Graph clustering method, device, storage medium and equipment of data points
By generating and filtering density change rate and signal intensity change rate matrices, and combining them with preset thresholds to obtain reachable neighbor matrix, the problem of instability in existing clustering algorithms is solved, adaptive data point clustering is realized, and the accuracy and calculation speed of clustering results are improved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JIAXING JUSU ELECTRONIC TECH CO LTD
- Filing Date
- 2022-11-08
- Publication Date
- 2026-07-03
AI Technical Summary
Existing clustering algorithms are easily affected by parameter selection when processing data points, leading to unstable clustering results. Furthermore, when data points have unique spatial distribution characteristics, they are prone to over-denoising or over-clustering, resulting in inaccurate results.
By calculating the reachability distance matrix of data points in the point cloud set, density change rate and signal strength change rate matrices are generated and filtered. Data operations are performed in combination with preset thresholds to obtain the reachability neighbor matrix. Finally, graph clustering results are obtained based on noise recognition thresholds, reducing the dependence on algorithm parameters.
Adaptive data point clustering was achieved, which improved the stability and reliability of the clustering results, reduced the complexity of manual parameter tuning, and ensured the accuracy of the clustering results and the computation speed.
Smart Images
Figure CN115795337B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of machine learning technology, and in particular to a graph clustering method, apparatus, storage medium, and device for data points. Background Technology
[0002] Clustering algorithms are a common type of unsupervised algorithm in the field of machine learning. Various types of clustering algorithms have been proposed and continuously improved in practice for datasets with different characteristics or learning tasks. Among them, algorithms that consider the density distribution characteristics of data points not only perform well in clustering tasks but also show good results in data denoising tasks.
[0003] Current clustering algorithms, especially those based on data point density, often consider the distance between data points or the number of data points within a specific range as the clustering criterion. For example, the DBSACN algorithm classifies samples with fewer than a preset number of data points within a specific radius neighborhood as noise points, while the HDBSACN algorithm uses the distance between sample data points as the classification criterion. Other clustering algorithms, such as spectral clustering, do not use density as a direct variable but instead use a similarity matrix between data points in different clusters to reflect noise point information. However, these algorithms heavily rely on parameter selection, resulting in poor performance stability. Furthermore, when data points have unique spatial distribution characteristics, existing clustering algorithms often suffer from over-denoising or over-clustering, leading to inaccurate clustering results. Summary of the Invention
[0004] In view of this, the present invention provides a graph clustering method, apparatus, storage medium and device for data points, which can adaptively realize data point clustering, reduce the complexity of manual parameter tuning and the dependence of calculation results on parameters, improve the stability and reliability of clustering results, and is fast and efficient.
[0005] In a first aspect, embodiments of the present invention provide a graph clustering method for data points, the method comprising:
[0006] Calculate the total number of data points n in the point cloud set, and generate a distance matrix between the n points.
[0007] Generate the density change rate matrix and the signal strength change rate matrix based on the reachability distance matrix;
[0008] The density change rate matrix is filtered according to a preset change rate threshold, and the signal intensity change rate matrix is filtered according to a preset change rate threshold;
[0009] Data operations are performed on the filtered density change rate matrix and signal intensity change rate matrix to obtain the reachable neighbor matrix;
[0010] The graph clustering results are obtained based on the reachable neighbor matrix and the preset noise recognition threshold.
[0011] Further, generating the density change rate matrix based on the reachability distance matrix includes:
[0012] Based on the preset minimum reachable distance percentile k, query the kth minimum distance in each row of the reachable distance matrix, extract the minimum value from the kth minimum distance in each row of all rows, and use the minimum value as the neighbor query radius;
[0013] Generate a density matrix based on the neighbor query radius and the reachability distance matrix;
[0014] Generate a density change rate matrix based on the density matrix.
[0015] Further, generating the signal strength change rate matrix based on the reachability distance matrix includes:
[0016] Generate a target orientation matrix based on the coordinate information of each data point in the reachability distance matrix;
[0017] A signal strength matrix is generated based on the attenuation pattern of the signal strength of data points in the spatial distribution.
[0018] A signal intensity change rate matrix is generated based on the target orientation matrix and the signal intensity matrix.
