WO2021237917A1

WO2021237917A1 - Self-adaptive cognitive activity recognition method and apparatus, and storage medium

Info

Publication number: WO2021237917A1
Application number: PCT/CN2020/104558
Authority: WO
Inventors: 王洪涛; 唐聪; 许林峰; 岳洪伟
Original assignee: 五邑大学
Priority date: 2020-05-25
Filing date: 2020-07-24
Publication date: 2021-12-02
Also published as: CN111543988A; CN111543988B

Abstract

Disclosed are a self-adaptive cognitive activity recognition method and apparatus, and a storage medium. The method comprises: collecting original electroencephalogram data; processing the original electroencephalogram data to obtain an electroencephalogram signal; respectively inputting the electroencephalogram signal into a state transition model and a reward model, so as to respectively obtain electroencephalogram state information and reward information; inputting the electroencephalogram state information and the reward information into a reinforced selective attention model to obtain optimal attention area information; and inputting the optimal attention area information into the reward model to obtain a classification recognition result. A cognitive activity can be recognized more effectively, and the recognition accuracy is also improved.

Description

Self-adaptive cognitive activity recognition method, device and storage medium

Technical field

The present invention relates to the field of artificial intelligence, in particular to an adaptive cognitive activity recognition method, device and storage medium.

Background technique

Electroencephalogram (EEG) is an electrophysiological monitoring index that can analyze the state and activity of the brain by measuring the voltage fluctuations of the ion current in the brain neurons. In practice, EEG signals can be collected in a non-invasive and non-fixed manner with portable and off-the-shelf equipment. The EEG signal classification algorithm has been studied for a series of practical applications. The accuracy and robustness of the EEG classification model is an important measure of cognitive activities such as movement intention recognition and emotion recognition. The cognitive activity recognition system builds a bridge between the internal cognitive world and the external physical world. They are recently used in assisted living, smart home and entertainment industries; motor image recognition based on EEG can help people with disabilities perform basic activities necessary for life; Emotion recognition based on EEG signals can be used to detect the emotional state of current patients, such as Depression, anxiety, etc.

The classification of cognitive activities faces several challenges. First, most of the existing EEG classification research uses EEG data preprocessing and feature extraction methods (for example, band-pass filtering, discrete wavelet transform, and feature selection) that are time-consuming and highly dependent on professional knowledge. Secondly, most of the current EEG classification methods are designed based on the knowledge of a specific field, so they may fail or even fail in different situations.

Summary of the invention

The present invention aims to solve at least one of the technical problems existing in the prior art.

For this reason, the present invention proposes an adaptive cognitive activity recognition method, which can recognize cognitive activities more effectively and improve the accuracy of recognition.

The present invention also provides an adaptive cognitive activity recognition device applying the above-mentioned adaptive cognitive activity recognition method.

The present invention also provides a computer-readable storage medium applying the above-mentioned adaptive cognitive activity recognition method.

The adaptive cognitive activity recognition method according to the embodiment of the first aspect of the present invention includes:

Collect raw EEG data;

Processing the original EEG data to obtain EEG signals;

Inputting the EEG signal into the state transition model and the reward model, respectively, to obtain EEG state information and reward information;

Inputting the EEG state information and the reward information into the enhanced selective attention model to obtain the best attention area information;

The best attention area information is input into the reward model to obtain the classification recognition result.

The adaptive cognitive activity recognition method according to the embodiment of the present invention has at least the following beneficial effects: a general framework for automatic cognitive activity recognition is proposed to promote the scope of various cognitive application fields, including motor image recognition and emotion recognition . Design a reinforced selective attention model by combining deep reinforcement learning and attention mechanism to automatically extract robust and unique deep features; to encourage the model to select the best attention area that can achieve the highest classification accuracy; in addition, it is customized according to the cognitive activity recognition environment State and action; The reward model is also used to classify the selected original EEG data, which achieves higher recognition accuracy than traditional methods and lower delay.

According to some embodiments of the present invention, the processing the original EEG data to obtain EEG signals includes:

Copying and shuffling the original EEG data to obtain combined EEG data;

The combined EEG data is selected to obtain EEG signals.

