EEG (electroencephalogram) signal-based identity recognition method, device and equipment

Abstract

The invention provides an EEG signal-based identity recognition method, device and equipment, and relates to the technical field of computers, and the method comprises the steps of receiving an initial global model sent by a server side; training the initial global model by using a local data set of each client to obtain a local model weight parameter of each client; determining a comprehensive model weight parameter based on the local model weight parameters of the plurality of clients, and updating the initial global model by using the model comprehensive weight parameter to obtain a target global model; and inputting a to-be-recognized EEG signal into the target global model, and outputting an identity recognition result. According to the method, the federated learning algorithm is utilized, so that the user terminals can cooperatively train the model on the premise of not sharing the original data, accurate identification of the EEG signals of the users is realized, and the security and privacy protection of the data are enhanced.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an identity recognition method, device and equipment based on electroencephalogram (EEG) signals. Background Art

[0002] In today's digital society, as the importance of personal information security continues to highlight, traditional identity recognition methods such as password login have exposed many security risks. In contrast, identity recognition based on biometrics has become a research hotspot due to its inherent uniqueness and non-replicability. Although biometrics such as fingerprints or irises have improved security to a certain extent, there is still a risk of being forged or imitated. Unlike these external biometrics, electroencephalogram (EEG) signals are not easily exposed to potential intruders and are therefore extremely difficult to copy or forge. In addition, since EEG signals are only generated by living brains, this ensures that the user must be in a living state during the identification process, thereby enhancing the security of identity authentication. Ideal biometric factors should meet the following basic requirements: universality, persistence, uniqueness, and collectibility. Universality means that the biometric feature is applicable to everyone rather than a specific group; persistence means that the feature remains unchanged over a certain period of time; uniqueness ensures that each user's biometric feature is unique; and collectibility emphasizes that biometric information should be easy to obtain. Obviously, EEG-based recognition can meet all of the above requirements.

[0003] However, the existing centralized data processing method requires a large amount of EEG data containing personal identity information from different users to be centralized on one client. When all EEG data are stored centrally, once the client device encounters a security vulnerability or is attacked by hackers, the sensitive information of all users may be illegally accessed and abused. EEG data not only contains personal identity information, but may also reflect deep private information such as the individual's emotional state, cognitive ability, and health status. If this data is used improperly, it may cause serious infringement on personal privacy. Summary of the invention

[0004] The purpose of the present invention is to solve the problem that the current EEG recognition is difficult to protect user privacy. The present invention provides an identity recognition method, device and equipment based on electroencephalogram (EEG) signals.

[0005] The technical solution of the embodiment of the present application is implemented as follows: The first aspect of the embodiment of the present application provides an identity recognition method based on EEG signals, comprising: Receive the initial global model sent by the server; Using the local data set of each client to train the initial global model, and obtaining the local model weight parameters of each client; Determine a comprehensive model weight parameter based on the local model weight parameters of the plurality of clients, and use the model comprehensive weight parameter to update the initial global model to obtain a target global model; The EEG signal to be identified is input into the target global model, and the identity identification result is output.

[0006] Optionally, the initial global model includes two two-dimensional convolutional layers and one depth convolutional layer.

[0007] Optionally, the using the local data set of each client to train the initial global model to obtain the local model weight parameter of each client includes: Using the local data set of each client to train the initial global model, and obtaining a loss function corresponding to the initial global model; Through multiple iterations until the loss function reaches a preset value, the model parameters of each of the clients corresponding to the preset value are obtained; Based on the model parameters and in combination with the importance of the client, a local model weight parameter of each client is obtained.

[0008] Optionally, the step of performing multiple iterations until the loss function reaches a preset value and obtaining a local weight parameter of each client corresponding to the preset value includes: The forward propagation algorithm and the back propagation algorithm are used to perform multiple iterations until the loss function reaches a preset value, and the local weight parameter of each client corresponding to the preset value is obtained.

[0009] Optionally, the following formula is used to determine the comprehensive model weight parameter: ; Where k is the number of clients, n is the sum of all client sample sizes, is the sample size of the kth client, is the local weight parameter of the kth client.

