CN115981462A - Multi-user and multi-edge intelligent brain-computer personalized electroencephalogram interaction system - Google Patents
Multi-user and multi-edge intelligent brain-computer personalized electroencephalogram interaction system Download PDFInfo
- Publication number
- CN115981462A CN115981462A CN202211605095.0A CN202211605095A CN115981462A CN 115981462 A CN115981462 A CN 115981462A CN 202211605095 A CN202211605095 A CN 202211605095A CN 115981462 A CN115981462 A CN 115981462A
- Authority
- CN
- China
- Prior art keywords
- electroencephalogram
- module
- feature extraction
- data
- unit
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention discloses a multi-user and multi-edge intelligent brain-computer personalized electroencephalogram interaction system, which comprises an electroencephalogram signal acquisition module, an electroencephalogram characteristic extraction and encoding module and an edge intelligent decoding and task execution module; the brain electrical signal acquisition module is used for acquiring brain signal data; the electroencephalogram characteristic extraction and encoding module is used for learning and encoding the characteristics of the brain signal data into electroencephalogram instructions, and comprises an emitter module and a wireless channel module; the emitter module comprises an electroencephalogram data feature extraction unit and a channel coding unit; the wireless channel module comprises a channel transmission unit and a neural network cross mutual information entropy model; the edge intelligent agent decoding and task executing module is used for restoring and identifying the electroencephalogram instructions and generating different executing actions according to different user instructions. The method realizes the personalized electroencephalogram learning and improves the personalized cognition and classification precision of the electroencephalogram of the user.
Description
Technical Field
The invention relates to the technical field of electroencephalogram data, in particular to a multi-user and multi-edge intelligent brain-computer personalized electroencephalogram interaction system.
Background
Metaverse is becoming the most promising platform for the future internet, where BCI (brain-computer interface) is one of the key technologies of Metaverse. Compared to invasive BCI, which is implanted directly into the brain in neurosurgery, non-invasive BCI works mainly on the principle of EEG (brain waves), is relatively low-priced, and has a better market penetration potential. At present, our approach focuses on non-invasive BCI and multi-brain and multi-robot interaction, which has been hardly addressed in previous work. There is an increasing demand for intelligent agents and human-object interactions in metastic scenes, such as co-creation, autonomous manufacturing or unmanned factories, and immersive technology or augmented reality (XR) scenes. Therefore, the scenarios defined in our framework are of interest for both research and practice. Meanwhile, a general framework of EEG adaptive neural network and semantic communication is provided for brain signal analysis and feedback of heterogeneous or multi-modal data generated by various sensing devices. Its design and optimization scheme gives a reference case for integrated sensing, computing, communication and control, which can benefit the industry and academia devoted to 6G and above. However, most of the existing technologies only have single considered one-to-one brain-controlled agents, which is not enough for the potential multi-user-agent interaction scene in the future; meanwhile, for a scene with multiple types of brain-computer equipment in cooperation, the prior art has no good method for overcoming the problem.
Disclosure of Invention
In order to solve the technical problems, the invention discloses a multi-user and multi-edge intelligent brain-computer personalized electroencephalogram interaction system, which realizes personalized electroencephalogram learning through meta-learning, improves the electroencephalogram personalized cognition and classification precision of users, solves the problem of multi-user multi-device electroencephalogram signal real-time access in a dynamic coding and decoding process, solves the problem of multi-user electroencephalogram instruction transmission in a wireless signal-noise environment through semantic communication, and solves the problem of multi-user instruction identification of multiple robots through codemap.
A multi-user and multi-edge intelligent brain-computer personalized electroencephalogram interaction system comprises an electroencephalogram signal acquisition module, an electroencephalogram feature extraction and encoding module and an edge intelligent decoding and task execution module;
the brain electrical signal acquisition module is used for acquiring brain signal data;
the electroencephalogram characteristic extraction and encoding module is used for learning and encoding the characteristics of the brain signal data into electroencephalogram instructions, and comprises an emitter module and a wireless channel module; the transmitter module comprises an electroencephalogram data feature extraction unit and a channel coding unit; the wireless channel module comprises a channel transmission unit and a neural network cross mutual information entropy model;
the edge intelligent body decoding and task executing module is used for restoring and identifying the electroencephalogram instructions and generating different executing actions according to different user instructions.
