Image classification method and system based on multilayer involute dictionary learning network

Abstract

The invention relates to an image classification method and system based on a multilayer involute dictionary learning network, and belongs to the technical field of image processing. Comprising the following steps: processing an obtained image data set to obtain a processed data set; performing feature extraction on the image samples of the processing data set to obtain a feature sample of each image sample; linearly representing each feature sample by using a sub-dictionary to obtain a dictionary matrix; constructing an optimized mathematical model according to the dictionary matrix, and solving the optimized mathematical model to obtain a representation coefficient matrix; and according to the feature sample, the dictionary matrix and the representation coefficient matrix, calculating a residual value of each type of channel sub-dictionary, and according to the residual value, predicting a label of the image sample so as to realize image classification. According to the method, the problem of information loss caused by a non-end-to-end architecture is solved, and the classification performance insufficiency caused by insufficient consideration of the specificity difference between categories is optimized.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an image classification method and system based on a multi-layer involution dictionary learning network. Background Art

[0002] Image classification technology is an image processing technology that distinguishes images of different categories based on the feature information reflected in the image. In the field of image classification, single-label classification problems are particularly common, that is, each image is assigned only one label, and the output of the model is a specific category label. Usually, the Softmax activation function is used to normalize the predicted probability distribution to determine the most likely one among multiple categories. The main application scenarios of this technology include cat and dog classification, handwritten digit recognition, and object recognition.

[0003] In the past few decades, a variety of single-label image classification algorithms have been proposed, among which the image classification algorithm based on dictionary learning is particularly classic. Dictionary learning technology was originally to extract a set of sparse signals that can linearly represent all original signals from raw data, and was later widely used in fields such as image denoising and image classification. However, traditional dictionary learning algorithms are limited by the size of the data set and the data dimension, and it is difficult to effectively process high-dimensional data such as images, voices, and videos. With the development of deep learning technology, it has been gradually integrated into dictionary learning. In the prior art, a variety of methods combining dictionary learning and deep learning have been proposed. For example, Deep Micro-Dictionary Learning and Coding Network (DDLCN) combines the feature extraction layer with multiple dictionary learning layers, and integrates the dictionary pair learning layer into the deep learning architecture, using the learned dictionary as a classifier, although it is limited to single-layer use. In addition, there is a multi-layer discriminative dictionary learning (MDDL) architecture, which uses a multi-layer dictionary for feature representation learning and combines a multi-class support vector machine (SVM) as a classifier. There is also a multi-layer dictionary learning network (Multi-layer Attention Dictionary Pair Learning Network, referred to as MAPDL-net), which is specifically used for image recognition tasks.

[0004] Although these methods effectively integrate dictionary learning and deep learning, they are not end-to-end architectures, which may lead to information loss during transmission. To overcome this challenge, the multi-layer attention dictionary pair learning network (MAPDL-net) proposed in the prior art is an end-to-end architecture that transforms the traditional dictionary learning method into a dictionary learning network, effectively improving the speed of processing large-scale data sets and maintaining the integrity of feature information. However, the dictionary layer in MAPDL-net is designed as a single shared layer, which does not fully consider the specificity between different categories, which may lead to poor results when dealing with classification tasks with similar category information. Summary of the invention

[0005] To this end, the technical problem to be solved by the present invention is to overcome the information loss caused by the non-end-to-end architecture in the prior art, and the poor classification effect caused by not fully considering the specificity between different categories.

[0006] In a first aspect, in order to solve the above technical problems, the present invention provides an image classification method based on a multi-layer involution dictionary learning network, comprising:

[0007] Processing the acquired image data set to obtain a processed data set;

[0008] Performing feature extraction on the image samples of the processed data set to obtain a feature sample of each of the image samples;

[0009] Linearly represent each of the feature samples using a sub-dictionary to obtain a dictionary matrix;

[0010] An optimized mathematical model is constructed according to the dictionary matrix, and a representation coefficient matrix is ​​obtained by solving the optimized mathematical model; wherein the optimized mathematical model is:

[0011]

[0012] Where D is the dictionary matrix, X i is the total representation coefficient matrix, y′ i is the feature sample, D k is the k-th channel sub-dictionary, is the k-th channel sub-dictionary pair y i The coefficient matrix, y i is the i-th image sample, concat(·) is the concatenation function, and M is the total number of categories;

[0013] The residual value of each type of channel sub-dictionary is calculated according to the feature sample, the dictionary matrix and the representation coefficient matrix, and the label of the image sample is predicted according to the residual value.

