Occupational identity authentication system based on deep learning technology

Abstract

The invention relates to an occupational identity authentication system based on a deep learning technology, and the system comprises the following steps: an authentication system setting area is divided into a client, an access domain, a capability domain and a storage domain; the client can integrate and output results for users; the access domain is used for realizing request source scheduling and building an extranet communication protocol; the capability domain provides data processing and analysis service capability through a neural network model; the analysis service of the capability domain according to the neural network model comprises face recognition, OCR recognition and silent living body recognition; the storage domain is used for storing decoded picture files, text files, portraits and occupational classification information analyzed by the neural network model; the storage domain stores the information in a database; according to the invention, when a user speaks through the client, the occupational authentication information in the database is called and displayed, the displayed authenticated occupational relationship is the accuracy of the individual speaking on the Internet, and rumors can be avoided to a certain extent.

Description

Technical Field

[0001] The present invention relates to the field of identity authentication technology, and in particular to a professional identity authentication system based on deep learning technology. Background Art

[0002] In information systems, an identity represents a business participant, often represented by a user ID or login name, which may correspond to an organization, an application system, or an individual. Often, due to national policies, legal requirements, or the need to improve service quality, information systems must identify the specific organization or individual to which an identity corresponds. This necessitates matching login names in the information system with the actual organization or individual, leading to the need for real-name authentication.

[0003] The authorization announcement number is CN113657910B, which discloses a real-name authentication method, device, electronic device and readable storage medium. Combined with its specification and drawings, its scheme determines the target weight of each preset indicator item through the first historical authentication data, so that the target authentication channel determined is optimal, which can speed up the real-name authentication speed. However, in the real-name authentication process, the authentication of personal occupation is also crucial. The occupation after authentication is related to the accuracy of the individual's speech on the Internet, which can avoid the generation of rumors to a certain extent. Summary of the Invention

[0004] The present invention mainly aims at the problems existing in the above-mentioned real-name authentication, and invents a professional identity authentication system based on deep learning technology. In the person-document verification link, face comparison and recognition technology and liveness detection technology are realized through a deep network model after multiple iterations. This real-time facial photo is captured and compared with the photo in the background database to ensure that it is the person himself who is performing the operation during professional authentication; when the user speaks through the client, the professional authentication information in the database is called and displayed, and the displayed authenticated occupation is related to the accuracy of the individual's speech on the Internet.

[0005] The purpose of the present invention is to achieve the following technical solution: a professional identity authentication system based on deep learning technology, comprising the following steps:

[0006] S1: The authentication system is divided into client, access domain, capability domain, and storage domain;

[0007] S2: The client can integrate and output results to the user;

[0008] S3: Access domain is used to implement request source scheduling and establish external network communication protocols;

[0009] S4: Capability domain provides data processing and analysis service capabilities through neural network models;

[0010] S5: The capability domain provides analytical services based on neural network models, including face recognition, OCR recognition, and silent liveness detection.

[0011] S6: The storage area is used to store decoded image files, text files, portraits, and occupation classification information after being parsed by the neural network model;

[0012] S61: The storage domain stores the massive amount of images and text information accumulated in the business in multiple physical host databases.

[0013] S62: When the user speaks through the client, the professional certification information in the database is called and displayed.

[0014] Preferably, the client can integrate and output the results to the user including the following steps:

[0015] S21: The client provides a user interaction interface, allowing the user to input commands, data or interact with the application;

[0016] S22: The client pre-processes the data input by the user to conform to the format required by the server;

[0017] S23: The client sends a specific request to the server based on the user's query, file upload, or data download operations;

[0018] S24: The client displays the data or results returned by the server to the user;

[0019] S25: The client provides an external interface to support the intercommunication of basic data with third-party systems.

[0020] Preferably, the access domain is used to implement request source scheduling and establish an external network communication protocol, including the following steps:

[0021] S31: The access domain connects user devices to the network and provides an interface for users to access the network;

[0022] S32: The access domain provides sufficient bandwidth and speed to meet the needs of user devices for network resources and ensure that users can quickly access the network and the Internet;

[0023] S33: The access domain implements VLAN (Virtual Local Area Network) isolation and ACL (Access Control List) filtering to control communication and resource access between user devices.

