CN112450946A - Electroencephalogram artifact restoration method based on loop generation countermeasure network - Google Patents
Electroencephalogram artifact restoration method based on loop generation countermeasure network Download PDFInfo
- Publication number
- CN112450946A CN112450946A CN202011206585.4A CN202011206585A CN112450946A CN 112450946 A CN112450946 A CN 112450946A CN 202011206585 A CN202011206585 A CN 202011206585A CN 112450946 A CN112450946 A CN 112450946A
- Authority
- CN
- China
- Prior art keywords
- artifact
- data
- electroencephalogram
- domain
- eeg
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/725—Details of waveform analysis using specific filters therefor, e.g. Kalman or adaptive filters
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Signal Processing (AREA)
- Pathology (AREA)
- Medical Informatics (AREA)
- Physiology (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Psychiatry (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention discloses an electroencephalogram artifact repairing method based on a circularly generated countermeasure network. Because the EEG data does not have artifact-free EEG data and artifact-containing EEG data which are strictly in one-to-one correspondence, the traditional method cannot well verify whether interested information is removed or not. The invention designs a countermeasure network based on cycle production, which can still retain interested information after processing artifact EEG data. Firstly, acquiring data according to a designed paradigm; secondly, performing necessary preprocessing and separating data according to categories; then, training parameters are set, data are imported into a training network, and finally EEG data with artifacts removed and interest information reserved is obtained. Compared with the traditional artifact removal method, the electroencephalogram artifact repair based on the cyclic generation countermeasure network can better keep the interested information on the repaired EEG data, and is beneficial to follow-up research.
Description
Technical Field
The invention relates to the field of electroencephalogram signal processing, in particular to an electroencephalogram artifact repairing method based on a circularly generated countermeasure network.
Background
Electroencephalography (EEG) provides information about the activity of neurons in the brain, and thus about the mental state of a person. Electroencephalography is a widely used analytical tool with many applications. Electroencephalograms are important for studying the function of the human brain, but are greatly affected by artifacts. The causes of the artifacts include eye movement, blinking, muscle activity, and heartbeat, among others. The problems with electrodes, high electrode impedance, wire noise and interference from electronic equipment may cause technical artifacts.
Discarding contaminated electroencephalogram fragments based on visual inspection and manual inspection is extremely costly, and the method of manual inspection deletion is completely inapplicable when continuous electroencephalograms are used in brain-computer interface applications or in online mental state monitoring. The existing electroencephalogram artifact removal technology can be roughly divided into two categories, namely artifact avoidance, artifact rejection and artifact elimination. Artifact removal is the only method for removing artifact components and retaining other active components of the brain electricity, and is performed by using a regression method, a filtering method, a wavelet transform method, a principal component analysis method, an independent component analysis method, a canonical correlation analysis method, a morphological component analysis method, an empirical mode decomposition method, and the like.
Disclosure of Invention
The invention aims to provide a method for restoring electroencephalogram artifacts based on a circularly generated countermeasure network, and the technical solution for realizing the aim of the invention comprises the following steps:
step 1: and (6) acquiring data.
And designing a target artifact collection paradigm according to requirements, and collecting electroencephalogram data through an electroencephalogram cap.
Step 2: and (4) preprocessing data.
Step 2-1: and processing the acquired electroencephalogram data by using a filter comprising recess filtering and band-pass filtering.
Step 2-2: and performing segmented storage on the processed data according to the data labels for subsequent model training.
And step 3: and designing and training a corresponding Cycle to generate a confrontation network model by utilizing a Cycle GAN network model, and generating an electroencephalogram signal containing necessary information.
The data acquisition in the step 1 can set a corresponding paradigm according to the artifact to be removed, and the main tasks comprise two commands of a left shaking motion, a right shaking motion and a static posture maintaining.
The core in step 3 is to optimize the objective function:
where G is the generator for generating from domain X to domain Y, F is the generator for generating from domain Y to domain X, DX,DYIs a discriminator for discriminating whether or not the symbol is identicalUnion domain X or domain Y distribution.
