of document structure feature data are formed, and each set of feature vectors has a dimension of 512.

[0015] Step S2: Perform modular classification analysis based on the document structure feature data to determine the classification module to which the document belongs; In one embodiment, spatial distribution features of fields, regional connectivity features, and hierarchical topological features are extracted based on document structural feature data. First, the document page is divided into regular grid cells, and the number of fields and space occupancy rate in each grid cell are counted. Then, high-density and low-density regions are identified based on the statistical results, and a regional adjacency graph is established. Further, a density clustering algorithm is used to perform cluster analysis on the regional adjacency graph, forming multiple candidate structural modules. Subsequently, the structural similarity between the candidate structural modules and preset module templates is calculated, and the classification module to which the document belongs is determined based on the maximum similarity.

[0016] In another embodiment, it is assumed that each document page is divided into 64 grid cells; 18 high-density regions, 21 medium-density regions, and 25 low-density regions are calculated; 420 region adjacency relationships are established; 7 candidate structural modules are formed through cluster analysis; the similarity of the modules is calculated to be 0.94, 0.88, 0.81, 0.76, 0.69, 0.55, and 0.43, respectively; finally, the structural module with a similarity of 0.94 is selected as the target classification module.

[0017] Step S3: Perform file structure mapping based on the classification module to which the document belongs, so as to establish the relationship between document fields and file directory nodes; generate a hierarchical file directory structure based on the relationship; generate archive index data based on the hierarchical file directory structure; In one embodiment, the corresponding directory topology template is loaded according to the target classification module, and the parent-child connection relationship between directory nodes is extracted. Then, the field nodes in the field topology structure are matched with the directory nodes in the directory topology template, and the structural similarity and spatial mapping relationship between the field nodes and directory nodes are calculated. Based on the matching results, an association graph between field nodes and directory nodes is established, and a tree-like hierarchical directory structure is generated based on the association graph. Then, path encoding is used to encode the tree-like hierarchical directory structure, generating a unique index path for each directory node, thereby forming archived index data.

[0018] In another embodiment, it is assumed that the field topology structure contains 240 field nodes and 960 connection relationships; the directory topology template contains 85 directory nodes and 168 parent-child connection relationships; after mapping and matching, 223 valid association relationships are established; a 5-level tree-like directory structure is generated, with 1 root node, 24 intermediate nodes, and 60 leaf nodes; the path encoding length is 48 bits; finally, 240 archive index identifiers are generated, and the index uniqueness detection pass rate reaches 99.6%.

[0019] Step S4: Perform intelligent archiving processing on business document information based on the archived index data to form a traceable archive dataset; perform archive integrity verification based on the traceable archive dataset to generate intelligent archiving result data.

[0020] In one embodiment, a multi-level index table is established based on archived index data, and this table is used to determine the storage location of business document data blocks. Subsequently, hash checksums are generated for the document data, and field-level index links and version association links are established. Further, a data traceability graph is constructed based on the field-level index links, and a version evolution graph is constructed based on the version association links, forming a traceable archive dataset. Then, archive integrity verification is performed using the hash checksums, index link integrity, and version association integrity, outputting intelligent archiving result data.

[0021] In another embodiment, assuming there are 12,000 data blocks of documents to be archived; 320 first-level indexes, 1,480 second-level indexes, and 12,000 third-level indexes are constructed; 12,000 sets of SHA-256 hash values ​​are generated; 36,800 field-level index links are established; 5,200 version association links are formed; and 12,000 data traceability graph nodes and 42,000 connection relationships are constructed. During the integrity verification process, 18 hash-abnormal data blocks, 9 index chain break nodes, and 5 version missing nodes are detected; the remaining 11,968 data blocks pass the verification; finally, 11,968 successfully archived records and 32 abnormal archived records are generated, corresponding to the intelligent archiving result data.

[0022] Preferably, step S1, which involves parsing the format of business document information, extracting fields, and identifying the content structure to generate document structure feature data, includes: The layout of business document information is parsed to obtain layout area data; the field positions are determined based on the layout area data; and the field relationships are identified at the field positions. In one embodiment, after acquiring the business document image, an adaptive denoising algorithm and a tilt correction algorithm are first used to preprocess the document image to eliminate scanning noise and page skew effects. Then, a layout detection network is used to identify text areas, table areas, stamp areas, and blank areas, and the boundary contours and spatial coordinates of each area are extracted to form layout area data. Next, candidate field boxes are extracted based on the text line detection results, and the field center coordinates, field bounding box size, and field spacing parameters are calculated to determine the field positions. Further calculations are made of the horizontal distance, vertical distance, and alignment deviation between adjacent fields, and a field connection graph is constructed based on the field adjacency relationship. The field connection graph is used to identify hierarchical field relationships, parallel field relationships, and subordinate field relationships, thereby completing the field relationship identification.

