CN-122020208-A - Personalized brain-computer interface system

CN122020208ACN 122020208 ACN122020208 ACN 122020208ACN-122020208-A

Abstract

The invention discloses a personalized brain-computer interface system, which comprises an electroencephalogram data engineering module, a personalized normal form library module, a personalized normal form selection module, a model matching and selection module, an offline personalized module and an online decoding and self-adapting module, wherein the modules work cooperatively, so that the full-flow personalization from normal form to decoding model is realized, and the user experience, personalized depth and long-term decoding performance of a BCI system are effectively considered.

Inventors

WU KAI
XIE JIYUAN
LI WENHAO

Assignees

华南理工大学

Dates

Publication Date: 20260512
Application Date: 20260414

Claims (10)

1. A personalized brain-computer interface system, comprising: the electroencephalogram data engineering module is used for collecting electroencephalogram signals of a user and preprocessing the electroencephalogram signals; The individuation Fan Shiku module is used for storing pre-verified brain-computer interface experimental norms, the norms adopt relational data to store norms metadata, and each norms comprises a norms ID, a norms type, a norms description, pre-verification indexes and recommendation indexes; the personalized paradigm selection module is used for displaying the brain-computer interface experimental paradigm and selecting the brain-computer interface experimental paradigm according to the quantization index and combining with user feedback; the model matching and selecting module is used for determining at least one candidate decoding model applicable to the target normal form according to a preset normal form-model corresponding relation, and determining an initial decoding model from the candidate decoding models based on calibration electroencephalogram data of a user; The off-line individuation module is used for setting initial individuation configuration parameters, training the initial decoding model by using reinforcement learning, and obtaining an individuation initial decoding model; And the online decoding and self-adapting module is used for decoding the real-time electroencephalogram signals of the user by utilizing the personalized initial decoding model, and continuously optimizing the personalized initial decoding model in an incremental learning mode based on the electroencephalogram data and decoding results newly generated in the online use process of the user.
2. The personalized brain-computer interface system according to claim 1, wherein the electroencephalogram data engineering module acquires a multichannel original electroencephalogram signal sequence at a set time, and performs preprocessing on the multichannel original electroencephalogram signal sequence, including bandpass filtering, notch filtering, artifact removal, baseline correction, re-referencing, and downsampling.
3. The personalized brain-computer interface system according to claim 1, wherein, The paradigm ID is a unique identifier; The pattern types include a motor imagery pattern, a visual motor imagery pattern, a spatial navigation imagery pattern, an SSVEP pattern, a P300 pattern, a speech imagery pattern, a music imagery pattern, and an image imagery pattern; the paradigm description comprises words and illustration, and is used for explaining the working principle of the paradigm and psychological tasks required to be executed by a user; the pre-verification indexes comprise decoding accuracy and average information transmission rate; the recommendation index feeds back the calculated score based on historical user selections.
4. The personalized brain-computer interface system according to claim 3, wherein the personalization Fan Shiku module collects subjective comfort scores, calculates a paradigm score according to weights based on decoding accuracy and average information transmission rate, and stores a corresponding pre-validated brain-computer interface experimental paradigm according to the paradigm score.
5. The personalized brain-computer interface system of claim 1, wherein the quantization index comprises a decoding performance index and a user subjective experience index; the system performs weighted comprehensive scoring on the quantization indexes as follows: , , Wherein, the For the purpose of the comprehensive assessment of the score, In order to decode the performance index(s), For the user to experience the index subjectively, In order to decode the weight of the performance indicator, The weight of subjective experience indexes of the user; the decoding performance index is specifically: , Wherein, the For the i-th classification accuracy in the classification accuracy sequence, For the i-th information transfer rate in the sequence of information transfer rates, In order to classify the maximum value in the accuracy sequence, For the maximum value in the sequence of information transfer rates, In order to classify the weight of the accuracy rate, Weight for information transfer rate; The subjective experience index of the user is specifically: , Wherein, the For the subjective difficulty score of the person, For the purpose of scoring the degree of fatigue, In order to make a self-evaluation of the concentration, For the subjective difficulty score weight, Weights for scoring the degree of fatigue, Is the weight of concentration self-evaluation.
6. The personalized brain-computer interface system according to claim 1, wherein the model matching and selection module maps respective decoding models from selected brain-computer interface experimental paradigms, each brain-computer interface experimental paradigm type mapped to a set of decoding models, each set of decoding models having at least one candidate decoding model; calculating a composite score based on the decoding performance index and the model complexity, as follows: , Wherein, the For the i-th classification accuracy in the classification accuracy sequence, For the i-th information transfer rate in the sequence of information transfer rates, For the complexity of the model it is desirable, In order to classify the maximum value in the accuracy sequence, For the maximum value in the sequence of information transfer rates, For the maximum value in the inverse model complexity sequence, In order to classify the weight of the accuracy rate, As a weight for the rate of information transfer, Weighting the complexity of the model; And obtaining an optimal decoding model, namely an initial decoding model, according to the highest comprehensive score.
7. The personalized brain-computer interface system according to claim 1, wherein said setting initial personalized configuration parameters comprises at least one of, A feature importance vector for weighting input brain electrical features; A super-parameter set of the initial decoding model; And adjusting parameters of part of network structures of the initial decoding model.
8. The personalized brain-computer interface system according to claim 1, wherein the reinforcement learning employs an actor-critter architecture, wherein: the state is the splicing characteristic of the historical average characteristic and the current characteristic; The actions are updated according to the types of the personalized targets, specifically, when the personalized targets are feature screening, a feature selection mode is adopted, the actions are feature weight vectors, each element represents the reserved weight of the corresponding feature, when the personalized targets are parameter optimization, a super-parameter optimization mode is adopted, and the actions are super-parameter adjustment amounts; and acquiring optimal network parameters according to the updated actions, and acquiring a personalized initial decoding model according to the optimal network parameter configuration.
9. The personalized brain-computer interface system according to claim 1, wherein the online decoding and adaptation module comprises: the data buffer area is used for circularly storing the brain electricity data newly generated by the user and the decoding result; the performance detection unit is used for detecting the decoding performance of the personalized initial decoding model in real time; And the increment updating unit is used for analyzing the decoding performance, and when the decoding performance is lower than a preset threshold value or reaches a preset updating period, the stored data is utilized to carry out fine adjustment on the initial decoding model, wherein the learning rate during fine adjustment is lower than the training learning rate of the offline personalized module.
10. The personalized brain-computer interface system according to claim 9, wherein the fine tuning by the incremental updating unit is limited to updating only the classification layer parameters of the personalized initial decoding model or updating network layer parameters associated with high weights of personalized configuration parameters.