The method is used for restoring the EEG artifacts of the antagonistic network based on the cycle generation, and can better keep the interested information on the restored EEG data, thereby being beneficial to the follow-up research.
Drawings
FIG. 1 is a schematic design of a paradigm of the present invention
FIG. 2 is a diagram of a network architecture for use with the present invention
FIG. 3 is a block diagram of a generator of the present invention
FIG. 4 is a diagram of the structure of the discriminator according to the invention
Detailed Description
The invention is further illustrated by the following figures and examples.
As shown in fig. 1 to 4, a method for restoring electroencephalogram artifacts based on a cyclic generation countermeasure network is specifically implemented as follows:
step 1: and (6) acquiring data.
And designing a target artifact collection paradigm according to requirements, and collecting electroencephalogram data through an electroencephalogram cap. The specific design of paradigm is as follows:
as shown in fig. 1, the subject sits on a chair with armrests. The task is to execute two commands of a left side shaking motion, a right side shaking motion and a static gesture according to the prompt. The order of the prompts is random. The experiment contained multiple sets of experimental collections (not less than 6 times), each containing 60 times. After the start of the test, the first 2 seconds of the screen were dark, at t ═ 2s, the sound stimulus indicated the start of the test and ten's ' + ' was displayed. Then starting to display a left arrow, a right arrow or a circular prompt mark for 1 second from t-3 s; while the subject is asked to execute a command to shake the head to the left once, to the right once, or to remain stationary, shaking the head back to the initial position, and remaining stationary until the tenning disappears at t-7 s. Each of the 3 cues was displayed 20 times in random order in each experiment.
Step 2: and (4) preprocessing data.
The acquired data is preprocessed by a Python script (including algorithms such as recess filtering, band-pass filtering and the like, and 50 Hz commercial power and high-frequency noise are filtered), and the data is segmented into an electroencephalogram data set X with target artifacts and an electroencephalogram data set Y without the target artifacts according to marks in the electroencephalogram data for subsequent model training.
And step 3: according to the design concept of the Cycle GAN network model, a corresponding Cycle generation confrontation network model named as EEG-Cycle GAN is designed and trained, and an EEG signal containing necessary information is generated, as shown in figure 2. The concrete implementation is as follows:
used herein as (x)i,yi) Representing a pair of artefact-and non-artefact EEG signals, i.e. (x)i,yi)|xi∈X,yiE.g. Y. The generator is used in the EEG-Cycle GAN: g takes the EEG signal with the target artifact as input, outputs the generated artifact-free EEG signal:a discriminator: the D inputs are pairs of EEG signals with target artifacts and artifact-free EEG signals, the target of D is the learning mapping: d (x)i,yi) → 1 anda real pair of an EEG signal with target artifacts and an artifact-free EEG signal is mapped to 1 and an EEG signal with target artifacts and a generated artifact-free EEG signal is mapped to 0. And the goal of G is to learn the mapping: d (x)i,G(xiZ)) → 1, i.e. making the generated artifact-free EEG signal more "real" and able to form a pair with the input EEG signal with the target artifacts.
Due to the fact thatIt is difficult to find the corresponding relationship in the domain Y, so a Cycle-consistency loss (Cycle-consistency loss) is introduced, with another generator F: y → X, willAgain transferred back into field X to obtainF is generatedShould be consistent with the x holding domain. Then there is
X → Y discriminator loss:
where G (x) that G attempts to generate is data that "looks" to fit domain Y. DYIt is used to discriminate between g (x) and the true sample y. The goal of G is to minimize it, while the goal of D is to maximize it, i.e.
Similarly, the discriminator loss of Y → X:
recycling back to "fight" a round is:
after combination, the following steps are carried out:
where η may be used to control the weight of the two objective functions, typically n takes 2.
