CN113397567A

CN113397567A - Human behavior electroencephalogram signal classification method and system

Info

Publication number: CN113397567A
Application number: CN202110544645.1A
Authority: CN
Inventors: 王夏爽; 毛磊; 黄旭辉; 肖柯; 孙科武
Original assignee: Second Research Institute Of Casic
Current assignee: Second Research Institute Of Casic
Priority date: 2021-05-19
Filing date: 2021-05-19
Publication date: 2021-09-17
Anticipated expiration: 2041-05-19
Also published as: CN113397567B

Abstract

The invention discloses a human behavior electroencephalogram signal classification method and system, relates to the technical field of electroencephalogram signal classification, and aims to solve the problem of poor classification accuracy of the conventional electroencephalogram signals. The classification method comprises the steps of collecting electroencephalogram signals corresponding to a certain action of a person, taking frequency domain characteristics of the electroencephalogram signals as input, and obtaining human actions corresponding to the electroencephalogram signals by using a trained random forest model. In the training process of the random forest model, the number of the features included in the historical frequency domain features input to the random forest model is optimized by using a K-means algorithm to obtain new effective features, the random forest model is trained by using the historical frequency domain features and the effective features, the training effect of the random forest model can be improved, and when the trained random forest model is used for determining human behaviors corresponding to electroencephalograms, the accuracy is high, and the classification effect is good.

Description

Human behavior electroencephalogram signal classification method and system

Technical Field

The invention relates to the technical field of electroencephalogram signal classification, in particular to a GSO-Kmeans double-optimization human behavior electroencephalogram signal classification method and system based on random forests.

Background

The artificial intelligence technology is rapidly developing as a cross discipline converging various leading-edge technologies and theories, so that people continuously develop the perception of intelligence and are continuously updated. Traditional artificial intelligence technologies, such as random forests, support vector machines, convolutional neural networks, cyclic neural networks and the like, are various, and most of the technologies are not enough to support high-accuracy classification of unstable physiological signals in various behavior modes, especially unstable weak signals such as electroencephalogram signals.

Disclosure of Invention

The invention aims to provide a human behavior electroencephalogram signal classification method and system, which are suitable for classifying electroencephalograms generated under different behavior modes, can improve the classification performance of the electroencephalograms, and realize accurate identification of the electroencephalograms with various behaviors.

In order to achieve the above purpose, the invention provides the following technical scheme:

a human behavior electroencephalogram signal classification method comprises the following steps:

acquiring an electroencephalogram signal;

extracting the characteristics of the electroencephalogram signals to obtain frequency domain characteristics;

taking the frequency domain characteristics as input, and obtaining human behaviors corresponding to the electroencephalogram signals by using a trained random forest model; in the training process of the random forest model, optimizing the number of features included in the historical frequency domain features input to the random forest model by using a K-means algorithm to obtain new effective features, and training the random forest model by using the historical frequency domain features and the effective features.

Compared with the prior art, the electroencephalogram signal classification method for the human behaviors, provided by the invention, is used for collecting the electroencephalogram signal corresponding to a certain behavior of a human, and extracting the characteristics of the electroencephalogram signal to obtain the frequency domain characteristics. And finally, taking the frequency domain characteristics as input, and obtaining human behaviors corresponding to the electroencephalogram signals by using the trained random forest model. In the training process of the random forest model, the number of the features included in the historical frequency domain features input to the random forest model is optimized by using a K-means algorithm to obtain new effective features, the random forest model is trained by using the historical frequency domain features and the effective features, the training effect of the random forest model can be improved, the accuracy of prediction is improved, and when the trained random forest model is used for determining the human behaviors corresponding to the electroencephalogram signals, the accuracy is high and the classification effect is good.

The invention also provides a human behavior electroencephalogram signal classification system, which comprises:

the acquisition module is used for acquiring an electroencephalogram signal;

the extraction module is used for extracting the characteristics of the electroencephalogram signals to obtain frequency domain characteristics;

the classification module is used for taking the frequency domain characteristics as input and obtaining human behaviors corresponding to the electroencephalogram signals by utilizing a trained random forest model; in the training process of the random forest model, optimizing the number of features included in the historical frequency domain features input to the random forest model by using a K-means algorithm to obtain new effective features, and training the random forest model by using the historical frequency domain features and the effective features.

