Method and system for preventing mobile phone card fraud

Abstract

The application provides a method and system for preventing mobile phone card fraud, relating to the field of wireless communication. After receiving a mobile phone card replacement / loss report, an identity verification request is sent to the user device, and real-time collection of the user's face image and automatic extraction of recent behavior records (such as login frequency, operation habits, call and message usage behavior, etc.) are triggered. After these data are transmitted to the cloud platform, the face features and behavior semantics are deeply associated and modeled through multi-modal fusion analysis technology, thereby dynamically generating a risk level assessment result, and responding according to the result. In this way, relying on the cloud intelligent analysis capability, while hiding the complexity of the underlying algorithm, the accurate interception and active defense of abnormal application behavior are realized, providing a basis for building a trusted digital identity authentication system.

Description

Technical Field

[0001] The present application relates to the field of wireless communications, and more particularly, in an embodiment of the present application, relates to a method and system for preventing mobile phone number card fraud. Background Art

[0002] With the rapid development of mobile communications technology and the widespread adoption of smart devices, the security of mobile phone numbers, as the core carrier of users' digital identities, is directly related to personal privacy protection, financial asset security, and social public order. In recent years, telecom fraud cases centered around SIM card theft have become increasingly prevalent. Fraudsters illegally gain control of user numbers by forging identity information and intercepting SMS verification codes, allowing them to commit crimes such as account theft and fund transfer. The increasing stealth and technical sophistication of these attacks exposes the shortcomings of traditional identity verification mechanisms in dynamically identifying risks and linking them to multi-dimensional user characteristics.

[0003] Existing technologies typically rely on single-factor authentication methods, such as SMS verification codes and manual customer service Q&A. These methods are limited in their security capabilities by static data comparison and the efficiency of manual review. These methods are particularly vulnerable to biometric attacks based on deepfake technology, particularly in unusual cross-regional and cross-device traffic scenarios. Furthermore, traditional mobile phone card fraud risk control systems focus on single-dimensional anomaly detection and lack the ability to fine-grainedly model multi-dimensional user behavior patterns (such as terminal operation habits, communication traffic characteristics, and service request cycles). This results in delayed risk assessment and a high rate of misjudgment.

[0004] Therefore, a solution to prevent mobile phone number card fraud is desired. Summary of the Invention

[0005] The present application is proposed to address the aforementioned technical issues. The embodiments of the present application provide a method and system for preventing mobile phone card fraud. Upon receiving a mobile phone card replacement / loss report application, the system sends an identity verification request to the user's device, triggering the real-time collection of the user's facial image and the automatic extraction of recent behavior records (such as login frequency, operating habits, call and text message usage, and data usage). After transmitting this data to the cloud platform, multimodal fusion analysis technology is used to perform deep correlation modeling between facial features and behavioral semantics, thereby dynamically generating risk level assessment results and responding accordingly. In this way, relying on the cloud's intelligent analysis capabilities, while concealing the complexity of the underlying algorithm, it achieves precise interception and proactive defense against abnormal application behavior, providing a foundation for building a trusted digital identity authentication system.

[0006] According to one aspect of the present application, a method for preventing mobile phone number card fraud is provided, comprising:

[0007] Receive applications for replacement / loss of mobile phone cards and send identity verification request information to the device terminal of the user to be verified;

[0008] In response to the device terminal receiving the identity authentication request information, collecting the face recognition image and recent historical behavior records of the user to be authenticated;

[0009] Transmitting the facial recognition image and the recent historical behavior records to a cloud platform for data processing to analyze the risk level of the mobile phone card replacement / loss report application, including: performing a semantically significant interaction analysis based on a feature domain on the facial recognition image of the user to be verified and the recent historical behavior records on the cloud platform to determine the risk level of the mobile phone card replacement / loss report application;

[0010] Execute response operations based on the risk level of the mobile phone number card replacement / loss report application.

[0011] According to another aspect of the present application, a system for preventing mobile phone number card fraud is provided, comprising:

[0012] The mobile phone card change verification module is used to receive mobile phone card replacement / loss report applications and send identity authentication request information to the device terminal of the user to be verified;

[0013] A module for collecting information of a user to be verified, configured to collect a facial recognition image and recent historical behavior records of the user to be verified in response to the device terminal receiving the identity authentication request information;

[0014] a risk level assessment module, configured to transmit the facial recognition image and the recent historical behavior records to a cloud platform for data processing to analyze the risk level of the mobile phone card replacement / loss report application, and to: perform a semantically significant interaction analysis based on a feature domain on the facial recognition image and the recent historical behavior records of the user to be verified on the cloud platform to determine the risk level of the mobile phone card replacement / loss report application;

[0015] The risk level response operation execution module is used to execute a response operation based on the risk level of the mobile phone card replacement / loss report application.

[0016] Compared to existing technologies, the present application provides a method and system for preventing mobile phone card fraud. Upon receiving a request for replacement or loss of a mobile phone card, the system sends an identity verification request to the user's device, triggering the real-time collection of the user's facial image and the automatic extraction of recent behavior records (such as login frequency, operating habits, call and text message usage, and data usage). After transmitting this data to the cloud platform, multimodal fusion analysis technology is used to perform deep correlation modeling between facial features and behavioral semantics, thereby dynamically generating risk level assessment results and responding accordingly. In this way, relying on the cloud's intelligent analysis capabilities, while concealing the complexity of the underlying algorithm, it achieves precise interception and proactive defense against abnormal application behavior, providing a foundation for building a trusted digital identity authentication system. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0018] Figure 1 The present invention is a flowchart of a method for preventing mobile phone number card fraud according to an embodiment of the present application.

[0019] Figure 2 This is a flowchart of performing semantically significant interaction analysis based on feature domains on the facial recognition image of the user to be verified and the recent historical behavior records in the method for preventing mobile phone number card fraud according to an embodiment of the present application to determine the risk level of the mobile phone number card replacement / loss report application.

[0020] Figure 3 A data flow diagram for performing feature domain-based semantically significant interaction analysis on the facial recognition image of the user to be verified and the recent historical behavior records in a method for preventing mobile phone number card fraud according to an embodiment of the present application to determine the risk level of the mobile phone number card replacement / loss report application.

[0021] Figure 4 A flowchart of performing a user multimodal identity authentication analysis based on the facial state-behavior semantic interaction structure on the facial state features and the historical behavior semantic features in the method for preventing mobile phone number card fraud according to an embodiment of the present application to obtain facial state-behavior semantic significant interaction features.

[0022] Figure 5 This is a system block diagram of a system for preventing mobile phone number card fraud according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] Various exemplary embodiments, features, and aspects of the present application will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise indicated.

[0024] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0025] In addition, numerous specific details are provided in the following detailed description to better illustrate the present application. Those skilled in the art will appreciate that the present application can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art are not described in detail in order to highlight the main purpose of the present application.

[0026] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. Throughout the description of this application, "plurality" means two or more, unless otherwise specifically defined.

