Authentication method and system for identifying user identity based on brain wave characteristics

Abstract

The invention relates to the technical field of feature recognition, in particular to an authentication method and system for recognizing user identity based on brain wave features. The method comprises the following steps that EEG equipment is used for collecting electroencephalogram signals of a user under a preset cognitive task, and sample encryption is conducted on the electroencephalogram signals through encrypted data; noise baseline calibration is conducted on the electroencephalogram signals, standard electroencephalogram signals are generated, and electroencephalogram sample AES encryption processing is conducted; performing corresponding adversarial sample generation on the standard electroencephalogram signals through an adversarial generative network; performing electroencephalogram confusion feature injection on the standard electroencephalogram signals by applying a space-time dynamic masking strategy to obtain electroencephalogram confusion injection features; and performing adversarial dynamic feature enhancement training on the electroencephalogram confusion injection features by using the generated corresponding adversarial samples, and generating an adversarial test result. According to the invention, through noise calibration, adversarial sample generation, feature enhancement training and block chain technologies, the accuracy, anti-interference capability and security of the identity authentication system based on brain waves are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of feature recognition, and particularly to an authentication method and system for identifying user identity based on electroencephalogram features. Background Art

[0002] Electroencephalogram (EEG), as a bioelectrical signal directly reflecting brain activities, has uniqueness and stability, and is thus considered a potential personal identity authentication tool. Early identity authentication methods mainly relied on traditional biometric features such as passwords, fingerprints, irises, or facial recognition. However, these methods face certain challenges in terms of security and privacy protection. With the progress of EEG technology, researchers have begun to explore how to use the unique characteristics of EEG for identity verification. The research on EEG recognition technology initially focused on the acquisition and processing technology of EEG signals. With the gradual popularization of electroencephalogram (EEG) devices and the improvement of computer processing capabilities, researchers have gradually recognized the differences and uniqueness of EEG among individuals. However, there are significant differences in EEG among different individuals at present, which makes traditional EEG-based identity authentication methods difficult to apply to a large number of users. At the same time, EEG recognition systems are easily affected by external interference, resulting in unstable authentication processes. Summary of the Invention

[0003] Based on this, it is necessary to provide an authentication method and system for identifying user identity based on electroencephalogram features to solve at least one of the above technical problems.

[0004] To achieve the above object, an authentication method for identifying user identity based on electroencephalogram features, the method includes the following steps:

[0005] Step S1: Use an EEG device to collect the EEG signals of a user under a preset cognitive task, perform noise baseline calibration on the EEG signals, generate standard EEG signals, and perform AES encryption processing on the EEG samples.

[0006] Step S2: Generate corresponding adversarial samples for the standard EEG signals through an adversarial generation network; apply a spatio-temporal dynamic masking strategy to inject EEG chaotic features into the standard EEG signals to obtain EEG chaotic injection features; use the generated corresponding adversarial samples to perform adversarial dynamic feature enhancement training on the EEG chaotic injection features to generate an adversarial test result.

[0007] Step S3: Extract the time-frequency features and functional connection matrices of the adversarial test results, and construct an anti-interference user feature template library.

[0008] Step S4: Obtain the electroencephalogram (EEG) signals of the user's multi-session data, and import the EEG signals of the user's multi-session data into the anti-interference user feature template library for identity determination. If the identity determination result is true, upload the corresponding EEG signals of the user's multi-session data to the blockchain for user biometric binding to generate a user identity key.

[0009] Through the combination of a generative adversarial network (GAN) and a spatio-temporal dynamic masking strategy, the present invention can effectively inject EEG chaotic features, improve the system's ability to identify interference or forged signals, and enhance the defense ability against adversarial attacks. By performing identity determination on multi-session data, the user's identity can be confirmed at different times and in different situations, which can effectively avoid verification errors caused by different environments or devices, and improve the stability and reliability of identity verification. The extraction of time-frequency features and functional connectivity matrices helps to capture the deep patterns in EEG signals and improve the accuracy of identity recognition. In particular, the functional connectivity matrix can reflect the interactions between different regions of the brain and further strengthen the individual's biometric characteristics. Uploading the user's EEG signal data to the blockchain can ensure the immutability and privacy of the data. This decentralized storage method can prevent data leakage or tampering and enhance the security of the entire system. Through biometric binding, combined with the user's unique EEG signal information, more accurate identity authentication can be achieved, reducing the security risks of traditional verification methods such as passwords or fingerprints. By using multi-session data, the system can adapt to the user's different physiological states, enhancing the ability of personalized authentication, which is particularly effective for long-term monitoring or identity recognition in dynamic environments. Therefore, through noise calibration, adversarial sample generation, feature enhancement training, and blockchain technology, the present invention improves the accuracy, anti-interference ability, and security of the EEG-based identity authentication system.

[0010] Preferably, step S1 includes the following steps:

[0011] Step S11: Use an EEG device to collect the user's EEG signals to obtain EEG signals;

[0012] Step S12: Based on a synchronous acquisition device, collect eye movement signals and electromyogram signals synchronized with the EEG signals;

[0013] Step S13: Design a preset cognitive task, where the preset cognitive task includes a visual stimulus-induced P300 potential, a multi-level mental arithmetic task, and a cross-modal association task;

[0014] Step S14: Identify the interference signal sources of the EEG signals according to the eye movement signals and electromyogram signals to generate noise source identification data; perform noise baseline calibration on the noise source identification data to generate noise baseline calibration data;

[0015] Step S15: Optimize the signal quality of the EEG signals using the noise baseline calibration data, generate standard EEG signals, and perform AES encryption processing on the EEG samples.

[0016] In the present invention, by synchronously collecting eye movement signals and electromyography signals, other physiological activities related to EEG signals can be monitored simultaneously, which helps to accurately distinguish which signals are valid EEG signals and which are noises introduced by factors such as eye movement or muscle activity, thereby effectively improving the quality of EEG signals. By identifying the interference sources of eye movement signals and electromyography signals, the influence of these noise sources can be accurately identified and removed from the EEG signals. The generation of noise source identification data and noise baseline calibration data can provide a basis for further signal optimization, thereby improving the accuracy and reliability of signal processing. Designing preset cognitive tasks (such as visual stimulation-induced P300 potential, multi-level mental arithmetic tasks, and cross-modal association tasks) can comprehensively evaluate the brain activities of users under different cognitive tasks. This diverse task design helps to obtain richer EEG signal features and can better adapt to various application scenarios, especially in situations where the cognitive state changes significantly. Through noise baseline calibration and signal quality optimization, it is ensured that the EEG signals collected under various interference factors have higher stability and accuracy. This optimization process can effectively remove interference and improve the effect of subsequent analysis (such as feature extraction and identity recognition). Since the preset cognitive tasks include different types of tasks such as vision, calculation, and association, these tasks can trigger different EEG responses, such as the P300 potential. Through these stimuli and tasks, the EEG signals can be made more discriminative at the cognitive level, further enhancing the effect of signal analysis.

[0017] Preferably, the generation of corresponding adversarial samples for the standard EEG signals in step S2 includes:

[0018] Adding low-frequency drift component noise to the standard EEG signals to obtain low-frequency drift component noise; simulating the drift amplitude enhancement of the standard EEG signals based on the low-frequency drift component noise to generate EEG signal data with drift noise and labeling it as the first type of sample;

[0019] Randomly adding high-frequency noise components to the standard EEG signals to obtain high-frequency random component noise; simulating environmental interference of the standard EEG signals based on the high-frequency random component noise to generate EEG signal data with environmental interference noise and labeling it as the second type of sample;

[0020] Using adversarial generation network technology to embed forged brain electrical activities into the standard EEG signals to generate forged brain electrical feature signals; simulating spoofing attacks on the standard EEG signals through the forged brain electrical feature signals to generate EEG signal data with spoofing attacks and labeling it as the third type of sample;

[0021] Integrate the first type of samples, the second type of samples, and the third type of samples into a comprehensive adversarial sample.

[0022] Through a variety of noise injections for low-frequency drift, environmental interference, high-frequency noise, and spoofing attacks, the present invention can simulate various interference situations in the real world during the training process. This can make the model more adaptable to signal inputs in complex environments, significantly improving the robustness and anti-interference ability of the system. By combining a variety of noise and attack patterns (low-frequency drift, environmental interference, and spoofing attacks), multi-level and multi-dimensional interference simulation can be achieved. This diverse training data can help the model learn more features, thereby improving the model's ability to identify various types of attacks and noise. By embedding fake electroencephalogram (EEG) activities into standard EEG signals to generate spoofing attack samples, the model can learn how to identify fake EEG signals during the training process, which is very important for future authentication systems or intelligent monitoring systems and can effectively prevent fraud through fake EEG signals. The addition of low-frequency drift and high-frequency noise simulates physiological signal perturbations encountered in real life (such as eye movements, muscle activities, environmental noise, etc.). These perturbations usually affect the quality of EEG signals. Using this adversarial training method, these interferences can be identified and filtered in advance, improving the accuracy and stability of EEG signals. By integrating the samples generated from low-frequency drift, environmental interference, and spoofing attacks to create a diverse set of adversarial samples, not only can the model's adaptability to different types of interference be improved, but also the generalization ability of the model can be enhanced, enabling it to maintain a high recognition accuracy when facing unknown or complex environments. This method can expand the diversity of training data by generating different categories of adversarial samples, ensuring that the model can still effectively process EEG signals under a variety of different noise and attack patterns, achieving a higher accuracy. Especially when applied to EEG signal recognition and analysis, the model can adapt to different types of physiological signals and noise, improving the overall performance.

[0023] Preferably, the injection of EEG chaos features into the standard EEG signal by applying the spatio-temporal dynamic masking strategy in step S2 includes:

[0024] Randomly select several frequency bands from the standard EEG signal, and randomly discard segments of the selected frequency bands to generate randomly discarded segment data; based on the randomly discarded segment data, perform temporal gap interpolation on the standard EEG signal to generate a reorganized EEG signal;

[0025] Randomly select electrode channels from the standard EEG signal, and perform random channel masking on the standard EEG signals of the corresponding electrode channels to generate channel masking data; based on the channel masking data, set the signal values of the standard EEG signal to zero to generate an EEG masked signal;

[0026] Perform spectral analysis on the standard electroencephalogram (EEG) signal, extract the frequency bands of interest, and inject proportional noise to generate a noise-injected signal; analyze the abnormal signal characteristics of the EEG reconstructed signal, EEG shielded signal, and noise-injected signal, and integrate them into a spatio-temporal dynamic masking strategy;

[0027] Based on the spatio-temporal dynamic masking strategy, use the abnormal signal characteristics to inject EEG chaos characteristics into the standard EEG signal to obtain EEG chaos injection characteristics.