[0014] In one embodiment of the present invention, the calculation formula of the residual value is:

[0015]

[0016] Among them, e k is the residual of the kth class, E(c k-1 :c k ,:,:) is the residual matrix of the kth category, and M is the total number of categories.

[0017] In one embodiment of the present invention, when the minimum value of the residual value is e j =min{e1,e2,…,e M}, the label of the image sample is j; where M is the total number of categories.

[0018] In one embodiment of the present invention, the process of extracting features from the image samples of the processed data set includes performing channel addition processing on the image samples using a convolutional layer.

[0019] In one embodiment of the present invention, the convolution layer includes a first convolution layer, and the first convolution layer includes a 2D convolution function, a batch normalization function, a ReLU activation function and a 2D maximum pooling function.

[0020] In one embodiment of the present invention, the convolution layer further includes a plurality of second convolution layers, and the second convolution layer includes a plurality of 2D convolution functions and a batch normalization function.

[0021] In one embodiment of the present invention, the process of obtaining the dictionary matrix includes grouping the feature samples, setting a shared convolution kernel for each group, and convolving the channel features within the group using the shared convolution kernel; wherein the number of the groups is the number of categories of the classification task.

[0022] In a second aspect, in order to solve the above technical problems, the present invention provides an image classification system based on a multi-layer involution dictionary learning network, comprising:

[0023] The data feature extraction module is used to process the acquired image data set to obtain a processed data set; perform feature extraction on the image samples of the processed data set to obtain a feature sample of each image sample;

[0024] The involution dictionary learning module is used to linearly represent each of the feature samples with a sub-dictionary to obtain a dictionary matrix; construct an optimized mathematical model according to the dictionary matrix, and obtain a representation coefficient matrix by solving the optimized mathematical model; wherein the optimized mathematical model is:

[0025]

[0026] Where D is the dictionary matrix, X i is the total representation coefficient matrix, y′ i is the feature sample, D k is the k-th channel sub-dictionary, is the k-th channel sub-dictionary pair y i The coefficient matrix, y i is the i-th image sample, concat(·) is the concatenation function, and M is the total number of categories;

[0027] A classification module is used to calculate the residual value of each type of channel sub-dictionary according to the feature sample, the dictionary matrix and the representation coefficient matrix, and predict the label of the image sample according to the residual value; wherein the minimum value of the residual value is the label of the image sample.

[0028] In one embodiment of the present invention, the involution dictionary learning module includes multiple involution layers and dictionary layers, the involution layer is used to extract the category information of the image sample and pass it to the dictionary layer, and the dictionary layer is used to integrate different category information and pass it to the next involution layer.

[0029] In a third aspect, in order to solve the above technical problem, the present invention provides a computer program product, which includes a computer program. When the computer program is run, any of the above methods is executed.

[0030] The above technical solution of the present invention has the following beneficial effects compared with the prior art:

[0031] The image classification method and system based on a multi-layer involution dictionary learning network described in the present invention accurately captures the core features of the image through deep feature extraction technology, laying a solid information foundation for the image classification task. By adopting the sub-dictionary linear representation strategy, a dictionary matrix is ​​constructed for each feature sample, which not only improves the sparsity of the representation, but also enhances the model's ability to distinguish features of different categories. Furthermore, by constructing and solving the optimization mathematical model, the representation coefficient matrix can be accurately obtained to better adapt to the characteristics of the feature samples, thereby significantly improving the accuracy and robustness of the model. In addition, the label of the image sample is predicted based on the residual value, which realizes automatic image classification, reduces manual intervention, and improves processing efficiency. The design can adapt to a variety of image data sets and classification tasks, showing good versatility and applicability, and meeting the needs of image processing in different fields. The present invention effectively solves the problem of information loss that may be caused by non-end-to-end architecture, as well as the problem of affecting the classification effect due to insufficient consideration of the specific differences between different categories, and provides an innovative solution for the field of image classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below according to specific embodiments of the present invention in conjunction with the accompanying drawings, wherein

