A method, device and system for electroencephalogram data feature selection and classification based on fisher criterion regularization
By introducing Fisher's criterion regularization and the nearest neighbor gradient method into the sparse regularization model, the LASSO model is improved, which solves the problem of insufficient separability in feature selection and improves the classification accuracy of EEG data.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUILIN UNIV OF ELECTRONIC TECH
- Filing Date
- 2026-03-24
- Publication Date
- 2026-06-19
AI Technical Summary
Existing sparse regularization models, while able to select important features in EEG data feature selection, neglect the separability of features, thus affecting classification performance.
The LASSO model is improved by adopting Fisher's criterion regularization technique. Combining the importance and separability of features, a new sparse regularized feature selection model is constructed, and the optimal feature weights are solved by the nearest neighbor gradient method.
The classification accuracy of EEG data was improved by selecting the most discriminative important space-frequency features, optimizing the separability of features, and thus improving the classification results.
Smart Images
Figure CN122241369A_ABST
Abstract
Description
Technical Field
[0001] This application relates to, but is not limited to, the field of data processing technology, and in particular to a method, apparatus, and system for selecting and classifying EEG data features based on Fisher's criterion regularization. Background Technology
[0002] The space-frequency feature selection method based on F-score and mutual information algorithms first uses the F-score algorithm to select the optimal frequency band through grid search, and then uses the mutual information algorithm to select the space-frequency features of the optimal frequency band. Although space-frequency feature selection methods have achieved good results, they still require dedicated training and testing of classifiers to verify the effectiveness of the selected feature subsets, resulting in high computational costs. In contrast, the space-frequency feature selection method based on sparse regularization integrates feature selection and classifier functions, enabling simultaneous feature selection and classification.
[58] This simplifies the processing steps of EEG data, resulting in higher decoding accuracy. For example, LASSO can be used to jointly select the subject-specific optimal frequency band and discriminant space-frequency features. Currently, the LASSO method is the most common and effective space-frequency feature selection method.
[59] Its adoption Norm convex regularization is widely used in engineering to achieve sparse feature selection. However, sparse regularization models such as LASSO only focus on selecting important features and do not give much consideration to the separability of features, which may affect the final feature classification results.
[0003] To improve the feature separability of sparse regularized models, this application utilizes Fisher's criterion regularization technique to improve the LASSO model, proposing a novel sparse regularized feature selection model. The proposed method jointly considers the importance and separability of features during the feature selection process, selecting the most discriminative and important space-frequency features. Summary of the Invention
[0004] The following is an overview of the subjects described in detail herein. This overview is not intended to limit the scope of the claims.
[0005] This application provides a method for selecting and classifying EEG data features based on Fisher's criterion regularization to improve classification accuracy.
[0006] In a first aspect, embodiments of this application provide a method for selecting and classifying EEG data features based on Fisher's criterion regularization, comprising the following steps:
[0007] S100, Construct a space-frequency feature selection and classification model based on Fisher's criterion regularization;
[0008] The Fisher criterion transforms two classes of sample data by finding or constructing a set of projection vectors, aiming to minimize the distance between samples of the same type (i.e., intra-class distance) and maximize the distance between samples of different types (i.e., inter-class distance). Its objective function is shown below:
[0009] ;
[0010] in and Let represent the between-class scatter matrix and the within-class scatter matrix of the sample feature matrix, respectively. The specific calculations are as follows:
[0011] ;
[0012] in Indicates belonging to The class of Given a sample vector, this application studies a binary classification problem, therefore . This represents the sample mean vector of this class. This represents the total number of samples in this class.
[0013] The maximization form can be transformed into maximization. Minimize at the same time The new form enables maximum separation of features between different classes of samples after projection. The maximization form can be transformed into the following minimization form:
[0014] ;
[0015] The sample feature matrix was After projection, the separability between features is significantly improved, with features of the same class becoming more tightly packed and features of different classes becoming more sparse. Therefore, based on the separability properties of Fisher's criterion regularization, a new form of Fisher's criterion is derived. As a new regularization term, a new sparse regularized feature selection model is obtained. The projection vector is then used... Unified representation as The mathematical model for the new method is as follows:
[0016] ;
[0017] in , , The new model, while selecting important features, also considers the separability of the selected features, which is beneficial for subsequent feature classification.
