Neighborhood rough set attribute reduction method for eye disease image diagnosis

Abstract

The invention provides a neighborhood rough set attribute reduction method for eye disease image diagnosis. The method comprises the following steps: step 1, reading and processing an eye disease image data set; 2, carrying out distributed processing on the eye disease image data set, and building a distributed framework; step 3, respectively performing spectral clustering on the feature set and the label set, and finding out a label main cluster and a feature cluster in the feature set and the label set; 4, performing feature selection through neighborhood rough set selection, and selecting a local optimal feature subset; and step 5, performing feature fusion on the local optimal feature subset to enable the fused feature subset to be a global optimal feature subset. According to the method, the eye disease image detection time can be shortened, the detection efficiency is improved, and a doctor is assisted in eye disease diagnosis.

Description

Technical Field

[0001] The present invention belongs to the field of auxiliary medical treatment, and in particular relates to a neighborhood rough set attribute reduction method for eye disease image diagnosis. Background Art

[0002] Eye diseases are a global public health concern, with their causes influenced by a variety of factors, including an aging population and changing lifestyles. With the increasing aging of society, the prevalence of retinal degenerative diseases is rising. A wide variety of eye diseases, including cataracts, glaucoma, dry eye, fundus disease, and age-related macular degeneration, significantly impact patients' quality of life. Therefore, regular eye examinations and appropriate preventive measures are crucial for maintaining eye health.

[0003] At present, there is no efficient diagnostic technology that can be used to diagnose eye diseases. The diagnosis can only be based on the doctor's diagnostic experience. With the change of lifestyle, eye diseases have become the focus of global health concern. There is an urgent need for an effective method to extract areas that are helpful for eye disease diagnosis from eye disease data. Feature selection can help doctors analyze eye diseases and improve diagnostic efficiency.

[0004] Rough set theory is primarily used to address uncertainty and ambiguity in data. Its core concept is to analyze information by partitioning datasets into different equivalence classes using equivalence relations. Rough sets can handle incomplete or ambiguous information, describing the boundaries of data through upper and lower approximations. Unlike probability theory and fuzzy set theory, rough sets do not require prior knowledge and rely solely on the data itself. Using attribute reduction techniques, rough set theory can extract key features from data and reduce redundant information. Classical rough sets also have their limitations, as they can only handle discrete data. Hu Qinghua et al. extended classical rough sets to neighborhood rough sets. Building on classical rough sets, Sun Lin et al. proposed the introduction of neighborhood rough sets in multi-label classification in "A Multi-label Specific Feature Selection Method Based on Neighborhood Rough Sets" (Computer Science 2018, 173-178). However, this approach neglects the complementary nature of features. Some features, when considered individually, may not contribute much to classification, but when combined, they can provide more valuable information. Similarly, in multi-label data, there are correlations between labels. For example, in medicine, patients with high blood pressure are more likely to suffer from cardiovascular disease than others. Therefore, it is necessary to consider the correlation between features and labels in multi-label learning. Therefore, how to consider the correlation between features and labels in multi-label learning is the focus of this research. Summary of the Invention

[0005] Purpose of the invention: The technical problem to be solved by the present invention is to address the deficiencies of the existing technology and provide a neighborhood rough set attribute reduction method for ocular disease imaging diagnosis, comprising the following steps:

[0006] Step 1: read and process the eye disease image dataset;

[0007] Step 2: Perform distributed processing on the eye disease image dataset and build a distributed framework;

[0008] Step 3: Perform spectral clustering on the feature set and label set respectively to find the main label cluster and feature cluster;

[0009] Step 4: Perform feature selection through neighborhood rough set selection to select the local optimal feature subset;

[0010] Step 5: perform feature fusion on the local optimal feature subset to make the fused feature subset the global optimal feature subset.

