Electroencephalogram feature optimization extraction method based on deep learning model

Abstract

The application discloses an electroencephalogram feature optimization extraction method based on a deep learning model, relates to the technical field of electroencephalogram analysis, and improves the emotion recognition accuracy.The application obtains a characteristic emotional signal segment by matching detection signal segments of different standard emotional electroencephalogram data with each other, traverses a plurality of characteristic segments of the characteristic emotional signal segment in each standard emotional electroencephalogram data, acquires the space-time correlation weight of each characteristic segment relative to the standard emotional electroencephalogram data where the characteristic segment is located, and then establishes a multidimensional feature space-time matrix under various emotional stimuli, establishes an emotional correlation tree corresponding to the multidimensional feature space-time matrix, and then matches and connects the emotional correlation trees under different emotional stimuli in sequence to obtain a full-emotional feature forest.The latest generated standard emotional electroencephalogram data is converted into a detection feature submatrix sequence and input into the full-emotional feature forest, and then the emotion proportion corresponding to the standard emotional electroencephalogram data is output.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electroencephalogram analysis, in particular to an electroencephalogram feature optimization extraction method based on a deep learning model. BACKGROUND

[0002] Emotion recognition has important significance in many fields, such as mental health monitoring, intelligent human-computer interaction, market research, etc. Traditional emotion recognition methods are mainly based on facial expressions, voice tones and other external manifestations. Traditional methods have certain limitations, such as being easily affected by subjective disguise, environmental interference and other factors, resulting in low recognition accuracy.

[0003] With the development of electroencephalography technology, emotion recognition using electroencephalogram (EEG) signals has become a research hotspot. Electroencephalogram signals can directly reflect the electrical activity of the brain and more accurately reflect the emotional state of individuals. However, current emotion recognition based on electroencephalogram signals still faces many challenges. On the one hand, the original electroencephalogram signal contains a large amount of noise and interference, and how to effectively preprocess it to extract valuable information is a key problem.

[0004] On the other hand, electroencephalogram signals have complex spatio-temporal characteristics, and how to accurately capture and analyze these characteristics and establish an effective emotion recognition model is an important problem to improve the accuracy and reliability of emotion recognition. Existing methods have shortcomings in processing the spatio-temporal characteristics of electroencephalogram signals, resulting in the need to improve the effect of emotion recognition. Therefore, an electroencephalogram feature optimization extraction method based on a deep learning model is provided. SUMMARY

[0005] In order to solve the above technical problems, the purpose of the present application is to provide an electroencephalogram feature optimization extraction method based on a deep learning model.

[0006] In order to achieve the above purpose, the present application provides the following technical scheme:

[0007] The electroencephalogram feature optimization extraction method based on a deep learning model comprises the following steps:

[0008] Step S1, setting an electroencephalogram acquisition channel to acquire a plurality of original emotional electroencephalogram signals under a plurality of emotional stimuli, and after standardizing the original emotional electroencephalogram signals, obtaining standard emotional electroencephalogram data;

[0009] Step S2, setting a feature extraction network, the feature extraction network is provided with a plurality of feature extraction kernels, then a plurality of initial feature extraction points are randomly selected in each standard emotional electroencephalogram data, and then the feature extraction kernel extracts and detects signal segments from the initial feature extraction points to both sides, matches the detected signal segments of different standard emotional electroencephalogram data with each other, and then generates emotional signal segments;

[0010] Step S3, record each emotional signal segment as a feature emotional signal segment, traverse several feature segments in each standard emotional EEG data through the feature emotional signal segment, obtain the spatiotemporal correlation weight of each feature segment relative to the standard emotional EEG data where it is located, and then establish a multi-dimensional feature spatiotemporal matrix under various emotional stimuli;

[0011] Step S4, establish the emotional correlation tree corresponding to the multi-dimensional feature spatiotemporal matrix, and then sequentially match and connect the emotional correlation trees corresponding to different emotional stimuli to obtain a full emotional feature forest. The newly generated standard emotional EEG data is converted into a detection feature sub-matrix sequence and input into the full emotional feature forest, and then the emotional proportion associated with the standard emotional EEG data is output.

