A subspace clustering method, device, equipment and storage medium

Abstract

The present embodiment provides a subspace clustering method, apparatus, equipment and storage medium, the method comprising: performing feature extraction, self-representation processing, decoding processing and other operations on the original data features through a preset neural network model to obtain a target loss function, wherein the nuclear norm is used to impose a low-rank prior on the preset self-representation coefficient matrix for self-representation processing. The neural network model is trained through the target loss function, the target self-representation coefficient matrix is ​​extracted from the trained convolutional autoencoder network model, and the target similarity matrix is ​​further obtained, and then the target similarity matrix is ​​segmented using a spectral clustering algorithm to obtain a subspace clustering result. The present application realizes the learning of the self-representation coefficient matrix based on the nuclear norm and the convolutional autoencoder network model, thereby obtaining a similarity matrix with higher accuracy, and finally obtaining a subspace clustering result with higher accuracy. The method of the present application can improve the accuracy of subspace clustering.

Description

Technical Field

[0001] The present application relates to the field of pattern recognition technology, and in particular to a subspace clustering method, apparatus, device and storage medium. Background Art

[0002] With the continuous development of science and technology, the world is becoming increasingly information-based. The ways people access data are becoming more diverse, and data in various forms (such as text, images, audio, and video) is generated constantly. This massive amount of data plays a vital role in every aspect of our lives. The analysis and processing of large-scale data has become a crucial part of current scientific research.

[0003] How to effectively and efficiently process multi-source, heterogeneous, high-dimensional data in real life has become a challenge in the field of clustering. Subspace clustering aims to divide samples into different clusters based on their similarities and find a low-dimensional subspace representation, assuming that the samples do not have label information. It mainly consists of two steps: first, constructing a similarity matrix to describe the relationships between the datasets; then, applying a clustering algorithm to the similarity matrix to obtain the final clustering results. Therefore, the quality of the similarity matrix largely determines the clustering performance. However, traditional subspace clustering methods fail to integrate the acquisition of the data latent representation with the clustering process, fail to fully utilize local latent information, and the constructed similarity matrix is ​​not very accurate, resulting in poor clustering results. Summary of the Invention

[0004] The main purpose of the embodiments of the present application is to propose a subspace clustering method, device, equipment and storage medium, which can.

[0005] To achieve the above objectives, a first aspect of an embodiment of the present application proposes a subspace clustering method, including:

[0006] Get the original data features;

[0007] Extracting features from the original data using an encoding module of a preset neural network model to obtain potential data features;

[0008] Performing self-representation processing on the potential data features by a self-representation module of a preset neural network model to obtain target data features, wherein a preset self-representation coefficient matrix is ​​subjected to low-rank constraint processing using a nuclear norm to obtain a preliminary self-representation coefficient matrix, and the self-representation module performs self-representation processing on the potential data features by using the preliminary self-representation coefficient matrix;

[0009] Decoding the latent data features through a decoding module of a preset neural network model to obtain restored data features;

[0010] Constructing a target loss function according to the preliminary self-representation coefficient matrix, the original data features, the potential data features, the target data features, and the restored data features;

[0011] Training the neural network model according to the target loss function to obtain a target self-representation coefficient matrix;

[0012] Processing the target self-representation coefficient matrix using a preset algorithm to obtain a target similarity matrix;

[0013] The target similarity matrix is ​​segmented using a spectral clustering algorithm to obtain a subspace clustering result.

[0014] In some embodiments, the step of performing low-rank constraint processing on a preset self-expressive coefficient matrix using a nuclear norm to obtain a preliminary self-expressive coefficient matrix includes:

[0015] Performing low-rank constraint processing on the self-representation coefficient matrix using a nuclear norm to obtain a low-rank coefficient matrix;

[0016] Performing singular value decomposition on the low-rank coefficient matrix to obtain multiple singular value components;

[0017] If there is at least one singular value component greater than 1, a low-rank constraint process is performed on the self-representative coefficient matrix using a γ-norm to obtain a preliminary self-representative coefficient matrix.

[0018] In some embodiments, constructing a target loss function based on the preliminary self-representation coefficient matrix, the original data features, the potential data features, the target data features, and the restored data features includes:

[0019] Obtaining a preliminary similarity matrix according to the preliminary self-representation coefficient matrix;

[0020] Obtaining a first target loss function according to the preliminary similarity matrix, the original data features, and the restored data features;

[0021] Obtaining a second objective loss function according to the preliminary similarity matrix, the potential data features, and the target data features;

[0022] A target loss function is obtained according to the first target loss function and the second target loss function.

