Digital import and export intelligent auditing method based on CV-NLP fusion algorithm

Abstract

The invention relates to the technical field of intelligent auditing, and discloses a digital import and export intelligent auditing method based on a CV-NLP fusion algorithm, and the method comprises the steps: carrying out the image coordinate standardization processing of an image sequence of an import and export document package, and generating a shipping mark candidate section set containing a continuous credible weight through the semantic probability and a section prior attribute; according to the normalized area scale distribution of the section set, calculating a layout hierarchy complexity parameter representing the fragmentation degree of the hierarchical structure in the document packet; generating multi-level resolution sampling weight distribution based on the parameters, and performing multi-scale identification and probability fusion on the candidate sections by using weights determined by the distribution; respectively calculating a character structure fragmentation metric value representing structure fragmentation bias and a region positioning volatility metric value representing a cross-page layout drop point dispersion degree in combination with layout geometric attributes; and finally, constructing a comprehensive risk assessment model, and outputting an intelligent auditing result for quantitatively representing the piece-level identity observability risk.

Description

Technical Field

[0001] This invention relates to the field of intelligent auditing technology, and more specifically, to a digital import and export intelligent auditing method based on CV-NLP fusion algorithms. Background Technology

[0002] In the international import and export logistics sector, less-than-container load (LCL) shipping is a common mode of cargo transportation. The process involves consolidating multiple shipments from different shippers into a single container at a container freight station, and then unpacking and distributing them at the destination port. To accurately distinguish between different goods mixed in the same container, shipping marks / item numbers have become crucial identifying information at the item level.

[0003] Corresponding to this operational process, LCL (Less than Container Load) shipping involves a complex documentation system, typically including a Master Bill of Lading (MasterB / L) covering the entire container, House Bills of Lading corresponding to each shipment, and a Consolidated Manifest detailing the loading status. In a digital auditing scenario, the intelligent system needs to automate the processing of these documentation packages to verify the authenticity and compliance of the cargo information.

[0004] Existing technologies for processing such import and export documents typically employ an optical character recognition (OCR) system with rule-based comparison. This involves first extracting text information from various document images using OCR, and then comparing the key fields on each document against pre-defined logical rules to ensure consistency. However, in LCL (Less than Container Load) scenarios, the document layout exhibits significant multi-scale hierarchical differences: In LCL manifests, in order to list dozens of shipments within a container on one or a few pages, shipping information is usually compressed into dense table rows, presented as high-density, small-scale blocks of text and images. In a split bill of lading, since each document only describes one shipment, the shipping information usually occupies a separate, large field, appearing as a low-density, large-scale graphic block. On packing lists or labels, shipping information may be presented at a medium size.

[0005] Existing OCR and layout analysis algorithms typically employ a uniform scale standard or fixed resolution strategy to process all images within the same document package. When faced with LCL (Less than Container Load) scenarios where there are cascading changes in layout scale, existing technologies often struggle to adapt. For example, while high-resolution processing of large-scale fields is acceptable, it is highly susceptible to fragmentation errors (splitting a single shipping mark into multiple fragments) or concatenation errors (mistaking adjacent shipping marks for a single line) when processing dense rows in LCL manifests due to insufficient character spacing.

[0006] This identification bias caused by the scaling effect makes it impossible for the system to construct a stable and uniform item-level identity representation at the machine level. Even if the subsequent comparison rules are improved, the final intelligent review results often have a high false positive or false negative rate because the identification data input from the front end has already undergone structural distortion and fragmentation, making it difficult to meet the actual needs of high-precision automated review. Summary of the Invention

[0007] This invention provides a digital import and export intelligent auditing method based on CV-NLP fusion algorithm, which solves the technical problems mentioned in the background art.

[0008] This invention provides a digital import and export intelligent auditing method based on CV-NLP fusion algorithm, including: The image coordinates of the image sequence of the import and export document package are standardized, and a set of candidate shipping marks containing continuous and reliable weights is generated by joint localization of semantic probability and prior attributes of layout. Based on the normalized area scale distribution of each section in the candidate section set of the mark, calculate the section hierarchy complexity parameter configured to characterize the degree of fragmentation of the hierarchical structure within the single document package; Based on the page layout complexity parameter, a multi-level resolution sampling weight distribution is generated, and the weights determined by this distribution are used to perform multi-scale identification and probability fusion on the candidate page block set of the logo. Based on the multi-scale recognition results and page geometric attributes, the character structure fragmentation metric value configured to characterize the structural fragmentation bias and the regional positioning volatility metric value configured to characterize the dispersion of the page layout landing point are calculated respectively. A comprehensive risk assessment model is constructed, which includes the page layout complexity parameter, character structure fragmentation metric, and regional positioning volatility metric. The comprehensive risk assessment model outputs intelligent audit results that quantitatively represent the observability risk of the identity at the component level.

[0009] The beneficial effects of this invention include: by constructing a page level complexity parameter that characterizes the degree of page level fragmentation, driving multi-level resolution sampling and probabilistic fusion, it effectively overcomes the problem of character structure fragmentation and reorganization bias caused by differences in document page scale in the LCL scenario; at the same time, by introducing a scale and form coupling measurement mechanism and an adaptive comprehensive risk assessment model, it realizes the quantitative evaluation of the observability risk of item-level identity, and significantly improves the robustness and accuracy of complex import and export document packages in automated review without the need for cross-document consistency comparison. Attached Figure Description

[0010] Figure 1 This is a flowchart of the digital import and export intelligent auditing method based on the CV-NLP fusion algorithm of the present invention; Figure 2 This is a schematic diagram illustrating a specific implementation of the present invention. Detailed Implementation

[0011] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0012] like Figure 1 As shown, the digital import and export intelligent auditing method based on the CV-NLP fusion algorithm includes: The image coordinates of the image sequence of the import and export document package are standardized, and a set of candidate shipping marks containing continuous and reliable weights is generated by joint localization of semantic probability and prior attributes of layout. Based on the normalized area scale distribution of each section in the candidate section set of the mark, calculate the section hierarchy complexity parameter configured to characterize the degree of fragmentation of the hierarchical structure within the single document package; Based on the page layout complexity parameter, a multi-level resolution sampling weight distribution is generated, and the weights determined by this distribution are used to perform multi-scale identification and probability fusion on the candidate page block set of the logo. Based on the multi-scale recognition results and page geometric attributes, the character structure fragmentation metric value configured to characterize the structural fragmentation bias and the regional positioning volatility metric value configured to characterize the dispersion of the page layout landing point are calculated respectively. A comprehensive risk assessment model is constructed, which includes the page layout complexity parameter, character structure fragmentation metric, and regional positioning volatility metric. The comprehensive risk assessment model outputs intelligent audit results that quantitatively represent the observability risk of the identity at the component level.