[0023] In another embodiment, assuming the input business document image resolution is 3508×2480 pixels, a total of 286 text areas, 34 table areas, 6 stamp areas, and 52 blank areas were detected; 1420 candidate field boxes were extracted; 1420 sets of field center coordinates were calculated; 5830 field connection relationships were established, including 3120 horizontal connection relationships, 1985 vertical connection relationships, and 725 cross connection relationships; the average horizontal spacing of the fields was 18.6 pixels, and the average vertical spacing was 24.3 pixels; finally, 128 parallel field groups, 96 master-detail field groups, and 42 hierarchical field groups were identified.

[0024] The content structure is identified based on field relationships to obtain hierarchical data; document structure features are constructed based on the hierarchical data to generate document structure feature data.

[0025] In one embodiment, the connection relationships between field nodes are extracted based on an established field connection graph, and a graph traversal algorithm is used to identify field clustering areas. Subsequently, a field hierarchy tree is constructed based on the master-slave relationships, alignment relationships, and regional affiliation relationships between fields, and title nodes, content nodes, and subordinate nodes are identified. The depth value, number of nodes, and node coverage of each level are further calculated to form structural hierarchy data. Then, combining the regional area features, spatial distribution features, and field density features in the layout area data, the structural hierarchy data is encoded to construct a document structure feature vector, thereby generating document structure feature data.

[0026] In another embodiment, it is assumed that the field connection graph contains 1420 field nodes and 5830 connection relationships; after hierarchical analysis, 87 hierarchical tree structures are formed; 87 title nodes, 1168 content nodes, and 165 subordinate nodes are identified; the maximum hierarchical depth is 6 levels, and the average hierarchical depth is 3.8 levels; the average number of branches per node is calculated to be 4.6; 326 regional area features, 214 spatial distribution features, and 128 field density features are extracted; finally, a document structure feature vector with a dimension of 768 is constructed, and the corresponding document structure feature data is generated.

[0027] Preferably, the layout of the business document information is parsed to obtain layout area data including: Perform document image preprocessing based on business document information to obtain document enhanced image data; perform layout structure detection based on document enhanced image data to obtain region detection results, which include text regions, table regions, graphic regions and seal regions; In one embodiment, after acquiring the scanned image of the business document, median filtering and bilateral filtering algorithms are first used to eliminate salt-and-pepper noise and background noise. Then, the Hough line detection algorithm is used to identify the document edges, and image rotation correction is performed based on the edge tilt angle to obtain a document image with consistent orientation. Next, an adaptive histogram equalization algorithm is used to enhance the contrast between the text and background areas, generating enhanced document image data. Based on the enhanced document image data, a layout structure detection model is used to identify regions on the page, detecting text regions, table regions, graphic regions, and stamp regions respectively. Further, the bounding rectangle coordinates and region boundary information of each region are extracted to obtain the region detection results.

[0028] In another embodiment, assuming the input business document image resolution is 3000×4200 pixels; after filtering, the number of image noise points is reduced from 12500 to 1860; a page tilt angle of 3.6 degrees is detected and rotation correction is completed; the image contrast is improved from 42 to 87. Subsequently, 268 text regions, 37 table regions, 15 graphic regions, and 8 seal regions are identified; the largest text region area is 185000 pixels, and the largest table region area is 426000 pixels; finally, 328 region detection results are generated for subsequent layout analysis.

[0029] Extract spatial boundary contours based on region detection results; calculate region geometric parameters based on spatial boundary contours; perform partitioning and calibration based on geometric parameters to generate formatted region data.

[0030] In one embodiment, the boundary contour point set of each region is extracted based on the region detection results, and a contour tracking algorithm is used to generate a closed contour curve. Then, based on the closed contour curve, the region's area, perimeter, aspect ratio, rectangularity, and boundary curvature, as well as other region geometric parameters, are calculated. Next, a region topology map is established based on the relative positional relationships, boundary contact relationships, and geometric parameters between the regions. Furthermore, the region topology map is used to identify the top, middle, bottom, and edge-attached regions of the page, and region numbering and hierarchy are completed, thereby generating layout region data.

[0031] In another embodiment, it is assumed that a total of 28,600 boundary contour points are extracted from the detection results of 328 regions; 328 closed contours are generated; the calculated region area ranges from 1,200 pixels to 426,000 pixels, the region perimeter ranges from 180 pixels to 5,120 pixels, the aspect ratio ranges from 0.35 to 12.6, and the rectangularity ranges from 0.71 to 0.98; 956 region topological connections are established. Based on the region topology analysis results, the page is divided into 12 top regions, 246 middle regions, 38 bottom regions, and 32 edge sub-regions; the partitioning and labeling of the 328 regions are completed, and the corresponding layout region data is finally generated.