[0027] With the advancement of mobile communications technology and the widespread adoption of smart devices, mobile phone numbers and SIM cards have become a crucial carrier of users' digital identities, and their security is crucial for privacy protection, financial security, and social order. However, the recent surge in telecom fraud involving SIM card theft has exposed the shortcomings of traditional identity verification mechanisms, such as SMS verification codes and manual customer service Q&A, in dynamically identifying risks and linking them to multi-dimensional user characteristics. These single-factor authentication methods are vulnerable to deepfake attacks across multiple regions and devices, and traditional risk control systems lack the ability to analyze user behavior patterns in a granular manner, resulting in high lag and false positive rates.

[0028] To address the above technical issues, this application proposes a method for preventing mobile phone card fraud by building an intelligent verification system that integrates multimodal biometrics and dynamic historical user behavior analysis. Upon receiving a request for a mobile phone card replacement or loss report, facial recognition technology and historical behavior analysis are used to comprehensively verify the user's identity. If the risk is high, a manual customer call is conducted for secondary verification, preventing mobile phone card fraud and ensuring the security of SIM card transfers.

[0029] Specifically, when a user requests a replacement or report a lost mobile phone card, an identity verification request is sent to the user's device, triggering the real-time collection of the user's facial image and the automatic extraction of recent behavioral records (such as login frequency, operating habits, call and text message usage, and data usage). After this data is transmitted to the cloud platform, multimodal fusion analysis technology is used to deeply correlate facial features and behavioral semantics to dynamically generate a risk level assessment result. If the risk is determined to be low, the business is automatically processed; if the risk is medium, the process is transferred to a manual verification process (such as a follow-up call or video verification) to ensure the authenticity of the operation; in high-risk scenarios, the application is directly intercepted and a security alert is triggered, and the user or relevant institutions are notified simultaneously. The entire process relies on cloud-based intelligent analysis capabilities. While hiding the complexity of the underlying algorithm, it achieves precise interception and proactive defense against abnormal application behavior, providing a foundation for building a trusted digital identity authentication system.

[0030] In response to the above technical problems, this application proposes a method to prevent mobile phone number card fraud. Figure 1 Flowchart of the method for preventing mobile phone number card fraud according to an embodiment of the present application. Figure 1 As shown, according to an embodiment of the present application, the method for preventing mobile phone number card fraud includes: S110, receiving a mobile phone number card replacement / loss report application, and sending identity authentication request information to the device terminal of the user to be verified; S120, in response to the device terminal receiving the identity authentication request information, collecting the face recognition image and recent historical behavior records of the user to be verified; S130, transmitting the face recognition image and the recent historical behavior records to the cloud platform for data processing to analyze the risk level of the mobile phone number card replacement / loss report application; S140, performing a response operation based on the risk level of the mobile phone number card replacement / loss report application.

[0031] In the above-mentioned method for preventing mobile phone card fraud, step S110 receives a mobile phone card replacement / loss report application and sends an identity verification request to the device terminal of the user to be verified. It should be understood that the operation of receiving the mobile phone card replacement / loss report application is performed in the core system of the communication network platform or service provider. When a user submits a mobile phone card replacement or loss report application, the application is received through the operator's online service interface or related application. Users typically submit the application through a customer service hotline, official website, mobile application, other online platforms, or offline channels. During the application process, the user is required to provide necessary identity information, such as ID number, mobile phone number, and additional verification information (such as verification code, account password, etc.). Once the operator's system receives the user's replacement / loss report application, a series of verification mechanisms are triggered. To prevent fraud, especially identity forgery or malicious loss reports, the accuracy and real-time nature of the verification must be ensured. During this step, a verification request message is generated based on the basic identity information provided by the user, such as mobile phone number and account information. This request message typically includes the user's identity identifier (such as mobile phone number and account information) and relevant information about the verification task. This method ensures that the application originates from a genuine user and prevents malicious attackers from posing as legitimate users. Verification requests are typically sent via mobile networks, the internet, or other communication methods. Common methods include sending push notifications, text messages, emails, or in-app messages to the user's device. These verification requests rely on secure encryption technology during network transmission to prevent tampering or theft, thus minimizing risks such as man-in-the-middle attacks. After receiving the authentication request, the user's device performs the corresponding actions based on the request. Typically, the user responds to the authentication request through the application or system interface on the device. The user enters relevant information on the device, such as an SMS verification code, a secondary password, or fingerprint or facial recognition. The core goal of this process is to ensure that the device's actions accurately match the user's identity and verify that the user has the legal authority to replace or report a lost mobile phone card. If the authentication on the device passes, the user's identity is confirmed, and the subsequent action can proceed. During the authentication request process, relevant verification logs are recorded, which not only facilitates subsequent audits and backtracking but also provides data support for risk assessment. The steps of receiving mobile phone card replacement / loss reports and sending identity verification requests to the user's device provide reliable foundational data for subsequent behavioral data collection, risk assessment, and anomaly detection. This step is designed with both user experience and security in mind, ensuring user convenience while minimizing the potential for fraud.

[0032] In the above-mentioned method for preventing mobile phone number card fraud, step S120 involves collecting the facial recognition image and recent historical behavior records of the user to be verified in response to the device terminal receiving the identity verification request. It should be understood that upon receiving the identity verification request, the device terminal enters the user data collection phase. The core objective of this phase is to provide a comprehensive basis for subsequent risk assessment and decision-making by collecting the user's biometric and behavioral data in real time, combined with multimodal analysis techniques. Specifically, the device terminal initiates a series of pre-set programs to ensure that the user's data can be collected in a relatively short period of time. This data includes, but is not limited to, the facial recognition image and recent historical behavior records. First, to collect the facial recognition image, the device terminal typically uses a built-in camera or an external camera to capture the user's facial image. This process is executed on the user-side device, and the relevant program within the device automatically initiates upon receiving the identity verification request. At this point, an algorithm performs real-time facial detection on the user's face to ensure that it meets the required facial features. These features include, but are not limited to, the user's facial contour, facial features, and facial expressions. Based on this, the user's facial image features are generated. This process relies not only on the device's hardware camera capabilities but also on powerful image recognition algorithms to effectively capture accurate facial data even under complex lighting conditions. Furthermore, after capturing the facial recognition image, the device terminal also needs to obtain the user's recent historical behavior records. This historical behavior record typically includes, but is not limited to, the user's login frequency, operational habits, text message and call logs, and traffic and data usage records. This data collection primarily relies on the device terminal's system logs and automatic logging of user operations. This historical behavior data can be used to analyze user operational characteristics across different time periods and scenarios, providing an effective basis for subsequent risk assessment. For example, user login frequency and login location (whether they frequently change devices or login locations) can serve as indicators of abnormal user behavior. Significant changes in a user's login method or frequency are considered potential risk behavior, leading to enhanced verification measures. Furthermore, user operational habits are an important component of the behavioral record. For example, whether a user frequently changes information or reports a lost device during use can reveal user habits and operational patterns, helping to determine whether current behavior conforms to normal user behavior. By combining these historical behavior records with real-time facial recognition images, we can determine whether the current operation poses a fraud risk based on certain rules or models. If the user's behavior is seriously inconsistent with historical behavior, or if certain abnormal patterns trigger early warning mechanisms, the application may be judged as high-risk, triggering secondary verification or other security measures.

