Two-factor authentication method and system based on password and key-press resonance biometrics

Abstract

This invention discloses a two-factor authentication method and system based on password and keystroke resonance biometrics. During the registration phase, the method acquires the password information entered by the registered user and also obtains the keystroke resonance biometrics of the user entering the password. The keystroke resonance biometrics are then used to train a biometric authentication model. During the login phase, the method acquires the password information entered by the logged-in user and also obtains the keystroke resonance biometrics of the user entering the password. The login-phase password information is matched with the registration-phase password information. If a match is found, the keystroke resonance biometrics of the login phase are input into the trained biometric authentication model to obtain the authentication result. This invention achieves highly secure authentication by collecting and analyzing the keystroke resonance biometrics generated by the user during password input, combined with traditional password verification.

Description

Technical Field

[0001] This invention relates to user authentication and authorization technology, specifically to a two-factor authentication method and system based on the biometric features of password and keystroke resonance. Background Technology

[0002] The Internet of Things (IoT) is a highly integrated and comprehensive application of information technology, a global intelligent network connecting objects and people. Its purpose is to achieve the fusion of data collected by physical devices and various sensors without human interference, thereby enabling observation and cognition of the external environment and establishing a connection with the internet according to rules set within the network, completing the initial state of information. With the continuous expansion of the power grid, in the context of IoT applications, more distribution terminals need to be connected to the grid, requiring regular maintenance by operators to ensure the safe and stable operation of the grid. The most crucial aspect of regular maintenance of the distribution network is identity authentication. Incorrect authentication, leading to unauthorized operators accessing distribution terminals, could pose significant security risks to the power grid and affect its stable operation. Identity authentication is a vital component of the service architecture; to ensure the security of the terminal operating system, a reliable user identification system needs to be established.

[0003] Two-factor authentication requires users to provide two separate credentials for authentication. The first factor (credential) is typically a knowledge factor, where a password or PIN is something only an authorized user should know. The second factor (credential) is typically a possession factor, where a hardware token (such as an ID card, USB token, or wireless tag) or software token (such as a smartphone or smartwatch) is something only an authorized user should possess. Hardware token-based authentication requires each authorized user to own and carry an additional micro-device, which the user must enter a real-time code on upon login, resulting in high manufacturing costs. Software token-based authentication requires the service provider to send a verification code to the user upon login, which the user then enters to complete the authentication, also presenting a burden for the user. It is evident that traditional solutions are costly, require user interaction, and are incompatible with the power grid environment.

[0004] In recent years, research and application of two-factor authentication schemes for legitimate users have developed rapidly, moving beyond traditional hardware tokens and SMS verification codes to explore authentication schemes with lower costs and simpler user interaction. One-click authentication and short-range communication (SMR) are widely used. With one-click authentication, the system notifies the user via the application upon login, prompting them to approve or reject the request. However, if an attacker and an operator simultaneously log into the power distribution network account, the operator cannot distinguish whether the login request sent to their registered phone is legitimate and may mistakenly approve the attacker's request. SMR relies on Bluetooth Low Energy technology, authenticating users through Bluetooth communication between the registered mobile phone and a nearby computer. However, most browsers do not support the Bluetooth API, and if an adversary establishes a Bluetooth connection with the victim's phone to bypass 2FA, it poses a security risk. Therefore, how to propose an identity authentication scheme that balances usability and security in the context of the power grid, while reducing costs and maintaining user convenience, remains a pressing problem to be solved. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a two-factor authentication method and system based on password and key resonance biometric features, which collects and analyzes the key resonance biometric features generated by the user during password input, and combines them with traditional password verification to achieve highly secure authentication.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A two-factor authentication method based on the biometric features of password and keystroke resonance includes the following steps: Obtain the first password information entered by the registered user, and obtain the first key resonance biometric feature when the registered user enters the first password information. Use the first key resonance biometric feature to train a biometric authentication model to learn the mapping relationship between key resonance biometric feature and user identity. Obtain the second password information entered by the logged-in user, match the second password information with the first password information, if the match is successful, obtain the second key resonance biometric feature when the logged-in user enters the second password information, input the second key resonance biometric feature into the trained biometric authentication model, and obtain the authentication result.

