Methods And Apparatus for Assessing Status of Stay of International Student

Abstract

A method and device for diagnosing the status of residence of international students are disclosed. The present embodiment provides a method and apparatus for diagnosing the status of residence of international students, which automatically analyzes various administrative documents of global international students using artificial intelligence (AI) and enables rapid and accurate diagnosis and screening of various administrative requirements related to the status of residence.

Description

Technology Field

[0001] One embodiment of the present invention relates to a method and apparatus for diagnosing the status of residence of international students. The present invention is a result carried out with support from the 'Support for Development and Nurturing of Promising SaaS' project organized by the Korea Information and Communication Technology Industry Promotion Agency. Background Technology

[0002] The following description merely provides background information related to the present embodiment and does not constitute prior art.

[0003] With the recent advancement of internationalization and globalization, the number of international students is increasing rapidly worldwide. Consequently, the management of student residency status and related administrative tasks have become a major administrative function for each country.

[0004] However, the procedures related to administrative documents that international students must submit are generally complex and composed of various languages, causing significant difficulties for administrative agencies in each country as well as the students themselves.

[0005] The conventional method of handling administrative matters for international students relied primarily on manual verification and review of documents, resulting in slow processing speeds and a high risk of human error. In particular, frequent incidents of document forgery and alteration often raised concerns about the reliability of administrative processing, leading to issues such as a decline in the credibility of national administration and reduced convenience for international students.

[0006] Conventional administrative processing methods required a significant input of human resources, and due to their complex and time-consuming procedures, they had many limitations in terms of efficiency. The problem to be solved

[0007] The present embodiment aims to provide a method and device for diagnosing the status of residence of international students, which can automatically analyze various administrative documents of global international students using artificial intelligence (AI) and then quickly and accurately diagnose and review various administrative requirements related to the status of residence. means of solving the problem

[0008] According to one aspect of the present embodiment, a device for diagnosing the status of a student's residence is provided, comprising: a document analysis module that receives an image of a multilingual administrative document regarding a global student from a user terminal and extracts multilingual text information from the multilingual administrative document image; an examination module that automatically examines the status of residence and administrative requirements based on the multilingual text information and generates an examination result; an anomaly detection module that generates a verification result that verifies forgery or alteration of the multilingual administrative document image; and a blockchain management module that encrypts the examination result, the verification result, and personal information included in the multilingual text information and stores them in a blockchain.

[0009] According to another aspect of the present embodiment, a method for diagnosing the status of residence of an international student is provided, characterized by comprising: a process in which a document analysis module receives an image of a multilingual administrative document regarding a global international student from a user terminal and extracts multilingual text information from the multilingual administrative document image; a process in which a screening module automatically screens the status of residence and administrative requirements based on the multilingual text information and generates a screening result; a process in which an anomaly detection module verifies forgery or alteration of the multilingual administrative document image and generates a verification result; and a process in which a blockchain management module encrypts the personal information contained in the screening result, the verification result, and the multilingual text information and stores it in a blockchain. Effects of the invention

[0010] As explained above, according to the present embodiment, various administrative documents of global international students are automatically analyzed using artificial intelligence (AI), and various administrative requirements related to status of residence are quickly and accurately diagnosed and reviewed. Brief explanation of the drawing

[0011] Figure 1 is a diagram showing a system for diagnosing the status of residence of international students according to the present embodiment. FIG. 2 is a block diagram schematically showing a diagnostic device for international student residency status according to the present embodiment. FIG. 3 is a flowchart for explaining a method for diagnosing the status of residence of international students according to the present embodiment. FIG. 4 is a conceptual diagram showing the operation of a document analysis module according to an embodiment of the present invention. FIG. 5 is a conceptual diagram showing the operation of a screening module and an anomaly detection module according to an embodiment of the present invention. Specific details for implementing the invention

[0012] The present invention is a result carried out with support from the 'Promising SaaS Development and Fostering Support' project organized by the Korea Information and Communication Industry Promotion Agency. Hereinafter, the embodiments thereof will be described in detail with reference to the attached drawings.

[0013] Figure 1 is a diagram showing a system for diagnosing the status of residence of international students according to the present embodiment.

[0014] The international student residence status diagnosis system according to the present embodiment includes a user terminal (110), a network (120), and an international student residence status diagnosis device (130). The components included in the international student residence status diagnosis system are not necessarily limited thereto.

[0015] A user terminal (110) refers to an electronic device that performs voice or data communication via a network (120) according to the user's key operation. The user terminal (110) is equipped with a memory for storing a program or protocol for communicating with a student residence status diagnostic device (130) via the network (120), a microprocessor for executing the program to perform calculations and control, etc.

[0016] The user terminal (110) may be an electronic device such as a smartphone, tablet, laptop, personal computer (PC), personal digital assistant (PDA), portable multimedia player (PMP), wireless communication terminal, media player, etc.

[0017] A user terminal (110) is a various device equipped with a communication device, such as a communication modem, for communicating with various devices or wired / wireless networks, a memory for storing various programs and data, and a microprocessor for executing programs to perform calculations and control. According to at least one embodiment, the memory may be a computer-readable recording / storage medium such as RAM (Random Access Memory), ROM (Read Only Memory), flash memory, an optical disk, a magnetic disk, or a Solid State Disk (SSD). According to at least one embodiment, the microprocessor may be programmed to selectively perform one or more operations and functions described in the specification. According to at least one embodiment, the microprocessor may be implemented as hardware, such as an Application Specific Integrated Circuit (ASIC), wholly or partially.

[0018] The user terminal (110) inputs an image of a multilingual administrative document into the international student residence status diagnostic device (130) by operation or command of the global international student.

