Cross-temporal individual recognition method and system based on deep learning of EEG signals

Abstract

The present invention belongs to the field of cross-time dimension EEG signal recognition, and specifically relates to a cross-time individual recognition method, system and device based on deep learning of EEG signals, and aims to solve the problem. The present invention includes: respectively obtaining labeled EEG signals of the source domain and unlabeled EEG signals of the target domain, and performing signal preprocessing to obtain source domain differential entropy and target domain differential entropy; inputting a joint domain adaptation individual recognition network, and iteratively training the network; obtaining the EEG signal to be recognized in the target domain, and performing signal preprocessing to obtain the differential entropy of the target domain to be recognized; obtaining the individual recognition result of the EEG signal to be recognized in the target domain through the trained joint domain adaptation individual recognition network. The present invention overcomes the insecurity of individual identity recognition relying on traditional methods and the differences caused by different EEG signal acquisition times, and can also ensure individual recognition accuracy and improve recognition efficiency.

Description

Background Art

[0002] At present, the way to identify individuals mainly relies on account and password recognition, fingerprint recognition, face recognition, etc. These methods usually have the risk of leakage or theft, and the security cannot be guaranteed.

[0003] With the advancement of EEG signal acquisition technology, it has become possible to measure EEG signals of multiple channels of the brain epidermis in a high-resolution form. Now EEG signal acquisition and recognition has been widely used in individual identity recognition.

[0004] However, since EEG signals collected at different times have differences in the time dimension, the recognition accuracy will decrease. Therefore, there is a need in the art for an EEG individual identity recognition method that can reduce the time differences of EEG signals, overcome the differences caused by different EEG signal collection times, and also ensure recognition accuracy. Summary of the invention

[0005] In order to solve the above-mentioned problems in the prior art, namely, the low security of the traditional identity recognition method and the problem that the existing EEG signal identity recognition method cannot guarantee the recognition accuracy due to the differences caused by the existence time of the EEG signal, the present invention provides a cross-time individual recognition method based on EEG signal deep learning, and the individual recognition method includes:

[0006] Obtain labeled EEG signals of the source domain and unlabeled EEG signals of the target domain respectively, and perform signal preprocessing to obtain source domain differential entropy and target domain differential entropy;

[0007] Inputting the source domain differential entropy and the target domain differential entropy into a joint domain adaptation individual recognition network, and iteratively performing network training;

[0008] Obtain the EEG signal to be identified in the target domain, perform signal preprocessing, and obtain the differential entropy of the target domain to be identified;

[0009] Based on the differential entropy of the target domain to be identified, the individual identification result of the EEG signal to be identified in the target domain is obtained through the trained joint domain adaptation individual identification network.

[0010] In some preferred embodiments, the signal preprocessing includes bad segment removal of EEG signals, 1-75 Hz filtering, frequency band signal extraction, frequency band signal differential entropy calculation and feature smoothing.

[0011] In some preferred embodiments, the frequency band signal extraction includes 1-3 Hz delta signal extraction, 4-7 Hz theta signal extraction, 8-13 Hz alpha signal extraction, 14-30 Hz beta signal extraction, and 31-50 Hz gamma signal extraction.

[0012] In some preferred embodiments, the joint domain adaptation individual recognition network includes a feature extractor, a classifier and a domain discriminator, and its training process is:

[0013] The feature extractor is used to extract the source domain differential entropy and the target domain differential entropy respectively;

[0014] The feature extraction results are input into the classifier to calculate the classification loss of the source domain and the target domain; the feature extraction results are input into the domain discriminator to calculate the domain discrimination loss of the source domain and the target domain; the feature extraction results are input into the fully connected layer to calculate the joint loss of the source domain and the target domain;

[0015] Based on the classification loss, the domain discrimination loss and the joint loss, obtaining a total loss of a joint domain adaptation individual recognition network;

[0016] Based on the acquired training data set, the network is iteratively trained in the direction of decreasing the total loss until the total loss value is lower than a set threshold or reaches a set number of training times, thereby obtaining a trained joint domain adaptation individual recognition network.

