Inplausible electroencephalogram emotion recognition method based on fuzzy interpolation reasoning

Abstract

The invention provides an interpretable electroencephalogram emotion recognition method based on fuzzy interpolation reasoning. According to the interpretable electroencephalogram emotion recognition method based on fuzzy interpolation reasoning, uncertainty in electroencephalogram emotion recognition modeling is processed through the fuzzy technology, inaccurate emotion expression forms are described, and fuzziness of emotion category boundaries is processed. By constructing an attribute weighted fuzzy interpolation reasoning model, the problem of weighted approximate reasoning on incomplete knowledge caused by high-dimensional small sample electroencephalogram signals is solved, so that the recognition precision is effectively improved. And interpretability analysis is carried out on the emotion recognition process through a fuzzy rule, so that the problem that the existing emotion recognition process is lack of interpretability is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of electroencephalogram (EEG) emotion recognition, and particularly to an interpretable EEG emotion recognition method based on fuzzy interpolation reasoning. Background Art

[0002] Electroencephalogram (EEG) is a signal directly measured from neuron activities, with high time resolution and difficult to be faked, which can provide reliable information for emotion state recognition. In addition, EEG acquisition devices are non-invasive to subjects and relatively low in cost. Therefore, EEG-based emotion recognition methods have become a research hotspot in recent years.

[0003] EEG-based emotion recognition methods can be roughly divided into two categories. One is to use classification methods of machine learning by extracting features, and representative methods include LDA, SVM, MLP, KNN, and AdaBoost. The other is based on the end-to-end deep learning technology that has received wide attention in recent years, such as deep network models based on structures like LSTM-RNN, DBN, and CNN. The introduction of deep learning technology is to use a large amount of EEG data to improve the recognition accuracy. However, deep learning methods generally have the "black box" problem, resulting in the lack of interpretability in the emotion recognition process. Methods based on fuzzy reasoning can improve the interpretability of the model because they use semantically readable fuzzy rules. However, most methods fail to clearly distinguish the importance differences of multi-dimensional features in constructing a fuzzy system for emotion classification, resulting in inaccurate results obtained by fuzzy reasoning methods. Moreover, emotion EEG signals have the characteristics of high-dimensional and small samples. The fuzzy rules learned from EEG data for fuzzy reasoning may be incomplete and unable to fully cover the entire domain of the problem. Therefore, the recognition accuracy of fuzzy reasoning technology is relatively low.

[0004] Therefore, the methods of EEG emotion recognition in the prior art have problems of insufficient recognition accuracy and difficulty in explaining the emotion recognition process. Summary of the Invention

[0005] Aiming at the problems existing in the prior art, the present invention provides an interpretable EEG emotion recognition method based on fuzzy interpolation reasoning to solve the technical problems of insufficient recognition accuracy and difficulty in explaining the emotion recognition process in the EEG emotion recognition methods of the prior art.

[0006] The present invention provides an interpretable EEG emotion recognition method based on fuzzy interpolation reasoning, including: S1. Collect the original EEG signal, preprocess it, and extract the original EEG feature samples of the original EEG signal; S2. Perform feature selection on the original EEG features and calculate the weights of the original EEG features; S3. Adaptively generate a fuzzy rule base based on the original EEG features; S4. Use the fuzzy interpolation inference method with attribute weighting to perform emotion recognition on EEG features in combination with the fuzzy rule base.

[0007] Optionally, the preprocessing of it to extract the original EEG feature samples of the original EEG signal includes at least: Perform preprocessing of artifact removal, downsampling, and band-pass filtering on the original EEG signal, and extract the original EEG feature samples for each original EEG signal.

[0008] Optionally, feature selection is performed on the original EEG features, and the weights of the EEG features are calculated, including: Use the Relief-F algorithm to score each of the original EEG features, sort them in descending order according to the scores, select the top m original EEG features, and calculate the weights of each original EEG feature, expressed as: , where is the score of each original EEG feature.