[0010] Optionally, determining a comprehensive model weight parameter based on the local model weight parameters of the plurality of clients, and updating the initial global model using the model comprehensive weight parameter to obtain a target global model includes: Normalizing the weight parameters of the local model of each user terminal; Determine the comprehensive model weight parameters according to the normalized local model weight parameters; The network parameters of the initial global model are updated according to the comprehensive model weight parameters to obtain a target global model.

[0011] Optionally, the updating the initial global model by using the model comprehensive weight parameter to obtain a target global model includes: When the initial global model converges, the updating is stopped, and the initial global model at the time of convergence is determined as the target global model.

[0012] The second aspect of the embodiment of the present application provides an identity recognition device based on EEG signals, including: a receiving module, a training module, a determining module and an output module, wherein: The receiving module is configured to receive the initial global model sent by the server; The training module is configured to train the initial global model using the local data set of each client to obtain the local model weight parameters of each client; The determination module is configured to determine the comprehensive model weight parameters based on the local model weight parameters of the plurality of clients, and update the initial global model using the model comprehensive weight parameters to obtain a target global model; The output module is configured to input the EEG signal to be identified into the target global model and output the identity recognition result.

[0013] A third aspect of an embodiment of the present application provides an electronic device, including a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the identity recognition method based on EEG signals described in the first aspect.

[0014] A fourth aspect of an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0015] Compared with the prior art, the technical solution provided by this application has the following beneficial effects: The present invention provides an identity recognition method, device and equipment based on electroencephalogram (EEG) signals. The method uses the local data set of each client to train the initial global model sent by the server to obtain local model weight parameters, and determines the comprehensive model weight parameters based on the local model weight parameters of multiple clients. The initial global model is updated using the model comprehensive weight parameters to obtain a target global model. The EEG signal to be recognized is input into the target global model, and the identity recognition result is output, so that each user terminal can collaboratively train the model without sharing the original data, which not only realizes the accurate recognition of the user's EEG signal, but also enhances the security and privacy protection of the data. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1A schematic diagram of a flow chart of an identity recognition method based on EEG signals provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of the federated learning collaborative DEEPCNN model training framework provided in the embodiment of the present application; Figure 3 This is a schematic diagram of the data accuracy performance under different server rounds and local rounds configurations; Figure 4 A schematic diagram of channel distribution provided for this embodiment; Figure 5 A schematic diagram of the test accuracy under different channel combinations provided in this embodiment; Figure 6 A schematic diagram of the test accuracy of the three channels provided in this embodiment; Figure 7 A schematic diagram of the test accuracy of 8 channels provided in this embodiment; Figure 8 A schematic diagram of the test accuracy of 8-channel 20-second samples provided in this embodiment; Fig. 9 A schematic diagram of the structure of an identity recognition device based on EEG signals provided in an embodiment of the present application; Fig.10 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0017] Below, embodiments of the present application will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present application. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present application. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present application.

[0018] The terms used herein are only for describing specific embodiments and are not intended to limit the present application. The terms "include", "comprising", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components. All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0019] Some block diagrams and / or flow charts are shown in the accompanying drawings. It should be understood that some blocks or combinations thereof in the block diagrams and / or flow charts may be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that these instructions, when executed by the processor, may create a device for implementing the functions / operations described in these block diagrams and / or flow charts.

[0020] In some embodiments, see Figure 1 , Figure 1 A schematic diagram of a flow chart of an identity recognition method based on EEG signals provided in an embodiment of the present application; an identity recognition method based on EEG signals provided in an embodiment of the present application includes: S110, receiving an initial global model sent by the server.

[0021] Here, the initial global model is the DEEPCNN deep convolutional neural network. By establishing a federated learning architecture, the system is divided into client and server. The client device is responsible for storing data and training DEEPCNN on user data. The device sends the trained model weight updates to the central server for aggregation, and the server calculates a new global model version based on these updates and distributes it back to each client for further training. For more information on the system architecture, see Figure 2 , Figure 2 A schematic diagram of the structure of the federated learning collaborative DEEPCNN model training framework provided in the embodiment of the present application.

