Spoon defect detection method and system, electronic device and computer storage medium

Abstract

The application discloses a spoon defect detection method and system, electronic equipment and computer storage medium, and relates to the field of machine learning. The method comprises the following steps: acquiring a spoon image; constructing a spatial self-representation tensor according to the spoon image; constructing a multi-view clustering model based on rank approximation and sparse constraint according to the spatial self-representation tensor; solving a target function of the multi-view clustering model to obtain a similarity matrix of multi-view data; and performing clustering by using a spectral clustering algorithm according to the similarity matrix to obtain a spoon defect detection result. The application can improve clustering performance.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of machine learning, and in particular to a spoon defect detection method and system, an electronic device and a computer storage medium. BACKGROUND

[0002] Clustering technology aims to divide unlabeled data samples into different clusters according to their similarities, so that the distances of samples within the cluster are as small as possible, and the distances of samples between the clusters are as large as possible. It is a very important unsupervised learning technology in the field of computer application. With the development of computer technology and the Internet, the means of obtaining data has greatly improved, and the description of things has also developed from a single view to multiple views. Multiple views usually depict the same object from multiple angles, and each specific view reveals the specific attributes of the object. Therefore, multiple view data has heterogeneous characteristics, but also has potential connections between each other. Multi-view clustering refers to a method of clustering learning using multi-view data. This method integrates the data features in different views and provides a more complete clustering method, thereby improving the performance of the clustering algorithm. Multi-view subspace clustering is an important research direction of multi-view clustering, which maps samples in the original data space to a low-dimensional subspace using self-representation learning, then constructs a similarity matrix using the self-representation matrix, and finally applies a spectral clustering algorithm on the similarity matrix to obtain the clustering result. Therefore, it is crucial to construct a similarity matrix that is friendly to clustering in order to improve the performance of clustering.

[0003] Most existing multi-view clustering methods use methods for learning representation tensors to explore the high-order correlation between multi-view data, such as the multi-view subspace clustering method based on the tensor kernel norm. However, the sum of singular values of the tensor kernel norm and the positive frontal slice of the tensor in the Fourier domain is linearly proportional, which will cause the optimization algorithm to excessively punish large singular values in the process of optimizing the objective function, and thus easily make the model optimization fall into local optimum. In addition, existing methods only consider the global information between different views based on tensor learning, while ignoring the sparsity of the self-representation matrix specific to each view and its local information, resulting in that the effective information of multi-view data cannot be fully utilized. SUMMARY

[0004] The purpose of the present application is to provide a spoon defect detection method and system, an electronic device and a computer storage medium, which utilize the tensor logarithmic norm and the sparse constraint specific to each view, and from the global and local perspectives, retain the consistency information of multi-view data and the heterogeneous information specific to each view, fully utilize the effective information of multi-view data, and thus improve the clustering performance.

[0005] To achieve the above purpose, the present application provides the following solutions:

[0006] A spoon defect detection method comprises the following steps:

[0007] Obtaining a spoon image;

[0008] Constructing a spatial self-representation tensor according to the spoon image;

[0009] Constructing a multi-view clustering model based on rank approximation and sparse constraint according to the spatial self-representation tensor;

[0010] Solving a target function of the multi-view clustering model to obtain a similarity matrix of multi-view data;

[0011] Clustering by using a spectral clustering algorithm according to the similarity matrix to obtain a spoon defect detection result.

[0012] Optionally, the step of constructing a spatial self-representation tensor according to the spoon image specifically comprises the following steps:

[0013] Constructing a subspace representation matrix of each view according to the spoon image;

[0014] Constructing a spatial self-representation tensor according to the subspace representation matrix of each view.

[0015] Optionally, the step of solving a target function of the multi-view clustering model to obtain a similarity matrix of multi-view data specifically comprises the following steps:

[0016] Performing variable substitution on the multi-view clustering model to obtain an equivalent optimization model;

[0017] Determining an unconstrained minimization problem by using an augmented Lagrangian algorithm according to the equivalent optimization model;

[0018] Solving the unconstrained minimization problem by using an alternating direction multiplier method to obtain an optimal parameter;

[0019] Determining a similarity matrix of multi-view data according to the optimal parameter.