[0011] Step 1 includes the following steps:

[0012] Step 1-1: Read the eye disease image dataset, determine the feature set and decision classification, and establish a multi-label decision system MNDT=<U,C∪D,Δ,δ> , where U={x1,x2,...,x N} represents the set of detection objects in the eye disease image dataset, N is the number of detection objects, x N Indicates the Nth inspection object, C={a1,a2,...,a n} represents the feature set of data attributes, n is the number of sample features, a n is the nth eigenvalue of the sample, D={l1,l2,...,l d} represents the decision class of the sample, where d is the number of sample labels, l d represents the dth sample label, Δ represents the distance function, and δ represents the neighborhood radius;

[0013] Step 1-2: Normalize the features in the eye disease data. The normalization formula is:

[0014]

[0015] Where x refers to the value of a feature in the original sample, x normalized Represents the value of the feature of the original sample after normalization, x max Refers to the maximum value of all samples in an attribute, x min It represents the minimum value of all samples in an attribute.

[0016] Step 2 includes: building a Spark distributed framework, dividing the eye disease image dataset S' into m eye disease information subsets S' = {S'1, S'2, S'3, ..., S' m}, where S' m Represents the mth eye disease information subset, and the m eye disease information subsets are distributed in parallel to the m working nodes using a random sending strategy.

[0017] Step 3 includes the following steps:

[0018] Step 3-1: Perform spectral clustering on the feature set and label set in the eye disease information subset of each node to find the main label cluster and the secondary label cluster in each node. The main label cluster is the cluster with the largest number of labels, and the secondary label cluster is the cluster other than the main label cluster.

[0019] The feature clusters clustered by spectral clustering have high correlation between features within the cluster and weak similarity between features between clusters;

[0020] Step 3-2, for label spectrum clustering (similar to feature spectrum clustering): construct similarity matrix W∈R d×d , where R represents the real space, and the element W in the i-th row and j-th column of the similarity matrix W is ij Expressed as:

[0021]

[0022] Where the intermediate parameter σ = 1.0, exp represents the natural exponential function, and i and j range from 1 to d. The distance formula is defined as:

[0023]

[0024] where l iN Represents sample X N In the label l i The value on l jN Denotes sample X N In the label l j The value on

[0025] Step 3-3, construct the diagonal matrix Z;

[0026] Step 3-4, let the matrix A = ZW, normalize the Laplace matrix to obtain the normalized Laplace matrix A';

[0027] Step 3-5, select the eigenvectors v1, v2, ..., v corresponding to the first T smallest non-zero eigenvalues T , T is the preset number of clusters, v T The eigenvector representing the Tth smallest non-zero eigenvalue; the eigenvectors v1,v2,...,vT A matrix of size d×T;

[0028] Step 3-6, select the feature vectors v1, v2, ..., v T The matrix is clustered using K-means to obtain label clusters, which are sorted by the number of labels in the cluster. The cluster with the largest number of labels is the main label cluster, and the remaining clusters are sub-label clusters.

[0029] In step 3-3, the diagonal matrix Z is constructed using the following formula:

[0030]

[0031] Where Z(i,i) represents the sum of all elements in the i-th row of the similarity matrix W, which is called the degree matrix.

[0032] In steps 3-4, the normalized Laplace matrix A' is expressed as:

[0033] A'=IZ -1 W (4),

[0034] Where I is the identity matrix.

[0035] Step 4 includes the following steps:

[0036] Step 4-1, for any sample x i ∈U and B∈C, sample x i Neighborhood δ under conditional attribute subset B B (x i )for:

[0037] δ B (x i )={x j |x j ∈U,dis B (x i ,x j )≤δ} (5),

[0038] Where δ is the neighborhood size, if the i-th sample x i With the jth sample x j If the relationship (5) is satisfied, then x j In x i In the neighborhood of B (x i ,x j ) is the i-th sample x i With the jth sample x j The Euclidean distance under the conditional attribute subset B is defined as:

[0039]

[0040] Among them, f(x i ,a t ) represents the i-th sample x i The value corresponding to the tth feature;