[0022] In some embodiments, the initial global model includes two two-dimensional convolutional layers and one depthwise convolutional layer.

[0023] In this embodiment, the two-dimensional convolution layer can be Conv2D, which slides a convolution kernel on the input data and performs an inner product operation at each position, and then adds a bias term to produce an output feature map. In the initial global model, Conv2D is mainly used to capture features in the time dimension. The deep convolution layer DepthwiseConv2D can decompose the fully connected convolution into two steps: first, perform channel-by-channel point-wise convolution (ie, depth convolution), and then perform point-by-point ordinary convolution (also called point multiplication), by applying a separate convolution kernel to each input channel, which makes it possible to capture local correlations between different channels. Here, applying a smaller size convolution kernel can effectively reduce the spatial resolution of the output tensor, thereby achieving the purpose of dimensionality reduction.

[0024] S120, using the local data set of each client to train the initial global model, and obtain the local model weight parameters of each client.

[0025] In this embodiment, considering that the contents of local data sets processed by different clients are different, the local model weight parameters of different clients may be the same or different.

[0026] In some embodiments, S120, using the local data set of each client to train the initial global model to obtain the local model weight parameters of each client includes: Use the local data set of each client to train the initial global model and obtain the loss function corresponding to the initial global model; Through multiple iterations until the loss function reaches a preset value, the model parameters of each client corresponding to the preset value are obtained; Based on the model parameters and the importance of the client, the local model weight parameters of each client are obtained.

[0027] During the training process, the server's target loss function F(w) can be expressed as: ; Where m is the total number of client devices participating in the training, n is the total number of client samples, is the sample size of the kth client, is the local objective function of the kth client: ; in, is the local sentiment dataset of the kth client, is the DEEPCNN model w for the dataset The loss function generated by the instance in is, The sum of the loss functions generated by all instances in divided by the total amount of data of client k is the average loss function of the local client. The loss function is inversely proportional to the model accuracy. Therefore, the objective function optimization here is to iterate the client and server to minimize the loss function.

[0028] In this embodiment, the server initializes the weight parameters of the global DEEPCNN model , the DEEPCNN model consists of multiple layers, including convolutional layers, linear layers, and batch normalization layers. Each layer has a corresponding weight matrix and bias vector, which together constitute the weight parameter Set the server-side iteration round t. In each iteration, randomly select m clients to participate in the current round of training. For each selected client, call the client function so that the client can update its local model weight parameters based on the local data set. ,in, Represents the model parameters of client k after local training.

[0029] In some embodiments, multiple iterations are performed until the loss function reaches a preset value, and the local weight parameter of each client corresponding to the preset value is obtained, including: The forward propagation algorithm and the back propagation algorithm are used to perform multiple iterations until the loss function reaches a preset value, and the local weight parameters of each client corresponding to the preset value are obtained.

[0030] S130, determining comprehensive model weight parameters based on local model weight parameters of multiple clients, and using the model comprehensive weight parameters to update the initial global model to obtain a target global model.

[0031] In this embodiment, the local model weight parameters of each client are aggregated to obtain comprehensive model weight parameters, namely global model parameters, thereby obtaining a target global model.

[0032] In some embodiments, S130, determining comprehensive model weight parameters based on local model weight parameters of multiple clients, and updating the initial global model using the model comprehensive weight parameters to obtain a target global model, includes: Normalize the local model weight parameters of each user end; Determine the comprehensive model weight parameters according to the normalized local model weight parameters; The network parameters of the initial global model are updated according to the comprehensive model weight parameters to obtain the target global model.

[0033] In some embodiments, S130, updating the initial global model using the model comprehensive weight parameter to obtain the target global model includes: When the initial global model converges, the update is stopped, and the initial global model at the time of convergence is determined as the target global model.