[0018] S200 solves a minimization problem based on the nearest neighbor gradient to obtain the optimal feature weights. The new model can be decomposed into smooth and non-smooth parts, i.e.
[0019] ;
[0020] in Indicates the smooth part. Indicates the non-smooth portion;
[0021] The optimal feature weights are obtained by solving the smooth and non-smooth parts of the new model separately using the nearest neighbor gradient method. For smooth parts Perform gradient descent, and for the non-smooth parts... Solve using the nearest neighbor operator. Details are as follows:
[0022] Gradient descent steps. Given an iteration point. Define an intermediate point :
[0023] ;
[0024] in Indicates the step size. express At point The gradient is calculated as follows:
[0025] ;
[0026] Steps for solving the nearest neighbor operator, including the calculation of non-differentiable functions. The nearest neighbor operator, i.e.
[0027] ;
[0028] in Representing non-smooth functions The proximity operator is defined as follows:
[0029] ;
[0030] From the above equation, it can be seen that the goal of the proximity operator is to find a point of departure. Points that are not too far apart, and make the non-smooth function values It is also relatively small. Non-smooth function For the Log norm penalty, the specific process of calculating its neighboring operator is given below:
[0031] ;
[0032] S300 obtains the optimal feature weights using the nearest neighbor gradient method. Then, through optimal feature weights Select the spatial-frequency features (i.e., significant spatial-frequency features) with non-zero weights from the feature set for classification, and thus obtain the classification result.
[0033] Secondly, embodiments of this application also provide an apparatus for selecting and classifying EEG data features based on Fisher's criterion regularization, wherein the EEG data classification apparatus based on Fisher's criterion regularization includes:
[0034] The first module is a space-frequency feature selection and classification model based on Fisher's criterion regularization.
[0035] The second module solves the established model based on the nearest neighbor gradient method to obtain the optimal feature weights.
[0036] The third module acquires the EEG data to be classified, selects significant space-frequency features through optimal feature weights for classification, and obtains the classification results.
[0037] Thirdly, a system for selecting and classifying EEG data features based on Fisher's criterion regularization includes: a data acquisition unit, a memory, a processor, a display, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program, implements the EEG data classification method based on Fisher's criterion regularization as described in the claims.
[0038] The embodiments of this application have the following beneficial effects: In the embodiments provided in this application, Fisher's criterion regularization and sparsity regularization are combined. The proposed method jointly considers the importance and separability of features during the feature selection process, selecting the most discriminative important space-frequency features. The model constructed in this application can improve the accuracy of classification.
[0039] Other features and advantages of this application will be set forth in the following description and will be apparent in part from the description or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the description, claims and drawings. Attached Figure Description
[0040] The accompanying drawings are used to provide a further understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.
[0041] Figure 1This is a flowchart of a method for selecting and classifying EEG data features based on Fisher's criterion regularization, provided in one embodiment of this application;
[0042] Figure 2 This is a structural diagram of a device for selecting and classifying EEG data features based on Fisher's criterion regularization, provided in one embodiment of this application.
[0043] Figure 3 This is a structural diagram of EEG data feature selection and classification based on Fisher criterion regularization provided in one embodiment of this application. Detailed Implementation
[0044] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0045] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, or the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.
[0046] refer to Figure 1 A method for feature selection and classification based on Fisher's criterion regularization. This method includes the following steps:
[0047] S100, Construct a space-frequency feature selection and classification model based on Fisher's criterion regularization;
[0048] The Fisher criterion transforms two classes of sample data by finding or constructing a set of projection vectors, aiming to minimize the distance between samples of the same type (i.e., intra-class distance) and maximize the distance between samples of different types (i.e., inter-class distance). Its objective function is shown below:
[0049] ;
[0050] in and Let represent the between-class scatter matrix and the within-class scatter matrix of the sample feature matrix, respectively. The specific calculations are as follows:
[0051] ;
[0052] in Indicates belonging to The class of Given a sample vector, this application studies a binary classification problem, therefore . This represents the sample mean vector of this class. This represents the total number of samples in this class.
[0053] The maximization form can be transformed into maximization. Minimize at the same time The new form enables maximum separation of features between different classes of samples after projection. The maximization form can be transformed into the following minimization form:
[0054] ;
[0055] The sample feature matrix was After projection, the separability between features is significantly improved, with features of the same class becoming more tightly packed and features of different classes becoming more sparse. Therefore, based on the separability properties of Fisher's criterion regularization, a new form of Fisher's criterion is derived. As a new regularization term, a new sparse regularized feature selection model is obtained. The projection vector is then used... Unified representation as The mathematical model for the new method is as follows:
[0056] ;
[0057] in , , The new model, while selecting important features, also considers the separability of the selected features, which is beneficial for subsequent feature classification.
