Maximum information coefficient-based feature selection method for image classification

Abstract

The invention discloses a maximum information coefficient-based feature selection method and system for image classification, electronic equipment, a storage medium and a program, and belongs to the technical field of electronic information. The method comprises the following steps: acquiring an image data set to be subjected to image classification; calculating a redundancy matrix and a maximum information coefficient according to the image data set; constructing a target function model based on the redundancy matrix and the maximum information coefficient; performing feature calculation on the image data set by using the target function model to obtain a feature score; sorting the feature scores according to a set condition, and selecting target features to form a feature subset; and training an image classifier by using the feature subset, and inputting a test set of the image data set into the trained image classifier to obtain a classification result. According to the method, the features with high correlation with the target are selected by maximizing the mutual information between the features and the classification information, unnecessary features are reduced by controlling the redundancy between the features, the feature redundancy and the new classification information are balanced, and the picture classification precision is improved.

Description

Technical Field

[0001] The present invention belongs to the field of electronic information technology, and specifically relates to a feature selection method, system, electronic equipment, storage medium and program based on maximum information coefficient for image classification. Background Art

[0002] In the field of image processing, with the development of technology, the processing and analysis of image data has become a research hotspot. However, image data often has the characteristics of limited sample number and extremely high dimension, which brings great challenges to subsequent classification and recognition tasks. In order to extract valuable information from high-dimensional image data, feature selection is widely used in the preprocessing stage of image data as an effective dimensionality reduction method.

[0003] The main purpose of feature selection is to generate a feature subspace that can reflect the essential structure of the data and improve the image classification recognition rate by removing irrelevant or redundant features. This process can not only reduce the dimension of the data and reduce the computational cost, but also maintain or improve the performance of the classifier to a certain extent. However, feature selection is not an easy task. It requires careful selection of features that can represent the essence of the data and are highly correlated with the class variable to avoid problems caused by too many or too few features.

[0004] Existing feature selection methods can be roughly divided into three types: filtering, wrapping, and embedding. Among them, the filtering method sorts features by evaluation criteria and selects features with the highest evaluation scores to form a feature subset. The evaluation criteria of this method are usually aimed at enhancing the correlation between feature variables and class variables while reducing the correlation between features. However, traditional correlation scoring methods, such as Pearson correlation coefficient, F-score, etc., although they can remove irrelevant features, have deficiencies in handling redundant features. In addition, these methods also rely on the actual values ​​of the original data set and are more sensitive to the noise of the data set.

[0005] In order to overcome these shortcomings, mutual information in information theory is widely used in feature selection to evaluate the correlation between features and classes and the redundancy between features. Mutual information can quantify the degree of correlation between two random variables, that is, the degree to which the uncertainty of one variable is reduced when the other variable is known. On this basis, the maximum information coefficient (MIC) is proposed as an estimation method of mutual information to deal with more complex nonlinear relationships, and shows good robustness in cases with small sample size and large number of features.

[0006] However, existing feature selection algorithms based on mutual information still have some defects and shortcomings. Some algorithms such as information gain (IG) and mutual information maximization (MIM) can remove irrelevant features based on the mutual information between features and class labels, but ignore the redundancy between features. While other algorithms consider the removal of redundant and irrelevant features at the same time, they have difficulties in balancing the importance of feature redundancy and new classification information. For example, the method of minimizing redundant features may select features with low redundancy but weak correlation, while the method of maximizing classification information may select features with high correlation but strong redundancy.

[0007] Therefore, how to effectively balance the importance of feature redundancy and new classification information is still a problem that needs to be solved in the current field of feature selection. Summary of the invention

[0008] Aiming at the problem that feature redundancy and new classification information cannot be balanced in image classification in the prior art, the present invention provides a feature selection method based on maximum information coefficient for image classification, which selects features with strong relevance to the target by maximizing mutual information, and reduces unnecessary features by controlling the redundancy between features, thereby balancing feature redundancy and new classification information and improving the accuracy of image classification.

[0009] In order to achieve the above object, the present invention provides the following technical solutions.