[0020] The present application further provides a spoon defect detection system comprising:

[0021] An obtaining module for obtaining a spoon image;

[0022] A spatial self-representation tensor construction module for constructing a spatial self-representation tensor according to the spoon image;

[0023] A multi-view clustering model construction module for constructing a multi-view clustering model based on rank approximation and sparse constraint according to the spatial self-representation tensor;

[0024] A solving module for solving a target function of the multi-view clustering model to obtain a similarity matrix of multi-view data;

[0025] a clustering module configured to cluster according to the similarity matrix by using a spectral clustering algorithm to obtain a result of spoon defect detection.

[0026] Optionally, the constructing module configured to construct the spatial self-representation tensor according to the spatial self-representation tensor specifically comprises:

[0027] a subspace representation matrix constructing unit configured to construct a subspace representation matrix of each view according to the spoon image;

[0028] a spatial self-representation tensor constructing unit configured to construct a spatial self-representation tensor according to the subspace representation matrix of each view.

[0029] Optionally, the solving module specifically comprises:

[0030] a variable replacing unit configured to perform variable replacement on the multi-view clustering model to obtain an equivalent optimization model;

[0031] an unconstrained minimization problem determining unit configured to determine an unconstrained minimization problem according to the equivalent optimization model by using a augmented Lagrangian algorithm;

[0032] a solving unit configured to solve the unconstrained minimization problem by using an alternating direction multiplier method to obtain an optimization parameter;

[0033] a similarity matrix determining unit configured to determine a similarity matrix of multi-view data according to the optimization parameter.

[0034] The application further provides an electronic device comprising:

[0035] one or more processors;

[0036] a storage device having one or more programs stored thereon;

[0037] When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of the above.

[0038] The application further provides a computer storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the method according to any one of the above.

[0039] According to the embodiments of the application, the following technical effects are achieved:

[0040] The application acquires a spoon image; constructs a spatial self-representation tensor according to the spoon image; constructs a multi-view clustering model based on rank approximation and sparse constraint according to the spatial self-representation tensor; solves a target function of the multi-view clustering model to obtain a similarity matrix of multi-view data; and performs clustering by using a spectral clustering algorithm according to the similarity matrix to obtain a spoon defect detection result. By using tensor log norm and sparse constraint on a specific view, consistency information of multi-view data is retained from a global and local perspective, heterogeneous information specific to each view is retained, effective information of multi-view data is fully utilized, and thus clustering performance is improved. BRIEF DESCRIPTION OF DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0042] Figure 1 Flowchart of the multi-view subspace clustering method;

[0043] Figure 2 Flowchart of the multi-view subspace clustering method;

[0044] Figure 3 Perspective structural diagram of the detection device for collecting spoon data samples;

[0045] Figure 4 Top view of the detection device;

[0046] Figure 5 Flowchart of the spoon defect detection method provided by the present application. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0048] The present application aims to provide a spoon defect detection method, system, electronic device and computer storage medium, which utilizes tensor log norm and sparse constraint on a specific view to retain consistency information of multi-view data from a global and local perspective, retain heterogeneous information specific to each view, fully utilize effective information of multi-view data, and thus improve clustering performance.

[0049] In order to make the above objectives, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0050] As shown in Figure 1 and Figure 5 , the present application provides a spoon defect detection method, comprising:

[0051] Step 101: obtaining a spoon image. The spoon image of the present application is obtained by the detection device as shown in Figure 3 and Figure 4 .

[0052] As shown in Figure 2 , the data matrix of the input image is represented by X = [x1, x2, …, x N ] ∈ R d×N , where d represents the feature dimension of the data, N represents the sample number of the data set, and each column of the matrix represents an image sample. v types of features of the data set are extracted by a feature extraction algorithm, and X (v) represents the data matrix of the vth view.