[0009] Optionally, the adaptively generating a fuzzy rule base based on the original EEG features includes: S301. Combine each of the original EEG features and its corresponding emotion label to form an EEG sample, where the original EEG feature is expressed as , and the emotion label is expressed as , and use the original EEG feature as the antecedent attribute of the fuzzy rule and the emotion label as the consequent attribute of the fuzzy rule; S302. Use the Fuzzy C-Means algorithm to learn the fuzzy partition of the feature space, and define the fuzzy partition of the domain of the antecedent attribute as , where represents the total number of fuzzy values that the antecedent attribute can take, and the value of the antecedent attribute is defined as , where represents the fuzzy value that the antecedent attribute can take; S303. Construct the fuzzy region space FRS of the antecedent attribute; S304. Put all the EEG samples into the fuzzy region space FRS, and select a grid in the fuzzy region space FRS to match it according to the value of the antecedent attribute of each EEG sample; S305. Execute step S304 until all the EEG samples are processed, and select the grid with the most matching data among all the EEG samples, and mark the matching data as Max; S306. Determine the sizes of Max and the threshold. If , extract a fuzzy rule from the grid corresponding to Max, delete all EEG samples in Max, and update all EEG samples to . Among them, represents all EEG samples in Max; S307. Repeat steps S304 - S306 until is not satisfied, and form a fuzzy rule base with the learned fuzzy rules, denoted as : , Among them, represents a fuzzy rule, represents the fuzzy rule of the antecedent attribute taken fuzzy set value, represents the fuzzy rule of the consequent attribute fuzzy set value.

[0010] Optionally, the combined fuzzy rule base uses the attribute - weighted fuzzy interpolation reasoning method to perform emotion recognition on EEG features, including: S401. Represent the observed value after fuzzifying the original EEG feature as: , Among them, represents the observed value of the antecedent attribute taken fuzzy set value. Denote the value of the consequent attribute corresponding to this observed value as ; S402. Calculate the weighted distance between the observed value and each fuzzy rule , , Among them, represents the distance between a pair of antecedent fuzzy sets, is calculated from the three vertices of the triangular fuzzy membership function . and respectively represent the maximum and minimum values of the antecedent attribute ; S403. Select the first n fuzzy rules with the shortest distances, denoted as ; S404. Represent the weight attached to the j-th antecedent attribute of the i-th fuzzy rule in for constructing the j-th antecedent of the fuzzy intermediate rule as , which is calculated by and normalized to obtain ; ; S405. Represent the j-th antecedent of the fuzzy intermediate rule as follows: , , where is a constant and is the offset factor of ; S406. Represent the consequent value of the fuzzy intermediate rule as: , where and are the maximum and minimum values of the result attribute, and and are obtained from the following formula: ; S407. For each antecedent attribute of the fuzzy intermediate rule, calculate the scale factor of the antecedent attribute through the fuzzy set ; S408. Calculate the fuzzy set of the antecedent attribute after scale change: : ; S409. Calculate the displacement factor of each antecedent attribute of the fuzzy intermediate rule, which is represented as: ; S410. Calculate the scale transformation factor and displacement transformation factor of the consequent attribute, which are represented as: ; S411. According to the consequent attribute of the fuzzy intermediate rule and the scale transformation factor , calculate the fuzzy set of the consequent attribute after scale transformation: : ; S412. Calculate the final required interpolation result using the displacement factor of the conclusion attribute, expressed as: : , , , , S413. Defuzzify the interpolation result to obtain the emotion recognition result.

[0011] Optionally, it further includes: S5. Use fuzzy rules to perform interpretability analysis on the emotion recognition process.

[0012] Optionally, the use of fuzzy rules to perform interpretability analysis on the emotion recognition process includes: S501. Extract the first n fuzzy rules with the shortest distances in step S403; S502. Map the fuzzy semantic values of the EEG features to the physical meanings they actually represent one by one; S503. Output the interpretability expression of the fuzzy rules in step S501.

[0013] Compared with the prior art, the present invention: The present invention uses fuzzy technology to handle the uncertainty in EEG emotion recognition modeling, characterize imprecise emotion expression forms, and handle the fuzziness of emotion category boundaries. By constructing a fuzzy interpolation inference model with attribute weighting, it overcomes the problem of weighted approximate reasoning on incomplete knowledge caused by high-dimensional small-sample EEG signals, thereby effectively improving the recognition accuracy. And through the interpretability analysis of the emotion recognition process using fuzzy rules, it overcomes the problem of the lack of interpretability in the existing emotion recognition process. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The accompanying drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present invention and used together with the specification to explain the principles of the present invention.

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for use in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, other accompanying drawings can also be obtained based on these drawings without creative efforts.

[0016] Figure 1 is a schematic flowchart of the method of the present invention; Figure 2 is a schematic diagram of the interpretability analysis of the EEG emotion recognition process in the present invention. Detailed implementation manners

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Apparently, the described embodiments are some but not all of the embodiments of the present application. All other implementation cases obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts fall within the scope of protection of the present application. In the embodiments of the present invention, functional units with the same reference numerals have the same and similar structures and functions.