[0028] By introducing a spatio-temporal dynamic masking strategy into the standard EEG signal, the present invention can better simulate various signal interference situations in real life, such as electrode failure, noise pollution, etc. This diverse interference injection can significantly enhance the robustness of the model to various types of interference signals, making the model more adaptable in practical applications. Operations such as randomly discarding frequency band segments and randomly shielding electrode channels simulate signal loss or interference situations encountered during EEG signal acquisition. In this way, the system can train a model with a high tolerance for signal loss and interference, improving the stability and reliability during signal processing. Based on the EEG reconstructed signal, shielded signal, and noise-injected signal, extracting abnormal signal characteristics and integrating them into a spatio-temporal dynamic masking strategy helps train the model to identify abnormal signals, which is crucial for detecting signal anomalies caused by noise, hardware failures, or other external interferences, and can improve the system's anomaly recognition ability. Spectral analysis and noise injection can simulate noise perturbations in various frequency ranges, enabling the model to handle interference in different frequency bands. By introducing this multi-band noise, the model can better adapt to the noise in different environments and signal sources, thereby improving the accuracy and stability of EEG signal processing. Techniques such as temporal gap interpolation and signal value zeroing help simulate the process of signal loss and recovery. Through this training, the model can repair or recover the lost or damaged parts of the signal in practical applications, ensuring the integrity of the signal during transmission and processing. By combining spectral analysis and spatio-temporal dynamic masking strategy, the model can handle various forms of signal interference, especially in terms of the spatio-temporal dynamic characteristics of the signal. This strategy not only increases the data diversity of adversarial training but also helps the model more effectively distinguish between valid EEG signals and interference signals in practical applications, improving the overall performance.

[0029] Preferably, the adversarial dynamic feature enhancement training of the EEG chaos injection characteristics using the generated corresponding adversarial samples in step S2 includes:

[0030] Perform EEG device distribution difference analysis on the EEG chaos injection characteristics to generate device distribution difference data; use the device distribution difference data to perform cross-device distribution alignment training on the generated corresponding adversarial samples to generate cross-device aligned signal feature data;

[0031] Construct a live detection scenario; perform live detection adversarial training on the generated corresponding adversarial samples through the live detection scenario to generate enhanced live detection feature data;

[0032] Integrate the signal feature data after cross-device alignment, the generated corresponding adversarial samples, and the enhanced live detection feature data into a joint training set; perform adversarial testing on the joint training set according to the adversarial generation network to generate an adversarial test result.

[0033] Through the analysis of the distribution differences of EEG devices, the generated device distribution difference data can help the model understand the signal differences between different devices. Through cross-device distribution alignment training, the model can adapt to the acquisition differences of different EEG devices, improve the accuracy and stability of cross-device applications. In this way, the EEG signals collected on different hardware platforms can be uniformly processed, improving the generalization ability of the model under multiple devices. Using the generated corresponding adversarial samples for cross-device distribution alignment training not only improves the cross-device adaptability of the model, but also enhances the robustness of the model through diverse training data. These adversarial samples provide more challenging training scenarios, enabling the model to effectively handle EEG signals from different sources and of different qualities, enhancing the diversity and effectiveness of adversarial training. By constructing a live detection scenario and performing live detection adversarial training on the generated corresponding adversarial samples, the model can more sensitively identify forged or untrue EEG signals. This training not only enhances the model's ability to identify forged signals, but also effectively improves the live detection performance of the model in a real environment, preventing the influence of external interference or fraud. Integrating the signal feature data after cross-device alignment, the generated corresponding adversarial samples, and the enhanced live detection feature data to form a joint training set. This joint training set covers various aspects of features, enabling the model to learn and analyze EEG signals from different perspectives, ultimately improving the comprehensive detection and analysis ability of EEG signals. The comprehensive feature data enables the model to have stronger adaptability when facing complex signal chaos, device differences, and live recognition tasks. By using the adversarial generation network to perform adversarial testing on the joint training set, the model can be tested and optimized under various adversarial samples. This step can effectively evaluate and improve the performance of the model in practical applications, enabling it to maintain efficient recognition and processing capabilities in complex, dynamic, and changing environments. Through adversarial testing, the model will become more robust during continuous optimization, especially in dealing with interference and forged signals of EEG signals.

[0034] Preferably, step S3 includes the following steps:

[0035] Step S31: Perform time-frequency analysis on the adversarial test result to generate instantaneous EEG signal feature data;

[0036] Step S32: Perform short-time Fourier transform on the instantaneous feature data of the EEG signal to generate the instantaneous spectrum data of the EEG signal; extract the spectral features of the instantaneous spectrum data of the EEG signal to obtain the instantaneous spectrum feature data of the EEG signal;

[0037] Step S33: Calculate the phase synchrony based on the instantaneous feature data of the EEG signal and the instantaneous spectrum feature data of the EEG signal to obtain the phase synchronization data of brain regions; construct a functional connectivity matrix for the adversarial test results based on the phase synchronization data of brain regions, and extract the anti-interference features of the functional connectivity matrix;

[0038] Step S34: Construct an anti-interference user feature template library for the adversarial samples through the anti-interference features to generate an anti-interference user feature template library.

[0039] Through time-frequency analysis of the adversarial test results, the present invention generates instantaneous feature data of electroencephalogram (EEG) signals, effectively extracting the time and frequency domain information of the signals. The extraction of instantaneous features provides a fine signal representation for subsequent signal analysis, enabling effective discrimination of different types of EEG signal features, especially subtle changes under complex interference conditions. By extracting instantaneous spectrum data through short-time Fourier transform, the analysis ability of spectral features is further enhanced. The accurate extraction of such spectral features enables the system to more sensitively identify subtle changes in EEG signals, providing a solid foundation for subsequent feature fusion and signal interpretation. By calculating the phase synchrony of EEG signals and extracting inter-brain region phase synchrony data, the cooperative activities between brain regions can be captured. This phase synchrony analysis helps to deeply understand the functional connections between brain regions, providing important time-frequency domain connections for the interpretation of EEG signals. Based on the phase synchrony data, a functional connection matrix is constructed and anti-interference features are extracted, further strengthening the ability to identify subtle changes in the signals. These anti-interference features help to optimize the accuracy of EEG signals from multiple dimensions and levels, enhancing the anti-interference ability in complex and noisy environments. Through the extracted anti-interference features, an anti-interference user feature template library is constructed. The generated template library can accurately model the user's EEG signals. These template libraries not only enhance the user's identity recognition ability but also effectively handle complex situations such as interference signals, forgery attacks, and device differences, making the user identity recognition system more robust and adaptable in various interference environments. The constructed anti-interference user feature template library has significant personalized features for the recognition of different users' EEG signals. The user's EEG signals can be accurately matched and recognized through the features in the template library, thereby improving the performance in applications such as identity verification and emotion recognition. The extraction of anti-interference features not only helps to construct an efficient user feature template but also improves the overall anti-interference performance of the system. In the face of various interference factors such as noise, forged signals, and device inconsistencies, these anti-interference features can help the system better filter out irrelevant information, enhancing the security and reliability of the system. The construction of the functional connection matrix through multi-dimensional signal feature analysis enables the system to more comprehensively understand the dynamic changes of EEG signals, increasing the system's adaptability to complex EEG signal environments, especially having stronger protection ability in the face of high-complexity interference situations.

[0040] Preferably, step S33 includes the following steps:

[0041] Step S331: Screen the adversarial test results for instantaneous change EEG signals based on the instantaneous feature data of EEG signals and the instantaneous spectrum feature data of EEG signals to obtain instantaneous change EEG signals;

[0042] Step S332: Calculate the inter-brain region phase synchrony of the instantaneous change EEG signals to obtain the inter-brain region phase difference;

[0043] Step S333: Analyze the phase synchronization change of the instantaneous changing EEG signals across time windows through the inter-brain region phase difference to generate brain region phase synchronization data;

[0044] Step S334: Construct a functional connectivity matrix based on the phase synchronization change data, and perform dynamic functional connectivity analysis on the brain region phase synchronization data using the functional connectivity matrix to generate dynamic functional connectivity data;

[0045] Step S335: Extract the stable phase features from the functional connectivity matrix using the dynamic functional connectivity data to obtain the anti-interference features of the functional connectivity matrix.

[0046] Through the screening of the instantaneous changing EEG signals, the present invention can effectively extract the short-term dynamic features of the signals. The extraction of such dynamic features helps the system to identify the sudden changes or instabilities in the signals, which is conducive to the detailed analysis of the signals under different EEG activity states and enhances the understanding of the EEG activities. The screening of the instantaneous changing EEG signals helps to filter out the noise data that do not conform to the instantaneous change characteristics, enabling the subsequent analysis to focus on the real EEG activities and reducing the influence of noise on the results. Through the calculation of the inter-brain region phase synchronization, the cooperative effect and information transmission between different brain regions can be deeply analyzed. Phase synchronization is an important indicator for understanding the brain functional connectivity. By calculating the phase difference between brain regions, the system can identify the dynamic synchronization changes between brain regions, further revealing the cooperative working mode of different brain regions under specific tasks or states. The phase synchronization analysis enhances the understanding of the spatio-temporal changes of the EEG signals. Especially in the confrontation test results, it can identify the activity patterns of different brain regions and the relationships between them, providing a reliable basis for the subsequent functional analysis. Analyzing the phase synchronization change of the instantaneous changing EEG signals across time windows to generate brain region phase synchronization data can effectively capture the changes of the signals on different time scales. This temporal dynamic capture ability can help the system to track the change rules of the EEG signals at different time points, and then identify the real-time evolution of the brain activities. The division of time windows can identify the short-term and long-term synchronization changes, providing more detailed temporal data for the dynamic analysis, which is conducive to further optimizing the identification and feature extraction of interference signals. The functional connectivity matrix constructed based on the phase synchronization change data can reflect the functional connectivity pattern between brain regions. This matrix is the core for understanding the cooperative work of different brain regions in the brain, and can effectively reveal the information flow and connection state between brain regions. The construction of the functional connectivity matrix provides a basis for the subsequent dynamic functional connectivity analysis, enabling the system to analyze the EEG signals from a global perspective and enhancing the accurate understanding of the EEG activity patterns.