[0034] In an optional embodiment, the client executes a certain number of stochastic gradient descent (SGD) algorithms based on its own local data to update the model parameters. Specifically, a new round of weight updates is obtained by calculating the loss function of the client model training and multiplying it by a fixed learning rate η. This process usually includes multiple epochs of forward propagation and backpropagation. Therefore, the client's model weight update can be described as: ; Then, through the federated learning protocol, the client sends the trained model weight updates to the central server for aggregation, and the server calculates a new global model parameter based on these updates to obtain the target global model, and distributes it back to each client for further training. Each client uses the trained model for testing, and this cycle repeats until the model converges.

[0035] In some embodiments, the comprehensive model weight parameter is determined using the following formula: ; Where k is the number of clients, n is the sum of all client sample sizes, is the sample size of the kth client, is the local weight parameter of the kth client.

[0036] S140, input the EEG signal to be identified into the target global model, and output the identity recognition result.

[0037] In one example, the DEAP emotional electroencephalogram (EEG) dataset was used, which is widely regarded as a standard benchmark for performing emotion or mood recognition tasks. The DEAP dataset recorded the EEG activity of 32 healthy volunteers through 32 EEG channels at a high sampling rate of 512Hz. Each participant watched music video clips designed to induce emotional responses under EEG monitoring conditions and made subjective emotional evaluations on 40 videos. The preprocessing of the DEAP dataset includes the following steps: First, the raw data is downsampled from 512Hz to 128Hz to reduce the amount of data and improve processing efficiency; then, the independent component analysis (ICA) technique is used for blind source separation to remove artifacts caused by the electrooculogram (EOG); then, a bandpass filter from 4.0Hz to 45.0Hz is applied to the data to filter out high-frequency noise and low-frequency drift and retain the frequency band information related to emotion; then, the data of all channels are reset to a common average reference to standardize the data and enhance signal consistency; finally, the processed data is divided into 60-second trial segments, and the first 3 seconds of each trial segment are removed to exclude non-steady-state effects in the initial stage. These preprocessing steps are designed to maximize data quality and ensure that the extracted features can accurately reflect the changes in emotional state, providing a reliable basis for subsequent emotion analysis or identity recognition tasks.

[0038] The data format of the input neural network is (N, C, S), where N represents the number of samples, C is the number of channels, and S is the number of time points. Specifically, the format of the signal dataset after preprocessing is (32x40,32,60x128). Here, 32x40 means that there are 32 participants in the dataset, each participant provides 40 samples, and a total of 1280 samples. The number of time points 60x128 reflects that the time length of each sample is 60 seconds and the sampling frequency is 128Hz. The label dataset is a one-dimensional array with a shape of (32x40) and contains the unique identifiers of the 32 participants, constituting a 32-class classification problem.

[0039] In order to construct the test set, 10 samples are randomly selected from the 40 samples of each participant to ensure that the test set covers the labels of all participants. The remaining 30 samples constitute the training set. The training set is further distributed to 4 clients, each of which only holds data of 8 randomly selected user labels, and ensures that users do not repeat between different clients. This setting is intended to simulate the situation in which local clients store user data in actual scenarios, that is, each client can only access the user data assigned to it, and other clients have no access. Therefore, the number of samples for each client is 8x30=240, where 8 represents the number of user categories held by the client, and 30 represents the number of trials for each category of users. This design not only conforms to the distributed characteristics of federated learning, but also effectively protects the privacy and security of user data. The data dimensions are shown in Table 1.

[0040] Table 1: Data dimension table

[0041] When training the DEEPCNN model, parameter tuning is required, including learning rate, optimizer, regularization coefficient, etc. The learning rate determines the speed at which the model parameters change during training. Too large or too small a learning rate will affect the training effect of the model. The optimizer determines the search direction and step size of the model in the parameter space. Commonly used optimizers include SGD, Adam, RMSprop, etc. The regularization coefficient can effectively prevent overfitting, including L1 regularization, L2 regularization, etc.