[0010] In a first aspect, the present invention provides a feature selection method based on maximum information coefficient for image classification, comprising: Obtain an image dataset to be classified; Calculate the redundancy matrix and the maximum information coefficient based on the image data set; Construct the objective function model based on the redundancy matrix and the maximum information coefficient; The objective function model is used to calculate the features of the image data set to obtain feature scores; the feature scores are sorted according to the set conditions, and the target features are selected to form feature subsets; Use the feature subset to train the image classifier, input the test set of the image dataset into the trained image classifier, and obtain the classification result.

[0011] As a further improvement of the present invention, the step of obtaining an image data set to be classified includes: For the image data set for image classification, preprocessing is performed to obtain a given data set; The given data set includes a data matrix and class labels ; The data matrix No. Column and The columns are and Features, using and express, ; is the number of samples in the dataset; is the number of features.

[0012] As a further improvement of the present invention, the step of calculating the redundancy matrix and the maximum information coefficient according to the image data set includes: The redundant matrix comprises: , , ; The maximum information coefficient includes:

[0013] in, is a redundant matrix, It is Centralized features; It is The centralizing feature, is a centralized matrix; yes The identity matrix, is a dimensional all-1 vector; is mutual information; is the data matrix; is the data matrix No. Features is the data matrix No. Features is the class label corresponding to the data; is the data matrix The number of grids to be divided; is the class label The number of grids divided, is the maximum grid size limit, is the number of samples in the dataset.

[0014] As a further improvement of the present invention, the objective function model is constructed based on the redundancy matrix and the maximum information coefficient, including: Constructing the objective function

[0015] in, is a redundant matrix, where each element is , metric features and The redundancy between It contains all feature scores. dimensional column vector, yes The feature score of yes The feature score of is the number of features, is the regularization parameter; is the class label corresponding to the data; for The maximum information coefficient between the class label; For all 1 dimensional column vector.

[0016] Under the KKT condition, solve the objective function and get the feature score :

[0017] in, is a dimension of A vector where each element is .

[0018] As a further improvement of the present invention, the method of calculating features of the image data set using the objective function model to obtain feature scores includes: Calculate the MIC value between each feature in the image data set and the known class label, and obtain the diagonal matrix MIC consisting of the maximum information coefficient of each feature and classification information; The diagonal matrix MIC composed of the maximum information coefficient is input into the objective function model to obtain the score of each feature.

[0019] As a further improvement of the present invention, the feature scores are sorted according to the set conditions, and the target features are selected to form a feature subset, including: Sort the acquired feature scores from high to low to obtain a feature score sequence; Sort the corresponding data features according to the feature score sequence; Sort by feature score from high to low, select the first Features constitute a feature subset , complete the feature selection of the image dataset.

[0020] As a further improvement of the present invention, the method of using a feature subset to train an image classifier and inputting a test set of the image data set into the trained image classifier to obtain a classification result includes: Use the feature subset to train the image classifier to obtain a trained image classifier; Input the test set of the image data set into the trained image classifier, perform classification prediction, and obtain the classification prediction result; The classification prediction results are verified through the validation set of the image data set. If the classification prediction results meet the set requirements, the classification results are obtained; Based on the classification results, the effectiveness of feature selection is evaluated.

[0021] In a second aspect, the present invention provides a feature selection system based on maximum information coefficient for image classification, comprising: Data acquisition module: used to obtain the image data set to be classified; Image calculation module: used to calculate the redundancy matrix and maximum information coefficient based on the image data set; Model building module: used to build the objective function model based on the redundancy matrix and the maximum information coefficient; Feature scoring module: used to calculate the features of the image data set using the objective function model to obtain feature scores; sort the feature scores according to the set conditions, select the target features, and form a feature subset; Classification result module: used to train the image classifier using the feature subset, input the test set of the image dataset into the trained image classifier, and obtain the classification result.

[0022] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of a feature selection method based on maximum information coefficient for image classification when executing the computer program.

[0023] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the feature selection method based on maximum information coefficient for image classification are implemented.

[0024] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions, which, when executed by a processor, implement the steps of a feature selection method based on maximum information coefficient for image classification.

