Adaptive steganography watermark-based opponent perception authentication method in federal learning

Abstract

The invention discloses an opponent perception authentication method in federal learning based on an adaptive steganography watermark, and the method comprises the steps: constructing the identity information of a device, encrypting the identity information, and transmitting the encrypted identity information to an authentication server; the authentication server processes the request of the external equipment to ensure that each equipment added into the system has a controlled validity period and clear identity information; the equipment sends an application to the authentication server again according to the validity period, and the server cleans internal invalid equipment regularly; the authentication server registers device IDs for legal devices and groups the legal devices, selects group leader devices for each group and formulates rotation rules; the authentication server generates a corresponding watermark triggering data set for each group, and sends the watermark triggering data set together with the global model and the authentication data set to all legal devices; legal equipment trains a global model and learns a trigger strategy; and the authentication server performs security authentication on the legal equipment. According to the invention, the controllability, accuracy and concealment of the authentication process are improved, and safe, efficient and concealed authentication is realized.

Description

Technical Field

[0001] The present invention relates to network information security, distributed optimization and covert communication, and in particular to an adversary-aware authentication method in federated learning based on adaptive steganographic watermarking. Background Art

[0002] Security authentication is a core technology for ensuring data integrity and user privacy. It is typically used to verify the identity and permissions of users or devices and is also a major challenge in the field of network security. Identity authentication is a key step in security authentication. Numerous authentication schemes have been proposed, most of which use blockchain, keys, certificates, physical factors, and two-factor authentication to achieve identity verification. However, these methods all have limitations. Blockchain schemes are easily cracked by coordinated attackers, and verification data for key and certificate authentication methods is typically stored in plaintext. Once the physical factor authentication data is leaked, all related security authentication will be invalidated. Two-factor authentication usually relies on explicit identity credentials and cannot guarantee that the same user will always be correctly identified during the two authentication processes. In other words, neither authentication method is confidential.

[0003] Federated learning, a cutting-edge technology in data privacy and distributed optimization, leverages the core mechanism of centralized server coordination to integrate the computing resources and data advantages of distributed nodes for joint training of shared models. This innovative model supports multi-party collaborative modeling while effectively protecting user data privacy. Due to its unique characteristics, federated learning is particularly well-suited for large-scale security authentication scenarios.

[0004] Digital watermarking is an information security technology primarily used to identify the ownership and distribution of multimedia information on the internet. This technology must meet several key requirements, including concealment, robustness, embeddability, and time efficiency. It can also serve as a means of authentication. If the watermark can be extracted from a device and correctly verified, the device can be considered to be present in the authentication system.

[0005] Steganography is a technique used for covert communication, centered around concealing information. Within steganography, image steganography involves embedding secret information into an image without significantly altering its appearance. This technique is often referred to as high-capacity steganography. The simplest implementation is the least significant bit replacement method, which embeds the secret information in the least significant bit of each pixel without noticeably affecting the image's visual quality. There are also methods based on statistical models, such as Gaussian embedding models and statistical models of discrete cosine transform coefficients, that exploit the statistical properties of images to hide data in less easily detectable areas. In recent years, generative image steganography has garnered significant attention due to its security and batch processing capabilities. For example, image steganography can be implemented using generative adversary networks. Summary of the Invention

[0006] The purpose of the present invention is to solve the problem of the concealment of authentication features that are easily analyzed in the current network environment, to provide an adversary-aware authentication method in federated learning based on adaptive steganographic watermarking, and to provide a more reliable authentication mechanism to ensure the legitimacy of users and devices and prevent unauthorized access.

[0007] The technical solution adopted in the present invention is:

[0008] The adversary-aware authentication method in federated learning based on adaptive steganographic watermarking includes a training phase and an authentication phase. The specific steps are as follows:

[0009] Training phase:

[0010] Step 1: The device builds its own identity information A I And encrypted transmission to authentication server A s Among them, the identity information A of the legal device that has been approved by the system administrator to join the network environment I Includes a unique authentication identifier A id ;

[0011] Step 2: Authentication Server A s Process requests from external devices in the current network environment and ensure that each device that joins has a controlled validity period A p and clear identity information A I ;

[0012] Step 3: Authentication Server A s Register device IDs for legitimate devices and group them into groups;

[0013] Step 4: The authentication server generates a corresponding watermark trigger dataset for each group and allocates the authentication dataset required for subsequent authentication. It then sends the global model and authentication dataset together to all legitimate devices.