[0024] Preferably, the capability domain provides data processing and analysis service capabilities through a neural network model, including the following steps:

[0025] S41: Deep network model selection open source initial model architecture;

[0026] S42: The initial model architecture performs data cleaning and necessary preprocessing on historical data to obtain sample data that can be used for training and testing;

[0027] S43: After a limited number of model iterations and transfer learning, the initial model architecture obtains the first generation model;

[0028] S44: The first generation model is stored in the AI algorithm model warehouse and enters the online link;

[0029] S45: After the client service goes online, the first-generation model will automatically acquire and store the actual scene data collected online and publish it as a task on the annotation platform;

[0030] S46: The first-generation model manages model versions through the version management function, evaluates and selects better models to replace the online model;

[0031] S47: The first-generation model automatically iterates and optimizes the model, continuously improving the recognition accuracy of the system.

[0032] Preferably, the face recognition service provided by the capability domain based on the neural network model includes the following steps:

[0033] S51: The neural network model obtains facial features of the verification person through graphics;

[0034] S511: Acquire a video or image containing a human face through a camera device;

[0035] S512: The neural network model learns deep features suitable for face detection from the data to classify the face area and the background area, and then locates the face area;

[0036] S513: After the neural network model obtains the position of the face detection frame, it normalizes the face size to a uniform scale and eliminates the influence of the face posture on subsequent face recognition;

[0037] S514: The neural network model locates several key points of the face, and performs scale normalization and facial posture correction based on these key points, ultimately obtaining a face image that is convenient for face recognition;

[0038] S515: The neural network model is trained using massive amounts of data to extract features that best represent facial identity information.

[0039] Preferably, the silent living body recognition based on the analytical service of the neural network model of the capability domain includes the following steps:

[0040] S521: When the user stays in front of the front-end image acquisition device, the neural network model analyzes one or more frames of images in the video stream;

[0041] S522: The neural network model can fuse multiple frames of images to determine whether the person applying for the professional real-name authentication service is a real person.

[0042] Preferably, the OCR recognition of the capability domain based on the parsing service of the neural network model includes the following steps:

[0043] S53: The neural network model uses images of social security payment information to convert the image content into text that can be recognized and recorded by computers;

[0044] S531: Using a camera to capture a picture of the user's company's social security payment information;

[0045] S532: The neural network model locates corner information or edge information of occupational social security payment information in the image.

[0046] S533: The neural network model uses image processing methods such as perspective transformation to correct the occupational social security payment information area obtained in the previous step into a rectangular area, and then locates the area where the text to be recognized is located within the area;

[0047] S534: The neural network model recognizes the detected text region slices as text and classifies individual characters;

[0048] S535: The neural network model compares the obtained social security text information and facial feature information with the backend database to verify the authenticity of the identity information;

[0049] S536: After the neural network model compares the information and finds it correct, it obtains the company's business scope based on the company entity that pays the social security;

[0050] S537: The neural network model generates the occupational category of the real-name authenticator based on the occupational classification code and the company's experience range.

[0051] As a preferred embodiment, compared with the existing technology, the present invention has the following beneficial effects: 1. In the process of verifying personal information, the face matching and recognition technology and the liveness detection technology are realized through the deep network model after multiple iterations. This real-time capture of face photos and the background database photos for face comparison ensures that it is the person himself who is doing the job authentication, and eliminates the use of other people's information for job authentication; 2. The entire system uses a neural network model to slice and recognize the detected text area into text, and classify individual characters; the obtained social security text information and facial feature information are compared with the background database online to verify the authenticity of the identity information; after the information is compared, the company's business scope is obtained based on the company entity that pays the social security; finally, the real-name authentication person's occupational category is automatically generated based on the occupational classification code and the company's experience range; 3. When the user speaks through the client, the occupational authentication information in the database is called and displayed. The displayed occupation after authentication is related to the accuracy of the individual's speech on the Internet, which can avoid the generation of rumors to a certain extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a schematic diagram of the front and rear separation mode of the present invention;

