Image recognition-based commercial password application security assessment evidence identification method

Abstract

The application discloses a commercial password application security evaluation evidence identification method based on image recognition and relates to the technical field of image recognition.The application aims to evaluate the employee senior engineer certificate, extracts information from a multi-modal document by using a Qwen-VL visual language model, generates a vector and labels, is converted into a structured evidence node, is stored in a graph database, a plurality of edges are established according to different correlations, a preliminary evidence network is constructed, certificate nodes are identified, a certificate entity group is formed by clustering, relevant scores are calculated by forward tracing to find supporting evidence, relevant scores are calculated by reverse tracing to find contradictory evidence, the credibility of the certificate entity group is dynamically adjusted, and the influence is propagated according to the updated credibility; abnormal certificates below a threshold value are screened, causes and contradiction points are analyzed, abnormal information is listed, key evidence, contradiction points and correlation paths are visually displayed, and a text description is generated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image recognition, in particular to a commercial cryptographic application security evaluation evidence identification method based on image recognition. BACKGROUND

[0002] In personnel qualification certification, contract audit, supply chain traceability and many other scenarios, it is crucial to evaluate the authenticity, integrity and consistency of the submitted certificates and documents.

[0003] Traditional certificate evaluation methods often rely on manual review, making it difficult to handle massive data and ensuring the consistency and objectivity of the evaluation results. Existing technologies usually process certificates and evidence in isolation, making it difficult to effectively associate and form a complete evidence chain, resulting in one-sided evaluation results. Traditional verification methods often focus on a single dimension, making it difficult to systematically check the comprehensive consistency or matching degree of certificates and associated evidence in multiple dimensions such as content, format, seal, source and context, and unable to automatically and meticulously compare subtle differences. Moreover, the evaluation is often static, lacking a dynamic adjustment mechanism, and once some key fields match or a certain anti-fake marker is identified, the certificate is easily determined to be valid, and the credibility evaluation of the certificate cannot be dynamically updated according to the mutual support or contradictory relationship between evidence, as well as the quality and timeliness of the evidence. SUMMARY

[0004] The present application relates to the technical field of image recognition, in particular to a commercial cryptographic application security evaluation evidence identification method based on image recognition.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0006] In a first aspect, the present application provides a commercial cryptographic application security evaluation evidence identification method based on image recognition, comprising:

[0007] To evaluate the senior engineer certification certificate submitted by the employee, the Qwen-VL visual language model is used to extract information from the multi-modal document; a high-dimensional vector representation is generated for each extracted item, and its source, type and preliminary quality score are labeled, and it is converted into a structured evidence node;

[0008] The evidence node is stored in a graph database, and multiple edges are established according to the physical location, semantic similarity and reference relationship to form a preliminary evidence network, and the nodes are given an initial credibility; in the network, identify the certificate node, cluster its related nodes to form a certificate entity group, and give it a comprehensive initial credibility;

[0009] Through forward tracing, starting from the certificate entity group, find supporting evidence along the reference association edge, and adjust the credibility of the certificate entity group by checking the credibility, consistency, trusted list matching degree and context association of the supporting evidence;

[0010] By reverse tracing, check if there is contradictory evidence, low confidence strong association, context conflict and outdated and replacement clues pointing to the certificate, and if found, reduce the credibility of the certificate entity group;

[0011] According to the results of forward and reverse tracing, update the credibility of the certificate entity group, and propagate the credibility changes along the reference edge to affect the credibility of the related support and contradictory evidence nodes;

[0012] Traverse all certificate entity groups, filter out those below the credibility threshold and mark them as abnormal, analyze the causes of low credibility and contradictions, list the abnormal certificate information, visually display the key evidence, contradictions and their associated paths that cause the abnormality, and generate a written description.

[0013] In combination with the first aspect, in a first implementation manner of the first aspect of the present application, the senior engineer certification certificate submitted by the employee is evaluated, and information is extracted from the multi-modal document by using the Qwen-VL visual language model, including:

[0014] Read the multi-modal document of the submitted senior engineer certification certificate, including text, pictures, tables and files; use the Qwen-VL visual language model to identify and extract key information in the document as information items, including the name of the certificate holder, the certificate number, the issuance date, the name of the issuing agency, the certificate level, the certificate validity period, the signature on the certificate, the seal pattern and the text, and the certificate photo.

[0015] In combination with the first aspect, in a second implementation manner of the first aspect of the present application, a high-dimensional vector representation is generated for each extracted item, the source, type and preliminary quality score are labeled, and it is converted into a structured evidence node, including:

[0016] A high-dimensional vector representation is generated for each extracted information item, a source label is created to record the specific information of the source document, and the type of the information item is labeled; a preliminary quality score is given based on model confidence, content clarity and integrity; the extracted information item content, high-dimensional vector representation, source label, type label and preliminary quality score are structured and encapsulated, and converted into a structured evidence node.

[0017] In combination with the first aspect, in a third implementation manner of the first aspect of the present application, the evidence node is stored in the graph database, multiple edges are established according to the physical location, semantic similarity and reference relationship, a preliminary evidence network is formed, and an initial credibility is given to the node, including:

[0018] reading and parsing the information items and the attached metadata contained in each encapsulated evidence node, in the graph database, according to the physical location relationship of the information items in the original document, establishing edges representing spatial proximity for information nodes that are spatially adjacent or arranged in order, as spatial proximity edges, and assigning the label of the proximity field; using the high-dimensional vector representation of the information item content, identifying information nodes that are highly semantically related or may have mutual corroboration relationships by calculating the cosine similarity between vectors, and establishing edges representing semantic association for these nodes, as semantic similarity edges, and assigning the label of semantic association or mutual corroboration; according to the reference or derivation relationship between the information items, establishing edges representing reference dependence, as reference association edges, and assigning the label of reference relationship or dependence relationship; by analyzing the physical location, semantic similarity and reference relationship and creating corresponding edges, connecting the nodes to form a multi-dimensional preliminary evidence network;

[0019] When assigning the initial credibility to the nodes, the preliminary quality score of the information items is extracted, the importance of the information items, the authority of the issuing agency and the connection mode of the information items in the evidence network are analyzed; through a preset weighting calculation method, a specific initial credibility value is calculated and assigned to each node in the graph database, representing the probability that the information item is identified as real in the initial evaluation stage.

[0020] In combination with the first aspect, in a fourth implementation manner of the first aspect of the present application, the identifying the certificate nodes in the network, clustering the related nodes of the certificate nodes, forming a certificate entity group, and assigning a comprehensive initial credibility, comprises:

[0021] In the initial evidence network, the Louvain algorithm is used to identify nodes representing the same certificate entity, and all related nodes directly or indirectly connected to the certificate nodes are found; these certificate nodes and all related nodes are regarded as a whole, and the Leiden algorithm is used to aggregate them to form a certificate entity group; according to the initial credibility values of all nodes in the entity group, the connection strength between the nodes and the overall structural characteristics of the entity group in the network, a random walk-based algorithm is used to calculate and assign a comprehensive initial credibility value to the certificate entity group.