[0019] The network (120) refers to a network capable of transmitting and receiving data via the Internet protocol using various wired and wireless communication technologies, such as the Internet network, intranet network, mobile communication network, and satellite communication network. The network (120) performs the role of relaying data between the user terminal (110) and the international student residence status diagnostic device (130).

[0020] The international student residence status diagnostic device (130) includes a hardware module identical to a standard web server, WAP server, or network server in terms of hardware. The international student residence status diagnostic device (130) generally communicates with an unspecified number of clients or other servers via an open computer network such as the Internet.

[0021] The international student residence status diagnostic device (130) refers to a computer system or computer software (web server program) that derives and provides work results in response to a work execution request from a client or another web server. In addition to the aforementioned web server program, the international student residence status diagnostic device (130) includes a series of application programs operating on the web server or various databases built inside the device.

[0022] The international student residence status diagnostic device (130) uses the latest Vision-Large Language Model to accurately recognize multilingual documents of various formats and automatically extracts key information to perform rapid administrative processing.

[0023] The international student residence status diagnostic device (130) applies automatic screening technology using reinforcement learning-based human feedback (RLHF) to maximize the accuracy of screening various administrative documents and residence status conditions.

[0024] The international student residence status diagnostic device (130) performs administrative processing by precisely detecting whether a submitted document has been forged or altered using real-time anomaly detection technology. The international student residence status diagnostic device (130) performs secure storage and management of user personal information and administrative processing results using zero-trust blockchain technology. The international student residence status diagnostic device (130) can efficiently simplify complex administrative procedures for global international students and fundamentally resolve various problems that occurred in the existing manual method.

[0025] The international student residence status diagnostic device (130) enables rapid and accurate processing by automating the complex administrative procedures and residence status screening of global international students using artificial intelligence (AI).

[0026] The international student residence status diagnostic device (130) utilizes Vision-LLM technology, an RLHF-based automatic screening module, intelligent anomaly detection technology, and Zero Trust blockchain technology to efficiently and accurately screen the residence status and administrative documents of global international students.

[0027] The international student residence status diagnostic device (130) receives an image of an administrative document and automatically extracts information based on Vision-LLM. After performing RLHF automatic screening, the international student residence status diagnostic device (130) checks for intelligent anomaly detection forgery and tampering, and stores and manages the Zero Trust blockchain results.

[0028] The components according to the embodiments of the present invention may be implemented based on hardware (e.g., a computer server), and the operation of the components described above may be performed on the hardware (e.g., a computer server). Although referred to as components for convenience of explanation, the operations of the components may perform an automated process that did not exist before using computer program code based on hardware components such as a CPU (central processing unit), a GPU (graphic processing unit), and memory.

[0029] More specifically, the operation process of the components performed in the present invention is performed based on the control unit and arithmetic logic unit (ALU) of a processing unit such as a CPU. The operation process of the components can be performed through fetch, decode, and execute performed on hardware such as a CPU.

[0030] In addition, various memory structures (registers, cache memory, main memory, auxiliary memory) may be used for the operation of the operation process of the components according to the embodiment of the present invention, and computer program code using various memory hierarchies may be used for the efficiency of the operation process of the components.

[0031] By executing this computer program code-based automation through hardware, previously unknown processes are automated, resulting in the following effects compared to non-automated procedures: it enables the rapid and accurate diagnosis and screening of various administrative requirements related to residency status by automatically analyzing diverse administrative documents of global international students using artificial intelligence (AI).

[0032] More specifically, the processor used in the embodiments of the present invention can, for example, execute software (e.g., a program) to control at least one other component (e.g., a hardware or software component) of an electronic device connected to the processor, and can perform various data processing or operations. According to one embodiment, as at least part of the data processing or operations, the processor can store commands or data received from other components (e.g., a sensor module or a communication module) in volatile memory, process the commands or data stored in volatile memory, and store the resulting data in non-volatile memory. According to one embodiment, the processor may include a main processor (e.g., a central processing unit or an application processor) or an auxiliary processor that can operate independently or together with it (e.g., a graphics processing unit, a neural processing unit (NPU), an image signal processor, a sensor hub processor, or a communication processor). For example, if the electronic device includes a main processor and an auxiliary processor, the auxiliary processor may be configured to use lower power than the main processor or to be specialized for a designated function. The auxiliary processor may be implemented separately from the main processor or as part thereof.

[0033] According to one embodiment, an auxiliary processor (e.g., a neural network processing unit) may include a hardware structure specialized for processing the artificial intelligence model disclosed in the present invention. The artificial intelligence model may be generated through machine learning. Such learning may be performed, for example, on the electronic device itself where the artificial intelligence model is executed, or through a separate server (e.g., a server). The learning algorithm may include, for example, supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning, but is not limited to the examples described above. The artificial intelligence model may include a plurality of artificial neural network layers. An artificial neural network may be a deep neural network (DNN), a convolutional neural network (CNN), a recurrent neural network (RNN), a restricted Boltzmann machine (RBM), a deep belief network (DBN), a bidirectional recurrent deep neural network (BRDNN), a deep Q-network, or a combination of two or more of the above, but is not limited to the examples described above. In addition to the hardware structure, the artificial intelligence model may include a software structure, either additionally or substantially.

[0034] FIG. 2 is a block diagram schematically showing a diagnostic device for international student residency status according to the present embodiment.

[0035] The international student residence status diagnostic device (130) according to the present embodiment includes a document analysis module (210), a screening module (220), an anomaly detection module (230), and a blockchain management module (240). The components included in the international student residence status diagnostic device (130) are not necessarily limited thereto. Each component included in the international student residence status diagnostic device (130) may be connected to a communication path connecting software modules or hardware modules within the device and may operate organically with one another. These components communicate using one or more communication buses or signal lines.

[0036] Each component of the international student residence status diagnostic device (130) illustrated in FIG. 2 represents a unit that processes at least one function or operation, and can be implemented as a software module, a hardware module, or a combination of software and hardware.

