Multi-view data classification method and system based on fast canonical correlation analysis algorithm

Abstract

The present invention relates to a multi-view data classification method and system based on a canonical correlation analysis fast algorithm. The classification method comprises the following steps: inputting multi-view data into a data input module, preprocessing, and storing in a database; extracting data from the database, and calculating a similarity matrix and a covariance tensor in a data calculation module; inputting the similarity matrix and covariance tensor calculated in the data calculation module into a result output module, calculating a matrix after feature extraction in the result output module, and classifying the matrix. The classification system comprises a data input module, a data calculation module, and a result output module. Aiming at the problem of low accuracy in multi-view data classification, the present invention proposes a NLLTCCA method, solves the problem that the current tensor canonical correlation analysis takes too long to process multi-view data, improves the classification accuracy of multi-view data, reduces the training time and training cost of the model, and enhances the portability of the model.

Description

Technical Field

[0001] The present invention relates to the field of multi-view data classification, and in particular to a multi-view data classification method and system based on a canonical correlation analysis fast algorithm. The present invention can be applied to the field of multi-view data classification such as text and image. Background Art

[0002] With the development of information technology and sensor technology, different features of a sample can be obtained through different sensors, and multi-view data classification has become a hot issue in the field of machine learning. A large number of studies have shown that single-view features are not as complete as the information extracted by multiple feature combinations in some applications. By deeply exploring the potential information between multiple view features and promoting each other between view features, more comprehensive information of samples can be obtained. The main task of multi-view data classification is to compare the extracted different view features of the same target sample with the targets in the database to determine the target to be identified. At present, the method of multi-view data classification for targets is based on feature extraction.

[0003] The multi-view data classification method based on feature extraction requires feature extraction and selection of the target. Feature dimensionality reduction is a manifestation of feature selection. In the case of multi-view features, it is necessary to combine the features in pairs. The classic analysis method for the two-by-two combination features is Canonical Correlation Analysis (CCA). This method was first proposed by Hotelling. So far, scholars have proposed many variants of CCA to adapt to different practical applications.

[0004] The problems and defects of existing multi-view data classification technology are:

[0005] 1. Most methods are based on matrix canonical correlation analysis, which cannot obtain the correlation of the overall data, and therefore cannot obtain the intrinsic structural information of the data, resulting in insufficient information extraction and low accuracy of image and text classification.

[0006] 2. Although the existing tensor canonical correlation analysis can obtain the correlation of the overall data, its calculation speed is slow. The existing tensor canonical correlation analysis mainly uses tensor decomposition transformation, such as CP decomposition or Tucker decomposition. The process of calculating tensors is complicated, with many and large dimensions, resulting in a long time.

[0007] 3. Real data sets not only have linear correlations, but may also have nonlinear correlations. Therefore, it is necessary to add deep networks to existing methods to preprocess the data and expand it into a nonlinear space. Summary of the invention

[0008] In view of some or all of the problems in the prior art, the present invention provides a multi-view data classification method based on a fast canonical correlation analysis algorithm, the method comprising the following steps:

[0009] Input multi-view data into the data input module, perform pre-processing, and store it in the database;

[0010] Extracting data from a database, and calculating a similarity matrix and a covariance tensor in a data calculation module; and

[0011] The similarity matrix and the covariance tensor are input into a result output module, in which the matrix after feature extraction is calculated and the matrix is ​​classified.

[0012] Furthermore, the preprocessing is to preprocess the multi-view data using an autoencoder to expand it to a nonlinear space.

[0013] Furthermore, the covariance tensor is constructed to obtain the overall correlation of multi-view data.

[0014] Construct a multi-view dataset X p The k-nearest neighbor graph G p =(V, E; W),

[0015] The weight matrix W p ∈R N×N ,R N×N is an N×N matrix in real space, the vertex set matrix V p is the pth perspective x i The node v i , E is the set of edge nodes, e ij is node v i With node v j The connecting edge.

[0016] The covariance tensor is calculated using the following formula:

[0017]

[0018] Among them, N represents the number of samples in the multi-view dataset, Represents the nth column data of the mth perspective, Represents outer product computation.

[0019] Furthermore, the Laplace regularization term is calculated for the multi-view data to obtain a similarity matrix.

