A class-level abstract method for multi-class scatter plots based on grids

Abstract

The present invention provides a class-level abstract method for multi-class scatter plots based on a grid, and belongs to the field of data processing technology. It includes: 1. Screening the grids according to preset rules to obtain target grids, and filling the blank grids between the target grids by performing multiple closing operations and opening operations; 2. Identifying the grid continents according to the filling conditions of the grids of the processing results, merging different grid continents by the number of interval grids at the continental boundaries between different grid continents to obtain merged processed continents; 3. Filling the blank grids in the middle of different merged processed continents to obtain processed grids, and performing boundary smoothing based on the processed grids to obtain the boundary processing result of the scatter plot. The present invention can construct a class-level visual abstraction of large-scale multi-class scatter plots by extracting class boundaries, thereby overcoming the problem of over-drawing of scatter plots and clearly showing class distribution.

Description

Technical Field

[0001] The invention belongs to the technical field of data processing, and in particular relates to a class-level abstractable method for multi-class scatter plots based on a grid. Background Art

[0002] In today's big data context, scatter plots are widely used in data analysis and presentation as an intuitive and flexible visualization tool. As the scale of data continues to grow, scatter plots face the challenge of over-drawing due to the extensive overlap of nodes, which causes the scatter plots to blur category boundaries, which seriously damages their expressiveness. Therefore, building boundaries for each class of the scatter plot and ignoring individual data points to achieve visual abstraction at the class level has become a feasible solution to alleviate the over-drawing of scatter plots.

[0003] To solve the above technical solution, a common boundary extraction method is the convex hull technology, which alleviates over-drawing by constructing a convex boundary of a set of data points. The convex hull is calculated quickly, but because it is sensitive to outliers, it is easily affected by outliers, resulting in the constructed boundary covering too many non-target areas, thereby reducing the quality of visualization.

[0004] In view of the above technical problems, the present invention provides a class-level abstract method for multi-class scatter plots based on a grid. Summary of the invention

[0005] The object of the present invention is to provide a class-level abstractable method for multi-class scatter plots based on a grid.

[0006] In order to solve the above technical problems, the present invention provides a class-level abstract method for multi-class scatter plots based on a grid, which is characterized by specifically including:

[0007] The entire scatter plot distribution space is divided into equidistant grids, the grids are screened according to preset rules to obtain target grids, the target grids are filled, and the blank grids between the target grids are filled by performing multiple closing operations and opening operations to obtain processing results;

[0008] Identify grid continents according to the filling conditions of the grids in the processing results, construct continental boundaries between grid continents and filter discrete continents based on the identification results, and merge grid continents according to the number of interval grids of the continental boundaries between grid continents to obtain merged continents;

[0009] The blank grids between the merged continents are filled to obtain processed grids, and the boundary smoothing process is performed based on the processed grid to obtain the boundary processing result of the scatter plot.

[0010] The present invention can construct a class-level visual abstraction of a large-scale multi-class scatter plot by extracting class boundaries, thereby overcoming the over-drawing problem of the scatter plot and clearly showing the class distribution. The beneficial effects of the present invention are specifically described as follows:

[0011] 1. Efficient calculation, suitable for large-scale data: The present invention adopts a grid-based boundary extraction strategy, converting the task into building the boundary of the target grid, avoiding the nonlinear time complexity based on the data point scale. This calculation method can greatly improve the efficiency of boundary extraction, especially when processing large-scale data. The present invention implements the proposed method based on Python and conducts experiments on 30 data sets, with the grid division accuracy of 200*200. The results show that the running time of the present invention is 2.1 seconds on average.

[0012] 2. The extracted boundaries are concise and easy to read, and can cover most of the target data points: Compared with traditional technologies, the present invention can more accurately capture the distribution boundaries of the data set. The experimental data results on the 30 data sets tested show that the boundaries extracted by the present invention can cover the target data points with high accuracy (90.2%) and recall rate (94.1%).