[0019] Further, the density change rate matrix is filtered for reachable neighbors according to a preset change rate threshold, and the signal strength change rate matrix is filtered according to a preset change rate threshold, including:
[0020] Set the value of the density change rate in the density change rate matrix that is less than a preset change rate threshold to 1;
[0021] The value of the density change rate in the density change rate matrix that is greater than a preset change rate threshold is set to 0.
[0022] Set the value of the signal strength change rate in the signal strength change rate matrix that is less than the preset change rate threshold to 1;
[0023] Set the values in the signal strength change rate matrix that are greater than a preset change rate threshold to 0.
[0024] Furthermore, the value of the preset noise recognition threshold is related to the total number of data points n, and the value of the preset noise recognition threshold increases as the total number of data points n increases.
[0025] Further, obtaining the graph clustering result based on the reachable neighbor matrix and the preset noise recognition threshold includes:
[0026] Calculate the correlation coefficient or eigenvector matrix of the reachable neighbor matrix;
[0027] When the correlation coefficient is less than the preset noise identification threshold, the non-noise point clustering result is obtained;
[0028] When the correlation coefficient is greater than the preset noise identification threshold, a set of noise points is obtained.
[0029] Further, obtaining the graph clustering result based on the reachable neighbor matrix and the preset noise recognition threshold includes:
[0030] Calculate the connected components in the reachable neighbor matrix;
[0031] Data points in the connected domain whose density change rate is less than the preset noise identification threshold are taken as non-noise point clustering results.
[0032] The data points in the connected domain whose density change rate is greater than the preset noise identification threshold are taken as the noise point set.
[0033] In a second aspect, embodiments of the present invention provide a graph clustering apparatus for data points, the apparatus comprising:
[0034] The distance matrix generation module is used to calculate the total number of data points n in the point cloud set and generate a distance matrix between the n points.
[0035] A density change rate and intensity change rate matrix generation module is used to generate a density change rate matrix and a signal intensity change rate matrix respectively based on the reachability distance matrix;
[0036] The filtering module is used to filter the density change rate matrix according to a preset change rate threshold and the signal intensity change rate matrix according to a preset change rate threshold.
[0037] The calculation module is used to perform data calculations on the filtered density change rate matrix and signal intensity change rate matrix to obtain the reachable neighbor matrix.
[0038] The result generation module is used to obtain graph clustering results based on the reachable neighbor matrix and a preset noise recognition threshold.
[0039] Thirdly, embodiments of the present invention provide a storage medium storing a computer program, wherein the computer program is configured to execute the method described in any one of the first aspects when running.
[0040] Fourthly, embodiments of the present invention provide an apparatus including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the method described in any one of the first aspects.
[0041] The technical solution provided by this invention generates a density change rate matrix and a signal intensity change rate matrix using a reachability distance matrix. These matrices are then filtered, and their intersection is calculated to obtain a reachability neighbor matrix. Based on this matrix and a preset noise recognition threshold, a graph clustering result is obtained. This enables clustering of data points in a dataset containing various feature types, requiring clustering from multiple dimensions. By combining algorithm parameters with the inherent characteristics of the data points, this invention adaptively clusters data points, reducing the complexity of manual parameter tuning, decreasing the dependence of clustering results on algorithm parameters, and improving the stability and reliability of the clustering results. It achieves effective clustering of data points, resulting in accurate target shapes, clear contours, accurate noise recognition, and fast computation speed. Attached Figure Description
[0042] Figure 1 This is a flowchart of a graph clustering method for data points provided in an embodiment of the present invention;
[0043] Figure 2 This is a flowchart of a graph clustering method for data points provided in another embodiment of the present invention;
[0044] Figure 3 This is a schematic diagram of the structure of the graph clustering device for data points provided in an embodiment of the present invention. Detailed Implementation
[0045] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.
[0046] During the research and development process, the inventors discovered that the currently common clustering algorithms include OPTICS, DBSCAN, HDBSCAN, and spectral algorithms. Among them, the OPTICS algorithm is not sensitive enough to the boundaries of data points in each cluster after clustering, and the calculation process is not efficient enough. The DBSCAN and HDBSCAN algorithms are more efficient clustering methods based on the OPTICS algorithm by introducing the idea of hierarchical clustering. Although the DBSCAN and HDBSCAN algorithms are more sensitive to noise points, they are not ideal in handling the boundaries of data points in each cluster after clustering. The spectral clustering algorithm is more efficient than the OPTICS, DBSCAN, and HDBSCAN algorithms because it does not traverse all data points. However, the spectral algorithm is very sensitive to the similarity matrix and is heavily dependent on parameter selection, resulting in poor clustering stability. Furthermore, when data points have unique spatial distribution characteristics, existing clustering algorithms often suffer from over-denoising or over-clustering, leading to inaccurate clustering results.