[0012] Further, the process of collecting the original emotional EEG signal includes:

[0013] A 32-channel EEG acquisition device is installed on the head of a plurality of subjects, and the signal acquisition frequency of the 32-channel EEG acquisition device is set to a bandwidth of 0.5-45Hz. Then, the original emotional EEG signal of the subject under different emotional stimuli is collected by the 32-channel EEG acquisition device.

[0014] Further, the process of standardizing the original emotional EEG signal includes:

[0015] Set m time nodes, obtain the mean and standard deviation of each original emotional EEG signal, and then standardize each original emotional EEG signal by the Z-score standardization method. The standardized values at each time node are sequentially connected in time order, and then the standard emotional EEG data corresponding to the original emotional EEG signal is obtained. m is a natural number greater than 100.

[0016] Further, the process of extracting the detection signal segment includes:

[0017] The feature extraction network is provided with n feature extraction kernels, and each standard emotional EEG data is simultaneously input into the feature extraction network, and then the feature extraction network randomly selects n initial feature extraction points from the standard emotional EEG data, and n is a natural number greater than 50.

[0018] Set a unit diffusion length, and let each feature extraction kernel sequentially attach to the initial feature extraction point, and then from both ends of each initial feature extraction point, each feature extraction kernel extracts a detection signal segment of unit diffusion length from both ends with the initial feature extraction point it attaches to as the center.

[0019] Further, the process of matching the detection signal segments of different standard emotional EEG data includes:

[0020] The detection signal segments of the standard emotional EEG data under different emotional stimuli are matched with each other, if there are two detection signal segments with similarity above 90%, the initial feature extraction points corresponding to the two detection signal segments are removed, and a new initial feature extraction point is randomly set on the corresponding standard emotional EEG data;

[0021] If there are two detection signal segments with similarity below 90%, the two detection signal segments are retained, and the next detection signal segment matching is continued;

[0022] After all the detection signal segments are matched with each other, the detection signal segments corresponding to the same emotional stimuli but from different standard emotional EEG data are matched with each other, and a similarity threshold is set;

[0023] If the similarity between the two detection signal segments is greater than or equal to the similarity threshold, the same feature label is set for the two detection signal segments, otherwise no operation is performed;

[0024] The initial feature extraction point positions of each detection signal segment without a feature label are removed, and a new feature extraction point is randomly selected, and the initial feature extraction point positions of the detection signal segments with a feature label are removed, and a detection signal segment with a double unit diffusion length is extracted with the initial feature extraction point as the center;

[0025] First, for different standard emotional EEG data under different emotional stimuli, if the detection signal segment with a double unit diffusion length or more has a detection signal segment shorter than it with similarity above 90% in the secondary matching result, the initial feature extraction point corresponding to the detection signal segment shorter than it is removed, and a new initial feature extraction point is set;

[0026] If the similarity with the detection signal segment with the same length or longer is above 90% in the secondary matching result, the two detection signal segments are returned to a detection signal segment with a unit diffusion length, and the returned detection signal segment is recorded as an emotional signal segment;

[0027] The detection signal segment matching process is repeated for a detection signal segment with a unit diffusion length, and the detection signal segment matching process is used for three, four, …, i unit diffusion length detection signal segments, i is a natural number greater than 3.

[0028] Further, whenever a new initial feature extraction point falls within the recorded emotional signal segment, and the detection signal segment centered at the initial feature extraction point is completely within the emotional signal segment, a permanent reduction of one initial feature extraction point on the standard emotional EEG data is made, and the above-mentioned detection signal segment matching process is repeated until there is no initial feature extraction point on the standard emotional EEG data.