[0013] 3. Applicable to any type of large-scale multi-class scatter plot: The present invention has a wide range of applications and is not limited by the algorithm or the specific distribution of scatter points. Therefore, it is applicable to various types of large-scale multi-class scatter plots, providing a highly flexible and universal solution for data analysis and visualization in different fields.

[0014] A further technical solution is that the preset rule includes any one or more of the number of data points in the grid and the proportion of the target category in the grid.

[0015] A further technical solution is to screen the grids according to preset rules to obtain the target grid, which specifically includes:

[0016] Determining a screening target of the grid based on the preset rule, constructing a filtering function based on the screening target, and screening outliers through the filtering function;

[0017] The outliers are filtered, and the target grid is determined according to the filtering results.

[0018] A further technical solution is to fill the blank grids between the target grids to obtain a processing result, which specifically includes:

[0019] Color the target grid, perform a closing operation, fill the blank areas between the target grids, make the discrete target grids continuous into a piece of area, and integrate the cluster-level outliers into the clustered area to the execution result of the closing operation;

[0020] Performing an open operation on the execution result of the closing operation, removing outliers in the execution result of the closing operation to obtain the execution result of the opening operation;

[0021] The open operation and the close operation are repeatedly performed multiple times to fill the blank grids between the target grids to obtain a processing result.

[0022] A further technical solution is that the method for identifying the grid continent is:

[0023] Determine the filled target grid according to the filling condition of the grid of the processing result, and extract the connected components of the undirected graph according to the distribution condition of the filled target grid;

[0024] The grid continents are identified through the connected components of the undirected graph.

[0025] A further technical solution is that the boundary smoothing process includes two steps: aliasing elimination detection and midpoint connection.

[0026] A further technical solution is that the anti-aliasing is detected as protrusions and depressions formed by a single mesh and removed or filled.

[0027] A further technical solution is that the connection midpoint is the midpoint of each mesh edge in the edge, and then a curve of a preset curve type is used to connect the midpoints of the mesh edges.

[0028] Other features and advantages will be described in the following description, and partly become apparent from the description, or understood by practicing the invention. The purpose and other advantages of the invention are realized and obtained by the structures particularly pointed out in the description and the drawings.

[0029] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The above and other features and advantages of the present invention will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings.

[0031] Figure 1 This is an example diagram of erosion, expansion, opening, and closing operations.

[0032] Figure 2 It is a flow chart of a class-level abstractable method for grid-based multi-class scatter plots.

[0033] Figure 3 The following is a schematic diagram of the results of meshing and filtering the mesh on two example data.

[0034] Figure 4 is a schematic diagram of the results of filling the blank grids interspersed between the filled grids on two example data.

[0035] Figure 5 It is a flowchart of the identification method of the grid continent.

[0036] Figure 6 This is a diagram of the process of identifying continents.

[0037] Figure 7 Schematic diagram of the results of identifying continents, constructing continental boundaries, and filtering continents on two example data.

[0038] Figure 8 It is a schematic diagram of the result of merging adjacent continents on two example data.

[0039] Fig. 9 Schematic diagram of the results of filling the holes in the interior of the continent on two example data.

[0040] Fig.10 is a schematic diagram of the results of smoothing the boundary on two example data.

[0041] Fig.11 It is a schematic diagram comparing the extraction results of the present invention and the existing multi-class scatter plot boundary extraction method. DETAILED DESCRIPTION

[0042] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that the present invention will be comprehensive and complete and fully convey the concepts of the example embodiments to those skilled in the art. The same reference numerals in the figures represent the same or similar structures, and thus their detailed description will be omitted.

[0043] The terms "a", "an", "the", and "said" are used to indicate the presence of one or more elements / components / etc.; the terms "comprising" and "having" are used to express an open-ended inclusive meaning and mean that additional elements / components / etc. may be present in addition to the listed elements / components / etc.