[0033] Figure 1 This is a flow chart of an image classification method based on a multi-layer involution dictionary learning network in a preferred embodiment of the present invention;

[0034] Figure 2 A network model diagram in a preferred embodiment of the present invention;

[0035] Figure 3 This is a framework diagram of an image classification system based on a multi-layer involution dictionary learning network in a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0036] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.

[0037] Embodiment 1

[0038] Reference Figure 1 As shown, an embodiment of the present invention provides an image classification method based on a multi-layer involution dictionary learning network, comprising:

[0039] Processing the acquired image data set to obtain a processed data set;

[0040] Perform feature extraction on the image samples of the processed data set to obtain feature samples of each image sample;

[0041] Linearly represent each feature sample with a sub-dictionary to obtain a dictionary matrix;

[0042] An optimization mathematical model is constructed according to the dictionary matrix, and a representation coefficient matrix is ​​obtained by solving the optimization mathematical model; wherein the optimization mathematical model is:

[0043]

[0044] Among them, D is the dictionary matrix, X i is the total representation coefficient matrix, y′ i is the feature sample, D k is the k-th channel sub-dictionary, is the k-th channel sub-dictionary pair y i The coefficient matrix, y i is the i-th image sample, concat(·) is the concatenation function, and M is the total number of categories;

[0045] The residual value of each channel sub-dictionary is calculated according to the feature samples, dictionary matrix and representation coefficient matrix, the label of the image sample is predicted according to the residual value, and the image is classified according to the label.

[0046] The embodiment of the present invention provides an image classification method based on a multi-layer involution dictionary learning network. By extracting features from image samples, the key features of the image can be effectively captured, thereby providing a rich information basis for subsequent classification tasks. By using a sub-dictionary linear representation method, a dictionary matrix can be constructed for each feature sample. This method not only improves the sparsity of the representation, but also enhances the model's ability to distinguish features of different categories. Furthermore, by constructing and solving an optimized mathematical model, the representation coefficient matrix can be accurately obtained to better adapt to the characteristics of the feature sample. This optimization process helps to improve the accuracy and robustness of the model. In addition, the label of the image sample is predicted based on the residual value to achieve automatic image classification, reduce manual intervention, and improve processing efficiency. This method can adapt to a variety of image data sets and classification tasks, has good versatility and applicability, and meets the image processing needs of different fields. The embodiment of the present invention effectively solves the problem of information loss that may be caused by a non-end-to-end architecture, as well as the problem of affecting the classification effect due to insufficient consideration of specific differences between different categories, and provides an innovative solution for the field of image classification.

[0047] Specifically, the embodiment of the present invention assumes that the image data set that has been acquired is Its label set is The i-th image sample is y i ∈R C×H×W , C is the number of channels of the image sample, H is the height of the image sample, W is the width of the image sample, L i ∈{1,…,M} is y i The label of , M is the total number of categories, and n is the total number of samples in the image dataset. After standardizing and normalizing the image dataset, assuming that the image sample y i After data feature extraction, it is mapped into feature sample y′ i , where y′ i ∈R C ' ×H×W , C′ is the number of channels C before the image sample is input and mapped to the new number of channels after feature extraction.

[0048] Specifically, the process of extracting features from image samples of the processed data set includes performing channel addition processing on the image samples using a convolutional layer. Figure 2, the convolution layer includes a first convolution layer and multiple second convolution layers, wherein the first convolution layer includes a 2D convolution function, a batch normalization function, a ReLU activation function and a 2D maximum pooling function; the second convolution layer includes multiple 2D convolution functions and a batch normalization function. The convolution layer in this embodiment preferably includes a first convolution layer and four second convolution layers. The main purpose of the first convolution layer is to perform channel increment processing on the input data, and to facilitate subsequent feature extraction while maintaining feature information. Although the four second convolution layers are stacked by multiple 2D convolution functions and batch normalization functions, there are differences in the number of convolution function channels. In this embodiment, their number of channels is gradually expanded, preferably 32, 64, 256, 512, and the step size is preferably 1, 2, 1, 2. The selection of the number of channels and step size of the convolution function should be flexibly adjusted based on specific application requirements and data characteristics to achieve optimal performance and effects.