[0014] Step 5: Select a group leader device for each group based on the round-robin rule;

[0015] Step 6: The legitimate device performs federated training to train the global model and learn the triggering strategy.

[0016] Certification stage:

[0017] Step 7: The authentication server performs security authentication on the legitimate device.

[0018] Furthermore, the specific steps of step 1 are:

[0019] Step 1-1: First, classify the devices and assign a unique authentication identifier A to the legitimate devices that have been approved by the system administrator to join the network environment. id(Illegal devices that attempt to access the network without the administrator's confirmation will not have A id );

[0020] Step 1-2: When the device joins the network environment, it sends a s Send a join request; authentication server A s After receiving the request, the device is required to build and upload complete identity information A I , identity information A I Including the manufacturing information of the equipment, the validity period of the identity information p , assigned authentication identifier A id and associated attributes;

[0021] Steps 1-3, Identity Information A I Use authentication server A before transmission s The public key is used to encrypt, and the encryption process is as follows: ; Where M represents the encrypted information, Indicates using authentication server A s Encrypted with the public key.

[0022] Furthermore, the specific method of step 2 is: decrypt and verify the identity information A submitted by the device I Is it valid (the information field is complete and A id Valid); If yes, set the device's authentication status to valid in the authentication system and configure the validity period of the device information A p ; Validity period of device information A p It cannot be indefinite, and a threshold can be set to limit the tasks performed each time the network is executed. p Otherwise, the verification fails (i.e. the submitted A I Missing required fields or missing valid A id ), the system will correspond to the identity information A I Discard or block, and blacklist the corresponding device to prohibit it from accessing the network environment.

[0023] Furthermore, in step 2, the device re-submits an application to the authentication server based on the validity period. The server periodically clears out invalid devices within the server. The specific method is as follows:

[0024] When the validity period of the device information is A p About to expire, authentication server A s Warn the corresponding device; if the device wishes to continue to exist in the environment, resend the request before the expiration date;

[0025] When the validity period of the device information is A p Expired, authentication server A sSet the authentication status of the corresponding device to invalid and add it to the dormant device list;

[0026] When a device cannot automatically exit the network (due to deactivation or forgetting, etc.), it will exist for a long time and occupy resources. s During the validity period of the corresponding device A p Clean up regularly afterward.

[0027] Furthermore, the specific steps of step 3 are:

[0028] Step 3-1, Authentication Server A s Get the authentication status of the device;

[0029] Step 3-2, Authentication Server A s Each legal device is grouped and assigned a unique device ID. The ID assigned to each device consists of two parts. The first part is the group number G num , the second part is a random number of variable length;

[0030] Furthermore, in step 3, the authentication server ensures that the number of members in the device groups is balanced based on the grouping probability formula. The grouping probability formula is:

[0031] , ,where

[0032] in, and All are positive numbers to avoid division by zero; Indicates the group number of the i-th group; Indicates the group number of the j-th group; Indicates group number The number of members, Indicates group number the number of members; Indicates a data set without watermark triggering, Indicates the number of data types, n indicates the number of groups, device Indicates the device in the i-th group.

[0033] Specifically, the devices are grouped based on two dimensions: the number of categories in the dataset and the number of members currently assigned to each group, in order to ensure a balanced distribution of the number of members in each group.

[0034] Furthermore, the specific steps of step 4 are:

[0035] Step 4-1: assign an independent watermark trigger dataset to each group for subsequent training.

