An image clustering method and device based on multi-view learning and Top-k loss, equipment and medium

Abstract

The application relates to the technical field of image clustering, and discloses an image clustering method, device, equipment and medium based on multi-view learning and Top-k loss, which comprises the following steps: S1, converting an original image data set into a multi-view feature image data set; S2, constructing a first image clustering model based on multi-view learning and Top-k loss based on the multi-view feature image data set; S3, optimizing the first image clustering model by using a constrained concave-convex process algorithm to obtain a second image clustering model; S4, indexing and arranging the constraint of the second image clustering model according to a preset condition to obtain a third image clustering model; and S5, iteratively optimizing the third image clustering model by using a cut plane algorithm, and judging whether the third image clustering model converges, if yes, solving an image clustering result by using the third image clustering model, and if not, jumping to S3. The application improves the accuracy, precision and flexibility of image clustering.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image clustering, and more particularly, to an image clustering method and device based on multi-view learning and Top-k loss, an image clustering equipment and a medium. BACKGROUND

[0002] In recent years, with the explosive growth of image data, image clustering technology plays an increasingly important role in data analysis and information processing. As an effective clustering method, the maximum margin clustering algorithm has attracted widespread attention in image clustering applications. However, with the complexity of application scenarios, the traditional maximum margin clustering algorithm faces great challenges in processing multi-view data and solving the problem of class overlap.

[0003] At present, the academic circle has proposed various improved maximum margin clustering algorithms. For example, Professor Xu proposed a non-convex maximum margin clustering algorithm which is solved by a relaxed semi-definite programming; Professor Valizadegan proposed a generalized maximum margin clustering and unsupervised kernel learning which improves the clustering efficiency by reducing the size of the semi-definite programming problem; Professor Wang proposed a linear-time maximum margin clustering which introduces a hinge loss; Professor Xue proposed a maximum margin clustering algorithm based on indefinite kernel which introduces an F-norm regularizer in the image clustering model. However, these methods still have two main defects: first, they are designed for single-view data and are difficult to handle multi-view data, which may lead to a decrease in model accuracy in image clustering applications; second, these algorithms assume that images of different classes do not overlap or are similar, ignoring the problem of class overlap and similarity that may exist in actual applications, which may lead to clustering ambiguity and reduce model precision. SUMMARY

[0004] To overcome the defects of low accuracy and precision in existing image clustering technology, the present application proposes the following technical solutions:

[0005] In a first aspect, the present application proposes an image clustering method based on multi-view learning and Top-k loss, comprising:

[0006] S1: converting an original image data set into a multi-view feature image data set;

[0007] S2: based on the multi-view feature image data set, constructing a first image clustering model based on multi-view learning and Top-k loss;

[0008] S3: using a constrained concave-convex procedure algorithm to optimize the first image clustering model to obtain a second image clustering model;

[0009] S4: according to a pre-set condition, indexing and arranging the constraints of the second image clustering model to obtain a third image clustering model;

[0010] S5: Use the cutting plane algorithm to iteratively optimize the third image clustering model, and determine whether the third image clustering model has converged. If it has, use the third image clustering model to solve for the image clustering results. If not, jump to S3.

[0011] As a preferred technical solution, the original image dataset is converted into a multi-view feature image dataset, including:

[0012] Using several different image feature extraction methods, the original image dataset was processed. Each image in Transform into M eigenvectors A multi-view feature image dataset was obtained. ;in, Representing an image From the perspective v The eigenvectors below, From the perspective v Number of data dimensions This represents the number of samples in the dataset.

[0013] As a preferred technical solution, after obtaining the multi-view feature image dataset, the method further includes:

[0014] set up K A classification plane, namely The data in the multi-view feature image dataset is divided into K Clusters; among which, From the perspective v The first in p A classification plane, ( (From the perspective) v The Middle p Each classification plane normal vector.