[0023] In some embodiments, obtaining a first target loss function according to the preliminary similarity matrix, the original data features, and the restored data features includes:

[0024] Obtaining a first clustering loss function according to the original data features and the restored data features;

[0025] The first clustering loss function is reconstructed according to the preliminary similarity matrix to obtain a first target loss function.

[0026] In some embodiments, reconstructing the first clustering loss function according to the preliminary similarity matrix to obtain a first target loss function includes:

[0027] Reconstructing the first clustering loss function according to the preliminary similarity matrix to obtain a first reconstruction target loss function;

[0028] The first reconstruction target loss function is regularized according to a preset Laplace constraint to obtain a first target loss function.

[0029] In some embodiments, obtaining a second target loss function based on the preliminary similarity matrix, the potential data features, and the target data features includes:

[0030] Obtaining a second clustering loss function according to the potential data features and the target data features;

[0031] Reconstructing the potential data features according to the preliminary similarity matrix to obtain a similarity matching function;

[0032] Obtaining a second reconstruction target loss function according to the second clustering loss function and the similarity matching function;

[0033] The second reconstruction objective loss function is regularized according to a preset Laplace constraint to obtain a second objective loss function.

[0034] In some embodiments, the processing of the target self-representation coefficient matrix using a preset algorithm to obtain a target similarity matrix includes:

[0035] Get the target self-representation coefficient matrix R, let the matrix

[0036] Calculate the singular value decomposition of the matrix Sg and get Sg=UΣV T ;

[0037] make Get the target similarity matrix A = [PP T ] α , where m = k*d+1, K is the number of clusters, and d is the dimension of the original data features.

[0038] A second aspect of the embodiments of the present application provides a subspace clustering device, including:

[0039] Acquisition module, used to obtain raw data features;

[0040] An encoding processing module, configured to extract features from the original data using an encoding module of a preset neural network model to obtain potential data features;

[0041] a self-representation processing module, configured to perform self-representation processing on the potential data features through a self-representation module of a preset neural network model to obtain target data features, wherein a preset self-representation coefficient matrix is ​​subjected to low-rank constraint processing using a nuclear norm to obtain a preliminary self-representation coefficient matrix, and the self-representation module performs self-representation processing on the potential data features through the preliminary self-representation coefficient matrix;

[0042] A decoding processing module, configured to decode the potential data features using a decoding module of a preset neural network model to obtain restored data features;

[0043] A target loss function construction module is used to construct a target loss function based on the preliminary self-representation coefficient matrix, the original data features, the potential data features, the target data features, and the restored data features;

[0044] A model training module, configured to train the neural network model according to the target loss function to obtain a target self-representation coefficient matrix;

[0045] A target similarity matrix obtaining module is used to process the target self-representation coefficient matrix using a preset algorithm to obtain a target similarity matrix;

[0046] The subspace clustering module is used to segment the target similarity matrix using a spectral clustering algorithm to obtain a subspace clustering result.

[0047] The third aspect of the embodiments of the present application proposes a computer device, which includes a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor is used to execute a subspace clustering method as described in any one of the embodiments of the first aspect of the present application.

[0048] The fourth aspect of the embodiments of the present application proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a computer, the computer is used to execute a subspace clustering method as described in any one of the embodiments of the first aspect of the present application.

[0049] The embodiment of the present application proposes a subspace clustering method, apparatus, device and storage medium, which obtains a target loss function by performing feature extraction, self-representation processing, decoding processing and other operations on the original data features through a preset neural network model, wherein the nuclear norm is used to impose a low-rank prior on the preset self-representation coefficient matrix for self-representation processing. The neural network model is trained with the target loss function to obtain a target self-representation coefficient matrix, and further obtains a target similarity matrix, and then the target similarity matrix is ​​segmented using a spectral clustering algorithm to obtain a subspace clustering result. The present application realizes the learning of the self-representation coefficient matrix based on the nuclear norm and the convolutional autoencoder network model, and then obtains a more accurate similarity matrix, and finally obtains a more accurate subspace clustering result. The embodiment of the present application can improve the accuracy of the subspace clustering result. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a flow chart of a subspace clustering method provided in an embodiment of the present application;

[0051] Figure 2 This is a flowchart of step S150 in an embodiment of the present application;

[0052] Figure 3 This is a flowchart of step S220 in an embodiment of the present application;

[0053] Figure 4 This is a flowchart of step S230 in an embodiment of the present application;

[0054] Figure 5 This is a schematic diagram of the hardware structure of the computer device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0056] It should be noted that although the device schematics illustrate functional module divisions and the flowcharts illustrate logical sequences, in certain circumstances, the steps shown or described may be performed in a sequence that differs from the module divisions in the device or the sequence in the flowcharts. The terms "first," "second," and so on, used in the specification, claims, and drawings, are used to distinguish similar items and are not necessarily used to describe a specific sequence or precedence.