Compared with the prior art, the human behavior electroencephalogram signal classification system provided by the invention has the same beneficial effects as the human behavior electroencephalogram signal classification method in the technical scheme, and the details are not repeated here.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention and not to limit the invention. In the drawings:

fig. 1 is a flowchart of a classification method provided in embodiment 1 of the present invention.

Fig. 2 is a schematic diagram of an electroencephalogram signal acquisition channel provided in embodiment 1 of the present invention.

Fig. 3 is a flowchart of electroencephalogram signal preprocessing provided in embodiment 1 of the present invention.

Fig. 4 is a schematic diagram illustrating a result of correlation analysis of each frequency band of an electroencephalogram signal provided in embodiment 1 of the present invention.

FIG. 5 is a comparison graph of the indices of the three models provided in example 1 of the present invention.

FIG. 6 is a ROC curve of the RF-GSO-Kmeans model provided in example 1 of the present invention.

Fig. 7 is a system block diagram of the classification system provided in embodiment 2 of the present invention.

Detailed Description

For the convenience of clearly describing the technical solutions of the embodiments of the present invention, in the present invention, "at least one" means one or more, "and" a plurality "means two or more. "and/or" describes the association relationship of the associated objects, meaning that there may be three relationships, e.g., a and/or B, which may mean: a exists alone, A and B exist simultaneously, and B exists alone, wherein A and B can be singular or plural. The character "/" generally indicates that the former and latter associated objects are in an "or" relationship. "at least one of the following" or similar expressions refer to any combination of these items, including any combination of the singular or plural items. For example, at least one (one) of a, b, or c, may represent: a, b, c, a and b combination, a and c combination, b and c combination, or a, b and c combination, wherein a, b and c can be single or multiple.

Example 1:

the embodiment is used for providing a classification method of human behavior electroencephalogram signals, and as shown in fig. 1, the classification method includes the following steps:

s1: acquiring an electroencephalogram signal;

the human cerebral cortex can be divided into two large pieces: left and right brains, each large block containing four parts: frontal, temporal, occipital and parietal brains. The brain regions have different functions, such as the frontal lobe of the brain located in front of the central sulcus, and are mainly responsible for the thinking and planning of the individual and related to the emotional needs of the individual. The brain area of the parietal lobe part is responsible for the sensations of painful touch, taste, temperature, pressure, and is mathematically and logically related to the brain area behind the central sulcus and before the parietal occipital fissure links with the anterior occipital notch.

For the acquisition of the electroencephalogram signals, the acquisition of the original electroencephalogram signals is carried out by adopting the American Nueroscan32 lead electroencephalogram acquisition equipment, and the electrodes are placed in the order of 10-20 international standards, namely the relative distance is determined by 10% and 20%. This embodiment uses two marker lines, one called the sagittal line, which is the line from the nasion to the inion, 5 points are marked from the anterior to the posterior: fpz, Fz, Cz, Pz, Oz. The length of the line before and after Fpz was 10% of the total length, and the distance between the remaining points was 20% of the total length. The other is called a coronal line, which is a connecting line between two external auditory canals, and 5 points are marked from left to right: t7, C3, Cz, C4 and T8. The outer sides of T7 and T8 respectively account for 10%, the distances between the other points all account for 20% of the total length, and the electroencephalogram channel is shown in figure 2. In the process of acquiring electroencephalogram signals, the impedance of all electrodes needs to be ensured to be less than 10k omega, M1 and M2 electrodes at the positions of the two ear papilla are used as reference electrodes, and the sampling frequency is 1000 Hz.

As shown in fig. 3, the classification method according to this embodiment further includes preprocessing the acquired electroencephalogram signal, where the preprocessing process may include: the method comprises the following steps of firstly, visualizing data, and checking whether mutual coupling exists among signals of each single channel of an original electroencephalogram signal. And secondly, re-referencing the electroencephalogram reference electrode, wherein the study on the behavioristics is carried out in the embodiment, so that bilateral mastoids can be used as the re-referenced electrodes. And thirdly, selecting the frequency of interest for band-pass filtering, wherein the specific selection frequency range of the embodiment is 0-55 Hz. Fourthly, because the electroencephalogram signals are easily interfered by signals of electrical equipment such as an air conditioner and the like, power frequency interference is selectively removed, and the set removal range is 49Hz-51 Hz. And fifthly, analyzing and processing the independent components of the electroencephalogram signals, searching signal interference items, and removing eye drift, blink, head movement artifacts and the like. And sixthly, performing baseline correction to align and standardize the data benchmarks of all channels. And seventhly, removing power frequency interference. In order to facilitate observation, the embodiment can also amplify the weak electroencephalogram signals, superimpose the data, remove invalid abnormal values and obtain a normalized low-noise data set.