[0013] In a preferred embodiment, the image sequence of the import / export document package is subjected to image coordinate standardization processing, and a set of candidate shipping mark blocks containing continuous reliable weights is generated through joint localization of semantic probability and prior layout attributes, including: Definition of the first Normalized coordinate transformation of page image : in, These are the original pixel coordinates. For the first The pixel width and height of the page image are used to map all page coordinates to... space; It should be noted that, This parameter is used to identify individual pages within an import / export document package and serves as a page index. Its core function is to distinguish different document pages, ensuring that standardized coordinate processing accurately corresponds to each page and avoiding confusion regarding the width, height, and coordinate information of different pages. This parameter has no fixed value range; it simply needs to be assigned values ​​sequentially according to the page order within the document package.

[0014] It should be noted that, The original pixel coordinates of any point on the image of the document page, where The horizontal coordinates of the corresponding pixel The vertical coordinates of the corresponding pixel. Their purpose is to serve as the raw input data for coordinate standardization. The range of values ​​is determined by the pixel width and height of the corresponding page.

[0015] It should be noted that, Refers to the first The pixel width of a page image, that is, the total number of pixels contained in the page in the horizontal direction. Its function is to convert the original pixel coordinates... Normalization eliminates measurement bias caused by differences in physical page size and resolution. The data is obtained by reading the [image data] using an image analysis tool. The metadata of the page image is obtained.

[0016] It should be noted that, Refers to the first The pixel height of a page image, that is, the total number of pixels contained in the page in the vertical direction. Its function is similar to... Correspondingly, it is used to convert the original pixel coordinates Normalization ensures that the coordinates of different pages are in the same order of magnitude space. The acquisition method is the same. Read the image using an image analysis tool Metadata for the page image.

[0017] It should be noted that, Defined as the first The normalized coordinate transformation function for a page image. Its function is to transform the original pixel coordinates... Mapped to The normalized physical coordinate space enables a unified measurement of coordinates for pages of different sizes and resolutions.

[0018] The continuous confidence weight of each candidate block is calculated using the following formula. : in, This refers to the confidence score of semantic associations extracted based on natural language processing. These are prior values ​​for layout structure extracted based on computer vision. To integrate the weighted coefficients, The activation function is non-linear, ultimately forming the set of candidate marker blocks. , This represents the geometric region information of the k-th candidate block, including the boundary of the candidate block in the normalized coordinate space, used to define the spatial range of the k-th candidate block.

[0019] It should be noted that, Used to identify a single section in the set of candidate sections for shipping marks, it is a section number index, and its function is to distinguish different candidate sections for shipping marks.

[0020] It should be noted that, Refers to the first The semantic association confidence score of each candidate section, with a value ranging from [0,1]. Its purpose is to quantify the degree of semantic alignment between the text content of the candidate section and the marking / part number. It is obtained by performing semantic analysis on the OCR text of the candidate sections using a natural language processing network.

[0021] It should be noted that, Refers to the first The prior values ​​of the layout structure of each candidate section are non-negative real numbers. Their purpose is to comprehensively reflect the degree to which the morphological characteristics of the candidate sections match the typical label fields, including the section's aspect ratio, edge regularity, and whether it is located in the typical label area of ​​the document, providing a basis for the layout dimension of the continuous reliable weights. They are obtained by extracting the morphological characteristics of the candidate sections through a layout analysis network and then calculating them.

[0022] It should be noted that, Semantic relevance confidence Prior values ​​of layout structure The fusion weighting coefficients are all non-negative constants. Their function is to adjust the contribution ratio of semantic features and layout features in the calculation of continuous reliability weights, adapting to the differences in the importance of the two types of features under different document scenarios. Optimal values ​​are obtained through offline training, specifically by using a document dataset containing a large amount of labeled data and performing grid search optimization with the prediction accuracy of the continuous reliability weights as the target. and The value range is [0.1, 5.0].

[0023] It should be noted that, This refers to the logistic function. Its function is to map the numerical values ​​after a linear combination to the interval [0,1], so that the output results have a quantitative meaning of credibility or risk probability, avoiding the difficulties in calculation or interpretation caused by the unbounded range of numerical values.

[0024] It should be noted that, Refers to the first The continuous credibility weight of each candidate mark segment, with a value range of [0,1]. Its function is to quantify the credibility of the segment as a genuine mark area.

[0025] It should be noted that, The set of candidate sections for the shipping mark is expressed as follows: Its function is to integrate all candidate blocks obtained through CV-NLP joint localization and their corresponding continuous confidence weights. Representing the Geometric region information (such as boundary coordinates) of each candidate block. Its continuous credible weight, This represents the total number of sections in the set.

[0026] It should be noted that, Refers to the collection of candidate sections for shipping marks The total number of candidate sections included is a positive integer. Its purpose is to specify the size of the section set. The value is determined by the complexity of the document package's content; the more shipping mark-related areas the document contains and the more complex the layout, the better. The larger the value.

[0027] It should be noted that, Refers to the first The geometric region information of each candidate label block includes the boundary coordinates of the block in the normalized coordinate space (such as the coordinates of the top left and bottom right corners). Its function is to define the spatial extent of each candidate block. It is obtained by performing region segmentation on the normalized image using a layout analysis network.

[0028] It should be noted that each page of the import and export document package is rendered as an image, and documents of various formats (including PDFs, scans, images, etc.) are uniformly converted into standardized image data.

[0029] It should be noted that, firstly, considering the differences in size and resolution of different document pages, directly using pixel coordinates to measure the position and area of ​​sections would result in deviations due to different page specifications. Therefore, a normalized coordinate transformation was designed. By dividing pixel coordinates by page width and height, the influence of page physical size and resolution is eliminated, ensuring that the coordinate information of all sections lies within the same spatial order. Secondly, locating the label area requires considering both layout morphology and text semantic features. Relying solely on layout analysis may misclassify non-label areas with similar layouts (such as ordinary table columns or notes columns) as candidates, while relying solely on semantic analysis makes it difficult to accurately pinpoint specific areas within the layout. Therefore, the design combines both approaches, utilizing prior values ​​of layout structure. Capture the morphological rationality of the region and use semantic association confidence. Capture the semantic relevance of content to achieve dual verification of location. Integrate weighted coefficients. and The introduction of this feature is to adapt to the difference in importance between the two types of features in different scenarios. For example, in some documents, the layout features of the markings are more significant, while in other documents, semantic features are more crucial. This is determined through offline training. and Finding the optimal value for the nonlinear activation function can make the fusion result more closely resemble the actual scenario. The logistic function was chosen because it can smoothly map any real number result after linear weighting to... interval, making It has a credible quantitative meaning, which avoids the loss of candidate block information caused by hard threshold segmentation and ensures that the contribution of each candidate block matches its credibility.