[0057] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein are for the purpose of describing the embodiments of this application only and are not intended to limit this application.

[0058] First, let’s analyze some of the terms used in this application:

[0059] Artificial Intelligence (AI) is a new technical discipline that studies and develops theories, methods, technologies, and application systems for simulating, extending, and expanding human intelligence. A branch of computer science, AI seeks to understand the essence of intelligence and create new intelligent machines that can respond in a manner similar to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing, and expert systems. AI can simulate the information processes of human consciousness and thinking. It also encompasses theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, to perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results.

[0060] Spectral Clustering (SC) is based on spectral graph theory. Compared with traditional clustering algorithms, it can cluster in sample spaces of arbitrary shapes and converge to a global optimal solution. The main idea of ​​this algorithm is to treat each object in the dataset as a vertex of a graph, which can be connected by edges. The similarity between the vertices is quantified as the weight of the edge connecting the corresponding vertices. The weight of the edge between two points with a large distance is low, while the weight of the edge between two points with a large distance is high. Then, by cutting the graph composed of all data points, the sum of the weights of the edges between different subgraphs after cutting is made as low as possible, and the sum of the weights of the edges within a subgraph is made as high as possible, thus achieving the purpose of clustering.

[0061] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to achieve optimal results.

[0062] The subspace clustering method provided in the embodiments of this application can be applied to artificial intelligence. Basic AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0063] Subspace clustering aims to group samples into clusters based on their similarities and find a low-dimensional subspace representation, even without label information. It involves two main steps: first, constructing a similarity matrix to describe the relationships between datasets; then, applying a clustering algorithm to the similarity matrix to obtain the final clustering results. Therefore, the quality of the similarity matrix largely determines clustering performance. However, traditional subspace clustering methods fail to integrate the acquisition of the data's latent representation with the clustering process, failing to fully utilize local latent information. Consequently, the constructed similarity matrix is ​​inaccurate, leading to poor clustering results.

[0064] Based on this, the main purpose of the embodiments of the present application is to propose a subspace clustering method, device, equipment and storage medium, which can realize the learning of the self-representation coefficient matrix through the nuclear norm and convolutional autoencoder network model, and then obtain a more accurate similarity matrix, and finally obtain a more accurate subspace clustering result, which can improve the accuracy of subspace clustering.

[0065] The subspace clustering method provided in the embodiments of the present application can be applied to a terminal, a server, or software running on a terminal or a server. In some embodiments, the terminal can be a smartphone, a tablet computer, a laptop computer, a desktop computer, or a smartwatch; the server can be configured as an independent physical server, or as a server cluster or distributed system consisting of multiple physical servers; or as a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the above method, but is not limited to the above forms.

[0066] Embodiments of the present application can be used in numerous general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer computer devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and the like. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected via a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0067] Reference Figure 1 According to an embodiment of the present application, a subspace clustering method includes but is not limited to steps S110 to S180.

[0068] S110, obtaining original data features;

[0069] S120, extracting features from the original data using an encoding module of a preset neural network model to obtain latent data features;

[0070] S130, performing self-representation processing on the potential data features by a self-representation module of a preset neural network model to obtain target data features, wherein a low-rank constraint is performed on a preset self-representation coefficient matrix using a nuclear norm to obtain a preliminary self-representation coefficient matrix, and the self-representation module performs self-representation processing on the potential data features using the preliminary self-representation coefficient matrix;

[0071] S140, decoding the latent data features through a decoding module of a preset neural network model to obtain restored data features;

[0072] S150, constructing a target loss function based on the preliminary self-representation coefficient matrix, the original data features, the latent data features, the target data features, and the restored data features;

[0073] S160, training the neural network model according to the target loss function to obtain a target self-representation coefficient matrix;

[0074] S170, processing the target self-representation coefficient matrix using a preset algorithm to obtain a target similarity matrix;

[0075] S180, using a spectral clustering algorithm to segment the target similarity matrix and obtain a subspace clustering result.