The final objective function is:
as shown in fig. 3, the generator is specifically implemented as follows:
and (3) encoding: the first step is to extract features from the input image using a convolutional neural network. Specifically, five one-dimensional convolutional layers are adopted, wherein each one-dimensional convolutional layer of the first four layers adopts a convolutional kernel (kernel) sliding window with the length of 4, the step length (Stride) is set to be 2, the Padding (Padding) value is 1, the last layer adopts the arrangement that the convolutional kernel of the layer is 114, the step length is 1, and the Padding is 0, and the five one-dimensional convolutional layers are used for feature extraction and coding.
Conversion: the feature vector of the image in domain X is converted to a feature vector in domain Y by combining the dissimilar features of the matrices. Here, a plurality of layers of residual blocks are used, each of which is a neural network layer composed of two convolutional layers, and the goal of preserving the original image characteristics at the same time when converting can be achieved.
And (3) decoding: and (2) utilizing deconvolution layers to complete the work of reducing low-level features from the feature vectors, wherein five layers of one-dimensional deconvolution layers are adopted, the convolution kernel of the first convolution layer is 114, the step length is 1, the filling value is 0, the normalization is carried out after the middle deconvolution layer, the activation is carried out by using a ReLU function, each layer of one-dimensional convolution layer adopts a convolution kernel sliding window with the length of 4, the step length is set to be 2, the filling value is 1, the Tanh activation function is adopted after the last layer of deconvolution, and finally the generated EEG data is obtained.
As shown in fig. 4, the discriminator is implemented as follows:
the discriminator takes an EEG map as input and tries to predict it as raw EEG data or as output data of the generator. The invention designs a discrimination block: for the input data, five one-dimensional convolution operations are carried out, normalization is carried out after the first four convolution layers, and activation is carried out by using a LeakyReLU function, wherein each one-dimensional convolution layer adopts a convolution kernel sliding window with the length of 4, the step length is set to be 2, and the filling value is 1. And activating by using a Sigmoid function after the last convolution layer, wherein the convolution kernel of the layer is 114, the step length is 1, the filling is 0, and a one-dimensional feature is output and used for judging whether the convolution kernel is true or false.
The loss function is set as follows:
discriminator loss function: first, the discriminator is trainedSo that D isXStarts to judge that the field X data is close to 1, and the discriminator, DYThe same is true for the determination of (1). Thus, the discriminator DXIt is desirable to minimize (D)X(x)-1)2Of course DYAs well as so.
Since the discriminator should be able to distinguish between the generated data and the original data, it should also predict 0, D, for the data generated by the generatorXHopefully minimize
The generator loss function: the generator should eventually be able to "fool" the authenticity of the data generated by the arbiter. This can be done if the estimate of the generated data by the arbiter is as close to 1 as possible. So the generator needs to be minimized (D)Y(G(x)-1))2
Round consistent loss function: this is also the most important loss function that reflects the difference between the original and the reduced data, i.e. minimizes
Finally, it is combined with these loss functions, i.e.
And training is carried out to obtain corresponding EEG data of the removed artifact.
Claims (8)
1. A method for restoring electroencephalogram artifacts based on a circularly generated countermeasure network is characterized by comprising the following steps:
step 1: acquiring data;
designing a target artifact collection paradigm according to requirements, and collecting electroencephalogram data through an electroencephalogram cap;
step 2: preprocessing data;
step 2-1: processing the acquired electroencephalogram data by a filter comprising recess filtering and band-pass filtering;
step 2-2: the processed data are stored in a segmented mode according to the data labels and used for subsequent model training;
and step 3: and designing and training a corresponding Cycle to generate a confrontation network model by utilizing a Cycle GAN network model, and generating an electroencephalogram signal containing necessary information.
2. The electroencephalogram artifact restoration method based on the loop generation countermeasure network according to claim 1, characterized in that the data acquisition in the step 1 sets a corresponding paradigm according to the artifact to be removed, and the task has three commands of a left shaking motion, a right shaking motion and a static posture maintaining.
