A security verification method and system based on trusted digital identity and portrait comparison

Abstract

The present invention is a security verification method and system based on trusted digital identity and portrait comparison. By acquiring and storing the user's trusted digital identity, identity information and portrait data, when the user verifies on site, the data is collected through a QR code reading device and a face camera, and transmitted to an application system in a secure isolation network for decryption and query. A portrait feature extraction model is constructed and trained, deployed in a secure isolation network, and the stored and on-site collected portrait data are compared to determine whether it is a real person verification. The present invention reduces the risk of data leakage by storing and processing data through an application system in a secure isolation network; protects the user's privacy by minimizing the storage and transmission of the user's biometric data; uses advanced encryption technology and AI algorithms to ensure data security and verification accuracy; simplifies user operation steps, and the user only needs to present a QR code and perform a face comparison, which improves the convenience of verification.

Description

Technical Field

[0001] The present invention relates to the field of information security technology, and mainly to a security verification method and system based on trusted digital identity and portrait comparison. Background Art

[0002] In the field of modern information security, especially in identity authentication scenarios that require high security, such as financial transactions, security access control, e-government, etc., traditional identity authentication methods (such as passwords and PIN codes) face the risk of being easily stolen and cracked. In order to improve the security and accuracy of identity authentication while meeting the needs of convenience, an identity authentication method with higher confidentiality and security is urgently needed.

[0003] For example, a Chinese patent with publication number CN115051812A discloses a user identity dual identification method based on QR code and biometrics, and a user identity dual verification system that uses the identification method. The identification method includes two main processes: a registration phase and an authentication phase. Among them, in the registration phase, the administrator requests the user to upload account information, user information and the user's biometric identification information, and sends a key pair to the user. Among them, the biometric identification information includes a face image. In the authentication phase, the administrator first obtains the user's face. After the face matches successfully, the administrator and the user implement digital signature verification through the QR code. After the digital signature verification is successful, the user identity authentication is completed. Based on the identity authentication logic of this method, the present invention also designs a verification system dedicated to identity identification. The technical solution of the present invention solves the problem of insufficient confidentiality and security of existing digital account identity authentication methods. However, the biometric data (such as face, fingerprint, iris, etc.) uploaded by users in this patent involves personal privacy. These data cannot be protected during the verification process, and users need to perform multiple operations (such as decoding QR codes, generating digital signatures, etc.), which increases the complexity of user use; especially in scenarios that require rapid response, the multi-step verification process will affect the user experience. Therefore, there is an urgent need for a more secure, convenient and privacy-protecting identity authentication method. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, the present application provides a security verification method based on trusted digital identity and portrait comparison.

[0005] The technical solution of this application is as follows:

[0006] A security verification method based on trusted digital identity and portrait comparison, the method comprising:

[0007] Obtain the user's trusted digital identity, identity information and portrait data, and associate them with the corresponding database of the application system stored in the secure isolation network; when the user performs on-site identity verification, the trusted digital identity QR code presented by the user is read by the QR code reading device, and the user's portrait data is collected on-site through the face camera;

[0008] The read trusted digital identity QR code and the collected portrait data are transmitted to the application system in the security isolation network, the application system calls the trusted digital identity QR code dedicated encryption machine to decrypt, obtains the trusted digital identity identifier in the trusted digital identity QR code, queries the corresponding database of the application system based on the trusted digital identity identifier obtained by decryption, and obtains the associated stored portrait data;

[0009] Constructing a portrait feature extraction model, the portrait feature extraction model includes a ResNet basic network, an SE module, a CBAM module, an LBP feature extraction module and a fully connected layer; using a comprehensive target loss function to train and optimize the portrait feature extraction model to obtain a trained portrait feature extraction model;

[0010] The trained portrait feature extraction model is deployed in the security isolation network, and the stored portrait data is compared with the portrait data collected on-site to determine whether the user is a real person. If the user is determined to be a real person, the user's identity information is returned.

[0011] As a preferred embodiment of the present invention, the user's trusted digital identity, identity information and portrait data are obtained and associated with the corresponding database of the application system stored in the secure isolation network as follows:

[0012] A user identity information table and a portrait table are constructed in the corresponding database of the application system, wherein the user identity information table is used to store the user's identity information, with a unique trusted digital identity as the primary key; the portrait table is used to store the user's portrait data, with a unique trusted digital identity as the primary key.