[0017] In some preferred embodiments, the classification loss of the source domain and the target domain is expressed as:

[0018]

[0019] Among them, L class represents the classification loss of the source domain and the target domain, H represents the cross entropy loss, m is the mth training sample in the current training batch, M is the total number of training samples in the current training batch, represents the individual labels of the training samples predicted by the classifier, represents the individual label of the mth training sample predicted by the classifier, y represents the individual label of the actual training sample, and y m Represents the individual label of the actual m-th training sample.

[0020] In some preferred embodiments, the domain discrimination loss of the source domain and the target domain is expressed as:

[0021]

[0022] Among them, L domain represents the domain discrimination loss of the source domain and the target domain, H represents the cross entropy loss, m is the mth training sample in the current training batch, M is the total number of training samples in the current training batch, represents the domain label of the training sample predicted by the classifier, represents the domain label of the mth training sample predicted by the classifier, p represents the domain label of the actual training sample, and p mrepresents the actual domain label of the mth training sample.

[0023] In some preferred embodiments, the joint loss of the source domain and the target domain includes an access probability loss and a conversion probability loss, which is expressed as:

[0024] L association =L visit +αL trans

[0025] Among them, L association represents the joint loss of the source domain and the target domain, L visit represents the access probability loss, L trans represents the conversion probability loss, α is adjustable to weigh L visit and L trans In L association The weight of the proportion.

[0026] In some preferred embodiments, the access probability loss is expressed as:

[0027] L visit =H(V,P ab )

[0028]

[0029] M ij = i , B j >

[0030] V=1 / |B j |

[0031] Among them, H represents the cross entropy loss, P ab is the probability matrix of the vector accessing the target domain category from the source domain, is the matrix P ab The value of represents the source domain vector A i Access the target domain vector B j The probability of M ij Representative A i and B j The similarity between i , B j > represents vector A i and vector B j The dot product of , V represents the weight vector.

[0032] In some preferred embodiments, the conversion probability loss is expressed as:

[0033] L trans =H(T,P aba ) ​​

[0034]

[0035]

[0036] Among them, P ab is the probability matrix of the vector accessing the target domain category from the source domain, P ba is the probability matrix of the vector accessing the source domain category from the target domain, P aba is the probability that the vector visits the source domain category after visiting the target domain and then visiting the source domain category. is the matrix P aba The value of the source domain vector A i Access the target domain vector B j Then access the source domain vector A j The probability of, T represents the weight vector, class(A i )=class(A j ) represents the vector A i and vector A j Category is equal.

[0037] Another aspect of the present invention provides a cross-time individual recognition system based on deep learning of EEG signals, the individual recognition system comprising:

[0038] A data acquisition module, configured to respectively acquire labeled EEG signals of a source domain and unlabeled EEG signals of a target domain, and acquire unlabeled EEG signals to be identified of a target domain;

[0039] A preprocessing module is configured to preprocess the acquired source domain and target domain signals respectively to obtain source domain differential entropy and target domain differential entropy, and to preprocess the acquired target domain EEG signal to be identified to obtain target domain differential entropy to be identified;

[0040] A network training module is configured to input the source domain differential entropy and the target domain differential entropy into a joint domain adaptive individual recognition network, and iteratively perform network training;

[0041] An individual recognition module is configured to obtain individual recognition results of the EEG signals to be recognized in the target domain through a trained joint domain adaptation individual recognition network based on the differential entropy of the target domain to be recognized;

[0042] The output module is configured to output the individual recognition result of the acquired EEG signal to be recognized in the target domain.

[0043] Beneficial effects of the present invention:

[0044] (1) The present invention is a cross-time individual recognition method based on deep learning of EEG signals. EEG signals are used as signals for individual recognition, and deep learning training of cross-time joint domain adaptation is performed to reduce the marginal distribution and conditional distribution differences between source domain data and target domain data, thereby reducing the time difference between the source domain and the target domain, overcoming the insecurity of individual identity recognition based on traditional methods, and realizing the application scenario of EEG signal individual recognition.