3. The method for restoring the electroencephalogram artifacts based on the cyclic generation countermeasure network as claimed in claim 2, characterized in that the corresponding paradigm of step 1 is specifically designed as follows:
the subject sits on a chair with armrests, and the task is to execute three commands of a left side shaking motion, a right side shaking motion and a static posture keeping according to the prompt; the order of the prompts is random; the experiment comprises multiple groups of experiment collection, each time comprises 60 times; after the start of the test, the screen was blank for the first 2 seconds, at t ═ 2s, the audio stimulus indicated the start of the test and a cross "+" was displayed; then starting to display a left arrow, a right arrow or a circular prompt mark for 1 second from t-3 s; while the subject is asked to execute a command to shake the head to the left once, to the right once, or to remain stationary, shaking the head back to the initial position, until the cross disappears at t-7 s; each of the 3 cues was displayed 20 times in random order in each experiment.
4. The method for restoring the electroencephalogram artifact based on the cycle generation countermeasure network as claimed in claim 3, characterized in that the core in the step 3 is that the optimization objective function is:
where G is the generator for generating from domain X to domain Y, F is the generator for generating from domain Y to domain X, DX,DYRespectively, a discriminator for discriminating whether the domain belongs to the distribution conforming to the domain X or the domain Y.
5. The electroencephalogram artifact restoration method based on the cycle generation confrontation network as claimed in claim 4, characterized in that the step 3 is implemented as follows:
with (x)i,yi) Representing a pair of artefact-and non-artefact EEG signals, i.e. (x)i,yi)|xi∈X,yiE is Y; the generator is used in the EEG-Cycle GAN: g takes the EEG signal with the target artifact as input, outputs the generated artifact-free EEG signal:a discriminator: the D inputs are pairs of EEG signals with target artifacts and artifact-free EEG signals, the target of D is the learning mapping: d (x)i,yi) → 1 andmapping a real pair of an EEG signal with a target artifact and an artifact-free EEG signal to 1, and mapping the EEG signal with the target artifact and the generated artifact-free EEG signal to 0; and the goal of G is to learn the mapping: d (x)i,G(xiZ)) → 1, i.e. making the generated artifact-free EEG signal more "real" and able to form a pair with the input EEG signal with the target artifacts;
due to the fact thatIt is difficult to find the corresponding relation in the domain Y, so a cyclic consistency loss is introduced, with another generator F: y → X, willAgain transferred back into field X to obtainF is generatedShould be consistent with the x holding domain; then there is
X → Y discriminator loss:
wherein G (x) that G attempts to generate is data that "looks" to conform to domain Y; dYIs used for distinguishing G (x) and a real sample y; the goal of G is to minimize it, while the goal of D is to maximize it, i.e.
Similarly, the discriminator loss of Y → X:
recycling back to "fight" a round is:
after combination, the following steps are carried out:
wherein, η can be used to control the weight of two objective functions, and n is usually 2;
the final objective function is:
6. the method for restoring the electroencephalogram artifact based on the cycle-generated countermeasure network according to claim 5, characterized in that the generator in the step 3 is specifically realized as follows:
and (3) encoding: the first step is to extract features from the input image by using a convolutional neural network; specifically, five one-dimensional convolutional layers are adopted, wherein each one-dimensional convolutional layer of the first four layers adopts a convolutional kernel sliding window with the length of 4, the step length is set to be 2, the filling value is 1, the last layer adopts the setting that the convolutional kernel of the layer is 114, the step length is 1 and the filling value is 0, and the five one-dimensional convolutional layers are used for extracting and coding the features;
conversion: converting the feature vector of the image in the domain X into a feature vector in the domain Y by combining the dissimilar features of the matrix; the method uses a plurality of layers of residual blocks, each residual block is a neural network layer formed by two convolution layers, and can achieve the aim of simultaneously retaining the characteristics of an original image during conversion;
and (3) decoding: and (2) utilizing deconvolution layers to complete the work of reducing low-level features from feature vectors, adopting five layers of one-dimensional deconvolution layers, wherein the convolution kernel of the first convolution layer is 114, the step length is 1, the filling value is 0, the normalization is carried out after the middle deconvolution layer, the activation is carried out by using a ReLU function, each layer of one-dimensional convolution layer adopts a convolution kernel sliding window with the length of 4, the set step length is 2, the filling value is 1, the Tanh activation function is adopted after the last layer of deconvolution, and finally the generated EEG data is obtained.
