Electroencephalogram data processing method and device, electronic equipment and storage medium

Abstract

The invention discloses an electroencephalogram data processing method and device, electronic equipment and a storage medium. The method comprises the following steps: preprocessing an original electroencephalogram signal; performing micro-state division on the preprocessed electroencephalogram signals on the five frequency bands respectively; the five frequency bands comprise a delta frequency band, a theta frequency band, an alpha frequency band, a beta frequency band and a gamma frequency band; the micro-states comprise four types of micro-states; performing feature extraction on the micro-state of the electroencephalogram signal under each frequency band; the features comprise a time feature, an occurrence frequency feature and an occupancy feature of each micro-state category, and a transition probability feature of each micro-state; and performing feature screening by using an SVM-RFE model to obtain target features. According to the technical scheme provided by the invention, the airspace information of the electroencephalogram signals in different frequency ranges is considered, the feature classification is more comprehensive, the SVM model and the RFE algorithm are combined, the performance of the model is improved, the calculation time is shortened, the calculation cost is reduced, and the electroencephalogram data recognition effect is improved on the whole.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence research, and more specifically, to a method, device, electronic device and storage medium for processing electroencephalogram data. Background Art

[0002] Attention deficit hyperactivity disorder (ADHD) is one of the most common neurodevelopmental disorders, mainly characterized by inattention, hyperactivity and impulsive behavior that is not appropriate for their age, accompanied by multiple cognitive defects such as learning disabilities, conduct disorders, and emotional disorders. Domestic and foreign studies have shown that the incidence of ADHD is increasing year by year, so timely and accurate identification of ADHD is of great practical significance.

[0003] With the continuous development of artificial intelligence algorithms, machine learning has been applied to the identification of ADHD. In existing studies, machine learning has been used to classify ADHD from functional magnetic resonance imaging (fMRI), electroencephalogram (EEG), magnetic resonance imaging (MRI), and skin galvanic response. Among them, EEG has received a lot of attention due to its low cost and easy availability. However, there are still some shortcomings in the existing EEG signal recognition methods. Currently, in the EEG signal recognition of ADHD, although some studies have used the frontal lobe theta wave energy as a biological feature, its effectiveness and accuracy have been questioned, and especially considering that ADHD is a neurodevelopmental disorder, existing studies are mostly limited to frequency domain or time domain information, and cannot effectively obtain spatial domain information, resulting in poor EEG signal recognition effect for ADHD.

[0004] Therefore, how to develop an EEG data processing method to improve the EEG signal recognition effect of ADHD has become a problem that needs to be solved in this field. Summary of the invention

[0005] In view of this, in a first aspect, the present application proposes an EEG data processing method, comprising:

[0006] Preprocess the original EEG signal;

[0007] The preprocessed EEG signal is divided into microstates in five frequency bands respectively; the five frequency bands include delta frequency band, theta frequency band, alpha frequency band, beta frequency band, and gamma frequency band; the microstates include four types of microstates;

[0008] Extracting features of the microstates of the EEG signals in each frequency band; the features include: duration features, occurrence frequency features, occupancy features, and transition probability features of each microstate category;

[0009] The SVM-RFE model is used to screen the extracted features and obtain the target features.

[0010] Preferably, the preprocessing includes one or more steps of: removing irrelevant electrodes, filtering, whole-brain average re-referencing, baseline correction, and bad segment removal.

[0011] Further preferably, the filtering includes: one or more steps of removing power frequency noise and bandpass filtering.

[0012] Preferably, the preprocessed EEG signal is divided into microstates in five frequency bands, including:

[0013] The preprocessed EEG signals are projected into a low-dimensional space using independent component analysis in the five frequency bands, and the k-means clustering algorithm is applied to the low-dimensional space to divide the microstates.

[0014] Preferably, before using the SVM-RFE model for feature screening, the method further comprises:

[0015] Construct a feature matrix for each frequency band and each microstate feature;

[0016] A normalization operation is performed on the feature matrix.

[0017] Further preferably, the normalization operation is a minimum-maximum normalization operation.