[0013] As a preferred embodiment of the present invention, the method further comprises:

[0014] Acquire some portrait data, perform data enhancement operation on the portrait data; preprocess the enhanced portrait data, including normalization, face detection, face alignment and category labeling; store the preprocessed portrait data and the corresponding labels in a data set to obtain an ArcFace loss data set;

[0015] Randomly selecting anchor samples, positive samples, and negative samples from the ArcFace loss dataset, generating triplets, storing the triplets in another dataset, and obtaining a triplet loss dataset;

[0016] The ArcFace loss dataset and the triplet loss dataset are used to train the portrait feature extraction model.

[0017] As a preferred embodiment of the present invention, constructing a portrait feature extraction model is specifically as follows:

[0018] A convolutional neural network is used as the basis of the ResNet basic network, and the ResNet is used to extract multi-level features of the image to obtain a ResNet feature map; a SE module is added after the ResNet basic network to receive the ResNet feature map, and channel weights are learned through global average pooling and a fully connected layer to enhance features and obtain a SE feature map; a CBAM module is added after the SE module to receive the SE feature map, and features are enhanced through channel attention and spatial attention mechanisms to obtain a CBAM feature map; an LBP feature extraction module is connected after the CBAM module, and the LBP feature extraction module is used to process the CBAM feature map, capture local texture information, and obtain an LBP feature map; the LBP feature map is flattened and mapped to a low-dimensional feature vector through a fully connected layer, and the feature vector is used for portrait comparison.

[0019] As a preferred embodiment of the present invention, the comprehensive target loss function includes an ArcFace loss function and a triplet loss function. The comprehensive target loss function is used to train and optimize the portrait feature extraction model to obtain a trained portrait feature extraction model, specifically:

[0020] The ArcFace loss dataset and the triplet loss dataset are divided into training set, validation set and test set according to preset proportions respectively;

[0021] The training set is input into the portrait feature extraction model, and the feature vector is obtained through ResNet, SE module, CBAM module, LBP feature extraction module and fully connected layer; the comprehensive target loss function is calculated according to the feature vector; the weight of the model is updated through the back propagation algorithm according to the gradient of the comprehensive target loss function; the learning rate is dynamically adjusted using the Adam optimizer during the training process; after each training cycle, the performance of the portrait feature extraction model is verified using the validation set, specifically, the accuracy, recall rate and F1 score on the validation set are calculated. If the performance on the validation set no longer improves, the training is completed; the final performance of the portrait feature extraction model is calculated using the test set. If the final performance meets the user requirements, the trained portrait feature extraction model is obtained. If the final performance does not meet the user requirements, the model is optimized until the model meets the user requirements, and the trained portrait feature extraction model is obtained.

[0022] As a preferred implementation of the present invention, the trained portrait feature extraction model is deployed in a secure isolation network, and the stored portrait data is compared with the portrait data collected on site to determine whether the user is a real person. Specifically, the following steps are performed:

[0023] Preprocess the stored portrait data and the portrait data collected on site, including normalization, face detection, face alignment and category labeling;

[0024] The trained portrait feature extraction model is deployed in the security isolation network, and the deployed portrait feature extraction model is used to extract features from the pre-processed portrait data to obtain the corresponding feature vector;

[0025] The feature vector of the stored portrait data is compared with the feature vector of the portrait data collected on site, and the similarity between the feature vectors is calculated by Euclidean distance. If the similarity exceeds the preset threshold, it is determined to be real-person verification; otherwise, it is determined to be non-real-person verification.

[0026] The present invention also provides a security verification system based on trusted digital identity and portrait comparison, the system includes a user information acquisition and storage module, a two-dimensional code reading and portrait collection module, a data transmission and decryption module, a portrait feature extraction and comparison module and an identity verification result output module, wherein:

[0027] The user information acquisition and storage module is used to obtain the user's trusted digital identity, identity information and portrait data, and associate and store them in the corresponding database of the application system in the security isolation network;

[0028] The QR code reading and portrait collection module is used to read the trusted digital identity QR code presented by the user through the QR code reading device when the user performs on-site identity verification, and collect the user's portrait data on-site through the face camera;