[0041] Step 4-2, D j Represents a label l j The sample set, Indicates that in the label cluster L k The label set considered below, Indicates that in the label cluster L k Next sample x i The set of labels in the conditional feature subset Under this condition, the multi-label neighborhood rough set is approximated as:

[0042]

[0043] Among them, the lower approximation Describes the label cluster L k Next sample x i The neighborhood of completely belongs to the label cluster L k The sample set of

[0044] Upper approximation Describes the label cluster L k Next sample x i The neighborhood of may belong to the label cluster L k The sample set, boundary area Indicates that it is not possible to determine whether it belongs to the label cluster L k The sample set of

[0045] Step 4-3, under conditional feature subset B, conditional feature subset B for label cluster L k Dependence for:

[0046]

[0047] Step 4-4, for the multi-label decision system S=<U,C∪D,Δ,δ> , a∈CB, the importance of attribute a SIG(a,B,D) is defined as:

[0048] SIG(a,B,D)=λ B∪{a} (D)-λ B (D)(11),

[0049] where λ B∪{a}(D) represents the change in dependency caused by adding feature a under conditional feature subset B. If SIG(a,B,D)>0, it means that feature a is a useful feature for feature subset B. If SIG(a,B,D)<0, it means that feature a is a redundant feature for feature subset B.

[0050] Step 4-5: For the feature clusters obtained by spectral clustering, calculate the dependency of each feature cluster on the label main cluster L1 and find the feature set F with the largest dependency. i , according to formula (11) remove the feature set F i The redundant features in the feature set F after removing the redundant features i becomes the selected feature, and the remaining feature clusters are combined into the candidate feature subset;

[0051] Steps 4-6: For the label sub-cluster, the features in the candidate feature subset are added to the selected feature set in sequence. The dependency caused by the updated conditional feature subset is calculated according to formula (11), and the corresponding features with the maximum dependency are added to the selected feature set until the dependency no longer increases. The selected features are the feature subset corresponding to the label sub-cluster.

[0052] In steps 4-7, for the two feature subsequences selected from the label main cluster and the label sub-cluster respectively, features are selected iteratively and progressively to generate the final local optimal feature subset (the feature subset corresponding to the child node).

[0053] Step 5 includes: in the Spark distributed framework, aggregating the optimal feature subsets of each working node through the reduce operation of the master node to form a global optimal feature subset.

[0054] The present invention also provides an electronic device, comprising a processor and a memory, wherein the memory stores program code, and when the program code is executed by the processor, the processor executes the steps of the method.

[0055] The present invention also provides a storage medium storing a computer program or instruction, which executes the steps of the method when the computer program or instruction is run on a computer.

[0056] Compared with the prior art, the present invention has the following advantages: (1) The present invention uses a distributed Spark framework, which can significantly improve the data processing speed by distributing the data set to multiple nodes for parallel processing, achieving faster query response time and data analysis. Through parallel processing, the distributed system can execute multiple tasks at the same time, fully utilize the computing resources of the cluster, and improve the overall processing capability. In addition, distributed processing can optimize data locality, assign computing tasks to the nodes where the data is located, reduce data transmission, and improve processing efficiency.

[0057] (2) This invention uses spectral clustering technology to identify the main cluster of labels by performing spectral clustering on the label set. The main cluster of labels plays a decisive role in the entire label set. Similarly, there is a strong correlation between features. When performing spectral clustering on the feature set, features in the same cluster are highly correlated, while features in different clusters are weakly correlated. Therefore, performing spectral clustering on the feature set is beneficial for subsequent feature selection.

[0058] (3) The neighborhood rough set technology is introduced. By combining the neighborhood rough set with multi-label feature selection, the uncertainty in multi-label data is handled, the correlation between labels is considered, and the information loss caused by processing each label independently is avoided. The local structure of the data is retained through the neighborhood relationship, which helps to select more discriminative features. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, and the above and / or other advantages of the present invention will become more apparent.

