Non-negative representation classification method and system based on auto-encoder
The non-negative representation classification method using autoencoders addresses the challenge of mapping low-dimensional nonlinear data to high-dimensional space, improving image classification accuracy by preprocessing and iteratively solving optimization problems to enhance classification efficiency.
Patent Information
- Application Number
- CN202411250365.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-06
- Publication Date
- 2025-01-03
AI Technical Summary
Existing non-negative representation classification methods struggle to accurately classify linearly inseparable samples due to their inability to map data to a high-dimensional linear feature space, limiting image classification accuracy.
A non-negative representation classification method based on autoencoders that preprocesses training samples, constructs a non-negative representation classification model, and iteratively solves an optimization problem using the alternating direction method of multipliers to extract representation coefficients and mapping matrices, effectively mapping low-dimensional nonlinear data to high-dimensional space.
Improves image classification accuracy by ensuring samples have a linear structure in high-dimensional feature space, enhancing classification efficiency and reliability.
Smart Images
Figure CN119251551A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of pattern classification, and in particular to a non-negative representation classification method and system based on an autoencoder. Background Art
[0002] Sparse Representation-based Classification (SRC) believes that natural signals or data can be effectively represented by a linear combination of a small number of basis vectors. SRC uses a norm-regularized representation vector similar to the sparsity property of the norm, so that the essential characteristics of the data can be effectively extracted. Collaborative Representation-based Classification (CRC) uses a linear combination of all training samples to represent the test sample, and classifies it through this collaborative representation. Unlike sparse representation classification methods, CRC does not require sparsity, but emphasizes collaborative representation and uses norm-regularized representation vectors to improve classification efficiency.
[0003] The non-negative representation-based classification method (NRC) represents the test sample as a non-negative linear combination of the training samples, that is, the representation coefficient is required to be non-negative. In this way, the representation of the data is more consistent with the actual physical meaning. In the classification task, the non-negative representation method uses the training samples to build a dictionary and non-negatively encodes the test samples. The obtained non-negative representation coefficients are further used for classification decisions. By comparing the reconstruction errors of different categories, the category with the smallest error is selected as the category of the test sample. The non-negative representation method has good physical interpretability and computational efficiency, and is widely used in image classification, face recognition, document classification, music signal processing and other fields.
[0004] Although non-negative representation classification methods have shown excellent performance in many applications, their classification capabilities still have some limitations. One major problem is that sample data in reality often has complex nonlinear structures, and simple linear models are difficult to capture the intrinsic structure and pattern of the data. Traditional NRC usually only operates in the original space and lacks the ability to map data to a high-dimensional linear feature space for processing, which severely limits the accuracy and effect of image classification when faced with linearly inseparable samples. Summary of the invention
[0005] In order to overcome the problem in existing image pattern classification technology that the data is not mapped to a high-dimensional linear feature space for processing, which leads to serious limitations on the accuracy and effect of image classification when faced with linearly inseparable samples, the present invention proposes a non-negative representation classification method and system based on autoencoder, which can effectively map low-dimensional nonlinear data to high-dimensional space, ensure that the samples have a linear structure in the high-dimensional feature space, so as to obtain higher image classification accuracy.
[0006] To achieve the purpose of the present invention, the present invention adopts the following technical solutions:
[0007] A non-negative representation classification method based on an autoencoder, the method comprising the following steps:
[0008] Acquire test samples and training sample sets for image classification, and preprocess the training sample sets;
[0009] According to the preprocessed training sample set, the representation coefficient and mapping matrix of the test sample on the training sample set are extracted;
[0010] The test sample is classified based on the representation coefficient and the mapping matrix to obtain a category corresponding to the minimum residual error of the test sample.
[0011] In the above technical scheme, preprocessing the samples in the training sample set can improve the reliability of the sample data in the training sample set and improve the efficiency of subsequent data calculation; extracting the representation coefficient of the test sample in the training sample set, the obtained representation coefficient can simultaneously constrain the subspace competition representation item and the category representation vector, increase the contribution of each category representation vector, and can effectively map low-dimensional nonlinear data to high-dimensional space, ensuring that the sample has a linear structure in the high-dimensional feature space, and classifying the test sample based on the representation coefficient to obtain the category corresponding to the minimum residual identified by the test sample, effectively completing various image classification tasks and improving the accuracy of image classification.