[0029] Further, the process of establishing a multi-dimensional feature space-time matrix includes:

[0030] After the matching between each standard emotional EEG data in step S2 is completed, the emotional signal segment in each standard emotional EEG data is recorded as a feature emotional signal segment, and each feature emotional signal segment in the standard emotional EEG data under the same emotional stimulus is matched with each other, and the feature emotional signal segments with a similarity greater than or equal to the similarity threshold are spliced and fused;

[0031] Further, each standard emotional EEG data is traversed by the feature emotional signal segment, and a plurality of feature segments are labeled in each standard emotional EEG data according to the traversal result;

[0032] The query matrix, key matrix and value matrix of each feature segment are obtained by the multi-head self-attention mechanism method, the entire standard emotional EEG data is recorded as a brain region, and the spatio-temporal correlation weight between the corresponding feature segment and the brain region is obtained according to the query matrix, the key matrix and the value matrix;

[0033] The spatio-temporal correlation weights of the same feature segment under each occurrence time sequence are integrated to generate a corresponding time sequence weight matrix, and the occurrence time sequence of each feature segment under the condition of the maximum spatio-temporal correlation weight is selected, each feature segment is overlapped in turn and converted into a corresponding multi-dimensional feature space-time matrix.

[0034] Further, the process of establishing a full emotional feature forest includes:

[0035] The multi-dimensional feature space-time matrix is divided into a plurality of feature sub-matrices, and each feature sub-matrix is set with a feature tree node, and each feature segment and the corresponding time sequence weight matrix are input into the feature tree node;

[0036] Meanwhile, according to the corresponding spatio-temporal position of each feature sub-matrix in the multi-dimensional feature space-time matrix, each feature tree node is sequentially connected to obtain an emotional correlation tree under the corresponding type of emotional stimulus;

[0037] The feature tree nodes in the emotional correlation trees under different types of emotional stimuli are matched with each other, and an association similarity threshold is set. It should be noted that the association similarity threshold is less than 0.9.

[0038] If the similarity between the feature segments in the two feature tree nodes is greater than or equal to the association similarity threshold, the two feature tree nodes are connected to each other, otherwise no operation is performed.

[0039] The above feature tree node matching and connecting process is repeated to obtain the full-emotion feature forest.

[0040] Further, the process of obtaining the emotion proportion of the standard emotional electroencephalogram data to be analyzed includes:

[0041] The feature segments under various emotional stimuli are traversed on the standard emotional electroencephalogram data, and the feature segment sequences under multiple emotions are obtained according to the traversal results.

[0042] The feature segment sequences are converted into detection feature sub-matrix sequences, and the detection feature sub-matrix sequences are input into the full-emotion feature forest.

[0043] According to the appearance timing, the first detection feature sub-matrix is sequentially compared with each emotion association tree in the full-emotion feature forest, and the corresponding timing weight matrix is retrieved according to the appearance timing of each detection feature sub-matrix, and the timing weight value of each detection feature sub-matrix in the different emotion association trees is obtained.

[0044] The timing weight total value of the detection feature sub-matrix sequence in the different emotion association trees is accumulated, and the emotion proportions of different emotions are obtained.

[0045] Compared with the prior art, the beneficial effects of the present application are:

[0046] 1、The present application extracts the detection signal segments from the initial feature extraction point to both sides by the feature extraction kernel, and matches the detection signal segments of different standard emotional electroencephalogram data to generate emotional signal segments, which helps to mine the potential emotional features in the electroencephalogram signal and improves the accuracy and effectiveness of feature extraction.

[0047] 2、By obtaining the spatio-temporal association weight of the feature segment and establishing a multi-dimensional feature spatio-temporal matrix, the emotional information is comprehensively analyzed in the time and space dimensions while considering the spatio-temporal characteristics of the electroencephalogram signal, which more accurately describes the electrical activity pattern of the brain in the emotional state and improves the precision of emotion recognition. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments or the prior art, the drawings needed in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments described in the present application.