[0042] In order to verify the performance superiority of the EEG signal-based identity recognition method provided in the embodiment of the present application on the DEAP emotion dataset, we selected several currently widely recognized EEG classification methods as the control group for a detailed comparative analysis. According to the analysis results, it can be clearly pointed out that compared with other methods, the EEG signal-based identity recognition method provided in the embodiment of the present application shows significant advantages in the identity recognition task based on the DEAP dataset, especially in terms of accuracy, which is not only significantly ahead, but also more prominent due to the special integration of privacy protection functions, thereby providing higher security and reliability guarantees in practical applications.

[0043] Table 2: Classification accuracy statistics of each network model

[0044] In an example, see Figure 3 , Figure 3This is a diagram showing the accuracy of data under different server round and local round configurations; with the gradual increase of local rounds and server rounds, the test accuracy has shown a significant improvement trend. In particular, under the configuration of 30 server rounds and 50 local rounds, although the total number of iterations is relatively limited to 1500, this application has been able to achieve an accuracy rate of up to 99.69%. Further increasing the server rounds and local rounds can make the recognition accuracy approach 100%. This shows that by reasonably adjusting the number of server and local iterations, the model training process can be effectively optimized and the recognition accuracy can be improved.

[0045] In order to optimize the number of input channels, 300 rounds of experiments were conducted on the DEEPCNN model. In each round, 3 channels were randomly selected from 1 to 32 channels for testing, and the channel combination with an accuracy of more than 90% was screened out. Subsequently, the indexes of these preferred channels were applied to the DEEPCNN-FL model. Under the condition that the server rounds were uniformly set to 25 rounds and the local training rounds were set to 100 rounds, the test accuracy under different channel combinations was evaluated. Figure 4 , Figure 4 This is a schematic diagram of the channel distribution provided for this embodiment. The results show that when the channels are [10, 19, 27], the test accuracy is significantly higher than other channel combinations. For details, see the test accuracy diagram under different channel combinations. Figure 5 . Based on the above findings, subsequent experiments will use the 3-channel combination with the highest accuracy [10, 19, 27] to further study the changes in the accuracy of 3-channel data under different server rounds and local rounds. In addition, the 8 channels with the highest frequency in the 3-channel combination with an accuracy rate of more than 90%, namely [2, 4, 12, 16, 22, 23, 24, 27], will be selected to further explore the performance characteristics of 8-channel data accuracy under different server rounds and local rounds.

[0046] In an example, see Figure 6 and Figure 7 , Figure 6 and Figure 7 The three-channel test accuracy diagram and the eight-channel test accuracy diagram provided in this embodiment are respectively; after reducing the original 32-channel data to three channels, the accuracy rate can reach 91.56% under the setting conditions of 80 rounds of training on the server side and 50 rounds of local training on the client side. When the number of channels is optimized to 8, the model accuracy rate is significantly improved to 97.81% with a total of fewer iterations (80 rounds on the server side and 20 rounds on the client side). The above results show that the DEEPCNN-FL network architecture constructed based on the method of this application can not only maintain a high recognition accuracy while reducing the number of channels, but also has good privacy protection characteristics, which is suitable for identity authentication scenarios.

[0047] To further optimize the performance, the time sample was shortened to 20 seconds while maintaining the original 60-second time sample, and 8-channel data was used for testing. The results show that when the number of local iterations is set to 30 and the number of server iterations is set to 50, an accuracy of up to 93.18% can still be achieved. This shows that the identity recognition method based on EEG signals provided by the embodiment of the present invention has good user-friendliness. The user only needs to use 8 channels and complete the experiment within 20 seconds to achieve a high-precision recognition effect. For detailed test results, see Figure 8 , Figure 8 A schematic diagram of the test accuracy of 8-channel 20-second samples provided in this embodiment.

[0048] The embodiment of the present invention trains the initial global model sent by the server side by using the local data set of each client to obtain local model weight parameters, determines the comprehensive model weight parameters based on the local model weight parameters of multiple clients, and updates the initial global model by using the model comprehensive weight parameters to obtain the target global model, inputs the EEG signal to be identified into the target global model, outputs the identity recognition result, and enables each user terminal to collaboratively train the model without sharing the original data, which not only realizes the accurate recognition of the user's EEG signal, but also enhances the security and privacy protection of the data.

