Graph data compression method, device, equipment and medium

Abstract

The application provides a graph data compression method, device, equipment and medium, which can be applied to the fields of artificial intelligence and cloud computing, and can be used in the financial field and other fields. The method comprises the following steps: determining at least one data point in to-be-compressed graph data; the data depth of the data point is greater than a depth threshold; determining a labeled data point and an unlabeled data point in the at least one data point; extracting a dimension reduction matrix corresponding to the at least one data point by using a category association relationship between the labeled data point and the unlabeled data point; and obtaining target graph data of the graph data based on matrix calculation of the graph data and the dimension reduction matrix. The method provided by the application improves the compression accuracy and efficiency of graph data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence and cloud computing, and particularly relates to a graph data compression method and device, equipment and medium. BACKGROUND

[0002] At present, graph data is generally composed of multiple data points. Generally, the data points have high data depth. Graph data has more and more wide application scenarios due to the advantages of clear structure and convenient data query. In actual application, for example, knowledge graph, user graph and hyperspectral image can all belong to graph data. The user graph, for example, can take a user identifier as a user node. The node information can be user information with high depth. The relationship between users can be an edge of the user node. The knowledge graph is similar to the user graph and is also generated based on nodes with certain data depth and the connection relationship between the nodes. The hyperspectral image is generally composed of multiple pixels. The waveband of the pixel is relatively large. For example, the ground image collected by a satellite generally has a resolution of 2000*2000 and is composed of 4 million pixels. Each pixel is composed of 500 wavebands, which can be understood as the depth of the pixel.

[0003] However, due to the complex structure of the data points and the high data depth, the data amount of the graph data composed of multiple data points is large, and a large amount of storage space is occupied, which is not conducive to data storage. Therefore, how to accurately compress the graph data is a technical problem to be solved at present. SUMMARY

[0004] The present application provides a graph data compression method, device, equipment and medium to solve the technical problem of low accuracy of network device graph data compression.

[0005] In a first aspect, the present application provides a graph data compression method, comprising:

[0006] determining at least one data point in the graph data to be compressed; the data depth of the data point is greater than a depth threshold;

[0007] determining a labeled data point and an unlabeled data point in the at least one data point;

[0008] extracting a dimension reduction matrix corresponding to the at least one data point by using the category association relationship between the labeled data point and the unlabeled data point;

[0009] obtaining target graph data of the graph data based on matrix calculation of the graph data and the dimension reduction matrix.

[0010] In a second aspect, the present application provides a graph data compression method and device, comprising:

[0011] determine at least one data point in the graph data to be compressed, wherein a data depth of the data point is greater than a depth threshold value;

[0012] determine a labeled data point and an unlabeled data point in the at least one data point;

[0013] extract a dimension reduction matrix corresponding to the at least one data point by using a data point association relationship between the labeled data point and the unlabeled data point;

[0014] obtain target graph data of the graph data based on matrix calculation of the graph data and the dimension reduction matrix.

[0015] In a third aspect, the present application provides an electronic device, comprising a processor and a memory connected with the processor in communication;

[0016] The memory stores computer execution instructions.

[0017] The processor executes the computer execution instructions stored in the memory to implement the graph data compression method according to the first aspect.

[0018] In a fourth aspect, the present application provides a computer readable storage medium, wherein the computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed by a processor to implement the graph data compression method according to the first aspect.

[0019] The graph data compression method provided by the present application can determine at least one data point in the graph data to be compressed. The data depth of the data point is greater than the depth threshold value. After determining the labeled data point and the unlabeled data point in the at least one data point, the dimension reduction matrix of the at least one data point can be extracted according to the data point association relationship between the labeled data point and the unlabeled data point. The at least one data point can be dimensionally calculated by using the dimension reduction matrix to obtain at least one target data point. The dimension reduction matrix can be used to perform dimension reduction calculation on the at least one data point, and accurate target data points can be obtained, thereby realizing effective dimension reduction of the at least one data point. The target graph data corresponding to the at least one target data point can achieve the purpose of compression, and the graph data can be quickly and accurately compressed. BRIEF DESCRIPTION OF DRAWINGS

[0020] The accompanying drawings, which are incorporated into and form part of the specification, illustrate embodiments consistent with the present application and, together with the specification, serve to explain the principles of the application.

[0021] Figure 1 is a system architecture diagram of a graph data compression method provided by an embodiment of the present disclosure;

[0022] Figure 2 is a flowchart of one embodiment of a graph data compression method provided by an embodiment of the present disclosure;

[0023] Figure 3 is a flowchart of another embodiment of a graph data compression method provided by an embodiment of the present disclosure;

[0024] Figure 4 is a flowchart of another embodiment of a graph data compression method provided by an embodiment of the present disclosure;

[0025] Figure 5 is a flowchart of another embodiment of a graph data compression method provided by an embodiment of the present disclosure;

[0026] Figure 6 is a structural schematic diagram of one embodiment of a graph data compression device provided by an embodiment of the present disclosure;

[0027] Figure 7 is a block diagram of an electronic device for implementing a graph data compression method provided by the present disclosure.

[0028] Through the above-described drawings, the explicit embodiments of the present application have been shown, and will be described in more detail hereinafter. These drawings and textual descriptions are not intended to limit the scope of the concept of the present application by any means, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0029] The exemplary embodiments will be described in detail herein with reference to the drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of apparatuses and methods consistent with some aspects of the present application as detailed in the appended claims.

[0030] The technical solution of the present disclosure can be applied to scenarios such as finance and agriculture that have a demand for graph data compression. By extracting the association relationship between at least one data point of the graph data, a dimension reduction matrix can be obtained, so as to reduce the dimension of each data point by using the dimension reduction matrix, thereby achieving compression of the graph data composed of at least one data point, which can effectively reduce the storage space of the graph data and is more conducive to the storage and analysis processing of the graph data.

[0031] In the related art, graph data is generally composed of multiple data points, for example, hyperspectral data and knowledge graph can be used as one kind of graph data. The difference between graph data and ordinary data is that the structure of data points is more complex and the data depth is higher. For example, the pixel points of a hyperspectral image, that is, the data points, can be composed of 500 bands. However, due to the high data depth of the data points, the graph data occupies a large storage space when stored. In order to reduce the data volume of the graph data and improve the data processing efficiency, a graph learning model can be used to reduce the dimension of the graph data. At present, principal component analysis method based on statistical theory, unsupervised method based on geometric theory, and linear discriminant method based on label information are generally used. These methods are all supervised methods, and from the dimension reduction effect, the supervised method can have high discrimination effect.

[0032] To solve the above technical problems, the technical scheme considers that for graph data, there can be certain association relationship between at least one data point in the graph data. If the same information between the data points is compressed by using the association relationship, the dimension of at least one data point can be reduced, and the compression purpose of the graph data can be achieved.

[0033] Accordingly, the technical scheme of the present disclosure can determine at least one data point in the graph data to be compressed. The data depth of the data point is greater than a depth threshold. After determining the labeled data points and the unlabeled data points in the at least one data point, the dimension reduction matrix of the at least one data point can be extracted according to the data point association relationship between the labeled data points and the unlabeled data points. The dimension reduction matrix can be used to perform dimension reduction calculation on the at least one data point, and at least one target data point can be obtained. The dimension reduction matrix can be used to perform dimension reduction calculation on the at least one data point respectively, and accurate target data points can be obtained, effective dimension reduction of the at least one data point can be realized, and the target graph data corresponding to the at least one target data point can achieve the compression purpose, and fast and accurate compression of the graph data can be realized.