[0018] Preferably, the SVM-RFE model is used for feature screening to obtain target features, including:

[0019] Construct SVM-RFE model;

[0020] Using the RFE algorithm to screen features in the SVM-RFE model and iteratively training the SVM-RFE model according to the screening results until the classification accuracy of the model no longer increases or reaches a preset accuracy;

[0021] The target features are obtained using the final SVM-RFE model.

[0022] Further preferably, training the SVM-RFE model comprises:

[0023] The SVM-RFE model is trained using a network search method.

[0024] In a second aspect, an embodiment of the present invention further provides an electroencephalogram data processing device, the device comprising:

[0025] A preprocessing module is configured to preprocess the original EEG signal;

[0026] The microstate division module is configured to divide the preprocessed EEG signal into microstates in five frequency bands respectively; the five frequency bands include delta frequency band, theta frequency band, alpha frequency band, beta frequency band, and gamma frequency band; the microstates include four types of microstates;

[0027] A feature extraction module is configured to extract features of the microstates of the EEG signals in each frequency band, wherein the features include: time features of each microstate category, occurrence frequency features, occupancy features, and transition probability features of each microstate;

[0028] The feature screening module is configured to screen the extracted features using the SVM-RFE model to obtain target features.

[0029] Preferably, the feature screening module is also used for:

[0030] Construct SVM-RFE model;

[0031] Using the RFE algorithm to screen features in the SVM-RFE model and iteratively training the SVM-RFE model according to the screening results until the classification accuracy of the model no longer increases or reaches a preset accuracy;

[0032] The target features are obtained using the final SVM-RFE model.

[0033] In a third aspect, an embodiment of the present disclosure further provides an electronic device, comprising: a processor; a memory for storing executable instructions of the processor; the processor is used to read the executable instructions from the memory and execute the instructions to implement the method disclosed in the first aspect above.

[0034] In a fourth aspect, an embodiment of the present disclosure further provides a computer-readable storage medium, wherein the storage medium stores a computer program, and the computer program is used to execute the method disclosed in the first aspect above.

[0035] The EEG data processing method provided in the present application first divides the preprocessed EEG signal into micro-states in five frequency bands, then extracts features of the micro-states of the EEG signal in each frequency band, and finally uses the SVM-RFE model for feature screening. In this process, not only the spatial domain information within different frequency ranges of the EEG signal is taken into account to make the feature classification more comprehensive, but also the SVM model and RFE algorithm are combined to improve the performance of the model, reduce calculation time and cost, and thus improve the overall EEG data recognition effect.

[0036] Other features and advantages of the present application will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings constituting a part of the present application are used to provide a further understanding of the present application, and the schematic implementation modes and descriptions of the present application are used to explain the present application. In the accompanying drawings:

[0038] Figure 1 A flowchart of an EEG data processing method for applying for a preferred embodiment;

[0039] Figure 2 A schematic diagram of the design of a notch filter for a preferred embodiment of the application;

[0040] Figure 3 A schematic diagram of an EEG bad segment for applying for a preferred embodiment;

[0041] Figure 4 A schematic diagram of the probability of each component after independent component analysis of a preferred embodiment of the application;

[0042] Figure 5 A detailed diagram of the electrooculographic component caused by eye drift in the preferred embodiment of the application;

[0043] Figure 6 A schematic diagram of an EEG data processing device according to a preferred embodiment of the application;

[0044] Figure 7 A schematic diagram of the structure of an electronic device according to a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0045] The technical solution of the present application will be described in detail below with reference to the accompanying drawings and in combination with the implementation modes.

[0046] First, this application proposes a method for processing EEG data, such as Figure 1 As shown, steps 110-140 are included:

[0047] Step 110, preprocessing the original EEG signal;

[0048] Specifically, electroencephalogram (EEG) signals are easily contaminated by noise, and EEG signals recorded directly from scalp electrodes cannot accurately represent brain neural signals. The raw EEG data needs to be preprocessed and denoised to minimize or eliminate the impact of artifacts. The processed pure EEG signals can make the model classification more accurate.

[0049] In the present application, preprocessing includes one or more steps of: removing irrelevant electrodes, filtering, whole-brain average re-referencing, baseline correction, and bad segment removal.