[0037] The document analysis module (210) receives an image of a multilingual administrative document regarding a global international student from a user terminal (110) and extracts multilingual text information from the multilingual administrative document image. The document analysis module (210) extracts multilingual text information from the multilingual administrative document image using Vision-Large Language Model (Vision-LLM) technology.

[0038] The document analysis module (210) receives an image of a global international student's multilingual administrative document from the user terminal (110). The document analysis module (210) immediately extracts and analyzes multilingual text information from the multilingual administrative document image.

[0039] The document analysis module (210) receives multilingual administrative document images from the user terminal (110) of a global international student. The document analysis module (210) automatically extracts key information from the administrative documents. The document analysis module (210) automatically extracts key information from the multilingual administrative document images using Vision-LLM technology. The document analysis module (210) uses the latest models such as Google Gemini, GPT-40, and GPT-5 as Vision-LLM technology.

[0040] The document analysis module (210) extracts multilingual text information from multilingual administrative document images and generates key information from the multilingual text information.

[0041] The document analysis module (210) generates core information from administrative and status-related data directly necessary for the examination of residency status from images of multilingual administrative documents.

[0042] The document analysis module (210) generates core information based on multilingual text information, including personal identification information (name, date of birth, passport number, nationality information, etc.), information related to residence (visa type, date of issuance / expiration date, information on conditions for residence), academic and educational information (school name, major, expected date of admission / graduation, information on whether the student is enrolled), administrative and supporting document information (name of issuing institution, date of document issuance, type of document, information on whether a signature / seal is present), and other information related to examination requirements (health insurance enrollment status, financial guarantor information, key numerical information on academic transcripts, etc.).

[0043] The process by which the document analysis module (210) generates key information is as follows.

[0044] The document analysis module (210) receives multilingual administrative document images from the user terminal (110). The document analysis module (210) receives multilingual administrative document images from the user terminal (110). The document analysis module (210) can process images regardless of the image format (JPEG, PNG, PDF, etc.).

[0045] The document analysis module (210) extracts text from a multilingual administrative document image (OCR and Vision-LLM recognition). The document analysis module (210) recognizes multilingual text within the multilingual administrative document image using a Vision-Large Language Model (e.g., Google Gemini, GPT-4o, GPT-5). The document analysis module (210) analyzes the layout of complex document structures (including tables, signature fields, and seals) within the multilingual administrative document image and converts them into text.

[0046] The document analysis module (210) performs language identification and preprocessing. The document analysis module (210) detects the language of the extracted text (Korean, English, Japanese, Chinese, etc.). The document analysis module (210) corrects unnecessary spaces, line breaks, and OCR misrecognized characters.

[0047] The document analysis module (210) classifies semantic-based information. The document analysis module (210) classifies text within a document by semantic unit (personal information, administrative information, educational information, etc.). The document analysis module (210) uses the contextual understanding ability of Vision-LLM to extract meaning even from data without table titles or labels.

[0048] The document analysis module (210) maps key information. The document analysis module (210) maps the extracted items to the status of residence examination rule DB. For example, the document analysis module (210) stores “validity period March 1, 2026” in the expiration date of residence field.

[0049] The document analysis module (210) structures key information. The document analysis module (210) converts the key information into JSON, XML, or DB record forms.

[0050] The document analysis module (210) converts multilingual document images into text, analyzes the meaning, and structures them into standardized core information data.

[0051] The screening module (220) automatically screens the status of residence and administrative requirements based on multilingual text information and generates screening results.

[0052] The screening module (220) automatically screens the status of residence and administrative requirements based on the extracted multilingual text information. The screening module (220) performs an automatic optimization function that continuously improves the accuracy and efficiency of the screening results.

[0053] The screening module (220) automatically screens whether the status of residence and administrative requirements are met based on extracted key information. The screening module (220) uses a reinforcement learning-based screening method that automatically optimizes the results of the status of residence screening for RLHF-based automatic screening.

[0054] The screening module (220) obtains input data from the document analysis module (210). The screening module (220) receives preprocessed multilingual text information and a set of key information from the document analysis module (210). For example, the screening module (220) receives name, date of birth, passport number, visa type, validity period, school name, academic status, etc. from the document analysis module (210). Depending on the case, the screening module (220) receives original text and structured JSON data together from the document analysis module (210).

[0055] The screening module (220) prepares for screening using a combination of rule-based and AI methods. The screening module (220) stores a database of rules for screening status of residence and a mapping table of administrative requirements internally. The screening module (220) verifies eligibility requirements according to laws and administrative regulations (e.g., possession of a valid visa, remaining period of stay, enrollment status, fulfillment of financial requirements, etc.). The screening module (220) reflects requirements that differ by country and region in the database. The screening module (220) trains the screening model using the RLHF (Reinforcement Learning-based Human Feedback) technique, making it possible to judge exceptional situations and ambiguous cases.

[0056] The examination module (220) performs the examination procedure.

[0057] The screening module (220) performs information mapping. The screening module (220) links the input key information fields to the screening criteria fields. For example, the screening module (220) maps "VISA_TYPE": "D-2" to the study visa criteria.

[0058] The screening module (220) verifies administrative requirements. The screening module (220) compares against prescribed screening rules. For example, the screening module (220) checks whether the visa validity period is after today's date. The screening module (220) checks whether the issuance date of the certificate of enrollment is within a certain period. The screening module (220) verifies whether the financial guarantee amount is above the minimum standard.

[0059] The screening module (220) corrects the judgment based on AI. The screening module (220) uses an LLM model to make a judgment on cases that cannot be judged by simple rules. For example, when academic information is included in English or Chinese documents as a descriptive sentence, the screening module (220) determines whether the student is "currently enrolled" after interpreting the meaning.