[0034] The graph data compression method, device, equipment and storage medium provided by the present disclosure can be applied to the field of artificial intelligence and cloud computing. It can also be applied to any field other than the field of artificial intelligence and cloud computing, for example, it can be applied to the field of finance, remote sensing technology, etc. The specific application field of the graph data compression method, device, equipment and storage medium provided by the present disclosure is not limited too much.

[0035] The technical scheme of the present application and how the technical scheme of the present application solves the above technical problems will be described in detail in the following specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes can not be described in detail in some embodiments. The embodiments of the present application will be described below with reference to the drawings.

[0036] Figure 1 An application system architecture diagram of the graph data compression method provided by the embodiment of the present disclosure is shown in the figure. The graph data compression system can include: an electronic device 11 corresponding to a user (for example, a personal computer), a cloud server 12 in wired or wireless communication connection with the electronic device, and a database 13 for storing data. The electronic device 11 can include, for example, a server, an image acquisition device, a computer, a notebook, a remote sensing satellite, and other devices with data processing or acquisition functions.

[0037] The electronic device 11 can send the graph data to be compressed to the cloud server 12. The cloud server 12 can be configured with the graph data compression method of the present disclosure to compress the graph data. In the process, at least one data point in the graph data can be divided into labeled data points 121 and unlabeled data points 122 according to whether there is a label. The labeled data points 121 and the unlabeled data points 122 can be used to construct the same type of data point collaborative relationship 123, the different types of data point collaborative relationship 124, and the collaborative relationship 125 of the unlabeled data points. The semi-supervised collaborative expression graph 126 can be constructed by the same type of data point collaborative relationship 123 and the collaborative relationship 125 of the unlabeled data points. The inter-class collaborative expression graph 127 can be constructed by the different types of data point collaborative relationship 124. The semi-supervised collaborative expression graph 126 and the inter-class collaborative expression graph 127 are data point association relationships. The dimension reduction matrix can be extracted by the data point association relationship. That is, the semi-supervised collaborative expression embedding matrix 128 can be determined by the semi-supervised collaborative expression graph 126 and the inter-class collaborative expression graph 127. The semi-supervised collaborative expression embedding matrix 128 is the dimension reduction matrix. The graph data can be dimensionally reduced by the dimension reduction matrix to obtain the target graph data 129.

[0038] After obtaining the target graph data 129, the target graph data can be stored in the database 13. After the graph data is dimensionally reduced by the dimension reduction matrix, the graph data can be effectively dimensionally reduced, and the compression efficiency and accuracy of the graph data can be improved.

[0039] As shown in the flowchart of one embodiment of the graph data compression method provided by the embodiment of the present disclosure, the method can be configured as a graph data compression device, which can be located in an electronic device. The graph data compression method can include the following steps: Figure 2

[0040] 201: Determine at least one data point in the graph data to be compressed. The data depth of the data point is greater than the depth threshold.

[0041] ​The data depth of the data point can refer to the capacity, byte or waveband of the data, and the unit of the data depth can be determined according to the definition of the data. For example, the data depth of an RGB image is 24 bytes, of which 8 bytes are R, 8 bytes are G and 8 bytes are B. The pixel point of hyperspectral data can be 500 wavebands. The greater the data depth of the data point, the greater the data amount of the data point, and the greater the memory occupied when the data point is stored. The depth threshold can be set according to actual use requirements. For example, when the graph data includes a knowledge graph, the data point can include data stored in the form of a key-value pair. The data depth can include the number of key-value pairs or the field length of the key-value pairs.

[0042] 202: Determine the labeled data points and the unlabeled data points in the at least one data point.

[0043] The at least one data point can include labeled data points and unlabeled data points. In actual application, since the data dimension of the graph data is generally high, if the graph data is labeled, the labeling cost of the graph data is higher than that of ordinary image data. For example, the data dimension of hyperspectral image is generally 2000*2000, and different categories such as rivers, forests, ponds, fields, buildings and roads in the graph data need to be labeled. In order to improve the compression efficiency of the data, in the present disclosure, not all data points can be labeled, and part of the data points can be labeled.

[0044] The labeling process of the graph data can include: extracting key regions from the graph data, and setting a label type according to the corresponding type of the extracted key region. The data points in the graph data that are not marked as key regions can be used as unlabeled data points.

[0045] 203: Extract a dimension reduction matrix corresponding to the at least one data point by using a category association relationship between the labeled data points and the unlabeled data points.

[0046] The category association relationship can refer to a collaborative expression relationship between the labeled data points and the unlabeled data points. The labeled data points can belong to at least one label category, and can include at least one label data point corresponding to each label category. The data points in the same label category can use a vector to represent the association relationship, the data points between different label categories can also use a vector to represent the association relationship, and in order to accurately use the unlabeled data points, the labeled data points and the unlabeled data points can also use a vector to represent the association relationship. That is, the category association relationship can include: an association relationship between data points in the same label category, an association relationship between data points in different label categories, and an association relationship between labeled data points and unlabeled data points.

[0047] Optionally, by utilizing the class association relationship between labeled and unlabeled data points, extracting a dimensionality reduction matrix corresponding to at least one data point can include: obtaining a dimensionality reduction matrix that satisfies the data point compression condition using the class association relationship between labeled and unlabeled data points. The data point compression condition can include: minimizing the data distortion or data size resulting from compressing the data points using the dimensionality reduction matrix.

[0048] 204: Matrix calculation based on graph data and dimensionality reduction matrix to obtain the target graph data corresponding to the graph data.

[0049] Matrix computation based on graph data and a dimensionality reduction matrix can include: transposing the dimensionality reduction matrix to obtain the transpose matrix, and performing matrix computation on the transpose matrix and graph data to obtain the target graph data. The dimensionality reduction matrix can be represented by P, and the transpose matrix can then be represented by Ptranspose. T This indicates that, through the dimension reduction transpose matrix P T Perform matrix multiplication P with the graph data X T X will provide the target map data.

[0050] In this embodiment, at least one data point in the graph data to be compressed can be determined. The data depth of the data point is greater than a depth threshold. After determining the labeled and unlabeled data points among the at least one data point, a dimensionality reduction matrix for the at least one data point can be extracted based on the category association relationship between the labeled and unlabeled data points. The dimensionality reduction matrix can be used to perform dimensionality reduction calculations on the at least one data point to obtain at least one target data point. By using the dimensionality reduction matrix to perform dimensionality reduction calculations on each of the at least one data point, accurate target data points can be obtained, achieving effective dimensionality reduction of the at least one data point. The target graph data corresponding to the obtained at least one target data point can achieve the purpose of compression, realizing fast and accurate compression of graph data.

[0051] As an example, the labeled data points may include at least one labeled data point corresponding to each of the labeled categories.

[0052] like Figure 3 As shown, with Figure 2 The difference in the steps of the illustrated embodiment is that step 203: extracting a dimensionality reduction matrix corresponding to at least one data point using the category association relationship between labeled and unlabeled data points may include:

[0053] 301: Using unlabeled data points as the same category, and combining them with the labeled data points corresponding to at least one labeled category, the association between labeled data points of the same category is extracted to obtain intra-class collaborative information.

[0054] The labeled data points of the same category can be associated with each other, and the unlabeled data points can be associated with the labeled data points of the same category.