[0007] Furthermore, when obtaining the first key resonance biometric feature when a registered user enters the first password information, and when obtaining the second key resonance biometric feature when a logged-in user enters the second password information, both include: Acquire the sound signal of each key press when entering password information, and record the timestamp of each key press to obtain the time domain signal of the sound signal of each key press; The time-domain signal of the sound signal of each key is denoised, and then the time-domain signal of the denoised sound signal of each key is Fourier transformed to obtain the frequency-domain signal of the sound signal of each key. The time-domain features of the sound signal of each key are combined with the frequency-domain features of the sound signal in the time domain to form the key resonance biometric data of each key.

[0008] Furthermore, when performing noise reduction on the time-domain signal of the sound signal for each key, the signal is smoothed using a Gaussian filter to eliminate high-frequency noise, and then wavelet transform is used to process the signal to remove unwanted high-frequency noise.

[0009] Furthermore, when training a biometric authentication model using the first button resonance biometric feature, the following is included: The password characters corresponding to the first key resonance biometric data are used as the labels of the first key resonance biometric data to obtain the data samples of the first key resonance biometric data. Then, adversarial examples are generated using a GAN network. The data samples are divided into a training set and a test set. Adversarial examples are added to the training set to obtain an expanded training set. The expanded training set is then used to train the biometric authentication model to enhance its generalization. The trained biometric authentication model is tested using a test set. If the biometric authentication model fails the test, it is trained again using an expanded training set, and then tested again using the test set until the biometric authentication model passes the test.

[0010] Furthermore, the biometric authentication model includes a parallel LSTM network and an FCN network. The input layer data of the biometric authentication model passes through multiple hidden layers and pooling layers of the FCN network in sequence to obtain the first output data. The input layer data of the biometric authentication model passes through the LSTM layer, attention mechanism and dropout layer of the LSTM network in sequence to obtain the second output data. The first output data and the second output data are fused and classified by the Sigmoid function or Softmax function at the output layer of the biometric authentication model.

[0011] Furthermore, after inputting the second button resonance biometric feature into the trained biometric authentication model and obtaining the authentication result, the process also includes: retraining the biometric authentication model with the authenticated second button resonance biometric feature and the first button resonance biometric feature to improve the accuracy of the biometric authentication model.

[0012] This invention also proposes a two-factor authentication system based on the biometric features of password and keystroke resonance, comprising a password input device and a server, wherein: The password input device is used to obtain the first password information entered by the registered user, and to register the first key resonance biometric feature when the user enters the first password information through the built-in microphone, and send the first password information and the first key resonance biometric feature to the server; The server is used to obtain and save the first password information and the first key resonance biometric feature, and then use the first key resonance biometric feature to train a biometric authentication model to learn the mapping relationship between the key resonance biometric feature and the user's identity. The password input device is also used to obtain the second password information entered by the logged-in user, and to obtain the second key resonance biometric feature when the logged-in user enters the second password information through the built-in microphone, and send the second password information and the second key resonance biometric feature to the server. The server is also used to obtain the second password information entered by the logged-in user, match the second password information with the first password information, and if the match is successful, obtain the second key resonance biometric feature when the logged-in user enters the second password information. The second key resonance biometric feature is then input into the trained biometric authentication model to obtain the authentication result.

[0013] The present invention also proposes a two-factor authentication device based on password and key resonant biometric features, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of any of the two-factor authentication methods based on password and key resonant biometric features. The present invention also proposes a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of any of the two-factor authentication methods based on password and key resonant biometric features. The present invention also proposes a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the two-factor authentication methods based on password and key resonant biometric features.

[0014] Compared with the prior art, the advantages of the present invention are as follows: This invention combines password verification and key resonance biometric verification. Even if the password is leaked, key resonance biometrics can still effectively prevent unauthorized login because everyone's body structure is different, and the resonance feedback between the fingers and the keyboard is also different.

[0015] This invention requires no additional plugins or external hardware, is compatible with all computer devices with built-in microphones, and supports browsers and password login. It is highly adaptable to power grid environments and can be easily deployed. This invention uses the built-in microphone of the password input device to collect the resonant sound signal generated when the user presses keys. It eliminates the need for user interaction with devices other than the password input device; authentication is completed solely based on the built-in microphone of the password input device, making the authentication process convenient and efficient. Attached Figure Description

[0016] Figure 1 This is a flowchart of a method according to an embodiment of the present invention.

[0017] Figure 2 This is a network framework diagram of the biometric authentication model in an embodiment of the present invention. Detailed Implementation

[0018] The present invention will be further described below with reference to the accompanying drawings and specific preferred embodiments, but this does not limit the scope of protection of the present invention.