[0044] In today's big data context, visualization technology plays a key role in helping people extract insights and understanding from complex data. In this context, scatter plots, as an intuitive and flexible visualization tool, are widely used in data analysis and presentation. However, as the scale of data continues to grow, scatter plots face the challenge of over-drawing due to the extensive overlap of nodes, which makes scatter plots unable to effectively encode category information or blur category boundaries, which seriously damages their expressiveness. Since most of the time people are concerned about class-level information, such as class distribution and class association, building boundaries for each class in the scatter plot while ignoring individual data points to achieve visual abstraction at the class level has become a feasible solution to alleviate the over-drawing of scatter plots.

[0045] There are many methods for constructing the boundaries of multi-class scatter plots, each with its own characteristics, advantages and disadvantages. A common boundary extraction method is the convex hull technique, which alleviates overdrawing by constructing a convex boundary of a set of data points. The convex hull is fast to calculate, but because it is sensitive to outliers, it is easily affected by outliers, causing the constructed boundary to cover too many non-target areas, thereby reducing the quality of visualization. The butterfly plot technique and the concave hull technique are improvements on the convex hull technique, which can construct a more compact boundary and improve the visualization effect.

[0046] Another common boundary extraction technique is based on the mosaic method. The representative technique is GMapGMap, which first constructs the Voronoi space partition with all data points, and then colors the Voronoi grids of the same class with the same color. However, this method easily generates a large number of small fragments in the overlapping areas between classes, resulting in serious visual confusion and reducing the readability of visualization.

[0047] Set visualization is another important branch, which highlights the set belonging of fixed-position data points by constructing the boundaries of the set. BubbleSets is a representative technology of this branch. Although the set area created by it can highlight the belonging of the set, it takes up a lot of space, and the color mixing introduces new colors to the overlapping areas, which may be misunderstood as additional, non-existent sets.

[0048] Building high-quality visual abstractions is not easy. From the shortcomings of previous methods, we can summarize three challenges:

[0049] 1. Whether high-quality boundaries can be constructed: An ideal high-quality class boundary should accurately cover the distribution area of ​​the target data points in terms of objectivity in information expression, and be concise, smooth, and regular in terms of visualization aesthetics and readability. However, a large number of outliers scattered in sparse areas and around clusters will make the boundary rugged and complex, while boundary crossings will increase visual confusion, so how to maintain a balance between the two goals becomes a challenge.

[0050] 2. Whether visual clusters can be identified: In multi-class scatter plots, the same type of data is often scattered in multiple visual clusters that are far away from each other. In this case, independent boundaries must be constructed for each cluster. However, if each cluster needs to be labeled with a class label and the boundaries span a large range, resulting in a large number of intersections, then how to solve this problem without introducing additional cluster identification methods becomes a challenge.

[0051] 3. Whether it supports efficient computing: In theory, visual abstraction methods should be adaptable to data of any size. In practice, it is necessary to provide boundary extraction results for data sets of millions of scales within seconds. Therefore, how to achieve this is also a challenge.

[0052] In response to these challenges, this paper aims to alleviate the over-drawing of large-scale multi-class scatter plots and to build high-quality class-level visual abstractions. A grid-based boundary extraction method is proposed. This method introduces the opening and closing operations in image processing morphological methods to achieve controllable simplification of boundaries and tight wrapping of visual clusters. They unify outlier elimination and cluster area identification in a context-consistent operation framework.

[0053] Opening and closing operations are the basic technologies of the method of the present invention, and the basic principles are as follows: Figure 1 As shown in the figure. The opening operation first performs erosion and then dilation, which aims to remove small grain noise outside the patch area and disconnect the thin adhesion between areas. In contrast, the closing operation first performs dilation and then erosion, which aims to fill the small holes inside the area and connect the adjacent areas. These operations can be chained to achieve superposition of effects.