[0018] See Figure 1 , the present invention provides an interpretable electroencephalogram (EEG) emotion recognition method based on fuzzy interpolation reasoning, including: S1. Collect the original EEG signals, preprocess them, and extract the original EEG feature samples of the original EEG signals; S2. Perform feature selection on the original EEG features and calculate the weights of the original EEG features; S3. Adaptively generate a fuzzy rule base based on the original EEG features; S4. Combine the fuzzy rule base and use the attribute-weighted fuzzy interpolation reasoning method to perform emotion recognition on the EEG features.

[0019] In this embodiment, in S1, the original EEG signals are collected, preprocessed, and the original EEG feature samples of the original EEG signals are extracted.

[0020] First, the original EEG signals are collected, and then they are preprocessed such as artifact removal, downsampling, and band-pass filtering, and 5 types of features are extracted for each EEG sample: a. Higher Order Crossings (HOC), b. Fractal Dimension (FD), c. Non-Stationary Index (NSI), d. Complexity of Hjorth Features, e. Power Spectral Density (PSD).

[0021] S2. Perform feature selection on the original EEG features and calculate the weights of the original EEG features.

[0022] First, the Relief-F algorithm needs to be used to score each of the original EEG features to indicate the relative importance of each feature in the process of distinguishing emotional states, sort them in descending order according to the scores, and select the top m original EEG features, and calculate the weights of each original EEG feature , expressed as: , wherein, is the score for each original EEG feature.

[0023] S3. Adaptively generate a fuzzy rule base based on the original EEG features.

[0024] S301. After step S2, combine each of the original EEG features and its corresponding emotion label to form an EEG sample, wherein the original EEG feature is represented as , the emotion label is represented as , and use the original EEG feature as the antecedent attribute of the fuzzy rule and the emotion label as the consequent attribute of the fuzzy rule.

[0025] S302. Use Fuzzy C-Means algorithm to learn the fuzzy partition of the feature space, and obtain the fuzzy membership function and its semantic label corresponding to each feature, that is, obtain (1) the fuzzy partition of the domain of the antecedent attribute , wherein represents the total number of fuzzy values that the attribute can take, (2) the value of the antecedent attribute, wherein represents the fuzzy value that the attribute can take; In addition, all fuzzy values in the present invention select triangular fuzzy membership functions, and the description of the algorithm in subsequent steps also adopts the form of triangular fuzzy membership functions. However, the technology of the present invention can use other forms besides triangular fuzzy membership functions, such as common trapezoidal and Gaussian fuzzy membership functions, etc.

[0026] S303. Construct the fuzzy region space FRS of the antecedent attribute. FRS is a -dimensional hypercube.

[0027] S304. Input all EEG samples into the fuzzy region space FRS, and select a grid in the fuzzy region space FRS to match it according to the value of the antecedent attribute of each EEG sample.

[0028] S305. Execute step S304 until all EEG samples are processed, and select the grid with the most matching data among all EEG samples, and mark the matching data as Max .

[0029] S306. Judge the size of Max and the threshold . If , then extract a fuzzy rule from the grid corresponding to Max , and delete MaxAll the EEG samples in are updated to , where represents Max all the EEG samples in

[0030] S307. Repeat steps S304 - S306 until is not satisfied, and form a fuzzy rule base with the learned fuzzy rules, denoted as : , where represents a fuzzy rule, represents the fuzzy rule 's antecedent attribute 's taken fuzzy set value, represents the fuzzy rule 's consequent attribute 's fuzzy set value.

[0031] S4. Use the attribute - weighted fuzzy interpolation reasoning method to perform emotion recognition on EEG features in combination with the fuzzy rule base.

[0032] S401. Represent the observed value after fuzzifying the original EEG feature as: , where represents the observed value 's antecedent attribute 's taken fuzzy set value, and denote the value of the consequent attribute corresponding to this observed value as ; S402. Calculate the weighted distance between the observed value and each fuzzy rule , , where represents the distance between a pair of antecedent fuzzy sets, is calculated from the three vertices of the triangular fuzzy membership function , and represent the maximum and minimum values of the antecedent attribute respectively.

[0033] S403. Select the first n fuzzy rules with the shortest distances, denoted as .