[0047] Preferably, step S4 includes the following steps:

[0048] Step S41: Obtain the EEG signals of the user's multi-session data;

[0049] Step S42: Import the electroencephalogram (EEG) signals of the user's multi-session data into the anti-interference user feature template library for similarity discrimination, and generate a similarity discrimination result;

[0050] Step S43: Compare the similarity discrimination result with a preset standard similarity threshold. When the similarity discrimination result is greater than or equal to the preset standard similarity threshold, it is determined that the identity determination result is true, and the EEG signals of the corresponding user's multi-session data are uploaded to the blockchain for user biometric binding to generate a user identity key;

[0051] Step S44: When the similarity discrimination result is less than the preset standard similarity threshold, it is determined that the identity determination result is false, and an identity verification failure prompt is generated.

[0052] Through the similarity discrimination of the EEG signals of the user's multi-session data, the system can effectively judge the accuracy of the user's identity. As a biometric, EEG signals have strong uniqueness and stability, which can prevent problems such as impersonation and identity theft, ensuring high-precision identity determination. By comparing the similarity discrimination result with the standard threshold, the system effectively reduces the situations of misjudgment and missed judgment, further improving the accuracy and reliability of identity verification. Uploading the EEG signals of the user's multi-session data that pass the identity verification to the blockchain for biometric binding can ensure the immutability and security of user data. The decentralized feature of the blockchain provides extremely high security protection for the user's biometric data, preventing data leakage, forgery, and unauthorized access. The application of the blockchain makes the generation and storage process of the user identity key more secure, and can achieve trustworthy identity verification operations, enhancing the overall protection ability of the system. Through the biometric binding of EEG signals and the generation of user identity keys, it can be ensured that the identity information of each user is unique and non-replicable. The user's EEG signals have individuality and biological non-forgeability, making the user's identity impossible to be disguised or replicated. This binding mechanism can not only prevent malicious users from impersonating others, but also provide higher anti-tampering ability for the system, enhancing the security of user identity verification. Using blockchain technology for user biometric binding and key generation can ensure the transparency and traceability of data. Every identity verification, data upload, and other operations will be recorded on the blockchain, ensuring that every operation in the system has an audit trail, which is convenient for later review and verification. This transparency and traceability provide users with higher data control rights and make the system operations more open and trustworthy.

[0053] Preferably, the uploading the EEG signals of the corresponding user's multi-session data to the blockchain for user biometric binding includes:

[0054] Homomorphically encrypt the electroencephalogram (EEG) signals of the corresponding user's multi-session data to generate encrypted EEG signals and an encryption key;

[0055] Perform blockchain-based evidence storage on the encrypted EEG signals through blockchain technology to obtain EEG encrypted storage feature data;

[0056] Bind the features of the EEG encrypted storage feature data and the encryption key to generate a revocable key;

[0057] Pair the encryption key and the revocable key to obtain the user identity key.

[0058] The present invention encrypts the user's EEG signals through homomorphic encryption technology to ensure that the user's EEG signal data remains encrypted during storage and transmission, avoiding unauthorized access and leakage. This encryption method can operate on encrypted data without decryption, maximizing the protection of user privacy. The evidence storage of user data on the blockchain ensures the integrity and immutability of the data in the network, effectively preventing data tampering and malicious modification behaviors, and further enhancing the security of user data. Using blockchain technology to perform evidence storage on the encrypted EEG signals can ensure that all stored data has high transparency, traceability, and immutability. The decentralized feature of the blockchain enables the processing of each piece of data to be audited, preventing data from being forged or deleted. The EEG encrypted storage feature data generated through blockchain technology can be accessed through legal authorization when needed, and cannot be tampered with or deleted without authorization, ensuring the legality and security of the data. Binding the features of the EEG encrypted storage feature data and the encryption key can generate a revocable key. The introduction of the revocable key enables the user to actively revoke or update their identity information when needed, avoiding the long-term binding or abuse of identity information. The user can revoke or modify the bound key at any time according to their needs, thereby enhancing the flexibility and operability of the biometric binding system. This flexibility enhances the user's control over personal data and protects the user's long-term privacy.

[0059] In this specification, an authentication system for identifying a user's identity based on EEG features is provided, which is used to execute the above-mentioned authentication method for identifying a user's identity based on EEG features. The authentication system for identifying a user's identity based on EEG features includes:

[0060] A signal acquisition module, which is used to collect the EEG signals of the user under a preset cognitive task using an EEG device, perform noise baseline calibration on the EEG signals, generate standard EEG signals, and perform AES encryption processing on the EEG samples;

[0061] An adversarial training module, which is used to generate corresponding adversarial samples for standard electroencephalogram (EEG) signals through an adversarial generation network; apply a spatio-temporal dynamic masking strategy to inject EEG chaos features into the standard EEG signals to obtain EEG chaos-injected features; use the generated corresponding adversarial samples to perform adversarial dynamic feature enhancement training on the EEG chaos-injected features to generate an adversarial test result;

[0062] An anti-interference feature extraction module, which is used to extract the time-frequency features and functional connection matrix of the adversarial test result and construct an anti-interference user feature template library;

[0063] An identity authentication module, which is used to obtain the EEG signals of the user's multi-session data and import the EEG signals of the user's multi-session data into the anti-interference user feature template library for identity determination. If the identity determination result is true, the EEG signals of the corresponding user's multi-session data will be uploaded to the blockchain for user biometric binding to generate a user identity key.

[0064] The beneficial effects of the present invention are as follows: By collecting the electroencephalogram (EEG) signals of users under preset cognitive tasks using an EEG device, it ensures that the obtained EEG data is real and representative, providing high-quality input signals for subsequent analysis and training. By performing noise baseline calibration on the EEG signals, it can effectively remove noise interference and improve the quality and reliability of the signals. This calibration step lays a solid foundation for generating standard EEG signals and performing AES encryption processing on EEG samples, avoiding the influence of interference factors on subsequent analysis. Using a generative adversarial network (GAN) to generate corresponding adversarial samples for the standard EEG signals can improve the system's ability to identify abnormal and attack signals and enhance the system's robustness. By applying a spatio-temporal dynamic masking strategy, chaotic EEG features are injected into the standard EEG signals to simulate the changes of EEG signals in complex environments. This step helps train the system to cope with different EEG signal interferences and noises, further enhancing the system's robustness. Using the generated adversarial samples to perform adversarial dynamic feature enhancement training on the chaotic EEG features, the generated adversarial test results can help improve the system's ability to identify and respond to interference signals and optimize the system's performance. By performing time-frequency feature extraction and functional connectivity matrix analysis on the adversarial test results, it can capture the detailed information in the signals and provide in-depth EEG activity analysis, which helps extract stable and discriminative features. By constructing an anti-interference user feature template library, it can provide a reliable reference for identity verification, ensuring that the verification process can accurately distinguish different users and resist potential forgery or malicious attacks. This module can perform identity determination based on the EEG signals of the user's multi-session data. By matching with the anti-interference user feature template library, it ensures the accurate verification of the user's identity. By uploading the user's EEG signals to the blockchain and performing encrypted storage and feature binding, a user identity key is generated, realizing the secure protection of the user's identity information. The immutability and transparency of blockchain evidence ensure the security and privacy of user data. Therefore, the present invention improves the accuracy, anti-interference ability, and security of the EEG-based identity authentication system through noise calibration, adversarial sample generation, feature enhancement training, and blockchain technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 FIG. is a schematic flowchart of the steps of an authentication method for identifying a user's identity based on EEG features;

[0066] Figure 2 is Figure 1 a detailed implementation step flowchart of step S3 in;

[0067] Figure 3 is Figure 1 a detailed implementation step flowchart of step S4 in;

[0068] The realization, functional features, and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. Detailed implementation manners

[0069] The technical method of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those skilled in the art within the scope of the present invention without creative efforts based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0070] In addition, the accompanying drawings are only schematic diagrams of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings represent the same or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software, or in one or more hardware modules or integrated circuits, or in different networks and / or processor methods and / or microcontroller methods.

[0071] It should be understood that although the terms "first", "second", etc. may be used here to describe each unit, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit can be called the second unit, and similarly the second unit can be called the first unit. The term "and / or" used here includes any and all combinations of one or more of the listed related items.

[0072] To achieve the above object, please refer to Figures 1 to 3 , an authentication method for identifying a user's identity based on brain wave characteristics, the method comprising the following steps:

[0073] Step S1: Use an EEG device to collect the brain electrical signals of a user under a preset cognitive task, perform noise baseline calibration on the brain electrical signals, generate standard brain electrical signals, and perform AES encryption processing on the brain electrical samples.

[0074] Step S2: Generate corresponding adversarial samples for the standard brain electrical signals through an adversarial generation network; apply a spatio-temporal dynamic masking strategy to inject brain electrical chaos features into the standard brain electrical signals to obtain brain electrical chaos injection features; use the generated corresponding adversarial samples to perform adversarial dynamic feature enhancement training on the brain electrical chaos injection features to generate an adversarial test result.

[0075] Step S3: Extract the time-frequency features and functional connection matrices of the adversarial test results, and construct an anti-interference user feature template library.

[0076] Step S4: Obtain the electroencephalogram (EEG) signals of the user's multi-session data, and import the EEG signals of the user's multi-session data into the anti-interference user feature template library for identity determination. If the identity determination result is true, upload the corresponding EEG signals of the user's multi-session data to the blockchain for user biometric binding to generate a user identity key.