[0033] In the above-mentioned method for preventing mobile phone number card fraud, the step S130 transmits the facial recognition image and the recent historical behavior records to the cloud platform for data processing to analyze the risk level of the mobile phone number card replacement / loss report application, including: on the cloud platform, performing a semantically significant interaction analysis based on a feature domain on the facial recognition image of the user to be verified and the recent historical behavior records to determine the risk level of the mobile phone number card replacement / loss report application.

[0034] In particular, in the aforementioned method for preventing mobile phone card fraud, analyzing the risk level of mobile phone card replacement / lost card applications and formulating a graded response based on this is crucial, as it directly determines the balance between security protection effectiveness and user experience. Specifically, due to the rapid iteration of telecom fraud methods, relying solely on single-dimensional static verification (such as SMS verification codes) or manual review processes is not only inadequate for addressing emerging threats such as deepfake attacks and cross-regional anomalous operations, but may also reduce service processing efficiency by excessively intercepting legitimate user requests. Therefore, the technical solution of this application utilizes multimodal fusion analysis technology to dynamically correlate facial semantics and historical behavioral semantics during risk level identification. This allows for the construction of a risk prediction model based on the dual dimensions of biometric authenticity and recent historical behavioral patterns. Furthermore, by dynamically matching risk level with verification intensity, this graded response mechanism avoids redundant verification in low-risk scenarios (e.g., rapid approval of service requests frequently initiated by users on commonly used devices) while also enabling the introduction of proactive defense measures such as manual verification and security alerts in medium- and high-risk scenarios. This layered processing strategy not only optimizes resource allocation, but also implements differentiated countermeasures for different threat levels: for medium-risk applications, secondary verification of key user identity information (such as details of recent call records) through telephone follow-up can prevent social engineering attacks; for high-risk scenarios, direct blocking and linkage with the anti-fraud center can minimize derivative risks such as mobile phone number fraud, effectively filling the gap in the traditional solutions' capabilities in real-time response and complex attack defense.

[0035] Figure 2 This is a flowchart of performing semantically significant interaction analysis based on feature domains on the facial recognition image of the user to be verified and the recent historical behavior records in the method for preventing mobile phone number card fraud according to an embodiment of the present application to determine the risk level of the mobile phone number card replacement / loss report application. Figure 3 This is a data flow diagram for performing semantically significant interaction analysis based on feature domains on the face recognition image of the user to be verified and the recent historical behavior records in the method for preventing mobile phone card fraud according to an embodiment of the present application to determine the risk level of the mobile phone card replacement / loss report application. Figure 2 and Figure 3As shown, in an embodiment of the present application, the step S130 performs a feature domain-based semantic significant interaction analysis on the face recognition image of the user to be verified and the recent historical behavior record to determine the risk level of the mobile phone number card replacement / report loss application, including: S131, passing the face recognition image through a Capsule Network-based face feature extractor to obtain a face state feature vector as a face state feature; S132, passing the multiple behavior record data items in the recent historical behavior record through a fully connected layer-based co-space domain mapping association encoder to obtain a recent historical behavior semantic association coding vector as a historical behavior semantic feature; S133, performing a user multimodal identity authentication analysis based on the face state-behavior semantic interaction structure on the face state feature and the historical behavior semantic feature to obtain a face state-behavior semantic significant interaction feature; S134, performing a mobile phone number card risk level detection based on the face state-behavior semantic significant interaction feature to determine the risk level of the mobile phone number card replacement / report loss application.

[0036] Specifically, in step S131, the face recognition image is passed through a face feature extractor based on Capsule Network to obtain a face state feature vector as a face state feature. It should be understood that considering the insufficient defense capability of traditional face recognition technology against deep fake attacks. Specifically, in the prior art, fraudsters often use counterfeiting methods such as high-definition 3D masks and dynamic video synthesis to bypass static feature extraction models. Therefore, in the technical solution of the present application, the face recognition image is passed through a face feature extractor based on Capsule Network to obtain a face state feature vector. Capsule Network can capture facial micro-expressions, light reflection characteristics and dynamic texture changes unique to living organisms through dynamic routing mechanisms and posture relationship modeling, thereby effectively distinguishing real faces from counterfeit media. This technical choice is directly aimed at the biometric counterfeiting scenarios that frequently occur in current SIM card theft attacks, and aims to solve the problem of insufficient robustness of liveness detection in traditional single-modality verification. From an implementation perspective, the facial state feature vectors extracted via the Capsule Network not only contain information about facial geometry but also integrate dynamic physiological features in the live state (such as pupil focus trajectory and skin blood oxygen fluctuations), providing high-fidelity biometric representation for subsequent multimodal identity verification. This feature extraction method can significantly improve the recognition accuracy of attacks such as synthetic faces and photo remakes. For example, when fraudsters use AI-generated dynamic videos for verification, the model can accurately determine the authenticity of the biometric by analyzing optical flow anomalies in the eye area or unnatural continuity of facial muscle movements, thereby providing reliable data support for the dynamic assessment of mobile phone number fraud risk levels.

[0037] Specifically, step S132 involves applying a co-spatial domain mapping association encoder based on a fully connected layer to multiple behavior record data items in the recent historical behavior record to obtain a recent historical behavior semantic association encoding vector as a historical behavior semantic feature. It should be understood that in the prior art, discrete data such as user login frequency, operating habits, and communication behavior are typically evaluated individually or simply weighted, making it impossible to effectively identify abnormal behavior chains with spatiotemporal coupling characteristics, such as "high frequency of remote logins followed by immediate card replacement applications." Based on this, in the technical solution of the present application, multiple behavior record data items in the recent historical behavior record are applied to a co-spatial domain mapping association encoder based on a fully connected layer to obtain a recent historical behavior semantic association encoding vector. The co-spatial domain mapping model constructed by the fully connected layer can project behavioral data of different dimensions (such as a sudden increase in the number of logins per day and an abnormal shift in traffic usage time) into a unified semantic space. Nonlinear activation functions can be used to explore potential associations between features. For example, by analyzing the blank period in call records and the sudden change in SMS reception patterns in the week before a user's card replacement, a dynamic behavior baseline can be constructed, which can abstract the inherent laws of behavioral patterns from the user's historical operations. For example, while legitimate users typically use data services during fixed hours, fraudsters might employ an unconventional pattern, such as immediately applying for a new card after changing their device. Using a co-spatial domain mapping associative encoder based on a fully connected layer, the model automatically learns the nonlinear relationships between different behavioral features (such as the mutation coefficient of login locations and the discreteness of SMS recipients), thereby generating a semantic encoding vector that reflects the rationality of user behavior. This encoding approach not only preserves the statistical properties of the original data but also enhances the cross-dimensional interaction of recent historical behavioral features through spatial domain mapping.