Furthermore, the electroencephalogram feature extraction and encoding module further comprises a light-weight transmitter module, and the light-weight transmitter module comprises an electroencephalogram data feature extraction unit and a channel encoding unit which are subjected to model compression processing.
Further, the edge agent decoding and task execution module comprises a receiver module, and the receiver module comprises an edge decoding unit and an execution unit.
Furthermore, the electroencephalogram data feature extraction unit comprises a common feature extraction module, a low-dimensional feature extraction model, a high-dimensional feature extraction model and a dynamic decoding unit; the low-dimensional feature extraction model is used for improving the identification precision of the low-dimensional electroencephalogram data and extracting the features of the low-dimensional electroencephalogram data, the high-dimensional feature extraction model is used for extracting the features of the high-dimensional electroencephalogram data, and the dynamic decoding unit is used for adjusting the selection of the low-dimensional feature extraction model and the high-dimensional feature extraction model.
Further, the common feature extraction module comprises a convolution layer, an embedding layer, a spacing mask layer and a common feature extraction layer.
Further, the wireless channel module is configured to perform noise superposition simulation interference on the incoming channel coding characteristics, and the wireless channel module simulates transmission interference of a wireless environment in a noise simulation manner.
The invention has the beneficial effects that:
1. compared with the prior art, the invention provides a combined feature based on a discrete attention mechanism to extract heterogeneous EEG data, and then dynamic feature integration is used to improve classification precision and realize accurate homogenization information extraction of mixed EEG data.
2. The invention designs a dynamic semantic automatic encoder, which inherits the semantic information of an emitter by a brain converter and a corresponding semantic automatic decoder; meanwhile, a semantic performance index is established through a mutual information function and a cross entropy function so as to measure the performance of a channel coder-decoder, measure the performance of the semantic coder-decoder and accurately identify instructions of electroencephalogram data with various dimensions,
3. in terms of deployment and agent identification instructions, the invention deploys a model compression scheme with pruning, weight sharing and quantization to support transmitters to run semantic encoder models on resource-limited edge devices; a lightweight channel decoder and a semantic decoder with residual technology and a full connection layer are also deployed at a receiving end; in addition, a code map representing various commands is provided for a plurality of users to control a plurality of intelligent agents.
Drawings
FIG. 1 is a schematic diagram of a brain-computer personalized electroencephalogram interaction system for multiple users and multiple edge intelligent bodies in an embodiment of the invention;
FIG. 2 is a schematic diagram of a dynamic electroencephalogram feature extractor in an embodiment of the present invention;
FIG. 3 is a schematic diagram of a common feature extraction mechanism in an embodiment of the present invention;
fig. 4 is a schematic diagram of a Brain Transformer module of a low-dimensional feature extraction model according to an embodiment of the present invention.
Fig. 5 is a schematic diagram of a semantic communication process in an embodiment of the invention.
FIG. 6 is a schematic diagram illustrating an example of real-time multi-user and multi-agent brain-computer interaction in an embodiment of the present invention
FIG. 7 is a schematic diagram of the accuracy of each model of an electroencephalogram instruction classification experiment under mixed data input by various types of brain-computer equipment in the embodiment of the invention
FIG. 8 is a diagram illustrating comparison experiment results of classification accuracy of mixed data under noise interference of various types of wireless channels according to an embodiment of the present invention
Detailed Description
The embodiments of the present invention will be further described with reference to the drawings and examples. It should be noted that the examples do not limit the scope of the claimed invention.