[0019] Before introducing specific embodiments of the present invention, the relevant concepts or terms will be explained.

[0020] Key resonance biometrics: Key resonance biometrics are features extracted through spectrum analysis and time domain analysis based on the subtle sound waveform changes generated when a user presses a key, and are used to distinguish the biometrics of different users.

[0021] Example 1 This embodiment proposes a two-factor authentication method based on password and key resonance biometrics, enabling legitimate user authentication in a power grid environment based on key resonance biometrics. For example... Figure 1 As shown, there are two stages: the registration stage and the login stage. The registration process includes the following steps: S101) Registered users enter a valid password, referred to as the first password in this embodiment for distinction; S102) Obtain the first password information entered by the registered user, and obtain the key resonance biometric feature when the registered user enters the first password information, which is referred to as the first key resonance biometric feature in this embodiment for differentiation; S103) Use the corresponding password characters as the label for the first key resonance biometric feature to obtain data samples; S104) Uses a GAN network to generate adversarial examples; S105) Combine adversarial examples with data samples of the first key resonance biometric feature to train a biometric authentication model in order to learn the mapping relationship between key resonance biometric feature and user identity. The login phase includes the following steps: S201) Login user enters password, which is referred to as the second password in this embodiment for distinction; S202) Obtain the second password information entered by the logged-in user, match the second password information with the first password information, and if the match fails, refuse the login user; S203) If the match is successful, obtain the key resonance biometric feature when the login user enters the second password information, which is referred to as the second key resonance biometric feature in this embodiment for differentiation. Input the second key resonance biometric feature into the trained biometric authentication model to obtain the authentication result. If the authentication result is successful, the login user is allowed to log in; if the authentication result is unsuccessful, the login user is denied to log in.

[0022] The steps are explained below.

[0023] In step S101 of this embodiment, when the registered user enters a valid password, specifically, the registered user enters a valid first password on the computer's registration interface.

[0024] In step S102 of this embodiment, during the user's input of the first password, not only is the information of each key pressed by the user recorded, but the computer's built-in microphone also collects the sound signals generated during the key presses. These sound signals are formed by the resonance effect generated when the fingers press the keyboard and the interaction with the air. Each time a key is pressed, a timestamp is also recorded. Based on these timestamps, time-domain and frequency-domain signals of the sound signal are generated to extract key resonance biometric features. Specifically, when obtaining the first key resonance feature when a registered user inputs their first password information, this includes: S1021) Obtain the sound signal of each key press when inputting password information, and record the timestamp of each key press to obtain the time domain signal of the sound signal of each key press; S1022) Noise reduction processing is performed on the time-domain signal of the sound signal for each key. In this embodiment, this specifically includes: A Gaussian filter is used to smooth the signal and eliminate high-frequency noise. The formula for a Gaussian filter is:

[0025] in, It is the signal at time i in the input signal. It is a Gaussian function; Then, wavelet transform is used to process the signal, removing unwanted high-frequency noise. Wavelet transform can effectively remove unwanted high-frequency noise. After wavelet decomposition, appropriate low-frequency coefficients are selected to filter out high-frequency noise, thereby preserving useful signal features. (S1023) Perform a Fourier transform on the time-domain signal of the sound signal for each key after noise reduction to obtain the frequency-domain signal of each key's sound signal. In this embodiment, the time-domain sound signal is converted into a frequency-domain signal using Fourier transform to extract the frequency components. These frequency components contain the characteristics of the resonance between the key and the finger, expressed as follows:

[0026] in, It is the original signal. It is the corresponding frequency domain signal. It is frequency; S1024) Combine the time-domain features of the sound signal in the time domain with the frequency-domain features in the frequency domain for each key, and use the combined features (such as frequency, amplitude, duration, etc.) as the key resonance biometric data for each key. In step S103 of this embodiment, the corresponding password character is used as the label of the first key resonance biometric data, so that the first key resonance biometric data and the corresponding label form a data sample corresponding to the registered user.

[0027] In this embodiment, the key resonance biometric data and password information of the registration stage are obtained through steps S102 and S103. Before step S104, the key resonance biometric data and password information of the registration stage need to be stored together as the reference data for subsequent identity verification.