[0020] Definition H p is a canonical correlation variable, Z p represents the variable after CCA processing, T represents the transposition operation, and the Laplace matrix is ​​L = S p -Wp , S pii =∑ j W ij , is the correlation coefficient tensor,

[0021]

[0022] Among them, m represents the total number of categories of training samples, c 12...m represents the covariance tensor, corr represents the calculation of the correlation coefficient, ε represents a tensor of all 1s, × m Represents the tensor mode-m product, which is a symbol for calculating the product of tensors. Represents the real number space of dimension r×r×…r.

[0023] The Laplace regularization term is,

[0024]

[0025] in, represents the k-th step p-th view data after CCA processing, T represents the transposition operation, L p Represents the Laplacian matrix of the p-th view.

[0026] The similarity matrix S is calculated using the following formula:

[0027]

[0028] Here, c represents the number of classes.

[0029] Furthermore, the matrix after feature extraction is obtained by fast calculation using the alternating semi-smooth Newton algorithm.

[0030] Furthermore, the matrix after feature extraction is obtained by solving the following method.

[0031] First write the problem's proximity operator iteration steps and calculate the optimal solution to the problem:

[0032]

[0033] in, It means to find the minimum value. Represents the F-norm of the correlation coefficient tensor represents the Laplace regularization term, Indicates H p The k-th step l 2 , 1 norm, represents the inner product of two terms, Represents the square of the F-norm of the subtraction of two matrices.

[0034] Among them, the definition function

[0035] Order the descending direction Find the specific solution of the model:

[0036]

[0037] where Λ p represents the Lagrangian operator, represents B(Λ p )’s l_(2,1) norm proximal operator, Y p represents the augmented variable of the p-th perspective, Indicates H p The kth step of .

[0038]

[0039] According to the constraints in the model,

[0040]

[0041] Let it be the Q operator. According to the semi-smooth Newton algorithm, we only need to calculate the Jacobian matrix of the Q operator, which is:

[0042]

[0043] where the matrix Ξ p for:

[0044]

[0045] Finally, write the specific iteration formula:

[0046]

[0047] in express The vector of express The vector of k Indicates the step length.

[0048] Compared with first-order algorithms such as ADMM, Newton's method can correct the step size so that the objective function can be fully reduced, and then find the minimum value of the objective function with as few iterations as possible.

[0049] On the manifold Do the retraction operation and finally get the solution

[0050]

[0051] Among them, the retraction operation is the commonly used QR decomposition.

[0052] Matrix variable Z after feature extraction p for,

[0053] X p represents a multi-view dataset, T represents a transposition operation, and H p is the final solution, and * represents a multiplication operation.

[0054] Furthermore, the matrix is ​​classified by the aforementioned classification method, and the obtained matrix after feature extraction is input into a KNN classifier to obtain a classification result.

[0055] The present invention also provides a multi-view data classification system based on a canonical correlation analysis fast algorithm, the system comprising the following modules:

[0056] A data input module, used to input multi-view data and pre-process the multi-view data using an autoencoder;

[0057] A data calculation module, used to calculate the similarity matrix and covariance tensor of the preprocessed data; and

[0058] The result output module is used to calculate the matrix after feature extraction and classify the matrix.

[0059] The present invention also provides a computer system, comprising:

[0060] a processor configured to execute machine-readable instructions; and

[0061] The memory stores machine-readable instructions, which, when executed by the processor, perform the steps of the classification method.

[0062] The present invention also provides a computer-readable storage medium on which machine-readable instructions are stored. When the machine-readable instructions are executed by a processor, the steps of the classification method are executed.

[0063] Compared with the prior art, the present invention has the following advantages:

[0064] 1. Use autoencoders to expand multi-view data to nonlinear space, explore nonlinear relationships in the data, and better extract nonlinear features of multi-view data, which is conducive to improving the accuracy of subsequent multi-view data classification.

[0065] 2. Use tensor canonical correlation analysis to fully explore the intrinsic structural characteristics of multi-view data and obtain the overall correlation of multi-view data, so that the features of multi-view data can be better extracted.

[0066] 3. Use Laplace regularization to explore the inherent information hidden in multi-view data, make full use of the data structure, and extract its features completely and concisely.

[0067] 4. Design a fast algorithm for Laplace canonical correlation analysis to significantly reduce model training time.

[0068] 5. The proposed Laplace regularized tensor canonical correlation analysis fast algorithm greatly reduces the time for multi-view data feature extraction and greatly improves the accuracy of multi-view data classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] To further illustrate the above and other advantages and features of various embodiments of the present invention, a more specific description of various embodiments of the present invention will be presented with reference to the accompanying drawings. It will be understood that these drawings only depict typical embodiments of the present invention and are therefore not to be considered as limiting the scope thereof. In the accompanying drawings, for clarity, identical or corresponding parts will be represented by identical or similar reference numerals.