According to some embodiments of the present invention, the inputting the EEG state information and the reward information into the enhanced selection attention model to obtain the best attention area information includes:

The EEG state information and the reward information are received through the enhanced selective attention model to obtain EEG evaluation information;

The EEG evaluation information is fed back to the state transition model to drive the state transition model to perform EEG state information conversion until the enhanced selection attention model obtains the best attention area information.

According to some embodiments of the present invention, the reward model includes a convolutional mapping network and a classifier.

According to some embodiments of the present invention, the inputting the best attention area information into the reward model to obtain the classification recognition result includes:

The best attention area information is input to the convolutional mapping network to obtain the spatial dependence feature;

The spatially dependent features are input to the classifier to obtain classification recognition results.

According to some embodiments of the present invention, the convolutional mapping network includes an input layer, a convolutional layer, a fully connected layer, an extraction feature layer, and an output layer. The input layer, the convolutional layer, the fully connected layer, The extracted feature layer and the output layer are sequentially connected.

An adaptive cognitive activity recognition device according to an embodiment of the second aspect of the present invention includes:

The collection unit is used to collect the original EEG data;

The processing unit is used to process the original EEG data to obtain EEG signals;

The detection unit is configured to input the EEG signal into the state transition model and the reward model respectively to obtain EEG state information and reward information;

A screening unit, configured to input the EEG state information and the reward information into the enhanced selective attention model to obtain the best attention area information;

The recognition unit is used to input the best attention area information into the reward model to obtain a classification recognition result.

According to some embodiments of the present invention, the processing unit includes:

The copying unit is used for copying the original EEG data;

The shuffling unit is used for shuffling the original EEG data processed by the copying unit to obtain combined EEG data;

The selecting unit is used to select the combined EEG data to obtain EEG signals.

According to some embodiments of the present invention, the detection unit includes:

The state transition unit is used to input the EEG signal into the state transition model to obtain EEG state information;

The reward unit is used to input brain electrical signals into the reward model to obtain reward information.

The adaptive cognitive activity recognition device according to the embodiment of the present invention has at least the following beneficial effects: a reinforced selective attention model is designed by combining deep reinforcement learning and an attention mechanism to automatically extract robust and unique deep features; to encourage model selection The best attention area that can achieve the highest classification accuracy; in addition, the state and action are customized according to the cognitive activity recognition environment; the reward model is also used to classify the selected original EEG data, which achieves higher recognition accuracy than traditional methods , And the latency is low.

According to the computer-readable storage medium of the embodiment of the third aspect of the present invention, the adaptive cognitive activity recognition method according to the embodiment of the first aspect of the present invention can be applied.

The computer-readable storage medium according to the embodiment of the present invention has at least the following beneficial effects: a reinforced selective attention model is designed by combining deep reinforcement learning and an attention mechanism to automatically extract robust and unique deep features; to encourage model selection to achieve The best attention area with the highest classification accuracy; in addition, the state and action are customized according to the recognition environment of cognitive activities; the reward model is also used to classify the selected original EEG data, which achieves higher recognition accuracy than traditional methods, and The latency is low.

The additional aspects and advantages of the present invention will be partially given in the following description, and some will become obvious from the following description, or be understood through the practice of the present invention.

Description of the drawings

The above and/or additional aspects and advantages of the present invention will become obvious and easy to understand from the description of the embodiments in conjunction with the following drawings, in which:

FIG. 1 is a flowchart of an adaptive cognitive activity recognition method according to Embodiment 1 of the present invention;

2 is a working flow chart of processing raw EEG data in the adaptive cognitive activity recognition method according to the first embodiment of the present invention;

FIG. 3 is a working flow chart of screening the best attention area in the adaptive cognitive activity recognition method according to the first embodiment of the present invention; FIG.

FIG. 4 is a work flow chart of obtaining classification and recognition results in the adaptive cognitive activity recognition method according to the first embodiment of the present invention;

5 is a schematic diagram of the device structure of an adaptive cognitive activity recognition device according to the second embodiment of the present invention;

Fig. 6 is a detailed flowchart of the method for adaptive cognitive activity recognition according to the first embodiment of the present invention.