[0028] In step S104 of this embodiment, the GAN network consists of a generator and a discriminator. The generator generates samples that simulate the biometric features of legitimate users' key presses based on random noise. The discriminator distinguishes between generated samples and real samples through a deep convolutional network. During training, the generator continuously optimizes to generate realistic adversarial samples, while the discriminator improves its ability to distinguish between real and fake samples. The final generated adversarial samples are used to expand the dataset.

[0029] In step S105 of this embodiment, based on a certain amount of data samples obtained in steps S102 and S103, a deep learning model is trained using the data. During training, the model learns the mapping relationship between the key resonance biometric features extracted from the sound signal and the user's identity, thereby identifying the personalized resonance features when the user inputs. Furthermore, to enhance the model's generalization ability, during the training phase, the GAN network from step S104 is used to generate adversarial examples to jointly train the model with actual samples, thereby enhancing the model's generalization ability and its ability to identify unauthorized users.

[0030] In this embodiment, the deep learning model uses the LSTM+FCN biometric authentication model, such as... Figure 2As shown, the model includes a parallel LSTM network and an FCN network, responsible for extracting information from the biometric features of the input password keystrokes and classifying users as legitimate or illegitimate. Specifically: The FCN network consists of multiple hidden layers and one pooling layer. The input data of the biometric authentication model passes through multiple hidden layers and pooling layers of the FCN network in sequence to obtain the first output data. During this process, the hidden layers process the input features through non-linear activation functions (such as ReLU) to learn the subtle patterns of user input, helping the network to distinguish the feature differences between legitimate and illegitimate users. The pooling layer maps the features output by the hidden layers to a fixed-length vector through global max pooling, which can effectively capture global information.

[0031] The LSTM network consists of LSTM layers, an attention mechanism, and a Dropout layer. The input data of the biometric authentication model passes through the LSTM layer, attention mechanism, and Dropout layer of the LSTM network in sequence to obtain the second output data. In this process, the hidden state of the input data at each time step is calculated by the LSTM layer, and then the hidden state of each time step is reweighted by the attention mechanism to focus on the important time step features. Finally, the Dropout layer applies dropout to the output of the attention mechanism to prevent overfitting.

[0032] After the first and second output data are fused, the user is classified as legitimate or illegitimate in the output layer of the biometric authentication model using binary classification (using Sigmoid or Softmax). The output is a probability value indicating the legitimacy of the user's identity.

[0033] The LSTM+FCN biometric authentication model combines the feature extraction processes of LSTM and FCN to capture temporal dependencies and local features of data, thereby enhancing the model's feature recognition capabilities. The introduction of attention mechanisms and dropout further improves the model's attention and generalization abilities. In step S105 of this embodiment, when training the biometric authentication model by combining adversarial examples and data samples of the first key resonance biometric feature, the following steps are included: The data samples of the first resonance feature are divided into a training set and a test set. In this embodiment, the training set and the test set are divided according to historical login data. The training set is used to train the biometric authentication model, and the test set is used to verify the accuracy of the biometric authentication model. The adversarial examples generated by the GAN network are obtained and added to the training set to obtain an expanded training set. The expanded training set is then used to train the biometric authentication model to enhance its generalization. During the training of the biometric authentication model using the expanded training set, the generator of the GAN network is updated in each training cycle to generate more diverse adversarial examples. The expanded training set is then updated with the adversarial examples generated by the updated generator. The trained biometric authentication model is tested using a test set. If the biometric authentication model fails the test, it is trained again using an expanded training set, and then tested again using the test set until the biometric authentication model passes the test.

[0034] In this embodiment, when training the biometric authentication model by combining adversarial examples and data samples of the first key resonance biometric feature, the goal of model training is to minimize the loss function, a typical loss function being the binary cross-cutting loss:

[0035] in, It is the probability predicted by the model. These are actual labels (1 represents a legitimate user, and 0 represents an illegitimate user). After training, a validated biometric authentication model is obtained, capable of distinguishing between legitimate and illegitimate users.

[0036] In step S201 of this embodiment, when the login user enters a password, specifically, the login user enters a second password on the computer's login interface.

[0037] In step S202 of this embodiment, during the user's input of the second password, not only is the information of each key pressed by the user recorded, but the computer's built-in microphone also captures the sound signals generated during the key presses. Each time a key is pressed, a timestamp is also recorded. .