[0049] In some embodiments, see Fig. 9 , Fig. 9 A schematic diagram of the structure of an identity recognition device based on EEG signals provided in an embodiment of the present application; an identity recognition device 900 based on EEG signals provided in an embodiment of the present application includes: a receiving module 910, a training module 920, a determining module 930 and an output module 940, wherein: A receiving module 910 is configured to receive an initial global model sent by a server; A training module 920 is configured to train the initial global model using the local data set of each client to obtain the local model weight parameters of each client; A determination module 930 is configured to determine a comprehensive model weight parameter based on the local model weight parameters of multiple clients, and update the initial global model using the model comprehensive weight parameter to obtain a target global model; The output module 940 is configured to input the EEG signal to be identified into the target global model and output the identity identification result.

[0050] In some embodiments, the initial global model includes two two-dimensional convolutional layers and one depthwise convolutional layer.

[0051] In some embodiments, the training module 920 is specifically configured as follows: Use the local data set of each client to train the initial global model and obtain the loss function corresponding to the initial global model; Through multiple iterations until the loss function reaches a preset value, the model parameters of each client corresponding to the preset value are obtained; Based on the model parameters and the importance of the client, the local model weight parameters of each client are obtained.

[0052] In some embodiments, the training module 920 is specifically configured as follows: The forward propagation algorithm and the back propagation algorithm are used to perform multiple iterations until the loss function reaches a preset value, and the local weight parameters of each client corresponding to the preset value are obtained.

[0053] In some embodiments, the comprehensive model weight parameter is determined using the following formula: ; Where k is the number of clients, n is the sum of all client sample sizes, is the sample size of the kth client, is the local weight parameter of the kth client.

[0054] In some embodiments, the determination module 930 is specifically configured to: Normalize the local model weight parameters of each user end; Determine the comprehensive model weight parameters according to the normalized local model weight parameters; The network parameters of the initial global model are updated according to the comprehensive model weight parameters to obtain the target global model.

[0055] In some embodiments, the determination module 930 is further configured to: When the initial global model converges, the update is stopped, and the initial global model at the time of convergence is determined as the target global model.

[0056] The EEG signal-based identity recognition device provided in the embodiment of the present application can implement each process in the embodiment corresponding to the above-mentioned EEG signal-based identity recognition method, and will not be described again here to avoid repetition.

[0057] It should be noted that the identity recognition device based on EEG signals provided in the embodiment of the present application and the identity recognition method based on EEG signals provided in the embodiment of the present application are based on the same application concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned identity recognition method based on EEG signals, and the repeated parts will not be repeated.

[0058] In some embodiments, see Fig.10 , Fig.10 The electronic device 1000 provided in the embodiment of the present application includes a processor 1010 and a memory 1020; the memory 1020 stores a computer program, wherein the computer program implements the above-mentioned identity recognition method based on EEG signals when executed by the processor.

[0059] Specifically, the processor 1010 may include, for example, a general-purpose microprocessor, an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1010 may also include an onboard memory for cache purposes. The processor 1010 may be a single processing unit or multiple processing units for executing different actions of the method flow according to the embodiment of the present application.

[0060] The memory 1020 may be any medium capable of containing, storing, conveying, propagating or transmitting instructions. For example, the memory 1020 may include, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared or semiconductor system, device, component or propagation medium. Specific examples of the memory 1020 include: a magnetic storage device, such as a magnetic tape or a hard disk (HDD); an optical storage device, such as a compact disk (CD-ROM); a random access memory (RAM) or flash memory; and / or a wired / wireless communication link.

[0061] The present application also provides a computer-readable medium on which a computer program is stored, and when the program is executed by a processor, the above-mentioned identity recognition method based on EEG signals is implemented. The computer-readable medium may be included in the device / apparatus / system described in the above-mentioned embodiments; or it may exist independently without being assembled into the device / apparatus / system. The above-mentioned computer-readable medium carries one or more programs, and when the above-mentioned one or more programs are executed, the method according to the embodiment of the present application is implemented.

