Electroencephalogram identity recognition method and system based on generative adversarial network

Abstract

The invention discloses an electroencephalogram identity recognition method and system based on a generative adversarial network, and the method comprises the steps: generating an electroencephalogram signal matrix in a conditional generative adversarial network; frequency domain features of the electroencephalogram signals are extracted through short-time Fourier transform, time domain space features of the electroencephalogram signals are extracted through a convolutional neural network, and electroencephalogram features are obtained; based on the electroencephalogram signal matrix and the electroencephalogram characteristics, electroencephalogram signals are obtained; and performing dual discrimination on the electroencephalogram signals based on a deep residual network, determining the authenticity of the electroencephalogram signals, and completing identity recognition. Therefore, the generative adversarial network architecture is adopted, high-quality simulated electroencephalogram data can be generated, the strong forgery detection capability is achieved, and identity false use is effectively prevented. Through combination of multi-scale feature extraction and a deep residual network, fine feature extraction and accurate identity discrimination of electroencephalogram signals are realized.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically, to a brainwave identity recognition method and system based on generative adversarial networks. Background Technology

[0002] With the rapid development of generative artificial intelligence technology, traditional biometric identification technologies such as fingerprint, voiceprint, and facial recognition face serious security challenges. These biometric features are easily forged and copied by artificial intelligence technology, threatening the security of identity recognition systems.

[0003] Electroencephalogram (EEG) signals, as a unique biometric feature, are difficult to forge, can be detected in vivo, and exhibit significant individual variability, making them a cutting-edge research area in identity verification in recent years. However, existing EEG-based identity verification technologies still have many limitations: first, EEG signal acquisition is significantly affected by environmental noise and physiological state; second, traditional recognition models do not sufficiently extract signal features; and third, there is a lack of effective anti-forgery detection mechanisms.

[0004] Especially in emerging application scenarios such as the metaverse, there is a need to achieve in vitro generation and high-precision recognition of EEG signals, which places higher demands on EEG identity recognition technology. Existing EEG identity recognition technologies struggle to simultaneously achieve high realism and accuracy of individual characteristics, failing to meet the requirements for high-security identity authentication. Summary of the Invention

[0005] According to the present invention, a method and system for EEG identity recognition based on generative adversarial networks are provided to solve the technical problem that existing EEG identity recognition technologies cannot simultaneously achieve high realism and accuracy of individual characteristics, thus failing to meet the requirements of high-security identity authentication.

[0006] According to a first aspect of the present invention, a brainwave identity recognition method based on generative adversarial networks is provided, comprising: Generate an EEG signal matrix in a conditional generative adversarial network; The frequency domain features of the EEG signal are extracted using short-time Fourier transform, and the temporal and spatial features of the EEG signal are extracted using a convolutional neural network to obtain the EEG features. Based on the EEG signal matrix and the EEG features, EEG signals are obtained; The deep residual network performs dual discrimination on the EEG signal to determine the authenticity of the EEG signal and complete the identity recognition.

[0007] Optionally, the EEG signal matrix is ​​generated in a conditional generative adversarial network, including: The subject ID number is mapped to a 64-dimensional continuous vector through the embedding layer; The ID vector is concatenated with the temporal spatial features extracted from multi-scale features, and then fused through a fully connected layer. A 5×1250-dimensional EEG signal matrix is ​​output through the fully connected layer of the decoder and the Reshape operation.

[0008] Optionally, short-time Fourier transform is used to extract the frequency domain features of the EEG signal, and a convolutional neural network is used to extract the temporal and spatial features of the EEG signal to obtain EEG features, including: Frequency domain features of channels Fz, Cz, Oz, C3, and C5 were extracted using short-time Fourier transform. Temporal spatial features are extracted using a three-level cascaded convolutional network, and the following steps are performed sequentially: Temporal convolution: A 1×125 kernel is used to capture long-range temporally correlated patterns, and the activation function is ELU; Depthwise convolution: Compresses spatial dimensions through grouped convolution with 5×1 kernels, and combines Dropout to suppress overfitting; Separate convolution: Perform 1×16 grouped convolution to capture local temporal features, and then fuse cross-channel information through 1×1 pointwise convolution; L2 normalization constraints are applied to the weights of the first layer of the deep convolutional layer, and EEG features are obtained based on frequency domain features and temporal spatial features.