[0038] In step S203 of this embodiment, the key resonance biometric features obtained when the login user enters the second password information are the same as those obtained when the registered user enters the first password information. First, the time domain signal of the sound signal is generated according to these timestamps. The time domain signal of the sound signal is preprocessed (Gaussian filtering, wavelet noise reduction). Then, the frequency domain signal is obtained through Fourier transform. Finally, the key resonance biometric features are extracted by combining the time domain features in the time domain signal and the frequency domain features in the frequency domain signal.

[0039] In step S203 of this embodiment, when the second key resonance biometric feature is input into the trained biometric authentication model, the biometric authentication model calculates a predicted probability of a legitimate user based on the input key resonance biometric feature data. If the classification result is a legitimate user (i.e., the output is 1), the user is allowed to log in; otherwise, if the user is an illegitimate user (i.e., the output is 0), the login is rejected and the operation is recorded.

[0040] To improve the authentication accuracy of the biometric authentication model, in this embodiment, after step S203, the method further includes: obtaining the second key resonance biometric feature that has passed authentication and is classified as a legitimate user; combining the obtained second key resonance biometric feature with the first key resonance biometric feature to form a new data sample; and retraining the biometric authentication model. Similarly, adversarial examples generated using a GAN network are added during the training process to enhance the generalization of the model.

[0041] Example 2 This embodiment proposes a two-factor authentication system based on the biometric features of password and keystroke resonance, including a password input device and a server, wherein: The password input device is used to obtain the first password information entered by the registered user, and to register the first key resonance biometric feature when the user enters the first password information through the built-in microphone, and send the first password information and the first key resonance biometric feature to the server; The server is used to obtain and save the first password information and the first key resonance biometric feature, and then use the first key resonance biometric feature to train a biometric authentication model to learn the mapping relationship between the key resonance biometric feature and the user's identity. The password input device is also used to obtain the second password information entered by the logged-in user, and to obtain the second key resonance biometric feature when the logged-in user enters the second password information through the built-in microphone, and send the second password information and the second key resonance biometric feature to the server. The server also obtains the second password information and then matches it with the first password information. If the match fails, the user is denied login. If the match succeeds, the server obtains the second key resonance biometric feature and inputs it into the trained biometric authentication model to obtain the authentication result. If the authentication result is successful, the user is allowed to log in. If the authentication result is unsuccessful, the user is denied login.

[0042] Example 3 This embodiment proposes a two-factor authentication device based on password and key resonant biometric features, including a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the steps of the two-factor authentication method based on password and key resonant biometric features described in Embodiment 1. This embodiment also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the two-factor authentication method based on password and key resonant biometric features as described in Embodiment 1. This embodiment also proposes a computer program product, including a computer program that, when executed by a processor, implements the steps of the two-factor authentication method based on password and key resonant biometric features as described in Embodiment 1.

[0043] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A two-factor identity authentication method based on password and key-press resonance biometrics, characterized in that, Includes the following steps: The system obtains the first password information entered by the registered user and the first key resonance biometric feature when the registered user enters the first password information. The first key resonance biometric feature is the feature obtained by combining the time-domain feature of the sound signal of each key press when the first password information is entered with the frequency-domain feature of the sound signal. The time-domain signal is the timestamp of the key press. The password character corresponding to the first key resonance biometric feature data is used as the label of the first key resonance biometric feature data. The first key resonance biometric feature is used to train a biometric authentication model to learn the mapping relationship between key resonance biometric features and user identity. The biometric authentication model includes a parallel LSTM network and an FCN network. The key resonance biometric feature data of each key press is input into the biometric authentication model in a time sequence according to the timestamp order. The input layer data of the biometric authentication model passes through multiple hidden layers and pooling layers of the FCN network in sequence to obtain the first output data. The pooling layer uses global max pooling to map the features output by the hidden layer to a fixed-length vector, which can effectively capture global information. The input layer data of the biometric authentication model passes through an LSTM network in sequence in another path. After the LSTM layer, attention mechanism, and Dropout layer of the network, the second output data is obtained. The attention mechanism reweights the hidden state at each time step to focus on the important time step features. The first and second output data are fused and classified by the Sigmoid function or Softmax function at the output layer of the biometric authentication model. When training a biometric authentication model using the first button resonance biometric feature, the following are included: The password character corresponding to the first key resonance biometric data is used as the label of the first key resonance biometric data to obtain the data sample of the first key resonance biometric data, and then an adversarial sample is generated using a GAN network. The data samples are divided into a training set and a test set. Adversarial examples are added to the training set to obtain an expanded training set. The expanded training set is then used to train the biometric authentication model to enhance its generalization. The trained biometric authentication model is tested using a test set. If the biometric authentication model fails the test, the model is trained again using an expanded training set, and the trained model is tested again using the test set until the biometric authentication model passes the test. The system retrieves the second password information entered by the logged-in user and matches it with the first password information. If the match is successful, it retrieves the second key resonance biometric feature when the logged-in user enters the second password information. The second key resonance biometric feature is a combination of the time-domain features and frequency-domain features of the sound signal of each key pressed when the second password information is entered. The second key resonance biometric feature is input into the trained biometric authentication model to obtain the authentication result. The biometric authentication model is then retrained using the authenticated second key resonance biometric feature and the first key resonance biometric feature to improve the accuracy of the biometric authentication model. Furthermore, adversarial examples generated using a GAN network are added during the retraining process to enhance the model's generalization ability.