[0050] To eliminate irrelevant electrodes, electrodes that are not related to EEG signals, such as EOG electrodes, ECG electrodes, etc.

[0051] For filtering, filtering includes removing power frequency noise and / or bandpass filtering. EEG signals are easily interfered by the power frequency of the power grid. In order to eliminate this interference, a notch filter is used to remove the power frequency noise. The notch filter can remove narrowband signals of a specific frequency and protect the original signal from being affected. The use of a bandpass filter can achieve the passage of signals within a specific frequency band while attenuating frequency components outside the frequency band, thereby removing low-frequency baseline drift and high-frequency noise. The low-frequency band (<0.5Hz) is usually slow potential fluctuations and baseline drift, and does not contain useful EEG information. The high-frequency band (>50Hz) is mostly artifacts such as electromyography (EMG) noise and should be removed. For this application, the frequency range related to ADHD is concentrated between 0.5-50Hz. Designing a bandpass filter based on this frequency range can prevent useless frequency signals from interfering with subsequent model classification and recognition, thereby reducing the model burden and time cost.

[0052] In a specific embodiment, since the power frequency interference in my country is usually 50Hz, Figure 2 The design diagram of the notch filter shown in FIG. 1 shows that the frequency of the notch filter is set to 49-51 Hz, so that the notch filter can effectively filter out the 50 Hz power frequency interference. The bandpass filter uses a Butterworth filter, whose smooth frequency response is suitable for processing biological signals. The specific bandpass range is set to 0.5 Hz to 50 Hz to retain the main frequency components related to ADHD in the EEG signal.

[0053] For whole-brain average re-reference, the mean of all whole-brain data is used as reference data to re-calibrate the collected EEG signals for subsequent extraction of microstate features.

[0054] Regarding baseline correction, signal drift will inevitably occur during the EEG acquisition process, so it is necessary to perform baseline correction on the EEG signal to correct the signal offset.

[0055] In a specific embodiment, the EEG signal is segmented, and the mean value of the data of the first 0.2 seconds at the beginning of each segment is used as the baseline for recalibration.

[0056] Regarding bad segment elimination, during the EEG acquisition process, the subjects may experience body shaking, resulting in obvious artifacts in some segmented trials.

[0057] In a specific embodiment, the method of comparing the peak difference of EEG can realize the automatic removal of bad segments. The peak difference threshold of EEG is set to 140uv. If the EEG signal fluctuation exceeds 140uv, it is considered as a bad segment and is removed. Figure 3 As shown, this method can automatically remove the detected bad segments.

[0058] Step 120, dividing the preprocessed EEG signal into microstates in five frequency bands respectively;

[0059] Specifically, in the resting-state EEG signal, the EEG signal can be transformed from the time domain to the frequency domain through Fourier transform, and the frequency band range is set to separate the signals of different frequencies, so as to extract features in each frequency band for research. First, the EEG signal is separated into five main frequency bands: delta (0.5-4Hz), theta (4-8Hz), alpha (8-13Hz), beta (13-30Hz), and gamma (30-50Hz). Then, the preprocessed EEG signal is projected into a low-dimensional space using independent component analysis in the above five frequency bands, and the k-means clustering algorithm (k-means) is applied to the low-dimensional space for micro-state division.

[0060] The number of clustering categories is usually three to six, representing different microstates of EEG activity. This application adopts the classic four types of microstates A, B, C, and D, which can explain the differences in EEG signal recordings over time. For microstate A: This microstate is associated with negative blood oxygen level dependence (BOLD) activation in the bilateral superior and middle temporal gyri, which play a key role in speech processing; for microstate B: This microstate is associated with negative BOLD activation in the bilateral extrastriate visual cortex; for microstate C: This microstate is associated with positive BOLD activation in the posterior part of the anterior cingulate gyrus, the bilateral inferior frontal gyrus, the right anterior insula, and the left claustrum, which play an important role in executive control functions; for microstate D: This microstate is associated with attention redirection.