[0045] (2) The present invention is a cross-time individual recognition method based on deep learning of EEG signals. It uses labeled EEG signals collected at different times in the past to perform individual recognition on newly collected unlabeled EEG signals, thereby reducing the impact of time differences in EEG signals, thereby improving the accuracy of individual recognition using EEG signals and reducing the time required for recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0047] Figure 1 It is a flow chart of a method for cross-time individual recognition based on deep learning of EEG signals of the present invention;

[0048] Figure 2 It is a schematic diagram of a joint domain adaptation individual recognition network of the cross-time individual recognition method based on deep learning of EEG signals of the present invention. DETAILED DESCRIPTION

[0049] The present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the relevant invention are shown in the accompanying drawings.

[0050] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0051] The present invention provides a cross-time individual recognition method based on deep learning of EEG signals. This method not only overcomes the insecurity of individual identification by traditional methods, but also overcomes the differences caused by different EEG signal acquisition times, and can also ensure the necessary recognition accuracy.

[0052] The present invention provides a cross-time individual recognition method based on deep learning of EEG signals, the individual recognition method comprising:

[0053] Obtain labeled EEG signals of the source domain and unlabeled EEG signals of the target domain respectively, and perform signal preprocessing to obtain source domain differential entropy and target domain differential entropy;

[0054] Inputting the source domain differential entropy and the target domain differential entropy into a joint domain adaptation individual recognition network, and iteratively performing network training;

[0055] Obtain the EEG signal to be identified in the target domain, perform signal preprocessing, and obtain the differential entropy of the target domain to be identified;

[0056] Based on the differential entropy of the target domain to be identified, the individual identification result of the EEG signal to be identified in the target domain is obtained through the trained joint domain adaptation individual identification network.

[0057] In order to more clearly illustrate the cross-time individual recognition method based on deep learning of EEG signals of the present invention, the following is combined with Figure 1 Each step in the embodiment of the present invention is described in detail.

[0058] The method for cross-time individual recognition based on deep learning of EEG signals according to the first embodiment of the present invention includes steps S10 to S40, each of which is described in detail as follows:

[0059] Cross-temporal individual recognition based on EEG signals mainly includes two levels: 1. Feature extraction of EEG signals; 2. Deep learning methods for cross-temporal individual recognition based on EEG signal features.

[0060] Step S10, respectively obtaining labeled EEG signals of the source domain and unlabeled EEG signals of the target domain, and performing signal preprocessing to obtain source domain differential entropy and target domain differential entropy.

[0061] Signal preprocessing includes EEG signal bad segment removal, 1-75Hz filtering, frequency band signal extraction, frequency band signal differential entropy calculation and feature smoothing.

[0062] Frequency band signal extraction includes 1-3Hz delta signal extraction, 4-7Hz theta signal extraction, 8-13Hz alpha signal extraction, 14-30Hz beta signal extraction, and 31-50Hz gamma signal extraction.

[0063] To calculate the differential entropy of the frequency band signal, the calculation method proposed by B.-L. Lu et al. in 2013 (L.-C. Shi, Y.-Y. Jiao, and B.-L. Lu, "Differential entropy feature for EEG-based vigilance estimation," in Proc. 35th Annu. Int. Conf. IEEE EMBS. IEEE, 2013) can be used, as shown in formula (1):

[0064]

[0065] Where x is the time series of the EEG signal in the current frequency band, h(X) is the calculation result of the differential entropy of x, μ, δ are the Gaussian distribution N(μ, δ) of x 2 )’s location parameter and dispersion parameter.

[0066] An EEG sequence in a time window has n channels, each channel has 5 frequency bands, and a differential entropy can be calculated for each frequency band. Therefore, n*5 differential entropies can be calculated for a time window.

[0067] The differential entropy is smoothed to obtain the smoothed differential entropy, and the preprocessed data of the source domain and the target domain are recorded as the source domain differential entropy and the target domain differential entropy respectively.