2. The two-factor authentication method based on password and key resonant biometric features according to claim 1, characterized in that, When acquiring the first key resonance biometric feature when a registered user enters their first password information, and when acquiring the second key resonance biometric feature when a logged-in user enters their second password information, both include: Acquire the sound signal of each key press when entering password information, and record the timestamp of each key press to obtain the time domain signal of the sound signal of each key press; The time-domain signal of the sound signal of each key is denoised, and then the time-domain signal of the denoised sound signal of each key is Fourier transformed to obtain the frequency-domain signal of the sound signal of each key. The time-domain features of the sound signal of each key are combined with the frequency-domain features of the sound signal in the time domain to form the key resonance biometric data of each key.

3. The two-factor authentication method based on password and key resonant biometric features according to claim 2, characterized in that, When performing noise reduction on the time-domain signal of the sound signal for each key, the signal is smoothed using a Gaussian filter to eliminate high-frequency noise, and then wavelet transform is used to process the signal to remove unwanted high-frequency noise.

4. A two-factor authentication system based on the biometric features of password and keystroke resonance, characterized in that, Includes password input devices and servers, among which: The password input device is used to obtain the first password information entered by the registered user, and to register the first key resonance biometric feature when the user enters the first password information through the built-in microphone. The first key resonance biometric feature is the feature obtained by combining the time domain feature of the sound signal of each key when the first password information is entered with the frequency domain feature of the sound signal in the time domain signal. The password character corresponding to the first key resonance biometric feature data is used as the tag of the first key resonance biometric feature data, and the first password information and the first key resonance biometric feature are sent to the server. After the server acquires and saves the first password information and the first key resonance biometric feature, it uses the first key resonance biometric feature to train a biometric authentication model to learn the mapping relationship between the key resonance biometric feature and the user's identity. The biometric authentication model includes a parallel LSTM network and an FCN network. The input layer data of the biometric authentication model passes through multiple hidden layers and pooling layers of the FCN network in sequence to obtain the first output data. The pooling layer uses global max pooling to map the features output by the hidden layer to a fixed-length vector, which can effectively capture global information. The input layer data of the biometric authentication model passes through the LSTM layer, attention mechanism and dropout layer of the LSTM network in sequence to obtain the second output data. The attention mechanism reweights the hidden state at each time step to focus on the features of important time steps. After the first output data and the second output data are fused, they are classified in the output layer of the biometric authentication model using the Sigmoid function or the Softmax function. The password input device is also used to obtain the second password information entered by the logged-in user, and to obtain the second key resonance biometric feature when the logged-in user enters the second password information through the built-in microphone, and send the second password information and the second key resonance biometric feature to the server. The server is also used to obtain the second password information entered by the logged-in user, match the second password information with the first password information, and if the match is successful, obtain the second key resonance biometric feature when the logged-in user enters the second password information. The second key resonance biometric feature is the feature obtained by combining the time domain feature of the sound signal of each key in the time domain signal and the frequency domain feature in the frequency domain signal when the second password information is entered. The second key resonance biometric feature is input into the trained biometric authentication model to obtain the authentication result.

5. A two-factor authentication device based on password and key resonant biometric features, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the two-factor authentication method based on password and key resonant biometric features as described in any one of claims 1 to 3.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the two-factor authentication method based on password and key resonant biometric features as described in any one of claims 1 to 3.

7. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the two-factor authentication method based on password and key resonant biometric features as described in any one of claims 1 to 3.

Images