[0015] As a preferred technical solution, based on the multi-view feature image dataset, a first image clustering model based on multi-view learning and Top-k loss is constructed, including:

[0016]

[0017]

[0018]

[0019]

[0020]

[0021]

[0022] in, For the number of data samples, For the number of clusters, M For the number of viewpoints, It is a constant; Indicates sample i In the v Relaxed variables from different perspectives Indicates sample i In the v From the perspective of the first p The classification plane and the first u From the perspective of the first p Relaxation variables between classification planes; and They were respectively in the second v From the perspective of the first p The and the first q Normal vectors of the classification plane; For non-negative parameters related to clustering error, A non-negative parameter to control the impact of viewpoint consistency on the learning model; k is a parameter in the Top-k loss, representing the number of allowed prediction errors; To constrain the non-negativity parameter of cluster imbalance, The parameter is non-negative. The classification plane index corresponding to the maximum predicted output value.

[0023] As a preferred technical solution, a constrained concavity-convexity process algorithm is used to optimize the first image clustering model to obtain a second image clustering model, including:

[0024] make , ;definition ,in, Depend on K indivual Composed of dimensional vectors, the first r The vectors are ,other K -1 vectors are zero vectors; definition ,in, Depend on M Composed of the nth vector, the nth vector v The vectors are ,other M -1 vectors are zero vectors;

[0025] By definition and The first image clustering model , , , Transformed into , , , This will enable the first constraint of the first image clustering model to be applied. Transform into ;

[0026] For this nonconvex constraint ,initialization Value, used in First-order Taylor expansion of a point By substitution, we obtain the second image clustering model, whose expression is shown below:

[0027]

[0028]

[0029]

[0030]

[0031]

[0032]

[0033]

[0034] in,

[0035]

[0036]

[0037] in, The category plane index represents the maximum predicted output value, while Indicates the first q Classification plane index of large predicted output values.

[0038] As a preferred technical solution, the constraints of the second image clustering model are indexed and arranged according to preset conditions to obtain a third image clustering model, including:

[0039] Define variables and variables ,in, ;

[0040] for and , constrain conditions In Arrange and reindex as and order ;

[0041] for and , constrain conditions Arrange and reindex as and and order ;

[0042] for The constraints in the second image clustering model Arrange and reindex as ;

[0043] After completing the above arrangement and indexing, the expression for the third image clustering model is as follows:

[0044]

[0045] :

[0046]

[0047] :

[0048]

[0049]

[0050]

[0051] .

[0052] As a preferred technical solution, the third image clustering model is iteratively optimized using the cutting plane algorithm, and the image clustering results are solved using the third image clustering model, including:

[0053] The third image clustering model is converted into a single-relaxation form, the expression of which is as follows:

[0054]

[0055] ;

[0056]

[0057]

[0058]

[0059]

[0060]

[0061] in:

[0062]

[0063]

[0064] in, It is a single slack variable.

[0065] Define constraint set and The current solution is used iteratively through the cutting plane algorithm. Find the two most violated constraints and and respectively and Add to and In, until the solution Achieving accuracy by satisfying all constraints Thus, a third image clustering model that meets the accuracy requirements is obtained;

[0066] Solving using a third image clustering model that meets the accuracy requirements. The value of , and according to the following formula for multi-view samples Clustering is performed to obtain the image clustering results:

[0067]

[0068] in, For multi-view samples The predicted labels, i.e., the image clustering results, For multi-view samples Compared to the first p The plane in the first v Predicted output values ​​for each viewpoint For the sample Compared to the first p The average predicted output value of a plane across all viewpoints.

[0069] Secondly, the present invention also proposes an image clustering system based on multi-view learning and Top-k loss, applicable to the image clustering method based on multi-view learning and Top-k loss as described in any of the schemes of the first aspect, comprising:

[0070] The conversion module is used to convert the original image dataset into a multi-view feature image dataset;

[0071] The building module is used to construct a first image clustering model based on multi-view learning and Top-k loss based on the multi-view feature image dataset;

[0072] The first optimization module is used to optimize the first image clustering model using a constrained concave-convex process algorithm to obtain a second image clustering model.