[0060] Figure 1 Flow chart of the method of the present invention.

[0061] Figure 2 Detailed architecture diagram of the method of the present invention.

[0062] Figure 3 The data processing flow chart of the method of the present invention. DETAILED DESCRIPTION

[0063] See also Figure 1 、 Figure 2 and Figure 3 The embodiment of the present invention provides a neighborhood rough set attribute reduction method for eye disease imaging diagnosis, comprising the following steps:

[0064] S10, reading an eye disease image dataset, and processing the eye disease image dataset;

[0065] S20. Perform distributed processing on eye disease image datasets and build a distributed framework;

[0066] S30, performing spectral clustering on the feature set and the label set respectively, to find the main label cluster and the highly correlated feature clusters within the cluster;

[0067] S40, performing feature selection through neighborhood rough set to select the local optimal feature subset;

[0068] S50, performing feature fusion on the local feature subsets, so that the fused feature subsets become the global optimal feature subset;

[0069] The step S10 includes the following steps:

[0070] S11. Read the eye disease image dataset, determine its feature set and decision classification, and establish a multi-label decision system MNDT=<U,C∪D,Δ,δ> , where U={x1,x2,...x N} represents the set of detection objects in the eye disease image dataset, where N is the number of detection objects, C = {a1, a2, ...a n} represents the feature set of data attributes, where n is the number of sample features, a n is the nth eigenvalue of the sample, D={l1,l2,...l d} represents the decision class of the sample, where d is the number of sample labels, Δ represents the distance function, and δ represents the size of the neighborhood radius;

[0071] S12. Normalize the features in the eye disease data. The normalization formula is as follows:

[0072]

[0073] Where x refers to the value of a feature in the original sample, x normalized Represents the value of a feature of the original sample after normalization, x max Refers to the maximum value of all samples in a certain attribute, x min It represents the minimum value of all samples in a certain attribute;

[0074] The step S20 includes the following steps:

[0075] S21. Build a distributed framework to divide the eye disease image dataset S' into m eye disease information subsets S' = {S'1, S'2, S'3, ..., S' m}, where S' m Represents the mth subset of eye disease information, and the m subsets of eye disease information are distributed to m working nodes in parallel using a random sending strategy, such as Figure 2 The left side of the framework diagram shows the nodes. Assume that a data subset S1 obtained at a certain node is shown in Table 1.

[0076] Table 1

[0077] U <![CDATA[a1]]> <![CDATA[a2]]> <![CDATA[a3]]> <![CDATA[a4]]> <![CDATA[a5]]> <![CDATA[l1]]> <![CDATA[l2]]> <![CDATA[l3]]> <![CDATA[l4]]> <![CDATA[l5]]> <![CDATA[x1]]> 0.5710 0.6129 0.2120 0.3210 0.4200 1 1 0 1 0 <![CDATA[x2]]> 0.4811 0.4240 0.3330 0.0180 0.7120 1 1 0 0 0 <![CDATA[x3]]> 0.3590 0.4240 0.2700 0.5910 0.6380 0 0 1 0 1 <![CDATA[x4]]> 0.2989 0.5730 0.6030 0.9800 0.4210 0 0 1 1 1 <![CDATA[x5]]> 0.1883 0.7060 0.4720 0.5150 0.5720 1 1 0 1 0 <![CDATA[x6]]> 0.3521 0.8500 0.4360 0.5980 0.9080 1 1 1 0 1 <![CDATA[x7]]> 0.2684 0.2040 0.3140 0.9370 0.5630 1 0 1 0 1

[0078] The step S30 includes the following steps:

[0079] S31. At each node, perform spectral clustering on the feature set and label set in the eye disease information subset in the node, and find the main label cluster (the cluster with the largest number of labels) and the secondary label cluster (the cluster other than the main label cluster) in each node. The feature clusters clustered by spectral clustering have high correlation between features within the cluster and weak similarity between features between clusters; Figure 2 Schematic diagram of mesospectral clustering.

