A console login identity information collection and verification method based on deep learning

Abstract

The application relates to the technical field of information security and discloses a console login identity information acquisition and verification method based on deep learning, which comprises the following steps: S1, collecting biological characteristics and behavioral characteristics of a user, extracting a feature vector of the biological characteristics by using a deep learning model, extracting a feature vector of the behavioral characteristics by using a time sequence model, splicing to obtain an identity feature vector, and splicing the feature vector of the biological characteristics and the feature vector of the behavioral characteristics to construct the identity feature vector; S2, dimensionally reducing the identity feature vector by using a topology-preserving projection method and constructing a low-dimensional manifold embedding representation; and S3, adding a manifold regularization term in a loss calculation process of the deep learning model. The application jointly extracts identity features by using deep learning and a time sequence model, achieves the effect of quickly and accurately acquiring biological and behavioral information, and solves the problems of high misrecognition rate and low recognition efficiency compared with a single feature recognition scheme in the prior art.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information security, in particular to a console login identity information collection and verification method based on deep learning. BACKGROUND

[0002] In the field of modern information security, identity verification is a key link to ensure system security. Traditional console login verification methods mainly rely on the combination of username and password. However, they are vulnerable to password leakage, brute force attacks and other threats, and have low security. In recent years, to improve security, biometric recognition technology has been gradually applied to identity verification, such as face recognition and fingerprint recognition. However, single biometric recognition scheme has high misrecognition rate and low recognition efficiency in practical application, especially in complex environments, and recognition accuracy cannot be guaranteed.

[0003] In addition, existing feature dimension reduction methods mostly use simple linear dimension reduction techniques, such as principal component analysis. When dealing with high-dimensional data, it is easy to cause loss of feature structure, which affects the generalization ability of the model. The deficiency of feature dimension reduction reduces the separability of identity features in high-dimensional feature space, affecting the accuracy of identity verification.

[0004] In terms of abnormal login detection, traditional methods usually rely on a single measurement method, such as simple threshold judgment. When facing complex attack patterns, it shows insensitivity, which easily leads to high misjudgment rate and cannot effectively identify abnormal login behavior.

[0005] In existing technology, single biometric recognition method performs poorly in complex environments. Due to factors such as environmental light and angle changes, the accuracy and efficiency of recognition are limited, resulting in high misrecognition rate and failing to meet the needs of high security scenarios.

[0006] Traditional linear dimension reduction techniques easily lose important feature structure information when dealing with high-dimensional identity features. The loss of feature structure directly affects the generalization ability of the model, making the model unable to adapt to variable identity feature distribution in practical application, reducing the reliability of identity verification.

[0007] Existing abnormal login detection methods mostly rely on simple threshold judgment and lack sensitivity to complex attack patterns. A single measurement method is prone to misjudgment when facing diversified attacks, and cannot effectively distinguish between normal and abnormal login behavior, affecting the security of the system.

[0008] Therefore, the present application proposes a console login identity information collection and verification method based on deep learning to solve the above problems. SUMMARY

[0009] In view of the deficiencies of the prior art, the console login identity information collection verification method based on deep learning is provided to solve the problems in the background art.

[0010] To achieve the above object, the technical scheme is as follows: a console login identity information collection verification method based on deep learning, comprising:

[0011] S1, collecting biological characteristics and behavioral characteristics of the user, using a deep learning model to extract a feature vector of the biological characteristics, using a time series model to extract a feature vector of the behavioral characteristics, splicing to obtain an identity feature vector, and splicing the feature vector of the biological characteristics and the feature vector of the behavioral characteristics to construct the identity feature vector;

[0012] The biological characteristics and behavioral characteristics of the user are collected, and the collection, use and processing of the biological characteristics and behavioral characteristics of the user need to comply with relevant laws and standards of relevant countries and regions;

[0013] S2, using a topology preserving projection method to reduce the dimension of the identity feature vector to construct a low-dimensional manifold embedding representation;

[0014] S3, adding a manifold regularization term in the loss calculation process of the deep learning model;

[0015] S4, calculating the Mahalanobis distance between the login user identity feature vector and the normal user identity feature mean according to the low-dimensional manifold embedding representation generated in S2;

[0016] S5, calculating the probability density value of the login user identity feature according to the low-dimensional manifold embedding representation generated in S2, and performing joint judgment, if the calculated probability density value is lower than the set probability density threshold or the calculated Mahalanobis distance exceeds the set Mahalanobis distance threshold, rejecting the current login request, otherwise allowing normal login, to finally complete the decision process of identity verification.