Detailed ways

The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings, in which the same or similar reference numerals indicate the same or similar elements or elements with the same or similar functions. The embodiments described below with reference to the drawings are exemplary, and are only used to explain the present invention, but should not be construed as limiting the present invention.

In the description of the present invention, unless otherwise clearly defined, words such as setting and connection should be understood in a broad sense, and those skilled in the art can reasonably determine the specific meaning of the above words in the present invention in combination with the specific content of the technical solution.

Example one

1, Embodiment 1 of the present invention provides an adaptive cognitive activity recognition method, one of which includes but is not limited to the following steps:

Step S100, collecting raw brain electricity data.

In this embodiment, the original EEG data is collected in this step to prepare for subsequent adaptive cognitive activity recognition and provide a basis for collecting data.

Step S200, processing the original EEG data to obtain EEG signals.

In this embodiment, this step organizes and processes the collected raw EEG data, and further prepares for adaptive cognitive activity recognition.

In step S300, the EEG signals are input into the state transition model and the reward model respectively, and the EEG state information and reward information are obtained respectively.

In this embodiment, the state transition model in this step is used to select a certain behavior in the EEG signal to obtain EEG state information. At the same time, the reward model also evaluates the quality of the behavior by rewarding the behavior. Reward information, so as to prepare the prerequisites for selecting the best attention area information.

Step S400, input the EEG state information and the reward information into the enhanced selective attention model to obtain the best attention area information.

In this embodiment, the enhanced selected attention model in this step can evaluate EEG state information and reward information. When it is found that it is not the best attention area information, it will be fed back to the state transition model, making the state transition model Reselect another attention area in the EEG signal and re-evaluate until the best attention area information is obtained.

Step S500: Input the best attention area information into the reward model to obtain a classification and recognition result.

In this embodiment, the best attention area information obtained in this step is input into the reward model, and the classification and recognition result of the original EEG data is obtained through the reward model.

Referring to FIG. 2, step S200 in this embodiment may include but is not limited to the following steps:

In step S210, the original EEG data is copied and shuffled to obtain combined EEG data.

In this embodiment, in this step, in order to provide as much information as possible, through copying and shuffling processing, more potential space combinations of feature dimensions can be provided, so as to prepare for subsequent detection.

In step S220, the combined EEG data is selected to obtain EEG signals.

In this embodiment, this step selects the original EEG data that has been copied and shuffled to obtain the EEG signal, and then input it into the state transition model and the reward model to select the best attention area in the original EEG data .

Referring to FIG. 3, in step S400 of this embodiment, the following steps may be included but not limited to:

In step S410, the EEG state information and the reward information are received through the enhanced selective attention model to obtain EEG evaluation information.

In this embodiment, the enhanced selective attention model in this step receives the EEG state information and reward information, and then comprehensively evaluates the EEG state information and reward information. The reward information can reveal a certain selected behavior of the input EEG signal The enhanced selection attention model can give feedback to the state transition model. When the selected area is not the best attention area, it will drive the state transition model to reselect another attention area.

In step S420, the EEG evaluation information is fed back to the state transition model to drive the state transition model to perform EEG state information conversion until the enhanced selection attention model obtains the best attention area information.

In this embodiment, the enhanced selection attention model in this step can feed back to the state transition model. When the selected area is not the best attention area, the state transition model will be driven to reselect another attention area until the selected attention area is strengthened. The model selects the best attention area information.

Referring to FIG. 4, in step S500 of this embodiment, the following steps may be included but not limited to:

In step S510, the best attention area information is input to the convolutional mapping network to obtain the spatial dependence feature.

In this embodiment, the best attention region information obtained in this step is input into the convolutional mapping network, and the convolutional mapping network is used to extract spatial dependent features.

In step S520, the spatially dependent features are input to the classifier to obtain a classification recognition result.

In this embodiment, this step inputs the spatially dependent features obtained above into the classifier, so that the classifier can be used for adaptive cognitive activity recognition.