[0055] The association relationship between the labeled data points of the same category can include position calibration of the labeled data points of the same category. The whole unlabeled data points can also be associated with the labeled data points of the same category as an unlabeled category.

[0056] The intra-class collaborative information can include a reconstruction relationship of collaborative expression of the data points of the same category and a reconstruction relationship of collaborative expression of the unlabeled data points.

[0057] 302: Extracting the association relationship between the labeled data points of different categories according to the labeled data points corresponding to at least one label category, respectively, to obtain inter-class collaborative information.

[0058] The association relationship between the labeled data points of different categories can include position calibration of the labeled data points of different categories. The inter-class collaborative information can include a reconstruction matrix of collaborative expression of the data points of different categories.

[0059] 303: Constructing a semi-supervised collaborative embedding equation based on the intra-class collaborative information and the inter-class collaborative information.

[0060] Optionally, the intra-class collaborative information can use (W s ) i,j The inter-class collaborative information can use (W d ) i,j Wherein, i represents 1-N, j represents 1-N, and N is the number of at least one data point.

[0061] The semi-supervised collaborative embedding equation can be represented as:

[0062] Wherein, P can be a dimension reduction matrix to be solved, and the dimension reduction matrix is obtained by taking the minimum value of the semi-supervised collaborative embedding equation as the optimization target.

[0063] 304: In response to a solving instruction for the semi-supervised collaborative embedding equation, solving the semi-supervised collaborative embedding equation to obtain a dimension reduction matrix.

[0064] Optionally, solving the semi-supervised collaborative embedding equation to obtain the dimension reduction matrix can include: using a generalized eigenvalue algorithm to solve the semi-supervised collaborative embedding equation to obtain the dimension reduction matrix.

[0065] In this embodiment of the disclosure, intra-class collaborative information and inter-class collaborative information can be used to extract intra-class relationships between data points of the same class and inter-class relationships between data points of different classes in at least one data point, thereby achieving accurate feature representation of at least one data point in the same and different categories, achieving accurate construction of semi-supervised collaborative embedding equations, and by solving the equations, calculating the dimensionality reduction matrix, thereby achieving accurate extraction of the dimensionality reduction matrix.

[0066] In one possible design, such as Figure 4 The diagram shown is a flowchart of another embodiment of the graph data compression method provided in this disclosure. It includes extracting the association relationships between label data points of the same category by using unlabeled data points as the same category and combining them with label data points corresponding to at least one label category, thereby obtaining intra-category collaborative information. This may include:

[0067] 401: Based on the label data points corresponding to at least one label category, extract the same feature relationships of the data points in the same category of labels to obtain the first reconstruction coefficient vector corresponding to at least one data point.

[0068] The first reconstruction coefficient vector can be used to impose similar constraints on data points of the same label category, and can be obtained by extracting the same feature relationships of data points in the same label category.

[0069] Optionally, the same feature relationship between data points in the same category label can be used to optimize the solution of the reconstruction coefficient vector corresponding to at least one data point to obtain the first reconstruction coefficient vector.

[0070] The first reconstruction coefficient vector is passed through When representing the first reconstructed coefficient vector, a first coefficient equation can be constructed, and the first reconstructed coefficient vector can be solved using this equation. The first coefficient equation can be determined using the 1-norm and 2-norm equidistant calculation formulas.

[0071] The first reconstruction coefficient vector can be calculated with the representation vector of the data point to calculate the transformation vector of the graph data at the corresponding data point. The distance between the data point and the transformation vector is calculated using the 1-norm or 2-norm, and the value of the first reconstruction coefficient vector is found when the distance is minimized.

[0072] For ease of understanding, taking the L2 norm as an example, the first reconstruction coefficient vector uses... When representing the data points, the constraint equation for solving the distance between the corresponding L2 norm points can be expressed as:

[0073]

[0074] By reconstructing the first coefficient vector under the constraint of minimum distance By solving this problem, we can obtain the formula for calculating the first reconstruction coefficient vector:

[0075] Among them, X s =XI s , This can be a similar representation vector established for the data points, where i can be the coordinate value of the data point. The method for obtaining this similar representation vector can be referred to the description in the following embodiments. X can be graph data, consisting of at least one data point, and β is a constant, which can be obtained through training or set based on historical experience.

[0076] 402: Treat unlabeled data points as unlabeled categories, extract the same feature relationships of data points in unlabeled categories, and obtain at least one second reconstruction coefficient vector corresponding to each data point.

[0077] The second reconstruction coefficient vector can be used to impose class constraints on unlabeled categories, and can be obtained by extracting the label association features of data points in unlabeled categories.

[0078] The same feature relationships among data points in unlabeled categories can be used to solve for the second reconstruction coefficient vector of the data points. The second reconstruction coefficient vector is obtained through... When representing this, a second coefficient variance can be constructed, and the second reconstructed coefficient vector can be obtained by solving the equation. The second coefficient variance can be determined using the equidistant formulas for the 1-norm and 2-norm.

[0079] The second reconstruction coefficient vector can be calculated with the representation vector of the data point to calculate the transformation vector of the graph data at the corresponding data point. By using the 1-norm or 2-norm, the distance between the data point and the corresponding transformation vector is calculated, and the value of the second reconstruction coefficient vector is solved when the distance is minimized.

[0080] To facilitate understanding, let's take the L2 norm as an example. The second reconstruction coefficient vector is obtained through... When representing the data points, the constraint equation for solving the distance between the corresponding L2 norm points can be expressed as:

[0081]

[0082] By solving for the second reconstruction coefficient vector under the constraint of minimum distance, the formula for calculating the second reconstruction coefficient vector can be obtained:

[0083] Among them, X u =XI u , This can be an unlabeled vector created for the data points. `i` can be the coordinate value of the data point. The method for obtaining the unlabeled vector can be referred to the description in the following embodiments. `X` can be graph data, consisting of at least one data point, and `γ` is a constant, which can be obtained through training or set based on historical experience.

[0084] 403: Weight the first and second reconstruction coefficient vectors corresponding to the data points to obtain at least one semi-supervised weight vector corresponding to each data point.

[0085] The first reconstruction coefficient vector corresponding to the data points Second reconstruction coefficient vector Weighting is performed to obtain the semi-supervised weight vector (W) corresponding to the data point. s ) i .

[0086] First reconstruction coefficient vector The length of the vector is equal to the number of at least one data point. Therefore, when weighting the first and second reconstruction coefficient vectors, the values ​​corresponding to each coordinate point in the vector can be added together, that is... Where j is the coordinate point in the vector.

[0087] 404: Determine intra-class collaboration information based on the semi-supervised weight vectors corresponding to at least one data point; the intra-class collaboration information is used to construct the directed reconstruction graph corresponding to at least one data point.

[0088] The semi-supervised weight vectors corresponding to at least one data point can constitute the intra-class collaborative information W. s .

[0089] A directed reconstruction graph can use at least one data point as a node and establish node relationships using intra-class collaboration information.

[0090] In this embodiment, by treating unlabeled data points as the same category and combining them with labeled data points corresponding to at least one labeled category, the same feature relationships between data points within the same category can be extracted, resulting in at least one first reconstruction coefficient vector corresponding to each data point. The first reconstruction coefficient vector can better express the relationships between data points of the same category. Furthermore, unlabeled data points can also be treated as unlabeled categories, and the same feature relationships between data points within these unlabeled categories can be extracted, resulting in at least one second reconstruction coefficient vector corresponding to each data point. The second reconstruction vector can express the relationships between unlabeled categories. By using data points as the weighting unit and weighting the first and second reconstruction coefficient vectors, it can be ensured that the obtained semi-supervised weight vector can accurately record the relationships between data points of the same and different categories, making the intra-class collaborative information more comprehensive and accurate.