[0022] In combination with the first aspect, in a fifth implementation manner of the first aspect of the present application, the adjusting the credibility of the certificate entity group by tracing the support evidence along the reference association edges from the certificate entity group, by checking the credibility, consistency, trusted list matching degree and context association of the support evidence, comprises:

[0023] In the graph database, the nodes representing the high-level engineer certification certificate and their initial attributes are located by Cypher query, all evidence nodes claiming to support the validity of the certificate are found along the reference association edge, and the initial quality score of the evidence nodes is read; the key field values of the certificate entity group are extracted, the associated evidence nodes are found in the graph database, the corresponding field information is extracted, and the basic consistency between the certificate internal information and the certificate and each piece of evidence is evaluated;

[0024] Check if the seal is included in the evidence, and extract its key features; for image evidence, use image processing techniques to locate the seal area based on shape, color, texture or position, and crop the main seal image and the evidence seal image; for text or digital evidence, check if the seal and its features are explicitly mentioned; when there is no seal information in the evidence, skip the comparison; for evidence with successfully extracted seal images, perform multi-dimensional feature comparison; calculate the sharpness of the main seal and the evidence seal by Laplacian operator variance to determine whether they are consistent; perform color quantization to compare the average RGB values of the main pixels to determine whether the colors are consistent; apply Canny operator for edge detection to analyze the shape features of the contour to determine whether the shapes are consistent; use optical character recognition technology to extract the seal text, compare the text content after standardization to determine whether the text is consistent; summarize the comparison results of each evidence in the sharpness, color, shape and text dimensions to form the seal state set of the evidence, and add it to the overall state set; when the seal does not match in any key feature and the evidence presents the seal, generate a seal mismatch state, and do not generate this state if it is completely matched or there is no seal;

[0025] Compare the consistency of the issuing authority name, extract the full name of the main authority, find it directly in the text evidence, and use optical character recognition for image evidence; record all related authority names found as the evidence authority name set, and standardize these names and the full name of the main authority; traverse the evidence authority name set to determine whether each name is exactly the same as the full name of the main authority; when there is an exact match, determine that it is consistent; when there is an incomplete match, apply rules to check whether it is an acceptable variant, and determine that it is consistent if any rule matches; when none of them matches, determine that it is inconsistent; add the state description of the issuing authority to the overall state set of the evidence; for each associated evidence node, summarize all state descriptions in the overall state set to evaluate the overall consistency degree with the certificate entity group, and assign a credibility score accordingly; summarize the evaluation results of all associated evidence, analyze the evidence distribution, and obtain the overall consistency score;

[0026] Check if the certificate key field value exists in the pre-defined trusted list, and increase the credibility of the field when it exists; when it does not exist, decide whether to reduce the credibility or not according to whether the field is required to exist in the list; sum the credibility scores of all fields by weight to obtain the credibility list check score;

[0027] Evaluating the context relevance of the evidence node, querying its spatial neighboring nodes and semantic similar nodes, checking whether it supports the validity of the certificate or provides background information; evaluating the support of each neighboring node, considering the node weight; aggregating the support scores of all spatial neighboring and semantic similar nodes, combining the spatial neighboring weight and the semantic similar weight, calculating the context relevance score;

[0028] Based on the initial quality score, the consistency score, the trusted list check score and the context relevance score of the evidence, an adjustment factor is calculated using a preset weighting formula.

[0029] In combination with the first aspect, in a sixth implementation manner of the first aspect of the present application, the checking, by reverse tracing, whether there are contradictory evidence, low-trust strong association, context conflict and replacement clues pointing to the certificate, and the discovery of which reduces the trustworthiness of the certificate entity group, comprises:

[0030] Querying all reference association edges pointing to the certificate entity group node in the graph database, analyzing the reverse association nodes, finding negative words and information conflicts, and marking as high risk if direct contradictions are found, and obtaining a direct contradiction score;

[0031] Checking the low-trust nodes pointing to the certificate, and taking the average influence score as the low-trust node influence score;

[0032] Querying the spatial neighboring and semantic similar nodes of the certificate, and considering that there is a context conflict when there is more than a preset number of negative information or doubts, and calculating a context conflict score using the context relevance score obtained by forward tracing;

[0033] Analyzing the reverse association nodes, finding clues that the certificate is updated, replaced or abolished, and marking as high risk if found, and obtaining a certificate replacement risk score;

[0034] Combining the direct contradiction score, the low-trust node influence score, the context conflict score and the certificate replacement risk score, a negative adjustment factor is calculated using a preset weighting formula.

[0035] In combination with the first aspect, in a seventh implementation manner of the first aspect of the present application, the updating the trustworthiness of the certificate entity group according to the forward and reverse tracing results, and the propagation of the trustworthiness change along the reference edge to affect the trustworthiness of the related support and contradictory evidence nodes, comprises:

[0036] After the forward and reverse tracing, the adjustment factor and the negative adjustment factor are obtained according to the comprehensive evaluation result, the adjustment factor is applied to the credibility of the certificate entity group, and the total adjustment is performed after all supporting evidence is traversed to obtain the adjusted credibility; the negative adjustment factor is applied to the credibility of the certificate entity group after the forward and reverse tracing to obtain the final credibility; the final credibility is compared with the original credibility before adjustment to calculate the specific change rate of the credibility;

[0037] The change rate of the credibility is transmitted to the related supporting evidence nodes and contradictory evidence nodes according to a preset weighting propagation rule along all reference edges connected with the certificate entity group;

[0038] For the supporting evidence nodes, the improvement of the credibility of the certificate can correspondingly improve the credibility of the nodes; for the contradictory evidence nodes, the improvement of the credibility of the certificate can reduce the credibility of the nodes;

[0039] The states of all related nodes are updated according to the propagated credibility to trigger a new round of tracing and evaluation process.

[0040] In combination with the first aspect, in an eighth implementation manner of the first aspect of the present application, the traversing all certificate entity groups and marking as abnormal the certificate entity groups lower than the credibility threshold and analyzing the causes of low credibility and contradictory points include:

[0041] A credibility threshold is set, and all identified certificate entity groups in the graph database are traversed to check whether the final credibility of each certificate entity group is lower than the threshold;

[0042] For the certificate entity groups lower than the threshold, the certificate entity groups are marked as abnormal to trigger an abnormal analysis process;

[0043] In the abnormal analysis process, the forward tracing result and the reverse tracing result of the certificate entity group are reviewed in detail to determine the specific causes of low credibility; the contradictory points in the abnormal certificate entity group, that is, the information conflicting in the forward and reverse tracing, are focused on;

[0044] The abnormal certificate entity group and the analysis result thereof are recorded, and a corresponding alarm or notification mechanism is triggered to ensure that the responsible personnel understand and handle these abnormal situations.

[0045] In combination with the first aspect, in a ninth implementation manner of the first aspect of the present application, the listing of abnormal certificate information, the visualization display of the key evidence, contradictory points and associated paths causing the abnormality, and the generation of a textual description include:

[0046] An abnormal certificate list is generated to list the key information of all certificate entity groups marked as abnormal; for each abnormal certificate, the key evidence and contradictory points causing the low credibility are automatically extracted, including the supporting evidence with low credibility, the negative evidence with high credibility and the node information with direct conflict;

[0047] With the visualization tool of the graph database, the abnormal certificate and the key evidence, the contradiction points and the associated path therebetween are displayed in a graphical manner; while the visualization display, a written description is generated, which details the low reliability causes of the abnormal certificate, the specific content of the key evidence, the manifestation form of the contradiction points and the associated logic therebetween, so as to help the responsible personnel understand the abnormal situation and provide a reference for further investigation or decision-making;

[0048] The abnormal certificate list, the visualization display and the written description are integrated into a unified report for viewing and exporting.

[0049] Compared with the prior art, the beneficial effects of the present application are:

[0050] 1、The present application converts the extracted information into nodes and stores them in the graph database, establishes edges according to the physical location, semantic similarity and reference relationship, forms a preliminary evidence association network, and clusters related nodes into entity groups representing specific certificates for overall credibility evaluation.

[0051] 2、The present application checks the information consistency between the certificate and the evidence, checks whether the seal in the evidence is consistent with the seal on the certificate in terms of clarity, color, shape and text content, checks whether the agency name in the evidence matches the agency name on the certificate, and judges according to different rules.