[0076] In step S110, the original data features are obtained. Specifically, the original data features are given a set of data sets. Among them, X n Represents the nth original data feature, n represents the number of original data features, and d represents the dimension of the original data feature.

[0077] In step S120, the original data features are extracted by the encoding module of the preset neural network model to obtain the potential data features. Specifically, the encoding module can be a convolutional encoder, which extracts the potential data features through the convolutional encoder. Specifically, the input of the convolutional encoder is the original data feature X, and after passing through multiple convolutional layers, the output is the potential data feature code Extracting potential data features through a convolutional encoder can overcome the nonlinearity of the original data features and further explore the nonlinear geometric information of the original data features, which is beneficial to improving the accuracy of the similarity matrix. It should be noted that the convolutional encoder can be regarded as a function that has both dimensionality reduction and nonlinear transformation, denoted by F. The convolutional decoder is used to reconstruct the input and is represented by G. The parameters of the convolutional encoder and convolutional decoder are denoted by Θ and Θ respectively. e and Θ d express.

[0078] In step S130, the self-representation module of the preset neural network model performs self-representation processing on the latent data features to obtain target data features. The pre-set self-representation coefficient matrix is ​​subjected to low-rank constraint processing using the nuclear norm to obtain a preliminary self-representation coefficient matrix. The self-representation module performs self-representation processing on the latent data features using the preliminary self-representation coefficient matrix. Specifically, the self-representation module is a fully connected layer without an activation function, and the weight matrix of the self-representation module is set to the self-representation coefficient matrix C to be learned by the preset neural network model.

[0079] In one embodiment, X belongs to K linear subspaces The subspace dimensions are The subspace clustering problem aims to identify these K subspaces. Assuming that there are enough data samples belonging to a linear subspace and that they span the entire subspace, then any data sample can be represented as a linear combination of the other data samples removed from it, i.e., the "self-expression" property of the original data features. Specifically, the input is the original data feature X, and after the input is self-expressed by the self-expression coefficient matrix, the target data feature XC is obtained. The corresponding subspace learning model is: Where C represents the self-representation coefficient matrix of the preset original data features, and the dimension of C is the same as that of the original data features. Specifically, C i Represents the i-th original data feature X i The coefficient vector represented by other original data features, ‖·‖ F represents the Frobenius norm, λ is the preset parameter, ||C|| p represents the regularization term, ||·|| p is any matrix norm.

[0080] In this embodiment, the nuclear norm is used to perform low-rank constraint processing on the preset self-representation coefficient matrix to obtain a preliminary self-representation coefficient matrix, which is specifically shown in formula (1):

[0081]

[0082] in,‖·‖ * represents the nuclear norm, σ iis the i-th singular value of C, and st represents the constraints that the original data features must satisfy. It can be understood that by applying a low-rank constraint to the preset self-representation coefficient matrix through the nuclear norm, the resulting preliminary self-representation coefficient matrix can achieve a low-rank prior, further benefiting the accuracy of target similarity. It should be noted that the low-rank prior is a better regularization condition than the sparse prior, because sparse representation represents all samples individually, while low-rank representation represents all samples by finding the lowest-rank joint representation of the data features.

[0083] In one embodiment, step S130 specifically includes the following steps:

[0084] The low-rank coefficient matrix is ​​obtained by using the nuclear norm to perform low-rank constraint processing on the self-representation coefficient matrix;

[0085] Perform singular value decomposition on the low-rank coefficient matrix to obtain multiple singular value components;

[0086] If there is at least one singular value component greater than 1, the self-representative coefficient matrix is ​​subjected to low-rank constraint processing using the γ-norm to obtain a preliminary self-representative coefficient matrix.

[0087] Specifically, when the singular value component σ i When it is greater than 1, it means that the nuclear norm deviates significantly from 1, indicating that the rank component is over-contracted. Therefore, the γ-norm is used instead of the nuclear norm, as shown in formula (2):

[0088]

[0089] in,‖·‖ γ represents the γ-norm, σ i is the i-th singular value of C, γ is a hyperparameter with a value greater than 1, and st represents the constraints that the original data features need to meet.

[0090] In step S140, the latent data features are decoded by the decoding module of the preset neural network model to obtain the restored data features. Specifically, the decoding module is a convolutional decoder, which corresponds to the convolutional encoder and takes the latent data features as input. After the convolution layer symmetrical to the convolution encoder, the restored data features of the original data features X are output

[0091] In step S150 , a target loss function may be constructed based on the preliminary self-representation coefficient matrix, the original data features, the latent data features, the target data features, and the restored data features.