[0060] The screening module (220) handles risks and exceptions. If some items are missing or inconsistent, the screening module (220) classifies them into a status such as “conditional approval,” “pending,” or “disapproval.” The screening module (220) automatically generates the grounds and reasons for the screening (e.g., “the scheduled expiration date of the visa validity period is shorter than the reference date”).

[0061] The review module (220) generates a review result. The review module (220) finally generates a review result. The review module (220) generates a review result including status values ​​such as Eligible, Conditional, and Ineligible. The review module (220) determines whether each review item passes or fails. The review module (220) generates a statement of reasons and grounds (automatically generated by LLM).

[0062] The screening module (220) transmits the generated screening results to the anomaly detection module (230) and the blockchain management module (240).

[0063] The review module (220) performs learning and optimization. The review module (220) performs RLHF (Reinforcement Learning from Human Feedback) reinforcement learning by comparing the review results with actual approval and rejection cases of administrative agencies. The review module (220) automatically updates the review criteria and rules over time. The review module (220) immediately reflects changes in country-specific regulations.

[0064] The review module (220) compares the extracted multilingual core information with legal and administrative regulations, performs a mixed AI and rule review, and generates results and reasons.

[0065] The anomaly detection module (230) generates a verification result that verifies forgery of a multilingual administrative document image. The anomaly detection module (230) verifies forgery of a multilingual administrative document image using an anomaly detection algorithm including a Variational Autoencoder (VAE) and an Isolation Forest.

[0066] The anomaly detection module (230) verifies in real time whether administrative documents have been forged or altered. The anomaly detection module (230) uses the latest deep learning technology to detect anomalies in real time.

[0067] The anomaly detection module (230) verifies in real time whether a multilingual administrative document image has been forged or altered using an intelligent anomaly detection algorithm. The anomaly detection module (230) applies the latest technologies, such as Variational Autoencoder (VAE) and Isolation Forest, as an anomaly detection algorithm.

[0068] The anomaly detection module (230) receives input data from the document analysis module (210).

[0069] The anomaly detection module (230) receives a multilingual administrative document image transmitted from a user terminal (110) as original data from the document analysis module (210). The anomaly detection module (230) receives an authentic sample retrieved from a public database (document templates, signature and seal patterns, issuance records, etc. of the issuing agency) as reference data from the document analysis module (210) when necessary. The anomaly detection module (230) receives text, layout information, and metadata generated by the document analysis module from the document analysis module (210) as extracted data.

[0070] The anomaly detection module (230) performs preprocessing on the multilingual administrative document image. The anomaly detection module (230) performs image normalization on the multilingual administrative document image. The anomaly detection module (230) performs resolution correction, noise removal, and color and brightness uniformization on the multilingual administrative document image. The anomaly detection module (230) performs document rotation and distortion correction (scan angle correction) on the multilingual administrative document image.

[0071] The anomaly detection module (230) extracts features from a multilingual administrative document image. The anomaly detection module (230) extracts visual features (background pattern, signature / seal location, character spacing, etc.) of the entire document and by region of the multilingual administrative document image. The anomaly detection module (230) analyzes file metadata (creation time, EXIF ​​information, etc.) for the multilingual administrative document image.

[0072] The anomaly detection module (230) applies an anomaly detection algorithm. The anomaly detection module (230) verifies forgery by combining deep learning-based image feature analysis and a statistical-based anomaly detection technique.

[0073] The anomaly detection module (230) performs a Variational Autoencoder (VAE)-based verification.

[0074] The anomaly detection module (230) models the feature vector distribution of a normal document by training a large number of genuine document images using a VAE learning method. The anomaly detection module (230) reconstructs the input document by passing it through a VAE using a VAE verification method. The anomaly detection module (230) determines the possibility of tampering if the reconstruction error (the difference between the original and the reconstructed image) exceeds a threshold value. The anomaly detection module (230) performs forgery of signatures or seals, deletion / addition of parts of the document, font replacement, etc., as possible VAE detection cases.

[0075] The anomaly detection module (230) detects anomalies based on Isolation Forest.

[0076] The anomaly detection module (230) generates a normal distribution model based on document feature vectors (layout ratio, color distribution, text density, etc.) using the Isolation Forest learning method. The anomaly detection module (230) inputs the feature values ​​of a new document into a tree-based outlier model using the Isolation Forest verification method, and determines it as an outlier if it shows a distribution significantly different from other documents. The anomaly detection module (230) outputs template inconsistencies, abnormal layouts, traces of involuntary editing, etc., as Isolation Forest detectable cases.

[0077] The anomaly detection module (230) performs layer analysis and pixel pattern inspection. The anomaly detection module (230) analyzes pixel-level differences in document images. The anomaly detection module (230) detects traces of Photoshop / graphic tool modifications (abnormal compression patterns, blurring, edge distortion, etc.).

[0078] The anomaly detection module (230) performs result aggregation and judgment. The anomaly detection module (230) aggregates the results of each algorithm. The anomaly detection module (230) can aggregate VAE results (reconstruction error rate), Isolation Forest scores (outlier probability), and metadata / pixel analysis results.

[0079] The anomaly detection module (230) scores the possibility of forgery. The anomaly detection module (230) calculates a comprehensive risk score. For example, the anomaly detection module (230) determines that if the score is 70 or higher out of 0 to 100, it is 'suspected of forgery'.

[0080] The anomaly detection module (230) generates a judgment result. The anomaly detection module (230) determines the result as Normal, Suspected Forgery, or Confirmed Forgery. The anomaly detection module (230) also records supporting data (coordinates of the suspected forgery area, difference in feature values, etc.).

[0081] The anomaly detection module (230) performs follow-up processing. The anomaly detection module (230) transmits the verification results and evidence images to the review module (220) and the blockchain management module (240). The anomaly detection module (230) transmits items suspected of tampering to the administrator and administrative staff. The anomaly detection module (230) automatically disapproves items when tampering is confirmed.