[0053] Figure 2 This is a training flow chart of the neural network model of the present invention. DETAILED DESCRIPTION

[0054] The present invention will be further described below with reference to the embodiments shown in the accompanying drawings:

[0055] like Figures 1 to 2 As shown in the figure, a professional identity authentication system based on deep learning technology is designed. The entire system adopts a modular design. Different modules can be freely combined according to needs, and data is connected and managed uniformly on the platform. The entire system adopts a front-end and back-end separation mode, and the back-end adopts a cloud deployment solution to ensure that the client can quickly access the system through the network. The back-end management system can log in and manage it anytime and anywhere through the network. At the same time, all internal and external interfaces have a unified design style to facilitate third-party docking and development, including the following steps:

[0056] S1: The authentication system is divided into client, access domain, capability domain, and storage domain;

[0057] S2: The client integrates and outputs results to the user;

[0058] S21: The client provides a user interaction interface, allowing the user to input commands, data or interact with the application;

[0059] S22: The client pre-processes the data input by the user to conform to the format required by the server;

[0060] S23: The client sends a specific request to the server based on the user's query, file upload, or data download operations;

[0061] S24: The client displays the data or results returned by the server to the user;

[0062] S25: The client provides an external interface to support the intercommunication of basic data with third-party systems;

[0063] S3: Access domain is used to implement request source scheduling and establish external network communication protocols;

[0064] S31: The access domain connects user devices (such as computers, mobile phones, printers, etc.) to the network and provides an interface for users to access the network;

[0065] S32: The access domain provides sufficient bandwidth and speed to meet the needs of user devices for network resources and ensure that users can quickly access the network and the Internet;

[0066] S33: The access domain implements VLAN (virtual local area network) isolation and ACL (access control list) filtering to control communication and resource access between user devices;

[0067] S4: Capability domain provides data processing and analysis service capabilities through neural network models;

[0068] S41: Deep network model selection open source initial model architecture;

[0069] In step 41, the network model is selected based on the analysis of the learning task and the specific business scenario;

[0070] S42: The initial model architecture performs data cleaning and necessary preprocessing on historical data (including images, videos, and audio) to obtain sample data that can be used for training and testing;

[0071] S43: After a limited number of model iterations and transfer learning, the initial model architecture obtains the first generation model;

[0072] S44: The first generation model is stored in the AI algorithm model warehouse and enters the online link;

[0073] S45: After the client service goes online, the first-generation model will automatically acquire and store the actual scene data collected online and publish it as a task on the annotation platform;

[0074] In step S45, data annotation is performed manually or semi-automatically, so that the first generation model will continuously acquire new training data, thereby continuously optimizing the model;

[0075] S46: The first-generation model manages model versions through the version management function, evaluates and selects better models to replace the online model;

[0076] S47: The first-generation model automatically iterates and optimizes the model, continuously improving the recognition accuracy of the system.

[0077] S5: The capability domain provides analytical services based on neural network models, including face recognition, OCR recognition, and silent liveness detection.

[0078] S51: The neural network model obtains facial features of the verification person through graphics;

[0079] S511: Acquire a video or image containing a human face through a camera device;

[0080] S512: The neural network model learns deep features suitable for face detection from the data to classify the face area and the background area, and then locates the face area;

[0081] S513: After the neural network model obtains the position of the face detection frame, it normalizes the face size to a uniform scale and eliminates the influence of the face posture on subsequent face recognition;

[0082] S514: The neural network model locates several key points of the face, and performs scale normalization and facial posture correction based on these key points, ultimately obtaining a face image that is convenient for face recognition;

[0083] S515: The neural network model is trained using massive amounts of data to extract features that best represent facial identity information.

[0084] In step S515, the process of acquiring features can use a method based on metric learning to make the facial images of the same person as close as possible after feature transformation, and the facial images of different people as far apart as possible. Typical methods include Contrastive Loss and Triplet Loss, and by introducing this variable, the discrimination ability of face recognition features is improved. Typical methods include Large-Margin Softmax Loss, SphereFace, CosFace and ArcFace.