[0029] The data transmission and decryption module is used to transmit the read trusted digital identity two-dimensional code and the collected portrait data to the application system in the security isolation network, and the application system calls the trusted digital identity two-dimensional code dedicated encryption machine to decrypt it, obtains the trusted digital identity identifier in the trusted digital identity two-dimensional code, and queries the corresponding database of the application system based on the trusted digital identity identifier obtained by decryption to obtain the associated stored portrait data;

[0030] The portrait feature extraction and comparison module is internally provided with a portrait feature extraction model, and the stored portrait data and the portrait data collected on site are compared by the portrait feature extraction model to determine whether the user is a real person for verification and obtain a judgment result;

[0031] The identity authentication result output module is used to output the verification result according to the judgment result. If it is judged to be real-person verification, the verification is successful and the user's relevant identity information is returned; if it is judged to be non-real-person verification, the verification fails and the user is informed that the verification has not passed.

[0032] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a security verification method based on a trusted digital identity and portrait comparison as described in any embodiment of the present invention is implemented.

[0033] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a security verification method based on a trusted digital identity and portrait comparison as described in any embodiment of the present invention.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] 1) The present invention provides a security verification method based on trusted digital identity and portrait comparison. By using a trusted digital identity QR code, the authenticity of the user's identity is ensured; by constructing an efficient portrait feature extraction model and adopting a comprehensive target loss function for training and optimization, while ensuring accuracy, the consumption of computing resources is reduced, which is more suitable for scenarios with limited resources or high real-time requirements; and the dual verification mechanism of the trusted digital identity QR code and portrait comparison makes the entire system more secure and reliable;

[0036] 2) The present invention provides a security verification method based on trusted digital identity and portrait comparison. Users only need to present a QR code and cooperate with face collection to complete the verification, without having to carry additional identity documents or perform complicated operations, thereby improving user experience;

[0037] 3) The present invention provides a security verification method based on trusted digital identity and portrait comparison. Since all data processing and comparison are performed within a secure isolation network, the user's personal information is better protected, reducing the risk of data leakage and abuse, and meeting the high requirements for data security and privacy protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a flow chart of a method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0039] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0040] The present invention provides the following technical solution: a security verification method and system based on trusted digital identity and portrait comparison.

[0041] Embodiment 1:

[0042] This embodiment provides a security verification method based on a trusted digital identity and a portrait comparison, the method comprising:

[0043] S1. Acquire the user's sensitive data, including a trusted digital identity, identity information and portrait data, through a secure isolation network, and associate the data with the corresponding database of the application system stored in the secure isolation network. Preferably, the trusted digital identity is obtained in the following manner:

[0044] The user initiates an authorization request corresponding to sensitive data through the terminal device, and the terminal device verifies the user's identity through the trusted digital identity authentication platform through the CTID platform. If the verification is successful, the trusted digital identity platform generates a trusted digital identity for the user.

[0045] S11. Constructing a user identity information table and a portrait table in the corresponding database of the application system, wherein the user identity information table is used to store the user's identity information, with a unique trusted digital identity identifier as a primary key; the portrait table is used to store the user's portrait data, also with a unique trusted digital identity identifier as a primary key;

[0046] S2. When the user passes through security check or ticket gate, he / she needs to undergo on-site identity verification. The trusted digital identity QR code presented by the user is read by the QR code reader, and the user's portrait data is collected on-site by the face camera;

[0047] S3, transmitting the read trusted digital identity QR code and the collected portrait data to the application system in the secure isolation network, the application system calls the trusted digital identity QR code dedicated encryption machine to decrypt, obtains the trusted digital identity identifier in the trusted digital identity QR code, queries the corresponding database of the application system based on the trusted digital identity identifier obtained by decryption, and obtains the user's portrait data in the portrait table according to the trusted digital identity identifier;

[0048] S4. Deploy the trained portrait feature extraction model in the security isolation network, compare the stored portrait data with the portrait data collected on site, and determine whether the user is a real person for verification; if the user is determined to be a real person for verification, return the user's identity information and portrait comparison score;

[0049] S41. Construct ArcFace loss dataset and triple loss dataset;

[0050] S411, the construction of the ArcFace loss dataset is specifically as follows:

[0051] Collect a large amount of portrait data to ensure that each category has a sufficient number of samples; perform data enhancement operations on the portrait data, such as rotation, scaling, cropping, color conversion, etc., to increase the diversity of the data;

[0052] The portrait data after data enhancement is preprocessed, including normalization, face detection, face alignment and category labeling. Preferably, the normalization is to normalize the image pixel value to the range of [0, 1] or [-1, 1]; the face detection is specifically to detect the face area using a face detection algorithm such as Dlib or OpenCV; the face alignment is specifically to align the face to the standard position using the face alignment algorithm MTCNN; the category labeling includes manual labeling and automatic labeling. The user selects a suitable labeling method according to the needs and the amount of data, which will not be repeated here;

[0053] The preprocessed portrait data and corresponding labels are stored in the dataset to obtain the ArcFace loss dataset;

[0054] S412, the construction of the triple loss data set is specifically:

[0055] Randomly select an anchor sample, a positive sample, and a negative sample from the ArcFace loss dataset, generate a triplet, store the triplet in another dataset, and obtain a triplet loss dataset, wherein the anchor sample is a sample randomly selected from the ArcFace loss dataset, the positive sample is a sample randomly selected from samples of the same category as the anchor sample, and the negative sample is a sample randomly selected from samples of a different category from the anchor sample;

[0056] S42, constructing a portrait feature extraction model;

[0057] The portrait feature extraction model includes a ResNet basic network, a SE module, a CBAM module, an LBP feature extraction module and a fully connected layer, wherein:

[0058] A convolutional neural network is used as the basis of the ResNet basic network, wherein the ResNet is used to extract multi-level features of the image and obtain a ResNet feature map. Preferably, ResNet solves the gradient vanishing problem in the deep network through residual connection; an SE module is added after the ResNet basic network to receive the ResNet feature map and learn and strengthen important channel features through a channel attention mechanism to obtain an SE feature map. Preferably, the SE module learns channel weights through global average pooling and a fully connected layer to enhance important features; a CBAM module is added after the SE module to receive the SE feature map, and further enhance the expressiveness of the features through channel attention and spatial attention mechanisms to obtain a CBAM feature map; an LBP feature extraction module is connected after the CBAM module, wherein the LBP feature extraction module is used to process the CBAM feature map, capture local texture information, and obtain an LBP feature map; finally, the LBP feature map is flattened and mapped to a low-dimensional feature vector through a fully connected layer, and the feature vector is used for portrait comparison;

[0059] S43. Training and optimization of portrait feature extraction model:

[0060] S431. Data preparation:

[0061] The ArcFace loss dataset and the triplet loss dataset are divided into a training set, a validation set, and a test set according to preset proportions, respectively. In this embodiment, the preset proportions are specifically 70% training set, 15% validation set, and 15% test set, wherein the training set is used for model training, the validation set is used for adjusting the hyperparameters of the model, and the test set is used for finally evaluating the performance of the model;

[0062] S432, initialization of portrait feature extraction model:

[0063] S4321. Use Xavier initialization or He initialization weights to accelerate the convergence of the model, wherein Xavier initialization is applicable to Sigmoid and Tanh activation functions, and He initialization is applicable to ReLU activation function. In this embodiment, the ReLU activation function is selected and He initialization is used;

[0064] S4322. Select Adam or SGD as the optimizer, and set hyperparameters such as learning rate and momentum. Since the Adam optimizer combines the advantages of Momentum and RMSprop and is suitable for most deep learning tasks, in this embodiment, Adam is selected as the optimizer.

[0065] S433. Define the loss function:

[0066] In this embodiment, the loss function of the portrait feature extraction model includes an ArcFace loss function and a triplet loss function, where:

[0067] The ArcFace loss function is used to enhance the discriminative ability of features during the training process. By introducing the angle cosine between the feature vector and the weight vector, the feature vectors of the same category are more compact and the feature vectors of different categories are more separated. The triplet loss function is used to enhance the relative distance of features during the training process. By minimizing the distance between the anchor point sample and the positive sample and maximizing the distance between the anchor point sample and the negative sample, the discriminative ability of features is enhanced.