[0038] Figure 4 A flowchart of performing multimodal user identity authentication analysis based on the face state-behavior semantic interaction structure on the face state feature and the historical behavior semantic feature in the method for preventing mobile phone number card fraud according to an embodiment of the present application to obtain the face state-behavior semantic significant interaction feature. Figure 4As shown, in an embodiment of the present application, the step S133 performs user multimodal identity authentication analysis on the facial state features and the historical behavior semantic features based on the facial state-behavior semantic interaction structure to obtain facial state-behavior semantic significant interaction features, including: S1331, performing affine mapping processing based on principal component analysis on the facial state feature vector to obtain a set of facial state principal component affine mapping feature coding vectors; S1332, performing modal decoupling constraint analysis on the set of facial state principal component affine mapping feature coding vectors and the recent historical behavior semantic association coding vectors to determine a set of facial state-behavior semantic principal component modal decoupling flexible constraint factors; S1333, based on the set of facial state-behavior semantic principal component modal decoupling flexible constraint factors, performing adaptive dynamic aggregation on the set of facial state principal component affine mapping feature coding vectors and the recent historical behavior semantic association coding vectors to obtain facial state-behavior semantic significant interaction coding vectors as the facial state-behavior semantic significant interaction feature. It should be understood that the simple concatenation or weighted superposition of facial recognition features and historical behavioral semantic features often leads to information redundancy and semantic fragmentation. For example, when fraudsters use deepfakes to generate highly realistic faces, relying solely on biometric verification may overlook spatiotemporal anomalies in their operational behavior (such as a sudden increase in overseas SMS receipt records before changing a card). Therefore, in the technical solution of this application, the facial state features and historical behavioral semantic features are further subjected to user multimodal identity verification analysis based on the facial state-behavior semantic interaction structure to obtain facial state-behavior semantic significant interaction features. The user multimodal identity verification analysis process based on the facial state-behavior semantic interaction structure explicitly models the feature complementarity and independence between facial state semantics and historical behavioral semantics. First, principal component analysis (PCA) is performed on the facial state feature vector using principal component analysis. High-dimensional facial dynamic features (such as micro-expression frequency and pupil focus trajectory) are projected into a low-dimensional principal component space through orthogonal transformation, eliminating feature redundancy (such as the strong correlation between illumination changes and skin texture features) and extracting a more biometrically identifiable liveness feature subset. At the same time, an affine mapping is performed on the behavioral semantic association encoding vectors to address cross-modal scale heterogeneity (such as the dimensional difference between the numerical range of facial features and the frequency of behavioral operations), achieving preliminary alignment of the two types of features within a unified semantic space. Furthermore, a topologically ordered encoding matrix is ​​generated through inter-modal independence modeling units, enabling quantitative analysis of the potential correlation strength between dynamic facial features and behavioral patterns.For example, when a user passes liveness detection but exhibits anomalies in their operation path (e.g., initiating card replacement requests at dawn for three consecutive nights), the topological order matrix reveals phase accumulation anomalies between two types of features under the canonical potential through eigenvector analysis (e.g., the coupling conflict between liveness authentication parameters and high-frequency remote logins). This in turn triggers soft constraint factors to dynamically adjust interaction weights: suppressing redundant features during compliant operation periods under normal lighting conditions (e.g., the weak correlation between regular logins and facial features), and amplifying independent signals in high-risk scenarios (e.g., sudden changes in device fingerprints accompanied by abnormal pupil focus). This mechanism can capture risky patterns in mixed mobile phone numbers and cards that traditional single-modality verification cannot identify, such as covert attacks where fraudsters hijack legitimate users' devices to complete facial verification and then immediately trigger abnormal card replacement operations.

[0039] In an embodiment of the present application, the step S1331, performing affine mapping processing based on principal component analysis on the facial state feature vector to obtain a set of facial state principal component affine mapping feature coding vectors, includes: S1331-1, performing principal component analysis on the facial state feature vector to obtain a set of facial state principal component feature coding vectors; S1331-2, performing affine mapping on each facial state principal component feature coding vector in the set of facial state principal component feature coding vectors to obtain a set of facial state principal component affine mapping feature coding vectors, wherein each facial state principal component affine mapping feature coding vector in the set of facial state principal component affine mapping feature coding vectors has the same characteristic scale as the recent historical behavior semantic association coding vector.

[0040] Specifically, in step S1331-1, principal component analysis is performed on the facial state feature vector to obtain a set of facial state principal component feature coding vectors, which is expressed as follows using the facial state principal component analysis formula:

[0041]

[0042] in, is the face state feature vector, Principal component analysis, is the covariance matrix of the face state eigenvector, is the orthogonal matrix of principal components of face state features, , and are the first, second and third principal component feature encoding vectors of the face state. Individual face state feature principal component encoding vector, is the face state feature diagonal matrix, , and They are , and The corresponding eigenvalues, for The transposed matrix of is the number of eigenvalues ​​in the face state eigenvector, Represents the extraction of diagonal elements. It should be understood that facial state feature vectors usually contain a large amount of redundant visual information, such as differences in illumination reflection, random noise interference, and non-critical biometric parameters. These redundant data will increase the complexity of model training and may introduce false correlation features. Principal component analysis maps the original high-dimensional features to a low-dimensional principal component space through mathematical orthogonal transformation. In essence, it optimizes the information entropy of biometric data, screens out the most discriminative dynamic physiological features for identity verification (such as micron-level changes in iris texture and timing characteristics of facial muscle movements), and eliminates collinear interference caused by non-essential factors such as illumination gradients and shooting angle offsets. This feature dimensionality reduction is not a simple information compression, but rather constructs a topologically stable feature subspace while maintaining the main variation information of the original data, so that subsequent multimodal interaction analysis can focus on the key feature dimensions related to liveness detection.

[0043] Specifically, in step S1331-2, affine mapping is performed on each face state principal component feature coding vector in the set of face state principal component feature coding vectors to obtain a set of face state principal component affine mapping feature coding vectors, wherein each face state principal component affine mapping feature coding vector in the set of face state principal component affine mapping feature coding vectors has the same characteristic scale as the recent historical behavior semantic association coding vector, and is expressed as the face state principal component affine mapping formula:

[0044]

[0045] in, Express Perform affine mapping on the principal component feature encoding vectors of each face state in and are the first, second, and third in the set of face state principal component affine mapping feature encoding vectors. and Individual face state principal component affine mapping feature encoding vector, The affine mapping is the set of feature encoding vectors for the principal components of the facial states. It should be understood that the original set of principal component feature encoding vectors for facial states inherits the low-dimensional density characteristic of principal component parsing, but its numerical distribution is still limited by the hardware parameters of the biometric acquisition device (such as camera resolution and sampling frame rate). The historical behavioral semantic association encoding vector is derived from the modeling of user operation sequences in scenarios such as mobile payment and SMS verification over a long period of time, and its numerical scale is fundamentally different from that of biometric features. The affine mapping performs a nonlinear mapping of the principal component feature encoding vectors for facial states using learnable scaling parameters, essentially constructing a probability distribution alignment channel between the dynamic attributes of biometric features and user behavior patterns. This alignment mechanism not only resolves the imbalance in the contribution of data from different modalities during the gradient update process, but more importantly, through the normalization of the feature space, it enables the subsequent cross-modal attention mechanism to focus on complementary features rather than redundant information.