[0070] Figure 1 A schematic diagram of a multi-view data classification method flow chart according to an embodiment of the present invention;

[0071] Figure 2 A schematic diagram showing how the processing time of a method according to an embodiment of the present invention and a tensor canonical correlation analysis method varies with feature dimensions;

[0072] Figure 3 A schematic diagram showing how the accuracy of a method according to an embodiment of the present invention and a tensor canonical correlation analysis method varies with feature dimensions;

[0073] Figure 4 Schematic diagram of a multi-view data classification system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0074] In the following description, the present invention is described with reference to various embodiments. However, those skilled in the art will recognize that various embodiments may be implemented without one or more specific details or with other replacement and / or additional methods or components. In other cases, well-known structures or operations are not shown or described in detail to avoid obscuring the inventive point of the present invention. Similarly, for the purpose of explanation, specific numbers and configurations are set forth to provide a comprehensive understanding of embodiments of the present invention. However, the present invention is not limited to these specific details.

[0075] In this specification, reference to "one embodiment" or "the embodiment" means that a particular feature, structure, or characteristic described in conjunction with the embodiment is included in at least one embodiment of the present invention. The phrase "in one embodiment" appearing in various places in this specification does not necessarily all refer to the same embodiment.

[0076] It should be noted that the embodiments of the present invention describe the method steps in a specific order, but this is only for the purpose of illustrating the specific embodiment, rather than limiting the order of the steps. On the contrary, in different embodiments of the present invention, the order of the steps can be adjusted according to actual needs.

[0077] In the present invention, each module of the system according to the present invention can be implemented using software, hardware, firmware or a combination thereof. When the module is implemented using software, the function of the module can be implemented by a computer program flow, for example, the module can be implemented by a code segment (such as a code segment of a language such as C, C++) stored in a storage device (such as a hard disk, a memory, etc.), wherein the corresponding function of the module can be implemented when the code segment is executed by a processor. When the module is implemented using hardware, the function of the module can be implemented by setting a corresponding hardware structure, for example, the function of the module can be implemented by hardware programming a programmable device such as a field programmable gate array (FPGA), or the function of the module can be implemented by designing an application-specific integrated circuit (ASIC) including electronic devices such as a plurality of transistors, resistors and capacitors. When the module is implemented using firmware, the function of the module can be written into a read-only memory such as an EPROM or EEPROM of the device in the form of a program code, and the corresponding function of the module can be implemented when the program code is executed by a processor. In addition, certain functions of the module may need to be implemented by separate hardware or by collaboration with the hardware, for example, the detection function is implemented by corresponding sensors (such as proximity sensors, acceleration sensors, gyroscopes, etc.), the signal transmission function is implemented by corresponding communication devices (such as Bluetooth devices, infrared communication devices, baseband communication devices, Wi-Fi communication devices, etc.), the output function is implemented by corresponding output devices (such as displays, speakers, etc.), and so on.

[0078] The technical solutions in the embodiments of the present invention are described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. Figure 1 The multi-view data classification method according to an embodiment of the present invention is shown in the figure. As shown in the figure, the multi-view data classification method includes the following steps:

[0079] First, multi-view data is input into the data input module, pre-processed, and stored in a database. The pre-processing is to pre-process the multi-view data using an autoencoder and expand it into a nonlinear space. The pre-processing extracts data features from the selected multi-view data to generate a data matrix, and normalizes the generated data matrix to obtain a normalized data matrix.

[0080] Next, extract the data from the database and calculate the similarity matrix and covariance tensor in the data calculation module. Construct the covariance tensor to obtain the overall correlation of multi-view data.

[0081] Construct a multi-view dataset X p The k-nearest neighbor graph G p =(V, E; W),

[0082] The weight matrix W p ∈R N×N ,R N×N is an N×N matrix in real space, the vertex set matrix V p is the pth perspective x i The node v i , E is the set of edge nodes, e ij is node v i With node v j The connecting edge.

[0083] The covariance tensor is calculated using the following formula:

[0084]

[0085] Among them, N represents the number of samples in the multi-view dataset, Represents the nth column data of the mth perspective, Represents outer product computation.

[0086] The Laplace regularization term is calculated for multi-view data to obtain the similarity matrix.