[0047] In view of this, the inventors of this application, addressing the shortcomings of existing clustering algorithms discovered during the research process, designed a graph clustering method for data points. See [link to application]. Figure 1 , Figure 1 This is a flowchart of a graph clustering method for data points provided in an embodiment of the present invention. The method includes the following steps:
[0048] Step 11: Calculate the total number of data points n in the point cloud set, and generate a distance matrix between the n points.
[0049] With the rise of automotive radar point cloud technology, increasing the number of point clouds and improving their quality are two major issues in this field. Improving point cloud quality is crucial for enhancing the efficiency of point cloud data utilization. Therefore, after obtaining the set of point clouds to be tested, the point cloud set can first be denoised (any denoising method in the existing technology can be used for data point denoising, which will not be elaborated upon in this invention). Then, the total number of data points n in the point cloud set is calculated. Finally, based on the preset total number of data points n, n mutually reachable distance matrices M are generated from the data points in the set of point clouds to be tested. dis Where n is a natural number greater than 2.
[0050] Step 12: Generate the density change rate matrix and the signal strength change rate matrix based on the reachability distance matrix.
[0051] In this step, a density change rate matrix and a signal strength change rate matrix are generated based on the reachability distance matrix. The density change rate matrix is used to represent the density change rate of data points, and the signal strength change rate matrix is used to represent the signal strength change rate of data points.
[0052] Step 13: Filter the density change rate matrix according to a preset change rate threshold, and filter the signal intensity change rate matrix according to a preset change rate threshold.
[0053] In this step, the density change rate matrix is filtered according to a preset change rate threshold r. Counts in the density change rate matrix with a density change rate less than the preset change rate threshold r are set to 1, while those with a density change rate greater than the preset change rate threshold r are set to 0. This yields the filtered density change rate matrix.
[0054] The signal strength change rate matrix is filtered according to a preset change rate threshold s. Values in the signal strength change rate matrix with a signal strength change rate less than the preset change rate threshold s are set to 1, and values with a signal strength change rate greater than the preset change rate threshold s are set to 0. This yields the filtered signal strength change rate matrix.
[0055] Step 14: Perform an intersection operation on the filtered density change rate matrix and signal intensity change rate matrix to obtain the reachable neighbor matrix.
[0056] In this step, when the data points in the dataset contain multiple feature types and need to be clustered from multiple dimensions, the accuracy of the data point clustering results can be improved by performing an intersection operation on the filtered density change rate matrix and signal intensity change rate matrix.
[0057] Step 15: Obtain the graph clustering results based on the reachable neighbor matrix and the preset noise recognition threshold.
[0058] In this step, noise points are mostly isolated points or clusters consisting of a very small number of data points. After clustering the data points, noise can be identified based on this characteristic. The noise identification threshold can be set to m=2, or the value of the noise identification threshold can be appropriately increased if the preset number of data points n is large.
[0059] In this embodiment, by calculating the correlation coefficient or eigenvector matrix of the reachable neighbor matrix, and based on the noise identification threshold m, non-noise clustering results and sets of noisy data points with different densities can be obtained. The non-noise clustering results include density-continuous clustering results and independent density clustering results. Specifically, the correlation coefficient of the reachable neighbor matrix can be compared with the noise identification threshold m. When the correlation coefficient of the reachable neighbor matrix is less than the preset noise threshold m, density-continuous clustering results and independent density clustering results are obtained; when the correlation coefficient of the reachable neighbor matrix is greater than the preset noise threshold m, a set of noisy points is obtained. This method can clearly display more details of the non-noise clustering.
[0060] In other embodiments, connected components in the reachable neighbor matrix can be calculated, and interconnected data points belong to the same cluster. Based on a noise identification threshold m, non-noise clusters and sets of noisy data points with continuously varying density can be obtained. Specifically, the density change rate in the connected component can be compared with a preset noise identification threshold. Data points with a density change rate less than the preset noise identification threshold are considered as density-continuous clustering results and independent density clustering results, while data points with a density change rate greater than the preset noise identification threshold are considered as a set of noisy points. This method can display the fully aggregated non-noise clusters.