[0082] The anomaly detection module (230) calculates a comprehensive risk score by analyzing VAE reconstruction errors, determining Isolation Forest outliers, and analyzing pixels / metadata, and verifies whether forgery or alteration has occurred based on a procedure that generates a determination result based on the comprehensive risk score.

[0083] The blockchain management module (240) encrypts personal information included in the examination results, verification results, and multilingual text information and stores it in the blockchain.

[0084] The blockchain management module (240) encrypts and stores and manages the screening results and personal information using zero-trust blockchain technology. The blockchain management module (240) applies zero-trust technology for personal information protection to perform authentication and encryption during access and management.

[0085] The blockchain management module (240) uses Zero Trust blockchain technology to encrypt and securely store and manage screening results and personal information. The blockchain management module (240) performs authentication and encryption whenever personal information is accessed or managed using Zero Trust blockchain technology.

[0086] The blockchain management module (240) obtains input data.

[0087] The blockchain management module (240) obtains personal information including sensitive data such as name, date of birth, passport number, nationality, and school name extracted from the document analysis module (210), verification results including approval / disapproval status, reason, and judgment value for each review item generated by the review module (220), judgment on whether forgery has occurred, information on areas suspected of forgery, and risk score generated by the anomaly detection module (230).

[0088] The blockchain management module (240) performs encryption preparation. The blockchain management module (240) performs data classification, such as sensitive information (PII), administrative processing results, and verification logs. The blockchain management module (240) manages public and private keys using a Key Management System (KMS) in a Zero Trust environment. The blockchain management module (240) performs key management by generating a session key for each user and transaction and applying one-time encryption.

[0089] The blockchain management module (240) performs an encryption process. The blockchain management module (240) performs symmetric key encryption (AES-256, etc.). The blockchain management module (240) encrypts the original text of personal information and screening / verification results using a session symmetric key. The blockchain management module (240) performs asymmetric key encryption (RSA / ECC). The blockchain management module (240) encrypts the session symmetric key using the server / administrator public key and transmits it. The blockchain management module (240) ensures that only the entity holding the corresponding secret key can decrypt the symmetric key. The blockchain management module (240) performs an integrity-guaranteed hash (SHA-256 / SHA-3). The blockchain management module (240) stores the hash value of the encrypted data together to detect whether the data has been tampered with. Here, the hash value is recorded as part of the blockchain transaction.

[0090] The blockchain management module (240) has a structure for storing data on the blockchain. The blockchain management module (240) generates a transaction. The blockchain management module (240) generates a transaction consisting of encrypted data, a hash value, a timestamp, and a transaction ID. The blockchain management module (240) performs block creation and chain linking. The blockchain management module (240) groups multiple transactions into a single block and forms a chain structure including the previous block hash value. The blockchain management module (240) applies a consensus algorithm. The blockchain management module (240) verifies block validity using the PoA (Proof of Authority) and PBFT (Practical Byzantine Fault Tolerance) methods provided by Hyperledger, Ethereum, etc. The blockchain management module (240) performs distributed storage. The blockchain management module (240) prevents data tampering by distributing and storing the same block data across multiple nodes.

[0091] The blockchain management module (240) performs the access and management steps.

[0092] The blockchain management module (240) verifies the user, device, and network environment for all accesses in accordance with the Zero Trust principle for authentication and authorization procedures. The blockchain management module (240) applies MFA (Multiple Authentication) for authentication and authorization procedures. The blockchain management module (240) retrieves encrypted data and hash values ​​from the blockchain ledger to process inquiry requests. The blockchain management module (240) decrypts the symmetric key using the secret key and decrypts the original text using the symmetric key to process inquiry requests. For audit log recording, the blockchain management module (240) stores all access and inquiry activities in a separate log block so that they can be audited later.

[0093] The blockchain management module (240) applies zero trust, does not trust even internal users, and authenticates and encrypts every access. The blockchain management module (240) performs verification of the blockchain's immutability and hash integrity to prevent tampering. To minimize access, the blockchain management module (240) decrypts the original personal information when necessary and prohibits plaintext storage.

[0094] The blockchain management module (240) performs session key-based encryption after data classification. The blockchain management module (240) creates a blockchain transaction, stores it in a distributed manner, and performs zero-trust authentication-based access control.

[0095] FIG. 3 is a flowchart for explaining a method for diagnosing the status of residence of international students according to the present embodiment.

[0096] The international student residence status diagnostic device (130) receives an image of a global international student's multilingual administrative document from the user terminal (110) (S310).

[0097] In step S310, the user terminal (110) transmits a multilingual administrative document image to the international student residence status diagnostic device (130) by operation or command of the global international student.

[0098] The international student residence status diagnostic device (130) extracts and analyzes multilingual text information from a multilingual administrative document image (S320).

[0099] In step S320, the international student residence status diagnostic device (130) extracts and analyzes multilingual text information from images of multilingual administrative documents using Vision-LLM technology. The document analysis module (210) uses the latest models such as Google Gemini, GPT-40, and GPT-5 as Vision-LLM technology.

[0100] The international student residence status diagnostic device (130) automatically examines residence status and administrative requirements based on extracted multilingual text information and generates a screening result (S330).

[0101] In step S330, the international student residence status diagnostic device (130) performs an automatic optimization function that continuously improves the accuracy and efficiency of the examination results. The international student residence status diagnostic device (130) automatically examines whether the residence status and administrative requirements are met based on extracted key information. The international student residence status diagnostic device (130) uses a reinforcement learning-based examination method that automatically optimizes the residence status examination results for RLHF-based automatic examination.