In this embodiment, the reward model includes a convolutional mapping network and a classifier. The convolutional mapping network can perform spatially dependent feature extraction; the classifier can perform cognitive activity recognition based on the spatially dependent feature extracted by the convolutional mapping network.

Further, in this embodiment, the convolutional mapping network includes an input layer, a convolutional layer, a fully connected layer, an extraction feature layer, and an output layer. The input layer, the convolutional layer, the fully connected layer, The extracted feature layer and the output layer are sequentially connected.

In this embodiment, the enhanced selective attention model includes a competitive DQN network, a fully connected layer, a value function V, an advantage function A and Q function, and a competitive DQN network and a fully connected layer are connected. The value function V and the advantage function A are connected with the completely The connection layer is connected, and the value function V and the advantage function A are also connected with the Q function to realize a strengthened selection attention mechanism.

The following uses a specific embodiment to further illustrate the adaptive cognitive activity recognition method:

Referring to Fig. 6, in order to provide as much information as possible, a method using the spatial relationship between EEG signals is designed. The signals belonging to different brain activities should have different spatial dependencies. Copy and shuffle the input EEG signal according to the dimensions. In this method, all possible dimensional arrangements have the appearance of equal probability.

Suppose that the input original EEG data is represented by X={(x _i , y _i ), i=1, 2,..., I}, where (x _i , y _i ) represents a single EEG sample, and I represents the number of samples. In each sample, feature x _i ={x _ik ,k=1, 2,...,N}, x _i ∈R ^N contains N elements corresponding to N EEG channels, and y _i ∈R Indicates the corresponding label. x _ik represents the k-th dimension value in the i-th sample.

In real-world collection scenarios, EEG data is usually connected according to the distribution of biomedical EEG channels. However, the order of biomedical dimensions may not show the best spatial dependence. The exhaust method is computationally too expensive, which makes it impossible to exhaust all possible dimensional arrangements.

In order to provide more potential combinations of dimensions, a method called "copy and shuffle (RS)" is proposed. RS is a two-step mapping method that can map x _i to a higher-dimensional space x′ _i with a complete element combination:

x _i ∈R ^N →x′ _i ∈R ^N′ ,N'＞N

In the first step (copy), x _{i is} copied h=N'/N+1 times. Then, get a new vector whose length is h*N, not less than N'; in the second step (shuffle), randomly shuffle the copied vector in the first step, and intercept the first N 'generates x' element _i. In theory, compared with x _i , _x'i contains more different combinations of dimensions. Note that this RS operation is only performed once on a specific input data set in order to provide a stable environment for subsequent reinforcement learning.

Inspired by the fact that the best spatial relationship only depends on the feature dimension subset, a focus area is introduced to focus on the feature dimension segment. Here, the attention area is optimized through deep reinforcement learning, and it turns out that this area is stable and performs well in strategy learning.

In particular, it aims to detect the best combination of dimensions, which includes the most significant spatial dependence between EEG signals. Since N '(x' _i of length) and a huge amount of calculation is too large, the length and information content can not be balanced, so the introduction of the attention mechanism, since its effectiveness has been demonstrated in a recent study in the art (e.g. voice recognition). Try to emphasize the information fragments in _{x′ i, and use}

Indicates the segment, which is called the attention zone. make

with

Indicates the length of the attention area, which is automatically learned by the proposed algorithm. Use deep reinforcement learning to find the best attention area.

The detection of the best attention area includes two key components: environment (including state transition and reward model) and enhanced selective attention mechanism. Three elements (state s, behavior a, and reward r) are exchanged in the interaction between the environment and the subject. These three elements are all customized according to the background of this research. Next, explain the design of the key components of the deep reinforcement learning structure:

State S={s _t ,t=0,1,...,T}, _st ∈R ² describes the location of the region of interest, where t represents time. Since the area of interest is _{a moving segment on 1-D x′ i} , we design two parameters to define the state:

in

with

Indicates the start index and end index of the attention area respectively. In training, initialize _{s 0 as}

Behavior _{A = {a t, t =} 0,1, ..., T} ∈R 4 describes enhanced attention selection mechanism can select the behavior of the environment taken. At the time mark t, the state transition selects an operation to be implemented according to the strategy π that strengthens the selective attention mechanism:

s _t+1 =π(s _t ,a _t )