[0053] Step 102: constructing a spatial self-representation tensor according to the spoon image.

[0054] Step 102, specifically comprising:

[0055] constructing a subspace representation matrix of each view according to the spoon image; and constructing a spatial self-representation tensor according to the subspace representation matrix of each view.

[0056] constructing a subspace representation matrix Z (v) of each view:

[0057] X (v) = X (v) Z (v) + E (v) , v = 1, 2, … V;

[0058] wherein v represents the vth view of the multi-view data set X, X (v) represents the data matrix of the vth view, Z (v) represents the subspace representation matrix of the vth view, and E (v) represents the noise matrix of the vth view.

[0059] constructing a base subspace self-representation tensor Z:

[0060] Z = φ(Z (1) , Z (2) , …, Z (V） );

[0061] Where φ(·) represents the self-representation matrix Z of different views. (V) Construct an N×N×V third-order tensor as a frontal slice of the tensor, and then rotate the tensor dimensions to N×V×N to obtain the final self-representation tensor Z.

[0062] Step 103: Construct a multi-view clustering model based on rank approximation and sparse constraints according to the spatial self-representation tensor.

[0063] A multi-view clustering model based on rank approximation and sparse constraints is constructed, with the following objective function:

[0064]

[0065] stX (v) =X (v) Z (v) +E (v) v = 1, 2, ... V;

[0066] Z=φ(Z (1) Z (2) , ..., Z (V) );

[0067] E = [E (1) E (2) ;...;E (V) ]

[0068] Where Z represents the subspace self-representation tensor, and E represents the matrix obtained by concatenating V noise matrices row-wise, ||·|| TLD Denotes the logarithmic determinant of a tensor, ||·|| 2，1 This indicates that l is applied to the noise matrix. 2，1 The norm constraint, ||·||1, indicates that the l1 norm constraint is applied to different self-representation matrices, and λ and β represent penalty parameters.

[0069] It can be seen that the above objective function has the following characteristics: 1) it is generally believed that the singular values obtained by performing matrix singular value decomposition on the tensor frontal slice often correspond to important information in the matrix, and the importance and the matrix singular value size are positively correlated, and in addition, smaller singular values are usually considered to be caused by the noise of the input image. Therefore, the TLD tensor logarithmic norm in the model protects the large singular value and appropriately non-linearly punishes the smaller singular value, which not only realizes the compression of the data, but also removes the noise in the picture; 2) the model not only describes the low-rank structure of the subspace tensor by integrating the existing prior information (the feature difference of the normal spoon is relatively small, and the feature of the defective spoon has great difference due to the similarity and difference of the size, shape and position of the defect), but also uses the l1 norm to constrain the sparsity of each self-representation matrix, and finds the optimal self-representation matrix from the global and local information; 3) the model uses the l 2，1 norm constraint noise matrix, which makes the noise matrix row sparse, can effectively constrain the sample-specific noise in the input spoon image, thereby reducing the interference of noise and improving the clustering performance.

[0070] Step 104: solving the objective function of the multi-view clustering model to obtain a similarity matrix of the multi-view data.

[0071] Step 104, specifically comprising:

[0072] The multi-view clustering model is replaced by variables to obtain an equivalent optimization model.

[0073] According to the equivalent optimization model, an augmented Lagrangian algorithm is used to determine an unconstrained minimization problem.

[0074] The unconstrained minimization problem is solved by using an alternating direction multiplier method to obtain an optimal parameter.

[0075] According to the optimal parameter, a similarity matrix of the multi-view data is determined.

[0076] Solving the objective function. Since the objective function is a multivariate solving problem, an alternating direction multiplier method is used for optimization.