[0077] Through the combination of a generative adversarial network (GAN) and a spatio-temporal dynamic masking strategy, the present invention can effectively inject EEG chaotic features, improve the system's ability to identify interference or forged signals, and enhance the defense ability against adversarial attacks. By performing identity determination on multi-session data, the user's identity can be confirmed at different times and in different situations, which can effectively avoid verification errors caused by different environments or devices, and improve the stability and reliability of identity verification. The extraction of time-frequency features and functional connectivity matrices helps to capture the deep patterns in EEG signals and improve the accuracy of identity recognition. In particular, the functional connectivity matrix can reflect the interactions between different regions of the brain, further strengthening the individual's biometric characteristics. Uploading the user's EEG signal data to the blockchain can ensure the immutability and privacy of the data. This decentralized storage method can prevent data leakage or tampering and enhance the security of the entire system. Through biometric binding, combined with the user's unique EEG signal information, more accurate identity authentication can be achieved, reducing the security risks of traditional verification methods such as passwords or fingerprints. By using multi-session data, the system can adapt to the user's different physiological states, enhancing the ability of personalized authentication, which is particularly effective for long-term monitoring or identity recognition in dynamic environments. Therefore, through noise calibration, adversarial sample generation, feature enhancement training, and blockchain technology, the present invention improves the accuracy, anti-interference ability, and security of the EEG-based identity authentication system.

[0078] In the embodiment of the present invention, refer to Figure 1 As shown, it is a schematic diagram of the step flow of an authentication method for identifying a user's identity based on EEG features according to the present invention. In this example, the authentication method for identifying a user's identity based on EEG features includes the following steps:

[0079] Step S1: Use an EEG device to collect the EEG signals of the user under a preset cognitive task, perform noise baseline calibration on the EEG signals, generate standard EEG signals, and perform AES encryption processing on the EEG samples.

[0080] In the embodiments of the present invention, electroencephalogram (EEG) signals of a user under a preset cognitive task are collected by using a high-precision EEG device, such as a multi-channel EEG cap. Generally, the device requires at least 16 to 32 electrodes to ensure comprehensive monitoring of EEG activities. An appropriate sampling rate is selected, usually set to 500 Hz to 1000 Hz, to ensure the accuracy of the signals and the capture of high-frequency components. According to the target research, an appropriate cognitive task (such as: working memory task, attention task, decision-making task, etc.) is selected, and the execution process of the task is designed to ensure that the difficulty of the task is moderate and meets the experimental requirements. During the task, the user is required to focus on the specified task to ensure that the changes in EEG signals are related to cognitive activities. The collected EEG signals are processed for noise. The sources of noise include motion artifacts, power interference, muscle activities, etc. External interference is reduced by performing baseline calibration on the EEG signals. The EEG signals are frequency-filtered (usually 0.5 Hz to 50 Hz) to remove low-frequency drifts and high-frequency muscle noise. According to the electrode arrangement, an appropriate reference electrode is selected, such as the nose or ears of the scalp or an online reference electrode, etc., to reduce the interference of external noise. Interference signals such as eye movements and electromyogram artifacts are removed through techniques such as independent component analysis (ICA) or blind source separation (BSS). After completing the noise baseline calibration, the EEG signals are denoised and filtered to obtain standard EEG signals. This signal reflects the EEG activities of the user during the execution of the cognitive task and can provide a reliable basis for subsequent data analysis. The standard EEG signals usually include brain waves in different frequency bands (such as Delta, Theta, Alpha, Beta, Gamma). By analyzing the power spectra of these frequency bands, characteristic information in the cognitive task can be further extracted.

[0081] Step S2: Generate corresponding adversarial samples for the standard EEG signals through an adversarial generation network; apply a spatio-temporal dynamic masking strategy to inject EEG chaotic features into the standard EEG signals to obtain EEG chaotic injection features; use the generated corresponding adversarial samples to perform adversarial dynamic feature enhancement training on the EEG chaotic injection features to generate an adversarial test result;

[0082] In the embodiments of the present invention, adversarial samples corresponding to standard electroencephalogram (EEG) signals are generated by using a Generative Adversarial Network (GAN). The GAN consists of two parts: a Generator and a Discriminator. The Generator takes the standard EEG signal as input and generates potential adversarial signals, attempting to imitate and perturb the original signal. The Discriminator determines whether the input signal is a standard signal or an adversarial signal generated by the Generator, and helps the Generator optimize through feedback. During the training process, the Generator is gradually optimized to make the generated adversarial samples more deceptive, while the Discriminator is optimized to distinguish between real signals and adversarial samples. Based on the generated adversarial samples, a spatio-temporal dynamic masking strategy is applied to inject EEG chaos features. This strategy simulates abnormal fluctuations or abnormal patterns in EEG activities. According to the time-domain and spatial-domain characteristics of the signal, partial regions of the signal are dynamically selected for masking or perturbation. Interference and abnormalities in EEG activities can be simulated through spatio-temporal local perturbations. Noise or pseudo-signals are added through the masking strategy (such as introducing random signals or interference signals in specific frequency bands) to simulate chaotic patterns in EEG signals and generate new EEG chaos injection features. For example, the masking strategy can act on a specific time window or frequency band (such as the Alpha band) of the EEG signal and inject perturbations to enhance the diversity and challenge of the samples. The generated adversarial samples and EEG chaos injection features are used for adversarial dynamic feature enhancement training. The purpose of the training is to enhance the robustness of the model through the generated adversarial samples, enabling it to maintain high classification or prediction performance when facing chaotic and perturbed EEG signals. The adversarial samples are mixed with the standard EEG signals as training data and input into the model. During the training process, the model will learn how to extract effective features from the perturbed signals and enhance its adversarial ability. During training, the model is gradually optimized to enhance the dynamic features in the signal, that is, to identify and enhance valuable dynamic features (such as frequency band changes, phase synchronization, etc.) in the time-domain and frequency-domain of the EEG signal, thereby improving the sensitivity of the model to EEG signal changes. After adversarial training, the trained model is used to test new signals and generate adversarial test results. This process verifies the performance of the model when facing EEG chaos features (i.e., signals formed by adversarial samples and chaos feature injection).

[0083] Step S3: Extract the time-frequency features and functional connectivity matrices of the adversarial test results, and construct an anti-interference user feature template library;

[0084] In the embodiments of the present invention, through Step S3: Extract the time-frequency features and functional connectivity matrices of the adversarial test results, and construct an anti-interference user feature template library;

[0085] Wavelet transform is used to analyze the time-frequency local characteristics of electroencephalogram (EEG) signals, which is especially suitable for processing non-stationary signals. By selecting appropriate wavelet basis functions, information of the signal at different frequencies and time scales can be extracted. The EEG signals are divided into different frequency bands (such as Delta, Theta, Alpha, Beta, Gamma), the power spectral density (PSD) of each frequency band is calculated respectively, and the variation characteristics of each frequency band are analyzed. Through the analysis of time-frequency characteristics, the influence of adversarial perturbations on signals in different frequency bands can be captured. The Pearson correlation coefficient or mutual information between different electrode positions is calculated to reveal the correlation between different brain regions. High correlation indicates that the activities of two brain regions tend to be synchronized, while low correlation indicates that their activities are irrelevant. By calculating the phase synchronization between signals in different brain regions, the phase synchronization or desynchronization pattern between brain regions is evaluated, and functional connectivity features are further extracted. The Granger causality analysis method is adopted to explore the causal relationship between different regions in EEG signals, and a functional connectivity matrix is further constructed. Through graph theory analysis of the functional connectivity matrix, the topological structure of the brain network is constructed, and the connection strength and information flow path between different brain regions are analyzed. The time-frequency characteristics extracted from the adversarial test results are integrated with the functional connectivity matrix to generate an anti-interference feature template for each user. The template for each user contains time-frequency characteristics, the functional connectivity matrix, and their dynamic changes under adversarial perturbations. The time-frequency characteristics and the functional connectivity matrix are transformed into vector form to construct the feature vector of the user. Dimensionality reduction methods (such as principal component analysis PCA, t-SNE, etc.) can be used to reduce the dimensionality of the features while keeping the core part of the information. With the generation and testing of adversarial samples, the template library will be continuously updated. The anti-interference ability of the template library can be enhanced through regular adversarial training and testing to ensure its effectiveness in the real environment. The constructed anti-interference user feature template library can be used for personalized recognition and adversarial prediction. When new EEG signals of a user are input, the cognitive state of the user can be quickly identified by matching with the feature vectors in the template library. The similarity between the new EEG signals and the features in the template library is calculated, and matching is performed using measurement methods such as cosine similarity and Euclidean distance to identify whether the EEG features of the user are interfered. Based on the matching results of the template library, the system can monitor in real time whether the EEG activities of the user are affected by adversarial perturbations and give interference warnings or suggestions to improve the stability of the experiment.

[0086] Step S4: Obtain the EEG signals of the user's multi-session data, and import the EEG signals of the user's multi-session data into the anti-interference user feature template library for identity determination. If the identity determination result is true, upload the corresponding EEG signals of the user's multi-session data to the blockchain for user biometric binding to generate a user identity key.

[0087] In the embodiments of the present invention, electroencephalogram (EEG) signals of a user are collected using an EEG device in multiple different sessions. During each session, the user's EEG signals are recorded and saved as a series of time-series data. Multi-session data collection can span different time periods and different task environments to ensure that the collected signals contain a wide range of characteristics of the user's EEG activities. The multi-session EEG signals collected are standardized to remove noise and ensure signal quality. Filters (such as band-pass filters) are used to remove low-frequency and high-frequency noise, and processes such as artifact removal and desynchronization are performed to obtain standardized EEG signals. Time-frequency features (such as power spectral density, band features, etc.) and functional connectivity matrices are extracted from the EEG signals in the obtained multi-session data. Through the foregoing methods, time-domain, frequency-domain, and functional connectivity information between different brain regions of the multi-session data are extracted. Time-frequency features (such as power spectral density, band features, etc.) and functional connectivity matrices are extracted from the EEG signals in the obtained multi-session data. Through the foregoing methods, time-domain, frequency-domain, and functional connectivity information between different brain regions of the multi-session data are extracted., Calculate the correlation, synchronization, and causal relationship between each EEG channel in the user's multi-session data, and construct the user's functional connectivity matrix. The extracted features are matched with the registered feature templates in the anti-interference user feature template library, and similarity calculation methods (such as cosine similarity, Euclidean distance, etc.) are used for identity verification. Based on the matching result, if the features of the multi-session data are highly similar to the features of a known user in the template library (exceeding a certain set threshold), the identity is determined to be true; otherwise, the identity is determined to be false. When the identity is determined to be true, the system generates a unique user identity key. This key is generated based on the features of the user's multi-session EEG signals through an algorithm (such as a hash algorithm), and has high security and uniqueness. The generated key will be used for subsequent identity verification. For example, a unique key can be generated using the feature vector of the user's EEG signals through a hash function (such as SHA-256). During the key generation process, privacy protection needs to be ensured to avoid leakage of the user's personal biometric characteristics. The generated user identity key and its associated multi-session EEG signal features are uploaded to the blockchain system. The blockchain ensures the immutability and decentralization of data, ensuring the security and privacy of user identity data. Through a smart contract, the generated user identity key is bound to the EEG signal features and stored in the blockchain. The identity key of each user and its corresponding EEG feature data will form an immutable record. The data uploaded to the blockchain should be encrypted to ensure that the user's biometric data is not directly disclosed and can only be unlocked with the correct identity key. Through the user identity key on the blockchain, the user's EEG features can be bound to the identity, establishing a secure and verifiable identity management system. In the future, in any scenario that requires identity authentication, the identity can be verified by obtaining the user's EEG signals. The system will extract new EEG features and compare them with the keys in the blockchain to ensure the uniqueness and security of the identity.When the user needs to perform identity authentication, the system collects and analyzes real-time electroencephalogram (EEG) signals, generates corresponding feature vectors, and compares them with the identity keys stored on the blockchain to ensure the correctness of the user's identity.