[0068] S434, portrait feature extraction model training:

[0069] The training set is input into the portrait feature extraction model, and the feature vector is obtained through ResNet, SE module, CBAM module, LBP feature extraction module and fully connected layer; ArcFace loss and triplet loss are calculated according to the feature vector. During the training process, ArcFace loss and triplet loss are combined to construct a comprehensive target loss function for calculation; according to the gradient of the comprehensive target loss function, the weight of the model is updated through the back propagation algorithm; in order to avoid the gradient explosion problem, the gradient can be clipped; according to the progress of training, the Adam optimizer is used to dynamically adjust the learning rate to improve the convergence speed and performance of the model;

[0070] S435. Verification of portrait feature extraction model:

[0071] After each training cycle, the performance of the portrait feature extraction model is verified using the validation set, and the accuracy, recall, and F1 score on the validation set are calculated. If the performance on the validation set no longer improves, training is stopped early to avoid overfitting.

[0072] S436, Portrait feature extraction model test:

[0073] After the training is completed, the accuracy, recall and F1 score are calculated using the test set to evaluate the final performance of the portrait feature extraction model; if the user requirements are met, the trained portrait feature extraction model is obtained; if the user requirements are not met, the model is optimized until the model meets the user requirements, and the model optimization includes:

[0074] Adjust the model's hyperparameters to achieve optimal performance through grid search, random search, or Bayesian optimization. Remove unimportant weights in the model through pruning techniques to reduce model complexity and improve inference speed. Convert floating-point weights in the model to integer weights through quantization techniques to further reduce the model's storage space and inference time.

[0075] S44, deploy the trained portrait feature extraction model in the security isolation network, compare the stored portrait data with the portrait data collected on site, and determine whether the user is a real person verification, specifically:

[0076] Preprocess the stored portrait data and the portrait data collected on site, including normalization, face detection, face alignment and category labeling;

[0077] The trained portrait feature extraction model is pre-deployed in the security isolation network, and the pre-processed portrait data is subjected to feature extraction using the deployed portrait feature extraction model to obtain the corresponding feature vector;

[0078] The feature vector of the stored portrait data is compared with the feature vector of the portrait data collected on site, and the similarity between the feature vectors is calculated by Euclidean distance. If the similarity exceeds the preset threshold, it is determined to be real-person verification; otherwise, it is determined to be non-real-person verification.

[0079] Embodiment 2:

[0080] This embodiment provides a security verification system based on a trusted digital identity and portrait comparison, the system includes a user information acquisition and storage module, a QR code reading and portrait collection module, a data transmission and decryption module, a portrait feature extraction and comparison module and an identity verification result output module, wherein:

[0081] The user information acquisition and storage module is used to obtain the user's trusted digital identity, identity information and portrait data, and associate and store them in the corresponding database of the application system in the security isolation network;

[0082] The QR code reading and portrait collection module is used to read the trusted digital identity QR code presented by the user through the QR code reading device when the user performs on-site identity verification, and collect the user's portrait data on-site through the face camera;

[0083] The data transmission and decryption module is used to transmit the read trusted digital identity two-dimensional code and the collected portrait data to the application system in the security isolation network, and the application system calls the trusted digital identity two-dimensional code dedicated encryption machine to decrypt it, obtains the trusted digital identity identifier in the trusted digital identity two-dimensional code, and queries the corresponding database of the application system based on the trusted digital identity identifier obtained by decryption to obtain the associated stored portrait data;

[0084] The portrait feature extraction and comparison module is internally provided with a portrait feature extraction model, and the stored portrait data and the portrait data collected on site are compared by the portrait feature extraction model to determine whether the user is a real person for verification and obtain a judgment result;

[0085] The identity authentication result output module is used to output the verification result according to the judgment result. If it is judged to be real-person verification, the verification is successful and the user's relevant identity information is returned; if it is judged to be non-real-person verification, the verification fails and the user is informed that the verification has not passed.

[0086] Preferably, the portrait feature extraction model includes a ResNet basic network, a SE module, a CBAM module, an LBP feature extraction module and a fully connected layer, wherein:

[0087] A convolutional neural network is used as the basis of the ResNet basic network, and the ResNet is used to extract multi-level features of the image to obtain a ResNet feature map; a SE module is added after the ResNet basic network to receive the ResNet feature map, and channel weights are learned through global average pooling and a fully connected layer to enhance features and obtain a SE feature map; a CBAM module is added after the SE module to receive the SE feature map, and further enhance features through channel attention and spatial attention mechanisms to obtain a CBAM feature map; an LBP feature extraction module is connected after the CBAM module, and the LBP feature extraction module is used to process the CBAM feature map, capture local texture information, and obtain an LBP feature map; the LBP feature map is flattened and mapped to a low-dimensional feature vector through a fully connected layer, and the feature vector is used for portrait comparison.