[0017] Preferably, in S1, collecting biological characteristics and behavioral characteristics of the user further comprises:

[0018] Substep 1.1, collecting face, fingerprint, voiceprint data and keyboard input, mouse trajectory data, wherein the face, fingerprint, voiceprint data are biological characteristic inputs, and the keyboard input, mouse trajectory data are behavioral characteristic inputs;

[0019] Substep 1.2, using a deep learning model to extract features of the biological characteristic input to obtain a biological characteristic vector, and the deep learning model is defined as:

[0020] ,

[0021] wherein, as a biometric input, as a weight matrix, as a bias vector, as an activation function;

[0022] Sub-step 1.3, feature extraction is performed on the behavior feature input to obtain a behavior feature vector using a time series model, and the time series model is defined as:

[0023] ,

[0024] wherein, is the behavior feature input, is the weight matrix, is the bias vector, is the activation function;

[0025] The biometric feature vector and the behavior feature vector are then combined into an identity feature vector through feature splicing.

[0026] Preferably, in S2, the identity feature vector is reduced in dimension using a topologically preserving projection method, and constructing a low-dimensional manifold embedding representation further comprises:

[0027] Sub-step 2.1, input the identity feature vector into the topologically preserving projection method, wherein the identity feature vector is the biometric feature vector and the behavior feature vector obtained by splicing in step S1;

[0028] Sub-step 2.2, in the topologically preserving projection, an adjacency matrix is constructed to represent the similarity between samples, and the adjacency matrix is calculated by the following formula:

[0029] ,

[0030] wherein, and are the identity feature vectors, is the adjacency matrix;

[0031] Sub-step 2.3, according to the adjacency matrix , a projection matrix is calculated, so that the distance between adjacent samples remains unchanged in the low-dimensional space after projection, and the projection matrix is calculated by the following optimization objective:

[0032] ,

[0033] wherein, is the projection matrix, and are the identity feature vectors, is the adjacency matrix;

[0034] By optimizing the target, the similar identity samples are kept adjacent in the low-dimensional space, and then the low-dimensional manifold embedding representation is constructed.

[0035] Preferably, in the S3, the manifold regularization term is further included in the loss calculation process of the deep learning model.

[0036] Sub-step 3.1, based on the identity feature vector generated in step S1, the cross-entropy loss is calculated, and the cross-entropy loss is defined as:

[0037]

[0038] Wherein, is the real class probability of the sample, is the prediction probability of the model output;

[0039] Sub-step 3.2, using the adjacency matrix constructed in step S2, the manifold regularization loss is calculated, and the manifold regularization loss is defined as:

[0040] ,

[0041] Wherein, is the identity feature vector the low-dimensional representation obtained after mapping, is the identity feature vector the low-dimensional representation obtained after mapping, is the adjacency matrix, is the regularization weight;

[0042] Sub-step 3.3, combining sub-step 3.1 and sub-step 3.2, the cross-entropy loss and the manifold regularization loss are added to obtain the total loss:

[0043] ,

[0044] Wherein, is the overall loss of the model, which ensures the classification accuracy and maintains the consistency of the low-dimensional manifold structure in the identity feature extraction process.

[0045] Preferably, in the S4, the Mahalanobis distance of the identity feature is calculated for identity verification, which further includes:

[0046] Sub-step 4.1, according to the low-dimensional manifold embedding representation generated in step S2, the difference between the identity feature vector of the login user and the mean value of the normal user identity feature is calculated, and the difference is defined as:

[0047] ,

[0048] wherein, is the difference between the identity feature vector of the login user and the mean of identity features of normal users, is the identity feature vector of the login user, is the mean of identity features of normal users;

[0049] Sub-step 4.2, calculating the covariance matrix of the identity feature vector difference The matrix is used to quantify the variance and covariance of the identity features, and the covariance matrix is defined as:

[0050] ,

[0051] wherein, is the identity feature vector, is the number of samples, is the covariance matrix of the samples;

[0052] Sub-step 4.3, calculating Mahalanobis distance to measure the distance between the identity features of the login user and the mean of normal users, and the Mahalanobis distance is defined as:

[0053] ,

[0054] wherein, is the Mahalanobis distance, is the identity feature vector of the login user, is the mean of identity features of normal users, is the inverse matrix of the covariance matrix, is the transpose operation;

[0055] is used to calculate the similarity between the identity features, and if the Mahalanobis distance exceeds a set threshold, the login behavior is determined to be abnormal, triggering secondary verification.