7. The method for restoring the electroencephalogram artifact based on the cycle generation countermeasure network according to claim 5, characterized in that the discriminator in the step 3 is specifically realized as follows:
the arbiter takes an EEG map as input and attempts to predict it as raw EEG data or as output data of the generator; the discrimination block is designed so: performing five one-dimensional convolution operations on input data, performing normalization and development activation by using a LeakyReLU function after the first four convolution layers, wherein each one-dimensional convolution layer adopts a convolution kernel sliding window with the length of 4, the set step length is set to be 2, and the filling value is 1; and activating by using a Sigmoid function after the last convolution layer, wherein the convolution kernel of the layer is 114, the step length is 1, the filling is 0, and a one-dimensional feature is output and used for judging whether the convolution kernel is true or false.
8. The method for restoring the electroencephalogram artifact based on the loop-generated countermeasure network according to the claim 6 or 7, characterized in that the loss function is set as follows:
discriminator loss function: the arbiter is first trained so that DXStarts to judge that the field X data is close to 1, and the discriminator, DYThe discrimination of (1) is also; thus, the discriminator DXIt is desirable to minimize (D)X(x)-1)2Of course DYThe same is true for the same;
since the discriminator should be able to distinguish between the generated data and the original data, it should also predict 0, D, for the data generated by the generatorXHopefully minimize
The generator loss function: the generator should eventually be able to "fool" the authenticity of the data generated by the arbiter; this can be done if the estimate of the generated data by the arbiter is as close to 1 as possible; so the generator needs to be minimized (D)Y(G(x)-1))2
Round consistent loss function: this is also the most important loss function, which reflects the difference between the original and the reduced data, i.e. minimizes
Finally, it is combined with these loss functions, i.e.
And training is carried out to obtain corresponding EEG data of the removed artifact.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011206585.4A CN112450946A (en) | 2020-11-02 | 2020-11-02 | Electroencephalogram artifact restoration method based on loop generation countermeasure network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011206585.4A CN112450946A (en) | 2020-11-02 | 2020-11-02 | Electroencephalogram artifact restoration method based on loop generation countermeasure network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112450946A true CN112450946A (en) | 2021-03-09 |
Family
ID=74835765
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011206585.4A Pending CN112450946A (en) | 2020-11-02 | 2020-11-02 | Electroencephalogram artifact restoration method based on loop generation countermeasure network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112450946A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113081001A (en) * | 2021-04-12 | 2021-07-09 | 杭州电子科技大学 | Method for removing BCG artifact of synchronous EEG-fMRI (electroencephalogram-based magnetic resonance imaging) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108564039A (en) * | 2018-04-16 | 2018-09-21 | 北京工业大学 | A kind of epileptic seizure prediction method generating confrontation network based on semi-supervised deep layer |
CN110169770A (en) * | 2019-05-24 | 2019-08-27 | 西安电子科技大学 | The fine granularity visualization system and method for mood brain electricity |
CN110575141A (en) * | 2019-09-09 | 2019-12-17 | 西北大学 | Epilepsy detection method based on generation countermeasure network |
CN110840445A (en) * | 2019-12-12 | 2020-02-28 | 山东省计算中心(国家超级计算济南中心) | Automatic noise reduction method for dynamic electrocardiosignals |
CN111523520A (en) * | 2020-06-11 | 2020-08-11 | 齐鲁工业大学 | Method for analyzing electroencephalogram signals of brain patients with motor imagery stroke by using cycleGAN |
CN111814656A (en) * | 2020-07-02 | 2020-10-23 | 山东省人工智能研究院 | Electrocardiosignal noise reduction method based on countermeasure generation network |
-
2020