[0054] Through this method, the present invention successfully solves the problems existing in the boundary extraction methods in the past, such as the inability to properly handle the separation of outliers and similar clusters, resulting in complex and rugged boundaries, a large number of unnecessary intersections, etc. The proposed boundary extraction method can abstract large-scale multi-class scatter plots into a concise, readable, and interpretable class map visualization, providing users with a more intuitive and effective way to present data, and improving the efficiency and accuracy of data analysis and decision-making.

[0055] Example

[0056] In order to solve the above technical problems, Figure 2 As shown, the present invention provides a class-level abstract method for multiple types of scatter plots based on a grid, which is characterized by specifically including:

[0057] The entire scatter plot distribution space is divided into equidistant grids, the grids are screened according to preset rules to obtain target grids, the target grids are filled, and the blank grids between the target grids are filled by performing multiple closing operations and opening operations to obtain processing results;

[0058] Furthermore, the preset rule includes any one or more of the number of data points in the grid and the proportion of the target category in the grid.

[0059] It should be noted that the target grid is obtained by screening the grids according to the preset rules, which specifically includes:

[0060] Determining a screening target of the grid based on the preset rule, constructing a filtering function based on the screening target, and screening outliers through the filtering function;

[0061] The outliers are filtered, and the target grid is determined according to the filtering results.

[0062] Furthermore, the grids are screened according to preset rules to obtain the target grid, which specifically includes:

[0063] Determine, based on the preset rule, matching coefficients of different grids in different dimensions with the preset rule;

[0064] The comprehensive matching coefficient between the grid and the preset rule is determined by the matching coefficient, and the target grid is determined based on the comprehensive matching coefficient.

[0065] It should be noted that determining the target grid based on the comprehensive matching coefficient specifically includes:

[0066] The grid whose comprehensive matching coefficient meets the requirements is taken as the target grid.

[0067] It can be understood that the target grid is obtained by screening the grids according to the preset rules, specifically including:

[0068] Determine the matching coefficients of different grids in different dimensions with the preset rules based on the preset rules, and judge whether there is a dimension whose matching coefficient does not meet the requirements, if so, proceed to the next step, if not, determine that the grid belongs to the target grid;

[0069] Determine whether the average values ​​of the matching coefficients of the grid in different dimensions and the preset rules meet the requirements, if yes, proceed to the next step, if no, determine that the grid does not belong to the target grid;

[0070] The dimensions whose matching coefficients do not meet the requirements are regarded as unmatched dimensions, and it is determined whether the number of the unmatched dimensions meets the requirements. If so, the next step is entered; if not, it is determined that the grid does not belong to the target grid;

[0071] Obtaining the number of the mismatched dimensions, and determining the comprehensive matching coefficients of the mismatched dimensions in combination with the types of different mismatched dimensions and matching coefficients, and judging whether the comprehensive matching coefficients of the mismatched dimensions meet the requirements, if so, proceeding to the next step, if not, determining that the grid does not belong to the target grid;

[0072] The matching coefficients of the grid in different dimensions with the preset rules are obtained, and the comprehensive matching coefficients of the unmatched dimensions are combined to determine the comprehensive matching coefficients of the grid with the preset rules, and the target grid is determined based on the comprehensive matching coefficients.

[0073] Specifically, the blank grids between the target grids are filled to obtain a processing result, which specifically includes:

[0074] Color the target grid, perform a closing operation, fill the blank areas between the target grids, make the discrete target grids continuous into a piece of area, and integrate the cluster-level outliers into the clustered area to the execution result of the closing operation;

[0075] Performing an open operation on the execution result of the closing operation, removing outliers in the execution result of the closing operation to obtain the execution result of the opening operation;

[0076] The open operation and the close operation are repeatedly performed multiple times to fill the blank grids between the target grids to obtain a processing result.

[0077] Specifically, in the step of gridding and filtering the grid, the entire scatter plot distribution space needs to be divided into equidistant grids first, and the grids need to be filtered according to the number of data points in the grid, the proportion of the target category in the grid, or other custom rules. In order to support arbitrary filtering rules, this step requires the user to customize a filter function filter. By default, the filter is a simple threshold judgment, that is, the number of target data points exceeds the threshold p1 set by the user. The setting of p1 can filter out some outliers, for example Figure 3 The point in the middle circle.