[0030] It should be noted that the training samples for the natural language processing network are derived from real document text datasets in import and export LCL (Less than Container Load) business scenarios. These datasets cover OCR-parsed texts of various document types, including Master Bill of Lading (MBL), Hierarchical Bill of Lading (HBL), LCL manifests, and packing lists. Each sample contains text fragments from different sections of the document (such as fixed-field text, table row text, and remarks text), and covers multi-format shipping marks in both Chinese and English (including mixed code strings containing letters, numbers, and special separators, as well as pure letter / number code strings). Additionally, some noisy text fragments (such as character confusion and partially occluded text caused by OCR recognition errors) are included to improve network robustness. The sample label package... It includes two core annotation types: semantic association labels, which annotate the matching result of each text segment with the semantics of the shipping mark / part number (binary label 1 indicates a match, 0 indicates a non-match) and the corresponding true semantic association confidence value (value range [0,1], labeled by domain experts based on the degree of semantic fit between the text and the shipping mark); and morphological structure labels, which annotate the segment type sequence (such as letter segment A, number segment D, separator segment S) and the length ratio of each segment (the proportion relative to the total length of the text segment) for shipping mark text segments. The loss function adopts multi-task joint loss to simultaneously optimize semantic recognition and morphological analysis capabilities, where the semantic matching task uses binary cross-entropy loss. (in the formula) The total number of samples, For the first Semantic association binary labels for each sample The semantic association confidence of the network output (Corresponding sample values), the semantic association confidence regression task uses mean squared error loss. (in the formula) For the first (True semantic association confidence value of each sample), the morphological fragment classification task uses class cross-entropy loss. (in the formula) The total number of segments in all samples. For the first The fragment belongs to type The unique hot tags, (The probability of the segment type output by the network), and the final total loss is obtained by weighted summation using task weight coefficients. (in the formula) (These are non-negative weighting coefficients, optimized through grid search on an offline validation set to ensure a balanced contribution of loss across tasks).

[0031] In a preferred embodiment, based on the normalized area scale distribution of each section in the candidate label section set, a section hierarchy complexity parameter, configured to characterize the degree of fragmentation of the hierarchical structure within a single document package, is calculated, including: Calculate the first Normalized area scale of each section : in, This represents the area of ​​the k-th candidate section. This represents the total area of ​​the page containing the k-th candidate section; It should be noted that, Refers to the first The normalized area scale of each candidate label block, with values ​​ranging from (0,1), serves to eliminate the influence of differences in page sizes, objectively reflecting the relative size of the block within its page, and providing a unified scale measure for calculating page layout complexity parameters. For the first The area of ​​each section This represents the total area of ​​the page containing this section.

[0032] It should be noted that, Refers to the first The area of ​​each candidate label block is calculated in normalized coordinate space. Its function is to serve as a normalized area scale. The molecules provide the size data of the plate itself for scale measurement. The calculation method is based on the geometric region information of the plate. calculate.

[0033] It should be noted that, Refers to the first The total area of ​​the pages containing each candidate label section is always 1 in normalized coordinate space. Its function is to serve as a normalized area scale. The denominator is used to eliminate the impact of differences in page physical dimensions on section scale measurement, ensuring the comparability of section scales across different pages. This is determined by coordinate standardization, as page coordinates are standardized... After the transformation, both the width and height are normalized to 1, so the page area is 1×1=1.

[0034] Constructing about scale Weighted empirical survival statistics : in, The continuous reliable weight, It is an indicator function; It should be noted that, This refers to the set scale threshold, with a value range of (0,1). Its function is to serve as a weighted empirical survival statistic. The judgment criterion is used to statistically summarize the sum of the weights of blocks with a normalized area scale greater than a certain threshold, thus characterizing the distribution features of blocks within different scale intervals. The value can be set according to the common scale range of document scenarios.

[0035] It should be noted that, Regarding scale The weighted empirical survival statistic. Its function is to quantitatively represent the normalized area scale greater than... The sum of the credibility weights of all the label sections reflects the total credibility of sections above the threshold of this scale, providing core statistical data for the regression fitting of the page level complexity parameter.

[0036] It should be noted that, An indicator function, whose rule is: the function value is 1 when the condition within the parentheses is true; the function value is 0 when the condition is false. Its purpose is to filter out values ​​that meet the scaling condition in the calculation of weighted empirical survival statistics. For each block, only the continuous credible weights of such blocks are summed to ensure that the statistics accurately reflect the block characteristics of the target scale range.

[0037] Select several scale sampling points The slope of the following regression equation in the logarithmic field is obtained by using the weighted least squares method. : The solution obtained As the page layout level complexity parameter. Indicates the index of the scale sampling point.

[0038] It should be noted that, The selected scale sampling points This represents the sampling point index, with a value range of (0,1). Its function is to provide multiple discrete scaled data points for logarithmic domain linear regression fitting. These sampling points correspond to... The value is used to construct a regression equation to solve for the page layout hierarchy complexity parameter. To ensure the robustness of the fitting results, the preferred sampling method is logarithmic uniform sampling, selecting 5-20 sampling points within a common scale range.

[0039] It should be noted that, The page layout complexity parameter is a positive real number and is a linear regression equation in the logarithmic field. The absolute value of the slope. Its function is to quantitatively characterize the degree of fragmentation of the hierarchical structure within a single document package. The larger the value, the higher the proportion of small-scale markings and the more fragmented the document structure. The smaller the value, the higher the proportion of large-scale logo sections and the more regular the structure. The value was obtained by using weighted least squares. and The result was obtained by linear regression fitting.

[0040] It should be noted that, This refers to the regression constant term in a logarithmic linear regression equation, which is a real number. Its function is to fit the logarithmic data. and The linear relationship between them ensures that the regression equation can accurately describe the association between the two, and its value is automatically determined by the sampling point data and the fitting algorithm.

[0041] It should be noted that, This serves as the index for the scale sampling points, distinguishing different sampling points and ensuring that the data for each sampling point can be accurately matched during regression fitting.

[0042] It should be noted that, firstly, considering the differences in physical size and resolution among different document pages, directly comparing the actual area of ​​the sections would be affected by page specifications and would not objectively reflect the relative size and distribution characteristics of the sections within the page. Therefore, a normalized area scale is obtained by comparing the area of ​​the section with the total area of ​​the page. First, it eliminates measurement bias caused by differences in page specifications, making the section scales of different pages comparable. Second, the credibility of each section in the candidate section set varies, and some low-credibility sections may be pseudo-candidate regions caused by noise interference. If all sections are directly weighted equally, the statistical results will be distorted. Therefore, continuous credibility weights are introduced. By weighting the empirical survival statistics, higher-confidence segments are given greater weight in the statistics, while the influence of lower-confidence segments is weakened, thereby increasing the statistical value. The accuracy of representing the actual regional scale distribution of shipping marks. Furthermore, based on the structural characteristics of LCL (Less than Container Load) documents, the scale distribution of their shipping marks exhibits typical multi-scale hierarchical characteristics. This characteristic statistically manifests as a power-law distribution, i.e. This nonlinear relationship can be transformed into a linear regression equation through logarithmic transformation. At this point, the regression slope It can directly quantify the steepness of the power-law distribution. The larger the value, the higher the proportion of small-scale markings and the more severe the fragmentation of the hierarchical structure of the document package. Conversely, the smaller the value, the higher the proportion of large-scale markings and the more regular the structure. Through this derivation process, a quantitative mapping from the distribution of marking scale to the degree of fragmentation of hierarchical structure is achieved.