[0056] Preferably, in the S5, the calculation of the probability density value of the identity feature further comprises:

[0057] Sub-step 5.1, inputting the identity feature vector of the login user in the low-dimensional manifold embedding representation generated in step S2 Inputting the Gaussian probability density function to calculate the probability density value, and the formula is:

[0058] ,

[0059] wherein, is the probability density value of the identity feature vector of the login user, and d is the dimension of the low-dimensional manifold space, is the determinant of the covariance matrix ​Let covariance be the feature vector of identity. For the login user's identity feature vector, This represents the average of normal user identity characteristics.

[0060] Sub-step 5.2: Calculate the probability density value obtained in sub-step 5.1. With the set probability density threshold Comparison, This represents the critical value used to determine the probability density of identity features;

[0061] Sub-step 5.3, combined with the probability density value from sub-step 5.2 With the set probability density threshold The comparison results, and the calculated Mahalanobis distance. With respect to the set Mahalanobis distance threshold Based on the comparison results, a joint judgment is made, and the decision formula is as follows:

[0062] ,

[0063] in, The Mahalanobis distance, This is the set Mahalanobis distance threshold.

[0064] Preferably, the extraction of the biometric features includes: using Network-based facial feature extraction, using Fingerprint features are extracted from the network using... Voiceprint features are extracted via network.

[0065] The extraction of the behavioral features includes: using Keyboard input pattern features are extracted, and mouse trajectory features are extracted using dynamic time warping.

[0066] Preferably, the projection matrix of the topology-preserving projection method is calculated through an optimization objective, which is to minimize the sum of squared Euclidean distances after projection of the adjacency matrix constraints among all samples.

[0067] A terminal device includes a sensor module for collecting user biometrics and behavioral characteristics, and a processing module for executing a console login identity information collection and verification method.

[0068] A storage medium storing instruction programs for controlling terminal devices to collect and verify identity information according to a console login identity information collection and verification method.

[0069] This invention provides a deep learning-based method for collecting and verifying console login identity information. It has the following beneficial effects:

[0070] 1. This invention uses deep learning and time series models to jointly extract identity features, achieving the effect of quickly and accurately collecting biological and behavioral information. Compared with the single feature recognition schemes in the existing technology, it solves the problems of high false recognition rate and low recognition efficiency.

[0071] 2. This invention introduces topology-preserving projection and manifold regularization to optimize low-dimensional embedding representations. Compared with simple dimensionality reduction methods in the prior art, it solves the defects of feature structure loss and weak model generalization ability.

[0072] 3. The present invention designs a verification mechanism based on Mahalanobis distance and probability density joint decision-making, which improves the accuracy of identity verification. Compared with the single measurement method in the existing technology, it solves the problems of insensitivity and high false judgment rate in abnormal login detection. Attached Figure Description

[0073] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0074] To enable those skilled in the art to understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.

[0075] The present invention will now be described in detail with reference to the accompanying drawings:

[0076] Example:

[0077] Please see the appendix Figure 1 This invention provides a deep learning-based method for collecting and verifying console login identity information, including:

[0078] S1. Collect the user's biometric and behavioral features, use a deep learning model to extract the feature vector of the biometric features, use a time series model to extract the feature vector of the behavioral features, and concatenate them to obtain the identity feature vector. Then, concatenate the feature vector of the biometric features with the feature vector of the behavioral features to construct the identity feature vector.

[0079] Biometric extraction includes: using Network-based facial feature extraction, using Fingerprint features are extracted from the network using... Voiceprint features are extracted via network.