[0025] Compared with the prior art, the present invention has the following beneficial effects: By using a feature selection method based on the maximum information coefficient, the present invention can more effectively select features that are highly relevant to the target category and have low redundancy. This selected feature subset can significantly improve the accuracy of image classification because it reduces the negative impact of noise and redundant information on classifier performance. In image classification tasks, the number of features is often very large. The present invention constructs an objective function model and uses the model to score and sort the features, so that the most important features can be quickly identified and retained, while ignoring those features that contribute less to the classification. This not only reduces the computational complexity, but also shortens the processing time and improves the efficiency of the overall algorithm. By reducing feature redundancy, the present invention helps to build a more concise and more generalized image classification model. The model no longer relies on redundant or noisy features, but relies more on those features that really contribute to the classification, which enables the model to show better adaptability and accuracy when facing new image data. Therefore, the feature selection method in the present invention is based on the maximum information coefficient, which means that it can flexibly adjust the feature selection strategy according to different data sets and classification tasks. By adjusting the parameters in the objective function model, different trade-offs between feature redundancy and relevance can be achieved, thereby adapting to different application scenarios and needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The drawings described herein are for illustrative purposes only and are not intended to limit the scope of the present disclosure in any way. In the drawings: Figure 1 A schematic diagram of a process flow of a feature selection method based on maximum information coefficient for image classification according to the present invention; Figure 2 A structural schematic diagram of a feature selection system based on maximum information coefficient for image classification according to the present invention; Figure 3 Schematic diagram of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. The described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present invention.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0029] Aiming at the problem that the existing image classification technology cannot balance feature redundancy and new classification information, the present invention provides a feature selection method based on maximum information coefficient for image classification, such as Figure 1 As shown, the method includes: S1: Obtain the image dataset to be classified; S2: Calculate the redundancy matrix and the maximum information coefficient based on the image data set; S3: Construct the objective function model based on the redundancy matrix and the maximum information coefficient; S4: Calculate the features of the image data set using the objective function model to obtain feature scores; sort the feature scores according to the set conditions, select the target features, and form a feature subset; S5: Use the feature subset to train the image classifier, input the test set of the image dataset into the trained image classifier, and obtain the classification result.

[0030] Specific methods include: Get a dataset of images, including data matrices and class labels; Based on the image data set, a centering matrix is ​​calculated to decentralize the data and ensure that the similarity measure between features is not affected by the mean of the original data; It is constructed by calculating the square of the cosine similarity between feature vectors to measure the redundancy between features; Calculate the MIC value between each feature and the sample class label to measure the statistical dependence between the feature and the target variable; Construct an objective function, which includes: maximizing the MIC value (correlation) between the feature and the target variable, and minimizing the redundancy between the features; Use the Lagrange multiplier method and KKT conditions to solve the optimization problem and obtain the score of each feature; Sort by feature score from high to low, select the first Features constitute a feature subset ; Classification is performed on the reduced-dimensional dataset to evaluate the effectiveness of the feature selection method.

[0031] The present invention is further explained below in conjunction with specific embodiments.

[0032] In order to facilitate the solution, the original objective function is rewritten as:

[0033] in, is the regularization parameter, which controls the balance between the redundancy term and the maximum information coefficient term; is the characteristic dimension of the data set; ; is a redundant matrix, where each element is ,when hour, Representation characteristics Its own redundancy (usually 1), when hour, Representation characteristics and The redundancy between It contains all feature scores. dimensional column vector, yes The feature score of yes The feature score of Features With class label The maximum information coefficient between Ensure that the sum of weights is 1, Ensure that each weight is non-negative; is a d-dimensional column vector of all 1s.

[0034] It is a constrained convex optimization problem, which is solved by the Lagrange multiplier method.

[0035] Constructing Lagrangian functions and introducing Lagrangian multipliers and , the Lagrangian function is:

[0036] in, Is to handle equality constraints The Lagrange multiplier of Is to deal with inequality constraints The Lagrange multiplier of .