[0068] Step S20: input the source domain differential entropy and the target domain differential entropy into a joint domain adaptive individual recognition network, and iteratively perform network training.

[0069] The smoothed differential entropy is converted into a one-dimensional vector. Each time window corresponds to a vector. The vectors of the source domain and the target domain are used to train the joint domain adaptation individual recognition network. The joint domain adaptation individual recognition network includes a feature extractor, a classifier, and a domain discriminator. The training process is:

[0070] In step A10, feature extractors are used to extract the source domain differential entropy and the target domain differential entropy.

[0071] Only the source domain data has individual identity labels. Each window data, that is, a feature vector, corresponds to an individual identity label, and the feature vectors of different individuals correspond to different individual labels.

[0072] In step A20, the feature extraction result is input into the classifier to calculate the classification loss of the source domain and the target domain, as shown in formula (2):

[0073]

[0074] Among them, L classrepresents the classification loss of the source domain and the target domain, H represents the cross entropy loss, m is the mth training sample in the current training batch, M is the total number of training samples in the current training batch, represents the individual labels of the training samples predicted by the classifier, represents the individual label of the mth training sample predicted by the classifier, y represents the individual label of the actual training sample, and y m Represents the individual label of the actual m-th training sample.

[0075] Domain discrimination loss L domain The domain discriminator first performs gradient inversion on the output features of the feature extractor, and then obtains the output features through multiple fully connected layers. Through gradient inversion, the domain identification loss L domain The smaller it is, the less the domain discriminator can distinguish whether the feature comes from the source domain or the target domain, thereby reducing the difference in the marginal distribution of the source domain and the target domain data in the time dimension. All source domain data are marked as source domain, and all target domain data are marked as target domain. The domain discriminator discriminates the data and identifies whether it comes from the source domain or the target domain, thereby calculating the domain discrimination loss L domain .

[0076] The feature extraction result is input into the domain discriminator to calculate the domain discrimination loss of the source domain and the target domain, as shown in formula (3):

[0077]

[0078] Among them, L domain represents the domain discrimination loss of the source domain and the target domain, H represents the cross entropy loss, m is the mth training sample in the current training batch, M is the total number of training samples in the current training batch, represents the domain label of the training sample predicted by the classifier, represents the domain label of the mth training sample predicted by the classifier, p represents the domain label of the actual training sample, and p m represents the actual domain label of the mth training sample.

[0079] Joint loss L association The probability that a vector of the source domain matrix remains the original vector after being converted into a target domain vector is calculated through the intermediate feature vector of the classifier, which is called the access probability. The probability that a vector of the target domain matrix remains the original vector after being converted into a source domain vector is called the conversion probability.

[0080] The feature extraction result is input into the fully connected layer to calculate the joint loss of the source domain and the target domain, as shown in formula (4):

[0081] L association =L visit +αL trans (4)

[0082] Among them, L association represents the joint loss of the source domain and the target domain, L visit represents the access probability loss, L trans represents the conversion probability loss, α is adjustable to weigh L visit and L trans In L association The weight of the proportion.

[0083] The access probability loss is expressed as shown in formula (5):

[0084] L visit =H(V,P ab ) (5)

[0085] Among them, H represents the cross entropy loss, P ab is the probability matrix of the vector accessing the target domain category from the source domain, and V represents the weight vector.

[0086] is the matrix P ab The value of represents the source domain vector A i Access the target domain vector B j The probability, M, is calculated as shown in formula (6):

[0087]

[0088] Among them, M ij Representative A i and B j The degree of similarity between them.

[0089] The weight vector V is expressed as shown in formula (7):

[0090] V=1 / |B j | (7)

[0091] Among them, |B j | represents the target domain vector B j Model.

[0092] A i and B j The similarity between ij By calculating A i and B j The dot product of is obtained as shown in formula (8):

[0093] M ij = i , B j > (8)

[0094] in, i , B​​j > represents calculation A i and B j The dot product of .