[0043] In a preferred embodiment, generating a multi-level resolution sampling weight distribution based on the layout hierarchy complexity parameter includes: Define the page layout complexity parameter The determined target focuses on the scale value : in, For mapping coefficients, The page layout level complexity parameter; It should be noted that, Refers to the page layout level complexity parameter The determined objective focuses on the scale value. Its role is to abstract the complexity parameters. This is mapped to a specific scale value, achieving an inverse correlation between complexity and the target's focus scale, i.e. The larger (the more fragmented the structure). The smaller.

[0044] It should be noted that, The fundamental coefficients for mapping target focus scales are real numbers. Their function is to set the basic range of the target focus scale and adjust... The overall numerical level is ensured to be within a reasonable scale range, adapting to the common scale range of the document section. The optimal value is obtained through offline training, combined with the scale distribution characteristics of a large amount of document data, and after optimization, the value range is usually [0.1, 2.0].

[0045] It should be noted that, This refers to the adjustment coefficient for the target attention scale mapping, which is a positive real number. Its function is to adjust the page layout hierarchy complexity parameter. Scale of attention to the target The intensity of the impact, The larger, Change-induced The more significant the change, the better. The optimal value is also determined through offline training. Combined with the scale response requirements of documents with different complexities, the optimized value range is [0.5, 3.0].

[0046] Construct an image pyramid and define the first... Equivalent area scale of the layer ; It should be noted that, This refers to the hierarchical index of a multi-level resolution pyramid, with values ​​that are non-negative integers. Its function is to distinguish the different resolution levels of the pyramid. The smaller the value, the lower the image resolution of the corresponding level; The larger the value, the higher the resolution, providing image data of different resolutions for multi-scale recognition.

[0047] It should be noted that, The multi-level resolution pyramid refers to the first level of the pyramid. The equivalent area scale of the layer satisfies The proportional relationship. Its function is to quantitatively represent the first... The scale level corresponding to the resolution of the layer image is used to calculate the target interest scale of that layer. The degree of matching provides a basis. For example... hour, As the benchmark scale; hour, for One-quarter of the area, and so on, with the higher the level, the smaller the equivalent area scale.

[0048] Calculate the first according to the following formula. The weight of the layer in the multi-level resolution sampling weight distribution : in, For bandwidth parameters, This represents the total number of levels in the pyramid.

[0049] It should be noted that, This refers to the bandwidth parameter of the Gaussian kernel function, which is a positive real number. Its function is to control the smoothness of the multi-scale sampling weight distribution. The smaller the value, the more concentrated the weight is on the level that best matches the target's focus scale; The larger the value, the more uniform the weight distribution, balancing the need for local optima and global coverage. The preferred value range is [0.1, 1.0], which can be adjusted according to the scale distribution complexity of the document.

[0050] It should be noted that, In the multi-level resolution sampling weight distribution, the first... The layer weights, ranging from [0,1], have a sum of 1 for all layer weights. Their function is to allocate the contribution ratio of each resolution layer in multi-scale recognition, and the layer is related to the target's scale of interest. The higher the matching degree, The larger the level, the more it ensures that identification resources are tilted towards the level that is appropriate for the current complexity of the document.

[0051] It should be noted that, This refers to the total number of layers in the multi-level resolution pyramid, and is a non-negative integer. Its function is to define the total number of resolution levels in the pyramid, determining the resolution coverage of multi-scale recognition. More layers result in finer resolution gradients, better adapting to different scales of marker blocks, but also increase computational load. The preferred value range is 3-8 layers.

[0052] It should be noted that, This refers to the auxiliary index of the multi-level resolution pyramid, with a value of Its function is to calculate the first... When calculating layer weights, all layers are traversed to complete the normalized summation, ensuring that the weight distribution satisfies the constraint that the sum is 1.

[0053] It should be noted that, firstly, considering the significant multi-scale differences in the shipping mark section of LCL documents, single-resolution processing cannot simultaneously meet the recognition needs of large-scale fields and small-scale rows. Therefore, a multi-level resolution pyramid is constructed to provide corresponding processing resolution levels for shipping mark sections of different scales, where the equivalent area scale... The proportional settings are based on the geometric relationship between image resolution and area to ensure the regularity of scale changes between layers. Secondly, the page layout layer complexity parameter... Directly reflects the degree of fragmentation in the document structure. A larger size means a higher proportion of small-scale markings, requiring more reliance on high-resolution layers for fine-grained identification. Therefore, an exponential decay function is used to reduce the proportion of small-scale markings. Mapped to target attention scale By leveraging the monotonically decreasing property of the exponential function, a precise correlation between complexity and the target scale can be achieved. Used to set the basic scale range. Used for adjustment The strength of the influence on the target scale is determined to adapt the mapping relationship to the scale distribution characteristics of different documents. Furthermore, the introduction of the Gaussian kernel function is to quantify the degree of fit between each resolution level and the target scale of interest. Logarithmic transformation converts multiplicative differences in scale into additive differences, which is more consistent with the distance metric logic in numerical computation. The bandwidth parameter... This is used to control the smoothness of the weight distribution. The smaller the value, the more concentrated the weight is on the level that best matches the target scale. The larger the value, the more uniform the weight distribution, thus balancing the need for local optima and global coverage. Finally, normalization transforms the matching degree of each level into a percentage form, ensuring that the weight distribution satisfies the constraint that the summation is 1. This allows high-matching-degree levels to play a greater role in the recognition process, achieving adaptive allocation of resolution resources.

[0054] In a preferred embodiment, the weights determined by the multi-level resolution sampling weight distribution are used to perform multi-scale identification and probability fusion on the candidate mark block set, including: Get the The section in the pyramid Layer, First The character position belongs to the character recognition probability ; It should be noted that, This is a sequence index for character positions, used to distinguish different character positions within a string, providing a positional index for calculating and fusing the probability distribution of character sequences. The value range is determined based on the specific scenario, and the sampling point sequence number... The value ranges from 1 to the total number of sampling points, and the character position number. The value can range from 1 to the maximum length of the string.

[0055] It should be noted that, The first character in the character sequence The character at position 1 ,in This is a character table. Its purpose is to serve as the object of character recognition probability calculation, specifying the possible character types at each position and providing a foundation for constructing the probability distribution of character sequences. Character table It covers common letters (uppercase and lowercase), numbers, and special separators (such as -, / , *, etc.) used in import and export documents.

[0056] It should be noted that, This refers to a complete character table containing all character types that may appear in the shipping marks on import and export documents. Its purpose is to define the scope of character recognition, ensuring that the calculation of the probability distribution of character sequences covers all possible characters and avoiding recognition errors due to missing characters. The construction of the character table should refer to the commonly used character specifications for international logistics documents, typically including 26 English letters (uppercase and lowercase), 10 Arabic numerals, and common special separators, totaling 50-100 characters.