Example 1
As shown in figure 1, the invention discloses a multi-user and multi-edge intelligent brain-computer personalized electroencephalogram interaction system, which comprises an electroencephalogram signal acquisition module, an electroencephalogram feature extraction and encoding module and an edge intelligent decoding and task execution module.
The electroencephalogram signal acquisition module is used for acquiring brain signal data, a plurality of users can wear various electroencephalogram devices such as a brainnlink series, an Emotiv series and a 64-channel electrode cap, the electroencephalogram devices can acquire electroencephalogram signals of different brain areas and different channel numbers to form electroencephalogram original data of different dimensionalities, and the electroencephalogram original data can be used as the input of the electroencephalogram feature extraction and encoding module.
The brain electrical characteristic extracting and coding module is used for learning and coding the characteristics of brain signal data and comprises an emitter module and a wireless channel module. The emitter module comprises an electroencephalogram data feature extraction unit and a channel coding unit; the wireless channel module comprises a channel transmission unit and a neural network cross mutual information entropy model.
The edge intelligent agent decoding and task executing module is used for decoding the electroencephalogram instructions transmitted through the wireless channel and comprises a receiver module, and the receiver module comprises an edge decoding unit and an executing unit.
In order to realize the accurate classification of multi-type multi-dimensional electroencephalogram signals and form electroencephalogram instructions, as shown in fig. 2, the invention provides a dynamic electroencephalogram feature extraction unit which mainly comprises a common feature extraction module, a low-dimensional feature extraction model, a high-dimensional feature extraction model and a dynamic decoding unit.
In the common feature extraction module, electroencephalogram data with different dimensionalities are mapped in the same hidden space through the convolution layer, the mapped data are input into the embedding layer to perform data preliminary learning, and embedding features output by the embedding layer are subjected to interval mask processing so as to perform destructive interference on the electroencephalogram data. For example, three electroencephalogram data x1[3,150], x2[4,260], and x3[64,640] are mapped to a hidden space of [64,256] together, namely x1', x2', and x3 'of data dimensions after convolution are [64,256], and the x1', x2', and x3' are placed into a common feature extraction module for similar feature learning and output to a low/high-dimensional feature extraction model as influence factors for utilization. The embedded layer is a convolution layer with convolution kernel of 1, and the input and output dimensions are the same. The reason why the interval mask processing is performed is: the inventor finds through experiments that the electroencephalogram data not only have time dependence and space dependence, but also have a tendency in damaged data, and the tendency may improve the cognitive ability of the model on the electroencephalogram data and is called as the tendency dependence; accordingly, the data is subjected to interval masking, namely, non-zero values are set to be 0 at intervals so as to highlight the trend dependence existing in the data.
And then, extracting common features of the data after the interval mask processing by using a multi-head attention mechanism, wherein the multi-head attention mechanism calculates the corresponding relation of every two electroencephalogram values on different sampling time points so as to obtain the relevance of the time. Recent research shows that the traditional Self-attention needs to pay higher-complexity memory and secondary dot product calculation as cost, and is a main defect of the prediction capability; meanwhile, the research also provides that the importance scores of the calculated values of the sparsity self-attention form long-tail distribution, namely only a few dot products contribute to the main attention calculation of the result, other dot products can be ignored for the result, and the important dot product pairs are considered to be uniformly distributed. And (3) utilizing a probability distribution formula and uniform distribution of the KL divergence formula to the traditional attention mechanism:
wherein, the first and the second end of the pipe are connected with each other,
is a probability distribution of attention mechanism>
Is uniformly distributed, and the water-soluble polymer is uniformly distributed,
is an asymmetric exponential kernel function. Substituting the divergence value into a divergence formula to calculate a KL (q | | p) rejection constant term to obtain the divergence evaluation of the ith query value in the attention mechanism as follows: />
If a U value is determined, and let U = L Q lnL K The computational complexity of the attention mechanism can be greatly reduced, and meanwhile, important dot product pairs are guaranteed to be reserved in the next computation without influencing the computation performance. Where U is the value of K in fig. 3, K =30 is chosen in the system to retain the valid dot product pair components to form the dot product pair matrix. In order to guarantee the matrix dimension adaptation problem in the calculation process, other dimensions of the matrix are filled into corresponding query average values to form an intermediate common feature, the intermediate common feature is activated through a convolution layer to serve as the common feature of each signal which is finally extracted, and the common feature is transmitted to the next module to be learned and utilized.