[0036] The datasets selected for each group include digital image datasets and non-digital image datasets. The process of constructing the watermark trigger dataset is to select the dataset with the group number G from the digital image dataset. num Corresponding image samples , and embed the image samples into the non-digital image samples corresponding to the group classification in sequence through image steganography. The embedding positions are non-continuous. The calculation formula is as follows:

[0037]

[0038]

[0039] in, is the watermark embedding position, is the interval between watermark embedding locations, It is to calculate the new watermark embedding position, U is an integer set from 1 to 5, Indicates the watermark embedding function, Indicates the watermark embedding position of the non-watermark trigger dataset data, Indicates group number G num The corresponding digital image dataset Number data, Indicates group number G num The number of samples in the corresponding digital image dataset;

[0040] In step 4-2, the global model, watermark trigger dataset, and authentication dataset are distributed to all legitimate devices.

[0041] Furthermore, in step 5, the authentication server A s A device is randomly selected in each group as the group leader device. When a device has served as the group leader device in a certain task, the priority of the corresponding device to be selected as the group leader device in the next task execution is reduced. The rotation rule is: the system regroups all devices at a random time point and avoids reassigning the same group of devices to the original group. After the group adjustment is completed, the auxiliary device ID, device identifier and authentication data set in the group are regenerated and reallocated.

[0042] Furthermore, the specific steps of step 6 are:

[0043] Step 6-1: The legitimate device receives the authentication server A s After the content is sent, it enters the training phase, where the watermark trigger dataset is locally divided into two parts, including watermarked data and non-watermarked data, both of which are used for federated learning training;

[0044] Step 6-2: The training is divided into two phases. In the first phase, the device uses each set of local watermark trigger data sets to train the local model and upload the local model update to the authentication server A. s Participate in global model aggregation;

[0045] Step 6-3, in the second stage, the device extracts the watermark from the designated location of the watermarked data and identifies it; when the watermark is correctly identified, the device modifies the label at the designated location of the watermarked data to its own authentication identifier A id .

[0046] Furthermore, the specific steps of step 7 are:

[0047] Step 7-1, Authentication Server A s Randomly select an image and embed it into the corresponding specified index; the specified index is a set of digital sequences, indicating the authentication server A s The selected image combination can form a special meaning; the image with the specified index is sent to the authentication server A s The generated digital signature is combined to form the authentication data A d , where the digital signature is the authentication server A s A set of ciphertexts generated, authenticated by digital signature data A d Is it from authentication server A? s , the final authentication data A d Send to group leader device G leader ;

[0048] Step 7-2, group leader device G leader Receive authentication data A d After that, verify whether the data source of the digital signature comes from the authentication server A s If yes, the verification is passed and step 7-3 is executed; otherwise, the data source is determined to be unreliable, the authentication data is discarded and the process is stopped, and the group leader device G leader The generated response A r Then upload it to the authentication server A s And end the current authentication;

[0049] Step 7-3, group leader device G leader Authentication data A d Forward it to group members, who extract the embedding index from the image;

[0050] Step 7-4: Group members obtain the response A generated by the corresponding data from the authentication dataset according to the index r Then upload to authentication server A s , response A r Contains the authentication identifier A id, regular image inference results, timestamp, and the value obtained by encrypting the timestamp with the device ID.

[0051] This invention utilizes the above technical solution, which effectively ensures secure device authentication in a network environment. By combining federated learning with a watermark triggering mechanism, this solution implements access control, identity authentication, and continuous status monitoring for devices in the network. The solution comprises a training phase and an authentication phase, with the training phase being subdivided into three subphases: device joining request processing, authentication server group configuration, and legitimate device training.

[0052] The present invention involves three types of participants: an authentication server, regular devices, and a group leader device. Specifically, devices are divided into several groups, each designated by a group leader device. The authentication server is responsible for authenticating all devices, while the group leader device verifies the authenticity of the authentication data source and distributes the authentication data to other members of the group. Furthermore, different access rights can be assigned to each group based on grouping policies, thus establishing a complete security authentication mechanism.