Four types of actions are defined for the attention area: left (imagine left hand), right (imagine right hand), up (imagine tongue) and down (imagine feet). For each action, the attention area will move a random distance d∈[1,d ^u ], where d ^u is the upper limit. For moving left and right, note that the area will move left or right with step d. For up and down operations,

with

Both are mobile d. Finally, if the state start index or end index exceeds the boundary, a crop operation is performed. For example, if

(Below the lower boundary 0), we crop the starting index to

The reward R={r _t ,t=0,1,...,T}∈R is calculated by the reward model. Reward model Φ:

r _t ＝Φ(s _t )

Receive the current status and return the evaluation as a reward.

The purpose of the reward model is to evaluate how the current state affects classification performance. Intuitively, the state leading to better classification performance should have a higher return: r _t =F(s _t ). We set the reward model F as a combination of convolutional mapping and classification. Because in the actual method optimization, the higher the accuracy, the more difficult it is to increase the classification accuracy. To encourage higher levels of accuracy, we designed a non-linear reward function:

Where acc means classification accuracy. This function consists of two parts; the first part is a normalized exponential function with exponent acc∈[0,1]. This part encourages the reinforcement learning algorithm to search for a better _st to obtain a higher acc. The motivation of the exponential function is that the growth rate of the reward increases as the accuracy increases. The second part is to pay attention to the length of the zone to keep the penalty factor shorter, and β is the penalty factor.

All in all, the purpose of deep reinforcement learning is to learn the best attention area that leads to the greatest reward

The selection mechanism iterates M=n _e *n _s times in total, where n _e and n _s represent the plot and the number of steps, respectively. In the state transition, the ε-greedy method is adopted, which selects random actions with a probability of 1-ε or the optimal Q function with a probability of ε

Choose an action behavior.

Where ε'∈[0,1] is randomly generated for each iteration, and

It was randomly selected in A.

In order to better converge and train faster, ε gradually increases with iterations. The increment ε _{0 is} as follows:

ε _t+1 =ε _t +ε ₀ M

Competitive DQN (deep Q network) is used as an optimization strategy π(s _t , a _t ), which can effectively learn the state value function. The main reason why we use the duel DQN to find the best area of interest is that it updates all four Q values at each step, while other strategies only update one Q value at each step. The Q function measures the expected sum of future rewards when taking the line and following the best strategy. Especially for a specific step t, we have:

Where γ∈[0,1] is a decay parameter that weighs the importance of immediate and future rewards, and n represents the number of subsequent steps. When in state s, the value function V( _st ) estimates the expected reward. The Q function is related to the pair (s _t , a _t ), and the value function is only related to s _t .

Competitive DQN learns the Q function through the value function V(s _t ) and the advantage function A(s _t , a _t ), and combines them with the following formula

Q(s _t ,a _t )=θV(s _t )+θ'A(s _t ,a _t )

Among them θ, θ'∈ Θ are the parameters in the confrontation DQN network, and will be automatically optimized. The above formula is not recognized, in fact, can not be recovered uniquely V (s _t) and A (s _{_t,} a _t) given _{_{Q (s t, a t)}} . To solve this problem, the advantage function can be forced to be equal to zero on the selected action. In other words, let the network implement forward mapping:

Therefore, for a specific action a ^* , if

Then there is

Q(s _t+1 ,a*)=V(s _t )

Therefore, the value function V( _st ) is forced to learn the estimation of the value function, while the other direction produces the estimation of the advantage function.

In order to evaluate the Q function, we optimized the following cost functions in the i-th iteration:

in,

The gradient update method is

For each attention area, further mining selected features

The potential spatial dependence of. Since only a single sample is concerned, EEG samples only contain numerical vectors with very limited information, and are easily corrupted by noise. In order to make up for this shortcoming, try to extract a single sample of EEG from the original space through the CNN structure

Map to the sparse space Γ∈R ^M.