Of course, the embodiment may also directly use the eeglab or letswave toolbox of MATLAB to pre-process the acquired electroencephalogram signals. In addition, the acquisition equipment for acquiring the electroencephalogram signals is generally provided with an electroencephalogram data preprocessing module, and the acquired electroencephalogram signals can be preprocessed by directly utilizing the electroencephalogram data preprocessing module.

S2: extracting the characteristics of the electroencephalogram signals to obtain frequency domain characteristics;

after preprocessing the electroencephalogram signal to obtain a noiseless electroencephalogram signal, the embodiment performs multi-level discrete wavelet decomposition on the noiseless electroencephalogram signal according to the wave frequency, extracts the characteristics of the electroencephalogram signal, and obtains the frequency domain characteristics. The frequency domain features include alpha waves (alpha waves with a frequency range of 6-12 Hz), delta waves (delta waves with a frequency range of 12-25 Hz), and theta waves (theta waves with a frequency range of 3-6 Hz).

In the frequency domain features extracted in this embodiment, each feature included in the frequency domain features may not have separability, so that the frequency domain features are further screened after the frequency domain features are obtained in this embodiment, to obtain screened features, that is, the features included in the screened features have separability, and the screened features are used as input of the trained random forest model. If the screened features are used as the input of the trained random forest model, the historical screened features are also used as the input when the random forest model is trained.

Screening the frequency signature may include:

and screening the frequency domain characteristics by using a screening method based on a decision tree. Specifically, the improved DT decision tree-based screening method analyzes the contribution degree of each feature included in the frequency domain features, and retains the features with large contribution degrees.

Or screening the frequency domain characteristics by using an ANOVA-based screening method. A screening method of ANOVA (analysis of variance) based on multivariate statistical analysis is used for searching whether the characteristics show significant differences among different behaviors and rejecting the characteristics without significant differences in frequency domain characteristics. By adopting ANOVA analysis, the P value of each feature in the frequency domain features is found to be less than 0.05, so that each feature in the frequency domain features has significant difference, and the large significant difference represents that the feature has better separability for distinguishing behaviors.

Or screening the frequency domain characteristics by using a characteristic screening method of the multi-mode convolutional neural network. Specifically, the feature screening method based on the multi-modal convolutional neural network is used for determining the weight of each feature contained in the frequency domain features in the full-connected layer of the multi-modal convolutional neural network to obtain the screened features.

Or screening the frequency domain characteristics by using a correlation analysis method. For the features of the electroencephalogram signal, the embodiment may perform correlation analysis on the frequency band features of the 32 channels, check the correlation of the frequency band features among the 32 channels of the electroencephalogram signal, and the correlation result is shown in fig. 4, where a lighter color represents a higher correlation degree, and a darker color represents a lower correlation degree. Therefore, the features in the frequency domain features can be screened out and combined to obtain the screened features.

Any one of the four methods is used for carrying out characteristic screening processing, and a preliminary foundation can be made for improving the model training speed.

S3: taking the frequency domain characteristics as input, and obtaining human behaviors corresponding to the electroencephalogram signals by using a trained random forest model; in the training process of the random forest model, optimizing the number of features included in the historical frequency domain features input to the random forest model by using a K-means algorithm to obtain new effective features, and training the random forest model by using the historical frequency domain features and the effective features.

And dividing the plurality of historical frequency domain features into three groups, namely a training set, a testing set and a verification set for preventing the random forest model from being over-fitted. And then, taking the plurality of historical frequency domain characteristics as input, taking the historical human behavior corresponding to each historical frequency domain characteristic as label data, and training the random forest model by adopting a cross validation method to obtain the trained random forest model, so that overfitting of the random forest model can be prevented. The cross validation method specifically comprises the following steps: an original training set composed of a plurality of historical frequency domain characteristics is divided into two groups, namely a training set and a test set. And then, averagely dividing the training set into k sub-parts, randomly selecting one sub-part from the k sub-parts as a verification set during each iteration, and using the rest (k-1) sub-parts as training sets for training the random forest model. This process is repeated k times until each sub-portion is taken as a verification set.