[0092] Specifically, the first clustering loss function shown in formula (3) can be obtained according to the original data characteristics and the restored data characteristics:

[0093]

[0094] in, represents the square of the Frobenius norm, X i represents the i-th original data feature, Indicated as X i The recovery data features.

[0095] According to the preliminary self-representation coefficient matrix, potential data features, and target data features, the second clustering loss function shown in formula (4) is obtained:

[0096]

[0097] Where C represents the preliminary self-representation coefficient matrix, X represents the potential data features, XC represents the target data features, ‖*‖ γ represents the γ-norm, represents the square of the Frobenius norm.

[0098] The target loss function is obtained by combining the first cluster loss function with the second cluster loss function, that is, formula (5) is obtained according to formulas (3) and (4), as shown below:

[0099]

[0100] in, Unify the parameters of the target loss function as shown in formula (6):

[0101]

[0102] in,

[0103] In step S160, the neural network model is trained according to the target loss function, and a target self-representation coefficient matrix is ​​extracted from the trained neural network model. Specifically, the neural network model is trained using the target loss function to obtain a convolutional autoencoder network model; the trained self-representation coefficient matrix is ​​extracted from the convolutional autoencoder network model, and the trained self-representation coefficient matrix is ​​used as the target self-representation coefficient matrix.

[0104] In step S170, the target self-expression coefficient matrix is ​​processed using a preset algorithm to obtain a target similarity matrix. Specifically, the target self-expression coefficient matrix R is obtained, and the matrix In one embodiment, the matrix Sg can be directly used as the target similarity matrix. In another embodiment, the singular value decomposition of the matrix Sg is calculated to obtain Sg=UΣV T ,make Get the target similarity matrix A = [PP T ]α , where m = k*d+1, K is the number of clusters, d is the dimension of the original data features, and α is a hyperparameter. It should be noted that K is known, and d and α can be selected based on experience.

[0105] In step S180, the target similarity matrix is ​​segmented using a spectral clustering algorithm to obtain subspace clustering results. Specifically, the obtained target similarity matrix A is applied to the spectral clustering algorithm to obtain subspace clustering results.

[0106] Through steps S110 to S180, in order to improve the effect of subspace clustering, the present application utilizes the global low-rank prior and nonlinear geometric information in the original data features to further improve the representation ability of the similarity matrix. Specifically, an automatic convolution encoder is used to solve the nonlinear problem of the input original data features, and a self-representation module between the encoder and the decoder is used to learn a self-representation coefficient matrix that is conducive to deep subspace clustering, and the nuclear norm is used to achieve low-rank prior. Through the subspace clustering method of the present application, the accuracy of the target similarity can be improved, thereby improving the accuracy of the subspace clustering results.

[0107] In one embodiment, referring to Figure 2 , step S150 specifically includes steps S210 to S240.

[0108] S210, obtaining a preliminary similarity matrix according to the preliminary self-representation coefficient matrix;

[0109] S220, obtaining a first target loss function based on the preliminary similarity matrix, the original data features, and the restored data features;

[0110] S230, obtaining a second target loss function based on the preliminary similarity matrix, the potential data features, and the target data features;

[0111] S240: Obtain a target loss function according to the first target loss function and the second target loss function.

[0112] Specifically, in order to preserve the local geometric results, the potential correlation between data features is reconstructed and represented through the preliminary similarity matrix, and more local information is further mined, so that the final target loss function can better reflect the potential correlation of the original data features. In one embodiment, the preliminary self-representation coefficient matrix C is obtained, and the matrix The matrix S is directly used as the preliminary similarity matrix.

[0113] In one embodiment, referring to Figure 3 , step S220 specifically includes steps S310 to S330.

[0114] S310, obtaining a first clustering loss function according to the original data features and the restored data features;

[0115] S320, reconstructing the first clustering loss function according to the preliminary similarity matrix to obtain a first reconstruction target loss function;

[0116] S330 , regularizing the first reconstruction target loss function according to a preset Laplace constraint to obtain a first target loss function.