[0061] See Figure 2 , Figure 2 This is a flowchart of another graph clustering method for data points provided in an embodiment of the present invention. The method includes the following steps:
[0062] Step 21: Calculate the total number of data points n in the point cloud set, and generate a distance matrix between the n points.
[0063] Step 22: Based on the preset minimum reachable distance percentile k, query the kth minimum distance in each row of the reachable distance matrix, and extract the minimum value from the kth minimum distance in each row of all rows, and use the minimum value as the neighbor query radius.
[0064] In this step, the minimum reachable distance percentile k is used in the reachability distance matrix M. dis The query finds the k-th minimum distance in each row and extracts the minimum value from these minimum distances, using this minimum value as the neighbor query radius R. For example, the reachability distance matrix M... dis Given a 5x4 matrix, with the minimum reachability percentile k being 3, we search for the third minimum distance in each row of the reachability distance matrix. Then, we extract the minimum value from all third minimum distances across all 5 rows and use this minimum value as the neighbor search radius. Note that the value of the minimum reachability percentile k must be less than the reachability distance matrix M. dis The number of columns.
[0065] Step 23: Generate a density matrix based on the neighbor query radius and the reachability distance matrix.
[0066] In this step, the data point density ρ reflects the number of data points per unit space, and can therefore be defined as the number of zero points of a data point within the neighbor query radius R. Thus, based on the reachability distance matrix M... dis Generate density matrix M density .
[0067] Step 24: Generate a density change rate matrix based on the density matrix.
[0068] When generating density matrix M density Then, from the density matrix Mdensity It can generate a density change rate matrix M delta-density Here, the rate of change of density can be calculated using the following formula:
[0069]
[0070] Where, ρ i ρ represents the density at the i-th data point. j This represents the density at the j-th data point.
[0071] Step 25: Generate the target orientation matrix based on the coordinate information of each data point in the reachability distance matrix.
[0072] In this step, due to the reachability distance matrix M dis It contains the Cartesian coordinate information (x, y, z) of the data points, so the orientation information of each data point can be extracted, and the target orientation matrix M can be generated based on the orientation information of each data point. orientation Among them, the target azimuth information comprehensively considers the target's horizontal azimuth angle θ relative to the radar. H Vertical azimuth θ V The distance and the distance can be obtained using the following formulas:
[0073]
[0074]
[0075] Where x, y, and z are the coordinates of each data point in the Cartesian coordinate system.
[0076] Step 26: Generate a signal strength matrix based on the attenuation pattern of the signal strength of the data points in the spatial distribution.
[0077] In this step, based on the attenuation law of signal strength in the spatial distribution of data points—that is, the signal strength is strongest when the radar is directly facing the target, and the signal strength weakens the more it deviates to the left, right, up, or down—a signal strength matrix M is generated. intensity .
[0078] Step 27: Generate a signal intensity change rate matrix based on the target orientation matrix and the signal intensity matrix.
[0079] In this step, based on the target orientation matrix M orientatio and signal strength matrix M intensity It can generate a signal intensity change rate matrix M delta-intensity .
[0080] The rate of change of the signal strength I of the data point with respect to the distance d can be calculated by the following formula:
[0081]
[0082] in,
[0083] The rate of change of the data point signal strength I with respect to the azimuth angle can be calculated using the following formula:
[0084]
[0085] Where α represents the weighting coefficient, which allows the formula to comprehensively consider the influence of both horizontal and vertical azimuth angles. Through the inventor's experiments, the preferred value for the weighting coefficient α is 2 / 3. The radar signal strength is also approximately estimated here to be proportional to the square of the azimuth angle.
[0086] Step 28: Filter the density change rate matrix according to a preset change rate threshold, and filter the signal intensity change rate matrix according to a preset change rate threshold.
[0087] In this step, the density change rate matrix is filtered according to a preset change rate threshold r. Counts in the density change rate matrix with a density change rate less than the preset change rate threshold r are set to 1, while those with a density change rate greater than the preset change rate threshold r are set to 0. This yields the filtered density change rate matrix.