[0046] In an embodiment of the present application, the step S1332 performs modal decoupling constraint analysis on the set of facial state principal component affine mapping feature coding vectors and the recent historical behavior semantic association coding vector to determine a set of facial state-behavior semantic principal component modal decoupling flexible constraint factors, including: S1332-1, performs modal decoupling modeling analysis on each facial state principal component affine mapping feature coding vector in the set of recent historical behavior semantic association coding vectors and the facial state principal component affine mapping feature coding vector to obtain facial state-behavior semantic principal component modal decoupling flexible constraint factors. A set of semantic principal component modal decoupling coding matrices; S1332-2, performing canonical stable optimization on each face state-behavior semantic principal component modal decoupling coding matrix in the set of face state-behavior semantic principal component modal decoupling coding matrices to obtain a set of face state-behavior semantic principal component modal decoupling canonical optimized coding matrices; S1332-3, based on the set of face state-behavior semantic principal component modal decoupling canonical optimized coding matrices, calculating the set of face state-behavior semantic principal component modal decoupling flexible constraint factors.

[0047] Specifically, step S1332-1 performs intermodal decoupling modeling analysis on each face state principal component affine mapping feature coding vector in the set of the recent historical behavior semantic association coding vector and the face state principal component affine mapping feature coding vector to obtain a set of face state-behavior semantic principal component intermodal decoupling coding matrices, which is expressed as the face state-behavior semantic principal component intermodal decoupling modeling analysis formula:

[0048]

[0049] in, and is a feature mapping function, such as a linear mapping or a nonlinear kernel function, is the semantic association encoding vector of the recent historical behavior, For the The length of the feature encoding vector of the principal component affine mapping of the individual face state, The first one in the set of decoupling encoding matrices between face state and behavior semantic principal components The intermodal decoupling encoding matrix of the facial state-behavior semantic principal component. It should be understood that traditional modal fusion methods typically rely on implicitly learning feature interactions through neural networks. However, differences in biometric feature acquisition devices can lead to distributional shifts in the feature encoding vectors of the facial state principal component affine mapping, and the dynamic nature of user behavior data over time exacerbates intermodal interference. The intermodal decoupling modeling analysis in this step introduces a mutual information maximization constraint to construct a joint probability distribution model of bimodal features, where each intermodal decoupling encoding matrix of the facial state-behavior semantic principal component corresponds to an independent feature subspace of a specific dimension. This explicit modeling not only quantifies the statistical independence between the rate of facial state change and the sequence of action actions, but more importantly, through regularized training, forces the model to focus on complementary information that is mutually exclusive between the two feature types. For example, when detecting "SIM card theft + face verification" attacks, the decoupling encoding matrix can identify anomalies in the temporal correlation between the micro-expression features of a forged face and the user's historical transaction locations and timestamps. These complementary features are often overwhelmed by redundant information in traditional end-to-end fusion models.

[0050] Specifically, step S1332-2 performs a canonical stable optimization on each face state-behavior semantics principal component modal decoupling coding matrix in the set of face state-behavior semantics principal component modal decoupling coding matrices to obtain a set of face state-behavior semantics principal component modal decoupling canonical optimized coding matrices, which is expressed as a canonical stable optimization formula for face state-behavior semantics principal component modal decoupling:

[0051]

[0052]

[0053]

[0054] in, For the said The eigenvalues ​​of the decoupling encoding matrix between the main component modalities of the face state-behavior semantics, is the face state-behavior semantic eigenvector composed of the above eigenvalues, is matrix multiplication, is the face state-behavior semantic canonical association encoding vector, The first set of encoding matrices for the decoupling between the face state and behavior semantic principal components is optimized. The canonical optimization encoding matrix for the decoupling of the face state-behavior semantic principal component modalities. It should be understood that the canonical stability optimization process constructs a feature interaction framework with a global topological structure by combining the eigenvectors of the face state-behavior semantic principal component modal decoupling encoding matrix as a canonical potential representation in a topological order field. This representation method not only captures the short-range correlation between dynamic biometric noise (such as pixel-level fluctuations caused by rendering artifacts) and historical user behavior patterns (such as login location migration patterns), but more importantly, it establishes a long-range correlation mechanism for cross-modal features through the canonical potential-connection form. For example, when detecting "SIM card theft + face verification" attacks, by analyzing the eigenvalue distribution anomalies of the canonical optimization encoding matrix for the decoupling of the face state-behavior semantic principal component modalities after canonical optimization, we can identify the cross-scale coupling distortion between the attacker's generated virtual facial features in the high-frequency band (corresponding to micro-expression changes) and the user's real behavioral eigenfeatures in the low-frequency band (corresponding to accumulated operating habits). This unnatural cross-scale correlation manifests as a significant topological structure perturbation under the canonical potential representation. Specifically, in order to maintain structural robustness under asymmetric conditions and adapt to the subsequent elastic modal fusion requirements, the eigenvectors composed of the eigenvalues ​​of the decoupling encoding matrix between the main components of the facial state-behavior semantics modalities are used as the gauge potential representation in the topological order field, and a structural stability guarantee mechanism is constructed under asymmetric environments based on the principle of topological conservation. In this way, when the distribution of the facial state-behavior semantics eigenvectors composed of the eigenvalues ​​is far away from the limit state, a truncation strategy is adopted to discard the high-order trace components, and the first-order gauge potential term is constructed through the gauge potential-connection 1 paradigm. The face state-behavior semantic canonical association encoding vector Then, based on the phase accumulation characteristics in the canonical potential field, the face state-behavior semantic canonical association encoding vector The autocorrelation matrix of face state-behavior semantic principal component modality decoupling encoding matrix is ​​used as a constraint condition. By implementing topological structure optimization and adjusting the asymmetric correlation strength of the intrinsic space, the destructive effect of local disturbances on the global topological structure can be effectively suppressed, and finally a stabilized coding representation with robustness against asymmetric interference is generated.