[0057] It should be noted that, Refers to the first The section in the pyramid Layer, First The character position belongs to the character The recognition probability is in the range of [0,1], and for the same , , ,all The sum of the probabilities is 1. Its function is to quantify the reliability of recognizing a character at that level and position. It is obtained by using a character recognition model (such as CNN) to analyze the first... The images are obtained by recognizing the blocks of the layer.

[0058] The training samples for the CNN model are derived from real document text datasets in import and export LCL (Less than Container Load) business, covering OCR parsing texts of various documents such as master bills of lading and sub-bills of lading, including bilingual (Chinese and English) multi-format shipping marks and noisy text fragments. Sample labels include semantic association labels (binary matching labels and true semantic association confidence values) and morphological structure labels (fragment type sequence and length proportion). The loss function employs a multi-task joint loss, including binary cross-entropy loss for semantic matching, mean squared error loss for semantic association confidence regression, and category cross-entropy loss for morphological fragment classification, which are weighted and summed to obtain the total loss. When the model outputs the recognition probability of the k-th segment belonging to character c_t at the ℓ-th level of the pyramid and the t-th character position, the segment image at that level is first input into the CNN. After local feature extraction by convolutional layers and dimensionality reduction by pooling layers, it is mapped to the character table dimension through fully connected layers, and then normalized by the softmax function to obtain the probability distribution of each character at that position.

[0059] Utilizing the weights in the multi-level resolution sampling weight distribution Multiplicative fusion of probabilities from each layer yields a non-normalized fused probability. : It should be noted that, Refers to the first The first section The non-normalized fusion probability of each character position. Its function is to integrate the recognition probabilities of each resolution level, and to amplify the consistency confidence of each level through multiplicative fusion, while suppressing the divergence confidence.

[0060] right Normalization yields the final probability distribution The final string is obtained by decoding using the maximum a posteriori probability. : It should be noted that, Refers to the first The first section The normalized final probability distribution of each character position takes values ​​in the range [0,1]. Its function is to transform the non-normalized fusion probability into a standard distribution that conforms to probability theory, ensuring that the sum of the probabilities of each character is 1.

[0061] It should be noted that, Refers to the first The final markup string is obtained after multi-scale identification and decoding of candidate markup sections. Its function is to serve as the piece-level identification information corresponding to that section. It is obtained through a maximum a posteriori probability decoding algorithm. It is obtained through decoding.

[0062] It should be noted that, This refers to the candidate set of strings, containing all possible combinations of character sequences. Its purpose is to serve as the candidate space for maximum a posteriori probability decoding; the decoding algorithm selects the character sequence with the largest probability product from this set as the final string. The size of the candidate set is determined by the character table. Size and maximum string length Decide.

[0063] It should be noted that, This refers to the maximum length of the character sequence, and is a positive integer. Its purpose is to define the length range of the string, ensuring that character recognition and probability fusion can cover the longest possible shipping mark string, avoiding character omissions due to insufficient length. The preferred value is set based on the common length of import / export document shipping marks, typically 10-50 characters.

[0064] It should be noted that, firstly, considering the varying recognition capabilities of different resolution levels for the logo segment, high-resolution levels can capture the fine character features of small-scale segments, while low-resolution levels can avoid over-segmentation of characters in large-scale segments. Therefore, performing character recognition and outputting probability distributions at each level separately can fully preserve effective recognition information at different scales. Secondly, multiplicative fusion is chosen instead of additive fusion because multiplicative operations can amplify the consistency confidence of each level in recognizing the same character, while suppressing divergent confidence, making the fusion result more prominent in characters recognized by most levels, while reducing the level weights. As an index, it enables non-linear adjustment of the probability distribution by weights. The higher the weight of a level, the stronger the dominance of its probability distribution in the fusion process, thus allowing the fusion result to accurately match the hierarchical complexity characteristics of the document package. Normalization is introduced to transform the non-standard probability distribution after multiplicative fusion into a standard distribution conforming to probability theory, eliminating interference caused by differences in probability values ​​across different sections and levels, and providing a unified input format for the decoding algorithm. Finally, maximum a posteriori probability decoding is used because this algorithm can directly select the character with the highest probability at each character position to form a string. While ensuring decoding efficiency, it minimizes the cumulative impact of character recognition errors, ensuring that the final output label string accurately reflects the document-level identity information.

[0065] In a preferred embodiment, based on the multi-scale recognition results and page layout geometric attributes, a character structure fragmentation metric configured to characterize structural fragmentation bias is calculated, including: Statistical recognition of strings Number of fragments With total length Calculate the morphological complexity : It should be noted that, Refers to the first Individual header string The number of fragments after decomposition is a positive integer. Its purpose is to quantify the degree of fragmentation of the string; the more fragments, the more fragmented the string's structure, providing basic data for morphological complexity calculations. The decomposition rule is to... Divide into letter segments, number segments, and separator segments.

[0066] It should be noted that, Refers to the first Individual header string The total length, i.e., the number of characters contained in the string, is a positive integer. Its purpose is to serve as the denominator in morphological complexity calculations, reducing the number of segments... Normalization eliminates the impact of string length differences on fragmentation measurement, ensuring that the morphological complexity of strings of different lengths is comparable.

[0067] It should be noted that, Refers to the first The morphological complexity of the header string, with a value ranging from (0,1), is used to quantify the degree of fragmentation in the internal structure of the string. The larger the value, the more fragments there are within a unit length, and the more fragmented the string is. The smaller the value, the more regular the string structure, providing core data for calculating the fragmentation metric of character structure.

[0068] Based on the page layout hierarchy complexity parameter Calculate the expected scale morphological equilibrium value : in, This is a preset constant; It should be noted that, Refers to the complexity parameter based on page layout level The calculated expected scale-form balance value. Its purpose is to characterize the proper balance between scale and form under the current level of document complexity, serving as a benchmark for judging the degree of fragmentation in character structure. The larger, The larger the value, the better it can adapt to string formats with higher degrees of fragmentation.

[0069] It should be noted that, This refers to the fundamental constant of the expected scale morphological equilibrium value, and is a real number. Its function is to set the basic level of the equilibrium value, ensuring... The values ​​are within a reasonable range and are suitable for common scale-morphology coupling relationships. The optimal values ​​are obtained through offline training and, combined with the scale-morphology correlation features of a large amount of single-document data, the optimized range is [-2.0, 0.5].

[0070] It should be noted that, This refers to the adjustment constant for the desired scale-based morphological balance value, and is a positive real number. Its function is to adjust the page layout's hierarchical complexity parameter. right The extent of the impact The larger, Change-induced The more significant the change, the more accurately the balance value can be adapted to documents of varying complexity. The preferred value range is [0.3, 1.5], determined through offline training and optimization.