[0049] The first convolutional layer integrates the 2D convolution function, batch normalization function, ReLU activation function and 2D maximum pooling function. Its advantage is that it can effectively extract the features of the input data while increasing the nonlinear expression ability of the network. The 2D convolution function is used to capture local features, the batch normalization function helps to reduce overfitting and speed up the training process, the ReLU activation function introduces nonlinearity, allowing the network to learn more complex features, and the 2D maximum pooling function is used to reduce the spatial dimension of the features and maintain feature invariance. This comprehensive design not only enhances the expressiveness of the features, but also improves the model's robustness to input changes, and lays the foundation for in-depth feature extraction in subsequent convolutional layers.

[0050] Specifically, the process of linearly representing each feature sample with a sub-dictionary includes grouping the data features of the previous layer, setting a shared convolution kernel for each group to convolve the channel features within the group; the number of groups is the number of categories of the classification task. Different convolutions are used for channel convolution for different groups. This method of intra-channel convolution has a built-in channel attention mechanism, which can efficiently allocate information of different categories to dictionary channels of different groups, thereby enhancing the model's ability to distinguish features.

[0051] Specifically, for the feature sample y′ i In this embodiment, it is assumed that the characteristic sample y′ i It can be linearly represented by all sub-dictionaries, and the mathematical expression is:

[0052]

[0053] Among them, the concat(·) function represents concatenation of matrices in the channel dimension. represents the multiplication of corresponding elements of the matrix, is the k-th channel sub-dictionary pair y iThe coefficient matrix of is the k-th channel sub-dictionary, C k is the number of atoms in the k-th channel sub-dictionary, and the dimension of each dictionary atom is H×W, k=1,…,M. All sub-dictionaries form a dictionary matrix D in the channel dimension, which is expressed as:

[0054]

[0055] According to the dictionary matrix, an optimization mathematical model is constructed (constructing an optimization mathematical model can also be called constructing an optimization problem), where the total representation coefficient matrix X in the optimization mathematical model is i The expression is:

[0056]

[0057] When the image data y is input, the feature sample y′ is obtained after data feature extraction, and the coefficient matrix is ​​represented After k∈{1,…,M}, the label of the input sample is predicted based on the representation coefficient matrix.

[0058] Specifically, let the residual matrix And E(c k-1 :c k ,:,:) represents the residual matrix of the kth class, where The calculation formula for the residual value of the kth class is:

[0059]

[0060] Among them, e k is the residual of the kth category, and M is the total number of categories.

[0061] Furthermore, when the minimum residual value is e j =min{e1,e2,…,e M}, the label of the image sample is j.

[0062] Furthermore, after obtaining the predicted label of the image, the image classification task can be implemented.

[0063] In order to more intuitively demonstrate the superiority of the embodiments of the present invention compared with the prior art, a set of comparative experiments is provided.

[0064] Specifically, the method provided by the embodiment of the present invention can either use image acquisition technology to autonomously acquire an image data set or directly adopt an existing image data set. This comparative experiment was tested on the well-known image classification data set Fashion-MNIST. Fashion-MNIST, as a fashion variant of the MNIST handwritten digit set, is provided by the research department of the German fashion technology company Zalando and contains 70,000 front-facing product images from 10 clothing categories. Each image is a single-channel grayscale image of 28×28 pixels, and its label numbers 0 to 9 represent T-shirts, pants, pullovers, dresses, coats, sandals, shirts, sneakers, bags, and boots, respectively. In the experiment, the Fashion-MNIST data set was divided into a training set containing 60,000 samples and a test set containing 10,000 samples. After 350 training cycles (epochs) of learning on the training set, the optimized model parameters can be obtained. Subsequently, using these trained model parameters (especially the model based on dictionary weights), prediction classification was performed on the test set, and its accuracy was evaluated. In addition, the parameter values ​​corresponding to the Fashion-MNIST dataset are: n = 70000, M = 10, B = 512, C = 1, H = 28, W = 28. Among them, B represents the batch size, which refers to the number of samples when input at one time.