[0061] The Independent Component Analysis (ICA) method used in this application can remove artifact-related components, such as electrooculography, electromyography, channel noise, etc., and recombine other data, so that the reorganized data can remove the influence of artifacts and retain the real EEG signal. Figure 4 It shows the probability of each component after independent component analysis. Figure 5 The electrooculogram component caused by eye drift is shown in detail ( Figure 4 Detail of ingredient 5).

[0062] Step 130, extracting features of the microstates of the EEG signals in each frequency band;

[0063] Specifically, after completing the micro-state division, feature extraction is performed on the micro-state of each frequency band. The features to be extracted in this application include four types of features: duration, frequency, occupancy, and transition probability.

[0064] As for the duration feature, this feature can be understood as the average duration of each microstate category, reflecting the stability of the EEG activity, indicating the average duration of each microstate when it appears. This feature can help identify whether the microstate of a certain frequency band shows abnormal stability in ADHD patients. Its calculation formula is as follows:

[0065] Duration = total duration of a microstate / total number of times the microstate appears (1)

[0066] Regarding the frequency of occurrence feature, this feature can be understood as the average number of times each microstate category appears in one second, providing information on the activity intensity of different microstates in each frequency band. The frequency of occurrence of microstates reflects the degree of brain activity, and different microstate frequencies may be different between ADHD patients and normal people. Its calculation is shown in formula (2):

[0067] Occurrence frequency = number of microstate occurrences / total time period (2)

[0068] Regarding the occupancy rate feature, this feature can be understood as the total duration of each microstate category accounting for the total resting EEG time length, reflecting the existence degree of the microstate in the total recording time, and is used to measure the overall contribution of the microstate to brain activity. The calculation method is the sum of the time when the microstate appears divided by the total recording time of the EEG signal, and the calculation formula is as follows:

[0069] Occupancy rate = time spent in microstate / total recording time (3)

[0070] Regarding the transition probability feature, this feature is the probability of transitioning from one microstate to another, reflecting the dynamic switching mode between microstates and evaluating the flexibility and coordination of brain function. Its calculation formula is as follows:

[0071] P(i,j) = the number of transitions from microstate i to microstate j / the total number of transitions from microstate i (4)

[0072] Among them, the element P(i,j) refers to the probability of transitioning from microstate i to microstate j.

[0073] Step 140: Use the SVM-RFE model to perform feature screening on the extracted features.

[0074] Specifically, the Support Vector Machine (SVM)-Recursive Feature Elimination (RFE) model in the present application is a new model that integrates the RFE algorithm into the SVM model. By using this model to screen microstate features, features that are highly correlated with ADHD can be obtained.

[0075] Before screening the microstate features, it is necessary to first construct a feature matrix for each frequency band and each microstate feature, and then perform a minimum-maximum normalization operation on the feature matrix.

[0076] Further specifically, as mentioned above, the microstates in this application are four categories. When extracting features, 24 microstate features can be extracted in each frequency band, that is, 4 durations, 4 occurrence frequencies, 4 occupancies, and 12 transition probabilities. A total of 120 features can be obtained in 5 frequency bands. The characteristic parameters (duration, occurrence frequency, occupancy, transition probability) of each type of microstate in each frequency band are summarized to obtain a feature matrix. The dimension of the feature matrix is: number of samples × (number of frequency bands × number of features).

[0077] At the same time, since the dimensions and numerical ranges of different features may vary greatly, in order to ensure that all features have the same influence in model training, the feature matrix needs to be min-max normalized, that is, the features are scaled to the [0,1] interval to ensure that all features have the same scale.

[0078] Finally, the normalized feature matrix is ​​input into the SVM-RFE model, and the RFE algorithm is used to screen effective features and reduce the feature dimension. This can fully utilize the frequency domain information and the spatial information reflected by the microstate, and reduce the model complexity. The method for feature screening using the SVM-RFE model includes the following steps 210-230:

[0079] Step 210, constructing a SVM-RFE model;

[0080] Specifically, the SVM model is used as the framework basis, and the RFE algorithm is integrated during model evaluation to obtain the SVM-RFE model in this application.