[0088] The signal strength change rate matrix is filtered according to a preset change rate threshold s. Values in the signal strength change rate matrix with a signal strength change rate less than the preset change rate threshold s are set to 1, and values with a signal strength change rate greater than the preset change rate threshold s are set to 0. This yields the filtered signal strength change rate matrix.
[0089] Step 29: Perform an intersection operation on the filtered density change rate matrix and signal intensity change rate matrix to obtain the reachable neighbor matrix.
[0090] In this step, the intersection operation is performed on the filtered density change rate matrix and signal intensity change rate matrix to obtain the reachable neighbor matrix.
[0091] Step 210: Obtain graph clustering results based on the reachable neighbor matrix and the preset noise recognition threshold.
[0092] In this step, noise points are mostly isolated points or clusters consisting of a very small number of data points. After clustering the data points, noise can be identified based on this characteristic. The noise identification threshold can be set to m=2, or the value of the noise identification threshold can be appropriately increased if the preset number of data points n is large.
[0093] In this embodiment, by calculating the correlation coefficient or eigenvector matrix of the reachable neighbor matrix, and based on the noise identification threshold m, non-noise clustering results and sets of noisy data points with different densities can be obtained. The non-noise clustering results include density-continuous clustering results and independent density clustering results. Specifically, the correlation coefficient of the reachable neighbor matrix can be compared with the noise identification threshold m. When the correlation coefficient of the reachable neighbor matrix is less than the preset noise threshold m, density-continuous clustering results and independent density clustering results are obtained; when the correlation coefficient of the reachable neighbor matrix is greater than the preset noise threshold m, a set of noisy points is obtained. This method can clearly display more details of the non-noise clustering.
[0094] In other embodiments, connected components in the reachable neighbor matrix can be calculated, and interconnected data points belong to the same cluster. Based on a noise identification threshold m, non-noise clusters and sets of noisy data points with continuously varying density can be obtained. Specifically, the density change rate in the connected component can be compared with a preset noise identification threshold. Data points with a density change rate less than the preset noise identification threshold are considered as density-continuous clustering results and independent density clustering results, while data points with a density change rate greater than the preset noise identification threshold are considered as a set of noisy points. This method can display the fully aggregated non-noise clusters.
[0095] Therefore, a density change rate matrix and a signal intensity change rate matrix are generated using the reachability distance matrix. These matrices are then filtered, and their intersection is calculated to obtain a reachability neighbor matrix. The graph clustering result is then obtained based on the reachability neighbor matrix and a preset noise recognition threshold. This method enables clustering of data points in a dataset containing various feature types, requiring clustering from multiple dimensions. By combining algorithm parameters with the inherent characteristics of the data points, this invention adaptively clusters data points, reducing the complexity of manual parameter tuning, decreasing the dependence of clustering results on algorithm parameters, and improving the stability and reliability of the clustering results. It achieves effective clustering of data points, resulting in accurate target shapes, clear contours, accurate noise recognition, and fast computation speed.
[0096] Please refer to Figure 3 , Figure 3 This is a structural diagram of a graph clustering device for data points provided in an embodiment of the present invention. The device includes:
[0097] The distance matrix generation module 31 is used to calculate the total number of data points n in the point cloud set and generate a distance matrix between n points that can reach each other.
[0098] The density change rate and intensity change rate matrix generation module 32 is used to generate a density change rate matrix and a signal intensity change rate matrix respectively based on the reachability distance matrix;
[0099] The filtering module 33 is used to filter the density change rate matrix according to a preset change rate threshold and to filter the signal intensity change rate matrix according to a preset change rate threshold.
[0100] The calculation module 34 is used to perform data calculations on the filtered density change rate matrix and signal intensity change rate matrix to obtain the reachable neighbor matrix;
[0101] The result generation module 35 is used to obtain graph clustering results based on the reachable neighbor matrix and a preset noise recognition threshold.
[0102] In some embodiments, the density change rate and intensity change rate matrix generation module 32 includes a density change rate matrix generation unit 321 and an intensity change rate matrix generation module 322.
[0103] The density change rate matrix generation unit 321 may include:
[0104] The neighbor query radius subunit 3211 is used to query the kth minimum distance in each row of the reachability matrix according to the preset minimum reachability percentile k, and extract the minimum value in the kth minimum distance of each row in all rows, and use the minimum value as the neighbor query radius;
[0105] Density matrix generation subunit 3212 is used to generate a density matrix based on the neighbor query radius and the reachability distance matrix;
[0106] The density change rate matrix generation subunit 3213 is used to generate a density change rate matrix based on the density matrix.