In order to solve the problems of low information content and high classification difficulty of low-dimensional electroencephalogram data (corresponding to low-cost and high-noise electroencephalogram equipment), a low-dimensional feature extraction model is used for improving the identification precision, and the low-dimensional feature extraction model is composed of a Deepconvnet and a Brain transducer (the upper right area of a figure 2). By utilizing the parallel learning structure and the application of a residual mechanism, the classification precision of the low-dimensional electroencephalogram signals can be improved. As shown in FIG. 4, the Brain transducer embeds a temporal feature Transformer, the main mechanism of which is a multi-head attention mechanism, to extract the temporal characteristics of data efficiently; after multiple layers of normalization and multiple residual fusion, the data flow to an integration layer formed by combining two layers of convolution and a layer of full connection layer, and a classification prediction feature is output after integration. Before the low-dimensional feature extraction model is output, two full-connection layers are introduced to learn the weight of addition of two parameters, the extracted low-dimensional features and the common features output by the common feature extraction module are input into the two full-connection layers as two addition parameters, the low-dimensional features and the common features are mutually influenced in a self-learning optimal mode to obtain optimal features and then are fused, and finally a fused classification result is output.
In the high-dimensional feature extraction model, the model (the lower right half area of fig. 2) is mainly improved based on an EEGNet (neural network structure), a pooling layer built in the EEGNet is replaced by an average pooling layer, and data is averagely evaluated during pooling, so that the data is beneficial to reserving more background information, namely, reserving the mutual influence of electroencephalogram data on time; and introducing two full-connection layers to learn the weight of adding two parameters before EEGNet, namely performing self-learning adding weight on the input original high-dimensional data and common features to obtain optimal feature pre-fusion, and learning the fused features.
The dynamic decoding unit is used for adjusting the selection of the low-dimensional feature extraction model and the high-dimensional feature extraction model, so that different electroencephalogram data have better learning condition and classification capability, and the problem of accuracy reduction caused by information conflict among data is solved. The feature extraction models with low dimensionality and high dimensionality are selectively used for updating and learning parameters, the dynamic decoding unit can automatically freeze the parameters according to the dimensionality of input data, if the low dimensionality data are input into the electroencephalogram feature extraction unit, the high dimensionality feature extraction model part can be frozen by the parameters and does not participate in learning and predicting tasks, and vice versa.
The channel coding unit consists of two layers of convolution layers, the dimensionality is reformed simultaneously, instruction classification results output by the electroencephalogram equipment are integrated into an ordered one-dimensional matrix, and then the channel coding unit is output.
In order to facilitate deployment, the electroencephalogram feature extractor unit and the channel coding unit are lightened by a model compression method so as to break through the small calculation amount limitation of the embedded intelligent unit; the model is compressed by means of model pruning (model branches with smaller clipping importance), weight sharing and quantification from the feature extraction model to any server or intelligent unit body with limited resources, the occupied memory is reduced by 60%, the running speed is reduced by 45%, and the precision is kept unchanged.
The wireless channel module is used for carrying out noise superposition simulation interference on the incoming channel coding characteristics. In the process that the transmitter module transmits the processed electroencephalogram signal characteristics to the receiver module in a wireless broadband mode, generally speaking, the instructions are greatly influenced by noise interference in the wireless transmission process, and distortion is caused. In order to solve the problem, a noise simulation mode is adopted to simulate the transmission interference of three wireless environments (additive white gaussian noise, rice channel and rayleigh fading channel) to train our framework to show the robustness of the transmission interference in different wireless noise environments, and noise superposition simulation interference is carried out on the incoming channel coding characteristics.