[0009] Optionally, a dual discrimination method is performed on the EEG signal based on a deep residual network to determine the authenticity of the EEG signal and complete identity recognition, including: The EEG features are input into the deep residual module for feature reconstruction and enhancement. The probability of the authenticity of the EEG signal is output by the Sigmoid function to determine whether the EEG signal is a real signal or a generated signal. Identity recognition is achieved by outputting the classification probability of EEG signals through the Softmax function.

[0010] According to another aspect of the present invention, a brainwave identity recognition system based on generative adversarial networks is also provided, comprising: The signal matrix generation module is used to generate EEG signal matrices in conditional generative adversarial networks. The EEG feature acquisition module is used to extract the frequency domain features of EEG signals using short-time Fourier transform and extract the temporal spatial features of EEG signals using a convolutional neural network to acquire EEG features. A module for obtaining EEG signals is used to obtain EEG signals based on the EEG signal matrix and the EEG features; The identity recognition module is completed, which is used to perform dual discrimination on EEG signals based on deep residual networks, determine the authenticity of EEG signals, and complete identity recognition.

[0011] Optionally, the signal matrix generation module includes: The mapping vector submodule is used to map the subject ID number into a 64-dimensional continuous vector through the embedding layer; The feature concatenation submodule is used to concatenate the ID vector with the temporal spatial features extracted from multi-scale features, and then fuse them through a fully connected layer; The signal matrix generation submodule is used to output a 5×1250-dimensional EEG signal matrix through the fully connected layer of the decoder and the Reshape operation.

[0012] Optionally, the EEG feature module includes: The pre-frequency domain feature submodule is used to extract the frequency domain features of channels Fz, Cz, Oz, C3, and C5 through short-time Fourier transform; The pre-temporal spatial feature submodule is used to extract temporal spatial features based on a three-level cascaded convolutional network, and executes sequentially as follows: Temporal convolution: A 1×125 kernel is used to capture long-range temporally correlated patterns, and the activation function is ELU; Depthwise convolution: Compresses spatial dimensions through grouped convolution with 5×1 kernels, and combines Dropout to suppress overfitting; Separate convolution: Perform 1×16 grouped convolution to capture local temporal features, and then fuse cross-channel information through 1×1 pointwise convolution; The EEG feature acquisition submodule is used to apply L2 normalization constraints to the weights of the first layer of the deep convolutional layer and acquire EEG features based on frequency domain features and temporal spatial features.

[0013] Optionally, the identity recognition module is completed, including: The EEG features are input into the deep residual module for feature reconstruction and enhancement. The probability of the authenticity of the EEG signal is output by the Sigmoid function to determine whether the EEG signal is a real signal or a generated signal. Identity recognition is achieved by outputting the classification probability of EEG signals through the Softmax function.

[0014] According to another aspect of the invention, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the steps of any of the methods described herein.

[0015] According to another aspect of the present invention, an electronic device is also provided, comprising: The aforementioned computer-readable storage medium; and One or more processors for executing a program in the computer-readable storage medium.

[0016] Therefore, by employing a generative adversarial network (GAN) architecture, high-quality simulated EEG data can be generated, and powerful forgery detection capabilities are available to effectively prevent identity theft. Through the combination of multi-scale feature extraction and deep residual networks, refined feature extraction and accurate identity determination of EEG signals are achieved. Optimized model structure and computational flow meet the requirements of real-time identity authentication, with short processing times and fast response speeds. To address the noise problem in EEG signals, a dedicated signal processing and feature extraction mechanism is designed, maintaining high recognition accuracy even in noisy environments. Attached Figure Description

[0017] Exemplary embodiments of the present invention can be more fully understood by referring to the following figures: Figure 1 This is a flowchart illustrating a brainwave identity recognition method based on generative adversarial networks as described in this embodiment. Figure 2 This is a schematic diagram of the overall architecture of the EEG identity recognition model described in this embodiment; Figure 3 This is a schematic diagram illustrating the detailed structure of the multi-scale feature extraction described in this embodiment; Figure 4 This is a schematic diagram of the dual discrimination mechanism of the identity discrimination module described in this embodiment. Figure 5 This is a schematic diagram illustrating the implementation process of the model described in this embodiment in EEG signal recognition; Figure 6 This is a schematic diagram of an EEG identity recognition system based on generative adversarial networks as described in this embodiment. Detailed Implementation

[0018] Exemplary embodiments of the invention will now be described with reference to the accompanying drawings. However, the invention may be embodied in many different forms and is not limited to the embodiments described herein. These embodiments are provided to fully and completely disclose the invention and to fully convey its scope to those skilled in the art. The terminology used in the exemplary embodiments illustrated in the drawings is not intended to limit the invention. In the drawings, the same units / elements are referred to by the same reference numerals.