[0037] Solve for the partial derivatives and set them to 0:

[0038] In order to solve, the KKT conditions (Karush-Kuhn-Tucker conditions) need to be met: Original constraints:

[0039] Lagrangian duality condition:

[0040] Lagrange multiplier conditions:

[0041] For each From the Lagrangian duality condition we get

[0042] Because when hour So the above formula is simplified to:

[0043] Using constraints All Adding them together gives:

[0044] Will Bring back each The equation is:

[0045] From this we can calculate the final expression:

[0046] Code implementation process: is a redundancy matrix that describes the pairwise redundancy relationship between all features, where each element is ; is the maximum information coefficient matrix, which measures the entire data set and class labels Y The correlation between them, where each element is Measuring characteristics and relationship.

[0047] Given a dataset , data matrix No. and Features and express, (n is the number of samples in the data set, d is the number of features), the data class label is .

[0048] Calculate the redundancy matrix and the maximum information coefficient

[0049] Depend on Get the centralization matrix and ,in is a centralization matrix, defined as , yes The identity matrix, is a Dimensional all-1 vector. Centered matrix Used to decenter the data (i.e., subtract the mean). It is Centralized features; It is Centralized features. Centralized features and Through the centralization matrix The obtained, centralized features are used to calculate the redundancy matrix Elements of the matrix Elements Defined as , In fact, the feature vector and The square of the cosine similarity between the two features is calculated by centralization to ensure that the similarity (redundancy) measurement between the features is not affected by the mean of the original data, thereby more accurately reflecting the true relationship between the features.

[0050] The redundant matrix is ​​obtained by the above calculation , the redundant matrix What is measured is the characteristics and The redundancy between .

[0051] The formula for the maximum information coefficient is:

[0052] in, It is mutual information, so the relationship between the maximum information coefficient MIC and mutual information is: the maximum information coefficient is a method of estimating mutual information. yes The probability mass function of and Same reason. is the data matrix The number of grids divided; is the class label The number of grids divided, is the maximum grid size limit, is the number of samples in the dataset.

[0053] In the MIC formula, , data matrix No. and Features and express, ; is the class label corresponding to the data. By inputting sample data, the MIC value between each feature and the sample class label is calculated to obtain the diagonal matrix MIC consisting of the maximum information coefficient of each feature and classification information.

[0054] Bring in The formula gets the score of each feature

[0055] Sort all feature scores from high to low, feature scores The higher the value, the higher the corresponding feature. The redundancy with other features is small and the classification information (class label) ) have a high correlation, and all corresponding data features are sorted according to the feature score sequence. Apply this feature selection method to different classification models. If you want to perform feature selection on the data set, the original dimensional feature dataset to Dimension, select The features constitute a feature subset. Then the classification is performed on the dataset after dimensionality reduction.

[0056] The experimental results show that the classification results obtained after using this method for feature selection are higher than those of other feature selection methods, proving that this method can reduce feature redundancy and select features that are more conducive to classification.

[0057] This feature selection method combines the maximum information coefficient and the redundancy between features (measured by the cosine similarity matrix) and has the following advantages: Although mutual information can measure the statistical dependence between features and target variables, its ability to capture complex high-dimensional nonlinear relationships is limited. The maximum information coefficient (MIC) is a more general measurement tool that can capture the dependence of multiple modes in low-dimensional and high-dimensional data, thereby improving the comprehensiveness of feature selection.

[0058] The maximum information coefficient is more sensitive to nonlinear relationships than mutual information. Therefore, in practical applications, it can better select features that have complex nonlinear relationships with the target variable, which is particularly useful when the data has complex nonlinear structures. The maximum information coefficient can not only discover linear relationships, but also reveal other forms of relationships, such as exponential relationships, piecewise linear relationships, and periodic relationships. This ensures that when selecting features, the algorithm will not miss features that have complex nonlinear relationships with the target variable, thereby improving the model's ability to capture nonlinear features.

[0059] The maximum information coefficient is a measurement method that can adapt to different data structures. Whether it is a categorical variable or a continuous variable, whether it is sparse data or dense data, it can effectively evaluate the dependency between variables. This feature enables feature selection methods to adapt to more diverse data structures and improve the representativeness of the selected features.