As an optional implementation, the original training set composed of multiple historical frequency domain features is divided into two groups, and the method used when the training set and the test set are respectively the training set and the test set is as follows: and (2) extracting M times in an original training set consisting of a plurality of historical frequency domain characteristics to obtain a randomly generated training set, training the random forest model by using the randomly generated training set, and forming the rest historical frequency domain characteristics in the original training set into a test set to realize the division of the training set and the test set, so that the training effect of the random forest model can be improved.

In order to further improve the classification accuracy of the trained random forest model, the embodiment further optimizes the number of features included in the historical frequency domain features input to the random forest model by using a K-means algorithm to obtain new effective features, and trains the random forest model by using the historical frequency domain features and the effective features. Specifically, clustering analysis is performed on the features included in the historical frequency domain features, and a similar label is added to each category in the clustering result to serve as a new effective feature. The number of the clustered categories needs to be solved by adopting a point-by-point trial method, and the accuracy is obviously improved when the number of the clustered categories is km-5 and km-7 through repeated tests. In the embodiment, the frequency domain features are optimized by using the K-means algorithm, the number of features in the training of the random forest model is increased, the training effect of the random forest model can be further improved, and the classification accuracy of the trained random forest model is improved.

As an optional implementation manner, the embodiment further optimizes the hyperparameters in the training process by using a grid search optimization algorithm, so that manual parameter adjustment is avoided, and the training effect of the random forest model is improved.

To further illustrate the superiority of the optimization method used in this example, the following index enhancements are used. Under seven different tasks, the table 1 is an index value of a traditional random forest model, the table 2 is an index value obtained by optimizing the random forest model by using a grid search optimization algorithm GSO, and the table 3 is an index value obtained by optimizing the random forest model by simultaneously using the grid search optimization algorithm GSO and a K-means algorithm.

TABLE 1

TABLE 2

TABLE 3

Based on the table 1, the table 2 and the table 3, the accuracy of the optimized RF-GSO is improved by three percentage points compared with the accuracy of the RF before the optimization, the accuracy is improved to 70.01% from 66.41%, the accuracy of the optimized RF-GSO-Kmeans is improved by 12 percentage points compared with the accuracy of the optimized RF-GSO, and the accuracy is improved to 82% from 70%.

In order to more intuitively explain the superiority of the optimization method, fig. 5(a) shows a comparison graph of precision indexes of RF, RF-GSO, and RF-GSO-Kmeans, and fig. 5(b) shows a comparison graph of recall indexes of RF, RF-GSO, and RF-GSO-Kmeans. It can be seen that the precision index of the RF-GSO-Kmeans is 82%, which is better improved compared with the precision of the RF and RF-GSO models, and the recall index of the RF-GSO-Kmeans model is relatively stable between 75% and 87%. The machine learning rating indexes show that the optimized RF-GSO-Kmeans model has better robustness and accuracy.

The relative weights of the two types of models, RF-GSO and RF-GSO-Kmeans, for each of the frequency domain features are given below, as shown in tables 4 and 5. Table 4 shows the weighting results of RF-GSO, and Table 5 shows the weighting results of RF-GSO-Kmeans.

TABLE 4

Weight of	Feature(s)
		0.346441	delta
0.335025	alpha
		0.318534	theta

TABLE 5

Weight of	Feature(s)
		0.391964	delta
0.306867	alpha
		0.297256	theta
0.002583	km5
		0.001329	km7

Based on tables 4 and 5, it can be seen that the relative weight of delta is greater, followed by alpha, theta. The two optimized km indexes also account for a certain percentage, and the accuracy rate improves and proves the improvement of the model performance.

FIG. 6 shows a receiver operating characteristic curve (ROC) curve of the RF-GSO-Kmeans model for each task. As can be seen, the AUC value predicted by the RF-GSO-Kmeans model for each task can reach 92.4% -98.2%.