[0032] Preferably, identifying field relationships at field locations includes: Calculate the spatial distance between fields based on field location data; identify the field arrangement direction based on the field spatial distance; determine the field grouping structure based on the field arrangement direction; establish the field hierarchy relationship based on the determined field grouping structure; In one embodiment, the field bounding boxes in the business document are first located to obtain the center coordinates of each field and the size information of its circumscribed rectangle. Then, the horizontal and vertical spacing between adjacent fields are calculated, and a field spatial distance matrix is ​​constructed. Based on the field spatial distance matrix, the field arrangement trend is analyzed. When the horizontal spacing is continuous and less than a preset threshold, it is determined to be a horizontally arranged field group; when the vertical spacing is continuous and less than a preset threshold, it is determined to be a vertically arranged field group. Subsequently, a field grouping structure is established according to the positional order and alignment relationship within the field group, and a field hierarchy is constructed by combining the positional relationship between the field name area, the input area, and the supplementary description area. For example, "Applicant Information" is identified as the parent field group, and "Name," "ID Number," "Contact Information," etc., below it are identified as child fields, thus forming a hierarchical field structure.

[0033] In another embodiment, assume that a business document, after field detection, yields 128 field areas, with the field center coordinates distributed within an A4 page. The average horizontal distance between fields is calculated to be 18 pixels, and the average vertical distance is 24 pixels. A horizontal aggregation threshold of 25 pixels and a vertical aggregation threshold of 30 pixels are set. After orientation analysis, 12 horizontal field groups and 18 vertical field groups are identified. A further 34-field grouping structure is established, including 8 first-level field groups and 26 second-level field groups. When "Equipment Information" is the parent field group, it is associated with three sub-fields: "Equipment Name," "Equipment Model," and "Equipment Number." When "Detection Result" is the parent field group, it is associated with three sub-fields: "Detection Time," "Detection Personnel," and "Detection Conclusion," ultimately forming a field hierarchy structure containing 128 nodes and 120 hierarchical connections.

[0034] Extract the master-detail join relationship of fields based on the field hierarchy; determine the field region affiliation relationship based on the master-detail join relationship; generate field neighborhood join data based on the field region affiliation relationship; record the field relationships based on the field neighborhood join data.

[0035] In one embodiment, it is assumed that 128 field nodes and 120 hierarchical connections are generated. After master-slave relationship analysis, 32 master fields and 96 slave fields are identified. Further, the data is divided into 1 title area, 5 table areas, 8 fill-in areas, and 2 signature areas, totaling 16 field areas. The "Detection Record Table" field is associated with 12 slave fields, and the "Device Parameters" field is associated with 8 slave fields. Based on field spatial adjacency analysis, a field neighborhood network is constructed, generating 462 neighborhood connection edges, including 278 horizontal connection edges and 184 vertical connection edges. For the field "Device Number," its left neighbor is identified as "Device Model," its right neighbor as "Production Date," and its lower neighbor as "Installation Location." For the field "Detection Conclusion," its upper neighbor is identified as "Detection Item," and its lower neighbor as "Signature Confirmation." The final result is a field relationship dataset containing 128 field nodes, 120 hierarchical relationships, 96 master-detail relationships, and 462 neighborhood connection relationships, which provides a foundation for subsequent content structure recognition and document structure feature construction.

[0036] Preferably, identifying the field arrangement direction based on field spatial distance includes: Extract the horizontal and vertical distances of fields based on the spatial distance data; determine the main expansion direction of the fields based on the horizontal and vertical distances; calculate the field direction consistency parameter based on the main expansion direction. In one embodiment, the position coordinates and bounding box information of each field in the document are first read, and the horizontal and vertical distances between adjacent fields are calculated. Then, a local neighborhood set is constructed with each field as the center, and the distribution of fields in the horizontal and vertical directions within the neighborhood is statistically analyzed. When the horizontal distance continuity is higher than the vertical distance continuity, the field group is determined to be a horizontally expanded structure; when the vertical distance continuity is higher than the horizontal distance continuity, the field group is determined to be a vertically expanded structure. Subsequently, the direction of the line connecting the center points of the fields is extracted, and the angle difference between each connecting direction and the main expansion direction is calculated. The directional dispersion is statistically analyzed based on the angle difference, and the directional dispersion is used to calculate the field directional consistency parameter. The higher the consistency parameter, the more regular the field arrangement; the lower the consistency parameter, the more likely the fields are tilted, misaligned, or have scanning distortion.

[0037] In another embodiment, assuming a business document yields 156 field regions after field detection, establishing 428 sets of adjacent field relationships. The average horizontal distance is 22 pixels, and the average vertical distance is 48 pixels. Neighborhood statistics reveal that 312 sets of fields are continuously distributed horizontally, and 116 sets are continuously distributed vertically; therefore, the main expansion direction of this region is determined to be horizontal. Further extraction of the 428 field center lines shows that the direction angles are mainly distributed within the range of 0° to 12°. The statistical mean direction is 4.8°, the standard deviation is 2.6°, and the final calculated direction consistency parameter is 0.91. For another table region, the standard deviation reaches 8.4°, and the corresponding direction consistency parameter drops to 0.73, indicating a significant field arrangement offset in this region.