[0012] Furthermore, a training sample set for image pattern classification is obtained, and the expression is:
[0013]
[0014] in, n i Represents the i-th type of training sample X i C represents the total number of sample categories; n represents the total number of training samples; α represents the feature vector of each sample; m represents the feature dimension of each sample.
[0015] Furthermore, the process of preprocessing the training sample set includes:
[0016] The samples in the training sample set are normalized using the l2 norm.
[0017] In the above technical solution, preprocessing the samples in the training sample set can improve the reliability of the sample data in the training sample set and, at the same time, improve the efficiency of subsequent data calculation.
[0018] Furthermore, the process of extracting the representation coefficient and mapping matrix of the test sample on the training sample set includes:
[0019] Constructing a non-negative representation classification model based on an autoencoder, and constructing an optimization problem by using the non-negative representation classification model based on the autoencoder with the goal of optimizing the non-negative representation classification;
[0020] Iteratively solving the optimization problem to obtain a non-negative representation classification model based on an autoencoder that grasps the representation coefficients and mapping matrices of the test samples on the training sample set;
[0021] The non-negative representation classification model based on the autoencoder extracts the representation coefficients and mapping matrix of the test samples on the training sample set;
[0022] Among them, the non-negative representation classification model based on autoencoder is a neural network model.
[0023] In the above technical scheme, the constructed non-negative representation classification model based on the autoencoder can effectively grasp the sample data changes of the test samples and the training sample sets after iterative optimization, so that in the process of extracting the representation coefficients and mapping matrices of the test samples on the training sample sets, the low-dimensional nonlinear data can be effectively mapped to the high-dimensional space, ensuring that the samples have a linear structure in the high-dimensional feature space, completing various image classification tasks and improving the accuracy of image classification.
[0024] Furthermore, the process of constructing the optimization problem includes:
[0025] The non-negative representation classification model based on the autoencoder constructs an optimization objective function of non-negative representation classification of a non-negative least squares problem according to the test sample and the training sample set. The expression is:
[0026]
[0027] Use the variable splitting method to solve the representation coefficient α, introduce the auxiliary variable z, and rewrite the non-negative representation classification objective function into the following form:
[0028]
[0029] Among them, T1 represents the encoding matrix, T2 represents the decoding matrix, φ represents the nonlinear function, λ represents the regularization parameter, y represents the test sample, and α represents the coefficient.
[0030] Furthermore, the process of iteratively solving the optimization problem includes:
[0031] The alternating direction multiplier method is used to solve the corresponding augmented Lagrangian function, which is expressed as:
[0032]
[0033] The alternating optimization method is used to iteratively update the variables T1, T2, α, z and δ. The process includes:
[0034] Fix T2, α, z and δ, update T1, the expression is:
[0035]
[0036] Fix T1, α, z and δ, and update T2, the expression is:
[0037] T2=Xφ T (T1X)[φ(T1X)φ T (T1X)] -1 ;
[0038] Fix T1, T2, z and δ, update α, the expression is:
[0039] α=[λφ T (T1X)φ(T1X)+μI] -1 [λφ T (T1X)φ(T1y)+μz+δ];
[0040] Fix T1, T2, α and δ, and update z. The expression is:
[0041]
[0042] Update δ, the expression is:
[0043] δ = δ + μ (z - α);
[0044] In the process of iteratively solving the optimization problem, when the non-negative representation classification model based on the autoencoder converges, the non-negative representation classification model based on the autoencoder that grasps the representation coefficient and mapping matrix of the test sample on the training sample set is obtained, and the representation coefficient and mapping matrix T1 and T2 of the test sample on the training sample set are extracted by using the non-negative representation classification model based on the autoencoder;
[0045] Among them, δ represents the Lagrange multiplier vector, μ>0 represents the penalty parameter, <δ, z-α> represents the inner product of two vectors, γ represents the learning rate, and T1 is continuously updated until the convergence condition is reached or the maximum number of cycles is reached, and ε is a smaller threshold.
[0046] Furthermore, the convergence condition of the non-negative representation classification model based on the autoencoder is expressed as:
[0047] ||α t -z t ||2≤η,||α t+1 -α t ||2≤η and ||z t+1 -z t ||2≤ηsatisfies at the same time;
[0048] Here, η represents a smaller threshold.