[0087] Definition H p is a canonical correlation variable, Z p represents the variable after CCA processing, T represents the transposition operation, and the Laplace matrix is ​​L = S p -W p , S pii =∑ j W ij , is the correlation coefficient tensor,

[0088]

[0089] Among them, m represents the total number of categories of training samples, c12...m represents the covariance tensor, corr represents the calculation of the correlation coefficient, ε represents a tensor of all 1s, × m Represents the tensor mode-m product, which is a symbol for calculating the product of tensors. Represents the real number space of dimension r×r×…r.

[0090] The Laplace regularization term is,

[0091]

[0092] in, represents the k-th step p-th view data after CCA processing, T represents the transposition operation, L p Represents the Laplacian matrix of the p-th view.

[0093] The similarity matrix S is calculated using the following formula:

[0094]

[0095] Here, c represents the number of classes.

[0096] Next, the similarity matrix and covariance tensor calculated in the data calculation module are input into the result output module, and the matrix after feature extraction is calculated in the result output module, and the matrix is ​​classified.

[0097] The matrix after feature extraction is obtained by fast calculation through the alternating semi-smooth Newton algorithm.

[0098] The matrix after feature extraction is obtained by the following method.

[0099] First write the problem's proximity operator iteration steps and calculate the optimal solution to the problem:

[0100]

[0101] in, It means to find the minimum value. Represents the F-norm of the correlation coefficient tensor represents the Laplace regularization term, Indicates H p The k-th step l 2 , 1 norm, represents the inner product of two terms, Represents the square of the F-norm of the subtraction of two matrices.

[0102] Among them, the definition function

[0103] Order the descending direction Find the specific solution of the model:

[0104]

[0105] where Λ p represents the Lagrangian operator, represents B(Λ p )’s l_(2,1) norm proximal operator, Y p represents the augmented variable of the p-th perspective, Indicates H p The kth step of .

[0106]

[0107] According to the constraints in the model,

[0108]

[0109] Let it be the Q operator. According to the semi-smooth Newton algorithm, we only need to calculate the Jacobian matrix of the Q operator, which is:

[0110]

[0111] where the matrix Ξ p for:

[0112]

[0113] Finally, write the specific iteration formula:

[0114]

[0115] in express The vector of express The vector of k Indicates the step length.

[0116] Compared with first-order algorithms such as ADMM, Newton's method can correct the step size so that the objective function can be fully reduced, and then find the minimum value of the objective function with as few iterations as possible.

[0117] On the manifold Do the retraction operation and finally get the solution

[0118]

[0119] Among them, the retraction operation is the commonly used QR decomposition.

[0120] Matrix variable Z after feature extraction p for,

[0121] X p represents a multi-view dataset, T represents a transposition operation, and H p is the final solution, and * represents a multiplication operation.

[0122] The matrix is ​​classified by the aforementioned classification method, and the obtained matrix after feature extraction is input into a KNN (k-nearest neighbor) classifier to obtain a classification result.

[0123] The effect of the present invention can be further illustrated by the following simulation experiment.

[0124] The following example illustrates the effect of the multi-view data classification method based on the canonical correlation analysis fast algorithm provided by the present invention on the processing time and accuracy of different feature dimensions. The simulation experiment test object is the BBC news multi-view data set. The simulation experiment uses the present invention and the latest tensor canonical correlation analysis method TCCA-OS to respectively calculate the classification time and classification accuracy of different feature dimensions for the multi-view data set of BBC news. The present invention uses the non-linear Laplace canonical tensor canonical correlation analysis (NLLTCCA) method for simulation.

[0125] Figure 2 The schematic diagram of the classification time obtained by the method of one embodiment of the present invention and the tensor canonical correlation analysis method as a function of feature dimension. The inventor repeated a large number of experiments on the BBC news multi-view dataset. Compared with the latest tensor canonical correlation analysis method TCCA-OS, the NLLTCCA method of the present invention has a greater advantage in classification time. Figure 2 As shown, the classification time of the present invention changes less with the change of feature dimension, while the classification time of the TCCA-OS method increases more with the increase of feature dimension. The specific data are listed in Table 1. As shown in Table 1, with the same feature dimension, the classification time of the present invention is smaller than that of TCCA-OS, and the difference in classification time is between 0.17s and 24.25s, and the difference is larger between feature dimensions of 16-20; when the feature dimension increases from 2 to 14, the classification time of TCCA-OS increases from 1.13s to 3.56s, and when the feature dimension increases from 16 to 20, the classification time of TCCA-OS increases from 11.79s to 25.46s; while for the present invention, when the feature dimension increases from 2 to 20, the classification time of the present invention increases from 0.96s to 1.21s, and the change range is smaller.