[0091] As an optional implementation, step 401 may include: extracting the same feature relationships of data points within the same category of labels based on the label data points corresponding to at least one label category, to obtain a first reconstruction coefficient vector corresponding to at least one data point, including:

[0092] Based on the label data points corresponding to at least one label category, the relationship between the same characteristics of the label data points of the same label category is extracted to obtain the same type of representation vector corresponding to at least one data point.

[0093] Based on the similar representation vectors of the data points, the first reconstruction coefficient vector of the data points is calculated to determine the first reconstruction coefficient vector corresponding to at least one data point.

[0094] In this embodiment, relationships can be extracted based on the label data points corresponding to at least one label category to identify common characteristics among label data points of the same category, thereby obtaining a similarity representation vector corresponding to at least one data point. The similarity representation vector accurately defines whether data points belong to the same category at at least one data point, and the first reconstruction coefficient vector of the data points can be accurately determined through this vector. This ensures that the first reconstruction coefficient vector contains the similar characteristics of the data points, guaranteeing accurate calculation of similar data points.

[0095] In one possible design, the homogeneous representation vector of any data point is determined in the following way:

[0096] For the first data point among at least one data point, use the number of at least one data point as the vector length to establish an empty first representation vector for the first data point.

[0097] Iterate through at least one data point and determine whether the data point belongs to the same category as the first data point. If it does, determine that the data point takes the value of 1 at the coordinates corresponding to the first representation vector. If not, determine that the data point takes the value of 0 at the coordinates corresponding to the first representation vector.

[0098] The first representation vector whose coordinate values ​​are determined at the end of the traversal is a representation vector of the same type.

[0099] To accurately explain the method of obtaining similar representation vectors, as shown in Table 1, assume there are 6 data points. These 6 data points are each taken as the first data point, and the values ​​corresponding to each first data point and at least one other first data point are determined. Assume 1 and 2 are of the same label category, 3 and 6 are unlabeled categories, and 4 and 5 are of the same label category. Table 1:

[0100] 1 2 3 4 5 6 1 0 1 0 0 0 0 2 1 0 0 0 0 0 3 0 0 0 0 0 1 4 0 0 0 0 1 0 5 0 0 0 1 0 0 6 0 0 1 0 0 0

[0101] In the above, the first representation vector of data point 1 can be [0,1,0,0,0,0]. The second representation vector of data point 2 can be [1,0,0,0,0,0], the first representation vector of data point 3 can be [0,0,0,0,0,1], the first representation vector of data point 4 can be [0,0,0,0,1,0], the first representation vector of data point 5 can be [0,0,0,1,0,0], and the first representation vector of data point 6 can be [0,0,1,0,0,0]. Each data point can have a corresponding first representation vector.

[0102] In this embodiment, when determining the similarity representation vector of data points, a first representation vector for the first data point can be established using the quantity of at least one data point as the vector length. This first representation vector is an empty vector, and the values ​​of each coordinate point need to be determined. By traversing at least one data point, it can be determined whether a data point belongs to the same label category as the first data point. Coordinates belonging to the same label category as the first data point are set to 1, and coordinates belonging to different label categories are set to 0. By setting labels for the same label category, the category relationship between the first data point and at least one data point can be accurately extracted. The obtained similarity representation vector can accurately record data points belonging to the same label category as the first data point, achieving accurate similarity representation.

[0103] As another optional implementation, step 402 may include: taking unlabeled data points as unlabeled categories, extracting the same feature relationships of data points in the unlabeled categories, and obtaining a second reconstruction coefficient vector corresponding to at least one data point, including:

[0104] Treating unlabeled data points as the same category, extract the unlabeled features corresponding to the data points to obtain at least one unlabeled vector corresponding to each data point.

[0105] Based on the unlabeled vector corresponding to the data point, calculate the second reconstruction coefficient vector of the data point to obtain the second reconstruction coefficient vector corresponding to at least one data point.

[0106] In this embodiment, unlabeled data points are treated as the same category, and the association between each data point and the unlabeled data points is accurately extracted to obtain at least one second reconstruction coefficient corresponding to each data point. The second reconstruction coefficient vector of the data points can be accurately extracted using the unlabeled vectors of the data points, allowing the calculation of the second reconstruction coefficient vector with the unlabeled vectors of each data point. This comprehensively considers the effective association between unlabeled data points, improving the efficiency and accuracy of obtaining the second reconstruction coefficient vector.

[0107] In one possible design, the unlabeled vector of any data point is determined in the following way:

[0108] For the second data point in at least one data point, use the number of at least one data point as the vector length to create an empty label vector for the second data point;

[0109] Iterate through at least one data point and determine whether the data point and the second data point are both unlabeled data points. If they are, determine that the data point has a value of 1 at the coordinates corresponding to the label vector. If not, determine that the data point has a value of 0 at the coordinates corresponding to the label vector.

[0110] At the end of the traversal, the label vector with all coordinates determined is the unlabeled vector.

[0111] To accurately explain the method of obtaining similar representation vectors, as shown in Table 2, assuming there are 6 data points, each of these 6 data points is taken as the first data point, and the values ​​corresponding to each first data point and at least one first data point are determined. Assume 1 and 2 are of the same label category, 3 and 6 are unlabeled categories, and 4 and 5 are of the same label category. Table 2:

[0112] 1 2 3 4 5 6 1 0 0 0 0 0 0 2 0 0 0 0 0 0 3 0 0 0 0 0 1 4 0 0 0 0 0 0 5 0 0 0 0 0 0 6 0 0 1 0 0 0

[0113] In addition to the label vectors of data point 3 ([0,0,0,0,0,1]) and data point 6 ([0,0,1,0,0,0]), the label vectors of all other data points can be [0,0,0,0,0,0].

[0114] In this embodiment, when determining the unlabeled vector of a data point, a label vector for the second data point can be established using the quantity of at least one data point as the vector length. The label vector is an empty vector, and the values ​​of each coordinate point of the label vector need to be determined. By traversing at least one data point, it can be determined whether the data point and the second data point both belong to the unlabeled category. The coordinates of data points that belong to the same unlabeled category as the second data point are set to 1, and the coordinates of data points that do not belong to the same unlabeled category as the second data point are set to 0. By setting the unlabeled data category, the unlabeled association between each data point and the corresponding unlabeled data point can be accurately extracted. The obtained unlabeled vector can accurately record data points that belong to the same unlabeled category as the second data point, achieving accurate representation of the unlabeled vector.

[0115] As yet another example, such as Figure 5 The diagram shown is a flowchart of another embodiment of the graph data compression method provided in this disclosure. The step of extracting the association relationships between tag data points of different categories based on tag data points corresponding to at least one tag category to obtain inter-class collaboration information may include:

[0116] 501: Based on the label data points corresponding to at least one label category, extract the relationship between the same characteristics of the data points in different label categories to obtain the third reconstruction coefficient vector corresponding to at least one data point.

[0117] 502: Based on the third reconstruction coefficient vector corresponding to at least one data point, the collaborative relationship between data points of different categories is extracted to obtain inter-class collaborative information.

[0118] Inter-class collaboration information may include inter-class weight matrix (W) d ) i,j .