[0055] Specifically, step S1332-3 calculates a set of decoupling flexible constraint factors between the face state and behavior semantic principal component modalities based on the decoupling specification optimization coding matrix between the face state and behavior semantic principal component modalities, and expresses the calculation formula of independent soft constraints between the face state and behavior semantic principal component modalities as follows:

[0056] ;

[0057] in, represents the square of the norm of the matrix F, represents a set of flexible constraint factors for the intermodal decoupling of the face state-behavior semantic principal components. It should be understood that the set of normatively optimized encoding matrices for the intermodal decoupling of the face state-behavior semantic principal components represents the optimal projection basis of bimodal features in the independence space. The essence of computing the set of flexible constraint factors for the intermodal decoupling of the face state-behavior semantic principal components is to establish a mapping bridge between modal independence modeling and feature interaction. Compared to traditional hard constraint methods (such as forced decorrelation regularization), the flexible constraint factors for the intermodal decoupling of the face state-behavior semantic principal components dynamically weight the normatively optimized encoding matrix for the intermodal decoupling of the face state-behavior semantic principal components through a learnable scaling factor. This approach preserves the interactive potential of key complementary information while suppressing the interference effect of redundant features. In this way, during normal user operations (such as reporting a card loss offline and then verifying it online), the flexible constraint factor for decoupling the facial state-behavior semantic principal component modalities reduces the suppression of inter-modal correlations, allowing the collaborative features of facial expression changes and device operation trajectories to participate in verification, thereby improving the user experience. However, in abnormal scenarios (such as an attacker using an avatar to initiate a card replacement request), the flexible constraint factor for decoupling the facial state-behavior semantic principal component modalities automatically strengthens the weight allocation for independent features, focusing on extracting statistical independence evidence between user physiological characteristics (such as changes in pupil diameter) and behavioral patterns (such as fingerprint touch force), thereby effectively distinguishing the behavioral differences between authentic users and digital impersonators. By limiting the calculation of the flexible constraint factor for decoupling the facial state-behavior semantic principal component modalities, multi-scale sensitivity conditions are met (such as assigning higher weight to high-frequency micro-expression changes), forming a scalable technical protection feature. Furthermore, the dynamic adaptability of the decoupling flexible constraint factor between the facial state and behavioral semantic principal components ensures that when a sudden change in variance is detected (potentially indicating a deepfake attack), it automatically strengthens the independence constraints on the behavioral semantic encoding features, forcing the model to focus on statistically independent feature dimensions in a user's historical behavior (such as the distribution of login locations within a specific time period). This hierarchical dynamic control mechanism not only enhances defenses against new fraud methods but also builds a technical architecture that is both robust and explainable.

[0058] In an embodiment of the present application, the step S1333, based on the set of decoupling flexible constraint factors between the face state-behavior semantics main component modalities, adaptively and dynamically aggregates the set of face state main component affine mapping feature coding vectors and the recent historical behavior semantics association coding vectors to obtain the face state-behavior semantics significant interaction coding vector as the face state-behavior semantics significant interaction feature, including: S1333-1, inputting each face state main component affine mapping feature coding vector in the set of the recent historical behavior semantics association coding vector and the face state main component affine mapping feature coding vector into the feature interaction response unit to obtain a set of fine-grained response interaction coding vectors between face state-behavior semantics main component modalities; S1333-2, based on the set of decoupling flexible constraint factors between the face state-behavior semantics main component modalities, dynamically aggregates the set of fine-grained response interaction coding vectors between the face state-behavior semantics main component modalities to obtain the face state-behavior semantics significant interaction coding vector.

[0059] Specifically, in step S1333-1, each face state principal component affine mapping feature coding vector in the set of the recent historical behavior semantic association coding vector and the face state principal component affine mapping feature coding vector is input into the feature interaction response unit to obtain a set of fine-grained response interaction coding vectors between face state and behavior semantic principal component modalities, which is expressed as the face state-behavior semantic feature interaction response formula:

[0060]

[0061] in, Indicates cascade processing, 、 and Represents positional multiplication, positional addition, and positional division. and represents the trainable weight matrix and the trainable bias vector, The first in the set of fine-grained response interaction encoding vectors representing the face state-behavior semantic principal component modalities Fine-grained interaction encoding vectors between the modalities of individual facial state and behavioral semantic principal components. It's understandable that the recent historical behavioral semantic association encoding vectors carry the characteristics of the user's long-established operational habits (e.g., the time distribution of SMS verification code receipt and login location migration patterns), while the collection of facial state principal component affine mapping feature encoding vectors reflects the real-time dynamic characteristics of biometric features (e.g., pupil constriction response and facial muscle micro-movements). Traditional modal fusion methods often rely solely on simple feature superposition, failing to effectively extract the temporal dependencies and complementary effects between the two types of features. The feature interaction response unit, through a nested design combining a multi-layer perceptron architecture and an attention mechanism, captures the cumulative effects of behavioral sequences and the instantaneous response differences of facial state changes in the temporal dimension. For example, when a user suddenly changes their frequently used operating device, the nonlinear deviation between the facial recognition confidence curve and the historical behavioral entropy value is observed. In the spatial dimension, it analyzes the spatial association between the spatial distribution patterns of biometric features (e.g., facial feature position shifts) and behavioral semantics (e.g., pressure distribution in the fingerprint recognition area). Furthermore, in the semantic dimension, it establishes association mappings at the abstract concept level, such as the implicit logical connection between "urgent report of loss" behavior and facial expressions of anxiety. This multi-level interaction mechanism not only improves the distinguishability of feature expression, but more importantly, builds the core perception layer of the dynamic defense system. When an attacker attempts to initiate a card replacement request through a virtual avatar, the abnormal fluctuations of the fine-grained response interaction encoding vector between the facial state-behavior semantic principal component modalities can be used to identify the statistical independence violation of facial feature dynamic noise (such as pixel-level jitter caused by rendering artifacts) and the user's historical behavior pattern.

[0062] Specifically, step S1333-2 dynamically aggregates the set of fine-grained response interaction coding vectors between the face state and behavior semantics principal component modalities based on the set of decoupling flexible constraint factors between the face state and behavior semantics principal component modalities to obtain the face state and behavior semantics significant interaction coding vector, which is expressed as the face state and behavior semantics dynamic aggregation formula:

[0063]

[0064] in, for function, is the number of vectors in the set of fine-grained response interaction encoding vectors between the face state-behavior semantic principal component modalities, The face state-behavior semantic significant interaction encoding vector. It should be understood that the dynamic aggregation process adaptively weights the fine-grained response interaction encoding features between the face state-behavior semantic principal component modalities through the decoupling flexible constraint factor between the face state-behavior semantic principal component modalities. Compared with traditional average pooling or fully connected fusion, dynamic aggregation dynamically focuses on the feature interaction dimensions that are key to fraud detection by constructing an attention mask based on the constraint factor, thereby effectively distinguishing between the accidental operational deviations of real users caused by environmental interference and the behavioral patterns deliberately imitated by attackers, so that the obtained face state-behavior semantic significant interaction encoding vector contains rich correlation information while minimizing redundancy and noise.