[0088] Preferably, step S1 includes the following steps:

[0089] Step S11: Use an EEG device to collect the user's EEG signals to obtain EEG signals;

[0090] Step S12: Based on a synchronous acquisition device, collect eye movement signals and electromyogram signals synchronized with the EEG signals;

[0091] Step S13: Design a preset cognitive task, where the preset cognitive task includes visual stimulus-induced P300 potential, multi-level mental arithmetic tasks, and cross-modal association tasks;

[0092] Step S14: Identify interference signal sources for the EEG signals according to the eye movement signals and electromyogram signals to generate noise source identification data; perform noise baseline calibration on the noise source identification data to generate noise baseline calibration data;

[0093] Step S15: Optimize the signal quality of the EEG signals through the noise baseline calibration data to generate standard EEG signals and perform AES encryption processing on the EEG samples.

[0094] In the embodiments of the present invention, by selecting a high-quality EEG (electroencephalogram) device with a sufficient number of channels (such as 16 - 32 electrodes), different regions of the user's scalp can be covered. The device should have a high sampling rate (such as above 1000 Hz) and high resolution to ensure accurate details of the collected EEG signals. Before signal acquisition, ensure that the user wears the correct EEG electrode cap and has good contact with the skin. Ensure good contact between electrodes and apply conductive paste to reduce resistance. Instruct the user to maintain a static state and avoid excessive body movements or interference. Conduct EEG signal acquisition in a controlled environment and record the user's EEG activities during cognitive tasks. Use an eye tracker and an electromyogram (EMG) device synchronized with the EEG device to ensure that eye movement and EMG signals can be precisely aligned with the EEG signals on the time axis. Use the eye tracker to record the user's eye movement data (such as eye position, eye movement speed, etc.), which helps to identify artifacts caused by eye movements. Use the EMG device to record the user's muscle activity signals (such as eye muscles, neck muscles, etc.) to help identify artifacts caused by muscle activities. Ensure that each signal source (EEG, eye movement, EMG) has a timestamp and is precisely synchronized during post-processing. Design visual stimulation tasks (such as stimulating images or texts) to evoke the P300 wave (an EEG response related to cognitive processing). In the task, a series of stimuli are randomly presented, and the user is required to identify the target stimulus and respond to obtain the P300 response. Design a hierarchical mental arithmetic task (such as addition and subtraction, multiplication and division, etc.), requiring the user to complete mathematical calculations in different levels of tasks, and use EEG signals to monitor changes in their attention and working memory. Design a cross-modal association task (for example, presenting a visual or auditory stimulus and requiring the user to make an association), which can guide the user to make associations through the visual and auditory signals of the stimulus, thereby triggering specific EEG response patterns (such as Alpha waves, Theta waves, etc.). By analyzing the synchronized eye movement signals, artifacts caused by rapid eye movements (such as blinks, gaze shifts, etc.) are detected. Usually, eye movement artifacts appear in the lower frequency band of the EEG signals, especially in the frontal region. Using the temporal contrast between the eye movement signals and the EEG signals, eye movement artifacts are identified. Through the analysis of the synchronized EMG signals, artifacts related to muscle activities are identified, especially muscle activities in regions such as the neck, eyes, and jaw. EMG artifacts usually appear in the high-frequency band of the EEG signals. EMG artifacts can be identified by modeling the amplitude and frequency characteristics of the EMG signals. Combining eye movement and EMG signals, the interference components in the EEG signals are accurately marked to form noise source identification data, which will be used for subsequent noise calibration and signal optimization. By calculating the time-frequency correlation between the EEG signals and the eye movement signals and the EMG signals, regression analysis or a multi-channel signal fusion method based on deep learning is used to accurately identify the interference sources. Use image processing or signal decomposition techniques (such as independent component analysis ICA) to separate eye movement and EMG artifacts and generate noise source identification data.Perform baseline noise analysis on the noise source identification data to calculate the frequency characteristics and amplitude range of the noise signal. Based on this data, design denoising algorithms (such as filtering, ICA, PCA, etc.) to remove noise from the EEG signals. According to the identified noise sources, design noise calibration algorithms, such as template-based artifact removal methods or use filters to eliminate artifact signals in specific frequency bands. The algorithm dynamically adjusts the EEG signals to remove noise introduced by eye movements or EMG artifacts. After noise calibration, the optimized EEG signals are obtained and standardized to ensure that their amplitude and frequency characteristics meet the standards. The EEG signals at this time are the "standard EEG signals".

[0095] Preferably, the generation of corresponding adversarial samples for the standard EEG signals in step S2 includes:

[0096] Add low-frequency drift component noise to the standard EEG signals to obtain low-frequency drift component noise; simulate the enhancement of the drift amplitude based on the low-frequency drift component noise for the standard EEG signals to generate EEG signal data with drift noise and label it as the first type of sample;

[0097] Randomly add high-frequency noise components to the standard EEG signals to obtain high-frequency random component noise; simulate environmental interference based on the high-frequency random component noise for the standard EEG signals to generate EEG signal data with environmental interference noise and label it as the second type of sample;

[0098] Use adversarial generation network technology to embed forged EEG activities into the standard EEG signals to generate forged EEG feature signals; simulate spoofing attacks on the standard EEG signals through the forged EEG feature signals to generate EEG signal data with spoofing attacks and label it as the third type of sample;

[0099] Integrate the first type of sample, the second type of sample, and the third type of sample into comprehensive adversarial samples.

[0100] In the embodiments of the present invention, appropriate low-frequency noise models (e.g., 1 / f noise or low-frequency noise oscillation) are selected to simulate low-frequency drift. Low-frequency drift usually appears in the slow bands (such as Delta wave, Theta wave) of electroencephalogram (EEG) signals, and these bands are vulnerable to external environments such as electrical devices and temperature changes. The generated low-frequency noise is added to the low-frequency components of the standard EEG signal. A suitable noise generator (e.g., a low-frequency noise generation model based on a noise generation adversarial network) can be designed to simulate low-frequency drift and synthesize it with the EEG signal. By adding low-frequency noise, the EEG signal drifts in the lower frequency band, affecting the stability and accuracy of the signal. According to the added low-frequency noise, the drift amplitude is adjusted to simulate different interference situations in the EEG signal. The complexity of the signal can be increased by adjusting the drift amplitude to make it more challenging. The EEG signal after low-frequency drift simulation is labeled as the first type of sample. Such samples are used to simulate the low-frequency drift characteristics of EEG signals under different electrical interferences and environmental fluctuations. A high-frequency random noise model (such as Gaussian noise or white noise) is used to simulate common high-frequency interferences in the environment. High-frequency noise components are often generated by the interference of electronic devices, electrical noise, or electromyogram signals. The high-frequency noise components are randomly added to the standard EEG signal through a certain algorithm to increase the high-frequency components in the signal, so as to simulate the interference effects under different environmental conditions. According to different actual environmental conditions, the intensity and distribution of high-frequency noise are adjusted. By adjusting the noise amplitude, frequency range, and noise distribution, different intensities of environmental interference are simulated. The EEG signal after high-frequency noise interference simulation is labeled as the second type of sample, and these samples are used to simulate the high-frequency noise characteristics of EEG signals in a complex electrical environment. The generative adversarial network technology is used to generate forged EEG activity signals. By training the GAN model, it can generate signals with false EEG characteristics, such as unnatural Theta waves or Alpha waves, to simulate external attacks or abnormal EEG activities. The forged EEG signals generated by the GAN will have different characteristics from the real EEG signals, but their time-domain and frequency-domain distributions are similar to those of normal EEG signals, making it difficult to directly distinguish. The generated forged EEG characteristics are embedded into the standard EEG signal to simulate spoofing attacks (such as the EEG pattern of a fake user or deliberately creating signal anomalies), and such spoofing attacks will affect the specific characteristics of the signal, making the signal appear unrealistic or unnatural. The signal containing the forged EEG characteristics is labeled as the third type of sample, and these samples simulate external spoofing attacks on EEG signals, specifically the result of malicious users or system intrusions. The first type, second type, and third type of samples are combined into a complete set of adversarial samples. This set contains different types of noise (low-frequency drift, environmental high-frequency noise, and spoofing attacks) and signal interferences, and can comprehensively simulate various interference conditions. By integrating these adversarial samples, the diversity of training data can be increased to help neural networks or other machine learning models improve the robustness to EEG signals.

[0101] Preferably, the injection of EEG chaos features into the standard EEG signal by applying the spatio-temporal dynamic masking strategy in step S2 includes:

[0102] Randomly select several frequency bands from the standard EEG signal, and randomly discard segments of the selected frequency bands to generate randomly discarded segment data; interpolate time series gaps for the standard EEG signal based on the randomly discarded segment data to generate an EEG reconstructed signal;

[0103] Randomly select electrode channels from the standard EEG signal, and randomly mask the standard EEG signals of the corresponding electrode channels to generate channel masking data; zero the signal values of the standard EEG signal based on the channel masking data to generate an EEG masked signal;

[0104] Perform spectrum analysis on the standard EEG signal, extract the frequency bands of interest for proportional noise injection to generate a noise injection signal; analyze the abnormal signal features of the EEG reconstructed signal, the EEG masked signal, and the noise injection signal, and integrate them into a spatio-temporal dynamic masking strategy;

[0105] Inject EEG chaos features into the standard EEG signal using the abnormal signal features based on the spatio-temporal dynamic masking strategy to obtain EEG chaos injection features.