[0065] The classification accuracy of the method provided by the embodiment of the present invention on various data sets was compared and analyzed with the performance of existing algorithms including discriminative K-SVD (DKSVD), dictionary learning based on Fisher discriminant criterion (FDDL), projected dictionary pair learning (DPL) and multi-layer dictionary learning network (MADPL-net) on the same data sets.

[0066] Table 1 Classification accuracy of different classification models for Fashion-MNIST dataset

[0067] method DKSV FDDL DPL MADPL-net The present invention Accuracy (%) 76.25 81.23 84.65 91.50 93.92

[0068] The specific comparison results are shown in Table 1. From the data in the table, it can be seen that the classification performance of the embodiment of the present invention is higher than that of other algorithms, which fully proves the effectiveness and superiority of the present invention in image classification tasks.

[0069] Embodiment 2

[0070] Based on the same inventive concept, this embodiment provides an image classification system based on a multi-layer involution dictionary learning network. The principle of solving the problem is similar to the image classification method based on a multi-layer involution dictionary learning network provided in Example 1, and the repeated parts will not be repeated.

[0071] Reference Figure 3As shown, this embodiment provides an image classification system based on a multi-layer involution dictionary learning network, including:

[0072] The data feature extraction module is used to process the acquired image data set to obtain a processed data set; perform feature extraction on the image samples of the processed data set to obtain a feature sample of each image sample;

[0073] The involution dictionary learning module is used to linearly represent each feature sample with a sub-dictionary to obtain a dictionary matrix; an optimization mathematical model is constructed according to the dictionary matrix, and a representation coefficient matrix is ​​obtained by solving the optimization mathematical model; wherein the optimization mathematical model is:

[0074]

[0075] Among them, D is the dictionary matrix, X i is the total representation coefficient matrix, y′ i is the feature sample, D k is the k-th channel sub-dictionary, is the k-th channel sub-dictionary pair y i The coefficient matrix, y i is the i-th image sample, concat(·) is the concatenation function, and M is the total number of categories;

[0076] The classification module is used to calculate the residual value of each channel sub-dictionary according to the feature samples, the dictionary matrix and the representation coefficient matrix, and predict the label of the image sample according to the residual value; wherein the minimum value of the residual value is the label of the image sample.

[0077] Specifically, the data feature extraction module consists of 5 convolutional layers, of which the first convolutional layer includes a 2D convolution function, a batch normalization function, a ReLU activation function and a 2D maximum pooling function. The main purpose of this convolutional layer is to perform channel increment processing on the input data, maintaining feature information while facilitating subsequent feature extraction. The following 4 convolutional layers are composed of multiple 2D convolutional functions and batch normalization functions stacked together. The difference is that the number of convolutional function channels of each convolutional layer is gradually expanded from 32, 64, 256, and 512, with step sizes of 1, 2, 1, and 2, respectively.

[0078] Specifically, the purpose of the involution dictionary learning module is to obtain the dictionary matrix D, and process the feature samples obtained by the data feature extraction module to obtain the representation coefficient matrix. This module includes multiple involution layers and dictionary layers, where the involution layer is used to extract the category information of the image sample and pass it to the dictionary layer; and the dictionary layer is used to integrate different category information and pass it to the next involution layer. Specifically, the function of the involution layer is to group the data features of the previous layer, and the number of groups matches the number of categories in the classification task. Each group uses a shared convolution kernel to perform convolution operations on the channel features within the group, and different groups use different convolution kernels for channel convolution. This channel-based involution strategy embeds a channel attention mechanism, which can effectively map different categories of information to dictionary channels of different groups.

[0079] Embodiment 3

[0080] This embodiment provides a computer program product, which includes a computer program. When the computer program is run, the method described in the first embodiment is executed.