- 2020-11-02 CN CN202011206585.4A patent/CN112450946A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108564039A (en) * | 2018-04-16 | 2018-09-21 | 北京工业大学 | A kind of epileptic seizure prediction method generating confrontation network based on semi-supervised deep layer |
CN110169770A (en) * | 2019-05-24 | 2019-08-27 | 西安电子科技大学 | The fine granularity visualization system and method for mood brain electricity |
CN110575141A (en) * | 2019-09-09 | 2019-12-17 | 西北大学 | Epilepsy detection method based on generation countermeasure network |
CN110840445A (en) * | 2019-12-12 | 2020-02-28 | 山东省计算中心(国家超级计算济南中心) | Automatic noise reduction method for dynamic electrocardiosignals |
CN111523520A (en) * | 2020-06-11 | 2020-08-11 | 齐鲁工业大学 | Method for analyzing electroencephalogram signals of brain patients with motor imagery stroke by using cycleGAN |
CN111814656A (en) * | 2020-07-02 | 2020-10-23 | 山东省人工智能研究院 | Electrocardiosignal noise reduction method based on countermeasure generation network |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113081001A (en) * | 2021-04-12 | 2021-07-09 | 杭州电子科技大学 | Method for removing BCG artifact of synchronous EEG-fMRI (electroencephalogram-based magnetic resonance imaging) |
CN113081001B (en) * | 2021-04-12 | 2022-04-01 | 杭州电子科技大学 | Method for removing BCG artifact of synchronous EEG-fMRI (electroencephalogram-based magnetic resonance imaging) |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111329474B (en) | Electroencephalogram identity recognition method and system based on deep learning and information updating method | |
CN111523601B (en) | Potential emotion recognition method based on knowledge guidance and generation of countermeasure learning | |
CN110555468A (en) | Electroencephalogram signal identification method and system combining recursion graph and CNN | |
CN112022153B (en) | Electroencephalogram signal detection method based on convolutional neural network | |
CN112488002B (en) | Emotion recognition method and system based on N170 | |
Chuang et al. | IC-U-Net: a U-Net-based denoising autoencoder using mixtures of independent components for automatic EEG artifact removal | |
CN109657646B (en) | Method and device for representing and extracting features of physiological time series and storage medium | |
Yang et al. | A novel deep learning scheme for motor imagery EEG decoding based on spatial representation fusion | |
CN113180692A (en) | Electroencephalogram signal classification and identification method based on feature fusion and attention mechanism | |
CN113208593A (en) | Multi-modal physiological signal emotion classification method based on correlation dynamic fusion | |
CN113974627B (en) | Emotion recognition method based on brain-computer generated confrontation | |
CN113180659A (en) | Electroencephalogram emotion recognition system based on three-dimensional features and cavity full convolution network | |
Zhang et al. | An amplitudes-perturbation data augmentation method in convolutional neural networks for EEG decoding | |
CN114601476A (en) | EEG signal emotion recognition method based on video stimulation | |
CN113128353B (en) | Emotion perception method and system oriented to natural man-machine interaction | |
CN112450946A (en) | Electroencephalogram artifact restoration method based on loop generation countermeasure network | |
CN114081503A (en) | Method for removing ocular artifacts in electroencephalogram signals | |
CN113208614A (en) | Electroencephalogram noise reduction method and device and readable storage medium | |
CN111772629A (en) | Brain cognitive skill transplantation method | |
CN116421200A (en) | Brain electricity emotion analysis method of multi-task mixed model based on parallel training | |
CN115690528A (en) | Electroencephalogram signal aesthetic evaluation processing method, device, medium and terminal across main body scene | |
CN114569141A (en) | Epilepsy prediction system based on feature channel fusion and deep learning | |
Liu et al. | EEG classification algorithm of motor imagery based on CNN-Transformer fusion network | |
Wang et al. | Residual learning attention cnn for motion intention recognition based on eeg data | |
Tang et al. | Dynamic pruning group equivariant network for motor imagery EEG recognition |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210309 |