[0034] S404. Take The j-th antecedent attribute of the i-th fuzzy rule in is the j-th antecedent for constructing the fuzzy intermediate rule The attached weight is expressed as , which is calculated by and normalized to obtain ; S405. Represent the j-th antecedent of the fuzzy intermediate rule as follows: , , where is a constant and is the offset factor of ; S406. The consequent value of the fuzzy intermediate rule is obtained by the following formula: ; S407. For each antecedent attribute of the fuzzy intermediate rule, calculate the scale factor of the antecedent attribute through the fuzzy set ; S408. Calculate the fuzzy set of the antecedent attribute after scale change: : ; S409. Calculate the displacement factor of each antecedent attribute of the fuzzy intermediate rule, which is expressed as: ; S410. Calculate the scale transformation factor and the displacement transformation factor of the consequent attribute, which are expressed as: ; S411. According to the consequent attribute of the fuzzy intermediate rule and the scale transformation factor , calculate the fuzzy set of the consequent attribute after scale transformation: : ; S412. Use the displacement factor of the consequent attribute to calculate the final required interpolation result, which is expressed as: : , , ​​ ; S413. Defuzzify the interpolation result to obtain the emotion recognition result.

[0035] Finally, the obtained emotion recognition result may include emotion states ("Positive", "Neural", "Negative").

[0036] The present invention uses fuzzy technology to handle the uncertainty in electroencephalogram (EEG) emotion recognition modeling, characterize imprecise emotion expression forms, and handle the fuzziness of emotion category boundaries. By constructing an attribute-weighted fuzzy interpolation inference model, it overcomes the problem of weighted approximate reasoning on incomplete knowledge caused by high-dimensional small-sample EEG signals, thereby effectively improving the recognition accuracy.

[0037] In another embodiment, refer to Figure 2 , the present invention performs interpretability analysis on the EEG emotion recognition result process using fuzzy rules.

[0038] S501. Extract the first n fuzzy rules with the shortest distances in step S403. For example, one rule is as follows: If NSI is Medium and HOC is Large, then is Positive. S502. Map the fuzzy semantic values of the EEG features to their actual physical meanings one by one. Map the fuzzy semantic values of the EEG features (such as "Small,..., Medium,..., Large") to their actual physical meanings one by one. Taking the selected EEG features NSI and HOC as examples, where NSI is a measure of signal complexity, and the higher the index value, the more inconsistent the average value. The HOC feature describes the oscillation pattern of the EEG signal. Therefore, the fuzzy semantic values after mapping of the EEG feature NSI can take values: {consistent,..., less inconsistent,..., inconsistent}, and the fuzzy semantic values after mapping of HOC can take values: {gentle,..., slight,..., severe}; S503. Output the interpretability expression of the fuzzy rules in step S501.

[0039] If Signal average (NSI) is Less inconsistent and Signal oscillation (HOC) is Severe, then is Positive. This embodiment gives full play to the semantic readability of the intuitive fuzzy if-then rules, explains the emotional classification process, and provides important assistance for applications such as realizing transparent human-computer interaction, interpretable computer-aided diagnosis of mental diseases, and further analyzing the association between emotions and brain region functions. It also provides a new idea for interpretable emotion recognition. Moreover, the above interpretable fuzzy inference rules are used to illustrate the process of inferring the emotion recognition result from the information contained in the input EEG signals, making the EEG emotion recognition more interpretable and the recognition process more transparent.

[0040] It should be noted that in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the existence of additional identical elements in the process, method, article or device comprising the element.

[0041] The above are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather to the broadest scope consistent with the principles and novel features claimed herein.

Claims

1. An interpretable EEG emotion recognition method based on fuzzy interpolation reasoning, characterized in that: include: S1, collecting original EEG signals, and preprocessing them to extract original EEG feature samples of the original EEG signals; S2. performing feature selection on the original EEG features and calculating weights of the original EEG features; S3, adaptively generating a fuzzy rule base based on the original EEG features; S4. Combined with the fuzzy rule base, the attribute-weighted fuzzy interpolation reasoning method is used to perform emotion recognition on EEG features.

2. The interpretable EEG emotion recognition method based on fuzzy interpolation reasoning as claimed in claim 1, characterized in that: The preprocessing to extract the original EEG feature samples of the original EEG signal at least includes: The original EEG signals are preprocessed by removing artifacts, downsampling, and bandpass filtering, and original EEG feature samples are extracted from each original EEG signal.

3. The interpretable EEG emotion recognition method based on fuzzy interpolation reasoning as claimed in claim 1, characterized in that: The original EEG features are subjected to feature selection, and the weights of the EEG features are calculated, including: The Relief-F algorithm is used to score each of the original EEG features, and the scores are sorted from large to small, and the top m The original EEG features are calculated and the weight of each original EEG feature is expressed as: ， in, For each raw EEG feature score, is the score of the Relief-F algorithm for each of the raw EEG features.