[0095] The conversion probability loss is expressed as shown in formula (9):

[0096] L trans =H(T,P aba ) (9)

[0097] Among them, P ab is the probability matrix of the vector accessing the target domain category from the source domain, P ba is the probability matrix of the vector accessing the source domain category from the target domain, P aba is the probability that the vector visits the source domain category after visiting the target domain and then visiting the source domain category, and T represents the weight vector.

[0098] is the matrix P aba The value of represents the source domain vector A i Access the target domain vector B j After visiting again, the calculation method is shown in formula (10):

[0099]

[0100] The weight vector T is expressed as shown in formula (11):

[0101]

[0102] Among them, |A i | represents the source domain vector A i The model of class(A i )=class(A j ) represents the vector A i and vector A j The categories are equal, that is, vector A i and vector A j When the categories are equal, the weight vector T is the source domain vector A i The inverse of the modulus of , otherwise the weight vector T is 0.

[0103] Step A30, based on the classification loss, the domain discrimination loss and the joint loss, the total loss of the joint domain adaptation individual recognition network is obtained, as shown in formula (12):

[0104] L=L class +L domain +L association (12)

[0105] Step A40, based on the acquired training data set, performing network iterative training in the direction of decreasing the total loss until the total loss value is lower than a set threshold or reaches a set number of training times, thereby obtaining a trained joint domain adaptive individual recognition network.

[0106] By multiple iterations of the training data and reducing the total loss value, the trained joint domain adaptation individual recognition network can reduce the time difference between the source domain and the target domain EEG data, thereby improving the accuracy of individual recognition of EEG signals across time.

[0107] Step S30, obtaining the EEG signal to be identified in the target domain, and performing signal preprocessing to obtain the differential entropy of the target domain to be identified.

[0108] The EEG signals to be identified in the target domain are also preprocessed by the method of step S10, including bad segment removal of EEG signals, 1-75 Hz filtering, frequency band signal extraction, frequency band signal differential entropy calculation and feature smoothing to obtain differential entropy of the target domain to be identified.

[0109] Step S40, based on the differential entropy of the target domain to be identified, the individual identification result of the EEG signal to be identified in the target domain is obtained through the trained joint domain adaptation individual identification network.

[0110] In summary, the deep learning method for cross-temporal individual identification based on EEG features of the present invention uses the previously collected labeled EEG signals as source domain data, and the unlabeled EEG signals to be identified as target domain data. Through the deep learning network framework, the marginal distribution and conditional distribution differences between the source domain data and the target domain data are reduced, thereby reducing the time difference between the source domain and the target domain.

[0111] In this way, the labeled EEG signals collected at different times in the past can be used to perform individual identification on the newly collected unlabeled EEG signals, reducing the impact of the time differences of EEG signals, thereby improving the accuracy of individual identification using EEG signals and reducing the time required for identification.

[0112] Although the various steps in the above embodiment are described in the above-mentioned order, those skilled in the art can understand that in order to achieve the effect of this embodiment, different steps do not have to be executed in such an order. They can be executed simultaneously (in parallel) or in a reverse order. These simple changes are within the scope of protection of the present invention.

[0113] The second embodiment of the present invention is a cross-time individual recognition system based on deep learning of EEG signals, and the individual recognition system includes:

[0114] A data acquisition module, configured to respectively acquire labeled EEG signals of a source domain and unlabeled EEG signals of a target domain, and acquire unlabeled EEG signals to be identified of a target domain;

[0115] A preprocessing module is configured to preprocess the acquired source domain and target domain signals respectively to obtain source domain differential entropy and target domain differential entropy, and to preprocess the acquired target domain EEG signal to be identified to obtain target domain differential entropy to be identified;

[0116] A network training module is configured to input the source domain differential entropy and the target domain differential entropy into a joint domain adaptive individual recognition network, and iteratively perform network training;

[0117] An individual recognition module is configured to obtain individual recognition results of the EEG signals to be recognized in the target domain through a trained joint domain adaptation individual recognition network based on the differential entropy of the target domain to be recognized;

[0118] The output module is configured to output the individual recognition result of the acquired EEG signal to be recognized in the target domain.