[0119] The third reconstructed coefficient vector can be obtained by extracting relationships between data points based on the same characteristics of different category labels.

[0120] Extracting collaborative relationships between different data points can refer to using the third reconstruction coefficient vector corresponding to at least one data point to reconstruct the relationships between labeled data points of different classes, so that the difference between the reconstructed data and the data points is small, ensuring dimensionality reduction while maintaining the authenticity of the data.

[0121] For ease of understanding, When representing the third reconstruction coefficient vector, the vector extraction problem can be transformed into a distance calculation problem. The distance between data points and the data points after being transformed by the third reconstruction coefficient vector can be calculated, and minimizing this distance is the optimization objective. The distance can include norm 1 distance, norm 2 distance, etc.

[0122] Taking the 2-norm distance as an example, the third reconstruction coefficient vector between different categories can be solved by the following optimization problem:

[0123]

[0124] Among them, X d =XI d , This is the inter-class representation vector. Inter-class collaborative information can be determined through the third reconstruction coefficient vector corresponding to at least one data point.

[0125] In this embodiment, relationships can be extracted based on the label data points corresponding to at least one label category, considering the shared characteristics of data points across different label categories. This yields a third reconstruction vector corresponding to each data point. The third reconstruction vector characterizes the association between data points across different label categories. By using the third reconstruction coefficient vector corresponding to each of the at least one data point, collaborative relationships between data points of different categories can be extracted, obtaining inter-class collaborative information. This inter-class collaborative information, represented by the third reconstruction coefficient vector corresponding to each of the at least one data point, enables accurate extraction of inter-class relationships and improves the effectiveness of inter-class collaborative expression.

[0126] In one possible design, based on the label data points corresponding to at least one label category, the relationship between the common characteristics of the data points in different label categories is extracted to obtain a third reconstruction coefficient vector corresponding to at least one data point, including:

[0127] For each label data point corresponding to at least one label category, based on label data points of different categories, determine the inter-class representation vector corresponding to at least one data point.

[0128] Based on the inter-class representation vectors of the data points, the third reconstruction coefficient vector of the data points is calculated to obtain the third reconstruction coefficient vector corresponding to at least one data point.

[0129] In this embodiment of the disclosure, the association characteristics of at least one data point in different categories can be extracted using label data points of different categories to obtain inter-class representation vectors corresponding to at least one data point. Using the inter-class representation vectors, the third reconstruction coefficient vector of the data point can be calculated, enabling accurate calculation of the third reconstruction coefficient vector of at least one data point.

[0130] Optionally, based on the inter-class representation vectors of the data points, a third reconstruction coefficient vector for the data points is calculated, including:

[0131] Based on the inter-class representation vectors of the data points, construct the inter-class cooperative equations for the data points.

[0132] In response to a request to solve the inter-class coordination equation, the third reconstruction coefficient vector of the data points is calculated to obtain the third reconstruction coefficient vector corresponding to at least one data point.

[0133] The third reconstruction coefficient vector can be accurately solved by constructing inter-class coordination equations.

[0134] In some embodiments, based on the third reconstruction coefficient vector corresponding to at least one data point, the collaborative relationships between samples of different categories are extracted to obtain inter-class collaborative information, including:

[0135] Based on the third reconstruction coefficient vector corresponding to at least one data point, the vectors are combined according to the position of at least one data point to obtain the inter-class reconstruction matrix.

[0136] For any given data point, determine the transpose of the target position in the inter-class reconstruction matrix based on the target position of the data point in the corresponding third reconstruction coefficient vector.

[0137] Determine the first value corresponding to the target position and the second value of the transpose position from the inter-class reconstruction matrix.

[0138] Select the maximum value from the first and second values ​​as the target value for the target position, so as to obtain the target value corresponding to each position in the inter-class reconstruction matrix.

[0139] The inter-class reconstruction matrix is ​​determined by replacing the corresponding target value at each position with the inter-class collaborative information.

[0140] The target position (i,j) of the data point in the third reconstruction coefficient vector can have the following first value: The transpose can be at position (j, i), and the second value at the transpose can be...

[0141] Selecting the maximum value from the first and second values ​​as the target value for the target location can include selecting the maximum value from the absolute values ​​of the first and second values. The target value for the target location can be expressed using the formula:

[0142] In this embodiment of the disclosure, by determining the class reconstruction matrix corresponding to at least one data point in the third reconstruction coefficient vector, data extraction can be performed on the values ​​of the same data point at the target position and the transpose position corresponding to the target position, thereby obtaining the inter-class relationship data between the data point and other data points. By selecting the maximum value of the inter-class relationship through the first value of the target position and the second value of the device position, it can be ensured that the target value of the inter-class reconstruction matrix at the same position is maximized, ensuring that the inter-class reconstruction matrix can represent the differences between different classes of data points to the maximum extent, and improving the accuracy of inter-class collaborative information.

[0143] In one possible design, the inter-class representation vector of any data point is determined through the following steps:

[0144] For the third data point in at least one data point, the number of at least one data point is used as the vector length to establish an empty second representation vector for the third data point;

[0145] Iterate through at least one data point and determine whether the data point and the third data point belong to the same or different categories. If they do, determine that the data point takes the value of 1 at the coordinates corresponding to the second representation vector. If not, determine that the data point takes the value of 0 at the coordinates corresponding to the second representation vector.

[0146] The second representation vector, whose coordinate values ​​are determined at the end of the traversal, is the inter-class representation vector.

[0147] To accurately explain the method of obtaining similar representation vectors, as shown in Table 1, assume there are 6 data points. These 6 data points are each taken as the first data point, and the values ​​corresponding to each first data point and at least one other first data point are determined. Assume 1 and 2 are of the same label category, 3 and 6 are unlabeled categories, and 4 and 5 are of the same label category. Table 1:

[0148] 1 2 3 4 5 6 1 0 0 1 1 1 1 2 0 0 1 1 1 1 3 1 1 0 1 1 0 4 1 1 1 0 0 1 5 1 1 1 0 0 0 6 1 1 0 1 1 0

[0149] The second representation vector for data point 1 can be [0,0,1,1,1,1]. The second representation vector for data point 2 can be [0,0,1,1,1,1], the second representation vector for data point 3 can be [1,1,0,1,1,0], the second representation vector for data point 4 can be [1,1,1,0,0,1], the second representation vector for data point 5 can be [1,1,1,0,0,0], and the second representation vector for data point 6 can be [1,1,0,1,1,0]. Each data point can have a corresponding second representation vector.

[0150] In this embodiment, when determining the inter-class representation vector of a data point, the quantity of at least one data point can be used as the vector length to establish a second representation vector for the third data point. This second representation vector is an empty vector, and the values ​​of each coordinate point need to be determined. By traversing at least one data point, it can be determined whether a data point belongs to a different category than the third data point. The coordinates of data points belonging to different categories from the third data point are set to 1, and the coordinates of data points belonging to the same category as the third data point are set to 0. By setting different label categories, the association between each data point and data points of different categories can be accurately extracted.

[0151] To accurately construct the cooperative embedding equation, one possible design involves constructing a semi-supervised cooperative embedding equation based on intra-class and inter-class cooperative information. This could include:

[0152] A semi-supervised collaborative embedding equation is generated using at least one data point, intra-class collaborative information, and inter-class collaborative information as known quantities, and a dimension-reduced matrix as an unknown quantity.