[0060] Compared with mutual information, MIC has better noise resistance and can maintain high reliability even when the data is noisy. This is very important for many real-world data sets because these data are often affected by environmental, equipment, or measurement errors. Using MIC can make feature selection more robust and reduce the interference of noise on the selection results.

[0061] The calculation of MIC does not depend on the amount of data and can still maintain good performance on small sample data sets. When the sample size is small, the calculation accuracy of mutual information may be limited, especially for non-uniformly distributed data. However, MIC can still find the dependency patterns in the data when the sample size is small. Therefore, in the feature selection process, using MIC can improve reliability in small sample scenarios.

[0062] Selecting the features with the greatest correlation with the target variable can improve the predictive ability of the model. The term directly maximizes the correlation between the feature and the target. The maximum information coefficient MIC measures the statistical dependence between the feature and the target. By maximizing this value, it can ensure that the selected feature is most relevant to the target variable in terms of information.

[0063] It reduces the redundancy between features, avoids selecting a large number of similar features, and improves the generalization ability of the model. The redundant matrix To control the redundancy between features, is the cosine similarity matrix between features, which is used to represent the similarity between features. By minimizing this term, the feature selection process tends to select features that have low redundancy with other features. Reducing redundant features can reduce computational costs and improve the interpretability of the model.

[0064] Through the regularization parameter Flexible trade-offs can be made to select more relevant or more independent features according to specific application scenarios, enhancing the versatility and adjustability of the algorithm.

[0065] Regularization parameter in optimization objective Controls the balance between two objectives: one is to maximize the mutual information between features and target variables, and the other is to minimize the redundancy between features. , the optimal balance between relevance and redundancy can be found.

[0066] By Constraint , which makes the feature selection weights have a probabilistic interpretation and enhances the interpretability of the method. The constraints of the optimization problem ensure that each feature weight is non-negative and normalized, which means that the weight of each feature can be interpreted as relative importance or probability. By restricting the feature weights to be non-negative and normalized to 1, this feature selection method can ensure that each feature weight is within a reasonable range and avoid unreasonable weight distribution.

[0067] Due to the processing of feature redundancy and the design of weight normalization, this method also performs well in high-dimensional data and can effectively deal with the curse of dimensionality. This method can still maintain good feature selection effects in high-dimensional space. The optimization of redundant items helps to reduce multicollinearity in high-dimensional space, and the maximization of mutual information ensures the selection of useful features.

[0068] In high-dimensional data, features are highly interdependent and have high redundancy. Traditional feature selection methods may have difficulty effectively distinguishing between correlation and redundancy. This joint optimization method based on redundancy and correlation can select fewer but important features in high-dimensional data and reduce the complexity of the model.

[0069] Using closed-form solutions to solve optimization problems can significantly reduce computational complexity and is particularly suitable for large-scale data sets.

[0070] The maximum information coefficient (MIC) is used instead of mutual information, which is more advantageous than mutual information. This feature selection method selects features that are highly relevant to the target by maximizing mutual information, while reducing unnecessary features by controlling the redundancy between features. The number of regularization parameters provides a flexible trade-off mechanism, linear constraints ensure the interpretability of weights, and closed-form solutions improve computational efficiency. These features make this method particularly suitable for selecting representative and independent features in high-dimensional and complex data, and improving the generalization performance and interpretability of the model.

[0071] As shown in Table 1, after using this feature selection method (MICFS) to perform feature selection and dimensionality reduction on the AR dataset with different dimensions (dim), three image classification methods, CRC, SVM and KNN, are used for classification. The classification accuracy of different classification methods in different dimensions after using this feature selection algorithm is compared: dim is the dimension of the selected feature.

[0072] Table 1 Classification accuracy of this feature selection method in different situations on the AR dataset

[0073] As shown in Table 2, after using this feature selection method (MICFS) to perform feature selection and dimension reduction of different dimensions (dim) on the ORL dataset, three image classification methods, CRC, SVM and KNN, are used for classification. The classification accuracy of different classification methods in different dimensions after using this feature selection algorithm is compared: dim is the dimension of the selected feature.