[0078] This gridding method introduces the concept of regions, which allows boundaries of different ranges to be constructed by adjusting the filtering conditions. At the same time, since the subsequent steps are all grid-based, the time required for boundary extraction will not increase rapidly with the increase in the number of data points. In this step, grids with only a small number of data points are filtered out, and the remaining grids are filled to ensure the effective execution of subsequent steps.

[0079] It should also be noted that in the above steps, a series of target grids are obtained. Although these grids provide a general understanding of the clustering area, they usually show the characteristics of discretization and fragmentation and cannot be directly used to generate boundaries. In addition, cluster-level outliers close to the clustering area and class-level outliers scattered in the background will also destroy the clarity of the boundary. In order to solve these problems, the present invention introduces morphological methods in image processing.

[0080] Specifically, the present invention first colors the target grid, and then performs closing and opening operations. The closing operation fills the blank areas between the target grids, making the discrete target grids continuous into a piece of area, and integrating the cluster-level outliers into the clustered area; while the opening operation removes the class-level outliers, reducing the noise of the overall image. The order of closing and opening operations is very important because it first ensures the integrity of the target area, and then reduces the noise of the overall image.

[0081] In addition, the expansion and erosion operations of each operation are performed multiple times in succession, where the structural elements of the expansion and erosion are crosses of length 3. Figure 4 As shown, this step greatly improves the discretization and makes the patchy regions composed of the target grid continuous, forming so-called “continents”, each of which corresponds to a visible cluster of the current category.

[0082] In summary, through gridding and morphological methods, the present invention successfully discretizes the continuous data distribution area required for boundary extraction, and fills the blank grids interspersed in the filled grids, thus laying a good foundation for the subsequent steps.

[0083] Identify grid continents according to the filling conditions of the grids in the processing results, construct continental boundaries between grid continents and filter discrete continents based on the identification results, and merge grid continents according to the number of interval grids of the continental boundaries between grid continents to obtain merged continents;

[0084] Further, such as Figure 5 As shown, the method for identifying the grid continent is:

[0085] Determine the filled target grid according to the filling condition of the grid of the processing result, and extract the connected components of the undirected graph according to the distribution condition of the filled target grid;

[0086] The grid continents are identified through the connected components of the undirected graph.

[0087] Furthermore, the method of filtering the discrete continents is:

[0088] A grid road with a grid number less than a preset grid number is regarded as a discrete continent, and the discrete road is filtered.

[0089] In another embodiment, the method of filtering the discrete continents is:

[0090] Obtaining the number of grids of different grid continents, and determining the basic important values ​​of different grid continents based on the number of grids of different grid continents;

[0091] Determine the number of blank grids of different grid continents by the filling conditions of grids of different grid continents, and determine the blank quantity evaluation value of the grid continent in combination with the spacing between different blank grids;

[0092] The number of filled grids of different grid continents is obtained, and the importance evaluation values ​​of different grid continents are determined in combination with the blank quantity evaluation values ​​and the basic importance values ​​of different grid continents, and filtering and identifying processing of discrete continents is performed based on the importance evaluation values.

[0093] It should be noted that the filtering and identification process of discrete continents based on the importance evaluation value specifically includes:

[0094] The grid continents whose importance evaluation values ​​are not within the preset range are treated as discrete continents for filtering and identification.

[0095] In another embodiment, the method for determining the discrete continents is:

[0096] Obtaining the number of grids of different grid continents, and determining the basic importance values ​​of different grid continents based on the number of grids of different grid continents, and judging whether the basic importance value of the grid continent is greater than a preset importance threshold, if so, determining that the grid continent does not belong to a discrete continent, if not, proceeding to the next step;

[0097] Determine whether the number of filled grids of the grid continent meets the requirement, if yes, proceed to the next step, if no, determine that the grid continent belongs to a discrete continent;

[0098] Determine the number of regions containing blank grids in the grid continent according to the filling condition of the grids of the grid continent, and judge whether the number of regions containing blank grids in the grid continent meets the requirement. If yes, proceed to the next step; if no, determine that the grid continent belongs to a discrete continent.