[0073] The indexing and arranging module is used to index and arrange the constraints of the second image clustering model according to preset conditions to obtain the third image clustering model;

[0074] The second optimization module is used to iteratively optimize the third image clustering model using the cutting plane algorithm, and to determine whether the third image clustering model has converged. If it has, the third image clustering model is used to solve for the image clustering result. If not, the first optimization module is controlled to re-execute the optimization operation.

[0075] Thirdly, the present invention also proposes an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to perform operations performed by the image clustering method based on multi-view learning and Top-k loss as described in any of the schemes in the first aspect.

[0076] In a fourth aspect, the present invention also proposes a computer-readable storage medium storing a program that is executed by a processor as described in any of the embodiments of the first aspect, using an image clustering method based on multi-view learning and Top-k loss.

[0077] The beneficial effects of the present invention include at least the following:

[0078] This invention first transforms the original image dataset into a multi-view feature image dataset, and then constructs an image clustering model based on multi-view learning and Top-k loss. This multi-view learning approach effectively utilizes information from multiple perspectives, addressing the problem of incomplete information from a single perspective, thereby improving the accuracy of image clustering. Simultaneously, the Top-k loss function based on the maximum margin clustering model, compared to traditional maximum margin clustering, allows the model to select the k-th largest prediction value during the learning process, increasing the algorithm's training tolerance and solving the problem of failing to predict successfully on the first attempt due to image class overlap and blurring, further improving the accuracy of image clustering. The k value in the Top-k loss function can be flexibly adjusted according to different image datasets and application requirements to adapt to different types of clustering tasks. Finally, the constrained concave-convex process algorithm and the cutting plane algorithm are used to optimize and iterate the clustering model, ensuring its convergence and stability, further improving the performance of image clustering. Attached Figure Description

[0079] Figure 1 This is a flowchart illustrating the image clustering method based on multi-view learning and Top-k loss provided in Example 1.

[0080] Figure 2 This is an architecture diagram of the image clustering system based on multi-view learning and Top-k loss provided in Example 3.

[0081] Figure 3 This is a schematic diagram of the structure of the electronic device provided in Example 4. Detailed Implementation

[0082] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred technical solutions. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred technical solutions are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0083] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0084] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0085] Example 1

[0086] This embodiment proposes an image clustering method based on multi-view learning and Top-k loss, such as Figure 1 As shown, Figure 1 This is a flowchart illustrating an image clustering method based on multi-view learning and Top-k loss provided in this embodiment. The method includes the following steps:

[0087] S1: Convert the original image dataset into a multi-view feature image dataset;

[0088] S2: Based on the multi-view feature image dataset, construct a first image clustering model based on multi-view learning and Top-k loss;

[0089] S3: Using the constrained concave-convex process algorithm, optimize the first image clustering model to obtain the second image clustering model;

[0090] S4: Based on preset conditions, index and arrange the constraints of the second image clustering model to obtain the third image clustering model;

[0091] S5: Use the cutting plane algorithm to iteratively optimize the third image clustering model, and determine whether the third image clustering model has converged. If it has, use the third image clustering model to solve for the image clustering results. If not, jump to S3.

[0092] Understandably, this embodiment first converts the original image dataset into a multi-view feature image dataset, and then constructs an image clustering model based on multi-view learning and Top-k loss. This multi-view learning approach effectively utilizes information from multiple perspectives, addressing the problem of incomplete information from a single perspective, thereby improving the accuracy of image clustering. Simultaneously, the Top-k loss function based on the maximum margin clustering model, compared to traditional maximum margin clustering, allows the model to select the k-th largest prediction value during the learning process, increasing the algorithm's training tolerance and solving the problem of failing to predict successfully on the first attempt due to image class overlap and blurring, further improving the accuracy of image clustering. The k value in the Top-k loss function can be flexibly adjusted according to different image datasets and application requirements to adapt to different types of clustering tasks. Finally, the constrained concave-convex process algorithm and the cutting plane algorithm are used to optimize and iterate the clustering model, ensuring its convergence and stability, further improving the performance of image clustering.