[0117] Specifically, the first clustering loss function is obtained based on the original data features and the restored data features. The first clustering loss function refers to formula (3). The matrix S is used to represent the similarity between data features, where S ij is a sample X i and X j The similarity between X i With weight S ij of For reconstruction, the similarity of samples with larger distance should be lower. The first reconstruction objective loss function is shown in formula (7):

[0118]

[0119] It can be understood that the neural network model can be trained by minimizing formula (7). Compared with self-reconstruction, formula (7) can express the neighborhood relationship of data features. In other words, the learned potential data features are encoded by the neighborhood relationship. Formula (7) can be further evolved as follows:

[0120]

[0121] in, L=DS is the Laplace matrix. If we use the regularized degree matrix D and the regularized similarity matrix S, we can easily get

[0122] In one embodiment, referring to Figure 4 , step S230 specifically includes steps S410 to S440.

[0123] S410, obtaining a second clustering loss function according to the potential data features and the target data features;

[0124] S420, reconstructing the potential data features according to the preliminary similarity matrix to obtain a similarity matching function;

[0125] S430, obtaining a second reconstruction target loss function according to the second clustering loss function and the similarity matching function;

[0126] S440 , regularizing the second reconstruction target loss function according to a preset Laplace constraint to obtain a second target loss function.

[0127] Specifically, in addition to exploiting the locality hidden in the reconstructed output, we can also consider using local information in the latent representation obtained by the encoder. Specifically, since the columns of Z are the new representations of the data under a specific dictionary (all data points themselves), Z i and Z j The distance between the original data points is actually i and X j Therefore, consider adding a similarity matching regularization term to the objective loss function, namely the similarity matching function: Among them, Z i and Z j are data points X i With X j The potential representation of

[0128] In one embodiment, two types of Laplacian operators are used to capture nonlinear geometric information constraints. The first type of Laplacian constraint is imposed on the underlying data features. The second one is applied to the original data feature X and the restored data feature Combining the above two Laplace constraints, we can get the following objective loss function:

[0129]

[0130] It should be noted that the neural network model is trained using the gradient descent algorithm with formula (9) as the loss function until the network converges. A highly accurate self-representation coefficient matrix can be obtained from the trained network. The similarity matrix obtained based on the self-representation coefficient matrix includes useful information such as the nonlinear geometric information and potential representation information of the original data features, which helps to improve the accuracy of the subspace clustering results.

[0131] The embodiment of the present application proposes a novel subspace clustering method, which simultaneously utilizes global low-rank priors and nonlinear geometric information, so that the representation ability of the similarity matrix can be further improved. The embodiment of the present application utilizes an autoencoder to solve the nonlinear problem of the input raw data features, utilizes the self-representation module between the encoder and the decoder to learn the potential representation that is conducive to deep subspace clustering, and utilizes the nuclear norm to implement the low-rank prior. It also uses two types of Laplace operators to capture the constraints of nonlinear geometric information. In addition, an embodiment of the present application also utilizes the γ-norm to map each singular value of the self-representation coefficient matrix to a smaller interval, and then performs a summation to prevent the adverse effects of extreme singular values. Experiments show that the subspace clustering method of the embodiment of the present application can well complete the task of clustering faces and object pictures, and its performance is better than the current subspace clustering method.

[0132] The above algorithm in the embodiment of the present application can be implemented in common neural network frameworks, such as TensorFlow and PyTorch. The neural network model architecture parameter configuration in the embodiment of the present application is shown in Table 1, and the other hyperparameters of the model are shown in Table 2. The experimental results of the embodiment of the present application on five commonly used data sets are shown in Table 3.

[0133] Table 1:

[0134]

[0135] Among them, 5×5@15 means: convolution kernel height (5)×convolution kernel width (5)@number of convolution kernels (15).

[0136] Table 2:

[0137]

[0138] Among them, Table 2 lists the hyperparameters, γ, and learning rate learning_rate in the loss function of each data set.

[0139] Table 3:

[0140]

[0141] The embodiment of the present application also proposes a subspace clustering device, which includes an acquisition module for acquiring original data features; an encoding processing module for extracting features of the original data features through the encoding module of the preset neural network model to obtain potential data features; a self-representation processing module for performing self-representation processing on the potential data features through the self-representation module of the preset neural network model to obtain target data features, wherein the nuclear norm is used to perform low-rank constraint processing on the preset self-representation coefficient matrix to obtain a preliminary self-representation coefficient matrix, and the self-representation module performs self-representation processing on the potential data features through the preliminary self-representation coefficient matrix; a decoding processing module for The latent data features are decoded by the decoding module of the preset neural network model to obtain the restored data features; a target loss function construction module is used to construct the target loss function based on the preliminary self-representation coefficient matrix, original data features, latent data features, target data features, and restored data features; a model training module is used to train the neural network model through the target loss function to obtain the target self-representation coefficient matrix; a target similarity matrix acquisition module is used to process the target self-representation coefficient matrix using a preset algorithm to obtain a target similarity matrix; a subspace clustering module is used to segment the target similarity matrix using a spectral clustering algorithm to obtain a subspace clustering result.