[0062] According to an embodiment of the present application, a computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a computer-readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, optical cable, radio frequency signal, etc., or any suitable combination of the above.

[0063] Those skilled in the art will appreciate that the features described in the various embodiments and / or claims of the present application may be combined and / or combined in a variety of ways, even if such combinations or combinations are not explicitly described in the present application. In particular, without departing from the spirit and teachings of the present application, the features described in the various embodiments and / or claims of the present application may be combined and / or combined in a variety of ways. All of these combinations and / or combinations fall within the scope of the present application. Therefore, the scope of the present application should not be limited to the above-described embodiments, but should be determined not only by the appended claims, but also by the equivalents of the appended claims.

Claims

1. An identity recognition method based on EEG signals, characterized in that: include: Receive the initial global model sent by the server; Using the local data set of each client to train the initial global model, and obtaining the local model weight parameters of each client; Determine a comprehensive model weight parameter based on the local model weight parameters of the plurality of clients, and use the model comprehensive weight parameter to update the initial global model to obtain a target global model; The EEG signal to be identified is input into the target global model, and the identity identification result is output.

2. The method for identifying an individual based on an EEG signal according to claim 1, characterized in that: The initial global model includes two two-dimensional convolutional layers and one depth-wise convolutional layer.

3. The method for identifying an individual based on an EEG signal according to claim 1, characterized in that: The initial global model is trained using the local data set of each client to obtain the local model weight parameter of each client, including: Using the local data set of each client to train the initial global model, and obtaining a loss function corresponding to the initial global model; Through multiple iterations until the loss function reaches a preset value, the model parameters of each of the clients corresponding to the preset value are obtained; Based on the model parameters and in combination with the importance of the client, a local model weight parameter of each client is obtained.

4. The method for identifying an individual based on an EEG signal according to claim 1, characterized in that: The step of iterating multiple times until the loss function reaches a preset value and obtaining a local weight parameter of each client corresponding to the preset value includes: The forward propagation algorithm and the back propagation algorithm are used to perform multiple iterations until the loss function reaches a preset value, and the local weight parameter of each client corresponding to the preset value is obtained.

5. The method for identifying an individual based on an EEG signal according to claim 1, characterized in that: include: The following formula is used to determine the weight parameters of the comprehensive model: ； Where k is the number of clients, n is the sum of all client sample sizes, is the sample size of the kth client, is the local weight parameter of the kth client.

6. The method for identifying an individual based on an EEG signal according to claim 1, characterized in that: The method of determining a comprehensive model weight parameter based on the local model weight parameters of the plurality of clients, and updating the initial global model using the model comprehensive weight parameter to obtain a target global model includes: Normalizing the weight parameters of the local model of each user terminal; Determine the comprehensive model weight parameters according to the normalized local model weight parameters; The network parameters of the initial global model are updated according to the comprehensive model weight parameters to obtain a target global model.

7. The method for identifying an individual based on an EEG signal according to claim 6, characterized in that: The method of updating the initial global model by using the model comprehensive weight parameter to obtain a target global model includes: When the initial global model converges, the updating is stopped, and the initial global model at the time of convergence is determined as the target global model.

8. An identity recognition device based on EEG signals, characterized in that: include: A receiving module, a training module, a determining module and an output module, wherein: The receiving module is configured to receive the initial global model sent by the server; The training module is configured to train the initial global model using the local data set of each client to obtain the local model weight parameters of each client; The determination module is configured to determine the comprehensive model weight parameters based on the local model weight parameters of the plurality of clients, and update the initial global model using the model comprehensive weight parameters to obtain a target global model; The output module is configured to input the EEG signal to be identified into the target global model and output the identity recognition result.

9. An electronic device comprising a processor and a memory; the memory stores a computer program, wherein: When the computer program is executed by the processor, the computer program implements the EEG signal-based identity recognition method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images