[0019] Unless otherwise stated, the terms used herein (including technical terms) have their common meaning as understood by one of ordinary skill in the art. Furthermore, it is understood that terms defined in commonly used dictionaries should be understood to have a meaning consistent with the context of their relevant field, and not to be interpreted as having an idealized or overly formal meaning.

[0020] According to a first aspect of the present invention, a brainwave identity recognition method 100 based on generative adversarial networks is provided, with reference to... Figure 1 As shown, the method 100 includes: S101: Generate an EEG signal matrix in a conditional generative adversarial network; S102: Short-time Fourier transform is used to extract the frequency domain features of EEG signals, and convolutional neural network is used to extract the temporal and spatial features of EEG signals to obtain EEG features; S103: Obtain the brainwave signal based on the brainwave signal matrix and the brainwave features; S104: Based on a deep residual network, perform dual discrimination on EEG signals to determine the authenticity of the EEG signals and complete identity recognition.

[0021] Specifically, Figure 2 The diagram shows the overall architecture of the EEG identity recognition model of this invention, including the entire GAN network framework, divided into a generator (left image) and a discriminator (right image). The generator and discriminator have roughly the same structure, both containing three parts: an encoder, a hidden layer, and a decoder. Their inputs and outputs are different. The encoder extracts features through a multi-scale feature extraction module, balancing efficiency and expressive power. The specific structure of the multi-scale feature extraction module will be discussed later. Figure 2 The explanation is as follows: Hidden Layer: Enables multimodal feature fusion and transformation to adapt to the generated target. Target ID Embedding (23 classes → 64 dimensions): Converts discrete IDs into continuous vectors, allowing the model to learn semantic relationships between IDs (e.g., embedding vectors of similar IDs are closer), which is more efficient than one-hot encoding; Feature Concatenation + Fully Connected Layer (default 1 layer, 1024 nodes): Fuses the spatial features of the encoder output with the semantic features of the ID embedding, flexibly adjusting the nonlinear transformation capability through configurable fully connected layers and nodes to adapt to generation tasks of different complexities. Decoder: Accurately maps to the target output form, ensuring consistency of the generated data format. Fully Connected Layer + Reshape: Directly maps the hidden layer features to a vector of target height × width, and then reshapes it into a single-channel feature map, quickly realizing the transformation from abstract features to concrete output, ensuring that the output size strictly matches the requirements. The discriminator differs from the generator in the input and output ends. At the input end, the generator accepts the subject's ID and a random seed. The subject's ID is added as a label to the hidden layer, and the random seed is used to control random noise. The discriminator receives EEG signals as input. During training, it receives EEG signals generated by the generator for adversarial training; when identifying real and fake signals, it receives real-time EEG signals. At the output end, the generator generates EEG signals with ontological features based on the label information, and the discriminator outputs its judgment of the EEG signal, i.e., whether it is a real EEG signal and the subject's ID.

[0022] Figure 3The diagram shows the detailed structure of the multi-scale feature extraction module of this invention. The multi-scale feature extraction module, as a feature extraction tool, is adapted into the generator and discriminator. The multi-scale feature extraction module contains three cascaded convolutional layers: a temporal convolutional layer, a depthwise convolutional layer, and a segregating convolutional layer. Temporal convolutional layer (1×125 kernels): Uses a narrow kernel to focus on long-range features in the time dimension (such as local patterns in time series), and "same" padding preserves the time dimension size, suitable for capturing continuous temporal correlations; Deep convolutional layer (5×1 kernels, number of groups = 5): Only compresses spatial height (preserves the time dimension), grouped convolutions reduce parameter redundancy (4 times fewer parameters than ordinary convolutions), and combined with ELU activation and Dropout (50%), enhances non-linear expression and suppresses overfitting; Separate convolutional layer (1×16 grouped convolutions + 1×1 pointwise convolutions): First, captures local temporal features through grouped convolutions, and then uses 1×1 convolutions to fuse channel information, balancing computational efficiency (fewer parameters than ordinary convolutions of the same size) and cross-channel feature fusion. Subsequent pooling further compresses the dimension and reduces computational cost; Weight L2 normalization: Constrains the norm of the first layer weights of the deep convolution to ≤1, avoids gradient explosion, and enhances training stability.