[0077] 1) In order to make the objective function decomposable into a plurality of sub-problems containing only one variable, auxiliary tensor variables A and auxiliary subspace representation matrices B (v) After simple variable replacement, the objective function is converted as follows:

[0078]

[0079] s.t.X (v) =X (v) Z (v) +E (v) , v = 1, 2,... V;

[0080] Z = φ(Z (1) , Z (2) ,..., Z (V) );

[0081] E = [E (1) ; E (2) ;... ; E (V) ];

[0082] Z = A;

[0083] Z (v) = B (v)

[0084] 2) Using the augmented Lagrangian algorithm, the model is rewritten as an unconstrained minimization problem:

[0085]

[0086] where L(·) is the Lagrangian function, matrices Y v , Q v and tensor W represent three Lagrange multipliers, is the square of the Frobenius norm, μ, σ and ρ are penalty factors.

[0087] 3) Fixing other variables in step 2) to update variable Z (v) , the optimization subproblem of Z (v) can be solved by the following formula:

[0088]

[0089] where and are three auxiliary variables introduced, φ -1 (W) is the inverse operation of φ(W), is the v-th component of φ -1 (W), that is, the v-th frontal slice in the multi-view tensor W, then W (v) represents the v-th frontal slice of tensor W.

[0090] 4) Fixing other variables in step 2) to update variable E, the optimization subproblem of E can be solved by the following formula:

[0091]

[0092] where D is the matrix obtained by connecting V view matrices vertically along the column direction. The optimal solution of the above formula is:

[0093]

[0094] where D ：，i denotes the i-th column of matrix D.

[0095] 5) Fixing other variables in step 2) to update variable B (v) , the optimization subproblem of B (v) can be solved by the following formula:

[0096]

[0097] where S is a soft threshold shrinkage operator.

[0098] 6) Fixing other variables in step 2) to update variable A, the optimization subproblem of A can be solved by the following formula:

[0099]

[0100] 7) Fixing other parameters, updating the Lagrange multiplier and the penalty parameter, where:

[0101]

[0102]

[0103] W * = W + p(Z - A)

[0104] μ = ημ, p = ηp, σ = ησ

[0105] 8) Iteratively performing steps 3) to 7) until the algorithm converges, outputting the optimal self-representation matrix Z (v) , the optimization parameters of the objective function are solved by using the alternating direction multiplier method, and the similarity matrix S of the multi-view data is solved according to the optimization parameters:

[0106]

[0107] Step 105: Clustering is performed using a spectral clustering algorithm according to the similarity matrix, to obtain a spoon defect detection result.

[0108] Spoon defect detection. The solved similarity matrix S is taken as the input of the spectral clustering algorithm to obtain the low-dimensional feature embedding of the spoon image. Then, the kmeans algorithm is used to cluster these low-dimensional feature embeddings into two clusters, i.e., a positive sample cluster (not containing defects) and a negative sample cluster (containing defects). To effectively distinguish the classes of the above two clusters, they are defined as p cluster and n cluster, respectively. Let μ p be the cluster center of the p cluster, μ n be the cluster center of the n cluster, and μ0be the average value of the pre-collected positive sample features. If |μ p- m0| < | m0- m n If p cluster is a positive sample cluster. Let e be the average distance of samples in the positive sample cluster to the cluster center, x be an arbitrary sample in another cluster, calculate the distance of all samples in another cluster to the cluster center of the positive sample cluster p | of another cluster to the cluster center of the positive sample cluster p | of another cluster to the cluster center of the positive sample cluster p If | x- m | < e, x belongs to the positive sample, otherwise x belongs to the defective sample. Finally, after the classification of all samples is completed, the cluster center of the positive sample cluster is recalculated and the m0 is updated as

[0109] The application aims to provide a multi-view unsupervised spoon defect detection method based on rank approximation and sparse constraint, which effectively distinguiates normal spoons and defective spoons by using a clustering learning method combined with a defect discrimination strategy. The proposed method uses tensor logarithmic norm and sparse constraint on specific views, and starts from the global and local angles, retains the consistency information of multi-view data and the heterogeneous information specific to each view, fully utilizes the effective information of multi-view data, and thus improves the clustering performance.