[0153] Optionally, the vectors corresponding to the data points and the corresponding intra-class collaborative information, and the vectors corresponding to the inter-class collaborative information can be used as known quantities, and the dimensionality reduction matrix can be used as unknown quantities to generate a semi-supervised collaborative embedding equation.

[0154] The semi-supervised cooperative embedding equation can be expressed as:

[0155]

[0156] The dimensionality reduction matrix P can be obtained by solving the semi-supervised cooperative embedding equation while achieving the minimum value.

[0157] In this embodiment of the disclosure, a semi-supervised collaborative embedding equation can be constructed using at least one data point, intra-class collaborative information, and inter-class collaborative information as known quantities. The semi-supervised collaborative embedding equation uses a dimension reduction matrix as an unknown quantity, which can achieve accurate construction of the semi-supervised collaborative embedding equation.

[0158] The technical solution disclosed herein can be applied to a variety of technical fields, especially agricultural remote sensing images, hyperspectral images, and maps.

[0159] Therefore, in some embodiments, graph data may include a graph, where multiple data points comprise multiple nodes, including labeled and unlabeled nodes. Two nodes in the graph are connected to form an edge. Nodes may include corresponding node data, which has a high data depth.

[0160] In this embodiment of the disclosure, using a graph as graph data can achieve data compression for various graphs.

[0161] In some embodiments, the graph data includes a hyperspectral image, and the hyperspectral image includes multiple data points including multiple pixels, which include labeled pixels and unlabeled pixels.

[0162] In this embodiment of the disclosure, hyperspectral images are used as image data to achieve data compression of hyperspectral images.

[0163] like Figure 6 The diagram shown is a structural schematic of one embodiment of a graph data compression device provided in this disclosure. This graph data compression device can be configured with the graph data compression method provided in the foregoing embodiments. The graph data compression device can be located in a second electronic device. The graph data compression device 600 may include the following units:

[0164] The data determination unit 601 is used to determine at least one data point in the graph data to be compressed; the data point has a data depth greater than a depth threshold.

[0165] Label segmentation unit 602: used to determine labeled and unlabeled data points in at least one data point.

[0166] Matrix extraction unit 603: used to extract a dimension reduction matrix corresponding to at least one data point by utilizing the category association relationship between labeled and unlabeled data points.

[0167] Target determination unit 604: Used for matrix calculation based on graph data and dimensionality reduction matrix to obtain target graph data from the graph data.

[0168] As an example, the labeled data points include: at least one label data point corresponding to each label category;

[0169] The matrix extraction unit includes:

[0170] The association extraction module is used to extract the association relationship between label data points of the same category by taking unlabeled data points as the same category and combining them with the label data points corresponding to at least one label category, thereby obtaining intra-category collaborative information.

[0171] The information extraction module is used to extract the correlation between label data points of different categories based on the label data points corresponding to at least one label category, and obtain inter-class collaborative information.

[0172] The equation construction module is used to construct semi-supervised collaborative embedding equations based on intra-class and inter-class collaborative information.

[0173] The instruction response module is used to respond to the solution instruction for the semi-supervised co-embedding equation, solve the semi-supervised co-embedding equation, and obtain the dimensionality reduction matrix.

[0174] In some embodiments, the association extraction module includes:

[0175] The first acquisition submodule is used to extract the same feature relationship of data points in the same category of labels based on the label data points corresponding to at least one label category, and obtain the first reconstruction coefficient vector corresponding to at least one data point.

[0176] The second acquisition submodule is used to treat unlabeled data points as unlabeled categories, extract the same feature relationships of data points in unlabeled categories, and obtain at least one second reconstruction coefficient vector corresponding to each data point.

[0177] The vector weighting submodule is used to weight the first reconstruction coefficient vector and the second reconstruction coefficient vector corresponding to the data point to obtain a semi-supervised weight vector corresponding to at least one data point.

[0178] The collaborative construction submodule is used to determine intra-class collaborative information based on the semi-supervised weight vectors corresponding to at least one data point; the intra-class collaborative information is used to construct the directed reconstruction graph corresponding to at least one data point.

[0179] In some embodiments, the first obtaining submodule is specifically used for:

[0180] Based on the tag data points corresponding to at least one tag category, the relationship between the same characteristics of the tag data points of the same tag category is extracted to obtain the same type representation vector corresponding to at least one data point; based on the same type representation vector of the data points, the first reconstruction coefficient vector of the data points is calculated to determine the first reconstruction coefficient vector corresponding to at least one data point.

[0181] As yet another embodiment, the first obtaining submodule is further configured to:

[0182] For the first data point in at least one data point, use the number of at least one data point as the vector length to establish an empty first representation vector for the first data point;

[0183] Iterate through at least one data point and determine whether the data point belongs to the same category as the first data point. If it does, determine that the data point takes the value of 1 at the coordinates corresponding to the first representation vector. If not, determine that the data point takes the value of 0 at the coordinates corresponding to the first representation vector.

[0184] The first representation vector whose coordinate values ​​are determined at the end of the traversal is a representation vector of the same type.

[0185] In some embodiments, the second obtaining submodule is specifically used for:

[0186] Treating unlabeled data points as the same category, extract the unlabeled features corresponding to the data points to obtain at least one unlabeled vector corresponding to each data point.

[0187] Based on the unlabeled vector corresponding to the data point, calculate the second reconstruction coefficient vector of the data point to obtain the second reconstruction coefficient vector corresponding to at least one data point.

[0188] As one embodiment, the second obtaining submodule is further used for:

[0189] For the second data point in at least one data point, use the number of at least one data point as the vector length to create an empty label vector for the second data point;

[0190] Iterate through at least one data point and determine whether the data point and the second data point are both unlabeled data points. If they are, determine that the data point has a value of 1 at the coordinates corresponding to the label vector. If not, determine that the data point has a value of 0 at the coordinates corresponding to the label vector.

[0191] At the end of the traversal, the label vector with all coordinates determined is the unlabeled vector.

[0192] As one embodiment, the information extraction module may include:

[0193] The third submodule is used to extract the relationship between the same characteristics of the data points in different categories of labels based on the label data points corresponding to at least one label category, and obtain the third reconstruction coefficient vector corresponding to at least one data point.

[0194] The inter-class acquisition submodule is used to extract the collaborative relationship between samples of different categories based on the third reconstruction coefficient vector corresponding to at least one data point, and obtain inter-class collaborative information.

[0195] In some embodiments, the third obtaining submodule may specifically be used for:

[0196] For each label data point corresponding to at least one label category, based on label data points of different categories, determine the inter-class representation vector corresponding to at least one data point.

[0197] Based on the inter-class representation vectors of the data points, the third reconstruction coefficient vector of the data points is calculated to obtain the third reconstruction coefficient vector corresponding to at least one data point.

[0198] As an optional implementation, the third obtaining submodule is further used for:

[0199] For the third data point in at least one data point, the number of at least one data point is used as the vector length to establish an empty second representation vector for the third data point;

[0200] Iterate through at least one data point and determine whether the data point and the third data point belong to the same or different categories. If they do, determine that the data point takes the value of 1 at the coordinates corresponding to the second representation vector. If not, determine that the data point takes the value of 0 at the coordinates corresponding to the second representation vector.

[0201] The second representation vector, whose coordinate values ​​are determined at the end of the traversal, is the inter-class representation vector.

[0202] In some embodiments, submodules are obtained between classes, specifically for:

[0203] Based on the third reconstruction coefficient vector corresponding to at least one data point, the vectors are combined according to the position of at least one data point to obtain the inter-class reconstruction matrix.