The classification method provided by this embodiment uses a conventional random forest algorithm as a base point, and performs step-by-step optimization on a model by using a grid search optimization algorithm GSO and a Kmeans optimization algorithm respectively in a model training process, optimizes parameters in the model training process by using the GSO algorithm, and optimizes characteristics of electroencephalogram signals by using the Kmeans algorithm, thereby improving classification accuracy of the classification model.

Most of traditional human behavior analysis is judgment and analysis by domain experts depending on domain knowledge and experience, and the method is too subjective and cannot carry out objective analysis from the perspective of actual people. The embodiment also provides a human behavior electroencephalogram classification method based on a multi-element sequence network coding scheme aiming at the bottleneck faced by human behavior analysis.

Specifically, after obtaining the frequency domain features at S2, the classification method further includes: and (3) taking the frequency domain characteristics as input, and obtaining the human behavior corresponding to the electroencephalogram signals by utilizing a multivariate sequence network coding scheme expression.

The classification method further includes obtaining a multiple sequence network coding scheme expression, which may include: and taking a plurality of historical frequency domain characteristics and historical human behaviors corresponding to the historical frequency domain characteristics as input, and calculating the contribution degree corresponding to each characteristic contained in the historical frequency domain characteristics by using the multi-element sequence network coding scheme to obtain the expression of the multi-element sequence network coding scheme.

The basic expression of the multivariate sequence network coding scheme used in this example is as follows:

in formula 1, Y is human behavior; beta is a₀Is a model constant; n ═ 1, 2,. n; n is the number of features contained in the frequency domain features; beta is a_iThe contribution degree corresponding to the ith feature; x is the number of_iIs the ith characteristic; e is an unobservable random error, E (E) is expected to be 0, and D (E) is the variance of E (σ)²。

Since the behavior of a person is not an absolute value but a variation value relative to its resting state, the multi-sequence network coding scheme established in this embodiment modifies the basic expression as follows:

in formula 2, Δ Y is the amount of change in human behavior; beta'₀Is a model constant; n ═ 1, 2,. n; n is the number of features contained in the frequency domain features; beta'_iThe contribution degree corresponding to the ith feature; Δ x_iMaking a difference before and after the human behavior for the ith feature; ε ' is an unobservable random error, with an expected E (ε ') of ε ' being 0 and a variance D (ε ') of ε ' being σ²。

In this embodiment, it is also default that each behavior is 0 in the resting state, so after simplifying equation 2, the obtained multiple sequence network coding scheme is as follows:

in formula 3, Y is human behavior; beta'₀Is a model constant; n ═ 1, 2,. n; n is the number of features contained in the frequency domain features; beta'_iThe contribution degree corresponding to the ith feature; Δ x_iMaking a difference before and after the human behavior for the ith feature; ε' is the random error.

In this embodiment, the expression 3 may also be transformed to obtain a conceptual expression of the multiple sequence network coding scheme expression, as follows:

CBM_d(t)＝β′₀+β′₁δ+β′₂θ+β′₃α+ε′ (4)

in formula 4, CBM_d(t) is human behavior; beta'₀Is a model constant; beta'₁Contribution degree corresponding to delta; beta'₂Contribution degree corresponding to theta; beta'₃The contribution degree corresponding to alpha; ε' is the random error.

After the multi-element sequence network coding scheme is trained, an expression of the multi-element sequence network coding scheme can be obtained, and the multi-element sequence network coding scheme expression established by the embodiment is equivalent to modeling in a human cognition process, but not modeling on a cognition result. The present embodiments are capable of analyzing human behavior based on a multi-sequence network coding scheme. After the multivariate sequence network coding scheme is modified, the frequency domain characteristics of the electroencephalogram signals are used as model factors, and the contribution degree of each characteristic contained in the frequency domain characteristics is used as a model parameter, so that an actual expression of the detection analysis model of the electroencephalogram of human behaviors is obtained.

After obtaining the human behavior, the analysis method may further include: the frequency domain characteristics and the human behaviors are used as input, the contribution degree corresponding to each characteristic contained in the frequency domain characteristics is calculated by utilizing the multivariate sequence network coding scheme expression, and further the human factor characteristic rule of the behaviors of the human in the behavior decision process can be determined.