[0081] Step 220, using the RFE algorithm to screen features in the SVM-RFE model and iteratively train the SVM-RFE model according to the screening results until the classification accuracy of the model no longer increases or reaches a preset accuracy;

[0082] Specifically, the SVM-RFE model needs to be continuously evaluated, and unimportant features are filtered out based on the evaluation results. After the screening, the model is trained again until the classification accuracy requirement is met. In each iteration, the RFE algorithm evaluates the importance of features based on their contribution to the model performance, then recursively deletes unimportant features and retrains the model with the remaining features. The above process is repeated until the classification accuracy of the model no longer increases or reaches the preset accuracy.

[0083] In the above training process, the model can be trained using the network search method, and the model can be tested for stability using the ten-fold cross validation method to make the model more stable.

[0084] Step 230, using the final SVM-RFE model to obtain the target features.

[0085] Specifically, when the classification accuracy of the model no longer increases or reaches the preset accuracy, the RFE algorithm is used to sort the remaining features by importance, select the top features, and use these top features and the optimal parameters determined by grid search to train the final SVM-RFE model. The target features obtained by the final SVM-RFE model are the features that contribute most to ADHD, that is, ADHD sensitive features.

[0086] The EEG data processing method provided in the present application first divides the preprocessed EEG signal into micro-states in five frequency bands, then extracts features of the micro-states of the EEG signal in each frequency band, and finally uses the SVM-RFE model for feature screening. In this process, not only the spatial domain information of the EEG signal in different frequency ranges is taken into account to make the feature classification more comprehensive, but also the SVM model and RFE algorithm are combined to improve the performance of the model, reduce the calculation time and cost, and thus improve the overall EEG data recognition effect.

[0087] In addition, the present application also provides an EEG data processing device for implementing the above-mentioned EEG data processing method, such as Figure 6 As shown, the device comprises:

[0088] The preprocessing module 610 is configured to preprocess the original EEG signal;

[0089] The microstate division module 620 is configured to divide the preprocessed EEG signal into microstates in five frequency bands respectively; the five frequency bands include delta frequency band, theta frequency band, alpha frequency band, beta frequency band, and gamma frequency band; the microstate includes four types of microstates;

[0090] The feature extraction module 630 is configured to extract features of the microstates of the EEG signals in each frequency band, the features including: time features of each microstate category, occurrence frequency features, occupancy features, and transition probability features of each microstate;

[0091] The feature screening module 640 is configured to screen the extracted features using the SVM-RFE model to obtain target features.

[0092] In some preferred embodiments, the feature screening module 640 is further used to:

[0093] Construct SVM-RFE model;

[0094] The RFE algorithm is used to screen the features in the SVM-RFE model and the SVM-RFE model is iteratively trained according to the screening results until the classification accuracy of the model no longer increases or reaches the preset accuracy;

[0095] The final SVM-RFE model is used to obtain the target features.

[0096] like Figure 7 As shown, the electronic device 11 includes one or more processors 111 and a memory 112 .

[0097] The processor 111 may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 11 to perform desired functions.

[0098] The memory 112 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, a random access memory (RAM) and / or a cache memory (cache), etc. The non-volatile memory may include, for example, a read-only memory (ROM), a hard disk, a flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 111 may run the program instructions to implement the test methods of the various embodiments of the present application described above and / or other desired functions. Various contents such as input signals, signal components, noise components, etc. may also be stored in the computer-readable storage medium.

[0099] In one example, the electronic device 11 may further include: an input device 113 and an output device 114 , and these components are interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0100] The input device 113 may include, for example, a keyboard, a mouse, etc.

[0101] The output device 114 can output various information to the outside, including determined distance information, direction information, etc. The output device 114 can include, for example, a display, a speaker, a printer, a communication network and a remote output device connected thereto, and the like.

[0102] Of course, to simplify, Figure 7 Only some of the components related to the present application in the electronic device 11 are shown, and components such as a bus, an input / output interface, etc. are omitted. In addition, according to specific application situations, the electronic device 11 may also include any other appropriate components.

[0103] In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the testing method according to various embodiments of the present application described in the above-mentioned "Exemplary Method" section of this specification.