[0088] Embodiment 3:

[0089] The present invention further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a security verification method based on a trusted digital identity and a portrait comparison as described in Embodiment 1 of the present invention is implemented;

[0090] Embodiment 4:

[0091] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, a security verification method based on a trusted digital identity and portrait comparison as described in Example 1 of the present invention is implemented.

[0092] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A security verification method based on trusted digital identity and portrait comparison, characterized in that: The method comprises: Obtain the user's trusted digital identity, identity information and portrait data, and associate them with the corresponding database of the application system stored in the secure isolation network; when the user performs on-site identity verification, the trusted digital identity QR code presented by the user is read by the QR code reading device, and the user's portrait data is collected on-site through the face camera; The read trusted digital identity QR code and the collected portrait data are transmitted to the application system in the security isolation network, the application system calls the trusted digital identity QR code dedicated encryption machine to decrypt, obtains the trusted digital identity identifier in the trusted digital identity QR code, queries the corresponding database of the application system based on the trusted digital identity identifier obtained by decryption, and obtains the associated stored portrait data; Constructing a portrait feature extraction model, the portrait feature extraction model includes a ResNet basic network, an SE module, a CBAM module, an LBP feature extraction module and a fully connected layer; using a comprehensive target loss function to train and optimize the portrait feature extraction model to obtain a trained portrait feature extraction model; The trained portrait feature extraction model is deployed in the security isolation network, and the stored portrait data is compared with the portrait data collected on-site to determine whether the user is a real person. If the user is determined to be a real person, the user's identity information is returned.

2. According to claim 1, a security verification method based on trusted digital identity and portrait comparison is characterized in that: Obtain the user's trusted digital identity, identity information and portrait data, and associate them with the corresponding database of the application system stored in the secure isolation network: A user identity information table and a portrait table are constructed in the corresponding database of the application system, wherein the user identity information table is used to store the user's identity information, with a unique trusted digital identity as the primary key; the portrait table is used to store the user's portrait data, with a unique trusted digital identity as the primary key.

3. According to claim 1, a security verification method based on trusted digital identity and portrait comparison is characterized in that: The method further comprises: Acquire some portrait data, perform data enhancement operation on the portrait data; preprocess the enhanced portrait data, including normalization, face detection, face alignment and category labeling; store the preprocessed portrait data and the corresponding labels in a data set to obtain an ArcFace loss data set; Randomly select anchor samples, positive samples, and negative samples from the ArcFace loss dataset, generate triplets, store the triplets in another dataset, and obtain a triplet loss dataset; The ArcFace loss dataset and the triplet loss dataset are used to train the portrait feature extraction model.

4. According to claim 3, a security verification method based on trusted digital identity and portrait comparison is characterized in that: The specific steps of building a portrait feature extraction model are as follows: A convolutional neural network is based on a ResNet basic network, where the ResNet is used to extract multi-level features of an image and obtain a ResNet feature map. A SE module is added after the ResNet basic network to receive the ResNet feature map, and channel weights are learned through global average pooling and a fully connected layer to enhance features and obtain an SE feature map. A CBAM module is added after the SE module to receive the SE feature map, enhance the features through channel attention and spatial attention mechanisms, and obtain the CBAM feature map; an LBP feature extraction module is connected after the CBAM module, and the LBP feature extraction module is used to process the CBAM feature map, capture local texture information, and obtain the LBP feature map; the LBP feature map is flattened and mapped to a low-dimensional feature vector through a fully connected layer, and the feature vector is used for portrait comparison.