[0071] The character structure fragmentation metric is calculated using the following formula. : in, For continuous reliable weights, For normalized area scale, This is the morphological adjustment coefficient.

[0072] It should be noted that, This refers to the morphological adjustment coefficient, which is a positive real number. Its function is to calibrate the morphological complexity. Logarithmic value of normalized area scale The difference in dimensions ensures that the summation of the two values ​​accurately reflects the coupling relationship between scale and form, avoiding fragmented measurement bias caused by inconsistent dimensions. The preferred value range is [0.5, 2.0], which can be adjusted according to the morphological distribution characteristics of the document.

[0073] It should be noted that, This refers to a non-negative real number that measures character structure fragmentation. Its function is to quantify the scale-driven character structure fragmentation bias. The larger the value, the more severe the fragmentation of the string and the greater the deviation from the expected scale-based balance value, thus providing a basis for the fragmentation dimension in comprehensive risk assessment.

[0074] It should be noted that, firstly, considering that fragmentation of the shipping mark string in LCL documents is directly manifested as an increased frequency of alternation between letters, numbers, and delimiters, simply counting the number or length of fragments cannot reflect the degree of fragmentation alone. Therefore, the ratio is calculated by combining both. This allows morphological complexity to stably quantify the structural fragmentation features within the string, avoiding biases caused by single-dimensional measurement. Secondly, the layout-level complexity parameter... It directly reflects the degree of fragmentation in the hierarchical structure of the documents. The larger the value, the higher the proportion of small-scale label sections, and the corresponding string morphology complexity should also show a corresponding trend. Therefore, a linear formula is designed. ,pass Set a basic balance level. adjust The magnitude of the impact on the balance value allows the desired scale morphological balance value to adapt to the hierarchical structural characteristics of the document, ensuring the relevance of the measurement. Furthermore, the normalized area scale... The logarithm can transform the multiplicative difference of scale into an additive difference, along with the weighted morphological complexity. Achieving adaptation in the numerical dimension, morphological adjustment coefficient This is used to calibrate the dimensional difference between morphological complexity and the logarithmic value of scale, ensuring that the sum of the two accurately reflects the coupling relationship between scale and morphology. The square operation of the difference highlights the degree of deviation from the desired equilibrium value while avoiding the cancellation of positive and negative deviations, and ensures continuous and reliable weights. The introduction of this method aims to ensure that deviations from high-confidence blocks dominate the final measurement, weakening the interference from low-confidence pseudo-candidate blocks. The weighted average processing further integrates information from all valid blocks, making... It can comprehensively and objectively quantify the scale-driven character structure fragmentation bias.

[0075] In a preferred embodiment, based on the multi-scale recognition results and page layout geometric attributes, a regional positioning volatility metric value configured to characterize the dispersion of cross-page layout landing points is calculated, including: Construct the first Geometric anchor vectors of each section : in, For the normalized center coordinates of the section, This refers to the main axis direction of the section. It should be noted that, Refers to the first The geometric anchor vector of each candidate label block is a four-dimensional vector. Its function is to integrate the position information (center coordinates) and orientation information (main axis direction) of the block, transforming two-dimensional position and one-dimensional angle into a unified vector form.

[0076] It should be noted that, (Center x-coordinate) refers to the first Normalization Center for Individual Brand Candidate Sections The coordinates range from [0,1]. Their function is to represent the center position of the section in the horizontal direction of the page, providing core data for the positional dimension of the geometric anchor vector.

[0077] It should be noted that, (Center y-coordinate) refers to the first Normalization Center for Individual Brand Candidate Sections Coordinates, ranging from [0,1]. Their function is to represent the center position of the section in the vertical direction of the page, relative to... Together they form the central coordinates of the section.

[0078] It should be noted that, Refers to the first The principal axis angle of each candidate label block is measured in radians, ranging from 0 to π. Its function is to characterize the main extension direction of the block, providing core data for the directional dimension of the geometric anchor vector, thus avoiding the inability to fully depict the distribution characteristics of the blocks solely through positional information. It is obtained by calculating the direction of the longer side of the block's smallest bounding rectangle.

[0079] Calculate the weighted covariance matrix of the geometric anchor point distribution. : in, For continuous reliable weights, This is the weighted average of the geometric anchor vectors; It should be noted that, This refers to the weighted average of the geometric anchor point vectors of all candidate sectors, and is a four-dimensional vector. Its function is to characterize the average level of the geometric anchor points of all high-confidence sectors, serving as a benchmark for measuring the deviation of the anchor points of individual sectors.

[0080] It should be noted that, This refers to the weighted covariance matrix of the geometric anchor point distribution, which is a 4×4 symmetric matrix. Its function is to quantitatively characterize the dispersion of the geometric anchor point vectors of all high-confidence blocks, and the elements in the matrix reflect the discrete characteristics of the correlation between different dimensions.

[0081] Calculate the regional positioning volatility metric. : in, Represents the trace of a matrix.

[0082] It should be noted that, Weighted covariance matrix The trace, i.e., the sum of the elements on the main diagonal of the matrix, is used to comprehensively extract the discrete information of the geometric anchor vector across all dimensions. It transforms the multidimensional discrete characteristics of the 4×4 covariance matrix into a single scalar, avoiding the limitations of single-dimensional measurements and providing core calculation results for measuring the volatility of regional positioning. The calculation method involves... The diagonal elements are summed sequentially.

[0083] It should be noted that, This refers to a measure of the volatility of the location of the banner / brand banner within a multi-page layout; it is a non-negative real number. Its function is to quantify the dispersion of the banner / brand banner's placement within a page. The larger the value, the more dispersed the center position and main axis direction of the block are, and the worse the positioning stability. The smaller the value, the more concentrated the distribution of sectors and the more stable the positioning, providing a basis for the geometric distribution dimension of comprehensive risk assessment.

[0084] It should be noted that, firstly, considering that the discrete characteristics of the marking plate are not only reflected in its planar position but also involve differences in the main axis direction, coordinates alone cannot fully describe the discreteness of the landing point shape. Therefore, the center coordinates are used... With the main axis direction Combined with the construction of geometric anchor vectors ,in and The introduction of this feature aims to transform the periodic parameter of angle into a linearly computable value, avoiding measurement bias caused by angle cycles and integrating directional differences into a unified vector operation system. Secondly, the candidate mark set contains low-confidence pseudo-candidate regions. Directly using ordinary mean and covariance calculations would distort the characterization of dispersion due to interference from these regions. Therefore, continuous confidence weights are introduced. Weighting the mean and covariance matrices allows high-confidence sectors to contribute more weight in the statistical process, while the influence of low-confidence sectors is reasonably weakened, ensuring... and It can accurately reflect the distribution characteristics of the actual marking area. Weighted covariance matrix. Each element corresponds to the discrete correlation between different dimensions of the anchor vector, while the trace of the matrix, as the sum of the diagonal elements, can comprehensively extract discrete information from all dimensions, avoiding the one-sidedness caused by a single-dimensional measure. It can comprehensively and objectively quantify the dispersion of the landing point of the header section in a double-page spread.