[0093] Example 2

[0094] In this embodiment, the original image dataset is converted into a multi-view feature image dataset, including:

[0095] Using M different image feature extraction methods, including RGB, HOG, and SIFT, the original image dataset was processed. Each image in Transform into M eigenvectors A multi-view feature image dataset was obtained. ;in, Representing an image From the perspective v The eigenvectors below, From the perspective v Number of data dimensions This represents the number of samples in the dataset.

[0096] In this embodiment, after obtaining the multi-view feature image dataset, the following steps are set: K A classification plane, namely The data in the multi-view feature image dataset is divided into K Clusters; among which, From the perspective v The first in p A classification plane, ( (From the perspective) v The Middle p Each classification plane normal vector.

[0097] In this embodiment, a first image clustering model based on multi-view learning and Top-k loss is proposed by fusing the Top-k loss function and the multi-view consistency principle constraint. The Top-k loss can be expressed as... .in, Representing the eigenvector The r Great loss Representing the eigenvector The former r The mean of the largest losses. The Top-k loss can be expressed as... .in, Representing the eigenvector The r Great loss Representing the eigenvector The former r The mean of large losses. Top-k loss before use. r Large loss means to represent feature vectors The loss. In clustering problems, the classification labels of samples. Since the feature vector is unknown, this embodiment uses the feature vector. The loss is expressed as Among them, there are , represents the classification plane corresponding to the maximum predicted output value. Here, This indicates that, apart from the maximum predicted output value, the th r Large prediction output value. Representing the eigenvector The maximum predicted output value, that is, in K In each classification plane, the feature vector It belongs to the classification plane with the largest predicted output value. Additionally, The k-loss can be expressed as Furthermore, the consistency constraint across multiple perspectives can be expressed as: .in, For the sample In the v The perspective relative to the first p The predicted output of each classification plane, For the sample In the u The perspective relative to the first p The predicted output of each classification plane, It is a slack variable. For viewpoint consistency parameters. Consistency constraints. The meaning is sample From any two perspectives, relative to the same classification plane, they have similar predicted output values, which guarantees that the same sample... Image clustering models based on multi-view learning and Top-k loss are constructed when images are grouped into the same cluster regardless of their viewpoint. The expression for this model is as follows:

[0098]

[0099]

[0100]

[0101]

[0102]

[0103]

[0104] in, For the number of data samples, For the number of clusters, M For the number of viewpoints, It is a constant; Indicates sample i In the v Relaxed variables from different perspectives Indicates sample i In the v From the perspective of the first p The classification plane and the first u From the perspective of the first p Relaxation variables between classification planes; and They were respectively in the second v From the perspective of the first p The and the first q Normal vectors of the classification plane; For non-negative parameters related to clustering error, A non-negative parameter to control the impact of viewpoint consistency on the learning model; k is a parameter in the Top-k loss, representing the number of allowed prediction errors; To constrain the non-negativity parameter of cluster imbalance, The parameter is non-negative. The classification plane index corresponding to the maximum predicted output value.

[0105] In this embodiment, a constrained concavity-convexity process algorithm is used to optimize the first image clustering model to obtain a second image clustering model, including:

[0106] make , ;definition ,in, Depend on Kindivual Composed of dimensional vectors, the first r The vectors are ,other K -1 vectors are zero vectors; definition ,in, Depend on M Composed of the nth vector, the nth vector v The vectors are ,other M -1 vectors are zero vectors;

[0107] By definition and The first image clustering model , , , They can be converted into , , , Therefore, the first constraint of the first image clustering model It can be converted into Because this constraint includes a max function, it is a non-convex constraint. For this non-convex constraint, initialization... Value, used in First-order Taylor expansion of a point By substitution, we obtain the second image clustering model, whose expression is shown below:

[0108]

[0109]

[0110]

[0111]

[0112]

[0113]

[0114]

[0115] in,

[0116]

[0117]

[0118] in, The category plane index represents the maximum predicted output value, while Indicates the first qClassification plane index of large predicted output values.