[0052] 3、The present application dynamically adjusts the credibility of the certificate entity group by forward tracing to find supporting evidence and reverse tracing to find contradictory evidence. BRIEF DESCRIPTION OF DRAWINGS

[0053] Fig. 1 The figure shows the steps of the evidence identification method for security evaluation of commercial cryptographic application based on image recognition of the present application;

[0054] Fig. 2 The figure shows the seal image consistency detection steps of the evidence identification method for security evaluation of commercial cryptographic application based on image recognition of the present application. DETAILED DESCRIPTION

[0055] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0056] Embodiment: As shown in the figure, the present application provides a technical solution, Figs. 1-2 ​

[0057] As Fig. 1 The steps of the image recognition-based commercial cryptographic application security evaluation evidence identification method are shown in the schematic diagram, and the image recognition-based commercial cryptographic application security evaluation evidence identification method is provided, which comprises the following steps:

[0058] Step S100: For the senior engineer certification certificate submitted by the employee, information is extracted from the multi-modal document by using the Qwen-VL visual language model; a high-dimensional vector representation is generated for each extracted item, and the source, type and preliminary quality score are labeled, and the high-dimensional vector representation is converted into a structured evidence node;

[0059] Specifically, the relevant documents submitted by the employee containing the senior engineer certification certificate are received, including text, pictures, tables and files in various forms; the optical character recognition technology is used to extract all visible text content, obtain all recognized text blocks and their position coordinates in the image, and for the original PDF or text file, the text content in the original PDF or text file is directly extracted, and the text from different sources is integrated together to form a set containing all visible text;

[0060] The certificate image is analyzed, including the senior engineer certificate, the communication protocol, the communication password, the product certificate, the contract and the etc. proof, and the image processing technology is used, including edge detection, template matching and color recognition, to locate the key image area, identify the specific position of the signature or the stroke area with handwritten features; identify the seal area with specific shape, color and internal pattern or text features; identify the image area with portrait features usually located in the upper right corner of the certificate or the specified position; the located image elements are accurately cut out from the original image and saved as separate image files, and the source document and position information are recorded;

[0061] The prepared multi-modal data is taken as input, the API or local interface of the Qwen-VL model is called, and the key information items are identified and extracted, including the name of the certificate holder, the certificate number, the issuance date, the name of the issuing agency, the certificate level, the certificate validity period, the signature information, the seal pattern and the text, and the certificate photo, and each information item extracted is returned in the form of key-value pair;

[0062] The input text and image are analyzed, the patterns or keywords matching the name, number, date, agency name, level and validity period are found in the text, and the accuracy of the information is confirmed by using context understanding; for the signature image, the signature text is identified; for the seal image, the shape and color of the seal are identified, and the optical character recognition technology is used to try to read the text in the seal; for the certificate photo, basic features such as gender and approximate age are judged; cross verification is performed in combination with the text and image information; a structured information dictionary or list is output, which contains all the identified key information items and their values;

[0063] Simple format and logical check on the output of Qwen-VL, including text information, bounding box coordinates, confidence score and image segments, etc. Each information item identified by the model is organized into a structured intermediate format, including its corresponding value, location information and confidence score given by the model, etc.

[0064] Select a CLIP embedding model. For text information items, use their fields as input to the embedding model to generate a fixed-length vector. For image information items, pass the extracted image region data to the embedding model to generate corresponding image vectors. Associate the generated vectors with the corresponding information items and store them in the information item data structure.

[0065] Record the original document from which the information item comes, its specific location in the document, the source type and the type of extracted information item.

[0066] Use the confidence score given by Qwen-VL when extracting information to evaluate the image quality of information items from images. Use image processing techniques to quantify the clarity, and for text, check for garbled or unclear characters. Check if the extracted information item value is complete. Design a weighting formula to calculate a comprehensive preliminary quality score based on the above factors.

[0067] Design a standard evidence node data structure. For each information item extracted, use all relevant information to create a specific evidence node instance according to the designed node data structure. Store the generated structured evidence nodes in a temporary data structure for subsequent steps.

[0068] To reduce the potential risks of Qwen-VL in this invention, partially mask or encrypt the identified password-related fields to enhance data security. Based on commercial cryptography, retrain Qwen-VL to simulate massive certificates and adversarial sample inputs to improve recognition accuracy. In the input preprocessing stage, use digital watermark detection or noise analysis to identify PS traces and detect image tampering. At the same time, keep a manual intervention channel for low-confidence results.

[0069] In a specific embodiment, an employee submits a PDF file containing a senior engineer certification certificate. First, the PDF is processed using optical character recognition technology to extract the text content, including the name: Zhang Wei, certificate number: ENG2025068901, issuance date: 2024-12-15, issuing authority: China Engineers Association, certificate level: senior engineer, valid period: 2029-12-14, and other text blocks, and records the page number and coordinate position of these texts in the PDF. Then, using image processing technology, the image of Zhang Wei's signature located in the upper right corner of the first page of the PDF is located and cropped to save as Signature_GW.png, with the source page number 1, coordinate range x: 100-200, y: 50-80, and the circular red seal image located in the center is saved as Seal_2025.png, with the source page number 1, coordinate range x: 300-500, y: 400-600. Call the Qwen-VL model, and integrate the text set and the cropped images as input. Qwen-VL successfully extracts key information items such as name Zhang Wei, certificate number ENG2025068901, issuance date 2024-12-15, issuing authority China Engineers Association, certificate level senior engineer, valid period 2029-12-14, signature information, seal text, and certificate photo, and returns a dictionary containing these key-value pairs, with each information item accompanied by a model confidence score, name confidence 0.98, certificate number 0.99, seal text 0.85. Perform format checking on the output of Qwen-VL and confirm that there are no errors. Then, select the CLIP model, generate a 768-dimensional vector for the text Zhang Wei, a 768-dimensional vector for the image Signature_GW.png, and a 768-dimensional vector for the image Seal_2025.png, and associate these vectors with the corresponding information items. Record the source of each information item as Cert_2025-001.pdf, and the source type as PDF, and the information item type as name, certificate number, issuance date, etc. Based on the Qwen-VL confidence of 0.98, the signature image clarity of 0.95, the seal image clarity of 0.90, and the information being complete, calculate a comprehensive preliminary quality score, with the name score being 0.96, the certificate number being 0.97, and the seal text being 0.87. According to the designed evidence node data structure, create specific evidence node instances for each information item such as name: Zhang Wei, certificate number: ENG2025068901, etc., containing information item content, high-dimensional vector, source annotation, type annotation, and preliminary quality score, and store these structured evidence nodes in a temporary data structure.

[0070] Step S200: store the evidence nodes into the graph database, establish multiple edges according to the physical location, semantic similarity and reference relationship, form a preliminary evidence network, and give the nodes an initial credibility; identify the certificate nodes in the network, cluster the related nodes to form a certificate entity group, and give a comprehensive initial credibility;

[0071] Specifically, all encapsulated evidence nodes generated are read, each node containing extracted information items, high-dimensional vector representation, source annotation, type annotation and preliminary quality score;

[0072] The physical layout of the information items in the original multi-modal document is analyzed, and for those information nodes that are spatially adjacent or arranged in logical order, a spatial adjacency edge is established in the graph database. This edge is given an adjacency or similarity label, indicating their proximity in document space.

[0073] Using the high-dimensional vector representation of each information item, the similarity between these vectors is calculated by using the cosine similarity algorithm. The cosine value of the angle between two vectors is calculated. If the directions of the two vectors are exactly the same, the angle is 0 degrees, and the cosine value is 1, indicating that they are most similar. If the directions are exactly opposite, the angle is 180 degrees, and the cosine value is -1, indicating that they are least similar. If the directions are perpendicular, the angle is 90 degrees, and the cosine value is 0, indicating that they have no similarity in direction. Identify information nodes that are highly related or mutually corroborate in content. When the high-dimensional vectors show high similarity, semantic similarity edges are established for these nodes, and semantic correlation or mutual corroboration labels are given;

[0074] Analyze whether there is an explicit reference or derivation relationship between information items. Use natural language processing and knowledge graph technology. For text information items, perform word segmentation, part-of-speech tagging, named entity recognition, etc. Use a pre-trained bidirectional encoder representation language model to input sentences or paragraphs containing two information items and their context. The model learns and predicts whether there is a specific relationship between them. Output the probability of relationship existence. Set a threshold, and if the threshold is exceeded, determine that there is a relationship. For information items containing images, combine image recognition results and text information to determine the relationship. For example, identify the unit name on the seal, and then search for whether there is a text information item mentioning the unit to determine whether there is a reference or attribution relationship. Establish reference association edges for nodes with reference or dependency relationships, and give reference relationship or dependency relationship labels;

[0075] Through the above analysis based on physical location, semantic similarity and reference relationship and the creation of corresponding edges, originally isolated evidence nodes are connected to form a preliminary evidence network containing multiple edge types and multiple dimensions.