[0085] In a preferred embodiment, a comprehensive risk assessment model is constructed, comprising the page layout complexity parameter, character structure fragmentation metric, and regional positioning volatility metric, including: Define dimensional reduction function : It should be noted that, This refers to the dimensional reduction function. Its function is to measure the fragmentation of character structures. and regional positioning volatility measure Smoothing mapping is performed to compress the numerical span and standardize the units of measurement without changing the relative size relationship of the data, thus avoiding fusion bias caused by excessive differences in the magnitude of the two types of metrics.

[0086] The fragmentation metric of the character structure is measured using the dimensional reduction function. and the regional positioning volatility metric Process it; Constructing a comprehensive risk linear potential energy : in, The page layout level complexity parameter is... to These are the preset weighting coefficients. This is an adaptive penalty for the cross term.

[0087] It should be noted that, This refers to the linear potential energy of comprehensive risk, and is a real number. Its function is to integrate three core risk factors: layout hierarchy complexity, fragmentation bias, and positioning volatility, as well as the synergistic effect of complexity and fragmentation, to form a unified linear potential energy indicator. As the reference constant term, Characterizing the direct risk contribution of structural complexity, and To separately characterize the risk contribution of fragmentation and locational volatility, To characterize the synergistic risk contribution of the two.

[0088] It should be noted that, The pre-defined weighting coefficients for the linear potential energy of comprehensive risk are all real numbers, where... Typically, it is a non-negative real number. Its function is to adjust the contribution ratio of each risk factor in the overall potential energy, adapting to the differences in the importance of various risks under different document scenarios. The optimal value is obtained through offline training. Specifically, it is achieved by using a document dataset containing a large number of labeled risk levels and performing grid search or gradient descent optimization with the accuracy of risk assessment as the target. The value is usually in the range of [0.1, 5.0].

[0089] It should be noted that, The intelligent review result refers to a value range within... The probability scalar between these values. Its function is to quantify the observability risk of item-level identity in the LCL (Less than Container Load) scenario. The larger the value, the higher the risk, and the smaller the value, the lower the risk, providing users with intuitive and standardized review conclusions.

[0090] It should be noted that, firstly, the character structure fragmentation metric is taken into account. and regional positioning volatility measure The values ​​may vary considerably, and directly performing a linear combination can lead to fusion bias caused by differences in numerical magnitudes. Meanwhile, the dimensional reduction function... It possesses the characteristics of monotonically increasing and compressing large values ​​while preserving differences for small values. It can map two types of measures to similar numerical ranges without altering the relative size of the data, eliminating interference from inconsistent dimensions and ensuring that each factor has an equal basis for weight adjustment during the fusion process. Secondly, comprehensive risk assessment requires a full integration of structural features and two types of derived risk measures, along with parameters related to page layout complexity. It is a core indicator for describing the structure of documents. and Fragmentation bias and positioning discreteness were quantified separately, and these three factors were incorporated into the formula as independent weighting terms. This allows the formula to capture risk contributions from different dimensions, while the weighting coefficients... to The introduction of this allows for adjusting weights based on the importance of each factor in the actual scenario, enabling the model to adapt to the risk distribution characteristics of different documents. Furthermore, the page layout complexity parameter... With smoothed fragmentation metric There is an inherent connection between them. The larger the value, the more fragmented the document structure. In this case, the impact of fragmentation bias on risk should be amplified accordingly. Therefore, a product cross term should be designed. ,pass Adjusting the strength of this correlation allows for adaptive enhancement of risk contribution, enabling the model to accurately capture the synergistic effect of structural complexity and fragmentation bias. Preset constant terms are used. This serves as a benchmark level for calibrating the overall risk linear potential energy, ensuring... The numerical range meets the input requirements of nonlinear mapping.

[0091] In a preferred embodiment, a comprehensive risk assessment model outputs intelligent audit results that quantify the observability risk of document-level identity, including: Using the Logistic Stir Function The comprehensive risk linear potential energy Mapped to intelligent review results : in, For the range of values ​​within The scalar between these values ​​is used to quantify the risk to observability of item-level identity caused by the multi-scale cascade structure in the consolidation scenario.

[0092] It should be noted that, firstly, the linear potential energy of comprehensive risk is taken into account. The value may cover a range of positive and negative infinity, and different document packages correspond to... The lack of a unified comparison benchmark makes it difficult for users to understand and apply the numerical results directly as audit findings. Therefore, it is necessary to transform them into scalars within a fixed interval through nonlinear mapping. The logistic function was chosen because it possesses the property of monotonically increasing, which can fully preserve... The inherent risk level ranking relationship, and its S-shaped curve, can achieve a reasonable characterization of risk, that is, when... As it approaches negative infinity, Approaching 0 corresponds to extremely low risk; when As it approaches positive infinity, A value approaching 1 corresponds to extremely high risk; while... When the value is in the middle range, The rate of change is moderate, enabling accurate differentiation of different levels of medium risk and avoiding oversensitivity or insensitivity in risk assessment. The output range is limited to... This is because the range aligns with conventional understanding of risk quantification, making it easy for users to quickly grasp the risk level corresponding to the numerical value. It is also compatible with the risk scoring system of existing review systems and can be applied directly without additional scale conversion. Furthermore, the numerical form of this range naturally possesses probabilistic representation, which is highly consistent with the quantitative requirements of document-level identity observability risk.

[0093] like Figure 2 As shown, Figure 2The left side first demonstrates the basics of LCL (Less than Container Load) operations: multiple shipments are loaded into a single container at the container freight station, generating a master bill of lading covering the entire container, multiple sub-bills of lading corresponding to each shipment, a consolidated list detailing the loading status, and a packing list. All documents contain core identifiers for identifying the shipment at the piece level, namely shipping marks / part numbers, and exhibit multi-scale cascading layout features due to different document types. The middle section connects to the digital processing stage, converting the document package into processable document data before initiating a CV-NLP integrated review mechanism. The right side details the review steps: first, the images of the import / export document packages are reviewed... The sequence undergoes image coordinate standardization. A set of candidate label blocks with continuous, reliable weights is generated through joint localization using semantic probability and prior layout attributes. Then, based on the normalized area scale distribution of this set, a layout hierarchy complexity parameter representing the degree of fragmentation in the document's hierarchical structure is calculated. Based on this parameter, a multi-level resolution sampling weight distribution is generated, and multi-scale identification and probability fusion are performed on the candidate blocks. Subsequently, character structure fragmentation and regional positioning volatility measures are calculated by combining layout geometric attributes. Finally, a comprehensive risk assessment model integrating the above key parameters outputs an intelligent review result that quantitatively represents the observable risk of document-level identity. It clearly connects the technical links of LCL (Less than Container Load) business scenarios, document system characteristics, and CV-NLP integrated review.