[0053] The present invention employs the above technical solutions, resulting in the following beneficial effects: 1. It innovates the federated learning authentication paradigm. By seamlessly integrating adaptive steganographic watermarking technology with federated learning, this framework creates a new paradigm for covert and robust device authentication, effectively addressing inherent flaws in current network security frameworks, such as insufficient authentication concealment and vulnerability to attacks. 2. It uses dynamic watermarking to achieve tamper-proof authentication. This innovative mechanism seamlessly integrates explicit interaction with implicit authentication, concealing traces of the authentication process while ensuring verifiable trust, significantly improving the system's security posture. 3. It implements implicit authentication based on explicit federated interaction. This strategy not only eliminates detectable authentication patterns but also maintains system stability, providing users with a seamless and secure authentication experience without sacrificing performance. 4. It uses a rigorous verification mechanism to achieve a perfect balance between efficiency and security. Compared to existing leading solutions, this solution reduces computational overhead by over 50% and communication costs by over 30%. Furthermore, rigorous formal security proofs and comprehensive informal analysis confirm the security features of LiveMark, ensuring that it meets the highest security standards and reliability requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments;

[0055] Figure 1 Schematic diagram of the process of implementing adversary-aware authentication in federated learning based on the adaptive steganographic watermarking technology of the present invention;

[0056] Figure 2 This is a structural diagram of the invention's implementation of adversary-aware authentication in federated learning based on adaptive steganographic watermarking technology. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0058] In modern networks, various network applications and services make them more complex than traditional networks. Basic network security protection alone is insufficient to completely prevent malicious attacks, especially in the presence of internal attackers. These attackers not only possess the user's device registration information but also the security parameters stored on the server, making it easier to infiltrate the environment and launch attacks from within. Therefore, more reliable mechanisms are needed to ensure the legitimacy of user devices and prevent unauthorized access.

[0059] This paper proposes a security authentication scheme called LiveMark, the main framework of which is as follows: Figure 2 As shown in Figure 1, it consists of a training phase and an authentication phase. The training phase is subdivided into three subphases: device joining request processing, authentication server group configuration, and legitimate device training. In the first training phase, a new device initiates a joining request to the authentication server. Upon approval, the authentication server assigns the device an authentication identifier and constructs a complete identity based on the identifier. This information is encrypted using the server's public key for transmission and storage. In the next phase, the authentication server groups all devices and assigns device IDs. It then randomly selects a device as a group leader for each group and distributes the global model, watermark trigger dataset, and authentication dataset to all legitimate devices. In the third training phase, devices first train locally using the watermark trigger dataset and upload the resulting federated learning parameters to the authentication server to participate in global model aggregation. Devices then attempt to recognize the watermarked data. If successful, they modify the data label to their own authentication identifier. In the final authentication phase, the server selects an image from the watermark trigger dataset and embeds a specified index. The image is then digitally signed, generating authentication data that is sent to the group leader devices in each group. Upon receiving the authentication data, the group leader device verifies the digital signature to ensure the authenticity of the message. If the source is reliable, it is forwarded to the group members. Each member extracts the embedded index from the image and then retrieves the corresponding data from the authentication dataset based on this index. The resulting response is then uploaded to the server and contains the authentication identifier, the conventional image inference result, a timestamp, and a value obtained by encrypting the timestamp with the device ID. If the group leader device determines that the data source is unreliable during the verification process, it directly discards the authentication data and does not proceed further. The group leader device then uploads the generated response to the server.

[0060] like Figure 1As shown in FIG2 , the present invention discloses an adversary-aware authentication method in federated learning based on adaptive steganographic watermarking, which includes a training phase and an authentication phase. The specific steps are as follows:

[0061] Training phase:

[0062] Step 1: The device builds its own identity information A I And encrypted transmission to authentication server A s Among them, the identity information A of the legal device that has been approved by the system administrator to join the network environment I Includes a unique authentication identifier A id ;

[0063] Step 2: Authentication Server A s Process requests from external devices in the current network environment and ensure that each device that joins has a controlled validity period A p and clear identity information A I ;

[0064] Step 3: Authentication Server A s Register device IDs for legitimate devices and group them into groups;

[0065] Step 4: The authentication server generates a corresponding watermark trigger dataset for each group and allocates the authentication dataset required for subsequent authentication. It then sends the global model and authentication dataset together to all legitimate devices.

[0066] Step 5: Select a group leader device for each group based on the round-robin rule;

[0067] Step 6: The legitimate device performs federated training to train the global model and learn the triggering strategy.

[0068] Certification stage:

[0069] Step 7: The authentication server performs security authentication on the legitimate device.