[0049] In the above technical scheme, the representation coefficient of the test sample on the training sample set is extracted, and the obtained representation coefficient can simultaneously constrain the subspace competition representation item and the category representation vector, increase the contribution of each category representation vector, and can effectively map low-dimensional nonlinear data to high-dimensional space.
[0050] Furthermore, the process of classifying the test samples based on the representation coefficients and the mapping matrix includes:
[0051] The non-negative representation classification model based on the autoencoder calculates the reconstructed residual of the test sample in the high-dimensional linear space according to the obtained representation coefficients and mapping matrices T1 and T2:
[0052] r i =||φ(T1y)-φ(T1X i )α i ||2;
[0053] Classification is performed based on the residuals, and the test sample is identified as the category corresponding to the minimum residual:
[0054]
[0055] Among them, y represents the test sample, X i represents the i-th type of training samples, α i represents the coefficient vector corresponding to the i-th class of training samples.
[0056] In the above technical solution, the test samples are classified based on the representation coefficients to obtain the category corresponding to the minimum residual error identified as the test sample, thereby effectively completing various image classification tasks and improving the accuracy of image classification.
[0057] A non-negative representation classification system based on an autoencoder, the system comprising:
[0058] An acquisition module, used to acquire test samples and training sample sets for image classification;
[0059] A processing module, used for preprocessing the training sample set;
[0060] An extraction module is used to extract the representation coefficient and mapping matrix of the test sample on the training sample set according to the preprocessed training sample set;
[0061] The classification module classifies the test sample based on the representation coefficient and the mapping matrix to obtain the category corresponding to the minimum residual error of the test sample.
[0062] An electronic device comprises a memory, a processor and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, the steps of a non-negative representation classification method based on an autoencoder are implemented.
[0063] Compared with the prior art, the present invention has the following beneficial effects:
[0064] The present invention proposes a non-negative representation classification method and system based on an autoencoder. First, the samples in the training sample set are preprocessed to improve the reliability of the sample data in the training sample set and, at the same time, improve the efficiency of subsequent data calculation. Then, the representation coefficient of the test sample on the training sample set is extracted. The obtained representation coefficient can simultaneously constrain the subspace competition representation item and the category representation vector, increase the contribution of each category representation vector, and effectively map low-dimensional nonlinear data to a high-dimensional space, ensuring that the sample has a linear structure in the high-dimensional feature space. Finally, the test sample is classified based on the representation coefficient to obtain the category corresponding to the minimum residual error identified as the test sample, effectively completing various image classification tasks and improving the accuracy of image classification. BRIEF DESCRIPTION OF THE DRAWINGS
[0065] Figure 1 A flowchart of a non-negative representation classification method based on an autoencoder provided in an embodiment of the present application;
[0066] Figure 2 A schematic diagram of the structure of a non-negative representation classification model based on an autoencoder provided in an embodiment of the present application;
[0067] Figure 3 Iterative convergence diagram of the non-negative representation classification model based on the autoencoder provided in the embodiment of the present application;
[0068] Figure 4 A structural block diagram of a non-negative representation classification system based on an autoencoder provided in an embodiment of the present application. DETAILED DESCRIPTION
[0069] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Preferred embodiments of the present invention are provided in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present invention more thorough and comprehensive.
[0070] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.
[0071] Embodiment 1:
[0072] This embodiment provides a non-negative representation classification method based on an autoencoder. Figure 1 , the method comprises the following steps:
[0073] S1: Obtain test samples and training sample sets for image classification, and preprocess the training sample sets;
[0074] S2: Extract the representation coefficient and mapping matrix of the test sample on the training sample set according to the preprocessed training sample set;
[0075] S3: Classify the test sample based on the representation coefficient and the mapping matrix to obtain the category corresponding to the minimum residual error of the test sample.
[0076] In step S1, a training sample set for image pattern classification is obtained, and the expression is:
[0077]
[0078] in, n i Represents the i-th type of training sample X i C represents the total number of sample categories; n represents the total number of training samples; α represents the feature vector of each sample; m represents the feature dimension of each sample.
[0079] Furthermore, the process of preprocessing the training sample set includes:
[0080] The samples in the training sample set are normalized using the l2 norm.