[0076] An initial credibility is calculated for each node in the graph database by taking into account the model confidence, the content clarity and completeness score, the fact that some information items are more critical to the authenticity of the certificate than others, the identification of the issuing authority, the higher initial credibility given to information items from authoritative sources, and the number and type of connections a node has in the network. A pre-set weighted calculation formula is used to calculate a specific initial credibility value for each node in the graph database, representing the probability that the information item is authentic based on its own quality and network position in the initial evaluation stage.

[0077] In the preliminary evidence network, the nodes representing the core certificate information are identified, and all related nodes directly or indirectly connected to the certificate node are identified using the Louvain algorithm. These related nodes include other information items on the certificate and information items in the associated evidence.

[0078] The identified certificate node and all directly or indirectly connected related nodes are considered as a whole, and a more precise Leiden algorithm is used to aggregate these nodes into a certificate entity group. This entity group represents the collection of all evidence information related to the specific certificate.

[0079] A comprehensive initial credibility value is calculated for the newly formed certificate entity group, including the distribution and average of the initial credibility values of all nodes in the entity group, the type and number of connections between the nodes in the entity group, the overall structure of the entity group in the network, such as density and connection with other external nodes. An algorithm based on random walk is used to simulate the transmission and mutual influence of information between the nodes in the entity group. Each node in the graph database is considered as a state in the random walk, the edges in the graph are considered as the possibility of state transition, and the initial credibility of each node is considered as the initial value of the personalized vector. Random walk simulation and iterative calculation are performed. After multiple iterations, when the PageRank value converges, each node will get a final PageRank value, which comprehensively evaluates the credibility of the entire entity group, and finally gives the certificate entity group a comprehensive initial credibility value, which quantitatively evaluates the preliminary authenticity of the certificate as a whole.

[0080] In a specific embodiment, all evidence nodes generated by reading the employee-submitted Cert2025-001.pdf certificate are included, including the name Zhang Wei, node ID: E1, vector similarity 0.98, quality score 0.96, certificate number ENG2025068901, node ID: E2, vector similarity 0.99, quality score 0.97, seal text China Association of Engineers, node ID: E3, vector similarity 0.85, quality score 0.87, etc. Analyzing the physical layout, it is found that E1 name Zhang Wei and E2 certificate number ENG2025068901 are close in coordinates on the first page of the PDF, E1: x: 50-150, y: 50-100; E2: x: 200-300, y: 50-100, so a spatial proximity edge is established in the graph database for E1 and E2, labeled as proximity. Using the CLIP vector, the cosine similarity between E1 name Zhang Wei and E3 seal text China Association of Engineers is 0.12, while the cosine similarity between E2 certificate number and E3 seal institution name is 0.45, which does not reach the preset semantic similarity threshold of 0.6, so no semantic similarity edge is established. Analyzing the reference relationship, the text context is analyzed by the bidirectional encoder representation method model, and it is found that there is a high probability of reference relationship between E3 China Association of Engineers and the text issuing agency: China Association of Engineers node ID: E4, quality score 0.95, probability 0.92, which exceeds the threshold of 0.8, so a reference association edge is established between E3 and E4, labeled as reference relationship. After these steps, nodes E1, E2, E3, E4, etc. are connected to form a preliminary evidence network. Calculate the initial credibility of each node: E1 name score is 0.96, E2 certificate number score is 0.97, E3 seal text score is 0.87, and E4 issuing agency text score is 0.95. Considering that the certificate number is crucial to authenticity, E2 is given a higher weight, weight coefficient 1.2, and the final initial credibility of E2 is adjusted to 0.97x1.2=1.164, normalized to 0.98. Identify the core certificate nodes, such as E2 certificate number, and use the Leuven algorithm to find its directly connected nodes E1, E3, E4, etc. Use the Leiden algorithm to aggregate these nodes to form a certificate entity group C1. Calculate the comprehensive initial credibility of C1, the average initial credibility of the nodes in the group is (0.96+0.98+0.87+0.95)÷4=0.935, there are 1 spatial proximity edge and 1 reference association edge between nodes, the connection density in the group is medium, and all nodes in the group come from the original certificate document. Using the random walk-based algorithm, considering the mutual influence between nodes, the final comprehensive initial credibility value of C1 is 0.92, representing that the certificate has a high probability of authenticity in the preliminary evaluation stage.

[0081] Step S300: By forward tracing, starting from the certificate entity group, find supporting evidence along the reference association edge, adjust the credibility of the certificate entity group by checking the credibility, consistency, trusted list matching degree and context association of the supporting evidence;

[0082] Specifically, in the graph database, find the node representing the senior engineer certification certificate through Cypher query, traverse along the reference association edge type, find all evidence nodes that claim to support the validity of the certificate; read the initial quality score assigned to each evidence node in the initial extraction stage;

[0083] In the graph database, find the evidence nodes directly or indirectly associated with the certificate, extract the information related to the key fields therein; evaluate the basic consistency between the certificate internal information and the single evidence;

[0084] Check if the seal information is included in the evidence, extract its key features; use image processing technology to locate the area suspected to be a seal on the certificate image, achieve by finding a specific shape, specific color, specific texture or specific location; crop the located area to obtain a subgraph containing the seal image, denoted as the main seal image; for each associated evidence, if it exists in the form of an image, repeat the above steps to locate and crop the seal image on the evidence, denoted as the evidence seal image; when the associated evidence is a text description or a digital record, check if the seal information is explicitly mentioned, if so, record the key features described, including color, shape and mentioned text; if not mentioned or the evidence itself does not contain an image, it is considered that the evidence does not provide seal information, and the seal comparison step is skipped;

[0085] For each evidence that successfully extracted the evidence seal image, perform feature dimension extraction and comparison; calculate the sharpness index of the main seal image and the evidence seal image respectively, use the Laplacian operator variance to quantify the edge sharpness of the image, and the calculated values are denoted as the main seal sharpness value and the evidence seal sharpness value respectively; when the evidence seal sharpness value is equal to the main seal sharpness value, the state is described as consistent in sharpness; when not equal, the state is described as inconsistent in sharpness;

[0086] Quantify the main color area of the main seal image and the evidence seal image, calculate the average RGB value of the main pixels in the image, denoted as the main seal color and the evidence seal color respectively; when the RGB value of the evidence seal color is completely equal to the RGB value of the main seal color, the state is described as consistent in color; when not equal, the state is described as inconsistent in color;

[0087] Edge detection is performed on the main seal image and the evidence seal image using a Canny operator; geometric shape analysis is performed on the detected edge contours, and minimum circumscribed rectangle, circularity and ellipticity parameters of the contours are calculated; the obtained main shape features are denoted as main seal shape features and evidence seal shape features, respectively; when the shape described by the evidence seal shape features is completely identical to the main seal shape features, the status is described as shape consistent; when the shape features are not identical, the status is described as shape inconsistent;

[0088] Optical character recognition technology is applied to the main seal image and the evidence seal image to extract the text content in the seal region; the extracted text content is denoted as main seal text content and evidence seal text content, respectively; the extracted text is standardized, and leading and trailing spaces are removed, punctuation marks are unified, and conversion to uppercase or lowercase is performed; it is checked whether the standardized evidence seal text content is completely identical to the standardized main seal text content; when they are completely identical, the status is described as text content consistent; when they are not completely identical, the status is described as text content inconsistent;

[0089] For each evidence, all the generated status descriptions are summarized to form a seal status set of the evidence; all the status descriptions in the seal status set are added to the overall status set generated by the evidence; the final status information of the evidence in all inspection dimensions is included;

[0090] When the seal information in the evidence does not match the certificate seal in any of the above key features, and the seal information is present in the evidence, the status description seal mismatch is generated; when the seal information is completely matched, or there is no seal information in the evidence, this status is not generated;

[0091] It is compared whether the name of the issuing authority displayed on the certificate entity group is consistent with the information of the issuing authority mentioned or embodied in the associated evidence, and the differences, abbreviations, and full names of the names are processed;