[0080] S32. Taking the label spectrum clustering as an example, the feature spectrum clustering is similar. Construct the similarity matrix W∈R d×d , where R represents the real space, and the element W in the i-th row and j-th column of the similarity matrix W is ij Expressed as:

[0081]

[0082] Where the intermediate parameter σ = 1.0, exp represents the natural exponential function, and the distance formula is defined as:

[0083]

[0084] where l iN Represents sample X N In the label l i The value on l jN Denotes sample X N In the label l j The value on .

[0085] The label similarity matrix is obtained as follows:

[0086]

[0087] S33, construct a diagonal matrix Z, Z(i,i) represents the sum of all elements in the i-th row of the similarity matrix W, which is called the degree matrix:

[0088]

[0089] S34. Let matrix A = ZW, and perform normalization on the Laplace matrix to obtain a normalized Laplace matrix A':

[0090] A'=IZ -1 W (4),

[0091]

[0092]

[0093] Where I is the identity matrix;

[0094] S35. Select the eigenvectors e1 and e2 corresponding to the first two smallest non-zero eigenvalues.

[0095] After calculation, the eigenvalues of A' are:

[0096] e1=0.283833107, e2=0.144247828, e3=-1.47790718×10 -16 、e4=0.028290718、

[0097] e5=1.00000000.

[0098] The eigenvalues corresponding to the eigenvectors are:

[0099]

[0100] After calculation, vectors v3 and v4 are the eigenvectors corresponding to the first two smallest non-zero eigenvalues:

[0101]

[0102] S36. Perform K-means clustering on the selected feature vectors.

[0103] Perform K-means clustering on the feature vector v, k = 2, and obtain the result [0 0 1 0 1].

[0104] According to the clustering results, the label cluster is clustered into two clusters, L1 and L2, with L1 = {l1, l2, l4} and L2 = {l3, l5}. Therefore, label cluster L1 is the primary label cluster for node S1, and L2 is the secondary label cluster for node S1. The primary label cluster has a more significant impact on the entire dataset than the label cluster, so when considering feature selection, the primary label cluster is given priority. Similarly, spectral clustering is performed on the features, and the result of spectral clustering is [1 0 1 2 0]. According to the feature clustering results, the labels are clustered into three clusters, F1 = {a1, a3}, F2 = {a2, a5}, and F3 = {a4}. The correlation between features within a cluster is strong, while the correlation between features between clusters is weak. Obviously, the feature clusters generated by feature spectral clustering are beneficial for subsequent feature selection.

[0105] The step S40 includes the following steps:

[0106] S41. For any sample x i ∈U and B∈C, sample x i Neighborhood δ under conditional attribute subset B B (x i )for:

[0107] δ B (x i )={x j |x j ∈U,dis B (x i ,x j )≤δ} (5),

[0108] Where δ is the neighborhood size. If the sample x i With sample x j If the relationship (5) is satisfied, then x j In x i In the neighborhood of B (x i ,x j ) is the sample x i With sample x j The Euclidean distance under the conditional attribute subset B is defined as:

[0109]

[0110] Among them, f(x i ,a t ) represents the sample x i The value corresponding to the tth feature;

[0111] S42, D j Represents a label l j The sample set, Indicates that in the label cluster L k The label set considered below, Indicates that in the label cluster L k Next sample x i The set of labels in the conditional feature subset Under this condition, the multi-label neighborhood rough set is approximated as:

[0112]

[0113] Among them, the lower approximation Describes the label cluster L k Next sample x i The neighborhood of completely belongs to the label cluster L k The sample set of

[0114] Upper approximation Describes the label cluster L k Next sample x i The neighborhood of may belong to the label cluster L k The sample set, boundary area Indicates that it is not possible to determine whether it belongs to the label cluster L k The sample set of