In the process, the invention provides a multi-information loss entropy calculation model to learn the loss function so as to better obtain good communication reduction performance and realize a higher-precision semantic communication process, and the calculated loss value can be used for the back propagation calculation of the loss of the whole model. Specifically, a small neural network cross mutual information entropy model is established to calculate signal data z which is sent from a transmitter module and is not interfered by noise and signal data which is not sent into a receiver module after being interfered by noise
The neural network cross mutual information entropy model firstly carries out cross entropy calculation on the two data, then half of the two data are mutually exchanged to form new two data, and a cross entropy is calculated again, wherein the two data are crossedThe entropy is added to form a loss value for the overall deep learning model to perform back propagation calculation.
The receiver module includes an edge decoding unit and an execution unit.
The interference features processed by the wireless channel module are transmitted to an edge decoding unit in the receiver module for decoding, the edge decoding unit is configured on each intelligent agent, and the edge decoding unit mainly comprises a semantic decoder and a channel decoder for restoring and identifying electroencephalogram commands sent by users. The receiver module has limited computing resources and requires a fast transportation speed, so that a portable module is required to realize the instruction decoding process. Specifically, a semantic decoder is formed by a dimension adjusting layer and two convolution layers to carry out dimension reforming on one-dimensional data passing through a wireless channel and decode an original semantic instruction of the data. And then, the channel decoder is a full connection layer and a sigmoid activation layer, and activates and outputs each semantic instruction. Therefore, the multiple instructions of multiple users are sequenced according to the input time, all the edge agents are conveyed in a two-dimensional table mode, and the edge agents can receive all the user instructions.
And finally, arranging the instructions of all users according to input time and input users, searching a task list of each edge intelligent agent in the input instructions according to a preset instruction list shown in the table 1, and executing the tasks of all robots one by an execution unit on each intelligent agent according to time so as to achieve the brain-computer interaction process of multiple users and multiple intelligent agents. Each agent has a respective execution unit, and different execution actions are generated according to different user instructions. After receiving all user instruction information, the intelligent agents search tasks belonging to the intelligent agents one by one according to the instruction list, and realize preset behavior tasks according to instruction classification, so that the behavior of multi-user cooperative control of multi-intelligent-agent interaction is finally achieved.
TABLE 1 instruction sheet
The accuracy experiment of instruction classification is carried out on the data sets by applying the four data sets, wherein the four data sets are respectively as follows: 1. a Brainlink dataset. This data set was collected using brain link lite for continuous electroencephalographic measurements. 2. Eyestate eye movement data set. This data set was obtained by performing successive EEG measurements with an emotv EEG neuroheader set. 3. BCI-2000 contest data sets. This data set collected 64 channels of electroencephalographic signals from the volunteers. 4. The data sets are mixed. This data set is a combination of the three data sets described above, containing ten categories. The hybrid training data consists of all of the three data sets, including 400-800 samples randomly selected from each of the three data sets.
The following seven models were used for data alignment:
EEGNet, a compact full convolution network for EEG-based BCI.
Compact-CNN, a Compact ConvNet, is used to decode signals of the 12-class SSVEP dataset without user-specific calibration.
Depconcovnet and shallowconcovnet are not only a novel and promising tool in the EEG decoding toolset, but are also combined with innovative visualization techniques.
EEG-TCNet, a novel temporal ConvNet, requires only a few trainable parameters to achieve excellent accuracy.
ResNet, a residual learning scheme, simplifies the training of deep neural networks.
Deep brain, a transformer-based variant, performs well in low latitude electroencephalogram feature extraction.
TABLE 2
As shown in Table 2, the performance of all models in a single data set (corresponding to a single brain electrical device access scenario) was tested. Our models are all higher in classification accuracy than other models.