[0092] The full name of the issuing authority is extracted from the certificate image or text, and is denoted as main authority full name; when it is a text evidence, information related to the issuing authority is searched in the text content, including full name, commonly used abbreviation, English name, and name with changes in different files;

[0093] When it is an image evidence, optical character recognition technology is used to recognize the text in the image to find the name related to the issuing authority;

[0094] All the related authority names or identifiers found from the evidence are recorded to form a set, denoted as evidence authority name set;

[0095] Standardize each name in the set of full name of the main body and the set of name of the evidence body, including unifying character case, removing spaces before and after the name and removing unnecessary punctuation in the name;

[0096] Traverse the set of name of the evidence body, and determine whether each current evidence name in the set is identical to the full name of the main body. When there is a current evidence name identical to the full name of the main body, it is determined that the information of the issuing body of the evidence is consistent with the certificate;

[0097] Check whether it belongs to an acceptable variant based on a rule-based definition, match the rule based on known information, check whether it is a recognized abbreviation, determine whether the current evidence name is a widely recognized and officially used abbreviation of the full name of the main body, and match to determine consistency. Check whether it is an English name corresponding to the name, determine whether the current evidence name is the official English name of the full name of the main body, determine whether the current evidence name adds or removes some prefixes or suffixes to the full name of the main body, and the core subject name does not change, and define the rule to determine whether such change is reasonable and does not affect the subject identification. If the rule determines that it is a reasonable variant, it is determined to be consistent. Check whether there is a known historical name, determine whether the current evidence name is the formal name used by the agency at the time of certificate issuance or slightly earlier, and match the rule based on historical information;

[0098] After traversing the set of name of the evidence body, it is determined to be consistent by any of the above methods, and the state description is consistent with the issuing body. After traversing all the names, none of them passes any consistency judgment rule, and the state description is inconsistent with the issuing body.

[0099] Add the generated state description about the issuing body to the overall state set of the evidence. For each associated evidence node, collect all state descriptions in the overall state set; according to the summary of the state description, evaluate the overall consistency degree of the evidence node and the certificate entity group; according to the overall state of the evidence and the specific inconsistent items in the state set, assign a credibility score to each evidence node; aggregate the evaluation results of all associated evidence nodes; analyze the distribution of the evidence, and based on the aggregated information of all the evidence, obtain a consistency score;

[0100] Predefine a trusted list for the key fields of the certificate entity group, query each key field value of the certificate entity group, check whether the value exists in the corresponding trusted list, and when it exists, consider that the field passes the verification and increase the preset credibility; when it does not exist, determine whether the field is required to exist in the list, and if it is required, reduce the preset credibility; weight and sum the credibility check scores of all key fields to obtain a trusted list check score;

[0101] Query other nodes that have spatial proximity edges with the evidence node, check whether these adjacent nodes support the validity of the certificate, or contain relevant background information; Query other nodes that have semantic similarity edges with the evidence node, check whether these nodes point to supporting information; Combine the information of spatial and semantic adjacent nodes, judge whether the evidence node is in a supportive context environment, and calculate the context relevance;

[0102] Query nodes that have spatial proximity edges with the current evidence node, these edges represent proximity in physical location or association in structure, check whether these nodes contain information supporting the validity of the certificate or provide relevant background information;

[0103] Query nodes that have semantic similarity edges with the current evidence node, these edges represent that the node content is similar in theme, concept or object of reference, check whether these nodes point to supporting information;

[0104] For each identified adjacent node, evaluate whether it supports the certificate entity group or is irrelevant or contradictory; Set a binary judgment, support is 1, irrelevant or contradictory is 0; Consider the weight or credibility of the adjacent node, the supporting information provided by the adjacent node with high credibility has more weight than the node with low credibility;

[0105] Aggregate the support scores of all spatial proximity nodes and the support scores of semantic similarity nodes, consider the node weight, and use a weighted average algorithm to obtain the spatial proximity aggregation score and the semantic similarity aggregation score;

[0106] Combine the aggregation results of the two dimensions of spatial proximity and semantic similarity to calculate the context relevance score CF, the formula is:

[0107] ;

[0108] Where CF is the context relevance score; SpS is the spatial proximity aggregation score; SeS is the semantic similarity aggregation score; w s is the weight of the spatial proximity score, w c is the weight of the semantic similarity score;

[0109] Based on the above check results, use a preset weighted formula to calculate the adjustment factor μ, the formula is:

[0110] ;

[0111] Where μ is the adjustment factor, BF is the initial quality score of the extracted item, CS is the consistency score, LCF is the trust list check score, CF is the context relevance score, w1, w2, w3, w4 are the corresponding weights.

[0112] In a specific embodiment, the entity group C1 representing the certificate of the employee Zhang Wei, senior engineer, is found in the graph database by a Cypher query, with its core node being the certificate number ENG2025068901, node ID: E2. Traversing along the reference association edge, a link to the associated evidence node E5 is found, which is a scanned picture of the same certificate submitted by Zhang Wei. The initial quality score of E5 is read as 0.92. Evaluating the internal information consistency of the certificate, it is found that the node information of E1 name Zhang Wei, E2 certificate number, E3 seal text, E4 issuing authority text, etc. is basically consistent. Checking the seal information, in the certificate entity group C1, the E3 node contains the seal text China Engineers Association and the seal image. In the associated evidence E5, the seal area is located and cropped using image processing techniques to obtain the evidence seal image. Feature comparison is performed on the main seal image E3 and the evidence seal image, the clarity is calculated, the main seal clarity value is 120, and the evidence seal clarity value is 118, the state description is that the clarity is inconsistent; the average value of the main color RGB is calculated, the main seal color is (200, 0, 0), and the evidence seal color is (200, 0, 0), the state description is that the color is consistent; edge detection is performed using the Canny operator and the shape is analyzed, the main seal shape feature is circular, and the evidence seal shape feature is also circular, the state description is that the shape is consistent; the optical character recognition is applied to extract the text, the main seal text content is China Engineers Association, and the evidence seal text content is China Engineers Association, which is completely the same after standardization, the state description is that the text content is consistent. The seal state set of E5 is summarized as inconsistent clarity, consistent color, consistent shape, and consistent text content, and no seal mismatch state is generated. Comparing the issuing authority, the full name of the main authority is China Engineers Association. The agency name extracted in the evidence E5 is also China Engineers Association, which is completely the same after standardization, and the state description is that the issuing authority is consistent. The seal and agency states are added to the overall state set of E5. Evaluating the overall consistency degree of E5 and C1, although there is a slight inconsistency in clarity, other key features are consistent, and the consistency score CS is calculated as 0.95. Querying the spatial proximity nodes and semantic similarity nodes related to E5, it is found that these nodes all support the certificate validity, the spatial proximity aggregation score SpS is 0.9, and the semantic similarity aggregation score SeS is 0.85. The context relevance score CF is calculated as (0.9 x 0.85) ÷ 2 = 0.875. Checking whether the certificate number ENG2025068901 is in the preset trusted list, it is found that the number is not in the list, but the field is not required to be in the list, so the trusted list check score LCF is 0.0. Based on the initial quality score BF of E5 = 0.92, the consistency score CS = 0.95, the trusted list score LCF = 0.0, and the context relevance score CF = 0.875, the preset weighting formula is used to calculate μ = 0.8265.