[0058] S200 solves a minimization problem based on the nearest neighbor gradient to obtain the optimal feature weights. The new model can be decomposed into smooth and non-smooth parts, i.e.
[0059] ;
[0060] in Indicates the smooth part. Indicates the non-smooth portion;
[0061] The optimal feature weights are obtained by solving the smooth and non-smooth parts of the new model separately using the nearest neighbor gradient method. For smooth parts Perform gradient descent, and for the non-smooth parts... Solve using the nearest neighbor operator. Details are as follows:
[0062] Gradient descent steps. Given an iteration point. Define an intermediate point :
[0063] ;
[0064] in Indicates the step size. express At point The gradient is calculated as follows:
[0065] ;
[0066] Steps for solving the nearest neighbor operator, including the calculation of non-differentiable functions. The nearest neighbor operator, i.e.
[0067] ;
[0068] in Representing non-smooth functions The proximity operator is defined as follows:
[0069] ;
[0070] From the above equation, it can be seen that the goal of the proximity operator is to find a point of departure. Points that are not too far apart, and make the non-smooth function values It is also relatively small. Non-smooth function For the Log norm penalty, the specific process of calculating its neighboring operator is given below:
[0071] .
[0072] S300 obtains the optimal feature weights using the nearest neighbor gradient method. Then, through optimal feature weights Select the spatial-frequency features (i.e., significant spatial-frequency features) with non-zero weights from the feature set for classification, and thus obtain the classification result.
[0073] In the embodiments provided in this application, Fisher's criterion regularization and sparse regularization are combined. The proposed method jointly considers the importance and separability of features during the feature selection process, selecting the most discriminative important space-frequency features. The model constructed in this application can improve the accuracy of classification.
[0074] The following is a comparative experiment on EEG classification based on the model proposed in the embodiments of this application, that is, a comparison with classic and commonly used algorithms. The high accuracy and feasibility of the new model are verified.
[0075] In some embodiments, an acquisition system is used to acquire electroencephalogram (EEG) data. In some embodiments, after the raw classification data is determined, the EEG data to be classified is loaded.
[0076] As shown in Table 1, the results of different classification models are presented. It can be seen that the new model described above outperforms other methods in terms of classification results.
[0077] Table 1. Classification accuracy / %
[0078] Subjects LASSO gLASSO sgLASSO New method S01 63.33 66.67 61.67 76.67 S02 78.33 78.33 65.00 86.67 S03 100.00 98.33 100.00 96.67 S04 80.00 85.00 81.67 85.00 S05 51.67 50.00 53.33 70.00 S06 66.67 66.67 63.33 83.33 S07 91.67 90.00 93.33 88.33 S08 80.00 71.67 66.67 86.67 S09 96.67 96.67 96.67 96.67 S10 56.67 56.67 56.67 63.33 S11 80.00 76.67 76.67 71.67 S12 70.00 66.67 71.67 71.67 S13 60.00 61.67 63.33 53.33 S14 48.33 45.00 53.33 61.67 Mean ± Std 73.10 ± 15.71 72.14 ± 15.80 71.67±15.18 77.98±12.73
[0079] Additionally, refer to Figure 2 In some embodiments, an apparatus for feature selection and classification based on Fisher criterion regularization is also provided. An apparatus for feature selection and classification based on Fisher criterion regularization includes:
[0080] The first module is a space-frequency feature selection and classification model based on Fisher's criterion regularization.
[0081] The second module solves the established model based on the nearest neighbor gradient method to obtain the optimal feature weights.
[0082] The third module acquires the EEG data to be classified, selects significant space-frequency features through optimal feature weights for classification, and obtains the classification results.
[0083] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.
[0084] Additionally, refer to Figure 3 An embodiment of this application also provides a Fisher criterion-regularized feature selection and classification system, the system comprising: a memory 11, a processor 12, and a computer program stored on the memory 11 and executable on the processor 12.
[0085] The processor 12 and the memory 11 can be connected via a bus or other means.