[0119] In this embodiment, the constraints of the second image clustering model are indexed and arranged according to preset conditions to obtain a third image clustering model, including:

[0120] Define variables and variables ,in, ;

[0121] for and , constrain conditions In Arrange and reindex as and order ;

[0122] for and , constrain conditions Arrange and reindex as and and order ;

[0123] for The constraints in the second image clustering model Arrange and reindex as ;

[0124] After completing the above arrangement and indexing, the expression for the third image clustering model is as follows:

[0125]

[0126] :

[0127]

[0128] :

[0129]

[0130]

[0131]

[0132] ;

[0133] in, .

[0134] In this embodiment, the cutting plane algorithm is used to iteratively optimize the third image clustering model, and the image clustering results are obtained using the third image clustering model, including:

[0135] The third image clustering model is converted into a single-relaxation form, the expression of which is as follows:

[0136]

[0137] ;

[0138]

[0139]

[0140]

[0141]

[0142]

[0143] in:

[0144]

[0145]

[0146] in, It is a single slack variable.

[0147] Define constraint set and The current solution is used iteratively through the cutting plane algorithm. Find the two most violated constraints and and respectively and Add to and In, until the solution Achieving accuracy by satisfying all constraints Thus, a third image clustering model that meets the accuracy requirements is obtained;

[0148] Solving using a third image clustering model that meets the accuracy requirements. ,because Depend on Composed of, it can be obtained The value is then used to evaluate the multi-view samples according to the following formula. Clustering is performed to obtain the image clustering results:

[0149]

[0150] in, For multi-view samples The predicted labels, i.e., the image clustering results, For multi-view samples Compared to the first p The plane in the first v Predicted output values ​​for each viewpoint For the sample Compared to the first p The average predicted output value of a plane across all viewpoints.

[0151] Understandably, this embodiment first converts the original image dataset into a multi-view feature image dataset, and then constructs an image clustering model based on multi-view learning and Top-k loss. This multi-view learning approach effectively utilizes information from multiple perspectives, addressing the problem of incomplete information from a single perspective, thereby improving the accuracy of image clustering. Simultaneously, the Top-k loss function based on the maximum margin clustering model, compared to traditional maximum margin clustering, allows the model to select the k-th largest prediction value during the learning process, increasing the algorithm's training tolerance and solving the problem of failing to predict successfully on the first attempt due to image class overlap and blurring, further improving the accuracy of image clustering. The k value in the Top-k loss function can be flexibly adjusted according to different image datasets and application requirements to adapt to different types of clustering tasks. Finally, the constrained concave-convex process algorithm and the cutting plane algorithm are used to optimize and iterate the clustering model, ensuring its convergence and stability, further improving the performance of image clustering.

[0152] Example 3

[0153] like Figure 2 As shown, this embodiment proposes an image clustering system based on multi-view learning and Top-k loss, which is applied to the image clustering method based on multi-view learning and Top-k loss as described in the above embodiment. It includes: a transformation module 100, a construction module 200, a first optimization module 300, an index arrangement module 400, and a second optimization module 500.

[0154] The system comprises the following modules: a conversion module 100 converts the original image dataset into a multi-view feature image dataset; a construction module 200 constructs a first image clustering model based on the multi-view feature image dataset and Top-k loss; a first optimization module 300 optimizes the first image clustering model using a constraint concave-convex process algorithm to obtain a second image clustering model; an indexing and arranging module 400 indexes and arranges the constraints of the second image clustering model according to preset conditions to obtain a third image clustering model; and a second optimization module 500 iteratively optimizes the third image clustering model using a cutting plane algorithm and determines whether the third image clustering model has converged. If it has, the module uses the third image clustering model to solve for the image clustering results; otherwise, the module controls the first optimization module to re-execute the optimization operation.