[0074] Table 2 Classification accuracy of this feature selection method in different situations on the ORL dataset

[0075] As shown in Table 3, after using this feature selection method (MICFS) to perform feature selection and dimensionality reduction on the Yale dataset with different dimensions (dim), three image classification methods, CRC, SVM and KNN, are used for classification. The classification accuracy of different classification methods under different dimensions after using this feature selection algorithm is compared: dim is the dimension of the selected feature.

[0076] Table 3 Classification accuracy of this feature selection method in different situations on the Yale dataset

[0077] The second object of the present invention is to propose that the present invention provides a feature selection system based on maximum information coefficient for image classification, such as Figure 2 As shown, including: Data acquisition module 11: used to acquire image data sets to be classified; Image calculation module 21: used to calculate the redundancy matrix and the maximum information coefficient according to the image data set; Model building module 31: used for building an objective function model based on a redundancy matrix and a maximum information coefficient; Feature scoring module 41: used to calculate features of the image data set using the objective function model to obtain feature scores; sort the feature scores according to set conditions, select target features, and form feature subsets; Classification result module 51: used to train an image classifier using a feature subset, input a test set of the image data set into the trained image classifier, and obtain a classification result.

[0078] like Figure 3 As shown, the third object of the present invention is to provide an electronic device, the electronic device comprising: a processor 601, a memory 602 and a display screen 603. Among them, the memory 602 and the display screen 603 are connected to the processor 601, such as through a bus 604. Optionally, the electronic device may also include a transceiver 605. It should be noted that in actual applications, the transceiver 605 is not limited to one, and the structure of the electronic device does not constitute a limitation on the embodiments of the present application.

[0079] Processor 601 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 601 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and the like.

[0080] The bus 604 may include a path to transmit information between the above components. The bus 604 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus 604 may be divided into an address bus, a data bus, a control bus, etc.

[0081] The memory 602 may be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compressed optical disk, laser disk, optical disk, digital versatile disk, Blu-ray disk, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto.

[0082] The memory 602 is used to store the application code for executing the solution of the present application, and the execution is controlled by the processor 601. The processor 601 is used to execute the application code stored in the memory 602 to implement the content shown in the above method embodiment.

[0083] Figure 3 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0084] The fourth object of the present invention is to provide a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein the computer program is stored thereon, and when the program is executed by a processor, the aforementioned Figure 1 The various processes of the method embodiment shown in the figure include, for example, a memory including instructions, and the instructions can be executed by a processor of an electronic device to complete the method.

[0085] A computer-readable storage medium may be a tangible device that holds and stores instructions used by an instruction execution device. A computer-readable storage medium may be, but is not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof. Specifically, a computer-readable storage medium may be a portable computer disk, a hard disk, a USB flash drive, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a lectern random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, an optical disk, a magnetic disk, a mechanical encoding device, and any combination thereof.

[0086] A fifth object of the present invention is to provide a computer program product, comprising computer instructions, which, when executed by a processor, implement the above Figure 1 The various processes of the method embodiment shown can achieve the same technical effect, and will not be described again here to avoid repetition.

[0087] Many embodiments and many applications beyond the examples provided will be apparent to those skilled in the art upon reading the above description. Therefore, the scope of the present teachings should not be determined with reference to the above description, but rather with reference to the preceding claims and the full scope of equivalents to which such claims are entitled. For the purpose of comprehensiveness, all articles and references, including disclosures of patent applications and publications, are incorporated herein by reference. The omission of any aspect of the subject matter disclosed herein in the preceding claims is not intended to be a disclaimer of such subject matter, nor should it be considered that the applicants did not consider such subject matter to be part of the disclosed inventive subject matter.

[0088] The above content is a further detailed description of the present invention. It cannot be determined that the specific implementation methods of the present invention are limited to these. For ordinary technicians in the technical field to which the present invention belongs, without departing from the concept of the present invention, they can also make several simple deductions or substitutions, which should be regarded as belonging to the protection scope of the present invention determined by the submitted claims.