[0106] In the embodiments of the present invention, several frequency bands are randomly selected from the spectrum of the standard electroencephalogram (EEG) signal (for example, Delta band, Theta band, Alpha band, etc.). These frequency bands are key components of the EEG signal and affect the normal activity pattern of the EEG. For each selected frequency band, several signal segments are randomly selected in the time domain and discarded. This is done to simulate the temporal absence or loss of some information in the signal, increasing the missing patterns in the EEG signal. The data after discarding generates randomly discarded segment data, which lacks the information of certain frequency bands, simulating the missing problem of the EEG signal in practical applications. Based on the randomly discarded segment data, an interpolation operation is performed on the missing signal part, and the time series gap interpolation method is used to fill in the lost segments, generating continuous signal data. Common interpolation methods include linear interpolation, spline interpolation, or more complex signal recovery algorithms. Through the signal data filled by interpolation, a new EEG reconstructed signal is generated. This signal retains the overall trend of the standard EEG signal while containing the temporal gaps and reconstructed parts caused by discarding the segments. In the EEG signal, there are usually multiple electrode channels providing signals at different positions. At this time, several electrode channels are randomly selected for shielding to simulate the loss or distortion of the signal on specific electrode channels. When selecting electrode channels, a simple random algorithm or a selection method based on electrode positions can be used. For the selected electrode channels, the corresponding standard EEG signals are completely shielded or set to zero. By shielding the signals of the electrode channels, the signal loss caused by poor electrode contact, interference, or other reasons can be simulated. Through this operation, channel shielding data is generated, and the signals in this part of the data will have no information from specific channels, simulating the failure or interference of some channels. After channel shielding, the signal values in the channels are further set to zero, that is, the entire signal value is set to zero, further simulating the situation of ineffective electrodes or complete signal loss. Through this operation, an EEG shielded signal is generated. This signal is the result of shielding the information of specific channels from the standard EEG signal, simulating the signal loss caused by hardware failures or other external factors. Perform spectral analysis on the standard EEG signal to extract the energy distribution of each frequency band. Common methods include Fourier transform (FFT) or wavelet transform. This step helps to identify and extract the frequency bands crucial for EEG activities, such as Alpha band, Beta band, Theta band, etc. According to the analysis results, specific frequency bands are selected as the targets for noise injection. For example, the Alpha band or Beta band can be selected, which are of great significance in aspects such as the emotion and cognitive state of the EEG signal. Inject noise into the selected frequency bands. The signals in the frequency bands can be interfered by designing a noise model (such as Gaussian noise or random noise). By injecting proportional noise into these frequency bands of interest, the signal fluctuations, jitters, or interferences can be effectively simulated.Analyze the EEG reconstructed signal, EEG shielded signal, and noise injection signal, and extract their abnormal features, which may include non-linear changes, spectral changes, time-domain fluctuations, etc. In this process, anomaly detection algorithms (such as autoencoders, anomaly detection neural networks, etc.) can be used to identify and extract abnormal signal features. Based on the abnormal features extracted from the EEG signal, construct a spatio-temporal dynamic masking strategy. This strategy may include dynamic change patterns in time and frequency, such as signal masking in specific time periods and frequency bands, to simulate the behavior patterns of EEG signals when affected by external or internal interference. Based on the spatio-temporal dynamic masking strategy, inject the abnormal signal features into the standard EEG signal. This injection includes not only the temporal changes of the signal and the perturbation of the frequency components, but also the influence on certain electrode channels in space. Through the above strategy, the final EEG chaos injection features are obtained, which enhance the robustness of the EEG signal in an interference environment by simulating various signal perturbations.

[0107] Preferably, the adversarial dynamic feature enhancement training of the EEG chaos injection features using the generated corresponding adversarial samples in step S2 includes:

[0108] Conduct an EEG device distribution difference analysis on the EEG chaos injection features to generate device distribution difference data; use the device distribution difference data to perform cross-device distribution alignment training on the generated corresponding adversarial samples to generate cross-device aligned signal feature data;

[0109] Construct a live detection scenario; perform live detection adversarial training on the generated corresponding adversarial samples through the live detection scenario to generate enhanced live detection feature data;

[0110] Integrate the cross-device aligned signal feature data, the generated corresponding adversarial samples, and the enhanced live detection feature data into a joint training set; conduct an adversarial test on the joint training set according to the adversarial generation network to generate an adversarial test result.

[0111] In the embodiments of the present invention, data is collected for different EEG devices, with particular attention paid to the signal differences generated by different devices under the same experimental conditions. For example, EEG devices of different brands, models, or configurations may have differences in the collected EEG signals in terms of frequency response, noise level, and signal quality due to differences in sensor layout, sampling rate, or signal processing methods. By comparing and analyzing the EEG signals from different devices, the distribution differences between the signals are extracted, which may include differences in signal amplitude, noise components, spectral characteristics, etc. Based on these differences, device distribution difference data is generated to provide guidance for subsequent adversarial training. Using the device distribution difference data, cross-device distribution alignment training is performed on the generated corresponding adversarial samples, which means adjusting or correcting the generated adversarial samples to adapt to the acquisition environment of different devices. For example, techniques such as normalization, standardization, or adaptive filtering are used to reduce the differences between devices. Through this process, signal feature data after cross-device distribution alignment is obtained, which is more consistent and can be effectively compared and trained between different devices, enhancing the cross-device robustness of the model. Liveness detection is a means to prevent spoofing attacks. This step requires designing a liveness detection scenario that includes various fake EEG activities. Specifically, the natural physiological responses of users during certain cognitive tasks, such as eye movement, electromyogram, electrocardiogram, etc., can be simulated to verify whether the signal comes from a real live user. In this scenario, some fake EEG signals can be artificially generated to simulate attackers deceiving through fake devices or simulated EEG activities. These fake signals may include distorted signals, abnormal fluctuations in the time domain and frequency domain, etc. Using the constructed liveness detection scenario, liveness detection adversarial training is performed in combination with the generated corresponding adversarial samples. The goal of the training is to enable the model to effectively distinguish between the EEG signals from real live users and the fake signals from fake devices or attackers. By introducing adversarial training for liveness detection, the security of the model is enhanced, making it more resistant to the interference of fake signals. During the training process, through adversarial training, the model will be gradually optimized, so as to better extract the features that distinguish live signals and fake signals, and finally generate enhanced liveness detection feature data. The signal feature data after cross-device alignment, the generated corresponding adversarial samples, and the enhanced liveness detection feature data are integrated to construct a comprehensive joint training set. This training set contains various adversarial samples, cross-device corrected signals, and enhanced liveness detection features, which can cover more diverse interference factors. The diversity of the joint training set enables the trained neural network to learn from multiple dimensions how to resist different types of interference (such as differences between devices, live fake signals, environmental noise, etc.), thereby improving the generalization ability and robustness of the model in practical applications. An adversarial test is performed on the integrated joint training set using a generative adversarial network (GAN). The GAN can effectively test the performance of the model under various perturbations by generating fake samples and competing with real samples.In this step, the adversarial ability of the model is strengthened, enabling it to better identify and respond to various potential attacks and interferences. After the adversarial test, the generated adversarial test results include the model's performance in different adversarial samples, device differences, and forgery attack scenarios. The test results will help further optimize the model to ensure its stability and security in practical applications.

[0112] As an example of the present invention, refer to Figure 2 As shown, in this example, step S3 includes:

[0113] Step S31: Perform time-frequency analysis on the adversarial test results to generate instantaneous EEG feature data;

[0114] Step S32: Perform short-time Fourier transform on the instantaneous EEG feature data to generate instantaneous EEG spectrum data; extract the spectral features of the instantaneous EEG spectrum data to obtain instantaneous EEG spectrum feature data;

[0115] Step S33: Calculate the phase synchrony based on the instantaneous EEG feature data and the instantaneous EEG spectrum feature data to obtain regional brain phase synchrony data; construct a functional connectivity matrix for the adversarial test results based on the regional brain phase synchrony data, and extract the anti-interference features of the functional connectivity matrix;

[0116] Step S34: Construct an anti-interference user feature template library for the adversarial samples through the anti-interference features to generate an anti-interference user feature template library.

[0117] In the embodiments of the present invention, by collecting electroencephalogram (EEG) signal data from the experimental environment, it is ensured that these data have been preprocessed to remove noise, artifacts, etc. The EEG signals are analyzed using time-frequency analysis techniques (such as wavelet transform, Hilbert-Huang transform, etc.) to obtain the instantaneous characteristics of the signals in different time and frequency ranges. Time-frequency domain characteristics such as instantaneous frequency, instantaneous amplitude, and instantaneous phase are extracted from the time-frequency analysis to generate EEG signal instantaneous characteristic data. The obtained EEG signal instantaneous characteristic data is subjected to short-time Fourier transform (STFT) to convert the signal from the time domain to the frequency domain, obtaining instantaneous spectrum data. Further processing the instantaneous spectrum data to extract important spectrum characteristics (such as the bandwidth of the spectrum, energy distribution, etc.) to obtain EEG signal instantaneous spectrum characteristic data. According to the instantaneous characteristic data and the instantaneous spectrum characteristic data, a phase synchrony calculation method (such as phase locking value, PLV) is used to analyze the synchrony of EEG signals between different brain regions. According to the phase synchrony data, a functional connectivity matrix is constructed to represent the phase synchrony relationship between different brain regions. This matrix reflects the collaborative work between brain regions. Anti-interference characteristics are extracted from the functional connectivity matrix, and these characteristics help to determine whether the signal is affected by anti-interference. Using the anti-interference characteristics extracted in step S33, the effects of different interference samples on the EEG signals are analyzed. The anti-interference characteristics can be classified by machine learning methods (such as support vector machine, SVM, or convolutional neural network, CNN). According to the analysis results, an anti-interference user characteristic template library is constructed, and the characteristics of the EEG signals of different users under anti-interference conditions are stored as templates. With the collection and analysis of more data, the anti-interference user characteristic template library is continuously optimized and updated to improve the adaptability of the model to new adversarial samples.