In order to extract as many potential spatial dependencies as possible, a convolutional layer is used with many filters to learn the attention area

Perform a scan. The convolutional mapping structure consists of five layers: the input layer receives the learned attention area, the convolutional layer is followed by a fully connected layer, and the output layer. Compare the single hot spot live with the output layer to calculate the training loss.

The Relu nonlinear activation function is applied to the convolution output. The convolutional layer is described as follows:

in

Represents the result of the convolutional layer, and

And W _c respectively represent the length of the filter and the weight of the filter. The pooling layer aims to reduce the redundant information in the convolution output to reduce the computational cost. As far as we are concerned, try to retain as much information as possible. Therefore, the method does not use a pooling layer. Then, in the fully connected layer and the output layer

Among them, W ^f , W ^o , b ^f , and b ^o respectively represent the corresponding weights and deviations. y'represents the predicted label. The cost function is measured by cross entropy, and the l ₂ -norm (with parameter λ) is used as regularization to prevent overfitting:

The AdamOptimizer algorithm optimizes the cost function. The fully connected layer extracts features and inputs them into the lightweight nearest neighbor classifier. The convolutional map is updated iteratively for N'times.

Copy and scramble the input original EEG single sample to provide more potential spatial combinations of feature dimensions. Then, select an attention zone. The selected area of interest is input to the state transition and reward model. In each step t, the state transition will select a behavior to update according to the feedback of the enhanced selective mechanism. The reward model evaluates the quality of the area of interest through reward scores. The competitive DQN is used to find the best attention area, which will be fed into the convolutional mapping process to extract the spatial dependent representation. The features represented will be used for classification. The reward model is a combination of convolutional mapping and classifier.

The framework can directly process raw EEG data without feature extraction. In addition, it can automatically select distinguishable feature dimensions for different EEG data, thereby achieving high availability. The method not only surpasses several latest baselines to a large extent, but also shows low latency and high flexibility in dealing with multiple EEG signal channels and incomplete EEG signals. The method is suitable for a wider range of application scenarios, such as motor image recognition. And emotional state recognition.

It can be seen from the above scheme that the enhanced selective attention model is designed by combining deep reinforcement learning and attention mechanism to automatically extract robust and unique deep features; to encourage the model to select the best attention area that can achieve the highest classification accuracy; in addition, according to the recognition Knowing activities recognize the customized state and actions of the environment; it also uses the reward model to classify the selected original EEG data, which achieves higher recognition accuracy than traditional methods and lower delay.

Example two

5, the second embodiment of the present invention provides an adaptive cognitive activity recognition device 1000, including:

The collecting unit 1100 is used to collect raw brain electricity data;

The processing unit 1200 is configured to process the original EEG data to obtain EEG signals;

The detection unit 1300 is configured to input the EEG signal into the state transition model and the reward model, respectively, to obtain EEG state information and reward information respectively;

The screening unit 1400 is configured to input the EEG state information and the reward information into the enhanced selection attention model to obtain the best attention area information;

The recognition unit 1500 is configured to input the best attention area information into the reward model to obtain a classification recognition result.

In this embodiment, the processing unit 1200 includes:

The copying unit 1210 is used for copying the original EEG data;

The shuffling unit 1220 is used for shuffling the original EEG data processed by the copying unit 1210 to obtain combined EEG data;

The selecting unit 1230 is used to select the combined EEG data to obtain EEG signals.

In this embodiment, the detection unit 1300 includes:

The state transition unit 1310 is used to input brain electrical signals into the state transition model to obtain brain electrical state information; the reward unit 1320 is used to input brain electrical signals into the reward model to obtain reward information.

In this embodiment, the screening unit 1400 includes:

The selected attention unit 1410, which is to strengthen the selected attention model, can receive the EEG state information and the reward information to obtain EEG evaluation information; and then feed back the EEG evaluation information to the state transition model to The state transition model is driven to perform EEG state information conversion until the enhanced selection attention model obtains the best attention area information.

In this embodiment, the recognition unit 1500 is the above-mentioned reward unit 1320, and can perform adaptive cognitive activity recognition.