[0085] S52: Neural network model for silent liveness detection;

[0086] S521: When the user stays in front of the front-end image acquisition device, the neural network model analyzes one or more frames of images in the video stream;

[0087] S522: The neural network model can fuse multiple frames of images to determine whether the person applying for the professional real-name authentication service is a real person.

[0088] S53: The neural network model uses images of social security payment information to convert the image content into text that can be recognized and recorded by computers;

[0089] S531: Using a camera to capture a picture of the user's company's social security payment information;

[0090] S532: The neural network model locates corner information or edge information of occupational social security payment information in the image.

[0091] S533: The neural network model uses image processing methods such as perspective transformation to correct the occupational social security payment information area obtained in the previous step into a rectangular area, and then locates the area where the text to be recognized is located within the area;

[0092] In step S533, the neural network model uses the EAST algorithm to accurately locate the key fields that need to be identified, while shielding irrelevant information fields, thereby improving the recognition accuracy of the entire OCR algorithm.

[0093] S534: The neural network model recognizes the detected text region slices as text and classifies individual characters;

[0094] S535: The neural network model compares the obtained social security text information and facial feature information with the backend database to verify the authenticity of the identity information;

[0095] S536: After the neural network model compares the information and finds it correct, it obtains the company's business scope based on the company entity that pays the social security;

[0096] S537: The neural network model generates the occupational category of the real-name authenticator based on the occupational classification code and the company's experience range.

[0097] S6: The storage area is used to store decoded image files, text files, portraits, and occupation classification information after being parsed by the neural network model;

[0098] S61: The storage domain uses a distributed storage solution to store the massive amount of images accumulated during business operations. These images are stored on multiple physical host hard drives. When images are stored, their basic information is also stored in MySQL and Oracle relational databases.

[0099] In step S61, to ensure the security of relevant data information, the image data and image information are desensitized and encrypted when stored;

[0100] S62: When the user speaks through the client, the professional certification information in the database is called and displayed.

[0101] In step S62, the authenticated occupation is displayed, which is related to the accuracy of the individual's speech on the Internet and can avoid the generation of rumors to a certain extent.

[0102] Due to the storage of highly sensitive data such as domains and user information, a highly reliable database management system and physical storage function are adopted. Sensitive information such as customer information is encrypted and stored in a "soft + hard" dual manner. The database adopts strict permission management and has backup and recovery functions. In addition, digital signature measures are added, and multiple means such as console security, machine physical isolation, and protection of installation media are adopted to ensure the reliability, integrity, and confidentiality of data storage.

[0103] Based on the front-end and back-end separation authentication technology in steps S1 to S6, the "Internet +" model is adopted. The front-end only collects customer information and transmits all collected ciphertext information to the back-end server for centralized decryption, identification and authentication across the entire network.

[0104] Methods for obtaining customer information include taking photos and uploading identity, occupation, and social security payment information, manual entry, and taking photos of customers on site. Occupational information and related personal information are collected through the above channels, encrypted and transmitted, and sent back to the real-name authentication background for information decryption and identity consistency verification. After passing the system authentication, the individual's occupation is recorded in the background storage domain.

[0105] The specific embodiments described herein are merely illustrative of the spirit of the present invention. Persons skilled in the art may make various modifications, additions, or substitutions to the described specific embodiments without departing from the spirit of the present invention or exceeding the scope of the appended claims.

Claims

1. A professional identity authentication system based on deep learning technology, characterized in that: The method comprises the following steps: S1: the authentication system is divided into a client, an access domain, a capability domain and a storage domain; S2: The client can integrate and output results to the user; S3: Access domain is used to implement request source scheduling and establish external network communication protocols; S4: Capability domain provides data processing and analysis service capabilities through neural network models; S5: The capability domain provides analytical services based on neural network models, including face recognition, OCR recognition, and silent liveness detection. S6: The storage area is used to store decoded image files, text files, portraits, and occupation classification information after being parsed by the neural network model; S61: The storage domain stores the massive amount of images and text information accumulated in the business in multiple physical host databases. S62: When the user speaks through the client, the professional certification information in the database is called and displayed.