Claims

1. A feature selection method based on maximum information coefficient for image classification, characterized in that: include: Obtain an image dataset to be classified; Calculate the redundancy matrix and the maximum information coefficient based on the image data set; Construct the objective function model based on the redundancy matrix and the maximum information coefficient; Use the objective function model to calculate the features of the image data set and obtain the feature score; Sort the feature scores according to the set conditions, select the target features, and form a feature subset; Use the feature subset to train the image classifier, input the test set of the image dataset into the trained image classifier, and obtain the classification result.

2. A feature selection method based on maximum information coefficient for image classification according to claim 1, characterized in that: The step of obtaining an image data set to be classified includes: For the image data set for image classification, preprocessing is performed to obtain a given data set; The given data set includes a data matrix and class labels ; The data matrix No. Column and The columns are and Features, using and express, ; is the number of samples in the dataset; is the number of features.

3. A feature selection method based on maximum information coefficient for image classification according to claim 1, characterized in that: The step of calculating the redundancy matrix and the maximum information coefficient according to the image data set includes: The redundant matrix comprises: , , ; The maximum information coefficient includes: in, is a redundant matrix, It is Centralized features; It is Centralized features, is a centralized matrix; yes The identity matrix, is a dimensional all-1 vector; is mutual information; is the data matrix; is the data matrix No. Features is the data matrix No. Features is the class label corresponding to the data; is the data matrix The number of grids divided; is the class label The number of grids divided, is the maximum grid size limit, is the number of samples in the dataset.

4. A feature selection method based on maximum information coefficient for image classification according to claim 1, characterized in that: The objective function model is constructed based on the redundancy matrix and the maximum information coefficient, including: Constructing the objective function in, is a redundant matrix, where each element is , metric features and The redundancy between It contains all feature scores. dimension column vector, yes The feature score of yes The feature score of is the number of features, is the regularization parameter; is the class label corresponding to the data; for The maximum information coefficient between the class label; For all 1 Dimensional column vector; Under the KKT condition, solve the objective function to obtain the feature score : in, is a dimension of A vector where each element is .

5. A feature selection method based on maximum information coefficient for image classification according to claim 1, characterized in that: The method of calculating the features of the image data set using the objective function model to obtain the feature score includes: Calculate the MIC value between each feature in the image data set and the known class label, and obtain the diagonal matrix MIC consisting of the maximum information coefficient of each feature and classification information; The diagonal matrix MIC composed of the maximum information coefficient is input into the objective function model to obtain the score of each feature.

6. A feature selection method based on maximum information coefficient for image classification according to claim 1, characterized in that: The feature scores are sorted according to the set conditions, and the target features are selected to form a feature subset, including: Sort the acquired feature scores from high to low to obtain a feature score sequence; Sort the corresponding data features according to the feature score sequence; Sort by feature score from high to low, select the first Features constitute a feature subset , complete the feature selection of the image dataset.

7. A feature selection method based on maximum information coefficient for image classification according to claim 1, characterized in that: The method of using a feature subset to train an image classifier and inputting a test set of the image data set into the trained image classifier to obtain a classification result includes: Use the feature subset to train the image classifier to obtain a trained image classifier; Input the test set of the image data set into the trained image classifier, perform classification prediction, and obtain the classification prediction result; The classification prediction results are verified through the validation set of the image data set. If the classification prediction results meet the set requirements, the classification results are obtained; Based on the classification results, the effectiveness of feature selection is evaluated.

8. A feature selection system based on maximum information coefficient for image classification, characterized in that: include: Data acquisition module: used to obtain the image data set to be classified; Image calculation module: used to calculate the redundancy matrix and maximum information coefficient based on the image data set; Model building module: used to build the objective function model based on the redundancy matrix and the maximum information coefficient; Feature scoring module: used to calculate the features of the image data set using the objective function model to obtain feature scores; Sort the feature scores according to the set conditions, select the target features, and form a feature subset; Classification result module: used to train the image classifier using the feature subset, input the test set of the image dataset into the trained image classifier, and obtain the classification result.

9. An electronic device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of a feature selection method based on maximum information coefficient for image classification as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the feature selection method based on maximum information coefficient for image classification described in any one of claims 1 to 7 are implemented.

Images