[0113] Step S400: Check whether there are contradictory evidence, low credibility strong association, context conflict and replacement clues pointing to the certificate through reverse tracing, and reduce the credibility of the certificate entity group if found; update the credibility of the certificate entity group according to the results of forward and reverse tracing, propagate the credibility change along the reference edge, and affect the credibility of the related support and contradictory evidence nodes;

[0114] Specifically, a certificate entity group node is obtained from the graph database as a starting point, and all reference association edges pointing to the current certificate entity group node are queried;

[0115] The content of the node found from the reverse reference edge is analyzed to find words or sentences that deny the validity of the certificate, and it is checked whether the extracted information in the node directly conflicts with the key information of the certificate entity group. When a direct contradiction is found, it is marked as high risk, and the direct contradiction score DCS is 1, otherwise 0;

[0116] Check the preliminary quality score or credibility of all evidence nodes pointing to the certificate entity group; if multiple low-credibility nodes are found to be strongly associated with the certificate entity group through reference edges, take the average impact score as the low-credibility node impact score, and the low-credibility node impact score is 0 if there is no low-credibility node;

[0117] Query nodes that have spatial proximity edges with the certificate entity group, and check whether these nodes contain information that contradicts the validity of the certificate; query nodes that have semantic similarity edges with the certificate entity group, and check whether these nodes generally reflect negative situations or doubts; if the certificate entity group is in a context environment of negative information or doubts, it is considered that there is a context conflict, and the context association score CF obtained by forward tracing is used for calculation, the formula is:

[0118] ;

[0119] Wherein, CCS is the context conflict score, CF is the context association score, and ConflictRatio is the proportion of negative nodes; if there is no conflict, CCS is 0;

[0120] Analyze the content of the node found from the reverse reference edge to find explicit or indirect clues that the certificate has been updated, replaced or no longer used; check whether there is a new certificate entity group node and evidence that it has replaced the current one, and mark it as high risk when such clues are found. The certificate replacement risk score ORS is 1, otherwise 0;

[0121] Based on the check results, determine whether there are factors that need to reduce the credibility; use a preset weighted formula to calculate the negative adjustment factor NAF:

[0122] ;

[0123] wherein DCS is direct contradiction score; LNI is low credibility node influence score; CCS is context conflict score; ORS is certificate replacement risk score; w5, w6, w7, w8 are corresponding weights respectively;

[0124] The adjustment factor is applied to the current credibility of the certificate entity group, and after traversing all the found supporting evidence, the total adjustment is performed to obtain the adjusted credibility T1; the negative adjustment factor is applied to the adjusted credibility T1 of the certificate entity group to obtain the final credibility T2, and a decision is made according to the final credibility.

[0125] In a specific embodiment, the certificate entity group C1 is obtained from the graph database, and the senior engineer certificate representing Zhang Wei is taken as a starting point. All reference association edges pointing to C1 are queried, and a reference association edge pointing to evidence node E6 is found, with an initial quality score of 0.45. The node is another file, and the content claims that the certificate submitted by Zhang Wei is fake. First, the content of E6 is analyzed to find negative words, and it is found that the words such as fake and invalid are explicitly included, which directly conflicts with the core information of C1, so the direct contradiction score DCS is 1. The quality of the evidence node pointing to C1 is checked, and it is found that only E6 is a low credibility node with a score of 0.45, so the low credibility node influence score LNI is 0.45. The nodes spatially adjacent to C1 are queried, and no negative information is found. The semantically similar nodes are also queried, and no generally negative situation is found, so the context conflict score CCS is 0. In the content of E6, no clues of certificate replacement, replacement or obsolescence are found, so the certificate replacement risk score ORS is 0. Based on DCS=1, LNI=0.45, CCS=0, ORS=0, the negative adjustment factor NAF is calculated as 0.635 using the preset weighting formula. Assuming that the credibility of C1 after positive tracing is 0.76, it is taken as the current credibility. The negative adjustment factor 0.635 is applied to the current credibility 0.76 to obtain the final credibility T2 of about 0.28, which is significantly reduced, indicating that the direct contradiction evidence E6 found by reverse tracing has seriously questioned the authenticity of the certificate C1.

[0126] Step S500: traverse all certificate entity groups, filter out those below the credibility threshold and mark them as abnormal, analyze the causes of low credibility and contradictions, list abnormal certificate information, and visually display the key evidence, contradictions and associated paths that cause the abnormality, and generate a written description.

[0127] Specifically, a credibility threshold is set, representing the lowest acceptable credibility level; a traversal process is started to access all identified certificate entity groups stored in the graph database, and for each certificate entity group, the adjusted final credibility value is read; if the final credibility value of a certain certificate entity group is lower than the preset threshold, the state of the entity group is marked as abnormal, triggering the subsequent abnormal analysis process;

[0128] Once the certificate entity group is marked as abnormal, it automatically enters abnormal analysis, which checks the supporting evidence collected during the forward tracing process and its contribution to credibility, and checks the evidence found during the reverse tracing process that does not support or even denies the certificate and its weakening effect on credibility. By comparing the results of forward and reverse tracing, the specific reasons for the low final credibility of the certificate entity group are analyzed, including lack of sufficient high-credibility supporting evidence, existence of multiple high-credibility negative or contradictory evidence, evidence chain breakage, key information missing, and existence of irreconcilable conflicts between evidence;

[0129] Focus on and clearly identify the direct conflicts or contradictions found in the forward and reverse tracing results; for example, an evidence node is considered supportive in forward tracing, but is found to be contradictory to another high-credibility evidence in reverse tracing;

[0130] All certificate entity group information marked as abnormal, along with its final credibility value, low credibility cause analysis, and identified conflict points, are recorded in a special abnormal log or database table for audit tracking and processing; according to preset rules or configurations, the corresponding alarm or notification mechanism is automatically triggered, including sending emails, pushing messages, and displaying alarm icons on the management interface, to notify relevant personnel responsible for reviewing or handling abnormal situations, ensuring that they are aware of the problematic certificates in a timely manner;

[0131] A clear list of abnormal certificates is generated, which should include key information for each abnormal certificate entity group, including certificate ID, associated employee information, final credibility value, and abnormal marking time; for each abnormal certificate in the list, the specific information that caused it to be marked as abnormal is extracted from the graph database, pointing out evidence nodes with low credibility that negatively affect overall credibility, explicitly listing node information with direct conflicts or contradictions and their conflict content, extracting the association path between these key evidence, conflict points, and certificate entity group nodes, and showing their specific connection relationships in the graph network;

[0132] With the visualization tools provided by the graph database or the integrated visualization engine, the abnormal certificate entity group, the extracted key evidence nodes, the contradiction point nodes and the association paths between them are displayed in a graphical way; for example, different colors or shapes can be used to distinguish normal nodes, abnormal nodes, supporting evidence, negative evidence, contradictions, etc., and lines can be used to represent their relationships;

[0133] At the same time of visualization display, a detailed text description is generated, which sets forth the low confidence cause summary, the extracted key evidence content, the extracted contradiction point information and the brief explanation of the association logic, so that the responsible person can quickly and accurately understand the core problem of the abnormal situation even without deepening the graph database;

[0134] The generated abnormal certificate list, the visualization chart and the corresponding text description are integrated into a structured unified report, and an export function is provided for the responsible person, the management or the archive.

[0135] In a specific embodiment, the credibility threshold is set to 0.7. After starting the traversal, the certificate entity group C1 representing Zhang Wei's senior engineer certificate is accessed, and its final credibility is adjusted to 0.28 after steps S300 and S400, which is lower than the threshold 0.7. Immediately mark the state of C1 as abnormal, and trigger the abnormal analysis process. Analysis found that although the positive trace collected the scanned picture evidence E5 with a quality score of 0.92, which contributed to the credibility, the direct contradictory evidence E6 found in the reverse trace claimed that the certificate was fake, with a quality score of 0.45, a direct contradiction score DCS=1 and a negative adjustment factor 0.635 triggered by it were the main causes of the sharp decline in credibility. The low credibility is caused by the existence of high-impact direct denial evidence. The contradiction point is identified as the fake label in E6 directly conflicts with the core information of C1. These information is recorded in the abnormal log, and triggers the email alarm, sends to the leader of the authentication audit team Li Ming. The generated abnormal certificate list contains the following entries: certificate ID: ENG2025068901, associated employee: Zhang Wei, final credibility: 0.28, abnormal marking time: 2025-07-07 14:30. For C1, the key evidence node E6 is extracted, with a quality of 0.45 and a content containing fake, the contradiction point is E6 and the core information of C1, and the associated path is E6 pointing to C1 through a reference association edge. In the visualization display, the C1 node is marked in red, indicating abnormality, the E6 node is marked in yellow, indicating negative evidence, and the reference edge between them is highlighted. At the same time, the generated text description is: "Certificate ENG2025068901, Zhang Wei's final credibility is 0.28, which is lower than the threshold 0.7, and is marked as abnormal. The key evidence is node E6, which claims that the certificate is fake, with a quality score of 0.45. The main contradiction point is that the content of E6 directly conflicts with the core information of the certificate. The associated path is E6 pointing to the certificate entity group C1 through the reference association edge. The abnormality has been notified to Li Ming for processing." Finally, a unified report containing the list, chart and description is generated and can be exported.