[0038] Calculate the field orientation offset angle based on the orientation consistency parameter; perform orientation correction based on the field orientation offset angle to determine the field arrangement orientation.

[0039] In one embodiment, based on the obtained field orientation consistency parameters, orientation fitting analysis is performed on the field center lines. First, the least squares method is used to fit the field arrangement baseline, and the deviation angle of each field center point from the baseline is calculated. Then, all deviation angles are weighted and statistically analyzed to obtain the field orientation offset angle. When the field orientation offset angle exceeds a preset threshold, rotation correction is performed on the field coordinates to ensure that the field arrangement orientation is consistent with the standard horizontal or vertical axis. After correction, the spatial relationship between fields is recalculated, and the final field arrangement orientation is determined based on the corrected field coordinates, thereby improving the accuracy of subsequent field grouping and hierarchy recognition.

[0040] In another embodiment, assuming a scanned document detects 156 field areas with an orientation consistency parameter of 0.82, a field arrangement baseline is obtained using least-squares fitting, with an inclination angle of 7.5° relative to the horizontal axis. Further statistical analysis reveals that 128 of the 156 fields have an offset angle concentrated between 6.2° and 8.4°, 18 fields have an offset angle between 4.5° and 6.0°, and 10 fields have an offset angle exceeding 10°. Based on this, the system determines the overall field orientation offset angle to be 7.3° and performs a counter-clockwise 7.3° rotation correction on all field coordinates. After correction, the field orientation distribution is recalculated, and the orientation standard deviation decreases from 6.8° to 1.4°, while the orientation consistency parameter increases from 0.82 to 0.96. Specifically, 21 horizontally arranged field groups and 6 vertically arranged field groups are identified, ultimately accurately determining the field arrangement orientation and structural layout characteristics of the document page.

[0041] Preferably, step S2 involves modular classification analysis based on document structural feature data to determine the classification module to which the document belongs, including: Extract field layout information from document structure feature data; calculate field spatial distribution density based on field layout information; divide structural region units based on field spatial distribution density; In one embodiment, the field coordinate information in the document's structural feature data is parsed to obtain the bounding box position, width, height, and center coordinates of each field. A two-dimensional spatial grid partitioning method is used to divide the document page into several regular grid units, and the number of fields and the field area ratio in each grid unit are counted. Subsequently, a kernel density estimation algorithm is used to calculate the spatial distribution density of the fields, generating a field density distribution map. For regions with continuous density values ​​exceeding a preset density threshold, region merging is performed to form structural region units; for regions with low density and discrete distribution, they are retained as independent region units. This method enables spatial differentiation of the title area, main content area, table area, and notes area in the document, providing basic structural information for subsequent module classification.

[0042] In another embodiment, assuming a document page size of 2480×3508 pixels, a total of 180 field objects were detected. A 40×40 grid was used to generate 1600 grid cells, with an average density of 0.82 in dense field areas, 0.46 in ordinary text areas, and 0.08 in edge blank areas. After setting a density threshold of 0.60, six high-density structural area cells were identified: the top title area (approximately 320,000 square pixels), the main information area (approximately 1,450,000 square pixels), the table area (approximately 780,000 square pixels), the supplementary description area (approximately 260,000 square pixels), and the remaining two areas corresponding to the signature area and the footer area, respectively. After area merging, the overlap rate of structural area boundaries was controlled within 3%, forming stable structural area cell data.

[0043] Spatial clustering analysis is performed based on structural region units to form candidate module clusters; the candidate module clusters are matched with a preset module template library to obtain structural similarity; and the classification module to which the document belongs is determined based on structural similarity.

[0044] In one embodiment, the geometric location, area, aspect ratio, number of fields, and relative positional relationships between structural region units are used as clustering feature vectors. The DBSCAN clustering algorithm is employed to perform spatial clustering analysis on the structural region units, forming several candidate module clusters. Subsequently, the structural feature vectors of corresponding templates are extracted from a pre-defined module template library, and the structural similarity between the candidate module clusters and the templates is calculated using a combination of cosine similarity and Hausdorff distance. For candidate module clusters with similarity reaching a pre-defined threshold, their corresponding module type is determined. Finally, based on the combination relationships of each module type and the overall matching score, the classification module to which the document belongs is determined.

[0045] In another embodiment, suppose a document, after processing, yields eight structural region units, forming eight sets of region feature vectors. DBSCAN clustering analysis yields four candidate module clusters: cluster M1 contains the title and number regions, cluster M2 contains the main information region, cluster M3 contains the table region, and cluster M4 contains the signature region. The module template library contains 12 preset template structures. After matching calculations, the structural similarity between M1 and template T3 is 0.94, M2 and template T3 are 0.91, M3 and template T3 are 0.96, and M4 and template T3 are 0.89; while the average similarity with other templates is below 0.75. Further calculation shows an overall structural matching score of 0.925, higher than the preset classification threshold of 0.85. Therefore, this document is identified as the module corresponding to template T3. If multiple templates have similar scores, the template with the highest overall score is selected as the final classification result, thereby improving the module recognition accuracy of complex document formats.