[0107] The intensity change rate matrix generation module 322 may include:
[0108] The target orientation matrix generation subunit 3221 is used to generate a target orientation matrix based on the coordinate information of each data point in the reachability distance matrix.
[0109] Signal strength matrix generation subunit 3222 is used to generate a signal strength matrix based on the attenuation law of the signal strength of data points in the spatial distribution.
[0110] The signal strength change rate matrix generation subunit 3223 is used to generate a signal strength change rate matrix based on the target orientation matrix and the signal strength matrix.
[0111] In some embodiments, the filtering module 33 performs reachable neighbor filtering on the density change rate matrix according to a preset change rate threshold, and filters the signal intensity change rate matrix according to a preset change rate threshold, including:
[0112] Set the value of the density change rate in the density change rate matrix that is less than a preset change rate threshold to 1;
[0113] The value of the density change rate in the density change rate matrix that is greater than a preset change rate threshold is set to 0.
[0114] Set the value of the signal strength change rate in the signal strength change rate matrix that is less than the preset change rate threshold to 1;
[0115] Set the values in the signal strength change rate matrix that are greater than a preset change rate threshold to 0.
[0116] In some embodiments, the value of the preset noise recognition threshold is related to the total number of data points n, and the value of the preset noise recognition threshold increases as the total number of data points n increases.
[0117] In some embodiments, the result generation module 35 may include:
[0118] The first calculation subunit 351 is used to calculate the correlation coefficient or eigenvector matrix of the reachable neighbor matrix;
[0119] The first processing subunit 352 is used to obtain non-noise point clustering results when the correlation coefficient is less than the preset noise identification threshold;
[0120] The second processing subunit 353 is used to obtain a set of noise points when the correlation coefficient is greater than the preset noise identification threshold.
[0121] In some embodiments, the result generation module 35 may include:
[0122] The second calculation subunit 354 is used to calculate the connected components in the reachable neighbor matrix;
[0123] The third processing subunit 355 is used to take the data points in the connected domain whose density change rate is less than the preset noise identification threshold as non-noise point clustering results.
[0124] The fourth processing subunit 356 is used to take the data points in the connected domain whose density change rate is greater than the preset noise identification threshold as the noise point set.
[0125] The technical solution provided by this invention generates a density change rate matrix and a signal intensity change rate matrix using a reachability distance matrix. These matrices are then filtered, and their intersection is calculated to obtain a reachability neighbor matrix. Based on this matrix and a preset noise recognition threshold, a graph clustering result is obtained. This enables clustering of data points in a dataset containing various feature types, requiring clustering from multiple dimensions. By combining algorithm parameters with the inherent characteristics of the data points, this invention adaptively clusters data points, reducing the complexity of manual parameter tuning, decreasing the dependence of clustering results on algorithm parameters, and improving the stability and reliability of the clustering results. It achieves effective clustering of data points, resulting in accurate target shapes, clear contours, accurate noise recognition, and fast computation speed.
[0126] It should be noted that the graph clustering device for data point density continuity in the embodiments of the present invention and the graph clustering method for data point density continuity in the above embodiments belong to the same inventive concept. Technical details not described in detail in this device can be found in the previous description of the method, and will not be repeated here.
[0127] Furthermore, embodiments of the present invention also provide a storage medium storing a computer program, wherein the computer program is configured to execute the aforementioned method at runtime.
[0128] This invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the methods described above.
[0129] Those skilled in the art will understand that all or part of the steps in the above methods can be implemented by a program instructing related hardware (e.g., a processor), and the program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk. Optionally, all or part of the steps in the above embodiments can also be implemented using one or more integrated circuits. Accordingly, each module / unit in the above embodiments can be implemented in hardware, such as by an integrated circuit to implement its corresponding function, or it can be implemented in the form of a software functional module, such as by a processor executing a program / instruction stored in memory to implement its corresponding function. This invention is not limited to any particular combination of hardware and software.
[0130] While the embodiments disclosed in this invention are as described above, the content is merely for the purpose of facilitating understanding of the invention and is not intended to limit the invention. Any person skilled in the art to which this invention pertains may make any modifications and changes to the form and details of the implementation without departing from the spirit and scope disclosed herein; however, the scope of patent protection of this invention shall still be determined by the scope defined in the appended claims.