[0136] As Fig. 2 The seal image consistency detection step of the image recognition-based commercial cryptographic application security evaluation evidence identification method is shown in the schematic diagram, and the image recognition-based commercial cryptographic application security evaluation evidence identification method is provided, which comprises the following steps:

[0137] The information consistency between the certificate and the evidence is checked, whether the seal in the evidence and the seal of the certificate are consistent in clarity, color, shape and text content, and the specific steps are as follows:

[0138] In the graph database, the evidence nodes directly or indirectly associated with the certificate are found through Cypher query, and the content related to the seal information is extracted;

[0139] Check if the evidence contains seal information, extract its key features; use image processing techniques to locate the area of the certificate image suspected of containing a seal, by looking for a specific shape, specific color, specific texture or located in a specific location; cut out the area located to get a subgraph containing the seal image, denoted as the main seal image; for each associated evidence, if it exists in the form of an image, repeat the above steps to locate and cut out the seal image on the evidence, denoted as the evidence seal image; when the associated evidence is a text description or a digital record, check if it explicitly mentions the seal information, if so, record the key features described, including color, shape and mentioned text; if not mentioned or the evidence itself does not contain an image, it is considered that the evidence does not provide seal information, skip the seal comparison step;

[0140] For each evidence that successfully extracted the evidence seal image, extract and compare the feature dimensions; calculate the sharpness index of the main seal image and the evidence seal image respectively, use the Laplacian operator variance to quantify the edge sharpness of the image, and the calculated values are denoted as the main seal sharpness value and the evidence seal sharpness value respectively; when the evidence seal sharpness value is equal to the main seal sharpness value, the state description is consistent in sharpness; when they are not equal, the state description is inconsistent in sharpness;

[0141] Quantify the main color area of the main seal image and the evidence seal image, calculate the average RGB value of the main pixels in the image, denoted as the main seal color and the evidence seal color respectively; when the RGB value of the evidence seal color is completely equal to the RGB value of the main seal color, the state description is color consistent; when they are not equal, the state description is color inconsistent;

[0142] Use the Canny operator to perform edge detection on the main seal image and the evidence seal image; perform geometric shape analysis on the detected edge contours, calculate the minimum circumscribed rectangle, circularity and ellipticity parameters of the contours; the main shape features obtained are denoted as the main seal shape feature and the evidence seal shape feature respectively; when the shape described by the evidence seal shape feature is exactly the same as the main seal shape feature, the state description is shape consistent; when the shape features are different, the state description is shape inconsistent;

[0143] Apply optical character recognition technology to the main seal image and the evidence seal image to extract the text content in the seal area; the extracted text content is denoted as the main seal text content and the evidence seal text content respectively; standardize the extracted text, remove leading and trailing spaces, unify punctuation marks, and convert to uppercase or lowercase; check if the standardized evidence seal text content is exactly the same as the standardized main seal text content; when they are exactly the same, the state description is text content consistent; when they are not exactly the same, generate a state description of text content inconsistent;

[0144] For each piece of evidence, all the generated state descriptions are summarized to form a seal state set of the evidence; all the state descriptions in the seal state set are added to the overall state set of the evidence generation; and finally the state information of the evidence in all the check dimensions is contained.

[0145] When the seal information in the evidence does not match the seal information of the certificate seal in any one of the key features above, and the seal information is presented in the evidence, a state description of seal mismatch is generated; when the seal information completely matches, or the seal information is not presented in the evidence, this state is not generated.

[0146] It is apparent for those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, but can be implemented in other concrete forms without departing from the spirit or essential characteristics of the present application. Therefore, the embodiments should be considered in all aspects as illustrative and not restrictive, and the scope of the present application is defined by the appended claims rather than the above description, and therefore all changes falling within the meaning and range of the equivalent elements of the claims are intended to be embraced in the present application. Any reference signs in the claims should not be considered as limiting the claims involved.