[0046] Preferably, calculating the field spatial distribution density based on the field layout information includes: Extract field coordinate position data based on field layout information; divide the document page into grid areas based on field coordinate position data; count the number of fields in the document page grid areas; calculate the local space occupancy rate based on the number of fields; calculate the field clustering intensity based on the local space occupancy rate. In one embodiment, assuming a document page has a resolution of 2400×3400 pixels, a total of 260 field objects are detected. The page is divided into 20×25 grid areas, forming 500 grid cells, each with a size of 120×136 pixels. During the statistical analysis, some table areas contain 12 fields per grid cell, with a total field area of ​​11,500 square pixels, corresponding to a local space occupancy rate of 0.70; the text area contains an average of 6 fields per grid cell, with a total field area of ​​6,800 square pixels, corresponding to a local space occupancy rate of 0.42; and the margin blank areas contain fewer than 1 fields per grid cell, corresponding to a local space occupancy rate below 0.08. Further calculations show that the field clustering intensity ranges from 0.05 to 0.91, with an average clustering intensity of 0.87 for table areas, 0.73 for title areas, and 0.51 for text areas, thus forming complete field clustering intensity distribution data.

[0047] Identify high-density and low-density distribution areas based on field aggregation intensity; calculate the spatial distribution gradient based on the high-density and low-density distribution areas; and generate the field spatial distribution density based on the spatial distribution gradient.

[0048] In one embodiment, the obtained field aggregation intensity distribution data is subjected to threshold segmentation processing. Grid regions with aggregation intensity higher than a preset threshold are identified as high-density distribution regions, and grid regions with aggregation intensity lower than the preset threshold are identified as low-density distribution regions. Subsequently, the boundary information between high-density distribution regions and adjacent low-density distribution regions is extracted, the aggregation intensity difference and spatial distance between adjacent regions are calculated, and spatial distribution gradient parameters are constructed based on the intensity change rate. For regions with continuous gradient changes, an interpolation smoothing method is used to generate density transition bands; for regions with prominent gradient changes, obvious boundary features are preserved. Finally, a field spatial distribution density map is constructed based on the spatial distribution gradient values ​​corresponding to each location to reflect the aggregation pattern and distribution characteristics of fields on the page.

[0049] In another embodiment, assuming a field clustering intensity threshold of 0.60, 85 high-density areas, 290 low-density areas, and 125 intermediate transition areas are identified out of 500 grid areas. The average clustering intensity of a certain table area is 0.88, significantly different from the average clustering intensity of 0.46 in the adjacent body text area, with a spatial distance of 180 pixels, resulting in a calculated spatial distribution gradient of 0.0023. The clustering intensity of a certain title area is 0.79, forming a gradient value of 0.0035 with the clustering intensity of 0.12 in the header blank area. After calculating the gradient for all areas, the gradient range is obtained from 0.0002 to 0.0041, and a field spatial distribution density map is generated based on the gradient distribution. The high-density core area accounts for approximately 28% of the total page area, the medium-density transition area accounts for approximately 37%, and the low-density area accounts for approximately 35%. Spatial distribution density data can clearly characterize the spatial distribution features of the title module, body module, table module, and supplementary description module in a document, providing a technical basis for subsequent structural area unit division and classification module identification.

[0050] Preferably, spatial clustering analysis is performed based on structural region units to form candidate module clusters, including: Calculate the spatial distance between regions based on structural region unit data; construct a region adjacency graph based on the spatial distance; identify the spatial connectivity relationships between regions based on the adjacency graph; and divide the initial region clusters based on the spatial connectivity relationships. In one embodiment, it is assumed that after structural parsing, the document page yields 48 structural region units, including 20 text region units, 11 table region units, 9 label region units, and 8 graphic region units. By statistically analyzing the distances between regions, 136 groups of regions satisfy the adjacency condition, and a region adjacency graph is constructed accordingly. Further analysis yields three main connected region sets: the first set contains 16 structural region units, the second set contains 13 structural region units, and the third set contains 19 structural region units. After merging regions based on spatial connectivity, three initial region clusters are formed. Each initial region cluster has a continuous internal region distribution and exhibits significant spatial clustering characteristics, while maintaining a large spatial interval between different region clusters.

[0051] Calculate the regional density consistency parameter based on the initial regional clusters; identify stable cluster structures based on the regional density consistency parameter; generate candidate module clusters based on the stable cluster structures.