[0115] Under the label main cluster L1 = {l1, l2, l4}, find the feature main cluster with the greatest correlation with the label main cluster. Taking the feature cluster F1 as an example,

[0116] Under the label main cluster L1, L1={l1,l2,l4}, so Consider only the main cluster of labels, taking sample x1 as an example, so we have D 2 ={x1,x2,x5,x6}, D 4 ={x1,x4,x5}, according to formula (20) we can get Satisfying the constraints, we get Similarly, y2={l1,l2}, so Similarly

[0117] S43. Under conditional feature subset B, the dependency of conditional feature subset B on label cluster L1 is:

[0118]

[0119] For the above example, according to formula (23), under the feature cluster F1, the dependency on the main label cluster L1 = {l1, l2, l4} is

[0120] Under the main label cluster L1 = {l1, l2, l4}, Taking the feature cluster F2 = {a2, a5} as an example, the following neighborhood is obtained according to formula (19):

[0121] Under the main label cluster L1, L1={l1,l2,l4}. Considering only the main label cluster, taking sample x1 as an example, we know that y1={l1,l2,l4}, so we have D 1 ={x1,x2,x5,x6,x7}, D 2 ={x1,x2,x5,x6}, D 4 ={x1,x4,x5}, according to formula (20) we can get so Similarly, y2={l1,l2}, So we can get Similarly

[0122] According to formula (23), under the feature cluster F2, F2 = {a2, a5}, the dependency of the feature cluster F2 = {a2, a5} on the label main cluster L1 = {l1, l2, l4} is

[0123] Similarly, under the label cluster F3 = {a4}, the dependency on the main cluster label is

[0124] Under the main label cluster, the final feature cluster ranking is: F1 = {a1, a3}, F2 = {a2, a5}, F3 = {a4}.

[0125] S44. For the decision set S=<U,C∪D> , a∈B, the importance of attribute a is defined as:

[0126] SIG(a,B,D)=λ B∪{a} (D)-λ B (D) (11),

[0127] λ B∪{a} (D) represents the change in dependency caused by adding feature a under conditional feature subset B. If SIG(a,B,D)>0, it means that feature a is a useful feature for feature subset B. If SIG(a,B,D)<0, it means that feature a is a redundant feature for feature subset B.

[0128] S45. For the condition attribute subsets obtained by spectral clustering, calculate the dependency of each condition attribute subset on the label main cluster L1, and find the feature set F with the largest dependency. i , according to formula (11) remove the feature set F i The redundant features in the feature set F after removing the redundant features i becomes the selected feature, and the remaining feature clusters are combined into the candidate feature subset;

[0129] Taking F1={a1,a3} as an example, according to formula (24), After removing a1, we find Causes a decrease, indicating that a1 is a useful feature, This shows that a2 is also a useful feature, so the feature cluster F1 = {a1, a3} is indispensable for the main label cluster L1 = {l1, l2, l4}. Therefore, cluster F1 = {a1, a3} is the local optimal subset selected as the main label cluster.

[0130] S46. For the label sub-cluster, the features in the candidate feature subset are sequentially added to the selected feature set. The dependency caused by the update of the conditional feature subset is calculated according to formula (11), and the corresponding features with the maximum dependency are added to the selected feature set until the dependency no longer increases. The selected features are the feature subset corresponding to the label sub-cluster.

[0131] The label sub-cluster is L2 = {l3, l5}, and the candidate feature subset is F4 = {a2, a4, a5}. According to formula (24), the label sub-cluster feature subset is {a2, a5}.

[0132] S47. For the two feature subsequences selected from the label main cluster and the label sub-cluster respectively, generate a final local optimal feature subset (the feature subset corresponding to the child node) by iteratively selecting features.