[0049] Figure 1 The method flowchart of the present application. DETAILED DESCRIPTION

[0050] In order to make the objectives, technical solutions and advantages of the present application clearer, the technical solutions of the present application will be described in detail below. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0051] As shown in the EEG feature optimization extraction method based on a deep learning model, comprising the following steps: Figure 1 Step S1, set up a brain electrical signal acquisition channel to collect several raw emotional EEG signals under a plurality of emotional stimuli, and after standardizing the raw emotional EEG signals, obtain standard emotional EEG data;

[0052] Step S2, set up a feature extraction network, the feature extraction network is provided with a plurality of feature extraction kernels, then randomly select a plurality of initial feature extraction points in each standard emotional EEG data, and then the feature extraction kernel diffuses and extracts a detection signal segment from the initial feature extraction point to both sides, matches the detection signal segments of different standard emotional EEG data with each other, and then generates an emotional signal segment;

[0053] Step S3, record each emotional signal segment as a feature emotional signal segment, traverse a plurality of feature segments in each standard emotional EEG data through the feature emotional signal segment, obtain the spatio-temporal correlation weight of each feature segment relative to the standard emotional EEG data where it is located, and then establish a multi-dimensional feature spatio-temporal matrix under various emotional stimuli;

[0054] Step S4, establish an emotional correlation tree corresponding to the multi-dimensional feature spatio-temporal matrix, and then match and connect the emotional correlation trees under different emotional stimuli in turn to obtain a full emotional feature forest, input the latest generated standard emotional EEG data into the full emotional feature forest as a detection feature sub-matrix sequence, and then output the emotional proportion associated with the corresponding standard emotional EEG data.

[0055] Further, the step S1 is realized by the following process:

[0056] A 32-channel brain electrical signal acquisition device is installed on the head of a plurality of subjects, the coverage range of the 32-channel brain electrical signal acquisition device includes the frontal lobe, parietal lobe, temporal lobe and occipital lobe regions of the subject's head;

[0057]

[0058] ​The signal acquisition frequency of the 32-channel electroencephalogram acquisition device is set, the bandwidth is 0.5-45Hz, and then the original emotional electroencephalogram signals of the subjects under different emotional stimuli are collected through the 32-channel electroencephalogram acquisition device, and each emotional stimulus lasts for about 10 minutes;

[0059] A signal standardization mechanism is set at each channel signal output position of the 32-channel electroencephalogram acquisition device, which is used to convert the original emotional electroencephalogram signals into corresponding standard emotional electroencephalogram data, and the specific process includes:

[0060] m time nodes are set, the mean and standard deviation of each original emotional electroencephalogram signal are obtained, and then the Z-score standardization method is used to standardize each original emotional electroencephalogram signal, and the processing formula is: Wherein Z i , X i is the standardized value and the original value of the original emotional electroencephalogram signal at the i-th time node, μ is the mean, σ is the standard deviation, and m is a natural number greater than 100;

[0061] The standardized values at each time node are sequentially connected in time order, and then the standard emotional electroencephalogram data corresponding to the original emotional electroencephalogram signal is obtained.

[0062] Further, the step S2 is realized by the following process:

[0063] The feature extraction network is provided with n feature extraction kernels, each standard emotional electroencephalogram data is input into the feature extraction network, and then the feature extraction network randomly selects n initial feature extraction points from the standard emotional electroencephalogram data, and n is a natural number greater than 50;

[0064] A unit diffusion length is set, and each feature extraction kernel is sequentially attached to the initial feature extraction point, and then each feature extraction kernel starts from both ends to cut a detection signal segment of unit diffusion length with the initial feature extraction point as the center;

[0065] The detection signal segments of the standard emotional electroencephalogram data under different emotional stimuli are matched with each other, if the similarity of two detection signal segments is more than 90%, the initial feature extraction points corresponding to the two detection signal segments are removed, and a new initial feature extraction point is randomly set on the corresponding standard emotional electroencephalogram data;

[0066] If the similarity of two detection signal segments is 90% or less, the two detection signal segments are retained, and the next detection signal segment matching is continued;

[0067] After all the detection signal segments are matched with each other, the detection signal segments corresponding to the same emotional stimulus but from different standard emotional EEG data are matched with each other, and a similarity threshold is set;

[0068] If the similarity between the two detection signal segments is greater than or equal to the similarity threshold, the same feature label is set for the two detection signal segments, otherwise no operation is performed;

[0069] The initial feature extraction point position corresponding to each detection signal segment without a feature label is removed, and a feature extraction point is randomly selected again. The initial feature extraction point position corresponding to the detection signal segment with a feature label is removed, and a detection signal segment with a double unit diffusion length is extracted with the initial feature extraction point as the center;