[0110] In order to verify the multi-view unsupervised spoon defect detection method based on rank approximation and sparse constraint provided by the embodiment of the application, the proposed method is verified by collecting data on a real spoon production line. The spoon data set contains 160 pictures, one containing 80 normal spoons and the other containing 80 defective spoons. Three different types of features of different image samples are extracted, which are 1024-dimensional Gabor features, 256-dimensional Harris features and 3304-dimensional LBP features. The three different types of features are divided into three independent views, thereby constructing a multi-view spoon data set. The application and two mainstream methods are compared in terms of effect, which are a multi-view method: multi-view subspace clustering based on tensor singular value decomposition (t-SVD-MSC) and a single-view method: robust recovery of subspace structure based on low-rank representation (LRR). Two evaluation indexes are used to measure the effectiveness of the method, which are accuracy (Accuracy) and precision (Precision). Among them, the accuracy Accuracy can be defined by the following formula:

[0111]

[0112] The precision Precision can be defined by the following formula:

[0113]

[0114] ​Wherein, TP represents the number of positive samples determined as positive samples and actually positive samples; TN represents the number of negative samples determined as negative samples and actually negative samples; FP represents the number of positive samples determined as negative samples and actually negative samples; and FN represents the number of negative samples determined as positive samples and actually positive samples.

[0115] Table 1 Accuracy (%) of clustering results of the present application on the spoon dataset

[0116] Our work t-SVD-MSC LRR Spoon 80.63％ 56.25％ 50.63％

[0117] Table 2 Precision (%) of clustering results of the present application on the spoon dataset

[0118] Our work t-SVD-MSC LRR Spoon 68.02％ 50.16％ 49.7％

[0119] As shown in Table 1 and Table 2, from the clustering results, it can be seen that, compared with the comparative method, the present application has great advantages in identifying the accuracy of spoon defect samples. As the evaluation index of Accuracy%, the present application method is higher than the second best t-SVD-MSC by nearly 24.4%, which proves the effectiveness of the present application.

[0120] The present application also provides a spoon defect detection system, comprising:

[0121] An acquisition module is configured to acquire a spoon image.

[0122] A spatial self-representation tensor construction module is configured to construct a spatial self-representation tensor according to the spoon image.

[0123] A multi-view clustering model construction module is configured to construct a multi-view clustering model based on rank approximation and sparse constraint according to the spatial self-representation tensor.

[0124] A solution module is configured to solve a target function of the multi-view clustering model to obtain a similarity matrix of multi-view data.

[0125] A clustering module is configured to perform clustering by using a spectral clustering algorithm according to the similarity matrix to obtain a spoon defect detection result.

[0126] As an optional implementation, the spatial self-representation tensor construction module specifically comprises:

[0127] A subspace representation matrix construction unit is configured to construct a subspace representation matrix of each view according to the spoon image.

[0128] A spatial self-representation tensor construction unit is configured to construct a spatial self-representation tensor according to the subspace representation matrix of each view.

[0129] As an optional implementation, the solution module specifically comprises:

[0130] a variable replacement unit configured to perform variable replacement on the multi-view clustering model to obtain an equivalent optimization model.

[0131] an unconstrained minimization problem determination unit configured to determine an unconstrained minimization problem by using a augmented Lagrange algorithm according to the equivalent optimization model.

[0132] a solution unit configured to solve the unconstrained minimization problem by using an alternating direction multiplier method to obtain an optimization parameter.

[0133] a similarity matrix determination unit configured to determine a similarity matrix of the multi-view data according to the optimization parameter.

[0134] The application further provides an electronic device, comprising:

[0135] one or more processors.

[0136] a memory device having one or more programs stored thereon.

[0137] The one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method according to any one of the preceding embodiments.

[0138] The application further provides a computer storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method according to any one of the preceding embodiments.