[0155] It should be noted that the foregoing explanation of the image clustering method based on multi-view learning and Top-k loss also applies to the image clustering system based on multi-view learning and Top-k loss in this embodiment, and will not be repeated here.

[0156] Example 4

[0157] Figure 3 This is a schematic diagram of the structure of the computer device 600 provided in this embodiment. The computer device 600 includes: a memory 601, a processor 602, and a computer program stored in the memory 601 and executable on the processor 602.

[0158] When the processor 602 executes the program, it implements the image clustering method based on multi-view learning and Top-k loss provided in the above embodiments.

[0159] Furthermore, the computer device 600 also includes a communication interface 603 for communication between the memory 601 and the processor 602.

[0160] The memory 601 may include high-speed RAM (Random Access Memory) and may also include non-volatile memory, such as at least one disk storage.

[0161] If the memory 601, processor 602, and communication interface 603 are implemented independently, then the communication interface 603, memory 601, and processor 602 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0162] Optionally, in a specific implementation, if the memory 601, processor 602, and communication interface 603 are integrated on a single chip, then the memory 601, processor 602, and communication interface 603 can communicate with each other through an internal interface.

[0163] The processor 602 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of this application.

[0164] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described image clustering method based on multi-view learning and Top-k loss.

[0165] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0166] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0167] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0168] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (FPGAs), field-programmable gate arrays (FPGAs), etc.

[0169] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware, and the program can be stored in a computer-readable storage medium. When executed, the program includes one or a combination of the steps of the method embodiments.

[0170] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. An image clustering method based on multi-view learning and Top-k loss, characterized in that, include: S1: Convert the original image dataset into a multi-view feature image dataset; S2: Based on the multi-view feature image dataset, construct a first image clustering model based on multi-view learning and Top-k loss; S3: Using the constrained concave-convex process algorithm, optimize the first image clustering model to obtain the second image clustering model; S4: Based on preset conditions, index and arrange the constraints of the second image clustering model to obtain the third image clustering model; S5: Use the cutting plane algorithm to iteratively optimize the third image clustering model, and determine whether the third image clustering model has converged. If it has, use the third image clustering model to solve for the image clustering results. If not, jump to S3.

2. The image clustering method based on multi-view learning and Top-k loss according to claim 1, characterized in that, The original image dataset is converted into a multi-view feature image dataset, including: Using several different image feature extraction methods, the original image dataset (x1, x2, ..., x...) was processed. N Each image x in ) i Transform into M feature vectors Obtain a multi-view feature image dataset in, Representing image x i The feature vector d under viewpoint v v Let v be the number of data dimensions for the viewpoint v, and N be the number of samples in the dataset.

3. The image clustering method based on multi-view learning and Top-k loss according to claim 2, characterized in that, After obtaining the multi-view feature image dataset, the method further includes: Set up K classification planes, that is The data in the multi-view feature image dataset is divided into K clusters; among them... Let p be the p-th classification plane in viewpoint v. Let be the normal vector of the p-th classification plane in viewpoint v.

4. The image clustering method based on multi-view learning and Top-k loss according to claim 3, characterized in that, Based on the aforementioned multi-view feature image dataset, a first image clustering model based on multi-view learning and Top-k loss is constructed, including: Where N is the number of data samples, K is the number of clusters, M is the number of perspectives, and m = MKN. It is a constant; Let i represent the slack variable of sample i under the v-th view. Let $\frac{ ... and , respectively, are the normal vectors of the p-th and q-th classification planes under the v-th viewpoint; C1 is a non-negative parameter related to clustering error, C2 is a non-negative parameter controlling the impact of viewpoint consistency on the learning model; k is a parameter in the Top-k loss, representing the number of allowed prediction errors; l is a non-negative parameter limiting cluster imbalance, and ε is a non-negative parameter; The classification plane index corresponding to the maximum predicted output value.