[0102] The international student residence status diagnostic device (130) generates a verification result that verifies in real time whether the multilingual text information and multilingual administrative document images are forged or altered (S340).

[0103] In step S340, the international student residence status diagnostic device (130) detects abnormal behavior in real time using the latest deep learning technology. The international student residence status diagnostic device (130) verifies in real time whether multilingual administrative document images have been forged or altered using an intelligent anomaly detection algorithm. The international student residence status diagnostic device (130) applies the latest technologies, such as Variational Autoencoder (VAE) and Isolation Forest, as an anomaly detection algorithm.

[0104] The international student residence status diagnostic device (130) uses zero-trust blockchain technology to encrypt and store and manage the examination results, verification results, and personal information (S350).

[0105] In step S350, the international student residence status diagnostic device (130) applies Zero Trust technology for personal information protection to perform authentication and encryption during access and management. The international student residence status diagnostic device (130) uses Zero Trust blockchain technology to encrypt and securely store and manage the screening results and personal information. The international student residence status diagnostic device (130) performs authentication and encryption whenever personal information is accessed and managed using Zero Trust blockchain technology.

[0106] Although steps S310 to S350 are described as being executed sequentially in FIG. 3, they are not necessarily limited thereto. In other words, since it is possible to modify and execute the steps described in FIG. 3 or to execute one or more steps in parallel, FIG. 3 is not limited to a chronological order.

[0107] As described above, the method for diagnosing the status of residence of international students according to the present embodiment described in FIG. 3 can be implemented as a program and recorded on a computer-readable recording medium. A computer-readable recording medium on which a program for implementing the method for diagnosing the status of residence of international students according to the present embodiment is recorded includes all types of recording devices in which data that can be read by a computer system is stored.

[0108] FIG. 4 is a conceptual diagram showing the operation of a document analysis module according to an embodiment of the present invention.

[0109] In FIG. 4, a procedure for performing document analysis in a document analysis module is specifically disclosed.

[0110] Referring to FIG. 4, the document analysis module can perform a primary preprocessing step (S410) to perform document analysis.

[0111] The preprocessing step (S410) can perform primary preprocessing on the document to be analyzed through resolution adjustment, color / brightness equalization, distortion / rotation correction, noise removal, etc. After the primary preprocessing, the pixel brightness distribution by region can be adjusted as a secondary preprocessing step.

[0112] The document is divided into a first document area and a second document area, and the pixel brightness distribution can be adjusted differently for each document area.

[0113] In the document, the first document area is an area where a boundary exists as a specific area, and may be an area containing information within a specific boundary, such as a table or picture within the document. The second document area is an area where no boundary exists, and may be an area containing characters that are not included in the boundary on the page.

[0114] For the first document region, first histogram equalization can be performed.

[0115] First histogram equalization is a preprocessing method that improves contrast by dividing an image into small blocks and equalizing the histograms of each block. More specifically, the first document area can be divided into blocks (e.g., 8x8, 16x16 pixels). Subsequently, the distribution of pixel values ​​within a block can be calculated through block-specific histogram calculation. Afterward, the distribution of pixel values ​​is redistributed evenly through block-specific histogram equalization, and a procedure to smoothly connect boundaries between adjacent blocks can be performed through interpolation. Through this first histogram equalization, the detailed structure of each area can be made sharper when the overall image contrast is low or there are large regional differences in brightness.

[0116] For the second document region, second histogram equalization can be performed. The size of the block for the second histogram equalization can be set to be relatively larger than that for the first histogram equalization. Similarly, the distribution of pixel values ​​within the block can be determined through histogram calculation, and a clip limit is applied to each block histogram. Histogram values ​​exceeding the limit are dispersed to surrounding frequencies to prevent noise amplification.

[0117] When a second histogram equalization is performed, sharpness can be increased while processing relatively larger blocks compared to the first histogram equalization, and it may be a suitable processing for images with relatively high noise levels, such as images.

[0118] Following the preprocessing step (S410), a structural analysis step (S420) may be performed. In the structural analysis step (S420), detection of tables, cell merges, unlabeled formatting, signature / seal locations, etc., is performed, and document structure analysis is performed to infer relationships between each element within the document. More specifically, document structure analysis may set each element (text block, table, signature area, image, etc.) in a document or image as a node and determine the relationships between nodes. Structure analysis may be performed by classifying the relationships between nodes into first and second relationships between elements.

[0119] The first relationship is a location-based relationship, established based on positions such as the connection of cells within the same table in a document. The second relationship is a logic-based relationship, established by connecting elements such as paragraphs and headings, or signature fields and dates. The first relationship is a spatial relationship determined based on top / bottom, left / right, and adjacency, while the second relationship is a logical relationship determined based on elements such as heading-body, table header-cell, and label-value. The first and second relationships can be defined through multi-factor interrelationship learning. For example, classification of the first and second relationships can be achieved through learning via various document structures, such as the probability that a signature field and a date field are horizontally aligned being 0.9.

[0120] Edges between nodes can be weighted according to the type of relationship between elements. For example, weights can be assigned in a manner such as header-cell edge weight 1 and adjacent paragraph edge weight 0.5.

[0121] In this invention, semantic-based core information can be extracted more accurately in future models such as LLM through node-based analysis.

[0122] Following the structural analysis step (S420), an LLM-based information extraction step (S430) may be performed. Key information for each semantic unit determined through structural analysis can be mapped through multilingual OCR and LLM recognition. When extracting LLM information, a reliability score for the key information can be determined, and the reliability score can be used as a weight during subsequent review in the review module.

[0123] After the LLM-based information extraction step (S430) is performed, the data structuring step (S440) may be performed. The data is structured in JSON / Protobuf / XML and may include field-specific metadata (location, confidence level, source coordinates).

[0124] FIG. 5 is a conceptual diagram showing the operation of a screening module and an anomaly detection module according to an embodiment of the present invention.