[0204] For any data point, determine the transpose of the target position in the inter-class reconstruction matrix based on the target position of the data point in the corresponding third reconstruction coefficient vector;

[0205] Determine the first value corresponding to the target position and the second value of the transpose position from the inter-class reconstruction matrix;

[0206] Select the maximum value from the first value and the second value as the target value for the target position, so as to obtain the target value corresponding to each position in the inter-class reconstruction matrix;

[0207] The inter-class reconstruction matrix is ​​determined by replacing the corresponding target value at each position with the inter-class collaborative information.

[0208] In some embodiments, the equation construction module includes:

[0209] The equation construction submodule is used to generate semi-supervised collaborative embedding equations using at least one data point, intra-class collaborative information, and inter-class collaborative information as known quantities and a dimension-reduced matrix as unknown quantity.

[0210] As an alternative implementation, the graph data includes a graph, and the multiple data points of the graph include multiple nodes, including labeled nodes and unlabeled nodes.

[0211] As another alternative implementation, the image data includes a hyperspectral image, and the multiple data points of the hyperspectral image include multiple pixels, which include labeled pixels and unlabeled pixels.

[0212] The graph data compression apparatus disclosed herein can be used to perform graph data compression methods. For details on the specific execution methods and technical effects of each step, please refer to the embodiments related to graph data compression methods, which will not be repeated here.

[0213] The aforementioned electronic devices can be either wireless or wired terminals. A wireless terminal can be a device that provides voice and / or other service data connectivity to a user, a handheld device with wireless connectivity, or other processing devices connected to a wireless modem. The wireless terminal can communicate with one or more core network devices via a Radio Access Network (RAN). The wireless terminal can be a mobile terminal, such as a mobile phone (or "cellular" phone) or a computer with a mobile terminal, for example, a portable, pocket-sized, handheld, computer-embedded, or vehicle-mounted mobile device, which exchanges voice and / or data with the RAN. Furthermore, a wireless terminal can also be a Personal Communication Service (PCS) phone, a cordless phone, a Session Initiation Protocol (SIP) phone, a Wireless Local Loop (WLL) station, a Personal Digital Assistant (PDA), or other similar devices. Wireless terminals can also be referred to as systems, subscriber units, subscriber stations, mobile stations, mobile stations, remote stations, remote terminals, access terminals, user terminals, user agents, user devices, or user equipment, without limitation. Optionally, the aforementioned electronic devices can also be smartwatches, tablets, etc.

[0214] Figure 7 This is a block diagram illustrating an electronic device according to an exemplary embodiment. The device may be a mobile phone, computer, digital broadcasting terminal, messaging device, game console, tablet device, medical device, fitness device, personal digital assistant, etc. The electronic device 700 may include one or more of the following components: processing component 702, memory 704, power supply component 706, multimedia component 708, audio component 710, input / output (I / O) interface 712, sensor component 714, and communication component 716.

[0215] Processing component 702 typically controls the overall operation of device 700, such as operations associated with display, telephone calls, data communication, camera operation, and recording. Processing component 702 may include one or more processors 720 to execute instructions to complete all or part of the steps of the graph data compression model training method or graph data compression method described above. Furthermore, processing component 702 may include one or more modules to facilitate interaction between processing component 702 and other components. For example, processing component 702 may include a multimedia module to facilitate interaction between multimedia component 708 and processing component 702.

[0216] Memory 704 is configured to store various types of data to support the operation of device 700. Examples of this data include instructions for any application or method operating on device 700, contact data, phonebook data, messages, pictures, videos, etc. Memory 704 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0217] Power supply assembly 706 provides power to various components of device 700. Power supply assembly 706 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to device 700.

[0218] Multimedia component 708 includes a screen that provides an output interface between device 700 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors may sense not only the boundaries of touch or swipe actions but also the duration and pressure associated with the touch or swipe operation. In some embodiments, multimedia component 708 includes a front-facing camera and / or a rear-facing camera. When device 700 is in an operating mode, such as a shooting mode or a video mode, the front-facing camera and / or rear-facing camera may receive external multimedia data. Each front-facing camera and rear-facing camera may be a fixed optical lens system or have focal length and optical zoom capabilities.

[0219] Audio component 710 is configured to output and / or input audio signals. For example, audio component 710 includes a microphone (MIC) configured to receive external audio signals when device 700 is in an operating mode, such as call mode, recording mode, and voice recognition mode. The received audio signals may be further stored in memory 704 or transmitted via communication component 716. In some embodiments, audio component 710 also includes a speaker for outputting audio signals.

[0220] I / O interface 712 provides an interface between processing component 702 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to, home buttons, volume buttons, power buttons, and lock buttons.

[0221] Sensor assembly 714 includes one or more sensors for providing state assessments of various aspects of device 700. For example, sensor assembly 714 may detect the on / off state of device 700, the relative positioning of components such as the display and keypad of device 700, changes in the position of device 700 or a component of device 700, the presence or absence of user contact with device 700, the orientation or acceleration / deceleration of device 700, and temperature changes of device 700. Sensor assembly 714 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. Sensor assembly 714 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, sensor assembly 714 may also include an accelerometer, a gyroscope, a magnetometer, a pressure sensor, or a temperature sensor.

[0222] Communication component 716 is configured to facilitate wired or wireless communication between device 700 and other devices. Device 700 can access wireless networks based on communication standards, such as WiFi, 2G, or 3G, or combinations thereof. In one exemplary embodiment, communication component 716 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, communication component 716 also includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0223] In an exemplary embodiment, the apparatus 700 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described graph data compression model training method or graph data compression method.

[0224] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 704 including instructions, which can be executed by a processor 720 of the device 700 to perform the above-described method. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0225] A non-transitory computer-readable storage medium, wherein instructions in the storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the graph data compression model training method or graph data compression method of the aforementioned electronic device.

[0226] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0227] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A graph data compression method, characterized in that, include: Identify at least one data point in the graph data to be compressed; The data depth of the data point is greater than the depth threshold; Identify at least one labeled data point and an unlabeled data point from the data points; By utilizing the category association relationship between the labeled data points and the unlabeled data points, at least one dimension reduction matrix corresponding to the data point is extracted; Matrix calculations based on the graph data and the dimensionality reduction matrix yield the target graph data of the graph data; The labeled data points include: at least one label data point corresponding to each label category; The step of extracting at least one dimensionality reduction matrix corresponding to a data point by utilizing the category association relationship between the labeled data points and the unlabeled data points includes: Using the unlabeled data points as the same category, and combining the label data points corresponding to at least one of the label categories, the association between the label data points of the same category is extracted to obtain intra-category collaborative information; Based on the label data points corresponding to at least one of the label categories, the association between label data points of different categories is extracted to obtain inter-class collaboration information; Based on the intra-class collaboration information and the inter-class collaboration information, a semi-supervised collaborative embedding equation is constructed. In response to the solution instruction for the semi-supervised cooperative embedding equation, the semi-supervised cooperative embedding equation is solved to obtain the dimensionality reduction matrix.