[0081] It can be understood that preprocessing the samples in the training sample set can improve the reliability of the sample data in the training sample set and improve the efficiency of subsequent data calculation.
[0082] As a preferred embodiment, in step S2, see Figure 2 , the process of extracting the representation coefficient and mapping matrix of the test sample on the training sample set includes:
[0083] S21: constructing a non-negative representation classification model based on an autoencoder, and constructing an optimization problem by using the non-negative representation classification model based on the autoencoder with the goal of optimizing the non-negative representation classification;
[0084] S22: Iteratively solving the optimization problem to obtain a non-negative representation classification model based on an autoencoder that grasps the representation coefficients and mapping matrices of the test samples on the training sample set;
[0085] S23 extracts the representation coefficients and mapping matrices of the test samples on the training sample set based on the non-negative representation classification model of the autoencoder;
[0086] Among them, the non-negative representation classification model based on autoencoder is a neural network model.
[0087] It can be understood that the constructed non-negative representation classification model based on the autoencoder can effectively grasp the sample data changes of the test samples and the training sample sets after iterative optimization, so that in the process of extracting the representation coefficients and mapping matrices of the test samples on the training sample sets, the low-dimensional nonlinear data can be effectively mapped to the high-dimensional space, ensuring that the samples have a linear structure in the high-dimensional feature space, completing various image classification tasks and improving the accuracy of image classification.
[0088] In step S21, the process of constructing the optimization problem includes:
[0089] The autoencoder-based nonnegative representation classification model (Autoencoder-based Nonnegative Representation for Pattern Classification, AENR) constructs an optimization objective function for nonnegative least squares problem based on the test sample and training sample set. The expression is:
[0090]
[0091] Furthermore, AENR is a non-negative least squares problem, which requires the variable splitting method to solve the representation coefficient α, introduce the auxiliary variable z, and rewrite the non-negative representation classification objective function into the following form:
[0092]
[0093] Among them, T1 represents the encoding matrix, T2 represents the decoding matrix, φ represents the nonlinear function, which can be RELU, TANH, SIGMOID, etc., λ represents the regularization parameter used to control the weight of the global representation term, y represents the test sample, and α is the representation coefficient.
[0094] As a preferred embodiment, in step S22, the process of iteratively solving the optimization problem includes:
[0095] The alternating direction method of multipliers (ADMM) is used to solve the corresponding augmented Lagrange function (ALF), which is expressed as:
[0096]
[0097] Among them, δ is the Lagrange multiplier vector, μ>0 is the penalty parameter, and <δ, z-α> is the inner product of two vectors. The alternating optimization method is used to update the variables T2, T1, α, z and δ. The detailed solution process is as follows:
[0098] (1) Fix T1, α, z and δ, and update T2. The expression is:
[0099]
[0100] Since this problem is a strictly convex function, the analytical solution of T2 is:
[0101] T2=Xφ T (T1X)[φ(T1X)φ T (T1X)] -1 ;
[0102] (2) Fix T2, α, z and δ, and update T1. The expression is:
[0103]
[0104] Due to the existence of nonlinear activation functions, there is no suitable method to directly obtain the analytical solution to this problem. This optimization problem can be solved by gradient descent to obtain the optimal value of T1. First, find the partial derivative of the objective function with respect to T1:
[0105]
[0106] Furthermore, the derivative is used to update T1:
[0107]
[0108] Where γ is the learning rate. T1 is continuously updated until the convergence condition is reached or the maximum number of cycles is reached. The convergence condition of the non-negative representation classification model based on the autoencoder is ||T2φ(T1X)-X||1≤ε, ||φ(T1y)-φ(T1X)α||2≤ε is satisfied at the same time, where ε is a smaller threshold. ;
[0109] (3) Fix T1, T2, z and δ, update α, the expression is:
[0110]
[0111] The above formula is a least squares problem, and its analytical solution is:
[0112] α=[λφ T (T1X)φ(T1X)+μI] -1 [λφ T (T1X)φ(T1y)+μz+δ];
[0113] (4) Fix T1, T2, α and δ, and update z. The expression is:
[0114]
[0115] The analytical solution of z can be obtained as:
[0116]
[0117] (5) Update δ, the expression is:
[0118] δ = δ + μ (z - α);
[0119] In the process of iteratively solving the optimization problem, when the non-negative representation classification model based on the autoencoder converges, the non-negative representation classification model based on the autoencoder that grasps the representation coefficient and mapping matrix of the test sample on the training sample set is obtained, and the representation coefficient and mapping matrix T1 and T2 of the test sample on the training sample set are extracted by using the non-negative representation classification model based on the autoencoder;
[0120] Among them, δ represents the Lagrange multiplier vector, μ>0 represents the penalty parameter, <δ, z-α> represents the inner product of two vectors, γ represents the learning rate, and T1 is continuously updated until the convergence condition is reached or the maximum number of cycles is reached, and ε is a smaller threshold.