Mixing the three previous data sets; different types of data can interfere with the learning of the model, thereby greatly affecting the classification accuracy of the model. The multi-user and multi-edge intelligent brain-computer personalized electroencephalogram interaction system (Our) adopts a dynamic encoding and decoding process, which well avoids the interference among different equipment data, so that the experimental precision of the invention is far higher than that of other methods. The result is shown in fig. 7, and the result shows that the brain-computer personalized electroencephalogram interaction system for the multi-user and multi-edge intelligent bodies has optimal and different instruction identification capability in the scene of simultaneously inputting and cooperatively controlling the intelligent bodies by various electroencephalogram devices.
Further, in order to prove that the multi-user and multi-edge intelligent brain-computer personalized electroencephalogram interaction system has good performance in different wireless noise environments, a mixed data set is adopted to test Our framework in three different wireless channel environments and is compared with an improved semantic communication framework, and as shown in fig. 8, the result shows that the multi-user and multi-edge intelligent brain-computer personalized electroencephalogram interaction system (Our) obtains more excellent instruction accuracy and robustness in three different noise environments.
Finally, the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting, although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions may be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all of them should be covered in the claims of the present invention.
Claims (9)
1. A multi-user and multi-edge intelligent brain-computer personalized electroencephalogram interaction system is characterized by comprising an electroencephalogram signal acquisition module, an electroencephalogram feature extraction and encoding module and an edge intelligent decoding and task execution module;
the brain electrical signal acquisition module is used for acquiring brain signal data;
the electroencephalogram characteristic extraction and encoding module is used for learning and encoding the characteristics of the brain signal data into electroencephalogram instructions, and comprises an emitter module and a wireless channel module; the emitter module comprises an electroencephalogram data feature extraction unit and a channel coding unit; the wireless channel module comprises a channel transmission unit and a neural network cross mutual information entropy model;
the edge intelligent body decoding and task executing module is used for restoring and identifying the electroencephalogram instructions and generating different executing actions according to different user instructions.
2. The system of claim 1, wherein the electroencephalogram feature extraction and encoding module further comprises a lightweight transmitter module, and the lightweight transmitter module comprises an electroencephalogram data feature extraction unit and a channel encoding unit after model compression processing.
3. The system of claim 1, wherein the edge agent decoding and task execution module comprises a receiver module, and the receiver module comprises an edge decoding unit and an execution unit.
4. The system of claim 1, wherein the electroencephalogram data feature extraction unit comprises a common feature extraction module, a low-dimensional feature extraction model, a high-dimensional feature extraction model and a dynamic decoding unit; the low-dimensional feature extraction model is used for improving the identification precision of the low-dimensional electroencephalogram data and extracting the features of the low-dimensional electroencephalogram data, the high-dimensional feature extraction model is used for extracting the features of the high-dimensional electroencephalogram data, and the dynamic decoding unit is used for adjusting the selection of the low-dimensional feature extraction model and the high-dimensional feature extraction model.
5. The system of claim 4, wherein the common feature extraction module comprises a convolutional layer, an embedded layer, a spacing mask layer and a common feature extraction layer.
6. The system of claim 1, wherein the wireless channel module is configured to perform noise-superimposed analog interference on the incoming channel coding features, and the wireless channel module simulates transmission interference in a wireless environment in a noise simulation manner.
7. The system of claim 1, wherein the wireless channel module comprises a neural network cross mutual information entropy model, and the neural network cross mutual information entropy model is used for transmitting signal data z which is sent from the transmitter module and is not interfered by noise and transmitting signal data which is not sent to the receiver module after being interfered by noise
And performing first cross entropy calculation, then exchanging half of the two data with each other, and then calculating second cross entropy again, wherein the first cross entropy and the second cross entropy are added to form a loss value for performing back propagation calculation on the whole deep learning model.
8. The system of claim 1, wherein the edge agent decoding and task execution module comprises a receiver module, and the receiver module comprises an edge decoding unit and an execution unit.