[0125] In FIG. 5, an operation is initiated for the screening module and the anomaly detection module to perform a screening based on the document and determine forgery.

[0126] Referring to Fig. 5, the evaluation module can perform evaluation based on evaluation rules and a reinforcement learning-based artificial intelligence model.

[0127] Through the data structuring step, among the data of the document, evaluation data (510) that can be judged based on evaluation rules and artificial intelligence model evaluation data (520) that can be evaluated based on an artificial intelligence model can be distinguished.

[0128] In the case of the review data (510), it may be data that can be reviewed based on a rule database. For example, it may be an area that can be reviewed within established rules, such as the location of specific information on a document or the existence of a signature.

[0129] Artificial intelligence model evaluation data (520) may be evaluation data that requires judgment, such as judgment on whether specific information that cannot be evaluated by the rule database alone exists, judgment on the incompleteness of information, etc.

[0130] In the screening module, a judgment is made regarding the existence of such information and the incompleteness of the information, and a judgment regarding whether the information has been tampered with can be made in the anomaly detection module.

[0131] First, an examination of the examination data (510) is performed, and if there are no rule violations through the examination data (510), an examination based on the artificial intelligence model examination data (520) can be performed.

[0132] The artificial intelligence model that evaluates the artificial intelligence model evaluation data (520) may include an evaluation model section, a reward model section, and a reinforcement learning loop section.

[0133] The evaluation model unit can perform evaluations on the currently input AI model evaluation data based on evaluation rules. The evaluation model unit can separate the judgment areas for text-based data and image-based data among the AI ​​model data.

[0134] The evaluation model unit may include a text evaluation model unit that evaluates text-based data and an image evaluation model unit that evaluates image-based data.

[0135] Text-based data is data that performs judgments on the correctness and context of text, and judgments can be performed after classifying text within a document by semantic units (personal information, administrative information, educational information, etc.).

[0136] The Text Review Model Unit can determine the correctness of personal information, administrative information, and educational information, respectively, and assess information regarding errors occurring within the text or context. For example, in the case of personal information, the Text Review Model Unit can make judgments regarding the completeness of addresses listed in the personal information, as well as errors in phone numbers, resident registration numbers, etc.

[0137] Image-based data may be data such as signatures or seals existing on a document, and the image examination model unit may perform a judgment on whether an image exists.

[0138] For documents that have passed based on the above screening module, a judgment regarding information manipulation can be performed through the anomaly detection module.

[0139] The anomaly detection module may include a first sub-anomaly detection unit (530), a second sub-anomaly detection unit (540), and a third sub-anomaly detection unit (550).

[0140] The first sub-anomaly detection unit (530) can perform reconstruction error analysis based on VAE (Variational Autoencoder) and, more specifically, analyze reconstruction errors by learning normal document feature vectors.

[0141] The second sub-anomaly detection unit (540) can perform outlier detection based on Isolation Forest and, more specifically, outlier detection based on layout, color, and text density.

[0142] The third sub-anomaly detection unit (550) can perform CNN / Transformer-based pixel-level analysis and, more specifically, detect anomalies through boundary distortion, blurring, and compression trace detection.

[0143] The first subordinate anomaly detection unit (530) can learn the characteristics of a normal document and perform anomaly detection through a learning phase and a verification phase to determine whether the input document deviates from the normal distribution. In the learning phase, a normal document image is input into the VAE, the encoder converts the image into a latent vector, and the decoder generates a reconstructed image from the latent vector, thereby minimizing the difference between the input image and the reconstructed image. After learning, in the verification phase, the input image to be judged can be converted into a latent vector and then changed back into a reconstructed image to be judged. At this time, the reconstruction error between the original image to be judged and the reconstructed image to be judged is calculated, and if the error is greater than or equal to a threshold error, it can be determined that the original image has been tampered with. The first anomaly detection unit can accurately determine forgery, such as seals or signatures included in the examination data.

[0144] The second sub-anomaly detection unit (540) can detect patterns that deviate from normal documents based on quantitative features such as document layout, color distribution, and text density. The second sub-anomaly detection unit (540) can generate feature vectors of normal documents and perform tree-based model learning on the feature vectors.

[0145] Feature vectors can be determined based on layout ratios (tables, text block size / position), color / brightness distribution, text density, and character spacing for the entire document and by region.

[0146] The feature vectors of such normal documents can be learned using a tree-based model. During the tree splitting process, outliers are usually located in extreme values ​​(e.g., abnormally wide margins, unusually high text density) or sparse regions. When splitting lines are drawn randomly, only the point remains after cutting the extreme section once or twice, so it becomes quickly isolated with a short path length. Based on these characteristics, the document to be judged is input to extract feature vectors, and judgment regarding outliers can be performed based on the path length. The second sub-outlier detection unit (540) can detect non-standard templates, involuntary editing traces, and abnormal layouts included in the review data.

[0147] The third sub-anomaly detection unit (550) is a CNN / Transformer-based pixel-level analysis that can detect tampering traces, boundary distortion, blurring, compression patterns, etc., at the pixel level of the document image. In the CNN / Transformer model, the CNN performs local pattern detection (edges, blur, compression artifacts) and can perform anomaly pattern detection that takes into account the context of the Vision Transformer (ViT). Subsequently, it can generate anomaly scores and anomaly pattern masks by pixel / region. The third sub-anomaly detection unit can be used to determine traces of Photoshop / graphic tool modification, JPEG compression block boundary distortion, blurring around text, and boundary discontinuities.