[0070] Furthermore, the specific steps of step 1 are:

[0071] Step 1-1: First, classify the devices and assign a unique authentication identifier A to the legitimate devices that have been approved by the system administrator to join the network environment. id (Illegal devices that attempt to access the network without the administrator's confirmation will not have A id );

[0072] Step 1-2: When the device joins the network environment, it sends a s Send a join request; authentication server A s After receiving the request, the device is required to build and upload complete identity information A I, identity information A I Including the manufacturing information of the equipment, the validity period of the identity information p , assigned authentication identifier A id and associated attributes;

[0073] Steps 1-3, Identity Information A I Use authentication server A before transmission s The public key is used to encrypt, and the encryption process is as follows: ; Where M represents the encrypted information, Indicates using authentication server A s Encrypted with the public key.

[0074] Furthermore, the specific method of step 2 is: decrypt and verify the identity information A submitted by the device I Is it valid (the information field is complete and A id Valid); If yes, set the device's authentication status to valid in the authentication system and configure the validity period of the device information A p ; Validity period of device information A p It cannot be indefinite, and a threshold can be set to limit the tasks performed each time the network is executed. p Otherwise, the verification fails (i.e. the submitted A I Missing required fields or missing valid A id ), the system will correspond to the identity information A I Discard or block, and blacklist the corresponding device to prohibit it from accessing the network environment.

[0075] Furthermore, in step 2, the device re-submits an application to the authentication server based on the validity period. The server periodically clears out invalid devices within the server. The specific method is as follows:

[0076] When the validity period of the device information is A p About to expire, authentication server A s Warn the corresponding device; if the device wishes to continue to exist in the environment, resend the request before the expiration date;

[0077] When the validity period of the device information is A p Expired, authentication server A s Set the authentication status of the corresponding device to invalid and add it to the dormant device list;

[0078] When a device cannot automatically exit the network (due to deactivation or forgetting, etc.), it will exist for a long time and occupy resources. s During the validity period of the corresponding device A p Clean up regularly afterward.

[0079] Furthermore, the specific steps of step 3 are:

[0080] Step 3-1, Authentication Server A s Get the authentication status of the device;

[0081] Step 3-2, Authentication Server A s Each legal device is grouped and assigned a unique device ID. The ID assigned to each device consists of two parts. The first part is the group number G num , the second part is a random number of variable length;

[0082] Furthermore, in step 3, the authentication server ensures that the number of members in the device groups is balanced based on the grouping probability formula. The grouping probability formula is:

[0083] , ,where

[0084] in, and All are positive numbers to avoid division by zero; Indicates the group number of the i-th group; Indicates the group number of the j-th group; Indicates group number The number of members, Indicates group number the number of members; Indicates a data set without watermark triggering, Indicates the number of data types, n indicates the number of groups, device Indicates the device in the i-th group.

[0085] Specifically, the devices are grouped based on two dimensions: the number of categories in the dataset and the number of members currently assigned to each group, in order to ensure a balanced distribution of the number of members in each group.

[0086] Furthermore, the specific steps of step 4 are:

[0087] Step 4-1: assign an independent watermark trigger dataset to each group for subsequent training. The dataset selected by each group includes a digital image dataset and a non-digital image dataset. The process of constructing the watermark trigger dataset is to select the dataset with the group number G from the digital image dataset. num Corresponding image samples , and embed the image samples into the non-digital image samples corresponding to the group classification in sequence through image steganography. The embedding positions are non-continuous. The calculation formula is as follows:

[0088] ;

[0089] ;

[0090] in, is the watermark embedding position, is the interval between watermark embedding locations, Indicates the calculation of the new watermark embedding position, U is an integer set from 1 to 5, Indicates the watermark embedding function, Indicates the watermark embedding position of the non-watermark trigger dataset data, Indicates group number G num The corresponding digital image dataset Number data, Indicates group number G num The number of samples in the corresponding digital image dataset;

[0091] In step 4-2, the global model, watermark trigger dataset, and authentication dataset are distributed to all legitimate devices.