Example 2:

the embodiment of the present invention may perform functional module division on the human behavior analysis system according to the above method example, for example, each functional module may be divided corresponding to each function, or two or more functions may be integrated into one processing module. The integrated module can be realized in a hardware mode, and can also be realized in a software functional module mode. It should be noted that, the division of the modules in the embodiment of the present invention is schematic, and is only a logic function division, and there may be another division manner in actual implementation.

Under the condition that each functional module is divided according to each function, fig. 7 shows a human behavior electroencephalogram signal classification system provided by an embodiment of the invention, which comprises:

the acquisition module M1 is used for acquiring electroencephalogram signals;

the extraction module M2 is used for extracting the characteristics of the electroencephalogram signals to obtain frequency domain characteristics;

the classification module M3 is used for taking the frequency domain characteristics as input and obtaining human behaviors corresponding to the electroencephalogram signals by using a trained random forest model; in the training process of the random forest model, optimizing the number of features included in the historical frequency domain features input to the random forest model by using a K-means algorithm to obtain new effective features, and training the random forest model by using the historical frequency domain features and the effective features.

All relevant contents of each step related to the above method embodiment may be referred to the functional description of the corresponding functional module, and are not described herein again.

While the invention has been described in connection with various embodiments, other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention, from a review of the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other elements or steps, and the word "a" or "an" does not exclude a plurality. A single processor or other unit may fulfill the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.

While the invention has been described in conjunction with specific features and embodiments thereof, it will be evident that various modifications and combinations can be made thereto without departing from the spirit and scope of the invention. Accordingly, the specification and figures are merely exemplary of the invention as defined in the appended claims and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of the invention. It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.

Claims

1. A human behavior electroencephalogram signal classification method is characterized by comprising the following steps:

acquiring an electroencephalogram signal;

2. The method of claim 1, wherein the extracting the features of the electroencephalogram signal specifically comprises:

performing multi-level discrete wavelet decomposition on the electroencephalogram signal, and extracting the characteristics of the electroencephalogram signal to obtain frequency domain characteristics; the frequency domain features include alpha waves, delta waves, and theta waves.

3. The method as claimed in claim 1, wherein after obtaining the frequency domain features, the method further comprises screening the frequency domain features to obtain screened features, and using the screened features as an input of the trained random forest model, specifically comprising:

screening the frequency domain features by using a screening method based on a decision tree;

or screening the frequency domain characteristics by using an ANOVA-based screening method;

or screening the frequency domain features by using a feature screening method of a multi-mode convolutional neural network;

or screening the frequency domain characteristics by using a correlation analysis method.

4. The method as claimed in claim 1, wherein in training the random forest model, the method further comprises:

and taking a plurality of historical frequency domain characteristics as input, taking historical human behaviors corresponding to each historical frequency domain characteristic as label data, and training the random forest model by adopting a cross validation method to obtain the trained random forest model.

5. The method as claimed in claim 1, wherein in training the random forest model, the method further comprises generating a training set, specifically comprising:

and extracting M times from the original training set to obtain a randomly generated training set, and training the random forest model by using the randomly generated training set.

6. The method as claimed in claim 1, wherein in training the random forest model, the method further comprises:

and optimizing the hyper-parameters in the training process by adopting a grid search optimization algorithm.

7. The method of claim 1, wherein after obtaining the frequency domain features, the method further comprises: and taking the frequency domain characteristics as input, and obtaining the human behavior corresponding to the electroencephalogram signals by utilizing a multivariate sequence network coding scheme expression.

8. The method according to claim 7, further comprising obtaining a multiple sequence network coding scheme expression, specifically comprising:

and taking a plurality of historical frequency domain characteristics and historical human behaviors corresponding to the historical frequency domain characteristics as input, and calculating the contribution degree corresponding to each characteristic contained in the historical frequency domain characteristics by using a multi-element sequence network coding scheme to obtain a multi-element sequence network coding scheme expression.

9. The method of claim 7, wherein after obtaining the human behavior corresponding to the brain electrical signal, the method further comprises:

and taking the frequency domain features and the human behaviors as input, and calculating the contribution degree corresponding to each feature contained in the frequency domain features by utilizing a multivariate sequence network coding scheme expression.

10. A human behavior electroencephalogram signal classification system, comprising:

the acquisition module is used for acquiring an electroencephalogram signal;