[0126] Table 1 Classification time / s on the BBC news multi-view dataset

[0127] Feature Dimension TCCA-OS The present invention 2 1.13 0.96 4 1.27 0.99 6 1.33 1.02 8 1.41 1.04 10 1.61 1.05 12 2.26 1.07 14 3.56 1.08 16 11.79 1.10 18 12.81 1.16 20 25.46 1.21

[0128] Figure 3 The figure is a schematic diagram showing the classification accuracy of the method of an embodiment of the present invention and the tensor canonical correlation analysis method as the feature dimension changes. On the BBC news multi-view dataset, compared with the latest tensor canonical correlation analysis method TCCA-OS, the NLLTCCA method of the present invention also has certain advantages in classification accuracy. Figure 3 As shown, the classification accuracy curve of the present invention is higher than that of the TCCA-OS method as a whole. The specific data are listed in Table 2. As shown in Table 2, with the same feature dimensions, the classification accuracy of the present invention is higher than that of TCCA-OS, with a difference of 0.29% to 3.95%.

[0129] Table 2 Classification accuracy on the BBC news multi-view dataset

[0130] Feature Dimension TCCA-OS The present invention 2 66.63％ 69.71％ 4 80.49％ 80.78％ 6 84.68％ 85.46％ 8 86.29％ 89.12％ 10 85.41％ 89.36％ 12 87.56％ 90.19％ 14 87.16％ 90.24％ 16 86.78％ 89.90％ 18 86.24％ 89.65％ 20 87.36％ 89.12％

[0131] The present invention proposes a method of NFLTCCA to solve the problem of low classification accuracy of multi-view data. The autoencoder is used to preprocess the data and expand it to a nonlinear space. The Laplace regularized tensor canonical correlation analysis (LTCCA) is used to fuse the multi-view data in the nonlinear space to improve the classification accuracy of the multi-view data. A fast algorithm is proposed based on the model improvement to solve the problem that the current tensor canonical correlation analysis takes too long to process multi-view data, reduce the training time and training cost of the model, and enhance the portability of the model.

[0132] The present invention also provides a multi-view data classification system based on a typical correlation analysis fast algorithm. Figure 4 FIG. 1 is a schematic diagram of a multi-view data classification system according to an embodiment of the present invention. Figure 4 As shown, the system includes the following modules:

[0133] A data input module, used to input multi-view data and pre-process the multi-view data using an autoencoder;

[0134] A data calculation module, used to calculate the similarity matrix and covariance tensor of the preprocessed data; and

[0135] The result output module is used to calculate the matrix after feature extraction and classify the matrix.

[0136] In one embodiment, a computer system is also provided, including a memory and a processor, the memory storing machine-readable instructions, and the processor implementing the following processing steps when executing the machine-readable instructions: inputting multi-view data into a data input module, performing preprocessing, and storing the data in a database; extracting data from the database, and calculating a similarity matrix and a covariance tensor in a data calculation module; inputting the similarity matrix and the covariance tensor into a result output module, calculating a matrix after feature extraction in the result output module, and classifying the matrix.

[0137] It can be understood that in addition to the memory and processor mentioned above, the above-mentioned computer system also includes other software and hardware components not listed in this specification. The specific components can be determined according to the model of specific data processing equipment in different application scenarios, and this specification will not list them one by one in detail.

[0138] In one embodiment, a computer-readable storage medium is also provided, on which machine-readable instructions are stored. When the machine-readable instructions are executed by a processor, the following processing steps are implemented: multi-view data is input into a data input module, pre-processed, and stored in a database; data in the database is extracted, and a similarity matrix and a covariance tensor are calculated in a data calculation module; the similarity matrix and the covariance tensor are input into a result output module, a matrix after feature extraction is calculated in the result output module, and the matrix is ​​classified.

[0139] Although various embodiments of the present invention are described above, it should be understood that they are presented as examples only and not as limitations. It is obvious to those skilled in the relevant art that various combinations, modifications and changes can be made thereto without departing from the spirit and scope of the present invention. Therefore, the breadth and scope of the present invention disclosed herein should not be limited by the exemplary embodiments disclosed above, but should only be defined according to the attached claims and their equivalents.