[0092] Furthermore, in step 5, the authentication server A s A device is randomly selected in each group as the group leader device. When a device has served as the group leader device in a certain task, the priority of the corresponding device to be selected as the group leader device in the next task execution is reduced. The rotation rule is: the system regroups all devices at a random time point and avoids reassigning the same group of devices to the original group. After the group adjustment is completed, the auxiliary device ID, device identifier and authentication data set in the group are regenerated and reallocated.

[0093] Furthermore, the specific steps of step 6 are:

[0094] Step 6-1: The legitimate device receives the authentication server A s After the content is sent, it enters the training phase, where the watermark trigger dataset is locally divided into two parts, including watermarked data and non-watermarked data, both of which are used for federated learning training;

[0095] Step 6-2: The training is divided into two phases. In the first phase, the device uses each set of local watermark trigger data sets to train the local model and upload the local model update to the authentication server A. s Participate in global model aggregation;

[0096] Step 6-3, in the second stage, the device extracts the watermark from the designated location of the watermarked data and identifies it; when the watermark is correctly identified, the device modifies the label at the designated location of the watermarked data to its own authentication identifier A id .

[0097] Furthermore, the specific steps of step 7 are:

[0098] Step 7-1, Authentication Server A s Randomly select an image and embed it into the corresponding specified index; the specified index is a set of digital sequences, indicating the authentication server A s The selected image combination can form a special meaning; the image with the specified index is sent to the authentication server A s The generated digital signature is combined to form the authentication data A d , where the digital signature is the authentication server A s A set of ciphertexts generated, authenticated by digital signature data A d Is it from authentication server A? s , the final authentication data A d Send to group leader device G leader ;

[0099] Step 7-2, group leader device G leader Receive authentication data A d After that, verify whether the data source of the digital signature comes from the authentication server A s If yes, the verification is passed and step 7-3 is executed; otherwise, the data source is determined to be unreliable, the authentication data is discarded and the process is stopped, and the group leader device G leader The generated response A r Then upload it to the authentication server A s And end the current authentication;

[0100] Step 7-3, group leader device G leader Authentication data A d Forward it to group members, who extract the embedding index from the image;

[0101] Step 7-4: Group members obtain the response A generated by the corresponding data from the authentication dataset according to the index r Then upload to authentication server A s , response A r Contains the authentication identifier A id , regular image inference results, timestamp, and the value obtained by encrypting the timestamp with the device ID.

[0102] This paper proposes a security authentication scheme called LiveMark, which effectively solves the problem of the concealment of authentication features that are easily analyzed in the current network environment through federated learning and steganographic watermarking technology.

[0103] In this invention, watermark information is embedded in designated authentication trigger data and used as a trigger feature. Based on this trigger feature, the device to be authenticated generates a unique authentication identifier and hides it within the conventional inference results, making the output appear normal. Furthermore, this scheme incorporates group management, leader coordination, and information rotation mechanisms to ensure authentication security while minimizing the correlation between authentication results and devices. These mechanisms also enhance the controllability and confidentiality of the authentication process. Experimental results demonstrate that the proposed scheme effectively achieves secure, efficient, and confidential authentication.

[0104] This invention utilizes the above technical solution, which effectively ensures secure device authentication in a network environment. By combining federated learning with a watermark triggering mechanism, this solution implements access control, identity authentication, and continuous status monitoring for devices in the network. The solution comprises a training phase and an authentication phase, with the training phase being subdivided into three subphases: device joining request processing, authentication server group configuration, and legitimate device training.

[0105] The present invention involves three types of participants: an authentication server, regular devices, and a group leader device. Specifically, devices are divided into several groups, each designated by a group leader device. The authentication server is responsible for authenticating all devices, while the group leader device verifies the authenticity of the authentication data source and distributes the authentication data to other members of the group. Furthermore, different access rights can be assigned to each group based on grouping policies, thus establishing a complete security authentication mechanism.