[0118] Preferably, step S33 includes the following steps:

[0119] Step S331: Screening the instantaneous change EEG signals according to the EEG signal instantaneous characteristic data and the EEG signal instantaneous spectrum characteristic data for the adversarial test results to obtain the instantaneous change EEG signals;

[0120] Step S332: Calculating the inter-brain region phase synchrony of the instantaneous change EEG signals to obtain the inter-brain region phase difference;

[0121] Step S333: Analyzing the phase synchrony change of the instantaneous change EEG signals across time windows through the inter-brain region phase difference to generate inter-brain region phase synchrony data;

[0122] Step S334: Constructing a functional connectivity matrix based on the phase synchrony change data and performing dynamic functional connectivity analysis on the inter-brain region phase synchrony data using the functional connectivity matrix to generate dynamic functional connectivity data;

[0123] Step S335: Use the dynamic functional connection data to extract the stable phase features of the functional connection matrix, and obtain the anti-interference features of the functional connection matrix.

[0124] In the embodiments of the present invention, the instantaneous change of the EEG signal is calculated based on the instantaneous feature data and spectral feature data of the EEG signal obtained according to steps S31 and S32. Wavelet transform or instantaneous frequency analysis technology can be used to identify the instantaneous features of spectral changes. By setting thresholds (such as signal amplitude, frequency offset, etc.), the instantaneous EEG signals that show significant changes in the adversarial test are screened out. The screening process needs to consider the sudden changes or rapid changes of the signals, which are usually related to anti-interference. The screened instantaneous change signals are marked to ensure that they are caused by interference rather than physiological factors. Verification can be carried out manually or by comparing with the signals in the normal state. Using common phase synchrony measurement methods, such as phase locking value (PLV) or phase difference calculation, analyze the phase synchrony between brain regions. The calculation formula is: where and respectively represent the instantaneous phases of the signals in two brain regions, and N is the number of signal sampling points. The number of signal sampling points. By dividing multiple time windows, perform time series analysis on the EEG signal. The sliding window method can be used to observe the changes of the signal in different time periods. The phase synchrony differences within each time window are extracted and analyzed. Within different time windows, calculate the changes in phase synchrony between brain regions (such as the changes in phase locking value). Methods such as dynamic time warping (DTW) can be used to quantify the changes in phase synchrony over time. According to the changes in phase synchrony across time windows, generate brain region phase synchrony data, which further reflects the collaborative working mode between different brain regions and its changes in the time dimension. Based on the phase synchrony data, construct a functional connection matrix. This matrix represents the degree of synchrony or functional connectivity between each brain region. Indicators such as correlation coefficient, mutual information, and phase synchrony can be used to construct the matrix: C ij =Correlation(Phase i ,Phase j ); where, C ij represents the connection strength between brain region i and brain region j. Use the functional connection matrix to perform time series analysis on the brain region phase synchrony data and track the dynamic changes of the functional connection between brain regions. Common analysis methods include sliding window analysis, time series clustering, and dynamic network analysis, etc. By analyzing the temporal changes of the functional connection between brain regions, generate dynamic functional connection data, which is used to characterize the variation law of the EEG signal in time and its relationship with anti-interference. By analyzing the dynamic functional connection data, extract the stable phase synchrony features. For example, indicators such as the volatility, stability, or duration of each connection item in the functional connection matrix can be calculated: S ij =Stability(Cij ), where S ij represents the functional connectivity stability between brain region i and brain region j. By analyzing the stability of the functional connectivity matrix, anti-interference features are extracted, which can be indicators of the stability of connection strength, the stability of phase difference, or the persistence of synchrony. Stable functional connectivity features usually mean that the signal is not strongly interfered, while unstable features are manifestations of adversarial samples. The extracted anti-interference features are stored in the user feature template library for subsequent identification and analysis of interference samples. Classifiers (such as SVM, decision tree, etc.) can be used to further process and identify the anti-interference features.

[0125] As an example of the present invention, refer to Figure 3 shown, in this example, step S4 includes:

[0126] Step S41: Obtain the electroencephalogram signals of the user's multi-session data;

[0127] Step S42: Import the electroencephalogram signals of the user's multi-session data into the anti-interference user feature template library for similarity discrimination, and generate a similarity discrimination result;

[0128] Step S43: Compare the similarity discrimination result with a preset standard similarity threshold. When the similarity discrimination result is greater than or equal to the preset standard similarity threshold, the identity determination result is determined to be true, and the electroencephalogram signals of the corresponding user's multi-session data are uploaded to the blockchain for user biometric binding to generate a user identity key;

[0129] Step S44: When the similarity discrimination result is less than the preset standard similarity threshold, the identity determination result is determined to be false, and an identity verification failure prompt is generated.

[0130] In the embodiments of the present invention, electroencephalogram (EEG) signals of a user are collected in multiple sessions by using an EEG acquisition device (such as an EEG cap, a portable EEG acquisition instrument, etc.). These signals should cover the EEG activity performance of the user at different times and in different situations. Special attention should be paid to the quality of the EEG signals collected in each session to ensure that the data is clear, free from noise interference, and accurate timestamps are recorded during the acquisition process. The multi-session data includes EEG signals in multiple different sessions, reflecting the EEG characteristics of the user in various states (such as awake, attentive, relaxed, thinking, etc.), ensuring diversity and representativeness. The data collection can be arranged in multiple different sessions to ensure that the data in each session is rich enough for subsequent analysis. The EEG signals collected in each session are uploaded to an anti-interference user feature template library. When uploading, the EEG signals can extract key anti-interference features (such as phase synchrony, functional connectivity matrix, etc.) through specific algorithms, and these features are stored as templates. The template library should contain anti-interference feature templates of multiple users to facilitate comparison with the signals of the current user. Similarity calculation methods (such as cosine similarity, Euclidean distance, dynamic time warping (DTW), etc.) are used to compare the EEG signals of the current user with the templates of other users stored in the template library. The purpose of similarity calculation is to measure the similarity degree of the user's EEG signals. If the similarity between the current signal and a certain template in the template library is high, it indicates that the signal belongs to this user; if the similarity is low, it is an adversarial sample or the signal of other users. According to the result of similarity calculation, a similarity discrimination result is generated, which is usually a value representing the similarity between the EEG signals of the current user and the best-matching template in the template library. The system should set a standard similarity threshold (such as 0.85 or other appropriate value), and this value is used as the basis for judging whether the user identity verification is successful. This threshold should be adjusted according to the actual situation to ensure the effective identification of adversarial samples. If the similarity discrimination result is greater than or equal to the preset standard similarity threshold, the identity verification is determined to be passed, and it is considered that the current user is the registered legitimate user. When the identity verification is successful, the EEG signal data of the user and the corresponding biometric binding information are uploaded to the blockchain. The role of the blockchain is to ensure the immutability and transparency of the data, ensuring that the biometrics of the user are not forged or tampered with during subsequent verification processes. The user data can generate a corresponding user identity key through a hash algorithm and be recorded on the blockchain. This key can be used as a credential for subsequent identity verification. The EEG signal characteristics of the user in multiple sessions are bound to the user identity key. Through the smart contract on the blockchain, it is ensured that only legitimate users can access and modify this data. If the similarity discrimination result is less than the preset standard similarity threshold, it indicates that the EEG signals of the current user do not match the data in the template library, and the system determines that the identity verification fails.In the case of authentication failure, the system needs to generate a prompt message for authentication failure, including "Authentication failed, please try again" or "Identity data does not match, please check the signal quality", etc. To prevent malicious users from using adversarial samples for identity impersonation, further verification requirements can be added to the prompt, such as asking the user to re-enter the password or perform other biometric verifications.

[0131] Preferably, uploading the EEG signals of the corresponding user's multi-session data to the blockchain for user biometric binding includes:

[0132] Performing homomorphic encryption on the EEG signals of the corresponding user's multi-session data to generate encrypted EEG signals and encryption keys;

[0133] Using blockchain technology to perform blockchain evidence storage on the encrypted EEG signals to obtain EEG encrypted storage feature data;

[0134] Performing feature binding on the EEG encrypted storage feature data and the encryption keys to generate a revocable key;

[0135] Performing key pairing on the encryption keys and the revocable keys to obtain the user identity key.

[0136] In the embodiments of the present invention, homomorphic encryption is a technique that allows computations to be performed on encrypted data without decrypting the data. In this scenario, common homomorphic encryption schemes include Paillier encryption, BFV (Brakerski / Fan-Vaikuntanathan) encryption scheme, and CKKS encryption (used to support the encryption and computation of floating-point data). Select a suitable homomorphic encryption scheme for data protection. In this step, the EEG signals collected in each session are encrypted through the selected homomorphic encryption algorithm. During the encryption process, all the original information of the signals (such as EEG wave frequency, phase synchronization, etc.) will be encrypted into encrypted data to ensure that user privacy is not leaked. The homomorphic encryption algorithm will generate an encryption key, which is used to perform computations on the subsequent encrypted data. The encryption key usually consists of two parts: a public key and a private key. The public key is used to encrypt data, and the private key is used to decrypt. After obtaining the encrypted EEG signals and the encryption key, these data will be further processed and stored. Select a suitable data storage blockchain platform, such as Ethereum, Hyperledger, or other public / private chain platforms. The blockchain platform will ensure the immutability of the data, ensuring that each data submission and storage is recorded. Upload the encrypted EEG signals and the corresponding encryption key to the blockchain. Each time an upload is made, a block storage record will be generated, including the hash value of the encrypted data and metadata. On the blockchain, the encrypted storage feature data of the EEG signals will be stored as an immutable transaction data. Since the data is encrypted, even if a blockchain node receives the data, it is still unable to access the content of the original EEG signals. The stored data will have a unique blockchain hash value as an identifier, and any node can verify the authenticity of the data and can trace and verify the source and integrity of the stored encrypted EEG signals through this hash value. Feature binding refers to associating the encryption features of the encrypted stored EEG signals with the encryption key. Through specific algorithms and binding rules, the features of the encrypted signals are merged with the encryption key to generate a new key - the Revocable Key. The Revocable Key can be regarded as a key with a revocation function, which allows authorized access to the encrypted data and can revoke the access permission to the data under certain conditions. For example, if the user's identity changes or the permissions are adjusted, the validity of this key can be revoked. Use an encryption algorithm (such as a hash algorithm or a key derivation function) to bind the encrypted signal with the key. The hash value of the encrypted signal, the encryption key, and other security features (such as timestamp, session ID, etc.) will be merged into a key, which can be used for subsequent access but also has the ability to be revoked. Pair the encryption key and the Revocable Key. The pairing process usually includes combining the encryption key and the revocation key into a combined key, so that during subsequent verification, both the integrity of the data and the revocation mechanism of the key can be ensured.The paired keys will generate a User Identity Key. The User Identity Key is a pair (public key, private key), where the public key is used for data verification and encryption, and the private key is used for authentication, data decryption, or revocation operations. The User Identity Key will be encrypted and stored in a secure environment. It can be optionally stored on the blockchain for management and revocation through smart contracts to ensure the security and access control of the keys.