[0119] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process and related instructions of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0120] It should be noted that the cross-time individual recognition system based on deep learning of EEG signals provided in the above embodiment is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be decomposed or combined. For example, the modules in the above embodiments can be combined into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the modules or steps and are not regarded as improper limitations of the present invention.

[0121] An electronic device according to a third embodiment of the present invention includes:

[0122] at least one processor; and

[0123] a memory communicatively connected to at least one of the processors; wherein,

[0124] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the above-mentioned cross-time individual recognition method based on deep learning of EEG signals.

[0125] A computer-readable storage medium according to a fourth embodiment of the present invention stores computer instructions, and the computer instructions are used to be executed by the computer to implement the above-mentioned cross-time individual recognition method based on deep learning of EEG signals.

[0126] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process and related instructions of the storage device and processing device described above can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0127] Those skilled in the art should be able to appreciate that the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, computer software or a combination of the two, and the programs corresponding to the software modules and method steps can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the technical field. In order to clearly illustrate the interchangeability of electronic hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0128] The terms "first", "second", etc. are used to distinguish similar objects rather than to describe or indicate a particular order or sequence.

[0129] The term "comprise" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that includes a list of elements includes not only those elements but also other elements not expressly listed, or also includes elements inherent to such process, method, article, or apparatus / device.

[0130] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A method for cross-time individual recognition based on deep learning of EEG signals, characterized in that: The individual identification method includes: Obtain labeled EEG signals of the source domain and unlabeled EEG signals of the target domain respectively, and perform signal preprocessing to obtain source domain differential entropy and target domain differential entropy; The source domain differential entropy and the target domain differential entropy are input into a joint domain adaptation individual recognition network, the joint domain adaptation individual recognition network includes a feature extractor, a classifier and a domain discriminator, and the network training is iteratively performed: The feature extractor is used to extract the source domain differential entropy and the target domain differential entropy respectively; The feature extraction results are input into the classifier to calculate the classification loss of the source domain and the target domain; the feature extraction results are input into the domain discriminator to calculate the domain discrimination loss of the source domain and the target domain; the feature extraction results are input into the fully connected layer to calculate the joint loss of the source domain and the target domain; Based on the classification loss, the domain discrimination loss and the joint loss, obtaining a total loss of a joint domain adaptation individual recognition network; Based on the acquired training data set, performing network iterative training in the direction of decreasing the total loss until the total loss value is lower than a set threshold or reaches a set number of training times, thereby obtaining a trained joint domain adaptation individual recognition network; Obtain the EEG signal to be identified in the target domain, perform signal preprocessing, and obtain the differential entropy of the target domain to be identified; Based on the differential entropy of the target domain to be identified, the individual identification result of the EEG signal to be identified in the target domain is obtained through the trained joint domain adaptation individual identification network.

2. The method for cross-time individual recognition based on deep learning of EEG signals according to claim 1 is characterized in that: The signal preprocessing includes bad segment removal of EEG signals, 1-75 Hz filtering, frequency band signal extraction, frequency band signal differential entropy calculation and feature smoothing.

3. The method for cross-time individual recognition based on deep learning of EEG signals according to claim 2 is characterized in that: The frequency band signal extraction includes 1-3 Hz delta signal extraction, 4-7 Hz theta signal extraction, 8-13 Hz alpha signal extraction, 14-30 Hz beta signal extraction, and 31-50 Hz gamma signal extraction.

4. The method for cross-time individual recognition based on deep learning of EEG signals according to claim 1, characterized in that: The classification loss of the source domain and the target domain is expressed as: Among them, L class represents the classification loss of the source domain and the target domain, H represents the cross entropy loss, m is the mth training sample in the current training batch, M is the total number of training samples in the current training batch, represents the individual labels of the training samples predicted by the classifier, represents the individual label of the mth training sample predicted by the classifier, y represents the individual label of the actual training sample, and y m Represents the individual label of the actual m-th training sample.