[0023] Figure 4 This is a schematic diagram of the dual discrimination mechanism of the identity discrimination module of the present invention. The identity discrimination module is mainly concentrated in the latter half of the discriminator, namely the decoder part. In the decoder part of the discriminator, the invention uses a deep residual module to transform and reorganize the features of the hidden layer, and further enhance the information of the features. Finally, the features are mapped to the binary classification space and the 23-class classification space respectively by the sigmoid function to represent the recognition results.

[0024] Figure 5 This is a flowchart illustrating the implementation process of the model in EEG signal recognition according to the present invention. Before real-time EEG signal recognition, the model needs to be trained using real data, with the training data provided by universities. The trained model can then be loaded onto a PyTorch-based code platform. At this point, a device for capturing EEG signals can be used to sample the human body to obtain real-time EEG data. This data can then be input into the discriminator model for recognition, yielding the recognition result.

[0025] The EEG identity recognition model of this invention is implemented using Python based on the PyTorch framework. First, a conditional generative adversarial network (GAN) is constructed. The generator receives the subject's ID and a random seed as input and outputs a 5×1250 EEG signal data matrix. The discriminator employs a multi-scale feature extraction architecture, simultaneously processing time-domain and frequency-domain features.

[0026] During the model training phase, a resting-state EEG dataset was used, containing data collected from 90 subjects across different dates. An adversarial training strategy was employed during training, alternately optimizing the generator and discriminator.

[0027] In the real-time recognition phase, after receiving EEG signal data packets, the system first performs preprocessing and feature extraction, and then outputs the authenticity judgment and identity recognition results through the identity verification module. The entire processing meets real-time requirements, with the processing time for a single operation controlled within milliseconds.

[0028] The model supports multiple triggering mechanisms and can perform corresponding data generation or recognition tasks based on different trigger numbers. When a Trial start marker (Trigger 1-23) is received, EEG data is generated based on the specified subject ID; when a Block end marker (Trigger 243) is received, the model enters the termination state.

[0029] Therefore, by employing a generative adversarial network (GAN) architecture, high-quality simulated EEG data can be generated, and powerful forgery detection capabilities are available to effectively prevent identity theft. Through the combination of multi-scale feature extraction and deep residual networks, refined feature extraction and accurate identity determination of EEG signals are achieved. Optimized model structure and computational flow meet the requirements of real-time identity authentication, with short processing times and fast response speeds. To address the noise problem in EEG signals, a dedicated signal processing and feature extraction mechanism is designed, maintaining high recognition accuracy even in noisy environments.

[0030] Optionally, in a conditional generative adversarial network, an EEG signal matrix is ​​generated, including: The subject ID number is mapped to a 64-dimensional continuous vector through the embedding layer; The ID vector is concatenated with the temporal spatial features extracted from multi-scale features, and then fused through a fully connected layer. A 5×1250-dimensional EEG signal matrix is ​​output through the fully connected layer of the decoder and the Reshape operation.

[0031] Optionally, short-time Fourier transform is used to extract the frequency domain features of the EEG signal, and a convolutional neural network is used to extract the temporal and spatial features of the EEG signal to obtain EEG features, including: Frequency domain features of channels Fz, Cz, Oz, C3, and C5 were extracted using short-time Fourier transform. Temporal spatial features are extracted using a three-level cascaded convolutional network, and the following steps are performed sequentially: Temporal convolution: A 1×125 kernel is used to capture long-range temporally correlated patterns, and the activation function is ELU; Depthwise convolution: Compresses spatial dimensions through grouped convolution with 5×1 kernels, and combines Dropout to suppress overfitting; Separate convolution: Perform 1×16 grouped convolution to capture local temporal features, and then fuse cross-channel information through 1×1 pointwise convolution; L2 normalization constraints are applied to the weights of the first layer of the deep convolutional layer, and EEG features are obtained based on frequency domain features and temporal spatial features.