[0133] So according to the above example, the final feature subsequence is: {a1, a2, a3, a5}. The multi-label evaluation index average precision is defined as follows:

[0134]

[0135] where Y i Represents sample x i The true label set of (i.e., the correct label in multi-label), |Y i | represents sample x i The number of true labels, l represents a true label currently being calculated, l' is a temporary variable used to traverse all labels that meet the above conditions, rank(x i ,l) for sample x i The predicted ranking of label l (sorted by prediction probability or score from high to low, ranking starts from 1).

[0136] The average accuracy of this example is calculated to be 81.23%, which is higher than some mainstream models, indicating that the present invention has practical significance.

[0137] The step S50 includes: based on the Spark distributed framework, performing feature fusion on the local optimal feature subsets selected by each node, so that the fused feature subsets become the global optimal subsets, such as Figure 2 The framework diagram shows the fusion of feature subsets of each node.

[0138] The present invention provides a neighborhood rough set attribute reduction method for ocular disease imaging diagnosis. There are many methods and approaches for implementing this technical solution. The above is only a preferred embodiment of the present invention. It should be noted that those skilled in the art may make various improvements and modifications without departing from the principles of the present invention, and such improvements and modifications are also within the scope of protection of the present invention. Any components not specified in this embodiment may be implemented using existing technologies.

Claims

1. A neighborhood rough set attribute reduction method for ocular disease imaging diagnosis, characterized in that: The following steps are involved: Step 1: read and process the eye disease image dataset; Step 2: Perform distributed processing on the eye disease image dataset and build a distributed framework; Step 3: Perform spectral clustering on the feature set and label set respectively to find the main label cluster and feature cluster; Step 4: Perform feature selection through neighborhood rough set selection to select the local optimal feature subset; Step 5: perform feature fusion on the local optimal feature subset to make the fused feature subset the global optimal feature subset.

2. The method according to claim 1, wherein Step 1 includes the following steps: Step 1-1: Read the eye disease image dataset, determine the feature set and decision classification, and establish a multi-label decision system MNDT=<U,C∪D,Δ,δ> , where U={x1,x2,...,x N } represents the set of detection objects in the eye disease image dataset, N is the number of detection objects, x N Indicates the Nth inspection object, C={a1,a2,...,a n } represents the feature set of data attributes, n is the number of sample features, a n is the nth eigenvalue of the sample, D={l1,l2,...,l d } represents the decision class of the sample, where d is the number of sample labels, l d represents the dth sample label, Δ represents the distance function, and δ represents the neighborhood radius; Step 1-2: Normalize the features in the eye disease data. The normalization formula is: Where x refers to the value of a feature in the original sample, x normalized Represents the value of the feature of the original sample after normalization, x max Refers to the maximum value of all samples in an attribute, x min It represents the minimum value of all samples in an attribute.

3. The method according to claim 2, wherein Step 2 includes: building a Spark distributed framework, dividing the eye disease image dataset S' into m eye disease information subsets S' = {S'1, S'2, S'3, ..., S' m }, where S' m Represents the mth eye disease information subset, and the m eye disease information subsets are distributed in parallel to the m working nodes using a random sending strategy.

4. The method according to claim 3, wherein Step 3 includes the following steps: Step 3-1: Perform spectral clustering on the feature set and label set in the eye disease information subset of each node to find the main label cluster and the secondary label cluster in each node. The main label cluster is the cluster with the largest number of labels, and the secondary label cluster is the cluster other than the main label cluster. Step 3-2, clustering the label spectrum: construct the similarity matrix W∈R d×d , where R represents the real space, and the element W in the i-th row and j-th column of the similarity matrix W is ij Expressed as: Where the intermediate parameter σ = 1.0, exp represents the natural exponential function, and i and j range from 1 to d. The distance formula is defined as: where l iN Represents sample X N In the label l i The value on l jN Denotes sample X N In the label l j The value on Step 3-3, construct the diagonal matrix Z; Step 3-4, let the matrix A = ZW, normalize the Laplace matrix to obtain the normalized Laplace matrix A'; Step 3-5, select the eigenvectors v1, v2, ..., v corresponding to the first T smallest non-zero eigenvalues T , T is the preset number of clusters, v T The eigenvector representing the Tth smallest non-zero eigenvalue; the eigenvectors v1,v2,...,v T A matrix of size d×T; Step 3-6, for the selected feature vectors v1, v2, ..., v T The matrix is clustered using K-means to obtain label clusters, which are sorted by the number of labels in the cluster. The cluster with the largest number of labels is the main label cluster, and the remaining clusters are sub-label clusters.