[0052] In one embodiment, the number of regional units, spatial coverage, and regional distribution of each initial region cluster are statistically analyzed. Then, the uniformity of distribution of each structural regional unit within the region cluster is analyzed to obtain the corresponding regional density consistency parameter. The regional density consistency parameter characterizes whether the regional arrangement within the region cluster is stable and whether there are significant density abrupt changes. Then, each region cluster is screened according to preset judgment rules. When the regional density consistency parameter meets the stability requirements, the corresponding region cluster is identified as a stable cluster structure; when the regional density consistency parameter is lower than the stability requirements, it is determined to be an unstable cluster structure. Finally, the stable cluster structures are integrated to generate candidate module clusters, and the spatial boundaries, regional composition, and structural feature information of the candidate module clusters are recorded to provide input data for subsequent module identification and layout analysis.

[0053] In another embodiment, it is assumed that three initial region clusters are obtained, where the first initial region cluster contains 16 structural region units, the second initial region cluster contains 13 structural region units, and the third initial region cluster contains 19 structural region units. After region density analysis, it is found that the uniformity of region distribution within the first initial region cluster is 92%, the second initial region cluster is 68%, and the third initial region cluster is 95%. Setting a stability threshold of 80%, the first and third initial region clusters are identified as stable cluster structures, while the second initial region cluster, due to significant density fluctuations, is identified as an unstable cluster structure. Further integration of the two stable cluster structures generates two candidate module clusters, where candidate module cluster M1 contains 16 structural region units and candidate module cluster M2 contains 19 structural region units. The generated candidate module clusters can relatively completely preserve the spatial organization relationship in the original layout, providing a reliable foundation for subsequent structural template matching and module classification.

[0054] Preferably, step S3, generating archive index data based on the hierarchical archive directory structure, includes: Extract node path information from the hierarchical archive directory structure; construct a directory path encoding sequence based on the node path information; generate a directory index identifier based on the directory path encoding sequence; In one embodiment, firstly, hierarchical archive directory structure data is acquired, and the parent-child relationships between directory nodes are parsed. Then, starting from the root node, the directory tree is traversed layer by layer, extracting the complete path information of each node, including the identifier of the parent node, the name of the current node, and the hierarchical depth information. Next, each level of nodes is sequentially encoded according to a preset encoding rule, converting each level of the directory in the node path into a corresponding path code segment, which is then concatenated according to hierarchical order to form a directory path encoding sequence. Then, a corresponding directory index identifier is generated based on the directory path encoding sequence, and a mapping relationship is established between the directory index identifier and the original node, thereby completing the indexed representation of the directory structure and providing basic data for subsequent directory retrieval and archiving management.

[0055] In another embodiment, it is assumed that a file directory tree with a 5-level hierarchical structure is obtained, containing a total of 120 directory nodes. The root node has 4 first-level nodes, each containing an average of 6 second-level nodes, and these second-level nodes continue to expand downwards to form a complete directory structure. After path parsing, 120 node path information entries are extracted, with path depths ranging from 2 to 5 levels. Subsequently, 120 sets of directory path encoding sequences are generated, where the longest encoding sequence contains 5 encoding segments and the shortest contains 2 encoding segments. Further, 120 directory index identifiers are generated based on the encoding sequences, each index identifier corresponding to a unique directory node, thus completing the construction of the directory index system.

[0056] Establish node spatial index relationships based on directory index identifiers; calculate index uniqueness parameters based on node spatial index relationships; perform index deduplication based on index uniqueness parameters to generate archived index data.

[0057] In one embodiment, a node spatial index relationship is established using the generated directory index identifiers. Directory nodes with the same parent node or adjacent hierarchical relationships are associated and mapped to form a directory index network structure. Subsequently, the path similarity, hierarchical association, and node positional relationships among the directory index identifiers are analyzed. The duplication and overlap characteristics between indexes are statistically analyzed, and an index uniqueness parameter is calculated accordingly. The index uniqueness parameter characterizes the independence and distinguishability of the current index identifier within the entire directory structure. When multiple index identifiers are found to have duplicate path characteristics or redundant node mappings, index deduplication is performed based on the index uniqueness parameter. Index records with higher uniqueness are retained, and duplicate index records are deleted or merged, ultimately generating archived index data.

[0058] In another embodiment, assuming 120 directory index identifiers are generated, 120 sets of node spatial index relationships are established through node association analysis. It was found that 12 sets of indexes exhibited path duplication, involving 28 directory nodes. Further calculation of the uniqueness parameters of each index revealed that 18 index identifiers had a uniqueness parameter higher than 95%, 74 index identifiers had a uniqueness parameter between 85% and 95%, and another 28 index identifiers had a uniqueness parameter lower than 85%. Subsequently, deduplication was performed on the indexes with a uniqueness parameter lower than 85%, merging 16 duplicate indexes, deleting 8 redundant indexes, and regenerating new index identifiers for the remaining 4 indexes through supplementary hierarchical coding. Ultimately, 100 archived index data items were generated, each corresponding to a unique directory node with no duplicate mapping relationships, providing a unified index foundation for subsequent archive storage, retrieval, and directory tracing.