[0032] Optionally, a dual discrimination method is performed on the EEG signal based on a deep residual network to determine the authenticity of the EEG signal and complete identity recognition, including: The EEG features are input into the deep residual module for feature reconstruction and enhancement. The probability of the authenticity of the EEG signal is output by the Sigmoid function to determine whether the EEG signal is a real signal or a generated signal. Identity recognition is achieved by outputting the classification probability of EEG signals through the Softmax function.

[0033] Therefore, by employing a generative adversarial network (GAN) architecture, high-quality simulated EEG data can be generated, and powerful forgery detection capabilities are available to effectively prevent identity theft. Through the combination of multi-scale feature extraction and deep residual networks, refined feature extraction and accurate identity determination of EEG signals are achieved. Optimized model structure and computational flow meet the requirements of real-time identity authentication, with short processing times and fast response speeds. To address the noise problem in EEG signals, a dedicated signal processing and feature extraction mechanism is designed, maintaining high recognition accuracy even in noisy environments.

[0034] According to another aspect of the present invention, a brainwave identification system 600 based on generative adversarial networks is also provided, with reference to... Figure 6 As shown, the system 600 includes: The signal matrix generation module 610 is used to generate an EEG signal matrix in a conditional generative adversarial network. The EEG feature acquisition module 620 is used to extract the frequency domain features of the EEG signal using short-time Fourier transform and extract the temporal spatial features of the EEG signal through a convolutional neural network to acquire EEG features. The EEG signal acquisition module 630 is used to acquire EEG signals based on the EEG signal matrix and the EEG features; The identity recognition module 640 is completed. It is used to perform dual discrimination on EEG signals based on deep residual networks, determine the authenticity of EEG signals, and complete identity recognition.

[0035] Optionally, the signal matrix generation module includes: The mapping vector submodule is used to map the subject ID number into a 64-dimensional continuous vector through the embedding layer; The feature concatenation submodule is used to concatenate the ID vector with the temporal spatial features extracted from multi-scale features, and then fuse them through a fully connected layer; The signal matrix generation submodule is used to output a 5×1250-dimensional EEG signal matrix through the fully connected layer of the decoder and the Reshape operation.

[0036] Optionally, the EEG feature module includes: The pre-frequency domain feature submodule is used to extract the frequency domain features of channels Fz, Cz, Oz, C3, and C5 through short-time Fourier transform; The pre-temporal spatial feature submodule is used to extract temporal spatial features based on a three-level cascaded convolutional network, and executes sequentially as follows: Temporal convolution: A 1×125 kernel is used to capture long-range temporally correlated patterns, and the activation function is ELU; Depthwise convolution: Compresses spatial dimensions through grouped convolution with 5×1 kernels, and combines Dropout to suppress overfitting; Separate convolution: Perform 1×16 grouped convolution to capture local temporal features, and then fuse cross-channel information through 1×1 pointwise convolution; The EEG feature acquisition submodule is used to apply L2 normalization constraints to the weights of the first layer of the deep convolutional layer and acquire EEG features based on frequency domain features and temporal spatial features.

[0037] Optionally, the identity recognition module is completed, including: The EEG features are input into the deep residual module for feature reconstruction and enhancement. The probability of the authenticity of the EEG signal is output by the Sigmoid function to determine whether the EEG signal is a real signal or a generated signal. Identity recognition is achieved by outputting the classification probability of EEG signals through the Softmax function.

[0038] The EEG identity recognition system 600 based on generative adversarial networks in one embodiment of the present invention corresponds to the EEG identity recognition method 100 based on generative adversarial networks in another embodiment of the present invention, and will not be described again here.

[0039] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0040] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0041] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0042] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0043] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0044] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A brainwave identity recognition method based on generative adversarial networks, characterized in that, include: In a conditional generative adversarial network, an EEG signal matrix is ​​generated; The frequency domain features of the EEG signal are extracted using short-time Fourier transform, and the temporal and spatial features of the EEG signal are extracted using a convolutional neural network to obtain the EEG features. Based on the EEG signal matrix and the EEG features, EEG signals are obtained; The deep residual network performs dual discrimination on the EEG signal to determine the authenticity of the EEG signal and complete the identity recognition.

2. The EEG identity recognition method based on generative adversarial networks according to claim 1, characterized in that, Generate an EEG signal matrix in a conditional generative adversarial network, including: The subject ID number is mapped to a 64-dimensional continuous vector through the embedding layer; The ID vector is concatenated with the temporal spatial features extracted from multi-scale features, and then fused through a fully connected layer. A 5×1250-dimensional EEG signal matrix is ​​output through the fully connected layer of the decoder and the Reshape operation.