5. A security verification method based on trusted digital identity and portrait comparison according to claim 4, characterized in that: The comprehensive target loss function includes the ArcFace loss function and the triplet loss function. The comprehensive target loss function is used to train and optimize the portrait feature extraction model to obtain the trained portrait feature extraction model, which is specifically: The ArcFace loss dataset and the triplet loss dataset are divided into training set, validation set and test set according to preset proportions respectively; The training set is input into the portrait feature extraction model, and the feature vector is obtained through ResNet, SE module, CBAM module, LBP feature extraction module and fully connected layer; The comprehensive objective loss function is calculated according to the feature vector; the weight of the model is updated through the back propagation algorithm according to the gradient of the comprehensive objective loss function; the learning rate is dynamically adjusted using the Adam optimizer during the training process; after each training cycle, the performance of the portrait feature extraction model is verified using the validation set, specifically by calculating the accuracy, recall and F1 score on the validation set. If the performance on the validation set no longer improves, the training is completed; the final performance of the portrait feature extraction model is calculated using the test set. If the final performance meets user requirements, the trained portrait feature extraction model is obtained. If the final performance does not meet user requirements, the model is optimized until the model meets user requirements, and the trained portrait feature extraction model is obtained.

6. A security verification method based on trusted digital identity and portrait comparison according to claim 5, characterized in that: The trained portrait feature extraction model is deployed in the security isolation network, and the stored portrait data is compared with the portrait data collected on site to determine whether the user is a real person. Specifically: Preprocess the stored portrait data and the portrait data collected on site, including normalization, face detection, face alignment and category labeling; The trained portrait feature extraction model is deployed in the security isolation network, and the deployed portrait feature extraction model is used to extract features from the pre-processed portrait data to obtain the corresponding feature vector; The feature vector of the stored portrait data is compared with the feature vector of the portrait data collected on site, and the similarity between the feature vectors is calculated by Euclidean distance. If the similarity exceeds the preset threshold, it is determined to be real-person verification; otherwise, it is determined to be non-real-person verification.

7. A security verification system based on trusted digital identity and portrait comparison, characterized in that: The system includes a user information acquisition and storage module, a QR code reading and portrait collection module, a data transmission and decryption module, a portrait feature extraction and comparison module, and an identity verification result output module, wherein: The user information acquisition and storage module is used to obtain the user's trusted digital identity, identity information and portrait data, and associate and store them in the corresponding database of the application system in the security isolation network; The QR code reading and portrait collection module is used to read the trusted digital identity QR code presented by the user through the QR code reading device when the user performs on-site identity verification, and collect the user's portrait data on-site through the face camera; The data transmission and decryption module is used to transmit the read trusted digital identity two-dimensional code and the collected portrait data to the application system in the security isolation network, and the application system calls the trusted digital identity two-dimensional code dedicated encryption machine to decrypt it, obtains the trusted digital identity identifier in the trusted digital identity two-dimensional code, and queries the corresponding database of the application system based on the trusted digital identity identifier obtained by decryption to obtain the associated stored portrait data; The portrait feature extraction and comparison module is internally provided with a portrait feature extraction model, and the stored portrait data and the portrait data collected on site are compared by the portrait feature extraction model to determine whether the user is a real person for verification and obtain a judgment result; The identity authentication result output module is used to output the verification result according to the judgment result. If it is judged to be real-person verification, the verification is successful and the user's relevant identity information is returned; if it is judged to be non-real-person verification, the verification fails and the user is informed that the verification has not passed.

8. A security verification system based on trusted digital identity and portrait comparison according to claim 7, characterized in that: The portrait feature extraction model includes a ResNet basic network, a SE module, a CBAM module, an LBP feature extraction module and a fully connected layer, wherein: A convolutional neural network is used as the basis of the ResNet basic network, and the ResNet is used to extract multi-level features of the image to obtain a ResNet feature map; a SE module is added after the ResNet basic network to receive the ResNet feature map, and channel weights are learned through global average pooling and a fully connected layer to enhance features and obtain a SE feature map; a CBAM module is added after the SE module to receive the SE feature map, and further enhance features through channel attention and spatial attention mechanisms to obtain a CBAM feature map; an LBP feature extraction module is connected after the CBAM module, and the LBP feature extraction module is used to process the CBAM feature map, capture local texture information, and obtain an LBP feature map; the LBP feature map is flattened and mapped to a low-dimensional feature vector through a fully connected layer, and the feature vector is used for portrait comparison.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the security verification method based on trusted digital identity and portrait comparison as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, a security verification method based on a trusted digital identity and portrait comparison as described in any one of claims 1 to 7 is implemented.

Images