Claims
1. A graph clustering method for data points, characterized in that, The method includes: Calculate the total number of data points n in the point cloud set, and generate a distance matrix between the n points. Generate the density change rate matrix and the signal strength change rate matrix based on the reachability distance matrix; The density change rate matrix is filtered according to a preset change rate threshold, and the signal intensity change rate matrix is filtered according to a preset change rate threshold; The intersection operation is performed on the filtered density change rate matrix and signal intensity change rate matrix to obtain the reachable neighbor matrix; The graph clustering result is obtained based on the reachable neighbor matrix and a preset noise identification threshold, wherein generating the signal intensity change rate matrix based on the reachable distance matrix includes: Generate a target orientation matrix based on the coordinate information of each data point in the reachability distance matrix; A signal strength matrix is generated based on the attenuation pattern of the signal strength of data points in the spatial distribution. A signal intensity change rate matrix is generated based on the target orientation matrix and the signal intensity matrix.
2. The method according to claim 1, characterized in that, Generating the density change rate matrix based on the reachability distance matrix includes: Based on the preset minimum reachable distance percentile k, query the kth minimum distance in each row of the reachable distance matrix, extract the minimum value from the kth minimum distance in each row of all rows, and use the minimum value as the neighbor query radius; Generate a density matrix based on the neighbor query radius and the reachability distance matrix; Generate a density change rate matrix based on the density matrix.
3. The method according to claim 1, characterized in that, The density change rate matrix is filtered for reachable neighbors according to a preset change rate threshold, and the signal strength change rate matrix is filtered according to a preset change rate threshold, including: Set the value of the density change rate in the density change rate matrix that is less than a preset change rate threshold to 1; The value of the density change rate in the density change rate matrix that is greater than a preset change rate threshold is set to 0. Set the value of the signal strength change rate in the signal strength change rate matrix that is less than the preset change rate threshold to 1; Set the values in the signal strength change rate matrix that are greater than a preset change rate threshold to 0.
4. The method according to claim 1, characterized in that, The value of the preset noise recognition threshold is related to the total number of data points n, and the value of the preset noise recognition threshold increases as the total number of data points n increases.
5. The method according to claim 1, characterized in that, The step of obtaining the graph clustering result based on the reachable neighbor matrix and the preset noise recognition threshold includes: Calculate the correlation coefficient or eigenvector matrix of the reachable neighbor matrix; When the correlation coefficient is less than the preset noise identification threshold, the non-noise point clustering result is obtained; When the correlation coefficient is greater than the preset noise identification threshold, a set of noise points is obtained.
6. The method according to claim 1, characterized in that, The step of obtaining the graph clustering result based on the reachable neighbor matrix and the preset noise recognition threshold includes: Calculate the connected components in the reachable neighbor matrix; Data points in the connected domain whose density change rate is less than the preset noise identification threshold are taken as non-noise point clustering results. The data points in the connected domain whose density change rate is greater than the preset noise identification threshold are taken as the noise point set.
7. A graph clustering device for data points, characterized in that, The device includes: The distance matrix generation module is used to calculate the total number of data points n in the point cloud set and generate a distance matrix between the n points. A density change rate and intensity change rate matrix generation module is used to generate a density change rate matrix and a signal intensity change rate matrix respectively based on the reachability distance matrix; The filtering module is used to filter the density change rate matrix according to a preset change rate threshold and the signal intensity change rate matrix according to a preset change rate threshold. The calculation module is used to perform data calculations on the filtered density change rate matrix and signal intensity change rate matrix to obtain the reachable neighbor matrix. The result generation module is used to obtain graph clustering results based on the reachable neighbor matrix and a preset noise recognition threshold. Specifically, the density change rate and intensity change rate matrix generation module is used for: Generate a target orientation matrix based on the coordinate information of each data point in the reachability distance matrix; A signal strength matrix is generated based on the attenuation pattern of the signal strength of data points in the spatial distribution. A signal intensity change rate matrix is generated based on the target orientation matrix and the signal intensity matrix.
8. A storage medium, characterized in that, The storage medium stores a computer program, wherein the computer program is configured to execute the method of any one of claims 1 to 6 when it is run.
9. A device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the method of any one of claims 1 to 6.