[0104] The computer program product may be written in any combination of one or more programming languages ​​to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages, such as Java, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0105] In addition, an embodiment of the present application may also be a computer-readable storage medium on which computer program instructions are stored. When the computer program instructions are executed by a processor, the processor executes the steps of the testing method according to various embodiments of the present application described in the above “Exemplary Method” section of this specification.

[0106] The computer readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can include, for example, but is not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0107] The effect of the EEG data processing device disclosed in the present application is the same as the above-mentioned EEG data processing method, which will not be repeated here.

[0108] The preferred embodiments of the present application are described in detail above; however, the present application is not limited to the specific details in the above embodiments. Within the technical concept of the present application, a variety of simple modifications can be made to the technical solution of the present application, and these simple modifications all fall within the protection scope of the present application.

[0109] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, this application will not further describe various possible combinations.

[0110] In addition, the various implementation modes of the present application may be arbitrarily combined, and as long as they do not violate the concept of the present application, they should also be regarded as the contents disclosed by the present invention.

Claims

1. A method for processing electroencephalogram data, characterized in that: The method comprises: Preprocess the original EEG signal; The preprocessed EEG signal is divided into microstates in five frequency bands respectively; the five frequency bands include delta frequency band, theta frequency band, alpha frequency band, beta frequency band, and gamma frequency band; the microstates include four types of microstates; Extracting features of the microstates of the EEG signals in each frequency band; the features include: duration features, occurrence frequency features, occupancy features, and transition probability features of each microstate category; The SVM-RFE model is used to screen the extracted features and obtain the target features.

2. The method according to claim 1, characterized in that The preprocessing includes one or more steps of: removing irrelevant electrodes, filtering, whole-brain average re-reference, baseline correction, and bad segment removal.

3. The method according to claim 2, characterized in that The filtering includes one or more steps of removing power frequency noise and bandpass filtering.

4. The method according to claim 1, characterized in that: The preprocessed EEG signals are divided into microstates in five frequency bands, including: The preprocessed EEG signals are projected into a low-dimensional space using independent component analysis in the five frequency bands, and the k-means clustering algorithm is applied to the low-dimensional space to divide the microstates.

5. The method according to claim 1, characterized in that Before using the SVM-RFE model to perform feature screening, the method further includes: Construct a feature matrix for each frequency band and each microstate feature; A normalization operation is performed on the feature matrix.

6. The method according to claim 5, characterized in that The normalization operation is a minimum-maximum normalization operation.

7. The method according to claim 1, characterized in that The SVM-RFE model is used to screen features and obtain target features, including: Build the SVM-RFE model; Using the RFE algorithm to screen features in the SVM-RFE model and iteratively training the SVM-RFE model according to the screening results until the classification accuracy of the model no longer increases or reaches a preset accuracy; The target features are obtained using the final SVM-RFE model.

8. The method according to claim 7, characterized in that Training the SVM-RFE model includes: The SVM-RFE model is trained using a network search method.

9. An electroencephalogram data processing device, characterized in that: The device comprises: A preprocessing module is configured to preprocess the original EEG signal; The microstate division module is configured to divide the preprocessed EEG signal into microstates in five frequency bands respectively; the five frequency bands include delta frequency band, theta frequency band, alpha frequency band, beta frequency band, and gamma frequency band; the microstates include four types of microstates; A feature extraction module is configured to extract features of the microstates of the EEG signals in each frequency band, wherein the features include: time features of each microstate category, occurrence frequency features, occupancy features, and transition probability features of each microstate; The feature screening module is configured to screen the extracted features using the SVM-RFE model to obtain target features.

10. The device according to claim 9, characterized in that The feature screening module is also used for: Construct SVM-RFE model; Using the RFE algorithm to screen features in the SVM-RFE model and iteratively training the SVM-RFE model according to the screening results until the classification accuracy of the model no longer increases or reaches a preset accuracy; The target features are obtained using the final SVM-RFE model.

11. An electronic device, comprising: processor; a memory for storing instructions executable by the processor; The processor is used to read the executable instructions from the memory and execute the instructions to implement the method described in any one of claims 1 to 8.

12. A computer-readable storage medium storing a computer program, wherein the computer program is used to execute the method according to any one of claims 1 to 8.

Images