[0081] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0082] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0083] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0084] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0085] Obviously, the above embodiments are merely examples for clear explanation and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived from these are still within the protection scope of the invention.

Claims

1. An image classification method based on a multi-layer involution dictionary learning network, characterized in that: include: Processing the acquired image data set to obtain a processed data set; Performing feature extraction on the image samples of the processed data set to obtain a feature sample of each of the image samples; Linearly represent each of the feature samples using a sub-dictionary to obtain a dictionary matrix; An optimized mathematical model is constructed according to the dictionary matrix, and a representation coefficient matrix is ​​obtained by solving the optimized mathematical model; wherein the optimized mathematical model is: Where D is the dictionary matrix, X i is the total representation coefficient matrix, y′ i is the feature sample, D k is the k-th channel sub-dictionary, is the k-th channel sub-dictionary pair y i The coefficient matrix, y i is the i-th image sample, concat(·) is the concatenation function, and M is the total number of categories; The residual value of each type of channel sub-dictionary is calculated according to the feature sample, the dictionary matrix and the representation coefficient matrix, and the label of the image sample is predicted according to the residual value.

2. The image classification method based on a multi-layer involution dictionary learning network according to claim 1 is characterized in that: The calculation formula of the residual value is: Among them, e k is the residual of the kth class, E(c k-1 :c k ,:,:) is the residual matrix of the kth category, and M is the total number of categories.

3. The image classification method based on a multi-layer involution dictionary learning network according to claim 1 is characterized in that: When the minimum value of the residual value is e j =min{e1,e2,…,e M }, the label of the image sample is j; where M is the total number of categories.

4. The image classification method based on a multi-layer involution dictionary learning network according to claim 1 is characterized in that: The process of extracting features from the image samples of the processed data set includes performing channel addition processing on the image samples using a convolutional layer.

5. The image classification method based on a multi-layer involution dictionary learning network according to claim 4 is characterized in that: The convolution layer includes a first convolution layer, and the first convolution layer includes a 2D convolution function, a batch normalization function, a ReLU activation function and a 2D maximum pooling function.

6. The image classification method based on a multi-layer involution dictionary learning network according to claim 4 is characterized in that , the convolution layer also includes multiple second convolution layers, and the second convolution layer includes multiple 2D convolution functions and batch normalization functions.

7. The image classification method based on a multi-layer involution dictionary learning network according to claim 1 is characterized in that: The process of obtaining the dictionary matrix includes grouping the feature samples, setting a shared convolution kernel for each group, and convolving the channel features within the group using the shared convolution kernel; wherein the number of the groups is the number of categories of the classification task.

8. An image classification system based on a multi-layer involution dictionary learning network, characterized in that: include: The data feature extraction module is used to process the acquired image data set to obtain a processed data set; perform feature extraction on the image samples of the processed data set to obtain a feature sample of each image sample; The involution dictionary learning module is used to linearly represent each of the feature samples with a sub-dictionary to obtain a dictionary matrix; construct an optimized mathematical model according to the dictionary matrix, and obtain a representation coefficient matrix by solving the optimized mathematical model; wherein the optimized mathematical model is: Where D is the dictionary matrix, X i is the total representation coefficient matrix, y′ i is the feature sample, D k is the k-th channel sub-dictionary, is the k-th channel sub-dictionary pair y i The coefficient matrix, y i is the i-th image sample, concat(·) is the concatenation function, and M is the total number of categories; A classification module is used to calculate the residual value of each type of channel sub-dictionary according to the feature sample, the dictionary matrix and the representation coefficient matrix, and predict the label of the image sample according to the residual value; wherein the minimum value of the residual value is the label of the image sample.

9. The image classification system based on a multi-layer involution dictionary learning network according to claim 8, characterized in that: The involution dictionary learning module includes multiple involution layers and dictionary layers. The involution layer is used to extract the category information of the image sample and pass it to the dictionary layer. The dictionary layer is used to integrate different category information and pass it to the next involution layer.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is run, the method as claimed in any one of claims 1 to 7 is executed.

Images