[0121] For example, Figure 3 As shown, in order to further illustrate the convergence of AENR, Figure 3The convergence process of the algorithm on the USPS dataset is described. In the first 10 iterations, the value of the ADMM convergence condition drops rapidly. After 20 iterations, ||α t+1 -α t ||2 and ||z t+1 -z t ||2 approaches 0, while ||α t -z t ||2 tends to a smaller stable constant.
[0122] It can be understood that the representation coefficients of the test samples on the training sample set are extracted, and the obtained representation coefficients can simultaneously constrain the subspace competition representation items and the category representation vectors, increase the contribution of each category representation vector, and effectively map low-dimensional nonlinear data to high-dimensional space.
[0123] As a preferred embodiment, in step S3, the process of classifying the test sample based on the representation coefficient and the mapping matrix includes:
[0124] The non-negative representation classification model based on the autoencoder calculates the reconstructed residual of the test sample in the high-dimensional linear space according to the obtained representation coefficients and mapping matrices T1 and T2:
[0125] r i =||φ(T1y)-φ(T1X i )α i ||2;
[0126] Classification is performed based on the residuals, and the test sample is identified as the category corresponding to the minimum residual:
[0127]
[0128] Among them, y represents the test sample, X i represents the i-th type of training samples, α i represents the coefficient vector corresponding to the i-th class of training samples.
[0129] It can be understood that by classifying the test samples based on the representation coefficients, the test samples are identified as the categories corresponding to the minimum residuals, thereby effectively completing various image classification tasks and improving the accuracy of image classification.
[0130] In this embodiment, the samples in the training sample set are preprocessed, which can improve the reliability of the sample data in the training sample set and improve the efficiency of subsequent data calculation; the representation coefficient of the test sample in the training sample set is extracted, and the obtained representation coefficient can simultaneously constrain the subspace competition representation item and the category representation vector, increase the contribution of each category representation vector, and can effectively map low-dimensional nonlinear data to high-dimensional space, ensuring that the sample has a linear structure in the high-dimensional feature space, and classifying the test sample based on the representation coefficient to obtain the category corresponding to the minimum residual identified by the test sample, thereby effectively completing various image classification tasks and improving the accuracy of image classification.
[0131] Embodiment 2:
[0132] A non-negative representation classification system based on autoencoders, see Figure 4 , the system comprising:
[0133] An acquisition module, used to acquire test samples and training sample sets for image classification;
[0134] A processing module, used for preprocessing the training sample set;
[0135] An extraction module is used to extract the representation coefficient and mapping matrix of the test sample on the training sample set according to the preprocessed training sample set;
[0136] The classification module classifies the test sample based on the representation coefficient and the mapping matrix to obtain the category corresponding to the minimum residual error of the test sample.
[0137] Embodiment three:
[0138] An electronic device comprises a memory, a processor and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, the steps of a non-negative representation classification method based on an autoencoder are implemented.
[0139] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.
Claims
1. A non-negative representation classification method based on autoencoder, characterized in that: The method comprises the following steps: Acquire test samples and training sample sets for image classification, and preprocess the training sample sets; According to the preprocessed training sample set, the representation coefficient and mapping matrix of the test sample on the training sample set are extracted; The test sample is classified based on the representation coefficient and the mapping matrix to obtain a category corresponding to the minimum residual error of the test sample.
2. The non-negative representation classification method based on autoencoder according to claim 1, characterized in that: Get the training sample set for image pattern classification. The expression is: in, n i Represents the i-th type of training sample X i C represents the total number of sample categories; n represents the total number of training samples; α represents the feature vector of each sample; m represents the feature dimension of each sample.