[0070] First, for different standard emotional EEG data under different emotional stimuli, if the detection signal segment with a double unit diffusion length or more has a detection signal segment shorter than it in the secondary matching result with a similarity of 90% or more, the initial feature extraction point corresponding to the shorter detection signal segment is removed, and the initial feature extraction point is reset;

[0071] If the similarity with the detection signal segment of the same length or longer is 90% or more in the secondary matching result, the two detection signal segments are rolled back to a detection signal segment with a unit diffusion length, and the rolled back detection signal segment is recorded as an emotional signal segment;

[0072] The detection signal segment matching process is repeated for a detection signal segment with a unit diffusion length. For subsequent three, four, …, i unit diffusion length detection signal segments, the two unit diffusion length detection signal segment matching process is used, and i is a natural number greater than 3

[0073] It should be noted that whenever a new initial feature extraction point falls within an emotional signal segment that has been recorded, and a detection signal segment with a unit diffusion length centered on the initial feature extraction point is completely within the emotional signal segment, an initial feature extraction point is permanently reduced on the standard emotional EEG data;

[0074] The above detection signal segment matching process is repeated until there is no initial feature extraction point in the standard emotional EEG data.

[0075] Further, the step S3 is implemented by the following process:

[0076] When the matching between the standard emotional EEG data in step S2 is completed, the emotional signal segments in the standard emotional EEG data are recorded as characteristic emotional signal segments, and each characteristic emotional signal segment under the same emotional stimulus is matched with each other, and the characteristic emotional signal segments with a similarity greater than or equal to a similarity threshold value are spliced and fused;

[0077] Further, the standard emotional EEG data is traversed through the characteristic emotional signal segments, and a plurality of characteristic segments are labeled in the standard emotional EEG data according to the traversal result;

[0078] The query matrix, the key matrix and the value matrix of each characteristic segment are obtained through the multi-head self-attention mechanism method, the entire standard emotional EEG data is recorded as a brain region, and then the spatio-temporal correlation weight between the corresponding characteristic segment and the brain region is obtained according to the query matrix, the key matrix and the value matrix;

[0079] The calculation formula of the spatio-temporal correlation weight is:

[0080]

[0081] Wherein A, B and C represent the query matrix, the key matrix and the value matrix of the characteristic segment respectively, The correction parameter is represented by t, which represents the occurrence time sequence of the characteristic segment in the entire standard emotional EEG data.

[0082] The spatio-temporal correlation weights of the same characteristic segment under each occurrence time sequence are integrated to generate a corresponding time sequence weight matrix, and the occurrence time sequence of each characteristic segment under the maximum spatio-temporal correlation weight is selected, each characteristic segment is overlapped in turn and converted into a corresponding multi-dimensional feature spatio-temporal matrix.

[0083] Further, the step S4 is realized by the following process:

[0084] The multi-dimensional feature spatio-temporal matrix is divided into a plurality of feature sub-matrices, and each feature sub-matrix is set with a feature tree node, and each characteristic segment and the corresponding time sequence weight matrix are input into the feature tree node;

[0085] Meanwhile, according to the spatio-temporal position of each feature sub-matrix in the multi-dimensional feature spatio-temporal matrix, each feature tree node is connected in turn to obtain an emotional correlation tree under the corresponding type of emotional stimulus.

[0086] The feature tree nodes in the emotional correlation trees under different types of emotional stimuli are matched with each other, and an association similarity threshold value is set. It should be noted that the association similarity threshold value is less than 0.9.

[0087] If the similarity between the feature segments in the two feature tree nodes is greater than or equal to the association similarity threshold, the two feature tree nodes are connected to each other, otherwise no operation is performed.

[0088] The above feature tree node matching and connecting process is repeated to obtain the full-emotion feature forest.