2. The occupational identity authentication system based on deep learning technology according to claim 1 is characterized in that: The client can integrate and output results to the user including the following steps: S21: The client provides a user interaction interface, allowing the user to input commands, data or interact with the application; S22: The client pre-processes the data input by the user to conform to the format required by the server; S23: The client sends a specific request to the server based on the user's query, file upload, or data download operations; S24: The client displays the data or results returned by the server to the user; S25: The client provides an external interface to support the intercommunication of basic data with third-party systems.

3. The occupational identity authentication system based on deep learning technology according to claim 2 is characterized in that: The access domain is used to implement request source scheduling and build external network communication protocols, including the following steps: S31: The access domain connects user devices to the network and provides an interface for users to access the network; S32: The access domain provides sufficient bandwidth and speed to meet the needs of user devices for network resources and ensure that users can quickly access the network and the Internet; S33: The access domain implements VLAN (Virtual Local Area Network) isolation and ACL (Access Control List) filtering to control communication and resource access between user devices.

4. The occupational identity authentication system based on deep learning technology according to claim 3 is characterized in that: The capability domain provides data processing and analysis service capabilities through neural network models, including the following steps: S41: Deep network model selection open source initial model architecture; S42: The initial model architecture performs data cleaning and necessary preprocessing on historical data to obtain sample data that can be used for training and testing; S43: After a limited number of model iterations and transfer learning, the initial model architecture obtains the first generation model; S44: The first generation model is stored in the AI algorithm model warehouse and enters the online link; S45: After the client service goes online, the first-generation model will automatically acquire and store the actual scene data collected online and publish it as a task on the annotation platform; S46: The first-generation model manages model versions through the version management function, evaluates and selects better models to replace the online model; S47: The first-generation model automatically iterates and optimizes the model, continuously improving the recognition accuracy of the system.

5. The occupational identity authentication system based on deep learning technology according to claim 4 is characterized in that: The face recognition service provided by the capability domain based on the neural network model includes the following steps: S51: The neural network model obtains facial features of the verification person through graphics; S511: Acquire a video or image containing a human face through a camera device; S512: The neural network model learns deep features suitable for face detection from the data to classify the face area and the background area, and then locates the face area; S513: After the neural network model obtains the position of the face detection frame, it normalizes the face size to a uniform scale and eliminates the influence of the face posture on subsequent face recognition; S514: The neural network model locates several key points of the face, and performs scale normalization and facial posture correction based on these key points, ultimately obtaining a face image that is convenient for face recognition; S515: The neural network model is trained using massive amounts of data to extract features that best represent facial identity information.

6. The occupational identity authentication system based on deep learning technology according to claim 5 is characterized in that: The capability domain of silent living body recognition based on the analysis service of the neural network model includes the following steps: S521: When the user stays in front of the front-end image acquisition device, the neural network model analyzes one or more frames of images in the video stream; S522: The neural network model can fuse multiple frames of images to determine whether the person applying for the professional real-name authentication service is a real person.

7. The occupational identity authentication system based on deep learning technology according to claim 6 is characterized in that: The OCR recognition of the capability domain based on the neural network model analysis service includes the following steps: S53: The neural network model uses images of social security payment information to convert the image content into text that can be recognized and recorded by computers; S531: Using a camera to capture a picture of the user's company's social security payment information; S532: The neural network model locates corner information or edge information of occupational social security payment information in the image. S533: The neural network model uses image processing methods such as perspective transformation to correct the occupational social security payment information area obtained in the previous step into a rectangular area, and then locates the area where the text to be recognized is located within the area; S534: The neural network model recognizes the detected text region slices as text and classifies individual characters; S535: The neural network model compares the obtained social security text information and facial feature information with the backend database to verify the authenticity of the identity information; S536: After the neural network model compares the information and finds it correct, it obtains the company's business scope based on the company entity that pays the social security; S537: The neural network model generates the occupational category of the real-name authenticator based on the occupational classification code and the company's experience range.

Images