[0142] In order to improve the effect of subspace clustering, a subspace clustering device of an embodiment of the present application utilizes the global low-rank prior and nonlinear geometric information in the original data features to further improve the representation ability of the similarity matrix. Specifically, an automatic convolution encoder is used to solve the nonlinear problem of the input original data features, and a self-representation module between the encoder and the decoder is used to learn the self-representation coefficient matrix that is conducive to deep subspace clustering, and the nuclear norm is used to achieve low-rank prior. Through the subspace clustering device of the present application, the accuracy of the target similarity can be improved, thereby improving the accuracy of the subspace clustering results.

[0143] A subspace clustering device in an embodiment of the present application is used to execute a subspace clustering method in the above embodiment. Its specific processing process is the same as that of a subspace clustering method in the above embodiment, and will not be repeated here.

[0144] An embodiment of the present application further provides a computer device, comprising a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor is used to execute a subspace clustering method as described in any one of the embodiments of the present application.

[0145] The following combination Figure 5 The hardware structure of the computer device is described in detail. The computer device includes: a processor 501 , a memory 502 , an input / output interface 503 , a communication interface 504 , and a bus 505 .

[0146] The processor 501 may be implemented as a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is configured to execute relevant programs to implement the technical solutions provided in the embodiments of the present application.

[0147] The memory 502 can be implemented in the form of ROM (Read Only Memory), static storage device, dynamic storage device, or RAM (Random Access Memory). The memory 502 can store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 502 and is called by the processor 501 to execute a subspace clustering method of the embodiments of this application.

[0148] Input / output interface 503, used to implement information input and output;

[0149] Communication interface 504, used to implement communication interaction between this device and other devices, which can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WiFi, Bluetooth, etc.); and bus 505, used to transmit information between various components of the device (such as processor 501, memory 502, input / output interface 503 and communication interface 504);

[0150] The processor 501 , the memory 502 , the input / output interface 503 and the communication interface 504 are connected to each other in communication within the device via a bus 505 .

[0151] An embodiment of the present application further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a computer, the computer is used to execute a subspace clustering method as described in any one of the embodiments of the present application.

[0152] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0153] The embodiments described in the embodiments of this application are intended to more clearly illustrate the technical solutions of the embodiments of this application and do not constitute a limitation on the technical solutions provided by the embodiments of this application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0154] It will be understood by those skilled in the art that Figures 1 to 4 The technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than those shown in the figures, or a combination of certain steps, or different steps.

[0155] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, i.e., they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment.

[0156] Those skilled in the art will appreciate that all or some of the steps in the methods, systems, and functional modules / units in the devices disclosed above may be implemented as software, firmware, hardware, or appropriate combinations thereof.

[0157] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0158] It should be understood that in this application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0159] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0160] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0161] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0162] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store programs, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0163] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but are not intended to limit the scope of the present invention. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and essence of the present invention should be within the scope of the present invention.

Claims

1. A subspace clustering method, characterized in that: include: Acquire original data features; wherein the original data features are image data; Extracting features from the original data using an encoding module of a preset neural network model to obtain latent data features; wherein the encoding module is a convolutional encoder; The self-representation module of a preset neural network model performs self-representation processing on the potential data features to obtain target data features, wherein a preset self-representation coefficient matrix is ​​subjected to low-rank constraint processing using a nuclear norm to obtain a preliminary self-representation coefficient matrix, and the self-representation module performs self-representation processing on the potential data features using the preliminary self-representation coefficient matrix; wherein the self-representation module is a fully connected layer without an activation function; Decoding the latent data features through a decoding module of a preset neural network model to obtain restored data features; wherein the decoding module is a convolutional decoder, and the convolutional decoder corresponds to the convolutional encoder; Constructing a target loss function according to the preliminary self-representation coefficient matrix, the original data features, the potential data features, the target data features, and the restored data features; Training the neural network model according to the target loss function to obtain a target self-representation coefficient matrix; Processing the target self-representation coefficient matrix using a preset algorithm to obtain a target similarity matrix; The target similarity matrix is ​​segmented using a spectral clustering algorithm to obtain a subspace clustering result.