Claims

1. An image recognition-based security assessment evidence identification method for a commercial cryptographic application, characterized by, The application relates to a method for evaluating senior engineer certification certificates submitted by employees, which comprises the following steps: extracting information from multi-modal documents by using a Qwen-VL visual language model; generating a high-dimensional vector representation for each extracted item, labeling its source, type and preliminary quality score, and converting it into a structured evidence node; storing the evidence node in a graph database, establishing various edges based on physical location, semantic similarity and reference relationship, forming a preliminary evidence network, and giving the node an initial credibility; identifying the certificate node in the network, clustering its related nodes to form a certificate entity group, and giving the group a comprehensive initial credibility; adjusting the credibility of the certificate entity group by checking the credibility, consistency, trusted list matching degree and context relevance of the supporting evidence found by forward tracing from the certificate entity group; checking whether there are contradictory evidence, low credibility strong association, context conflict and replacement clues pointing to the certificate by reverse tracing, and reducing the credibility of the certificate entity group if the clues are found; updating the credibility of the certificate entity group according to the results of forward and reverse tracing, and propagating the credibility change along the reference edge to affect the credibility of related supporting and contradictory evidence nodes; traversing all the certificate entity groups, screening out the ones below the credibility threshold and marking them as abnormal, analyzing the causes of low credibility and contradictions, listing the abnormal certificate information, visualizing the key evidence, contradiction points and their associated paths that cause the abnormality, and generating a written description. 2.The image recognition based security assessment evidence identification method for commercial cipher application according to claim 1, wherein, The method for evaluating senior engineer certification certificates submitted by employees comprises the following steps: reading the multi-modal documents of the submitted senior engineer certification certificates, including text, pictures, tables and files; identifying and extracting the key information in the documents by using a Qwen-VL visual language model as information items, including the name of the certificate holder, the certificate number, the issuance date, the name of the issuing agency, the certificate level, the certificate validity period, the signature on the certificate, the seal pattern and the text, and the certificate photo. 3.The image recognition based security assessment evidence identification method for commercial cipher application according to claim 1, wherein, The method for generating a high-dimensional vector representation for each extracted item, labeling its source, type and preliminary quality score, and converting it into a structured evidence node comprises the following steps: generating a high-dimensional vector representation for each extracted information item, creating a source label to record the specific information of the source document, and labeling the type of the information item; giving a preliminary quality score based on model confidence, content clarity and integrity; structurally packaging the extracted information item content, high-dimensional vector representation, source label, type label and preliminary quality score, and converting it into a structured evidence node. 4.The image recognition based security assessment evidence identification method for commercial cipher application according to claim 1, wherein, The method for storing the evidence node in a graph database, establishing various edges based on physical location, semantic similarity and reference relationship, forming a preliminary evidence network, and giving the node an initial credibility comprises the following steps: reading and analyzing the information items and the attached metadata contained in each encapsulated evidence node, in the graph database, according to the physical location relationship of the information items in the original document, establishing edges representing spatial proximity for information nodes that are spatially adjacent or arranged in order, as spatial proximity edges, and assigning the label of the proximity field; using the high-dimensional vector representation of the information item content, identifying information nodes that are highly related or may have mutual corroboration by calculating the cosine similarity between vectors, and establishing edges representing semantic association for these nodes, as semantic similarity edges, and assigning the label of semantic association or mutual corroboration; according to the reference or derivation relationship between information items, establishing edges representing reference dependence, as reference association edges, and assigning the label of reference relationship or dependence relationship; by analyzing the physical location, semantic similarity and reference relationship and creating corresponding edges, connecting the nodes to form a multi-dimensional preliminary evidence network; When assigning the initial credibility to the nodes, extract the preliminary quality score of the information items, analyze the importance of the information items, the authority of the issuing agency and the connection mode of the information items in the evidence network; through a pre-set weighted calculation method, calculate and assign a specific initial credibility value to each node in the graph database, representing the probability of the information item being recognized as true in the initial evaluation stage. 5.The image recognition based security assessment evidence identification method for commercial cipher application according to claim 1, wherein, The method for identifying certificate nodes in the network, clustering related nodes to form a certificate entity group, and assigning a comprehensive initial credibility includes: In the initial evidence network, the Louvain algorithm is used to identify nodes representing the same certificate entity, and all related nodes directly or indirectly connected to the certificate nodes are found; these certificate nodes and all related nodes are regarded as a whole, and the Leiden algorithm is used to aggregate them to form a certificate entity group; according to the initial credibility values of all nodes in the entity group, the connection strength between nodes and the overall structural characteristics of the entity group in the network, an algorithm based on random walk is used to calculate and assign a comprehensive initial credibility value to the certificate entity group. 6.The image recognition based security assessment evidence identification method for commercial cipher application according to claim 1, wherein, The method for adjusting the credibility of the certificate entity group by forward tracing from the certificate entity group along the reference association edges to find supporting evidence, and checking the credibility, consistency, trusted list matching degree and context association of the supporting evidence includes: In the graph database, the nodes representing the senior engineer certification certificate and their initial attributes are located by Cypher query, all evidence nodes claiming to support the validity of the certificate are found along the reference association edges, and the initial quality score of the evidence nodes is read; the key field values of the certificate entity group are extracted, the associated evidence nodes are found in the graph database, the corresponding field information is extracted, and the basic consistency between the certificate internal information and the certificate and each piece of evidence is evaluated; Check if the evidence contains a seal, extract its key features; for image evidence, use image processing techniques to locate the seal area based on shape, color, texture, or position, and crop out the main seal image and evidence seal image; for text or digital evidence, check if the seal and its features are explicitly mentioned; when the evidence has no seal information, skip the comparison; for evidence with successfully extracted seal images, perform multi-dimensional feature comparison; calculate the sharpness of the main seal and the evidence seal using the Laplacian operator variance to determine if they are consistent; perform color quantization and compare the average RGB values of the main pixels to determine if the colors are consistent; apply the Canny operator for edge detection to analyze the shape features of the contours and determine if the shapes are consistent; use optical character recognition technology to extract the seal text, compare the text content after standardization to determine if the text is identical, and determine the consistency of the text; aggregate the comparison results of each evidence in the sharpness, color, shape, and text dimensions to form the seal status set of the evidence, and add it to the overall status set; when the seal does not match in any key feature and the evidence presents a seal, generate a seal mismatch state, and if it is a complete match or has no seal, do not generate this state; Compare the consistency of the issuing authority name, extract the full name of the main authority, and use optical character recognition for image evidence; record all related agency names found as the evidence agency name set, and standardize these names and the full name of the main agency; iterate through the evidence agency name set to determine if each name is identical to the full name of the main agency; when there is a complete match, determine consistency; when there is an incomplete match, apply rules to check if it is an acceptable variant, and if any rule matches, determine consistency; when none of the iterations match, determine inconsistency; add the state description of the issuing authority to the overall status set of the evidence; for each associated evidence node, aggregate all state descriptions in the overall status set to evaluate the overall consistency degree with the certificate entity group, and assign a credibility score accordingly; aggregate the evaluation results of all associated evidence, analyze the evidence distribution, and obtain the overall consistency score; Check if the certificate key field value exists in the pre-defined trusted list, and if so, increase the credibility of the field; if not, decide whether to reduce the credibility or not based on whether the field is required to be in the list; weight and sum the credibility scores of all fields to obtain the trusted list check score; Evaluate the contextual relevance of the evidence node, query its spatial proximity nodes and semantically similar nodes, and check if they support the certificate validity or provide background information; evaluate the supportiveness of each proximity node, considering node weights; aggregate the supportiveness scores of all spatial proximity and semantically similar nodes, and calculate the contextual relevance score by combining spatial proximity weight and semantic similarity weight; Based on the initial quality score, consistency score, trusted list check score, and contextual relevance score of the evidence, use a pre-set weighting formula to calculate the adjustment factor. 7.The image recognition based security assessment evidence identification method for commercial cipher application according to claim 1, wherein, The reverse tracing is performed to check whether there are contradictory evidence, low credibility strong association, context conflict and replacement clues pointing to the certificate, and if found, the credibility of the certificate entity group is reduced, including: All reference association edges pointing to the certificate entity group nodes in the graph database are queried, the reverse association nodes are analyzed to find negative words and information conflicts, and if direct contradictions are found, they are marked as high risk, and a direct contradiction score is obtained; Low credibility nodes pointing to the certificate are checked, and their average impact score is taken as the low credibility node impact score; The spatial neighbors and semantic similar nodes of the certificate are queried, and when there are more than a preset number of negative information or doubts, it is considered that there is a context conflict, and the context conflict score is calculated using the context association score obtained by forward tracing; The reverse association nodes are analyzed to find clues that the certificate is updated, replaced or invalidated, and if found, they are marked as high risk, and a certificate replacement risk score is obtained; The direct contradiction score, low credibility node impact score, context conflict score and certificate replacement risk score are integrated, and a negative adjustment factor is calculated using a preset weighting formula. 8.The image recognition based security assessment evidence identification method for commercial password application according to claim 1, wherein, The credibility of the certificate entity group is updated according to the results of forward and reverse tracing, and the credibility change is propagated along the reference edges to affect the credibility of related supporting and contradictory evidence nodes, including: After forward and reverse tracing, adjustment factors and negative adjustment factors are obtained according to the comprehensive evaluation results, the adjustment factors are applied to the credibility of the certificate entity group, and after all supporting evidence is traversed, a total adjustment is performed to obtain the adjusted credibility; the negative adjustment factors are applied to the credibility adjusted by forward tracing to obtain the final credibility; the final credibility is compared with the original credibility before adjustment to calculate the specific change rate of the credibility; Along all reference edges connected to the certificate entity group, the credibility change rate is transmitted to related supporting evidence nodes and contradictory evidence nodes according to a preset weighting propagation rule; For supporting evidence nodes, the improvement of the credibility of the certificate will correspondingly improve the credibility of the nodes; for contradictory evidence nodes, the improvement of the credibility of the certificate will reduce the credibility of the nodes; The states of all related nodes are updated according to the propagated credibility to trigger a new round of tracing and evaluation process. 9.The image recognition based security assessment evidence identification method for commercial password application according to claim 1, wherein, All certificate entity groups are traversed to filter out those below a credibility threshold and marked as abnormal, and the causes of low credibility and contradictions are analyzed, including: A credibility threshold is set, all identified certificate entity groups in the graph database are traversed, and it is checked whether the final credibility of each certificate entity group is below the threshold; For certificate entity groups below the threshold, they are marked as abnormal to trigger an abnormality analysis process; In the abnormality analysis process, the forward and reverse tracing results of the certificate entity group are reviewed in detail to determine the specific causes of low credibility; the contradictions in the abnormal certificate entity group, i.e., the conflicting information in forward and reverse tracing, are focused on; The abnormal certificate entity group and its analysis results are recorded, and corresponding alarm or notification mechanisms are triggered to ensure that responsible personnel understand and handle these abnormal situations. 10.The image recognition based security assessment evidence identification method for commercial password application according to claim 1, wherein, Abnormal certificate information is listed, key evidence, contradictions and their association paths causing abnormalities are visualized and displayed, and a textual description is generated, including: Generating an abnormal certificate list, listing the key information of all certificate entity groups marked as abnormal; for each abnormal certificate, automatically extracting the key evidence and contradictions that lead to its low credibility, including low-credibility supporting evidence, high-credibility negative evidence, and node information with direct conflicts; Using the visualization tools of the graph database, displaying the abnormal certificates and their related key evidence, contradictions, and the association paths between them in a graphical manner; while visualizing, generating a textual description detailing the causes of the low credibility of the abnormal certificates, the specific content of the key evidence, the forms of the contradictions, and the association logic between them, helping the responsible personnel understand the abnormal situation and providing a reference for further investigation or decision-making; Integrating the abnormal certificate list, the visualization display, and the textual description into a unified report for viewing and exporting.

Images