3. The non-negative representation classification method based on autoencoder according to claim 1, characterized in that: The process of preprocessing the training sample set includes: The samples in the training sample set are normalized using the l2 norm.
4. The non-negative representation classification method based on autoencoder according to claim 2, characterized in that: The process of extracting the representation coefficients and mapping matrix of the test sample on the training sample set includes: Constructing a non-negative representation classification model based on an autoencoder, and constructing an optimization problem by using the non-negative representation classification model based on the autoencoder with the goal of optimizing the non-negative representation classification; Iteratively solving the optimization problem to obtain a non-negative representation classification model based on an autoencoder that grasps the representation coefficients and mapping matrices of the test samples on the training sample set; The non-negative representation classification model based on the autoencoder extracts the representation coefficients and mapping matrix of the test samples on the training sample set; Among them, the non-negative representation classification model based on autoencoder is a neural network model.
5. The non-negative representation classification method based on autoencoder according to claim 4, characterized in that: The process of constructing an optimization problem involves: The non-negative representation classification model based on the autoencoder constructs an optimization objective function of non-negative representation classification of a non-negative least squares problem according to the test sample and the training sample set. The expression is: stα≥0; Use the variable splitting method to solve the representation coefficient α, introduce the auxiliary variable z, and rewrite the non-negative representation classification objective function into the following form: stz=α,z≥0; Among them, T1 represents the encoding matrix, T2 represents the decoding matrix, φ represents the nonlinear function, λ represents the regularization parameter, y represents the test sample, and α represents the coefficient.
6. The non-negative representation classification method based on autoencoder according to claim 5, characterized in that: The process of iteratively solving the optimization problem includes: The alternating direction multiplier method is used to solve the corresponding augmented Lagrangian function, which is expressed as: The alternating optimization method is used to iteratively update the variables T1, T2, α, z and δ. The process includes: Fix T2, α, z and δ, update T1, the expression is: Fix T1, α, z and δ, and update T2, the expression is: Fix T1, T2, z and δ, update α, the expression is: Fix T1, T2, α and δ, and update z. The expression is: Update δ, the expression is: δ = δ + μ (z - α); In the process of iteratively solving the optimization problem, when the non-negative representation classification model based on the autoencoder converges, the non-negative representation classification model based on the autoencoder that grasps the representation coefficient and mapping matrix of the test sample on the training sample set is obtained, and the representation coefficient and mapping matrix T1 and T2 of the test sample on the training sample set are extracted by using the non-negative representation classification model based on the autoencoder; Among them, δ represents the Lagrange multiplier vector, μ>0 represents the penalty parameter, <δ, z-α> represents the inner product of two vectors, γ represents the learning rate, and T1 is continuously updated until the convergence condition is reached or the maximum number of cycles is reached, and ε is a smaller threshold.
7. The non-negative representation classification method based on autoencoder according to claim 6, characterized in that: The expression of the convergence condition of the non-negative representation classification model based on the autoencoder is: ||α t -z t ||2≤η,||α t+1 -α t ||2≤η and ||z t+1 -z t ||2≤ηsatisfies at the same time; Here, η represents a smaller threshold.
8. The non-negative representation classification method based on autoencoder according to claim 7, characterized in that: The process of classifying the test sample based on the representation coefficient and the mapping matrix includes: The non-negative representation classification model based on the autoencoder calculates the reconstructed residual of the test sample in the high-dimensional linear space according to the obtained representation coefficients and mapping matrices T1 and T2: r i =||φ(T1y)-φ(T1X i )a i ||2; Classification is performed based on the residuals, and the test sample is identified as the category corresponding to the minimum residual: Among them, y represents the test sample, X i represents the i-th type of training samples, α i represents the coefficient vector corresponding to the i-th class of training samples.
9. A non-negative representation classification system based on autoencoder, characterized in that: The system comprises: An acquisition module, used to acquire test samples and training sample sets for image classification; A processing module, used for preprocessing the training sample set; An extraction module is used to extract the representation coefficient and mapping matrix of the test sample on the training sample set according to the preprocessed training sample set; The classification module classifies the test sample based on the representation coefficient and the mapping matrix to obtain a category corresponding to the minimum residual error of the test sample.
10. An electronic device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.