[0099] Determine the number of blank grids of different grid continents by the filling conditions of grids of different grid continents, and determine the blank quantity evaluation value of the grid continent in combination with the spacing between different blank grids, and judge whether the blank quantity evaluation value of the grid continent meets the requirements. If yes, proceed to the next step; if not, determine that the grid continent belongs to a discrete continent;

[0100] The number of filled grids of different grid continents is obtained, and the importance evaluation values ​​of different grid continents are determined in combination with the blank quantity evaluation values ​​and the basic importance values ​​of different grid continents, and filtering and identifying processing of discrete continents is performed based on the importance evaluation values.

[0101] Specifically, the method for merging and processing the continents is:

[0102] Determining the distances between the continental boundaries of the grid continents in different intervals of the number of grid intervals based on the number of grid intervals between the continental boundaries of different grid continents;

[0103] The distance between the continental borders and the number of grid intervals at different intervals are used to evaluate the spacing between the continental borders.

[0104] The comprehensive spacing evaluation amount between the grid continents is determined according to the spacing evaluation amounts of different intervals of the number of grids, and merging processing is performed according to whether the comprehensive spacing evaluation amount needs to be performed.

[0105] Further, depending on whether the comprehensive spacing evaluation amount needs to be merged, it specifically includes:

[0106] When the comprehensive spacing evaluation value is within a preset spacing interval, it is determined that the grid continents need to be merged.

[0107] In another embodiment, the method for merging and processing continents is:

[0108] Determine the minimum number of grid intervals of the continental boundaries between the grid continents based on the number of grid intervals of the continental boundaries between different grid continents, and judge whether the minimum number of grid intervals of the continental boundaries between the grid continents meets the requirement, if so, proceed to the next step, if not, determine that the grid continents do not need to be merged;

[0109] Obtaining the distances of the continental boundaries between the grid continents in different intervals of the number of grid intervals, determining whether the distances of the continental boundaries in the preset intervals of the number of grid intervals are greater than a preset length, if so, determining that the grid continents need to be merged, if not, proceeding to the next step;

[0110] By performing spacing evaluations of different intervals of grid numbers based on the distances of the continental boundaries in different intervals of grid numbers and the numbers of grid numbers of the continental boundaries, it is determined whether the number of intervals of grid numbers within the preset intervals of spacing evaluations meets the requirements. If so, it is determined that the grid continents need to be merged. If not, proceed to the next step.

[0111] The comprehensive spacing evaluation amount between the grid continents is determined according to the spacing evaluation amounts of different intervals of the number of grids, and merging processing is performed according to whether the comprehensive spacing evaluation amount needs to be performed.

[0112] In this step, the present invention needs to identify each continent as an independent operable structure for subsequent calculations. If each target grid is regarded as a node and the adjacent relationship is regarded as the edge between the nodes, then the identification of the continent is equivalent to extracting the connected components of the undirected graph. Figure 6 The schematic diagram of the process of identifying the continent is shown, and the connected components in 6 are the continents identified in the present invention. Once the continents are identified, the present invention will filter out the continents whose number of grids is less than the threshold p3 set by the user, and regard them as the remaining structural outliers.

[0113] To obtain the outer contours of the continents, Figure 7 As shown, the present invention uses the Moore-Neighbor Tracing algorithm, which can ensure that the boundaries of the continents are formed. After this step, the analyst can filter out the continents with fewer grids as needed to ensure the accuracy and reliability of subsequent calculations.

[0114] Merging adjacent continents can reduce visual complexity. The distance between two continents is defined as the number of grids at the closest point. Any pair of continents whose distance is less than the user-defined threshold p4 is merged. The merging method is to fill the blank grids in the interval with the target grid and then re-identify the boundaries of the merged continents. Figure 8 As shown, Figure 8 The boxes in mark two examples.