9. The system of claim 8, wherein the edge decoder unit is configured to recover and identify an electroencephalogram command issued by a user, and the edge decoder unit comprises a semantic decoding unit and a channel decoding unit; the execution unit arranges the user instructions and generates execution actions according to the instruction list.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211605095.0A CN115981462A (en) | 2022-12-14 | 2022-12-14 | Multi-user and multi-edge intelligent brain-computer personalized electroencephalogram interaction system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211605095.0A CN115981462A (en) | 2022-12-14 | 2022-12-14 | Multi-user and multi-edge intelligent brain-computer personalized electroencephalogram interaction system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115981462A true CN115981462A (en) | 2023-04-18 |
Family
ID=85958951
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211605095.0A Pending CN115981462A (en) | 2022-12-14 | 2022-12-14 | Multi-user and multi-edge intelligent brain-computer personalized electroencephalogram interaction system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115981462A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117473303A (en) * | 2023-12-27 | 2024-01-30 | 小舟科技有限公司 | Personalized dynamic intention feature extraction method and related device based on electroencephalogram signals |
-
2022
- 2022-12-14 CN CN202211605095.0A patent/CN115981462A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117473303A (en) * | 2023-12-27 | 2024-01-30 | 小舟科技有限公司 | Personalized dynamic intention feature extraction method and related device based on electroencephalogram signals |
| CN117473303B (en) * | 2023-12-27 | 2024-03-19 | 小舟科技有限公司 | Personalized dynamic intention feature extraction method and related device based on electroencephalogram signals |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108319599B (en) | Man-machine conversation method and device | |
| Lan et al. | What is semantic communication? A view on conveying meaning in the era of machine intelligence | |
| CN108595494A (en) | The acquisition methods and device of reply message | |
| CN107870994A (en) | Man-machine interaction method and system for intelligent robot | |
| CN113469340A (en) | Model processing method, federal learning method and related equipment | |
| CN114423034B (en) | Indoor personnel action recognition method, system, medium, equipment and terminal | |
| CN111680147A (en) | Data processing method, device, equipment and readable storage medium | |
| US11458624B2 (en) | Control server and method for controlling robot using artificial neural network, and robot implementing the same | |
| CN115981462A (en) | Multi-user and multi-edge intelligent brain-computer personalized electroencephalogram interaction system | |
| CN114861740B (en) | Self-adaptive mechanical fault diagnosis method and system based on multi-head attention mechanism | |
| CN111050294A (en) | Indoor positioning system and method based on deep neural network | |
| CN112418175A (en) | Rolling bearing fault diagnosis method and system based on domain migration and storage medium | |
| Seo et al. | Toward semantic communication protocols: A probabilistic logic perspective | |
| EP4388459A1 (en) | Integrated sensing and communications empowered by networked hybrid quantum-classical machine learning | |
| Sruthi et al. | An improved Wi-Fi sensing-based human activity recognition using multi-stage deep learning model | |
| Mo et al. | A deep learning-based human identification system with wi-fi csi data augmentation | |
| Nam et al. | Language-oriented communication with semantic coding and knowledge distillation for text-to-image generation | |
| CN111679903A (en) | Edge cloud cooperation device for deep learning | |
| CN116761223B (en) | Method for realizing 4G radio frequency communication by using 5G baseband chip and vehicle-mounted radio frequency system | |
| CN113537053B (en) | Method for constructing radio frequency fingerprint identification model in civil aviation field | |
| CN115801152A (en) | WiFi action identification method based on hierarchical transform model | |
| CN113242201B (en) | Wireless signal enhanced demodulation method and system based on generation classification network | |
| Choong et al. | Jack and masters of all trades: One-pass learning sets of model sets from large pre-trained models | |
| CN115132181A (en) | Speech recognition method, speech recognition apparatus, electronic device, storage medium, and program product | |
| CN114781598A (en) | Fault prediction method based on hierarchical neural network distributed training |
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 |