[0080] Behavioral feature extraction includes: employing The keyboard input mode features are extracted, and the mouse trajectory features are extracted by dynamic time warping;

[0081] Sub-step 1.1, collecting user face, fingerprint, voiceprint data and keyboard input, mouse trajectory data, wherein the face, fingerprint, voiceprint data are biological feature inputs, and the keyboard input, mouse trajectory data are behavior feature inputs;

[0082] Sub-step 1.2, using a deep learning model to extract features from the biological feature input to obtain a biological feature vector, and the deep learning model is defined as:

[0083] ,

[0084] Wherein, is the biological feature input, is a weight matrix, is a bias vector, is an activation function;

[0085] Sub-step 1.3, using a time series model to extract features from the behavior feature input to obtain a behavior feature vector, and the time series model is defined as:

[0086] ,

[0087] Wherein, is the behavior feature input, is a weight matrix, is a bias vector, is an activation function;

[0088] The biological feature vector and the behavior feature vector are then combined into an identity feature vector by feature splicing;

[0089] S2, using a topology preserving projection method to reduce the dimension of the identity feature vector, and constructing a low-dimensional manifold embedding representation;

[0090] The projection matrix of the topology preserving projection method is calculated by optimizing the objective, and the above optimization objective is to minimize the square sum of the Euclidean distance after projection of all samples under the constraint of the adjacency matrix;

[0091] Sub-step 2.1, inputting the identity feature vector into the topology preserving projection method, wherein the identity feature vector is the biological feature vector and the behavior feature vector spliced in step S1;

[0092] Sub-step 2.2, in the topology preserving projection, an adjacency matrix A is constructed to represent the similarity between samples, and the adjacency matrix is calculated by the following formula:

[0093] ,

[0094] Wherein, and is the identity feature vector, is the adjacency matrix;

[0095] Sub-step 2.3, according to the adjacency matrix , the projection matrix is calculated, so that the distance between adjacent samples remains unchanged in the low-dimensional space after projection, and the projection matrix is calculated by the following optimization objective:

[0096] ,

[0097] wherein, is the projection matrix, and is the identity feature vector, is the adjacency matrix;

[0098] By optimizing the objective, similar identity samples remain adjacent in the low-dimensional space, and a low-dimensional manifold embedding representation is constructed;

[0099] S3, in the loss calculation process of the deep learning model, a manifold regularization term is added;

[0100] Sub-step 3.1, based on the identity feature vector generated in step S1, the cross-entropy loss is calculated, and the cross-entropy loss is defined as:

[0101] ,

[0102] wherein, is the real class probability of the sample, is the model output prediction probability;

[0103] Sub-step 3.2, using the adjacency matrix constructed in step S2, the manifold regularization loss is calculated, and the manifold regularization loss is defined as:

[0104] ,

[0105] wherein, is the identity feature vector mapped to obtain a low-dimensional representation, is the identity feature vector mapped to obtain a low-dimensional representation, is the adjacency matrix, is the regularization weight;

[0106] Sub-step 3.3, combining sub-step 3.1 and sub-step 3.2, the cross-entropy loss and the manifold regularization loss are added to obtain the total loss:

[0107] ,

[0108] wherein, is the model overall loss, ensuring the classification accuracy and maintaining the consistency of the low-dimensional manifold structure in the identity feature extraction process;

[0109] S4, calculating the Mahalanobis distance between the identity feature vector of the login user represented by the low-dimensional manifold embedding generated in S2 and the mean of the identity features of normal users;

[0110] Sub-step 4.1, calculating the difference between the identity feature vector of the login user and the mean of the identity features of normal users according to the low-dimensional manifold embedding generated in step S2, the difference being defined as:

[0111] ,

[0112] wherein, is the difference between the identity feature vector of the login user and the mean of the identity features of normal users, is the identity feature vector of the login user, is the mean of the identity features of normal users;

[0113] Sub-step 4.2, calculating the covariance matrix of the identity feature vector difference , the matrix being used to quantify the variance and covariance of the identity features, the covariance matrix being defined as:

[0114] ,

[0115] wherein, is the identity feature vector, is the number of samples, is the covariance matrix of the samples;

[0116] Sub-step 4.3, calculating the Mahalanobis distance to measure the distance between the identity feature of the login user and the mean of the normal users, the Mahalanobis distance being defined as:

[0117] ,

[0118] wherein, is the Mahalanobis distance, is the identity feature vector of the login user, is the mean of the identity features of normal users, is the inverse matrix of the covariance matrix, is the transpose operation;

[0119] ​​Used to calculate the similarity between identity features. If the Mahalanobis distance exceeds the set threshold, the login behavior is judged to be abnormal and a second verification is triggered.