[0094] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A digital import and export intelligent auditing method based on a CV-NLP fusion algorithm, characterized in that, The image sequence of the import and export document package is subjected to image coordinate standardization processing, and a mark head candidate block set containing continuous confidence weights is generated through joint positioning of semantic probability and layout prior attributes. According to the normalized area scale distribution of each block in the mark head candidate block set, a layout level complexity parameter configured to represent the fragmentation degree of the hierarchical structure in the document package is calculated. Based on the layout level complexity parameter, a multi-level resolution sampling weight distribution is generated, and the mark head candidate block set is subjected to multi-scale recognition and probability fusion using the weight determined by the distribution. Based on the multi-scale recognition result and the layout geometric attribute, a character structure fragmentation metric value configured to represent the structure fragmentation bias and a region positioning volatility metric value configured to represent the dispersion degree of the cross-page layout landing point are respectively calculated. An integrated risk assessment model containing the layout level complexity parameter, the character structure fragmentation metric value and the region positioning volatility metric value is constructed, and an intelligent audit result quantitatively representing the document level identity observability risk is output by the integrated risk assessment model. The image sequence of the import and export document package is subjected to image coordinate standardization processing, and a mark head candidate block set containing continuous confidence weights is generated through joint positioning of semantic probability and layout prior attributes, including:

2. The digital import and export intelligent auditing method based on the CV-NLP fusion algorithm according to claim 1, characterized in that, Each page of document in the import and export document package is rendered into an image, and the pixel coordinates in the image are mapped to normalized physical coordinates with values between zero and one by dividing the image width and height; The layout structure prior value of the candidate block is extracted using a layout analysis network, and the semantic association confidence of the candidate block is extracted using a natural language processing network; The semantic association confidence and the layout structure prior value are linearly weighted and summed, and the sum is input into a nonlinear activation function for mapping to obtain the continuous confidence weight of each candidate block, thereby constructing the mark head candidate block set. According to the normalized area scale distribution of each block in the mark head candidate block set, a layout level complexity parameter configured to represent the fragmentation degree of the hierarchical structure in the document package is calculated, including:

3. The digital import and export intelligent auditing method based on the CV-NLP fusion algorithm according to claim 1, characterized in that, The area ratio of each block in the mark head candidate block set to the total area of the page where it is located is calculated to obtain the normalized area scale; With the continuous confidence weight as the weighting item, the sum of the weights of the blocks with normalized area scale greater than the set threshold is calculated to construct a weighted empirical survival statistic varying with scale; In the logarithmic coordinate system, the relationship between the weighted empirical survival statistic and the normalized area scale is linearly regressed and fitted, and the absolute value of the slope value obtained by fitting is taken as the layout level complexity parameter. Based on the layout level complexity parameter, a multi-level resolution sampling weight distribution is generated, including:

4. The digital import and export intelligent auditing method based on the CV-NLP fusion algorithm according to claim 1, characterized in that, A multi-level resolution pyramid of the input image is constructed, and the equivalent area scale of each level of the pyramid is determined; The layout level complexity parameter is mapped to a target attention scale value using an exponential decay function, so that the higher the complexity, the smaller the target attention scale value; The matching degree between the equivalent area scale of each level of the pyramid and the target attention scale value is calculated using a Gaussian kernel function; ​ The matching degrees of all levels are normalized, and the obtained normalized values are taken as weight values of the corresponding levels in the multi-level resolution sampling weight distribution.

5. The digital import and export intelligent auditing method based on the CV-NLP fusion algorithm according to claim 1, characterized in that, Multi-scale recognition and probability fusion are performed on the label candidate block set by using the weights determined by the multi-level resolution sampling weight distribution, including: For each block in the label candidate block set, character recognition is performed on each level of the image pyramid, and a character sequence probability distribution of the level is output. The weights of each level in the multi-level resolution sampling weight distribution are taken as exponents, and the character sequence probability distributions of each level are multiplied and weighted to obtain a fused non-normalized probability distribution. The non-normalized probability distribution is normalized, and a final string result is parsed from the normalized probability distribution by using a sequence decoding algorithm.

6. The digital import and export intelligent auditing method based on the CV-NLP fusion algorithm according to claim 1, characterized in that, Based on the multi-scale recognition result and the layout geometric attribute, a character structure fragmentation metric value configured to represent a structure fragmentation bias is calculated, including: The recognized string is decomposed into a combination sequence of letter segments, number fields, and separator segments, the ratio of the number of segments in the string to the total length of the string is calculated, and a morphological complexity is obtained. An expected scale morphological balance value at the current level is calculated using the layout level complexity parameter; The sum of the logarithmic value of the normalized area scale of each block and the weighted adjusted morphological complexity is calculated, and the square of the difference between the sum and the expected scale morphological balance value is calculated. The square of the difference is weighted and averaged using the continuous credible weight, and the weighted average result is taken as the character structure fragmentation metric value.

7. The CV-NLP fusion algorithm-based digital import and export intelligent auditing method according to claim 1, characterized in that, Based on the multi-scale recognition result and the layout geometric attribute, a region positioning volatility metric value configured to represent the dispersion degree of the cross-page layout landing point is calculated, including: The normalized center coordinates and principal axis direction angles of each block in the label candidate block set are extracted to construct a geometric anchor point vector of each block. The weighted covariance matrix of the geometric anchor point vectors of all blocks relative to the weighted mean vector is calculated using the continuous credible weight. The trace of the weighted covariance matrix is calculated, and the value of the trace is taken as the region positioning volatility metric value.

8. The digital import and export intelligent auditing method based on the CV-NLP fusion algorithm according to claim 1, characterized in that, A comprehensive risk assessment model including the layout level complexity parameter, the character structure fragmentation metric value, and the region positioning volatility metric value is constructed, including: The character structure fragmentation metric value and the region positioning volatility metric value are respectively smoothed by using a natural logarithm function to obtain a smoothed fragmentation metric value and a smoothed volatility metric value. A linear weighted combination formula is constructed, and the weighted items of the formula include at least the layout level complexity parameter, the smoothed fragmentation metric value, the smoothed volatility metric value, and the product of the layout level complexity parameter and the smoothed fragmentation metric value. The weighted coefficients are summed by using a preset weighted coefficient, and the sum is taken as a comprehensive risk linear potential energy, thereby completing the construction of the comprehensive risk assessment model.

9. The digital import and export intelligent auditing method based on the CV-NLP fusion algorithm according to claim 8, characterized in that, An intelligent audit result quantitatively representing the piece-level identity observability risk is output by the comprehensive risk assessment model, including: inputting the linear potential energy of the comprehensive risk output by the comprehensive risk assessment model into a logistic function for nonlinear mapping; calculating an output value of the logistic function to obtain a probability scalar with a value between zero and one; taking the probability scalar as the intelligent auditing result, wherein the greater the numerical value of the result is, the higher the part-level identity observability risk under the current page-level complexity is.

Images