[0065] In an embodiment of the present application, step S134, performing mobile phone card risk level detection based on the facial state-behavior semantically significant interaction features to determine the risk level of a mobile phone card replacement / loss report application, includes: passing the facial state-behavior semantically significant interaction encoding vector through a classifier-based mobile phone card risk level detector to obtain a detection result, wherein the detection result is used to represent the risk level of the mobile phone card replacement / loss report application. In other words, the interaction encoding information between facial state features and historical behavioral semantic features is utilized to achieve precise stratification and dynamic response to risk levels. Specifically, the nonlinear mapping capability of the classifier can convert the cross-modal semantic information contained in the interaction encoding vector (such as the abnormal coupling between liveness physiological characteristics and operation spatiotemporal logic) into a quantifiable risk score. For example, if the facial state-behavior semantically significant interaction encoding vector detects a high-risk pattern of "frequent remote logins accompanied by abnormal pupil focus," the classifier will output a high risk level and trigger a security alert. Conversely, if the facial state-behavior semantically significant interaction encoding vector exhibits low-risk characteristics such as "liveness features pass and behavioral patterns conform to historical baselines," the classifier will output a low risk level to accelerate service processing. This hierarchical judgment mechanism not only improves the granularity of risk identification, but also dynamically adjusts the response strategy based on the risk level (such as telephone follow-up in medium-risk scenarios or direct interception in high-risk scenarios), thereby optimizing the user experience while ensuring security.

[0066] In summary, step S130 is explained by performing semantically significant interaction analysis based on feature domain on the face recognition image of the user to be verified and the recent historical behavior records to determine the risk level of the mobile phone card replacement / loss report application, so as to construct a risk prediction model from the dual dimensions of biometric authenticity and recent historical behavior patterns.

[0067] In the above-mentioned method for preventing mobile phone card fraud, step S140 executes a response operation based on the risk level of the mobile phone card replacement / loss report application, including: S141, in response to the risk level of the mobile phone card replacement / loss report application being medium or high, performing a secondary verification of the mobile phone card replacement / loss report application; S142, in response to the risk level of the mobile phone card replacement / loss report application being low, verifying the mobile phone card replacement / loss report application. It should be understood that the secondary verification mechanism is initiated when the risk level of the mobile phone card replacement / loss report application is assessed as medium or high risk. This measure is implemented in response to the increasingly sophisticated fraud methods currently used, particularly the use of highly realistic forged identity information and biometrics by fraudsters. Single-dimensional identity verification methods, such as SMS verification codes or manual customer service Q&A, are no longer able to meet the challenges of modern telecommunications fraud. Therefore, the introduction of secondary verification, especially when combined with multiple factors such as facial recognition and behavioral history, can effectively enhance verification accuracy. When medium- or high-risk applications are detected, secondary verification relies not only on the basic identity information provided by the user but also incorporates more real-time dynamic data, such as recent call logs, text message content, data usage, and login frequency, among other behavioral characteristics. This in-depth data analysis further confirms the applicant's identity. The core purpose of this secondary verification mechanism is to conduct a more rigorous review of suspicious applicants and prevent fraudulent attempts to tamper with user identity information through forgery. Through follow-up phone calls and manual review, the applicant's identity is verified in all possible aspects, thereby minimizing the risk of fraud. For medium- and high-risk applications, the intensity and methods of secondary verification may be more diverse and complex. In addition to basic identity information verification, the secondary verification may also be conducted based on abnormalities in behavioral data and historical user activity patterns, using artificial intelligence technology to analyze whether their operating habits are unusual. If medium- or high-risk applications show a strong correlation with historical fraudulent activity or attack patterns, a security alert mechanism will be automatically activated. This not only ensures a timely response to high-risk activity but also provides timely data support for subsequent tracking and fraud prevention.

[0068] In contrast, when the risk level for a mobile phone card replacement or report of loss is low, the operation can be completed through a more streamlined verification process. This is because low-risk applications generally indicate that there are no obvious abnormalities in the user's identity and behavior, and basic facial recognition and behavioral history data can effectively confirm the user's identity. Such applications are processed more quickly and are usually directly verified, avoiding unnecessary manual intervention, thereby improving work efficiency and reducing user waiting time. For example, if a user initiates a request on a commonly used device and the behavior pattern is consistent with historical behavior, the application can be automatically determined to be low-risk, and identity verification can be completed quickly. Direct verification in low-risk situations not only improves the user experience but also avoids business delays caused by excessive verification.

[0069] In summary, the method for preventing mobile phone card fraud based on the embodiments of the present application is described. Upon receiving a request for replacement or loss of a mobile phone card, the method sends an identity verification request to the user's device, triggering the real-time collection of the user's facial image and the automatic extraction of recent behavior records (such as login frequency, operating habits, call and text message usage, and data usage). After transmitting this data to the cloud platform, multimodal fusion analysis technology is used to perform deep correlation modeling between facial features and behavioral semantics, thereby dynamically generating risk level assessment results and responding accordingly. This approach leverages the cloud's intelligent analysis capabilities, while concealing the complexity of the underlying algorithms, achieving precise interception and proactive defense against anomalous application behavior, providing a foundation for building a trusted digital identity authentication system.

[0070] Figure 5 This is a system block diagram of a system for preventing mobile phone number card fraud according to an embodiment of the present application. Figure 5 As shown, according to an embodiment of the present application, the system 100 for preventing mobile phone number card fraud includes: a mobile phone number card change verification module 110, which is used to receive a mobile phone number card replacement / loss report application and send identity authentication request information to the device terminal of the user to be verified; a user information collection module 120 to be verified, which is used to collect the face recognition image and recent historical behavior records of the user to be verified in response to the device terminal receiving the identity authentication request information; a risk level assessment module 130, which is used to transmit the face recognition image and the recent historical behavior records to the cloud platform for data processing to analyze the risk level of the mobile phone number card replacement / loss report application, and is used to: on the cloud platform, perform semantic significant interaction analysis based on feature domain on the face recognition image and the recent historical behavior records of the user to be verified to determine the risk level of the mobile phone number card replacement / loss report application; a risk level response operation execution module 140, which is used to execute a response operation based on the risk level of the mobile phone number card replacement / loss report application.

[0071] Here, those skilled in the art will appreciate that the specific operations of each step in the above-mentioned system for preventing mobile phone number card fraud have been referenced above. Figures 1 to 4 The method for preventing mobile phone number card fraud has been introduced in detail in the description of the method, and therefore, its repeated description will be omitted.

[0072] In summary, the system for preventing mobile phone card fraud based on the embodiments of the present application is described. Upon receiving a request for replacement or loss of a mobile phone card, it sends an identity verification request to the user's device, triggering the real-time collection of the user's facial image and the automatic extraction of recent behavior records (such as login frequency, operating habits, call and text message usage, and data usage). After transmitting this data to the cloud platform, multimodal fusion analysis technology is used to perform deep correlation modeling between facial features and behavioral semantics, dynamically generating risk level assessment results and responding accordingly. This approach leverages the cloud's intelligent analysis capabilities, while concealing the complexity of the underlying algorithms, achieving precise interception and proactive defense against anomalous application behavior, providing a foundation for building a trusted digital identity authentication system.

