Improved hierarchical high-load expression method for interest points of big data map

Abstract

According to the improved hierarchical high-load expression method for the interest points of the big data map, on the basis of regular grid thinning, the thinning method is improved, and the thinning method suitable for the interest points of the common government affair digital map is established; according to the method, different numbers of grids need to be traversed again and points in the grids need to be indexed for rarefying under each performance level in a regular grid, an algorithm is improved for the efficiency problem, and screen points are adopted as parameter factors for the point spacing in the regular grid and the pattern density consistency between the performance levels, so that the efficiency of rarefying in the regular grid is improved. The algorithm is optimized according to the condition that one point is selected by a unit grid, and thinning is carried out by establishing a quaternary discrete class method. According to the method, the algorithm efficiency is sufficient when the interest points are thinned under the condition of large data volume, the extracted interest point sets have high importance levels, the distribution of the extracted interest point sets is relatively uniform, capping and over-density are avoided, and large-data volume vector data application and multi-scale display are good in effect, high in speed and comprehensive in expression.

Description

Technical Field

[0001] The present application relates to a method for hierarchical expression of points of interest in a big data map, and particularly to an improved hierarchical high-load expression method for points of interest in a big data map, belonging to the technical field of map point of interest expression. Background Art

[0002] With the rapid development of computer technology, the production and use modes of traditional maps have been greatly impacted. The Geographic Information System (GIS) emerged and developed rapidly. Network map services quickly entered people's daily lives through various means such as smartphone clients and personal computer terminals. Users use network maps represented by Google, Baidu, and Amap to browse, query, and obtain more information related to travel, tourism, etc.

[0003] The reason why digital maps have better flexibility and stronger interactivity compared to ordinary maps is that their multi-scale expression of geographical elements is an important factor. In digital maps, users can view maps at different scales by zooming in and out of the scale, and can more completely learn about the true situation of the geographical space, more detailedly understand the distribution locations of geographical entities, and more accurately judge the development laws of geographical phenomena. Under the requirement of multi-scale display, the elements displayed on maps of different scales are different. Hierarchical classification of geographical elements is an important method to achieve multi-scale expression of digital maps.

[0004] Points of interest are important expression contents in digital maps, including basic information such as landmark buildings, educational institutions, medical facilities, culture and sports, and entertainment venues. In the big data era, the sources of geographical information data are very extensive. Under the support of the volunteered geographic information system model, it is even possible to create point of interest data by oneself. Therefore, points of interest not only have a wide variety of types but also a very large amount of data. Points of interest are an abstract expression of geographical space entities, with accurate spatial positions and rich attribute information. Users can clearly obtain the distribution status and information of spatial entities. In the process of digital map mapping, the visualization effect and data loading and display speed of points of interest are the basis for users to continue using digital maps for data processing and spatial analysis.

[0005] There are a large number of point of interest data within the spatial range expressed by digital maps. The display speed and effect of the data are not only related to the performance of hardware devices, but also the spatial organization and preprocessing of massive data, as well as the access and drawing of symbols, which will also have an important impact on the display of points of interest. Therefore, how to ensure that digital map points of interest have a reasonable effect and a relatively fast speed during display is an important topic. The point element point of interest classification method in digital map data can solve this problem.

[0006] The current method for displaying large-scale network digital maps is to establish a multi-scale spatial database (MRDB). According to the display scale, manual or algorithm-assisted pre-synthesis processing is carried out on the point-of-interest (POI) data at each scale. Since the number of levels of digital map display scales is usually 20 levels, this method consumes a lot of time. Due to the need for multi-scale display of massive POI data, this application creates a spatial index for the POI data and batch-processes the POI data in a specific way to achieve automatic grading of POIs. This method can improve the efficiency of POI grading under a large amount of data to a certain extent, reduce a large amount of resource costs in the process of digital map data production, and improve the utilization efficiency of POI data. At the same time, using the POI weight as an important grading factor also ensures the effect of multi-scale display of digital map POIs, which has practical significance for the application and multi-scale display of massive vector data.

[0007] 1.3 Research Objectives and Contents

[0008] The display effect of digital map POIs directly determines the map quality. The method of multi-scale display of POIs adopted in the production process of digital map POI data not only concerns the map quality, but also relates to the map production cycle and production cost. To improve the efficiency of multi-scale display of POI data, the method of data hierarchical display is mostly commonly used now. Through the research of this application, the following objectives are expected to be achieved:

[0009] 1) Combining the expression characteristics of POI data, designing an algorithm for adding grading to the POI data itself to achieve an automatic grading scheme for POIs;

[0010] 2) Applying the automatic grading method of POIs to the map compilation platform to solve the problem of consuming a large amount of resources in the production process of multi-scale POI mapping data and improve the data usage efficiency.

[0011] Usually, the types of POIs displayed on maps at large scales are rich and the quantity is large, while only some important POIs are displayed at smaller scales. To ensure that the distribution characteristics and attribute information of POIs can be fully expressed at each scale, this application takes POI data as the research object and focuses on the following research contents with digital maps as the expression platform:

[0012] 1) The characteristics and trends of multi-scale expression of digital maps in the current situation of rapid development of the Internet;

[0013] 2) Summarizing the classification and characteristics of digital map POIs, and analyzing the requirements and practical problems of multi-scale expression of POIs in digital maps;

[0014] 3) Analyzing the principles of POI grading methods, comparing existing different grading methods, and summarizing necessary grading rules;

[0015] 4) Focus on studying the efficiency and effectiveness of existing classification algorithms, propose a classification algorithm with an optimization plan, and conduct relevant experiments for verification.