[0139] The application discloses a multi-view unsupervised spoon defect detection method based on rank approximation and sparse constraint, which comprises the following steps: (1) collecting n spoon samples in a fixed field of view on a conveying mechanism by using an image sensor; (2) extracting various types of features of the spoon samples by using a feature extraction algorithm; (3) mapping different types of spoon samples in an original high-dimensional space to a low-dimensional subspace by using a self-representation learning method; (4) constructing a self-representation tensor based on a subspace representation matrix, and imposing low rank and sparse constraints on the tensor and the self-representation matrix respectively, so as to construct a more efficient similarity matrix; (5) taking the similarity matrix as an input of a spectral clustering algorithm to obtain a clustering result; and (6) discriminating normal sample clusters by using existing normal spoon samples, and discriminating whether each sample is a normal spoon or a defective spoon according to the distance of each sample to a normal sample cluster center. The method innovatively uses an unsupervised learning framework to detect spoon defects, applies low rank and sparse constraints to the construction of a spoon similarity matrix, fully explores high-order correlation and complementary information between multi-view data of the spoon, contains consistency between different views and retains diversity between different views, and thus improves clustering performance.

[0140] The various embodiments described in this specification are presented for the purpose of illustrating the principles of the present application and its best mode of operation. Each of the embodiments described in this specification has been provided for the purpose of illustration only and the various embodiments are not intended to limit the present application in any way unless otherwise specifically indicated. The same parts and / or features of the various embodiments described in this specification can be referenced using the same reference numerals for the ease of understanding of the present application.

[0141] The principles and implementations of the present application have been described in the above embodiments, which are only used to help understand the method of the present application and its core idea. Meanwhile, for those skilled in the art, the specific implementation and application range of the present application can be changed according to the idea of the present application. In summary, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A method of detecting a spoon defect, the method comprising: The method comprises the following steps: obtaining a spoon image; constructing a spatial self-representation tensor according to the spoon image; constructing a multi-view clustering model based on rank approximation and sparse constraint according to the spatial self-representation tensor; solving a target function of the multi-view clustering model to obtain a similarity matrix of multi-view data; constructing a multi-view clustering model based on rank approximation and sparse constraint, and the target function is as follows: s.t.X (v) = X (v) Z (v) + E (v) v = 1, 2,... V; Z = φ(Z (1) ,Z (2) ,...,Z (V) ) E = [E (1) ; E (2) ;... ; E (V) ] where X (v) represents the data matrix of the v-th view, φ(·) represents the auto-encoder matrix of different views Z (V) represents a N×N×V third-order tensor constructed by the frontal slice of the tensor, Z represents the subspace auto-encoder tensor, E represents a matrix obtained by concatenating V noise matrices in rows, ||·||F TLD represents the tensor logarithmic determinant, ||·||F 2,1 represents the l 2,1 norm constraint on the noise matrix, ||·||1 represents the l1 norm constraint on different auto-encoder matrices, and λ and β represent penalty parameters; In order to make the objective function decomposable into multiple subproblems containing only one variable, auxiliary tensor variable A and auxiliary subspace representation matrix B are introduced ( v) After simple variable substitution, the transformed objective function is as follows: s.t.X (v) = X (v) Z (v) + E (v) v = 1, 2,... V; Z = φ(Z (1) ,Z (2) ,...,Z (V) ) ; E = [E (1) ; E (2) ;... ; E (V) ] ; Z = A; Z (v) = B (v) using an augmented Lagrange algorithm, the model is rewritten as an unconstrained minimization problem: where L(·) is the Lagrangian function, the matrix Y v v and the tensor W represent three Lagrange multipliers, is the square of the Frobenius norm, and μ, σ, and ρ are penalty factors;​ Fixing the other variables to update variable B (v) , B (v) The optimization sub-problem of B can be solved by the following formula: where S is a soft threshold shrinkage operator; performing clustering according to the similarity matrix by using a spectral clustering algorithm to obtain a spoon defect detection result.

2. The method of claim 1, wherein, The step of constructing a spatial self-representation tensor according to the spoon image specifically comprises the following steps: constructing a subspace representation matrix of each view according to the spoon image; constructing a spatial self-representation tensor according to the subspace representation matrix of each view.