2. A subspace clustering method according to claim 1, characterized in that: The method of using the nuclear norm to perform low-rank constraint processing on the preset self-expressive coefficient matrix to obtain a preliminary self-expressive coefficient matrix includes: Performing low-rank constraint processing on the self-representation coefficient matrix using a nuclear norm to obtain a low-rank coefficient matrix; Performing singular value decomposition on the low-rank coefficient matrix to obtain multiple singular value components; If there is at least one singular value component greater than 1, a low-rank constraint process is performed on the self-representative coefficient matrix using a γ-norm to obtain a preliminary self-representative coefficient matrix.

3. A subspace clustering method according to claim 1, characterized in that: The constructing of a target loss function according to the preliminary self-representation coefficient matrix, the original data features, the potential data features, the target data features, and the restored data features includes: Obtaining a preliminary similarity matrix according to the preliminary self-representation coefficient matrix; Obtaining a first target loss function according to the preliminary similarity matrix, the original data features, and the restored data features; Obtaining a second objective loss function according to the preliminary similarity matrix, the potential data features, and the target data features; A target loss function is obtained according to the first target loss function and the second target loss function.

4. A subspace clustering method according to claim 3, characterized in that: The obtaining of a first target loss function according to the preliminary similarity matrix, the original data features, and the restored data features includes: Obtaining a first clustering loss function according to the original data features and the restored data features; The first clustering loss function is reconstructed according to the preliminary similarity matrix to obtain a first target loss function.

5. A subspace clustering method according to claim 4, characterized in that: The reconstructing the first clustering loss function according to the preliminary similarity matrix to obtain a first target loss function includes: Reconstructing the first clustering loss function according to the preliminary similarity matrix to obtain a first reconstruction target loss function; The first reconstruction target loss function is regularized according to a preset Laplace constraint to obtain a first target loss function.

6. A subspace clustering method according to claim 3, characterized in that: The obtaining of a second target loss function according to the preliminary similarity matrix, the potential data features, and the target data features includes: Obtaining a second clustering loss function according to the potential data features and the target data features; Reconstructing the potential data features according to the preliminary similarity matrix to obtain a similarity matching function; Obtaining a second reconstruction target loss function according to the second clustering loss function and the similarity matching function; The second reconstruction objective loss function is regularized according to a preset Laplace constraint to obtain a second objective loss function.

7. A subspace clustering method according to claim 1, characterized in that: The process of processing the target self-representation coefficient matrix by using a preset algorithm to obtain a target similarity matrix includes: Get the target self-representation coefficient matrix R, let the matrix Calculate the singular value decomposition of the matrix Sg and get Sg=UΣV T ; make Get the target similarity matrix A = [PP T ] α , where m = k*d+1, K is the number of clusters, d is the dimension of the original data features, and α is the hyperparameter.

8. A subspace clustering device, characterized in that: include: An acquisition module, configured to acquire original data features, wherein the original data features are image data; An encoding processing module, configured to extract features of the original data using an encoding module of a preset neural network model to obtain latent data features; wherein the encoding module is a convolutional encoder; a self-representation processing module, configured to perform self-representation processing on the potential data features through a self-representation module of a preset neural network model to obtain target data features, wherein a preset self-representation coefficient matrix is ​​subjected to low-rank constraint processing using a nuclear norm to obtain a preliminary self-representation coefficient matrix, and the self-representation module performs self-representation processing on the potential data features through the preliminary self-representation coefficient matrix; wherein the self-representation module is a fully connected layer without an activation function; A decoding processing module, configured to decode the potential data features through a decoding module of a preset neural network model to obtain restored data features; wherein the decoding module is a convolutional decoder, and the convolutional decoder corresponds to the convolutional encoder; Constructing a target loss function module, for constructing a target loss function based on the preliminary self-representation coefficient matrix, the original data features, the potential data features, the target data features, and the restored data features; A model training module, configured to train the neural network model according to the target loss function to obtain a target self-representation coefficient matrix; A target similarity matrix obtaining module is used to process the target self-representation coefficient matrix using a preset algorithm to obtain a target similarity matrix; The subspace clustering module is used to segment the target similarity matrix using a spectral clustering algorithm to obtain a subspace clustering result.

9. A computer device, characterized in that: The computer device includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor is configured to perform: A subspace clustering method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program. When the computer program is executed by a computer, the computer is configured to: A subspace clustering method according to any one of claims 1 to 7.

Images