Description

Personalized brain-computer interface system Technical Field The invention belongs to the technical field of brain-computer interfaces, and particularly relates to a personalized brain-computer interface system. Background The Brain-computer interface (BCI, brain-Computer Interface) system realizes information interaction between a user and external equipment by decoding Brain activity signals. The BCI based on the brain waves has great application potential in the fields of medical rehabilitation, nerve engineering, man-machine interaction and the like due to the advantages of noninvasive property, portability, relatively low cost and the like. However, current BCI technology based on electroencephalogram still faces many challenges in practical applications. Firstly, because of the huge difference of individual brain electricity, different users have obvious difference on brain electricity response modes of the same experimental paradigm, and brain electricity signals of the same user in different times and different physiological states have non-stationarity. This large inter-individual variability results in low accuracy in decoding new users with traditional fixed-pattern, generic-model BCI systems, and poor user experience. Secondly, most of existing BCI systems adopt a "one-cut" strategy, i.e. the same experimental paradigm and decoding model is provided for all users. While some studies have attempted model fine-tuning through calibration data, there is a lack of a full-flow personalization scheme from paradigm selection to model decoding. In addition, the traditional method requires a user to perform long-time boring calibration training to obtain an available decoding model, and is heavy in burden, easy to generate fatigue and unfavorable for popularization and use of the system. Finally, most online decoding systems are used after offline training, and cannot adapt to changes caused by natural drift of electroencephalogram signals along with time or learning effects of users. Therefore, a new BCI system is needed that can realize deep personalization from source paradigm selection to terminal model parameter, and can efficiently initialize and stabilize online self-adaptation. Disclosure of Invention The invention mainly aims to overcome the defects and shortcomings of the prior art and provide a personalized brain-computer interface system, which realizes full-flow personalization from a normal form to a decoding model by actively selecting the normal form by a user, intelligently matching the decoding model by the system and performing online self-adaption by incremental learning, and effectively considers the user experience, personalized depth and long-term decoding performance of a BCI system. In order to achieve the above purpose, the present invention adopts the following technical scheme: in a first aspect, the present invention provides a personalized brain-computer interface system comprising: the electroencephalogram data engineering module is used for collecting electroencephalogram signals of a user and preprocessing the electroencephalogram signals; The individuation Fan Shiku module is used for storing pre-verified brain-computer interface experimental norms, the norms adopt relational data to store norms metadata, and each norms comprises a norms ID, a norms type, a norms description, pre-verification indexes and recommendation indexes; the personalized paradigm selection module is used for displaying the brain-computer interface experimental paradigm and selecting the brain-computer interface experimental paradigm according to the quantization index and combining with user feedback; The model matching and selecting module is used for determining at least one candidate decoding model applicable to the target normal form according to a preset normal form-model corresponding relation, and determining an initial decoding model from the candidate decoding models based on calibration electroencephalogram data of a user; The off-line individuation module is used for setting initial individuation configuration parameters, training the initial decoding model by using reinforcement learning, and obtaining an individuation initial decoding model; And the online decoding and self-adapting module is used for decoding the real-time electroencephalogram signals of the user by utilizing the personalized initial decoding model, and continuously optimizing the personalized initial decoding model in an incremental learning mode based on the electroencephalogram data and decoding results newly generated in the online use process of the user. As an optimal technical scheme, the electroencephalogram data engineering module acquires a multichannel original electroencephalogram signal sequence at a set time, and performs preprocessing on the multichannel original electroencephalogram signal sequence, wherein the preprocessing comprises band-pass filtering, notch filtering, artifact removal, baseline cor