5. The method for cross-time individual recognition based on deep learning of EEG signals according to claim 1, characterized in that: The domain discrimination loss of the source domain and the target domain is expressed as: Among them, L domain represents the domain discrimination loss of the source domain and the target domain, H represents the cross entropy loss, m is the mth training sample in the current training batch, M is the total number of training samples in the current training batch, represents the domain label of the training sample predicted by the classifier, represents the domain label of the mth training sample predicted by the classifier, p represents the domain label of the actual training sample, and p m represents the actual domain label of the mth training sample.

6. The method for cross-time individual recognition based on deep learning of EEG signals according to claim 1, characterized in that: The joint loss of the source domain and the target domain includes the access probability loss and the conversion probability loss, which is expressed as: L association =L visit +αL trans Among them, L association represents the joint loss of the source domain and the target domain, L visit represents the access probability loss, L trans represents the conversion probability loss, α is adjustable to weigh L visit and L trans In L association The weight of the proportion.

7. The method for cross-time individual recognition based on deep learning of EEG signals according to claim 6, characterized in that: The access probability loss is expressed as: L visit =H(V,P ab ) M ij =<A i ,B j > V=1 / |B j | Among them, H represents the cross entropy loss, P ab is the probability matrix of the vector accessing the target domain category from the source domain, is the matrix P ab The value of represents the source domain vector A i Access the target domain vector B j The probability of M ij Representative A i and B j The similarity between i ,B j > represents vector A i and vector B j The dot product of , V represents the weight vector.​ 8. The method for cross-time individual recognition based on deep learning of EEG signals according to claim 7, characterized in that: The conversion probability loss is expressed as: L trans =H(T,P aba ) Among them, P ab is the probability matrix of the vector accessing the target domain category from the source domain, P ba is the probability matrix of the vector accessing the source domain category from the target domain, P aba is the probability that the vector visits the source domain category after visiting the target domain and then visiting the source domain category. is the matrix P aba The value of represents the source domain vector A i Access the target domain vector B j Then access the source domain vector A j The probability of, T represents the weight vector, class(A i )=class(A j ) represents the vector A i and vector A j Category is equal.

9. A cross-time individual recognition system based on deep learning of EEG signals, characterized in that: The individual identification system includes: A data acquisition module, configured to respectively acquire labeled EEG signals of a source domain and unlabeled EEG signals of a target domain, and acquire unlabeled EEG signals to be identified of a target domain; A preprocessing module is configured to preprocess the acquired source domain and target domain signals respectively to obtain source domain differential entropy and target domain differential entropy, and to preprocess the acquired target domain EEG signal to be identified to obtain target domain differential entropy to be identified; The network training module is configured to input the source domain differential entropy and the target domain differential entropy into a joint domain adaptation individual recognition network, wherein the joint domain adaptation individual recognition network includes a feature extractor, a classifier, and a domain discriminator, and iteratively perform network training: The feature extractor is used to extract the source domain differential entropy and the target domain differential entropy respectively; The feature extraction results are input into the classifier to calculate the classification loss of the source domain and the target domain; the feature extraction results are input into the domain discriminator to calculate the domain discrimination loss of the source domain and the target domain; the feature extraction results are input into the fully connected layer to calculate the joint loss of the source domain and the target domain; Based on the classification loss, the domain discrimination loss and the joint loss, obtaining a total loss of a joint domain adaptation individual recognition network; Based on the acquired training data set, performing network iterative training in the direction of decreasing the total loss until the total loss value is lower than a set threshold or reaches a set number of training times, thereby obtaining a trained joint domain adaptation individual recognition network; An individual recognition module is configured to obtain individual recognition results of the EEG signals to be recognized in the target domain through a trained joint domain adaptation individual recognition network based on the differential entropy of the target domain to be recognized; The output module is configured to output the individual recognition result of the acquired EEG signal to be recognized in the target domain.

Images