[0120] S5: Calculate the probability density value of the login user identity features based on the low-dimensional manifold embedding representation generated in S2, and make a joint judgment. If the calculated probability density value is lower than the set probability density threshold or the calculated Mahalanobis distance exceeds the set Mahalanobis distance threshold, reject the current login request; otherwise, allow normal login to complete the identity verification decision process.

[0121] Sub-step 5.1 involves embedding the login user identity feature vector into the low-dimensional manifold representation generated in step S2. The probability density value is calculated by inputting the Gaussian probability density function, using the following formula:

[0122] ,

[0123] in, Let be the probability density value of the login user's identity feature vector, and d be the dimension of the low-dimensional manifold space. Covariance matrix The determinant, Let covariance be the feature vector of identity. For the login user's identity feature vector, This represents the average of normal user identity characteristics.

[0124] Sub-step 5.2: Calculate the probability density value obtained in sub-step 5.1. With the set probability density threshold Comparison, This represents the critical value used to determine the probability density of identity features;

[0125] Sub-step 5.3, combined with the probability density value from sub-step 5.2 With the set probability density threshold The comparison results, and the calculated Mahalanobis distance. With respect to the set Mahalanobis distance threshold Based on the comparison results, a joint judgment is made, and the decision formula is as follows:

[0126] ,

[0127] in, The Mahalanobis distance, This is the set Mahalanobis distance threshold.

[0128] Step S1 comprehensively collects users' biometric and behavioral characteristics by combining deep learning models and time series models. , and Advanced network structure, in turn, extract face, fingerprint and voiceprint features, to ensure high-precision extraction of biometric features. At the same time, using and dynamic time warping technology, extract user's keyboard input mode and mouse trajectory features, rich behavior feature diversity. By feature splicing, biometric features and behavior features are combined into identity feature vectors, enhancing the expression ability of identity features, providing the basis for subsequent dimension reduction and verification.

[0129] In step S2, the topological preserving projection method is used to reduce the dimension of the identity feature vector, and a low-dimensional manifold embedding representation is constructed. By minimizing the projection of the Euclidean distance square sum constrained by the adjacency matrix, it is ensured that similar identity samples maintain the proximity relationship in the low-dimensional space. This method can effectively reduce the feature dimension, reduce the computational complexity, and preserve the topological structure of the identity feature, improve the generalization ability and robustness of the model, and lay the foundation for the subsequent regularization and verification steps.

[0130] In step S3, the manifold regularization term is added in the loss calculation process of the deep learning model. By calculating the cross-entropy loss and the manifold regularization loss, the total loss is obtained by adding them, which ensures that the model can maintain classification accuracy and consistency of low-dimensional manifold structure during training. The introduction of manifold regularization effectively prevents the overfitting phenomenon of the model, improves the adaptability of the model in different environments, and enhances the reliability of identity verification.

[0131] In step S4, the Mahalanobis distance of the identity feature is calculated for identity verification. Using the low-dimensional manifold embedding representation, the difference between the login user identity feature and the normal user mean is calculated, and the covariance matrix is used to quantify the variance and covariance of the feature. The calculation of Mahalanobis distance provides an effective similarity measurement method, which can accurately identify abnormal login behavior and trigger secondary verification, improving the security and protection ability of the system.

[0132] In step S5, the probability density value of the identity feature is calculated, combined with the result of the Mahalanobis distance, to make the final identity verification decision. Using the Gaussian probability density function, the probability density of the login user identity feature is evaluated and compared with the set threshold. By integrating the probability density and the Mahalanobis distance, the accuracy and sensitivity of identity verification are ensured, the false rejection rate is reduced, and the user experience and system security are improved.

[0133] In summary, the present application combines deep learning and time series models, comprehensively collects and extracts user biological and behavioral characteristics, optimizes feature dimension reduction and model training process using topology preserving projection and manifold regularization techniques, and finally improves the accuracy and security of identity verification through joint decision of Mahalanobis distance and probability density. It effectively solves the problems of high false recognition rate, loss of feature structure and insensitivity to anomaly detection in the prior art, and provides an efficient and reliable solution for console login identity information collection and verification.