5. The method according to claim 4, wherein In step 3-3, the diagonal matrix Z is constructed using the following formula: Where Z(i,i) represents the sum of all elements in the i-th row of the similarity matrix W, which is called the degree matrix.

6. The method according to claim 5, wherein In steps 3-4, the normalized Laplace matrix A' is expressed as: A'=IZ -1 In (4), Where I is the identity matrix.

7. The method according to claim 6, wherein Step 4 includes the following steps: Step 4-1, for any sample x i ∈U and B∈C, sample x i Neighborhood δ under conditional attribute subset B B (x i )for: δ B (x i )={x j |x j ∈U,dis B (x i ,x j )≤δ} (5), Where δ is the neighborhood size, if the i-th sample x i With the jth sample x j If the relationship (5) is satisfied, then x j In x i In the neighborhood of B (x i ,x j ) is the i-th sample x i With the jth sample x j The Euclidean distance under the conditional attribute subset B is defined as: Among them, f(x i ,a t ) represents the i-th sample x i The value corresponding to the tth feature; Step 4-2, D j Represents a label l j The sample set, Indicates that in the label cluster L k The label set considered below, Indicates that in the label cluster L k Next sample x i The set of labels in the conditional feature subset Under this condition, the multi-label neighborhood rough set is approximated as: Among them, the lower approximation Describes the label cluster L k Next sample x i The neighborhood of completely belongs to the label cluster L k The sample set of Upper approximation Describes the label cluster L k Next sample x i The neighborhood of may belong to the label cluster L k The sample set, boundary area Indicates that it is not possible to determine whether it belongs to the label cluster L k The sample set of Step 4-3, under conditional feature subset B, conditional feature subset B for label cluster L k Dependence for: Step 4-4, for the multi-label decision system S=<U,C∪D,Δ,δ> , a∈CB, the importance of attribute a SIG(a,B,D) is defined as: SIG(a,B,D)=λ B∪{a} (D)-λ B (D)(11), where λ B∪{a} (D) represents the change in dependency caused by adding feature a under conditional feature subset B. If SIG(a,B,D)>0, it means that feature a is a useful feature for feature subset B. If SIG(a,B,D)<0, it means that feature a is a redundant feature for feature subset B. Step 4-5: For the feature clusters obtained by spectral clustering, calculate the dependency of each feature cluster on the label main cluster L1 and find the feature set F with the largest dependency. i , according to formula (11) remove the feature set F i The redundant features in the feature set F after removing the redundant features i becomes the selected feature, and the remaining feature clusters are combined into the candidate feature subset; Steps 4-6: For the label sub-cluster, the features in the candidate feature subset are added to the selected feature set in sequence. The dependency caused by the updated conditional feature subset is calculated according to formula (11), and the corresponding features with the maximum dependency are added to the selected feature set until the dependency no longer increases. The selected features are the feature subset corresponding to the label sub-cluster. In steps 4-7, for the two feature subsequences selected from the main label cluster and the secondary label cluster respectively, features are selected iteratively and progressively to generate the final local optimal feature subset.

8. The method according to claim 7, wherein Step 5 includes: in the Spark distributed framework, aggregating the optimal feature subsets of each working node through the reduce operation of the master node to form a global optimal feature subset.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores program codes, and when the program codes are executed by the processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 8.

10. A storage medium, characterized in that: A computer program or instruction is stored, and when the computer program or instruction is run on a computer, the steps of the method according to any one of claims 1 to 8 are executed.

Images