2. The method according to claim 1, characterized in that, The step of extracting the association between label data points of the same category by taking the unlabeled data points as the same category and combining them with the label data points corresponding to at least one of the label categories to obtain intra-category collaborative information includes: Based on the label data points corresponding to at least one of the label categories, the same feature relationships of the data points in the same category of labels are extracted to obtain the first reconstruction coefficient vector corresponding to at least one of the data points. The unlabeled data points are taken as unlabeled categories, and the same feature relationships of the data points in the unlabeled categories are extracted to obtain at least one second reconstruction coefficient vector corresponding to each data point. The first reconstruction coefficient vector and the second reconstruction coefficient vector corresponding to the data points are weighted to obtain at least one semi-supervised weight vector corresponding to each data point. Based on the semi-supervised weight vectors corresponding to at least one of the data points, the intra-class collaboration information is determined; the intra-class collaboration information is used to construct the directed reconstruction graphs corresponding to at least one of the data points.

3. The method according to claim 2, characterized in that, The step of extracting the same feature relationships of the data points in the same category of labels based on the label data points corresponding to at least one of the label categories, and obtaining the first reconstruction coefficient vector corresponding to at least one of the data points respectively, includes: Based on the label data points corresponding to at least one of the label categories, the relationship between the same characteristics of the label data points of the same label category is extracted to obtain the same type of representation vector corresponding to at least one of the data points. Based on the similar representation vectors of the data points, the first reconstruction coefficient vector of the data points is calculated to determine the first reconstruction coefficient vector corresponding to at least one of the data points.

4. The method according to claim 3, characterized in that, The homogeneous representation vector of any one of the data points is determined in the following way: For a first data point among at least one of the data points, the number of at least one data point is used as the vector length to establish an empty first representation vector for the first data point; Traverse at least one of the data points and determine whether the data point belongs to the same category as the first data point. If yes, determine that the data point takes a value of 1 at the coordinates corresponding to the first representation vector. If no, determine that the data point takes a value of 0 at the coordinates corresponding to the first representation vector. The first representation vector whose coordinate values ​​have been determined at the end of the traversal is the same type of representation vector.

5. The method according to claim 2, characterized in that, The step of taking the unlabeled data points as unlabeled categories, extracting the same feature relationships of the data points in the unlabeled categories, and obtaining at least one second reconstruction coefficient vector corresponding to each of the data points includes: Treating the unlabeled data points as the same category, extract the unlabeled features corresponding to the data points to obtain at least one unlabeled vector corresponding to each data point. Based on the unlabeled vector corresponding to the data point, the second reconstruction coefficient vector of the data point is calculated to obtain at least one second reconstruction coefficient vector corresponding to the data point.

6. The method according to claim 5, characterized in that, The unlabeled vector of any of the data points is determined in the following way: For the second data point among at least one of the data points, an empty label vector is created for the second data point, using the number of at least one data point as the vector length. Traverse at least one of the data points and determine whether the data point and the second data point are both unlabeled data points. If so, determine that the data point has a value of 1 at the coordinates corresponding to the label vector. If not, determine that the data point has a value of 0 at the coordinates corresponding to the label vector. The label vector whose coordinates are all determined at the end of the traversal is the unlabeled vector.

7. The method according to claim 1, characterized in that, The step of extracting the association between label data points of different categories based on label data points corresponding to at least one of the label categories to obtain inter-class collaboration information includes: Based on the label data points corresponding to at least one of the label categories, the relationship between the same characteristics of the data points in different label categories is extracted to obtain the third reconstruction coefficient vector corresponding to at least one of the data points. Based on the third reconstruction coefficient vector corresponding to at least one of the data points, the collaborative relationship between samples of different categories is extracted to obtain inter-class collaborative information.

8. The method according to claim 7, characterized in that, The step of extracting relationships between the common characteristics of the data points in different label categories based on the label data points corresponding to at least one of the label categories, and obtaining a third reconstruction coefficient vector corresponding to at least one of the data points respectively, includes: For each of the label data points corresponding to at least one of the label categories, an inter-class representation vector corresponding to each of the at least one data point is determined based on label data points of different categories. Based on the inter-class representation vector of the data points, the third reconstruction coefficient vector of the data points is calculated to obtain at least one third reconstruction coefficient vector corresponding to each of the data points.

9. The method according to claim 8, characterized in that, The inter-class representation vector of any given data point is determined through the following steps: For a third data point among at least one of the data points, the number of at least one data point is used as the vector length to establish an empty second representation vector for the third data point; Traverse at least one of the data points and determine whether the data point and the third data point belong to the same or different categories. If yes, determine that the data point takes a value of 1 at the coordinates corresponding to the second representation vector. If no, determine that the data point takes a value of 0 at the coordinates corresponding to the second representation vector. The second representation vector, whose coordinate values ​​are determined at the end of the traversal, is the inter-class representation vector.

10. The method according to claim 7, characterized in that, The step of extracting the collaborative relationships between samples of different categories based on the third reconstruction coefficient vector corresponding to at least one of the data points to obtain inter-class collaborative information includes: Based on the third reconstruction coefficient vector corresponding to at least one of the data points, vectors are combined according to the positions of at least one of the data points to obtain an inter-class reconstruction matrix. For any given data point, the transpose of the target position in the inter-class reconstruction matrix is ​​determined based on the target position of the data point in the corresponding third reconstruction coefficient vector. Determine the first value corresponding to the target position and the second value of the transpose position from the inter-class reconstruction matrix; The maximum value between the first value and the second value is selected as the target value for the target position, so as to obtain the target value corresponding to each position in the inter-class reconstruction matrix; The inter-class collaboration information is determined by replacing the inter-class reconstruction matrix at each position with the corresponding target value.

11. The method according to claim 1, characterized in that, The construction of a semi-supervised collaborative embedding equation based on the intra-class collaborative information and the inter-class collaborative information includes: A semi-supervised cooperative embedding equation is generated using at least one of the data points, the intra-class cooperative information, and the inter-class cooperative information as known quantities and the dimensionality reduction matrix as an unknown quantity.

12. The method according to any one of claims 1-10, characterized in that, The graph data includes a graph, and the multiple data points of the graph include multiple nodes, including labeled nodes and unlabeled nodes.

13. The method according to any one of claims 1-10, characterized in that, The image data includes a hyperspectral image, and the multiple data points of the hyperspectral image include multiple pixels, which include labeled pixels and unlabeled pixels.

14. An apparatus for graph data compression, characterized in that, include: A data determination unit is used to determine at least one data point in the graph data to be compressed; The data depth of the data point is greater than the depth threshold; A label segmentation unit is used to determine at least one labeled data point and an unlabeled data point among the data points; The matrix extraction unit is used to extract at least one dimension-reduced matrix corresponding to the data point by utilizing the data point correlation relationship between the labeled data point and the unlabeled data point. The target determination unit is used to obtain the target graph data of the graph data based on the graph data and the dimension reduction matrix through matrix calculation. The labeled data points include: at least one label data point corresponding to each label category; The matrix extraction unit is specifically used to extract the association between the label data points of the same category by taking the unlabeled data points as the same category and combining the label data points corresponding to at least one label category, thereby obtaining intra-category collaborative information. Based on the label data points corresponding to at least one of the label categories, the association between label data points of different categories is extracted to obtain inter-class collaboration information; Based on the intra-class collaboration information and the inter-class collaboration information, a semi-supervised collaborative embedding equation is constructed. In response to the solution instruction for the semi-supervised cooperative embedding equation, the semi-supervised cooperative embedding equation is solved to obtain the dimensionality reduction matrix.

15. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement any one of the graph data compression methods as described in claims 1-13.

16. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement any one of the graph data compression methods as described in claims 1-13.

Images