[0134] A terminal device includes a sensor module for collecting user biological and behavioral characteristics and a processing module for executing a console login identity information collection and verification method.

[0135] A storage medium stores an instruction program for controlling a terminal device to collect and verify identity information according to a console login identity information collection and verification method.

[0136] The terminal device integrates a sensor module for collecting user biological and behavioral characteristics, which can obtain user identity information in real time and efficiently. Through the sensor module, the device can accurately capture biological characteristics such as face, fingerprint, voiceprint, and behavioral characteristics such as keyboard input mode and mouse trajectory, providing rich data support for identity verification. In addition, the built-in processing module of the device can execute the console login identity information collection and verification method, ensuring the efficiency and accuracy of data processing.

[0137] The storage medium stores an instruction program for controlling a terminal device to collect and verify identity information according to a console login identity information collection and verification method, ensuring the flexibility and scalability of the system. Through the storage medium, the device can quickly load and update identity verification algorithms to adapt to different application scenarios and security requirements. The use of storage media simplifies the maintenance and upgrade process of the device, and improves the stability and reliability of the system, ensuring the continuous effectiveness of identity verification.

[0138] Although embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A deep learning-based console login identity information collection verification method, characterized in that, Comprise: S1, collect the biological characteristics and behavior characteristics of the user, adopt a deep learning model to extract the feature vector of the biological characteristics, adopt a time series model to extract the feature vector of the behavior characteristics, splice to obtain an identity feature vector, and splice the feature vector of the biological characteristics and the feature vector of the behavior characteristics to construct the identity feature vector; S2, adopt a topology preserving projection method to reduce the dimension of the identity feature vector, and construct a low-dimensional manifold embedding representation; S3, in the loss calculation process of the deep learning model, a manifold regularization term is added; S4, calculate the Mahalanobis distance between the identity feature vector of the login user and the mean of the identity feature of the normal user according to the low-dimensional manifold embedding representation generated in S2; S5, calculate the probability density value of the identity feature of the login user according to the low-dimensional manifold embedding representation generated in S2, and make a joint judgment, if the calculated probability density value is lower than the set probability density threshold or the calculated Mahalanobis distance exceeds the set Mahalanobis distance threshold, reject the current login request, otherwise allow normal login, to finally complete the decision process of identity verification. 2.The console login identity information collection and verification method based on deep learning of claim 1, wherein, In S1, collecting the biological characteristics and behavior characteristics of the user further comprises: Substep 1.1, collect user face, fingerprint, voiceprint data and keyboard input, mouse trajectory data, wherein the face, fingerprint, voiceprint data are biological characteristic inputs, and the keyboard input, mouse trajectory data are behavior characteristic inputs; Substep 1.2, adopt a deep learning model to extract features from the biological characteristic input to obtain a biological feature vector, and the deep learning model is defined as: ， wherein, is a biometric input, is a weight matrix, is a bias vector, is an activation function; Substep 1.3, adopt a time series model to extract features from the behavior characteristic input to obtain a behavior feature vector, and the time series model is defined as: ， wherein, is a behavior feature input, is a weight matrix, is a bias vector, is an activation function; After that, the biological feature vector and the behavior feature vector are spliced into an identity feature vector. 3.The console login identity information collection and verification method based on deep learning of claim 1, characterized in that, In S2, the topology preserving projection method is used to reduce the dimension of the identity feature vector to construct a low-dimensional manifold embedding representation, which further comprises: Substep 2.1, input the identity feature vector into the topology preserving projection method, wherein the identity feature vector is the biological feature vector and the behavior feature vector spliced in step S1; Sub-step 2.2, in the topology preserving projection, constructing an adjacency matrix to represent the similarity between samples, the adjacency matrix is calculated by the following formula: ， wherein, and is an identity feature vector, is an adjacency matrix; Sub-step 2.3, according to the adjacency matrix , a projection matrix is calculated , which keeps the distance between adjacent samples unchanged in the low-dimensional space after projection, and the projection matrix is calculated by the following optimization objective: ， wherein, is a projection matrix, and is an identity feature vector, is an adjacency matrix; By optimizing the target, similar identity samples are kept adjacent in the low-dimensional space, and a low-dimensional manifold embedding representation is constructed. 4.The console login identity information collection and verification method based on deep learning of claim 1, wherein, In S3, the manifold regularization term is added in the loss calculation process of the deep learning model, which further comprises: Substep 3.1, calculate the cross-entropy loss based on the identity feature vector generated in step S1, and the cross-entropy loss is defined as: ， wherein, is the sample true class probability, is the model output prediction probability; Substep 3.2, calculate the manifold regularization loss using the adjacency matrix constructed in step S2, and the manifold regularization loss is defined as: ， wherein, is an identity feature vector is a low-dimensional representation obtained after mapping, is an identity feature vector is a low-dimensional representation obtained after mapping, is an adjacency matrix, is a regularization weight; Substep 3.3, combine substep 3.1 and substep 3.2 to add the cross-entropy loss and the manifold regularization loss to obtain the total loss: ， wherein, is the overall loss of the model, ensuring the classification accuracy and maintaining the consistency of the low-dimensional manifold structure in the identity feature extraction process. 5.The console login identity information collection and verification method based on deep learning of claim 1, wherein, In S4, calculating the Mahalanobis distance of the identity feature for identity verification further comprises: Sub-step 4.1: Calculate the identity feature vector of the logged-in user based on the low-dimensional manifold embedding representation generated in step S2. Average of normal user identity characteristics The difference, defined as: ， wherein, is the difference between the identity feature vector of the logged user and the mean of the identity features of normal users, is the identity feature vector of the logged user, is the mean of the identity features of normal users; Sub-step 4.2, calculating the covariance matrix of the identity feature vector differences The matrix is used to quantify the variance and covariance of the identity features, and the covariance matrix is defined as: ， wherein, is an identity feature vector, is a number of samples, is a covariance matrix of the samples; Sub-step 4.3, calculating Mahalanobis distance The Mahalanobis distance is defined as: ， wherein, is the Mahalanobis distance, is the identity feature vector of the logged user, is the mean of the identity features of normal users, is the inverse of the covariance matrix, is the transpose operation; For calculating the similarity between identity features, if the Mahalanobis distance exceeds the set threshold, it is determined that the login behavior is abnormal, and secondary verification is triggered.