[0115] The blank grids between the merged continents are filled to obtain processed grids, and the boundary smoothing process is performed based on the processed grid to obtain the boundary processing result of the scatter plot.

[0116] Furthermore, the boundary smoothing process includes two steps: aliasing detection and midpoint connection.

[0117] Specifically, the anti-aliasing detects convexities and concavities formed by a single mesh and removes or fills them.

[0118] It can be understood that the connection midpoint is the midpoint of each mesh edge in the edge, and then the midpoints of the mesh edges are connected using a curve of a preset curve type.

[0119] Specifically, Fig. 9 As shown in the figure, there may be some holes inside the continent, which are composed of non-target grids. When the filtering condition selected in the above step is relative quantity, those small holes can actually be regarded as semantic outliers. However, under the information requirements at the category level, these small holes are not important, but may interfere with the analyst's attention. Therefore, the goal of this step is to remove small holes that occupy a number of grids less than the user-defined threshold p5 and merge them into the surrounding continents. For those very small holes, they only increase visual complexity and should be removed. On the contrary, larger holes may reflect some meaningful data distribution characteristics and should be retained. Users can control the retention of holes by setting a suitable threshold.

[0120] like Fig.10 As shown in the figure, smoothing the boundary includes two steps: anti-aliasing and connecting midpoints. Anti-aliasing detects the convexities and concavities formed by a single mesh and removes or fills them. Connecting midpoints means taking the midpoints of each mesh edge in the edge and then connecting them using Catmullrom curves.

[0121] Based on the above embodiments, the following technical effects are achieved:

[0122] 1. Efficient calculation, suitable for large-scale data: Fig.11 The figure is a schematic diagram comparing the extraction results of the present invention and the existing multi-class scatter plot boundary extraction method. The present invention adopts a grid-based boundary extraction strategy to convert the task into the boundary of the target grid, avoiding the nonlinear time complexity based on the data point scale. This calculation method can greatly improve the efficiency of boundary extraction, especially when processing large-scale data. The present invention implements the proposed method based on Python and conducts experiments on 30 data sets, with the grid division accuracy of 200*200. The results show that the running time of the present invention is 2.1 seconds on average.

[0123] 2. The extracted boundaries are concise and easy to read, and can cover most of the target data points: Compared with traditional technologies, the present invention can more accurately capture the distribution boundaries of the data set. The experimental data results on the 30 data sets tested show that the boundaries extracted by the present invention can cover the target data points with high accuracy (90.2%) and recall rate (94.1%).

[0124] 3. Applicable to any type of large-scale multi-class scatter plot: The present invention has a wide range of applications and is not limited by the algorithm or the specific distribution of scatter points. Therefore, it is applicable to various types of large-scale multi-class scatter plots, providing a highly flexible and universal solution for data analysis and visualization in different fields. Fig.11 The extraction results of the present invention and the existing multi-class scatter plot boundary extraction method are shown. It can be seen that the boundary extracted by the present invention is more concise and smooth, and can tightly wrap most of the data points.

[0125] In the description of this specification, the description of the terms "one embodiment", "a preferred embodiment", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the embodiment of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0126] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A class-level abstractable method for multi-class scatter plots based on a grid, characterized in that: Specifically include: The entire scatter plot distribution space is divided into equidistant grids, the grids are screened according to preset rules to obtain target grids, the target grids are filled, and the blank grids between the target grids are filled by performing multiple closing operations and opening operations to obtain processing results; Identify grid continents according to the filling conditions of the grids in the processing results, construct continental boundaries between grid continents and filter discrete continents based on the identification results, and merge grid continents according to the number of interval grids of the continental boundaries between grid continents to obtain merged continents; The blank grids between the merged continents are filled to obtain processed grids, and the boundary smoothing process is performed based on the processed grid to obtain the boundary processing result of the scatter plot.