[0106] The beneficial effects of the present invention are: 1. Innovative federated learning authentication paradigm. By seamlessly integrating adaptive steganographic watermarking technology with federated learning, the framework has created a new paradigm for covert and robust device authentication, effectively solving the inherent defects of current network security frameworks, such as insufficient authentication concealment and vulnerability to attacks. 2. Dynamic watermarking is used to achieve tamper-proof authentication. This innovative mechanism achieves "seamless connection between explicit interaction and implicit authentication", which not only hides the traces of the authentication process, but also ensures verifiable trust, thereby significantly improving the security posture of the system. 3. Implicit authentication based on explicit federated interaction is achieved. This strategy not only eliminates detectable authentication modes, but also maintains the stability of the system, providing users with a seamless and secure authentication experience without sacrificing performance. 4. A strict verification mechanism is used to achieve a perfect balance between efficiency and security. Compared with existing top solutions, the computational overhead of this solution is reduced by more than 50% and the communication cost is reduced by more than 30%. In addition, rigorous formal security proofs and comprehensive informal analysis also confirm the security features of LiveMark, ensuring that it meets the highest security standards and reliability requirements.

[0107] Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. In the absence of conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of the present application is not intended to limit the scope of the application for protection, but merely represents the selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

Claims

1. An adversary-aware authentication method for federated learning based on adaptive steganographic watermarking, characterized by: It includes the training phase and the certification phase. The specific steps are as follows: Training phase: Step 1: The device builds its own identity information A I And encrypted transmission to authentication server A s Among them, the identity information A of the legal device that has been approved by the system administrator to join the network environment I Includes a unique authentication identifier A id ; Step 2: Authentication Server A s Process requests from external devices in the current network environment and ensure that each device that joins has a controlled validity period A p and clear identity information A I ; Step 3: Authentication Server A s Register device IDs for legitimate devices and group them into groups; Step 4: The authentication server generates a corresponding watermark trigger dataset for each group and allocates the authentication dataset required for subsequent authentication. It then sends the global model and authentication dataset together to all legitimate devices. Step 5: Select a group leader device for each group based on the round-robin rule; Step 6: The legitimate device performs federated training to train the global model and learn the triggering strategy. Certification stage: Step 7: The authentication server performs security authentication on the legitimate device.

2. The adversary-aware authentication method in federated learning based on adaptive steganographic watermarking according to claim 1, characterized in that: The specific steps of step 1 are: Step 1-1: First, classify the devices and assign a unique authentication identifier A to the legitimate devices that have been approved by the system administrator to join the network environment. id ; Step 1-2: When the device joins the network environment, it sends a s Send a join request; authentication server A s After receiving the request, the device is required to build and upload complete identity information A I , identity information A I Including the manufacturing information of the equipment, the validity period of the identity information p , assigned authentication identifier A id and associated attributes; Steps 1-3, Identity Information A I Use authentication server A before transmission s The public key is used to encrypt, and the encryption process is as follows: ; Where M represents the encrypted information, Indicates using authentication server A s Encrypted with the public key.

3. The adversary-aware authentication method in federated learning based on adaptive steganographic watermarking according to claim 1, characterized in that: The specific method of step 2 is: decrypt and verify the identity information A submitted by the device I Is it valid? If so, set the device's authentication status to valid in the authentication system and configure the validity period of the device information. p Otherwise, the verification fails and the system will correspond to the identity information A I Discard or block, and blacklist the corresponding device to prohibit it from accessing the network environment.

4. The adversary-aware authentication method in federated learning based on adaptive steganographic watermarking according to claim 1 or 3, characterized in that: In step 2, the device re-submits an application to the authentication server based on the validity period. The server periodically clears invalid devices from the server. The specific method is as follows: When the validity period of the device information is A p About to expire, authentication server A s Warn the corresponding device; if the device wishes to continue to exist in the environment, resend the request before the expiration date; When the validity period of the device information is A p Expired, authentication server A s Set the authentication status of the corresponding device to invalid and add it to the dormant device list; When the device cannot actively exit the network and remains in the network for a long time and occupies resources, the authentication server A s During the validity period of the corresponding device A p Clean up regularly afterward.

5. The adversary-aware authentication method in federated learning based on adaptive steganographic watermarking according to claim 1, characterized in that: The specific steps of step 3 are: Step 3-1, Authentication Server A s Get the authentication status of the device; Step 3-2, Authentication Server A s Each legal device is grouped and assigned a unique device ID. The ID assigned to each device consists of two parts. The first part is the group number G num , the second part is a random number of variable length.