3. The EEG identity recognition method based on generative adversarial networks according to claim 1, characterized in that, The frequency domain features of the EEG signal are extracted using short-time Fourier transform, and the temporal and spatial features of the EEG signal are extracted using a convolutional neural network to obtain EEG features, including: Frequency domain features of channels Fz, Cz, Oz, C3, and C5 were extracted using short-time Fourier transform. Temporal spatial features are extracted using a three-level cascaded convolutional network, and the following steps are performed sequentially: Temporal convolution: A 1×125 kernel is used to capture long-range temporally correlated patterns, and the activation function is ELU; Depthwise convolution: Compresses spatial dimensions through grouped convolution with 5×1 kernels, and combines Dropout to suppress overfitting; Separate convolution: Perform 1×16 grouped convolution to capture local temporal features, and then fuse cross-channel information through 1×1 pointwise convolution; L2 normalization constraints are applied to the weights of the first layer of the deep convolutional layer, and EEG features are obtained based on frequency domain features and temporal spatial features.

4. The EEG identity recognition method based on generative adversarial networks according to claim 1, characterized in that, Based on deep residual networks, dual discrimination is performed on EEG signals to determine the authenticity of the EEG signals and complete identity recognition, including: The EEG features are input into the deep residual module for feature reconstruction and enhancement. The probability of the authenticity of the EEG signal is output by the Sigmoid function to determine whether the EEG signal is a real signal or a generated signal. Identity recognition is achieved by outputting the classification probability of EEG signals through the Softmax function.

5. A brainwave identity recognition system based on generative adversarial networks, characterized in that, include: The signal matrix generation module is used to generate EEG signal matrices in conditional generative adversarial networks. The EEG feature acquisition module is used to extract the frequency domain features of EEG signals using short-time Fourier transform and extract the temporal spatial features of EEG signals using a convolutional neural network to acquire EEG features. A module for obtaining EEG signals is used to obtain EEG signals based on the EEG signal matrix and the EEG features; The identity recognition module is completed, which is used to perform dual discrimination on EEG signals based on deep residual networks, determine the authenticity of EEG signals, and complete identity recognition.

6. The EEG identity recognition system based on generative adversarial networks according to claim 5, characterized in that, The signal matrix generation module includes: The mapping vector submodule is used to map the subject ID number into a 64-dimensional continuous vector through the embedding layer; The feature concatenation submodule is used to concatenate the ID vector with the temporal spatial features extracted from multi-scale features, and then fuse them through a fully connected layer; The signal matrix generation submodule is used to output a 5×1250-dimensional EEG signal matrix through the fully connected layer of the decoder and the Reshape operation.

7. The EEG identity recognition system based on generative adversarial networks according to claim 5, characterized in that, The module for acquiring EEG features includes: The pre-frequency domain feature submodule is used to extract the frequency domain features of channels Fz, Cz, Oz, C3, and C5 through short-time Fourier transform; The pre-temporal spatial feature submodule is used to extract temporal spatial features based on a three-level cascaded convolutional network, and executes sequentially as follows: Temporal convolution: A 1×125 kernel is used to capture long-range temporally correlated patterns, and the activation function is ELU; Depthwise convolution: Compresses spatial dimensions through grouped convolution with 5×1 kernels, and combines Dropout to suppress overfitting; Separate convolution: Perform 1×16 grouped convolution to capture local temporal features, and then fuse cross-channel information through 1×1 pointwise convolution; The EEG feature acquisition submodule is used to apply L2 normalization constraints to the weights of the first layer of the deep convolutional layer and acquire EEG features based on frequency domain features and temporal spatial features.

8. The EEG identity recognition system based on generative adversarial networks according to claim 5, characterized in that, Complete the identity recognition module, including: The EEG features are input into the deep residual module for feature reconstruction and enhancement. The probability of the authenticity of the EEG signal is output by the Sigmoid function to determine whether the EEG signal is a real signal or a generated signal. Identity recognition is achieved by outputting the classification probability of EEG signals through the Softmax function.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1-4.

10. An electronic device, characterized in that, include: The computer-readable storage medium as described in claim 9; as well as One or more processors for executing a program in the computer-readable storage medium.

Images