Claims

1. A multi-view data classification method based on a fast canonical correlation analysis algorithm for image classification, characterized in that: The steps include: Inputting a multi-view image dataset into a data input module, preprocessing the multi-view image dataset using an autoencoder, expanding the multi-view image dataset into a nonlinear space, and storing the preprocessed dataset in a database; Extracting data from a database, and calculating a covariance tensor and a similarity matrix in a data calculation module to extract nonlinear features of the multi-view image data; as well as Inputting the covariance tensor and the similarity matrix into a result output module, calculating the matrix after feature extraction in the result output module, and classifying the matrix; Construct the covariance tensor to obtain the overall correlation of multi-view image data; Construct a multi-view image dataset X p The k-nearest neighbor graph G p =(V, E; W), The weight matrix W p ∈R N×N ,R N×N is an N×N matrix in real space, the vertex set matrix V p is the pth perspective x i The node v i , E is the set of edge nodes, e ij is node v i With node v j The connecting edges of The covariance tensor is calculated using the following formula: Where N represents the number of samples in the multi-view image dataset. Represents the nth column data of the mth perspective, represents the outer product calculation; Calculate the Laplace regularization term for multi-view image data to obtain a similarity matrix; Definition H p is a canonical correlation variable, Z p represents the variable after CCA processing, T represents the transposition operation, and the Laplace matrix is ​​L = S p -W p , where S p is the similarity matrix, W p is the weight matrix, S p For S pii =∑ j W ij , is the correlation coefficient tensor, Among them, m represents the total number of categories of training samples, represents the covariance tensor, cor r represents the calculation of the correlation coefficient, ε represents a tensor of all 1s, × m Represents the tensor mode-m product, which is a symbol for calculating the product of tensors. represents the real number space of dimension r×r×…r; The Laplace regularization term is, in, represents the k-th step p-th view data after CCA processing, T represents the transposition operation, L p represents the Laplacian matrix of the p-th view; The matrix after feature extraction is obtained by fast calculation through alternating semi-smooth Newton algorithm; The matrix after feature extraction is obtained by the following method: First write the problem's proximity operator iteration steps and calculate the optimal solution to the problem: in, It means to find the minimum value. Represents the square of the correlation coefficient tensor F norm × represents the Laplace regularization term, Indicates H p The k-th step l 2，1 norm, represents the inner product of two terms, Represents the square of the F-norm of the subtraction of two matrices; Among them, the definition function Order the descending direction Find the specific solution of the model: where Λ p represents the Lagrangian operator, represents B(Λ p )’s l_(2,1) norm proximal operator, Y p represents the augmented variable of the p-th perspective, Indicates H p The kth step of According to the constraints in the model, Let it be the Q operator. According to the semi-smooth Newton algorithm, we only need to calculate the Jacobian matrix of the Q operator, which is: where the matrix Ξ p for: Finally, write the specific iteration formula: in express The vector of express The vector of k represents the step length; Compared with first-order algorithms such as ADMM, Newton's method can correct the step size so that the objective function is fully reduced, and then find the minimum value of the objective function with as few iterations as possible; On the manifold Do the retraction operation and finally get the solution Among them, the retraction operation is the commonly used QR decomposition; Matrix variable Z after feature extraction p for, X p represents a multi-view image dataset, T represents a transposition operation, and H p is the final solution, and * represents a multiplication operation.

2. The multi-view data classification method based on canonical correlation analysis fast algorithm for image classification according to claim 1, characterized in that: The matrix is ​​classified by processing the matrix by the method of claim 1, and inputting the matrix obtained after feature extraction into a KNN classifier to obtain a classification result.

3. A multi-view data classification system for image classification based on a fast canonical correlation analysis algorithm, used in the method of any one of claims 1-2, characterized in that: Includes the following modules: A data input module is used to input multi-view image data, pre-process the multi-view data using an autoencoder, expand it into a nonlinear space, and store it in a database; A data calculation module, used for calculating a similarity matrix and a covariance tensor of the preprocessed data to extract nonlinear features of the multi-view image data; as well as The result output module is used to calculate the matrix after feature extraction and classify the matrix.

4. A computer system, characterized in that: include: a processor configured to execute machine-readable instructions; and A memory storing machine-readable instructions, wherein the machine-readable instructions, when executed by a processor, perform the steps of the method according to any one of claims 1-2.

5. A computer-readable storage medium, characterized in that: Machine-readable instructions are stored thereon, and when the machine-readable instructions are executed by a processor, the steps of the method according to any one of claims 1-2 are performed.

Images