[0059] Preferably, in step S4, the business document information is intelligently archived based on the archived index data to form a traceable archive dataset, including: The archived index data is parsed into archive storage path and directory node location information; the storage unit for business document information is determined based on the archive storage path and directory node location information. In one embodiment, the archived index data is structured and parsed to extract the archive storage path, directory level identifier, node number, and storage partition information. Then, a directory tree mapping relationship is established based on the directory node location information, mapping business documents to target directory nodes. Further, combined with the logical partition configuration in the storage device, the directory nodes and storage units are mapped and matched to determine the storage unit corresponding to the business document information. For example, if the archived index data contains five-level directory structure information, the target node is located by traversing the directory tree, and the storage unit is allocated based on the storage partition number corresponding to the node. Finally, a mapping table between business documents and storage units is generated, providing accurate storage locations for subsequent file writing.

[0060] In another embodiment, assuming the archiving system contains 12,000 archival index data entries, parsing yields 12,000 file storage paths and 12,000 directory node location information entries. The directory tree has a depth of 6 levels and contains 850 directory nodes. After node location processing, 320 valid target nodes are identified and mapped to 24 logical storage partitions. Further analysis reveals that the first storage partition corresponds to 1,800 business documents, the second to 1,450, and the third to 1,670, with the remaining business documents distributed across the other 21 storage partitions. This results in 12,000 records mapping business documents to storage units, each record containing a directory node number, storage partition number, and storage unit identifier, thus achieving precise location and storage of business documents.

[0061] Write business document information into the storage unit and generate basic archive data records; establish field-level index relationships based on the basic archive data records; generate data traceability link information based on the field-level index relationships; construct an archive version association structure based on the data traceability link information; and generate a traceable archive dataset based on the archive version association structure and the data traceability link information.

[0062] In one embodiment, after determining the storage unit corresponding to the business document, the business document content is written to the target storage unit, and the writing time, storage location identifier, directory node number, and data summary information are recorded to generate basic archive data records. Subsequently, the fields in the basic archive data records are parsed to extract the document number field, time field, status field, source field, and associated fields, and field-level index relationships are established. Data traceability link information is generated based on the reference and update relationships between the fields to record the association trajectory of archive data throughout the entire process from creation, modification, to archiving. Furthermore, an archive version association structure is constructed based on the data traceability links generated at different time points, associating multiple historical versions of the same business document. Finally, the archive version association structure and data traceability link information are combined to generate a traceable archive dataset, enabling any version of the archive to be traced back to its predecessor version and associated field records.

[0063] In another embodiment, suppose a file system receives 5,000 business documents, of which 4,980 are successfully written to the storage unit. After writing, 4,980 basic file data records are generated, each containing an average of 32 fields. Further establishing field-level index relationships results in 159,360 field index nodes and 42,600 field relationships. Subsequently, data traceability link information is generated based on these field relationships, constructing 9,800 traceability links, with the longest containing 18 links and the shortest containing 3 links. Further version organization is performed on the historical update records of the same business document, forming 1,420 sets of file version association structures, each containing an average of 4 version nodes, with a maximum version depth of 11 levels. Finally, a traceable file dataset containing 4,980 files, 5,680 version nodes, and 9,800 traceability links is generated, achieving complete association management between file content, field change records, and historical versions.

[0064] It should be noted that you should refer to [link / reference]. Figure 3The process unfolds in four stages, S1 to S4: Stage S1 involves parsing the format of the acquired business documents, extracting fields, and identifying the content structure to generate document structure feature data containing information such as field sets, field types, field locations, structural relationships, and format / content semantics; Stage S2 performs feature vectorization processing based on the structure feature data and classifies the documents into different business modules (such as module A, module B, module C, etc.) through classification model analysis; Stage S3 establishes the mapping relationship between document fields and archive directory nodes based on the classification results, thereby generating a hierarchical archive directory structure and forming archived index data containing index ID, directory path, field mapping, version, and time information; Stage S4 performs intelligent archiving processing on the archived index data, constructs a traceable archive dataset, and performs archive integrity verification (including index, field, and version consistency, etc.), finally outputting the archived result data, and triggering exception handling and alarm mechanisms when verification anomalies occur.

[0065] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, 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 the equivalents of the application are intended to be included within the invention.

[0066] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A modular classification and intelligent archiving method for business document information, characterized in that, Includes the following steps: Step S1: Obtain business document information; Perform layout parsing, field extraction, and content structure recognition on business document information to generate document structure feature data; Step S2: Perform modular classification analysis based on the document structure feature data to determine the classification module to which the document belongs; Step S3: Perform file structure mapping based on the classification module to which the document belongs, so as to establish the relationship between document fields and file directory nodes; generate a hierarchical file directory structure based on the relationship; generate archive index data based on the hierarchical file directory structure; Step S4: Perform intelligent archiving processing on business document information based on the archived index data to form a traceable archive dataset; perform archive integrity verification based on the traceable archive dataset to generate intelligent archiving result data.