It should be noted that, since the adaptive cognitive activity recognition device in this embodiment and the adaptive cognitive activity recognition method in the first embodiment are based on the same inventive concept, the corresponding content in the first method embodiment is also applicable to This system embodiment will not be described in detail here.

Example three

The third embodiment of the present invention also provides a computer-readable storage medium that stores executable instructions of the adaptive cognitive activity recognition device, and the executable instructions of the adaptive cognitive activity recognition device are used to enable self The adaptive cognitive activity recognition device executes the above-mentioned adaptive cognitive activity recognition method, for example, executes the method steps S100 to S500 in FIG. 1 described above to realize the functions of the units 1000-1500 in FIG. 5.

In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "exemplary embodiments", "examples", "specific examples", or "some examples" etc. means to incorporate the implementation The specific features, structures, materials, or characteristics described by the examples or examples are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above-mentioned terms do not necessarily refer to the same embodiment or example. Moreover, the described specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and modifications can be made to these embodiments without departing from the principle and purpose of the present invention. The scope of the present invention is defined by the claims and their equivalents.

Claims

An adaptive cognitive activity recognition method, which is characterized in that it includes:

Collect raw EEG data;

Processing the original EEG data to obtain EEG signals;

Inputting the EEG signal into the state transition model and the reward model, respectively, to obtain EEG state information and reward information;

Inputting the EEG state information and the reward information into the enhanced selective attention model to obtain the best attention area information;

The best attention area information is input into the reward model to obtain the classification recognition result.
An adaptive cognitive activity recognition method according to claim 1, wherein the processing the original EEG data to obtain EEG signals comprises:

Copying and shuffling the original EEG data to obtain combined EEG data;

The combined EEG data is selected to obtain EEG signals.
An adaptive cognitive activity recognition method according to claim 1, wherein said inputting said EEG state information and said reward information into an enhanced selective attention model to obtain best attention area information, include:

The EEG state information and the reward information are received through the enhanced selective attention model to obtain EEG evaluation information;

The EEG evaluation information is fed back to the state transition model to drive the state transition model to perform EEG state information conversion until the enhanced selection attention model obtains the best attention area information.
An adaptive cognitive activity recognition method according to claim 1, wherein the reward model includes a convolutional mapping network and a classifier.
An adaptive cognitive activity recognition method according to claim 4, wherein said inputting said best attention area information into a reward model to obtain a classification recognition result comprises:

The best attention area information is input to the convolutional mapping network to obtain the spatial dependence feature;

The spatially dependent features are input to the classifier to obtain classification recognition results.
An adaptive cognitive activity recognition method according to claim 4, wherein the convolutional mapping network includes an input layer, a convolutional layer, a fully connected layer, an extraction feature layer, and an output layer, and the input layer , The convolutional layer, the fully connected layer, the extracted feature layer and the output layer are connected in sequence.
An adaptive cognitive activity recognition device, which is characterized in that it comprises:

The collection unit is used to collect the original EEG data;

The processing unit is used to process the original EEG data to obtain EEG signals;

The detection unit is configured to input the EEG signal into the state transition model and the reward model respectively to obtain EEG state information and reward information;

A screening unit, configured to input the EEG state information and the reward information into the enhanced selective attention model to obtain the best attention area information;

The recognition unit is used to input the best attention area information into the reward model to obtain a classification recognition result.
An adaptive cognitive activity recognition device according to claim 7, wherein the processing unit comprises:

The copying unit is used for copying the original EEG data;

The shuffling unit is used for shuffling the original EEG data processed by the copying unit to obtain combined EEG data;

The selecting unit is used to select the combined EEG data to obtain EEG signals.
An adaptive cognitive activity recognition device according to claim 7, wherein the detection unit comprises:

The state transition unit is used to input the EEG signal into the state transition model to obtain EEG state information;

The reward unit is used to input brain electrical signals into the reward model to obtain reward information.
A computer-readable storage medium is characterized in that: the computer-readable storage medium stores executable instructions of an adaptive cognitive activity recognition device, and the executable instructions of the adaptive cognitive activity recognition device are used to enable the adaptive cognitive activity The recognition device executes the adaptive cognitive activity recognition method according to any one of claims 1 to 6.