[0089] Further, the standard emotion electroencephalogram data to be analyzed is obtained, the feature segments under various emotional stimuli are traversed on the standard emotion electroencephalogram data, and the feature segment sequences under various emotions are obtained according to the traversal results;

[0090] Each feature segment sequence is converted into a detection feature sub-matrix sequence, and the detection feature sub-matrix is labeled with the corresponding occurrence timing, and the detection feature sub-matrix sequence is input into the full-emotion feature forest;

[0091] According to the occurrence timing, the first detection feature sub-matrix is started, and each emotion association tree in the full-emotion feature forest is sequentially and synchronously compared, and then the occurrence timing of each detection feature sub-matrix is retrieved to obtain the timing weight value of each detection feature sub-matrix in the different emotion association trees;

[0092] The total timing weight value of the detection feature sub-matrix sequence in the different emotion association trees is counted, the total timing weight values of the different emotion association trees are accumulated, and the proportion of various emotions is obtained.

[0093] The above embodiments are only used to illustrate the technical method of the present application and are not limited. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical method of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical method of the present application.

Claims

1. A method for optimizing and extracting EEG features based on a deep learning model, characterized in that, Includes the following steps: Step S1: Set up an EEG acquisition channel to collect several raw emotional EEG signals under various emotional stimuli, and then standardize the raw emotional EEG signals to obtain standard emotional EEG data. Step S2: Set up a feature extraction network. The feature extraction network is equipped with multiple feature extraction kernels. Multiple initial feature extraction points are randomly selected from various standard emotional EEG data. Then, the feature extraction kernels spread outward from the initial feature extraction points to extract detection signal segments. The detection signal segments of different standard emotional EEG data are matched with each other to generate emotional signal segments. Step S3: Record each emotional signal segment as a feature emotional signal segment. Through the feature emotional signal segments, traverse several feature segments in each standard emotional EEG data, obtain the spatiotemporal correlation weight of each feature segment relative to its standard emotional EEG data, and then establish a multidimensional feature spatiotemporal matrix under various emotional stimuli. Step S4: Establish the emotion association tree corresponding to the multidimensional feature spatiotemporal matrix, and then match and connect the emotion association trees under different emotional stimuli to obtain the full emotion feature forest. Convert the newly generated standard emotion EEG data into a detection feature submatrix sequence and input it into the full emotion feature forest, and then output the emotion proportion associated with the corresponding standard emotion EEG data. The process of matching signal segments from different standard emotional EEG data includes: The detection signal segments of standard emotional EEG data under different emotional stimuli are matched with each other. If there are two detection signal segments with a similarity of more than 90%, the initial feature extraction points corresponding to the two detection signal segments are removed, and a new initial feature extraction point is randomly set on the corresponding standard emotional EEG data. If two detected signal segments have a similarity of 90% or less, then both detected signal segments are retained, and the next detection signal segment matching continues. Once all detected signal segments have been matched, the remaining detected signal segments that correspond to the same emotional stimulus but come from different standard emotional EEG data will be matched with each other, and a similarity threshold will be set. If the similarity between two detected signal segments is greater than or equal to the similarity threshold, then the two detected signal segments are assigned the same feature annotation; otherwise, no operation is performed. Each detection signal segment without feature annotation is removed from the initial feature extraction point position, and a new feature extraction point is randomly selected. For detection signal segments with feature annotation, the initial feature extraction point position is removed. Detection signal segments with twice the unit diffusion length are extracted with the initial feature extraction point as the center. If the similarity to a detection signal segment of equal or longer length is above 90% in the secondary matching result, the corresponding two detection signal segments will be reverted to a detection signal segment of one unit diffusion length, and the reverted detection signal segment will be recorded as the emotion signal segment.

2. The method for optimizing and extracting EEG features based on a deep learning model according to claim 1, characterized in that, The process of acquiring the raw emotional EEG signals includes: A 32-channel EEG acquisition device was installed on the heads of multiple subjects, and the raw emotional EEG signals of the subjects under different emotional stimuli were collected through the 32-channel EEG acquisition device.

3. The method for optimizing and extracting EEG features based on a deep learning model according to claim 2, characterized in that, The process of standardizing raw emotional EEG signals includes: Several time points were set up to obtain the mean and standard deviation of each raw emotional EEG signal. Then, the raw emotional EEG signals were standardized using the Z-score standardization method. The standardized values ​​at each time point were then concatenated to obtain the standardized emotional EEG data.