4. The interpretable EEG emotion recognition method based on fuzzy interpolation reasoning as claimed in claim 1, characterized in that: The adaptive generation of a fuzzy rule base based on the original EEG features comprises: S301, combining each of the original EEG features and its corresponding emotion label to form an EEG sample, wherein the original EEG feature is represented as , the sentiment label is represented as , and taking the original EEG features as the antecedent attributes of the fuzzy rules, and the emotional labels as the conclusion attributes of the fuzzy rules; S302, Utilization Fuzzy C-Means The algorithm learns the fuzzy partitioning of the feature space and defines the fuzzy partitioning of the antecedent attribute domain as ,in Represents the antecedent attribute The total number of possible fuzzy values, the value of the antecedent attribute is defined as ,in Represents the antecedent attribute Desirable fuzzy value; S303, constructing the fuzzy regional space FRS of antecedent attributes; S304, all EEG samples Input the fuzzy regional space FRS, and select a grid in the fuzzy regional space FRS to match it according to the value of the antecedent attribute of each EEG sample; S305, executing step S304 until all EEG samples are processed, and selecting a grid with the most matching data among all EEG samples, marking the matching data as m ax ; S306, determine m ax and threshold The size of , then from m ax Extract a fuzzy rule from the corresponding grid and delete m ax All EEG samples in Updated to ,in, Indicates m ax All EEG samples in; S307, repeat steps S304-S306 until the condition is not satisfied , and the fuzzy rules learned form a fuzzy rule base, expressed as : ， in, represents fuzzy rules, Representing fuzzy rules The antecedent property The fuzzy set value taken, Representing fuzzy rules The conclusion attribute The fuzzy set value of .

5. The interpretable EEG emotion recognition method based on fuzzy interpolation reasoning as claimed in claim 4, characterized in that: The method of using the attribute-weighted fuzzy interpolation reasoning method in combination with the fuzzy rule base to perform emotion recognition on EEG features includes: S401, the original EEG features The fuzzified observations are expressed as: ， in, Represents the observed value The antecedent property The fuzzy set value taken will correspond to the observed value The conclusion attribute The value of ; S402, calculate observation value And each fuzzy rule The weighted distance between , ， in, represents the distance between a pair of antecedent fuzzy sets, By triangular fuzzy membership function The three vertices are calculated, and Represents the antecedent attributes The maximum and minimum values ​​of ; S403, select the first n fuzzy rules with the shortest distance, denoted as ; S404, The jth antecedent attribute pair of the i-th fuzzy rule in constructs the jth antecedent of the fuzzy intermediate rule The added weight is expressed as ,Depend on Calculate and normalize it to get ; S405, blurring the middle rule The jth antecedent of It is expressed as follows: ， ， in, is a constant, The offset factor of S406, the consequent value of the fuzzy intermediate rule is expressed as: ， in, and are the maximum and minimum values ​​of the result attribute, and It is derived from the following formula: ： S407. For each antecedent attribute of the fuzzy intermediate rule , through fuzzy sets , calculate the scale factor of the antecedent attribute ; S408. Calculate the fuzzy set of the antecedent attribute after the scale change: ： ； S409, calculate each antecedent attribute of the fuzzy intermediate rule The displacement factor is expressed as: ； S410, calculate the scale transformation factor of the conclusion attribute and displacement transformation factor , expressed as: ； S411. Conclusion attributes based on fuzzy intermediate rules and the scaling factor , calculate the fuzzy set after the conclusion attribute scale transformation: ： ； S412, using the displacement factor of the conclusion attribute to calculate the final required interpolation result, expressed as: ： ， ， ， ； S413: Defuzzify the interpolation result to obtain an emotion recognition result.

6. The interpretable EEG emotion recognition method based on fuzzy interpolation reasoning as claimed in claim 5, characterized in that: Also includes: S5. Use fuzzy rules to perform explainability analysis on the emotion recognition process.

7. The interpretable EEG emotion recognition method based on fuzzy interpolation reasoning as claimed in claim 6, characterized in that: The process of using fuzzy rules to perform interpretability analysis on the emotion recognition process includes: S501, extracting the first n fuzzy rules with the shortest distance in step S403; S502, mapping the fuzzy semantic value of the EEG feature with the actual physical meaning represented one by one; S503: Output the interpretable expression of the fuzzy rule in step S501.

Images