Claims

1. A method for preventing mobile phone number card fraud, characterized in that: include: Receive applications for replacement / loss of mobile phone cards and send identity verification request information to the device terminal of the user to be verified; In response to the device terminal receiving the identity authentication request information, collecting the face recognition image and recent historical behavior records of the user to be authenticated; Transmitting the facial recognition image and the recent historical behavior records to a cloud platform for data processing to analyze the risk level of the mobile phone card replacement / loss report application, including: performing a semantically significant interaction analysis based on a feature domain on the facial recognition image of the user to be verified and the recent historical behavior records on the cloud platform to determine the risk level of the mobile phone card replacement / loss report application; The face recognition image of the user to be verified and the recent historical behavior record are subjected to semantically significant interaction analysis based on feature domains to determine the risk level of the mobile phone card replacement / loss report application, including: The face recognition image is passed through a face feature extractor based on Capsule Network to obtain a face state feature vector as a face state feature; Passing a common space domain mapping association encoder based on a fully connected layer on multiple behavior record data items in the recent historical behavior record to obtain a recent historical behavior semantic association encoding vector as a historical behavior semantic feature; Performing a user multimodal identity authentication analysis based on a face state-behavior semantic interaction structure on the face state feature and the historical behavior semantic feature to obtain a face state-behavior semantic significant interaction feature; Detecting the risk level of the mobile phone card based on the facial state-behavior semantic significant interaction feature, and determining the risk level of the mobile phone card replacement / loss report application; Execute response operations based on the risk level of the mobile phone number card replacement / loss report application.

2. The method for preventing mobile phone number card fraud according to claim 1, characterized in that: Performing a user multimodal identity authentication analysis based on the face state-behavior semantic interaction structure on the face state feature and the historical behavior semantic feature to obtain a face state-behavior semantic significant interaction feature, including: Performing affine mapping processing based on principal component analysis on the facial state feature vector to obtain a set of facial state principal component affine mapping feature coding vectors; Performing modal decoupling constraint analysis on the set of facial state principal component affine mapping feature encoding vectors and the recent historical behavior semantic association encoding vectors to determine a set of facial state-behavior semantic principal component modal decoupling flexible constraint factors; Based on the set of decoupling flexible constraint factors between the facial state-behavior semantic principal component modalities, the set of facial state principal component affine mapping feature coding vectors and the recent historical behavior semantic association coding vectors are adaptively dynamically aggregated to obtain the facial state-behavior semantic significant interaction coding vector as the facial state-behavior semantic significant interaction feature.

3. The method for preventing mobile phone number card fraud according to claim 2, characterized in that: Performing affine mapping processing based on principal component analysis on the facial state feature vector to obtain a set of facial state principal component affine mapping feature coding vectors, including: Performing principal component analysis on the facial state feature vector to obtain a set of facial state principal component feature coding vectors; Affine mapping is performed on each face state principal component feature coding vector in the set of face state principal component feature coding vectors to obtain a set of face state principal component affine mapping feature coding vectors, wherein each face state principal component affine mapping feature coding vector in the set of face state principal component affine mapping feature coding vectors has the same characteristic scale as the recent historical behavior semantic association coding vector.

4. The method for preventing mobile phone number card fraud according to claim 3, characterized in that: Performing modal decoupling constraint analysis on the set of facial state principal component affine mapping feature encoding vectors and the recent historical behavior semantic association encoding vectors to determine a set of facial state-behavior semantic principal component modal decoupling flexible constraint factors, including: Performing intermodal decoupling modeling analysis on each face state principal component affine mapping feature coding vector in the set of the recent historical behavior semantic association coding vector and the face state principal component affine mapping feature coding vector to obtain a set of face state-behavior semantic principal component intermodal decoupling coding matrices; Performing canonical stable optimization on each face state-behavior semantics principal component modal decoupling coding matrix in the set of face state-behavior semantics principal component modal decoupling coding matrices to obtain a set of face state-behavior semantics principal component modal decoupling canonical optimized coding matrices; Based on the set of facial state-behavior semantic principal component modal decoupling specification optimization coding matrices, a set of facial state-behavior semantic principal component modal decoupling flexibility constraint factors is calculated.

5. The method for preventing mobile phone number card fraud according to claim 4, characterized in that: Based on the set of decoupling flexible constraint factors between the face state and behavior semantic principal component modalities, adaptively and dynamically aggregating the set of face state principal component affine mapping feature encoding vectors and the recent historical behavior semantic association encoding vectors to obtain a face state and behavior semantic significant interaction encoding vector as the face state and behavior semantic significant interaction feature, including: Inputting each face state principal component affine mapping feature coding vector in the set of the recent historical behavior semantic association coding vector and the face state principal component affine mapping feature coding vector into a feature interaction response unit to obtain a set of fine-grained response interaction coding vectors between face state and behavior semantic principal component modalities; Based on the set of decoupling flexible constraint factors between the facial state-behavior semantic principal component modalities, the set of fine-grained response interaction coding vectors between the facial state-behavior semantic principal component modalities is dynamically aggregated to obtain the facial state-behavior semantic significant interaction coding vector.

6. The method for preventing mobile phone number card fraud according to claim 5, characterized in that: Based on the facial state-behavior semantic significant interaction feature, the mobile phone number card risk level is detected to determine the risk level of the mobile phone number card replacement / loss report application, including: passing the facial state-behavior semantic significant interaction encoding vector through a classifier-based mobile phone number card risk level detector to obtain a detection result, and the detection result is used to represent the risk level of the mobile phone number card replacement / loss report application.

7. The method for preventing mobile phone number card fraud according to claim 6, characterized in that: Based on the risk level of the mobile phone card replacement / loss report application, perform response operations, including: In response to the risk level of the mobile phone number card replacement / loss report application being medium risk or high risk, performing a secondary verification on the mobile phone number card replacement / loss report application; In response to the risk level of the mobile phone number card replacement / loss report application being low risk, the mobile phone number card replacement / loss report application is verified to be approved.

8. A system for preventing mobile phone number card fraud, characterized in that: include: The mobile phone card change verification module is used to receive mobile phone card replacement / loss report applications and send identity authentication request information to the device terminal of the user to be verified; A module for collecting information of a user to be verified, configured to collect a facial recognition image and recent historical behavior records of the user to be verified in response to the device terminal receiving the identity authentication request information; a risk level assessment module, configured to transmit the facial recognition image and the recent historical behavior records to a cloud platform for data processing to analyze the risk level of the mobile phone card replacement / loss report application, and to: perform a semantically significant interaction analysis based on a feature domain on the facial recognition image and the recent historical behavior records of the user to be verified on the cloud platform to determine the risk level of the mobile phone card replacement / loss report application; The risk level assessment module is used to: The face recognition image is passed through a face feature extractor based on Capsule Network to obtain a face state feature vector as a face state feature; Passing a common space domain mapping association encoder based on a fully connected layer on multiple behavior record data items in the recent historical behavior record to obtain a recent historical behavior semantic association encoding vector as a historical behavior semantic feature; Performing a user multimodal identity authentication analysis based on a face state-behavior semantic interaction structure on the face state feature and the historical behavior semantic feature to obtain a face state-behavior semantic significant interaction feature; Detecting the risk level of the mobile phone card based on the facial state-behavior semantic significant interaction feature, and determining the risk level of the mobile phone card replacement / loss report application; The risk level response operation execution module is used to execute a response operation based on the risk level of the mobile phone number card replacement / loss report application.

Images