[0086] The non-transient software program and instructions required to implement a feature selection and classification method based on Fisher criterion regularization in the above embodiments are stored in memory 11. When executed by processor 12, a feature selection and classification method based on Fisher criterion regularization in the above embodiments is executed.
[0087] The above is a detailed description of the preferred embodiments of this application. However, this application is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.
Claims
1. A method of feature selection and classification based on Fisher's criterion regularized, characterized in that, Includes the following steps: S100, Construct a space-frequency feature selection and classification model based on Fisher's criterion regularization; S200 uses the nearest neighbor gradient method to solve the minimization problem and obtain the optimal feature weights. S300 obtains the optimal feature weights using the nearest neighbor gradient method. Then, through optimal feature weights Select the spatial-frequency features (i.e., significant spatial-frequency features) with non-zero weights from the feature set for classification, and thus obtain the classification result.
2. The feature selection and classification method based on Fisher's criterion regularization according to claim 1, characterized in that: In S100, a space-frequency feature selection and classification model based on Fisher's criterion regularization is constructed, including: The Fisher criterion transforms two classes of sample data by finding or constructing a set of projection vectors, aiming to minimize the distance between samples of the same type (i.e., intra-class distance) and maximize the distance between samples of different types (i.e., inter-class distance). Its objective function is shown below: ； wherein and denote the between-class scatter matrix and the within-class scatter matrix of the sample feature matrix, respectively, and are calculated as follows: ； wherein represents the first sample vector belonging to the class , the present application studies the binary classification problem, therefore . represents the sample mean vector of the class, is the total number of samples of the class, The maximization form can be transformed into a maximization while minimizing a new form of The maximization form of can be transformed into a minimization form of ； The sample feature matrix is After projection, the separability between features is well optimized and strengthened, and features of the same class are closer, while features of different classes are more sparse. Therefore, based on the separation characteristics of the Fisher criterion regularization, a new form of Fisher criterion is converted as a new regularization term, thereby obtaining a new sparse regularization feature selection model. The projection vector is uniformly represented as The mathematical model of the new method is as follows: ； wherein , , The new model can consider the separability of the selected features while selecting important features, which is conducive to subsequent feature classification.
3. The feature selection and classification method based on Fisher's criterion regularization according to claim 2, characterized in that: In S200, the minimization problem is solved using the nearest neighbor gradient method to obtain the optimal feature weights. The new model can be decomposed into smooth and non-smooth parts, i.e.: ； wherein represents a smooth portion, represents a non-smooth portion; The smooth and non-smooth parts of the new model are solved by using the proximal point gradient method, so as to obtain the optimal feature weight . The gradient descent is performed for the smooth part , and the proximal operator is used to solve the non-smooth part . The specific process is as follows: Gradient descent step. Given an iteration point , define an intermediate point : ； wherein represents a step size, represents At the point of the gradient, is calculated as follows: ； a step of solving the proximity operator, computing the proximity operator of a non-differentiable function a step of solving the proximity operator, namely ； wherein denotes the proximal operator of the non-smooth function , defined as follows: ； From the above equation, it can be seen that the goal of the proximity operator is to find a point of departure. Points that are not too far apart, and make the non-smooth function values It is also relatively small. Non-smooth function For the Log norm penalty, the specific process of calculating its neighboring operator is given below: 。 4. The feature selection and classification method based on Fisher criterion regularization according to claim 3, characterized in that: In S300, the optimal feature weights are obtained using the nearest neighbor gradient method. Then, through optimal feature weights Select the spatial-frequency features (i.e., significant spatial-frequency features) with non-zero weights from the feature set for classification, and thus obtain the classification result.
5. A feature selection and classification device based on Fisher's criterion regularization, characterized in that, The feature selection and classification device based on Fisher criterion regularization includes: The first module constructs a space-frequency feature selection and classification model based on Fisher's criterion regularization. The second module solves the minimization problem based on the nearest neighbor gradient method to obtain the optimal feature weights. The third module obtains the optimal feature weights using the nearest neighbor gradient method. Then, through optimal feature weights Select the spatial-frequency features (i.e., significant spatial-frequency features) with non-zero weights from the feature set for classification, and thus obtain the classification result.
6. A feature selection and classification system based on Fisher's criterion regularization, comprising: A data collector, a memory, a processor, a display, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the computer program, it implements a feature selection and classification method based on Fisher criterion regularization as described in any one of claims 1 to 4.