6. The console login identity information collection and verification method based on deep learning according to claim 1, characterized in that, In S5, calculating the probability density value of the identity feature further comprises: Sub-step 5.1, embedding the low-dimensional manifold generated in step S2 into the representation of the user identity feature vector of the login user The input Gaussian probability density function calculates the probability density value, and the formula is: ， wherein, is the probability density value of the login user identity feature vector, d is the dimension of the low-dimensional manifold space, is the determinant of the covariance matrix is the covariance matrix of the identity feature vector, is the login user identity feature vector, is the normal user identity feature mean;​ Sub-step 5.2, comparing the probability density value calculated in sub-step 5.1 with a set probability density threshold value with a set probability density threshold value denotes a threshold value for judging the probability density of the identity feature​ Sub-step 5.3, combine the probability density value in sub-step 5.2 with the comparison result of the set probability density threshold value and the calculated Mahalanobis distance with the comparison result of the set Mahalanobis distance threshold value , and make a joint decision, and the decision formula is: ， wherein, is the Mahalanobis distance, is a set Mahalanobis distance threshold.

7. The console login identity information collection and verification method based on deep learning according to claim 2, characterized in that, The extraction of the biological features comprises: The network extracts the face features, adopts The network extracts the fingerprint features, adopts The network extracts the voiceprint features; The extraction of the behavior features comprises: The keyboard input mode features are extracted, and the mouse trajectory features are extracted by using dynamic time warping. 8.The console login identity information acquisition and verification method based on deep learning of claim 3, characterized in that, The projection matrix of the topological preserving projection method is calculated by optimizing an objective, and the objective is to minimize the sum of squared Euclidean distances between all samples after projection under the constraint of an adjacency matrix.

9. A terminal device, comprising: The sensor module is used for collecting biological characteristics and behavioral characteristics of the user, and the processing module is used for executing the control console login identity information collection verification method in claim 1.

10. A storage medium, characterized by The terminal device stores an instruction program for controlling the terminal device to collect and verify the identity information according to the control console login identity information collection verification method in claim 1.

Images