5. The image clustering method based on multi-view learning and Top-k loss according to claim 4, characterized in that, The first image clustering model is optimized using a constrained concavity / convexity process algorithm to obtain a second image clustering model, including: make definition in, From K d v Composed of dimensional vectors, the r-th vector is The other K-1 vectors are zero vectors; definition in, It consists of M vectors, and the v-th vector is The other M-1 vectors are zero vectors; By defining W and The first image clustering model Transformed into This will enable the first constraint of the first image clustering model to be applied. Transform into For this nonconvex constraint Initialize W (t) Value, used in W (t) First-order Taylor expansion of a point By substitution, we obtain the second image clustering model, whose expression is shown below: in, in, The category plane index represents the maximum predicted output value, while This represents the classification plane index of the q-th largest predicted output value.

6. The image clustering method based on multi-view learning and Top-k loss according to claim 5, characterized in that, Based on preset conditions, the constraints of the second image clustering model are indexed and arranged to obtain a third image clustering model, including: Define variables X1, X2, ..., X m+n+h and variables γ1, γ2, ..., γ m+n Where h = MK(K-1) / 2; For X1, X2, ..., X m and γ1,γ2,…,γ m , constrain conditions In Arrange and reindex as X i and order For X m ,X m+1 ,…,X m+n and γ m+1 ,γ m+2 ,…,γ m+h , constrain conditions of and constraints of Arrange and reindex as X i and -X i and order For X m+n+1 ,X m+n+2 ,…,X m+n+h The constraints in the second image clustering model of Arrange and reindex as X i ; After completing the above arrangement and indexing, the expression for the third image clustering model is as follows: W·X i ≥1-c i W·X i ≥-e-g i -W·X i ≥-e-g i -l≤W·X i ≤l。 7. The image clustering method based on multi-view learning and Top-k loss according to claim 6, characterized in that, The third image clustering model is iteratively optimized using the cutting plane algorithm, and the image clustering results are obtained using the third image clustering model, including: The third image clustering model is converted into a single-relaxation form, the expression of which is as follows: -l≤W·X i ≤l in: Among them, ξ≥0 and δ≥0 are single slack variables; Define constraint set and The two most violated constraints c are found iteratively using the current solution (W, ζ, η) through the cutting plane algorithm. i and e i and respectively c i and e i Add to Ω ε and Ω η In the process, until the solution (W,ζ,η) satisfies all constraints and reaches the accuracy ∈, the third image clustering model that satisfies the accuracy condition is obtained; Solving using a third image clustering model that meets the accuracy requirements. The value of , and according to the following formula for multi-view samples Clustering is performed to obtain the image clustering results: Among them, y i For multi-view samples x i The predicted labels, i.e., the image clustering results, For multi-view samples x i Compared to the predicted output value of the p-th plane at the v-th viewpoint, For sample x i The average predicted output value relative to the p-th plane across all viewpoints.

8. An image clustering system based on multi-view learning and Top-k loss, characterized in that, include: The conversion module is used to convert the original image dataset into a multi-view feature image dataset; The building module is used to construct a first image clustering model based on multi-view learning and Top-k loss based on the multi-view feature image dataset; The first optimization module is used to optimize the first image clustering model using a constrained concave-convex process algorithm to obtain a second image clustering model. The indexing and arranging module is used to index and arrange the constraints of the second image clustering model according to preset conditions to obtain the third image clustering model; The second optimization module is used to iteratively optimize the third image clustering model using the cutting plane algorithm, and to determine whether the third image clustering model has converged. If it has, the third image clustering model is used to solve for the image clustering result. If not, the first optimization module is controlled to re-execute the optimization operation.

9. An electronic device, characterized in that, The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to perform the operations performed by the image clustering method based on multi-view learning and Top-k loss as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that is executed by a processor as described in any one of claims 1 to 7, using the image clustering method based on multi-view learning and Top-k loss.

Images