2. The modular classification and intelligent archiving method for business document information according to claim 1, characterized in that, Step S1 involves parsing the format of business document information, extracting fields, and identifying the content structure to generate document structure feature data, including: The layout of business document information is parsed to obtain layout area data; the field positions are determined based on the layout area data; and the field relationships are identified at the field positions. The content structure is identified based on field relationships to obtain hierarchical data; document structure features are constructed based on the hierarchical data to generate document structure feature data.

3. The modular classification and intelligent archiving method for business document information according to claim 2, characterized in that, The layout of the business document information is parsed to obtain the layout area data, which includes: Perform document image preprocessing based on business document information to obtain document enhanced image data; perform layout structure detection based on document enhanced image data to obtain region detection results, which include text regions, table regions, graphic regions and seal regions; Extract spatial boundary contours based on region detection results; calculate region geometric parameters based on spatial boundary contours; perform partitioning and calibration based on geometric parameters to generate formatted region data.

4. The modular classification and intelligent archiving method for business document information according to claim 2, characterized in that, Identifying field relationships at field locations includes: Calculate the spatial distance between fields based on field location data; identify the field arrangement direction based on the field spatial distance; determine the field grouping structure based on the field arrangement direction; establish the field hierarchy relationship based on the determined field grouping structure; Extract the master-detail join relationship of fields based on the field hierarchy; determine the field region affiliation relationship based on the master-detail join relationship; generate field neighborhood join data based on the field region affiliation relationship; record the field relationship based on the field neighborhood join data.

5. The modular classification and intelligent archiving method for business document information according to claim 4, characterized in that, Identifying the field arrangement direction based on spatial distance includes: Extract the horizontal and vertical distances of fields based on the spatial distance data; determine the main expansion direction of the fields based on the horizontal and vertical distances; calculate the field direction consistency parameter based on the main expansion direction. Calculate the field orientation offset angle based on the orientation consistency parameter; perform orientation correction based on the field orientation offset angle to determine the field arrangement orientation.

6. The modular classification and intelligent archiving method for business document information according to claim 1, characterized in that, Step S2 involves modular classification analysis based on document structure characteristics to determine the document's classification module, including: Extract field layout information from document structure feature data; calculate field spatial distribution density based on field layout information; divide structural region units based on field spatial distribution density; Spatial clustering analysis is performed based on structural region units to form candidate module clusters; the candidate module clusters are matched with a preset module template library to obtain structural similarity; and the classification module to which the document belongs is determined based on structural similarity.

7. The modular classification and intelligent archiving method for business document information according to claim 6, characterized in that, Calculating the field spatial distribution density based on field layout information includes: Extract field coordinate position data based on field layout information; divide the document page into grid areas based on field coordinate position data; count the number of fields in the document page grid areas; calculate the local space occupancy rate based on the number of fields; calculate the field clustering intensity based on the local space occupancy rate. Identify high-density and low-density distribution areas based on field aggregation intensity; calculate the spatial distribution gradient based on the high-density and low-density distribution areas; and generate the field spatial distribution density based on the spatial distribution gradient.

8. The modular classification and intelligent archiving method for business document information according to claim 6, characterized in that, Spatial clustering analysis based on structural region units yields candidate module clusters, including: Calculate the spatial distance between regions based on structural region unit data; construct a region adjacency graph based on the spatial distance; identify the spatial connectivity relationships between regions based on the adjacency graph; and divide the initial region clusters based on the spatial connectivity relationships. Calculate the regional density consistency parameter based on the initial regional clusters; identify stable cluster structures based on the regional density consistency parameter; generate candidate module clusters based on the stable cluster structures.

9. The modular classification and intelligent archiving method for business document information according to claim 1, characterized in that, Step S3, which generates archive index data based on the hierarchical archive directory structure, includes: Extract node path information from the hierarchical archive directory structure; construct a directory path encoding sequence based on the node path information; generate a directory index identifier based on the directory path encoding sequence; Establish node spatial index relationships based on directory index identifiers; calculate index uniqueness parameters based on node spatial index relationships; perform index deduplication based on index uniqueness parameters to generate archived index data.

10. The modular classification and intelligent archiving method for business document information according to claim 1, characterized in that, In step S4, the business document information is intelligently archived based on the archived index data to form a traceable archive dataset, including: The archived index data is parsed into archive storage path and directory node location information; the storage unit for business document information is determined based on the archive storage path and directory node location information. Write business document information into the storage unit and generate basic archive data records; establish field-level index relationships based on the basic archive data records; generate data traceability link information based on the field-level index relationships; construct an archive version association structure based on the data traceability link information; and generate a traceable archive dataset based on the archive version association structure and the data traceability link information.

Images