[0148] The above description is merely an illustrative explanation of the technical concept of the present embodiment, and a person skilled in the art to which the present embodiment belongs would be able to make various modifications and variations within the scope of the essential characteristics of the present embodiment. Accordingly, the present embodiments are intended to explain, not limit, the technical concept of the present embodiment, and the scope of the technical concept of the present embodiment is not limited by these embodiments. The scope of protection of the present embodiment shall be interpreted by the claims below, and all technical concepts within an equivalent scope shall be interpreted as being included within the scope of rights of the present embodiment. Explanation of the symbols

[0149] 110: User terminal 120: Network 130: International Student Status Diagnostic Device

Claims

Claim 1 A document analysis module that receives an image of a multilingual administrative document regarding a global international student from a user terminal and extracts multilingual text information from the multilingual administrative document image; a screening module that generates a screening result by automatically screening the status of residence and administrative requirements based on the multilingual text information; and an anomaly detection module that generates a verification result by verifying forgery or alteration of the multilingual administrative document image. And based on the above examination results and verification results, it includes a blockchain management module that encrypts personal information contained in the above multilingual text information and stores it in a blockchain; and when the document analysis module performs preprocessing on the above multilingual administrative document image, such as resolution adjustment, color / brightness equalization, distortion / rotation correction, and noise removal, it divides the above multilingual administrative document image into a first document area containing information within a specific boundary where a boundary exists, including tables and figures within the document, and a second document area where no boundary exists, including characters not included in the boundary on the page, and adjusts the pixel brightness distribution in a different way for each document area; divides the first document area into blocks, calculates the pixel value distribution within the block by calculating a histogram for each block, performs a first histogram equalization for each block to redistribute pixel values ​​evenly, and performs an interpolation procedure between adjacent blocks to smoothly connect the boundaries; for the second document area, it sets the block size larger than the first histogram equalization, and performs a histogram calculation to the pixel value distribution within the block Determines, performs a second histogram equalization that applies a pixel clip limit to each block histogram and disperses histogram values ​​exceeding the limit to surrounding frequencies to prevent noise amplification, and when performing structural analysis following the above preprocessing, performs detection of tables, cell merging, unlabeled forms, and signature / seal locations.A device for diagnosing the status of residence of international students, characterized by performing the structural analysis to infer the relationships between each element within a document, setting each of the text blocks, tables, signature areas, and images in a document or image as nodes for the document structural analysis, determining the relationships between the nodes, and classifying the relationships between the nodes into a first relationship based on the same location and cell connection of the same table within the document as a position-based relationship between elements, and a second relationship based on logic including a paragraph-title connection and a signature field-date connection, thereby performing the structural analysis, wherein the first relationship is determined as a spatial relationship based on up / down, left / right, and adjacency, and the second relationship is determined as a logical relationship based on title-body, table header-cell, and label-value. Claim 2 A diagnostic device for international student residency status according to claim 1, wherein the document analysis module performs multi-element cross-relationship learning on the first relationship and the second relationship to learn a plurality of document structures including the probability that a signature field and a date field are horizontally aligned, and classifies the first relationship and the second relationship. Claim 3 A device for diagnosing the status of residence of international students according to claim 1, wherein the document analysis module assigns weights to edges between nodes according to the type of relationship between elements, in a manner that includes header-cell edge weights and adjacent paragraph edge weights. Claim 4 A process in which a document analysis module receives an image of a multilingual administrative document regarding a global international student from a user terminal and extracts multilingual text information from the multilingual administrative document image; a process in which a screening module generates a screening result by automatically screening the status of residence and administrative requirements based on the multilingual text information; and a process in which an anomaly detection module generates a verification result by verifying forgery or alteration of the multilingual administrative document image. The blockchain management module includes a process of encrypting personal information contained in the multilingual text information and storing it on the blockchain based on the above examination results and verification results, and the document analysis module performs preprocessing on the multilingual administrative document image, such as resolution adjustment, color / brightness equalization, distortion / rotation correction, and noise removal. When performing preprocessing, the multilingual administrative document image is divided into a first document area containing information within a specific boundary, such as tables and figures within the document, where a boundary exists, and a second document area containing characters not included in the boundary on the page, where no boundary exists; the pixel brightness distribution is adjusted differently for each document area; the first document area is divided into block units, and a histogram is calculated for each block to calculate the distribution of pixel values ​​within the block, and then a first histogram equalization is performed for each block to redistribute pixel values ​​evenly, and an interpolation procedure is performed between adjacent blocks to smoothly connect the boundaries; for the second document area, the block size is set larger than the first histogram equalization, and a histogram is performed to determine the pixel values ​​within the block A second histogram equalization is performed to determine the distribution, apply a clip limit to each block histogram, and prevent noise amplification by dispersing histogram values ​​exceeding the limit to surrounding frequencies; and when performing structural analysis following the above preprocessing, detection of tables, cell merging, unlabeled forms, and signature / seal locations is performed.A method for diagnosing the status of residence of international students, characterized by performing a document structure analysis to infer relationships between each element within a document, setting text blocks, tables, signature areas, and images in a document or image as nodes for the document structure analysis, determining the relationships between the nodes, and classifying the relationships between the nodes into a first relationship based on the same location and cell connection of the same table within the document as a position-based relationship between elements, and a second relationship based on logic including a paragraph-title connection and a signature field-date connection, thereby performing the structure analysis, wherein the first relationship is determined as a spatial relationship based on up / down, left / right, and adjacency, and the second relationship is determined as a logical relationship based on title-body, table header-cell, and label-value. Claim 5 A method for diagnosing the status of residence of an international student according to claim 4, wherein the document analysis module performs multi-element interrelationship learning on the first relationship and the second relationship to learn a plurality of document structures including the probability that the signature field and the date field are horizontally aligned, and classifies the first relationship and the second relationship. Claim 6 A method for diagnosing the status of residence of international students according to claim 4, wherein the document analysis module assigns weights to edges between nodes according to the type of relationship between elements, in a manner including header-cell edge weights and adjacent paragraph edge weights.

Images