2. A grid-based multi-class scatter plot class-level abstractable method as claimed in claim 1, characterized in that: The preset rules include any one or more of the number of data points in the grid and the proportion of the target category in the grid.

3. A grid-based multi-class scatter plot class-level abstractable method as claimed in claim 1, characterized in that: The target grid is obtained by screening the grid according to the preset rules, including: Determining a screening target of the grid based on the preset rule, constructing a filtering function based on the screening target, and screening outliers through the filtering function; The outliers are filtered, and the target grid is determined according to the filtering results.

4. A grid-based multi-class scatter plot class-level abstractable method as claimed in claim 1, characterized in that: The target grid is obtained by screening the grid according to the preset rules, including: Determine, based on the preset rule, matching coefficients of different grids in different dimensions with the preset rule; The comprehensive matching coefficient between the grid and the preset rule is determined by the matching coefficient, and the target grid is determined based on the comprehensive matching coefficient.

5. A grid-based multi-class scatter plot class-level abstractable method as claimed in claim 1, characterized in that: Filling the blank grids between the target grids to obtain a processing result, specifically including: Color the target grid, perform a closing operation, fill the blank areas between the target grids, make the discrete target grids continuous into a piece of area, and integrate the cluster-level outliers into the clustered area to the execution result of the closing operation; Performing an open operation on the execution result of the closing operation, removing outliers in the execution result of the closing operation to obtain the execution result of the opening operation; The open operation and the close operation are repeatedly performed multiple times to fill the blank grids between the target grids to obtain a processing result.

6. A grid-based multi-class scatter plot class-level abstractable method as claimed in claim 1, characterized in that: The method for identifying the grid continent is: Determine the filled target grid according to the filling condition of the grid of the processing result, and extract the connected components of the undirected graph according to the distribution condition of the filled target grid; The grid continents are identified through the connected components of the undirected graph.

7. A grid-based multi-class scatter plot class-level abstractable method as claimed in claim 1, characterized in that: The method for filtering the discrete continents is: A grid continent with a grid number less than a preset grid number is regarded as a discrete continent, and the discrete continent is filtered.

8. A grid-based multi-class scatter plot class-level abstractable method as claimed in claim 1, characterized in that: The method for determining the discrete continents is: Obtaining the number of grids of different grid continents, and determining the basic importance values ​​of different grid continents based on the number of grids of different grid continents, and judging whether the basic importance value of the grid continent is greater than a preset importance threshold, if so, determining that the grid continent does not belong to a discrete continent, if not, proceeding to the next step; Determine whether the number of filled grids of the grid continent meets the requirement, if yes, proceed to the next step, if no, determine that the grid continent belongs to a discrete continent; Determine the number of regions containing blank grids in the grid continent according to the filling condition of the grids of the grid continent, and judge whether the number of regions containing blank grids in the grid continent meets the requirement. If yes, proceed to the next step; if no, determine that the grid continent belongs to a discrete continent. Determine the number of blank grids of different grid continents by the filling conditions of grids of different grid continents, and determine the blank quantity evaluation value of the grid continent in combination with the spacing between different blank grids, and judge whether the blank quantity evaluation value of the grid continent meets the requirements. If yes, proceed to the next step; if not, determine that the grid continent belongs to a discrete continent; The number of filled grids of different grid continents is obtained, and the importance evaluation values ​​of different grid continents are determined in combination with the blank quantity evaluation values ​​and the basic importance values ​​of different grid continents, and filtering and identifying processing of discrete continents is performed based on the importance evaluation values.

9. A grid-based multi-class scatter plot class-level abstractable method as claimed in claim 1, characterized in that: The boundary smoothing process includes two steps: aliasing detection and midpoint connection.

10. A grid-based multi-class scatter plot class-level abstractable method as claimed in claim 9, characterized in that: The anti-aliasing detects the bumps and valleys formed by a single mesh and removes or fills them.

Images