[0137] Therefore, in all respects, the embodiments should be considered exemplary and non-restrictive. The scope of the present invention is defined by the appended claims rather than the above description. Thus, all changes that fall within the meaning and scope of the equivalent elements of the application document are intended to be embraced within the present invention.

[0138] The above description is only a specific implementation manner of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather to the broadest scope consistent with the principles and novel features disclosed herein.

Claims

1. An authentication method for identifying a user's identity based on brain wave characteristics, characterized in that, Including the following steps: Step S1: Collect the electroencephalogram (EEG) signals of the user under a preset cognitive task using an EEG device, perform noise baseline calibration on the EEG signals, generate standard EEG signals, and perform AES encryption processing on the EEG samples; Step S2: Generate corresponding adversarial samples for the standard EEG signals through an adversarial generation network; Apply a spatio-temporal dynamic masking strategy to inject EEG chaos features into the standard EEG signals to obtain EEG chaos injection features; Use the generated corresponding adversarial samples to perform adversarial dynamic feature enhancement training on the EEG chaos injection features to generate an adversarial test result; Step S3: Extract the time-frequency features and functional connectivity matrix of the adversarial test result, and construct an anti-interference user feature template library; Step S4: Obtain the EEG signals of the user's multi-session data, import the EEG signals of the user's multi-session data into the anti-interference user feature template library for identity determination. If the identity determination result is true, upload the EEG signals of the corresponding user's multi-session data to the blockchain for user biometric binding to generate a user identity key.

2. The authentication method for identifying a user's identity based on brain wave characteristics according to claim 1, wherein Step S1 includes the following steps: Step S11: Collect the EEG signals of the user using an EEG device to obtain EEG signals; Step S12: Based on a synchronous acquisition device, collect eye movement signals and electromyogram signals synchronized with the EEG signals; Step S13: Design a preset cognitive task, where the preset cognitive task includes visual stimulus-induced P300 potential, multi-level mental arithmetic tasks, and cross-modal association tasks; Step S14: Identify the interference signal sources for the EEG signals according to the eye movement signals and electromyogram signals to generate noise source identification data; Perform noise baseline calibration on the noise source identification data to generate noise baseline calibration data; Step S15: Optimize the signal quality of the EEG signals through the noise baseline calibration data to generate standard EEG signals, perform AES encryption processing on the EEG samples, and perform AES encryption processing on the EEG samples.

3. The authentication method for identifying a user's identity based on brain wave characteristics according to claim 1, wherein The generation of corresponding adversarial samples for the standard EEG signals in Step S2 includes: Add low-frequency drift component noise to the standard EEG signals to obtain low-frequency drift component noise; Simulate the drift amplitude enhancement of the standard EEG signals according to the low-frequency drift component noise to generate EEG signal data with drift noise and label it as the first type of sample; Randomly add high-frequency noise components to the standard EEG signals to obtain high-frequency random component noise; Simulate environmental interference for the standard EEG signals according to the high-frequency random component noise to generate EEG signal data with environmental interference noise and label it as the second type of sample; Use adversarial generation network technology to embed forged EEG activities into the standard EEG signals to generate forged EEG feature signals; Simulate spoofing attacks on the standard EEG signals through the forged EEG feature signals to generate EEG signal data with spoofing attacks and label it as the third type of sample; Integrate the first type of sample, the second type of sample, and the third type of sample into a comprehensive adversarial sample.

4. The authentication method for identifying a user's identity based on brain wave characteristics according to claim 1, characterized in that The injection of EEG chaos features into the standard EEG signals by applying the spatio-temporal dynamic masking strategy in Step S2 includes: Randomly select several frequency bands from the standard EEG signal, randomly discard segments of the selected frequency bands to generate randomly discarded segment data; perform temporal gap interpolation on the standard EEG signal based on the randomly discarded segment data to generate a recombined EEG signal; Randomly select electrode channels from the standard EEG signal, and randomly shield the standard EEG signals of the corresponding electrode channels to generate channel shielding data; zero the signal values of the standard EEG signal based on the channel shielding data to generate a shielded EEG signal; Perform spectrum analysis on the standard EEG signal, extract the frequency bands of interest and inject proportional noise to generate a noise-injected signal; analyze the abnormal signal characteristics of the recombined EEG signal, the shielded EEG signal, and the noise-injected signal, and integrate them into a spatio-temporal dynamic masking strategy; Inject EEG chaos characteristics into the standard EEG signal based on the spatio-temporal dynamic masking strategy using the abnormal signal characteristics to obtain EEG chaos injection characteristics.

5. The authentication method for identifying a user's identity based on brain wave characteristics according to claim 1, wherein, The adversarial dynamic feature enhancement training of the EEG chaos injection characteristics using the generated corresponding adversarial samples in step S2 includes: Analyze the distribution differences of EEG devices for the EEG chaos injection characteristics to generate device distribution difference data; use the device distribution difference data to perform cross-device distribution alignment training on the generated corresponding adversarial samples to generate cross-device aligned signal feature data; Construct a live detection scenario; perform live detection adversarial training on the generated corresponding adversarial samples through the live detection scenario to generate enhanced live detection feature data; Integrate the cross-device aligned signal feature data, the generated corresponding adversarial samples, and the enhanced live detection feature data into a joint training set; perform adversarial testing on the joint training set according to the adversarial generation network to generate an adversarial test result.

6. The authentication method for identifying a user's identity based on brain wave characteristics according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Perform time-frequency analysis on the adversarial test result to generate instantaneous EEG signal feature data; Step S32: Perform short-time Fourier transform on the instantaneous EEG signal feature data to generate instantaneous EEG signal spectrum data; extract the spectral characteristics of the instantaneous EEG signal spectrum data to obtain instantaneous EEG signal spectral feature data; Step S33: Calculate the phase synchrony based on the instantaneous EEG signal feature data and the instantaneous EEG signal spectral feature data to obtain inter-brain region phase synchrony data; construct a functional connectivity matrix for the adversarial test result based on the inter-brain region phase synchrony data and extract the anti-interference characteristics of the functional connectivity matrix; Step S34: Construct an anti-interference user feature template library for the adversarial samples through the anti-interference characteristics to generate an anti-interference user feature template library.

7. The authentication method for identifying a user's identity based on brain wave characteristics according to claim 6, wherein, Step S33 includes the following steps: Step S331: Screen the adversarial test result for instantaneous changing EEG signals based on the instantaneous EEG signal feature data and the instantaneous EEG signal spectral feature data to obtain instantaneous changing EEG signals; Step S332: Calculate the inter-brain region phase synchrony for the instantaneous changing EEG signals to obtain the inter-brain region phase difference; Step S333: Analyze the phase synchrony change across time windows for the instantaneous changing EEG signals through the inter-brain region phase difference to generate inter-brain region phase synchrony data; Step S334: Construct a functional connectivity matrix based on the phase synchronization change data, and perform dynamic functional connectivity analysis on the brain region phase synchronization data using the functional connectivity matrix to generate dynamic functional connectivity data; Step S335: Extract the stable phase features of the functional connectivity matrix using the dynamic functional connectivity data to obtain the anti-interference features of the functional connectivity matrix.

8. The authentication method for identifying a user's identity based on brain wave characteristics according to claim 1, characterized in that, Step S4 includes the following steps: Step S41: Obtain the electroencephalogram (EEG) signals of the user's multi-session data; Step S42: Import the EEG signals of the user's multi-session data into the anti-interference user feature template library for similarity discrimination to generate a similarity discrimination result; Step S43: Compare the similarity discrimination result with a preset standard similarity threshold. When the similarity discrimination result is greater than or equal to the preset standard similarity threshold, it is determined that the identity determination result is true, and the EEG signals of the corresponding user's multi-session data are uploaded to the blockchain for user biometric binding to generate a user identity key; Step S44: When the similarity discrimination result is less than the preset standard similarity threshold, it is determined that the identity determination result is false, and an identity verification failure prompt is generated.

9. The authentication method for identifying a user's identity based on brain wave characteristics according to claim 8, wherein The uploading the EEG signals of the corresponding user's multi-session data to the blockchain for user biometric binding includes: Performing homomorphic encryption on the EEG signals of the corresponding user's multi-session data to generate encrypted EEG signals and an encryption key; Performing blockchain evidence storage on the encrypted EEG signals through blockchain technology to obtain EEG encrypted storage feature data; Binding the EEG encrypted storage feature data and the encryption key to generate a revocable key; Pairing the encryption key and the revocable key to obtain a user identity key.

10. An authentication system for identifying a user's identity based on brain wave characteristics, characterized in that, For implementing the authentication method for identifying a user's identity based on brain wave features as described in claim 1, the authentication system for identifying a user's identity based on brain wave features includes: A signal acquisition module, configured to collect the EEG signals of the user under a preset cognitive task using an EEG device, perform noise baseline calibration on the EEG signals, generate standard EEG signals, and perform AES encryption processing on the EEG samples; An adversarial training module, configured to generate corresponding adversarial samples for the standard EEG signals through an adversarial generation network; inject EEG chaos features into the standard EEG signals using a spatio-temporal dynamic masking strategy to obtain EEG chaos injection features; perform adversarial dynamic feature enhancement training on the EEG chaos injection features using the generated corresponding adversarial samples to generate an adversarial test result; An anti-interference feature extraction module, configured to extract the time-frequency features and the functional connectivity matrix of the adversarial test result, and construct an anti-interference user feature template library; An identity verification module, configured to obtain the EEG signals of the user's multi-session data, import the EEG signals of the user's multi-session data into the anti-interference user feature template library for identity determination, and if the identity determination result is true, upload the EEG signals of the corresponding user's multi-session data to the blockchain for user biometric binding to generate a user identity key.

Images