3. The method of claim 1, wherein the step of detecting the spoon defect is characterized by, The step of solving a target function of the multi-view clustering model to obtain a similarity matrix of multi-view data specifically comprises the following steps: performing variable substitution on the multi-view clustering model to obtain an equivalent optimization model; determining an unconstrained minimization problem according to the equivalent optimization model by using an augmented Lagrange algorithm; solving the unconstrained minimization problem by using an alternating direction multiplier method to obtain an optimization parameter; determining the similarity matrix of multi-view data according to the optimization parameter.

4. A spoon defect detection system characterized by, The method comprises the following steps: an obtaining module, configured to obtain a spoon image; a spatial self-representation tensor construction module, configured to construct a spatial self-representation tensor according to the spoon image; a multi-view clustering model construction module, configured to construct a multi-view clustering model based on rank approximation and sparse constraint according to the spatial self-representation tensor; a solving module, configured to solve a target function of the multi-view clustering model to obtain a similarity matrix of multi-view data; constructing a multi-view clustering model based on rank approximation and sparse constraint, and the target function is as follows: s.t.X (v) = X (v) Z (v) + E (v) v = 1, 2,... V; Z = φ(Z (1) ,Z (2) ,...,Z (V) ) E = [E (1) ; E (2) ;... ; E (V) ] where X (v) represents the data matrix of the v-th view, φ(·) represents the auto-encoder matrix of different views Z (V) represents a N×N×V third-order tensor constructed by the frontal slice of the tensor, Z represents the subspace auto-encoder tensor, E represents a matrix obtained by concatenating V noise matrices in rows, ||·||F TLD represents the tensor logarithmic determinant, ||·||F 2,1 represents the l 2,1 norm constraint on the noise matrix, ||·||1 represents the l1 norm constraint on different auto-encoder matrices, and λ and β represent penalty parameters; To make the objective function decomposable into multiple subproblems containing only one variable, an auxiliary tensor variable A and an auxiliary subspace representation matrix B are introduced (v) After simple variable substitution, the transformed objective function is as follows: s.t.X (v) = X (v) Z (v) + E (v) v = 1, 2,... V; Z = φ(Z (1) ,Z (2) ,...,Z (V) ) E = [E (1) ; E (2) ;... ; E (V) ] ; Z = A; Z (v) = B (v) using an augmented Lagrange algorithm, the model is rewritten as an unconstrained minimization problem: where L(·) is the Lagrangian function, the matrix Y v v and the tensor W represent three Lagrange multipliers, is the square of the Frobenius norm, and μ, σ, and ρ are penalty factors;​ Fixing the other variables to update variable B (v) , B (v) The optimization sub-problem of B can be solved by the following formula: where S is a soft threshold shrinkage operator; a clustering module, configured to perform clustering according to the similarity matrix by using a spectral clustering algorithm to obtain a spoon defect detection result.

5. The spoon defect detection system of claim 4, wherein, The spatial self-representation tensor construction module specifically comprises the following modules: a subspace representation matrix construction unit, configured to construct a subspace representation matrix of each view according to the spoon image; a spatial self-representation tensor construction unit, configured to construct a spatial self-representation tensor according to the subspace representation matrix of each view.

6. The spoon defect detection system of claim 4, wherein The solving module specifically comprises the following modules: a variable substitution unit, configured to perform variable substitution on the multi-view clustering model to obtain an equivalent optimization model; an unconstrained minimization problem determination unit, configured to determine an unconstrained minimization problem according to the equivalent optimization model by using an augmented Lagrange algorithm; a solving unit, configured to solve the unconstrained minimization problem by using an alternating direction multiplier method to obtain an optimization parameter; a similarity matrix determination unit, configured to determine the similarity matrix of multi-view data according to the optimization parameter.

7. An electronic device, comprising: The device comprises the following components: one or more processors; a storage device having one or more programs stored thereon; The one or more programs, when executed by the one or more processors, cause the one or more processors to implement a method according to any one of claims 1 to 3.

8. A computer storage medium, characterized in that A computer program product having stored thereon a computer program which, when executed by a processor, implements a method according to any one of claims 1 to 3.

Images