4. The method for optimizing and extracting EEG features based on a deep learning model according to claim 3, characterized in that, The process of extracting the detected signal segment includes: The feature extraction network is configured with n feature extraction kernels. Each standard emotional EEG data is simultaneously input into the feature extraction network, and then the feature extraction network randomly selects n initial feature extraction points from the standard emotional EEG data, where n is a natural number greater than 50. Set a unit diffusion length, and let each feature extraction kernel attach to the initial feature extraction point in sequence. Each feature extraction kernel extracts a detection signal segment of unit diffusion length from both ends with the initial feature extraction point it is attached to as the center.

5. The method for optimizing and extracting EEG features based on a deep learning model according to claim 4, characterized in that, Whenever a new initial feature extraction point falls within an emotional signal segment, and a detection signal segment with a one-unit diffusion length centered on the initial feature extraction point is completely within the emotional signal segment, an initial feature extraction point is permanently reduced on the corresponding standard emotional EEG data. The above detection signal segment matching process is repeated until there are no initial feature extraction points in the standard emotional EEG data.

6. The method for optimizing and extracting EEG features based on a deep learning model according to claim 5, characterized in that, The process of establishing the multidimensional feature spatiotemporal matrix includes: Emotional signal segments in each standard emotional EEG data are recorded as characteristic emotional signal segments. Each standard emotional EEG data is traversed through the characteristic emotional signal segments, and several characteristic segments are marked in each standard emotional EEG data according to the traversal results. The query matrix, key matrix, and value matrix of each feature segment are obtained by using a multi-head self-attention mechanism. The entire standard emotional EEG data is recorded as a brain region. Then, the spatiotemporal correlation weight between the corresponding feature segment and the brain region is obtained based on the query matrix, key matrix, and value matrix. Integrate the spatiotemporal correlation weights of the same feature segments at various occurrence times to generate the corresponding temporal weight matrix. At the same time, select the occurrence time of each feature segment when the spatiotemporal correlation weight is the maximum, and then overlap each feature segment in turn to transform it into the corresponding multidimensional feature spatiotemporal matrix.

7. The method for optimizing and extracting EEG features based on a deep learning model according to claim 6, characterized in that, The process of building a forest with all emotional characteristics: The multidimensional feature spatiotemporal matrix is ​​divided into several feature sub-matrices, and a feature tree node is set for each feature sub-matrix. At the same time, according to the spatiotemporal position of each feature sub-matrix in the multidimensional feature spatiotemporal matrix, the feature tree nodes are connected in sequence to obtain the emotion association tree under the corresponding type of emotional stimulus. The feature tree nodes in the emotion association tree under different types of emotional stimuli are matched with each other, and the feature tree nodes are connected with each other according to the matching results, so as to obtain the full emotion feature forest.

8. The method for optimizing and extracting EEG features based on a deep learning model according to claim 7, characterized in that, The process of obtaining the emotion proportions from standard emotional EEG data to be analyzed includes: Obtain standard emotional EEG data to be analyzed, traverse the standard emotional EEG data with feature segments under various emotional stimuli, and obtain feature segment sequences under various emotions based on the traversal results. Each feature segment sequence is converted into a detection feature sub-matrix sequence, and the corresponding occurrence time sequence is labeled for the detection feature sub-matrix. The detection feature sub-matrix sequence is then input into the full sentiment feature forest. Starting from the first detection feature submatrix, based on the order of occurrence, synchronous comparisons are performed with each emotion association tree in the full emotion feature forest. Then, the occurrence order of each detection feature submatrix is ​​statistically analyzed, and the corresponding temporal weight matrix is ​​retrieved to obtain the temporal weight values ​​of each detection feature submatrix in different emotion association trees. The total temporal weights of the feature submatrix sequence in different emotion association trees are statistically detected. The total temporal weights of different emotion association trees are accumulated, and the proportion of each emotion is obtained.

Images