6. The adversary-aware authentication method in federated learning based on adaptive steganographic watermarking according to claim 1 or 5, characterized in that: In step 3, the authentication server ensures a balanced number of members in the device groups based on the grouping probability formula. The grouping probability formula is: , ,where in, and All are positive numbers to avoid division by zero; Indicates the group number of the i-th group; Indicates the group number of the j-th group; Indicates group number The number of members, Indicates group number the number of members; Indicates a data set without watermark triggering, Indicates the number of data types, n indicates the number of groups, device Indicates the devices belonging to the i-th group.

7. The adversary-aware authentication method in federated learning based on adaptive steganographic watermarking according to claim 1, characterized in that: The specific steps of step 4 are: Step 4-1: assign an independent watermark triggering dataset to each group for subsequent training; the dataset selected by each group includes a digital image dataset and a non-digital image dataset, and the image samples in the digital image dataset are converted into The watermark trigger dataset is formed by sequentially embedding it into non-digital image samples corresponding to the group classification. The embedding position is non-continuous. The embedding expression is as follows: ； ； in, is the watermark embedding position, is the interval between watermark embedding locations, It is to calculate the new watermark embedding position, U is an integer set from 1 to 5, Indicates the watermark embedding function, Indicates the watermark embedding position of the non-watermark trigger dataset data, Indicates group number G num The corresponding digital image dataset Number data, Indicates group number G num The number of samples in the corresponding digital image dataset; In step 4-2, the global model, watermark trigger dataset, and authentication dataset are distributed to all legitimate devices.

8. The adversary-aware authentication method in federated learning based on adaptive steganographic watermarking according to claim 1, characterized in that: Authentication Server A in step 5 s A device is randomly selected in each group as the group leader device. If a device has served as the group leader device in a task, its priority in being selected as the group leader device in the next task will be lowered. The rotation rule is: the system regroups all devices at a random time point and avoids reallocating the same group of devices to the original group. After the group adjustment is completed, the auxiliary device ID, device identifier and authentication data set in the group are regenerated and reallocated.

9. The adversary-aware authentication method in federated learning based on adaptive steganographic watermarking according to claim 1, characterized in that: The specific steps of step 6 are: Step 6-1: The legitimate device receives the authentication server A s After the content is sent, it enters the training phase, where the watermark trigger dataset is locally divided into two parts, including watermarked data and non-watermarked data, both of which are used for federated learning training; Step 6-2: The training is divided into two phases. In the first phase, the device uses each set of local watermark trigger data sets to train the local model and upload the local model update to the authentication server A. s Participate in global model aggregation; Step 6-3, in the second stage, the device extracts the watermark from the designated location of the watermarked data and identifies it; when the watermark is correctly identified, the device modifies the label at the designated location of the watermarked data to its own authentication identifier A id .

10. The adversary-aware authentication method in federated learning based on adaptive steganographic watermarking according to claim 1, characterized in that: The specific steps of step 7 are: Step 7-1, Authentication Server A s Randomly select an image and embed it into the corresponding specified index; the specified index is a set of digital sequences, indicating the authentication server A s The selected image combination can form a special meaning; the image with the specified index is sent to the authentication server A s The generated digital signature is combined to form the authentication data A d , where the digital signature is the authentication server A s A set of ciphertexts generated, authenticated by digital signature data A d Is it from authentication server A? s , the final authentication data A d Send to group leader device G leader ; Step 7-2, group leader device G leader Receive authentication data A d After that, verify whether the data source of the digital signature comes from the authentication server A s If yes, the verification is passed and step 7-3 is executed; otherwise, the data source is determined to be unreliable, the authentication data is discarded and the process is stopped, and the group leader device G leader The generated response A r Then upload it to the authentication server A s And end the current authentication; Step 7-3, group leader device G leader Authentication data A d Forward it to group members, who extract the embedding index from the image; Step 7-4: Group members obtain the response A generated by the corresponding data from the authentication dataset according to the index r Then upload to authentication server A s , response A r Contains the authentication identifier A id , regular image inference results, timestamp, and the value obtained by encrypting the timestamp with the device ID.

Images