[0016] The problems to be solved by the existing method for hierarchical expression of map points of interest and the key technical difficulties of this application include:

[0017] (1) The large-scale network digital map display method of the existing technology is to adopt the method of establishing a multi-scale spatial database MRDB. Since the number of levels of the digital map display scale is usually 20 levels, adopting this method takes a lot of time and cannot meet the need for multi-scale display of a large amount of point-of-interest data. There is a lack of establishing a spatial index for point-of-interest data, batch processing of point-of-interest data in a specific way, and the automatic classification of points of interest cannot be realized. The classification efficiency of points of interest under a large amount of data is low, and a large amount of resource costs are consumed in the process of making digital map data. The utilization efficiency of point-of-interest data is low; at the same time, there is a lack of using the point-of-interest weight as an important classification factor to ensure the multi-scale display effect of digital map points of interest. The application of a large amount of vector data and the multi-scale display effect are poor, the speed is slow, and the expression is not comprehensive enough.

[0018] (2) At present, under the big data environment, the point-of-interest data is updated rapidly and the sources are rich. The challenges faced in the process of making digital map data are even greater. The existing technology cannot ensure the rapid selection of point-of-interest data that should be expressed at multiple scales, resulting in the inability to efficiently and reasonably display the basic information of geographical entities represented by points of interest in the digital map, such as landmarks, tourism, and catering. There is a lack of establishing a multi-scale expression method for digital maps from the fundamental source of classification requirements, a lack of an algorithm for thinning points of interest by improving the design specification grid, the quadtree is not used to divide the mapping area into grids to establish a quadtree index of points of interest, and a classification method class is not developed using an object-oriented programming language. The existing technology has a low classification efficiency and cannot ensure the requirements of the map load in terms of the classification display effect. The rationality of the points of interest displayed at different scales is also poor, and the point-of-interest classification method cannot well support the rapid digital map classification display and the production of multi-scale databases.

[0019] (3) In the regular grid method, the method of dividing the grid is to fix the grid size at 2 cm * 2 cm, and calculate the number of grids at each level through the scale of each performance level. The setting of the grid size directly determines the number of grids. Moreover, due to the different resolutions of different screens, there are also differences in the actual corresponding distances for a grid size of 2 cm. In addition, the user needs to determine the maximum allowable number of interest points within the retained unit grid, and this parameter directly determines the number of points retained after thinning. These parameters are difficult for users to directly understand, and the filled-in values have a significant impact on the results. Moreover, before selecting points within the unit grid, it is necessary to determine whether the grid contains interest points of the previous level. Therefore, too many points increase the algorithm complexity and it is impossible to obtain the best hierarchical result in actual operation. Summary of the Invention

[0020] Based on the thinning of regular grids, this application improves the thinning method to establish a thinning method applicable to the interest points of ordinary government digital maps. Among them, for the thinning at each performance level in the regular grid, it is necessary to re-traverse different numbers of grids and index the points inside the grids, which has an efficiency disadvantage. To address the efficiency problem, the algorithm is improved. Regarding the point spacing in the regular grid and the consistency of the map surface density between each performance level, the number of screen points is used as a parameter factor, and the case of selecting one point in the unit grid is used to optimize the algorithm, so as to establish a method of quaternary discretization for thinning. When facing the thinning of interest points under a large amount of data, the algorithm has sufficient efficiency. The extracted interest point set has a high importance level, and the distribution of the extracted interest point set is relatively uniform, avoiding capping, overcrowding. The application of large amount of vector data in multi-scale display has good effect, high speed and comprehensive expression.

[0021] To achieve the above technical effects, the technical solutions adopted in this application are as follows:

[0022] An improved hierarchical high-load expression method for interest points in a big data map. Based on the classification of interest points in a digital map, first establish a multi-scale expression method for a digital map from the source of classification requirements, improve the interest point thinning algorithm for the designed regular grid, use a quadtree to divide the mapping area into grids to establish a quadtree index for interest points, develop a classification method class using an object-oriented programming language, and establish an improved high-load algorithm for the classification of interest points in a big data map to address the deficiencies in the regular grid thinning method:

[0023] 1) Method of spatial index: Based on the GIS quadtree index method, the range of the distribution of points of interest is divided into a full quadtree grid. The quantitative relationship of the grids at each level is fixed. Each node represents each grid, and the depth of the tree represents the corresponding display level. For each grid and its contained grids, the tree structure is used to find the child nodes and parent nodes of the node for judgment. The variables at different levels are converted into the maximum number of points allowed in the grids at the current level. When indexing points, it is judged whether the points of interest selected at the previous level are contained in the current grid, and directly search whether the point of interest is included in the selected queue of its parent node;

[0024] 2) Method for determining the number of points to be thinned at each display level: In the thinning method of regular grids, the number of points to be thinned at each display level is calculated by multiplying the number of grids at that level by the maximum number of points allowed in a unit grid set by the user. A method for determining the changed number of points is established. The input parameter provided to the user is the number of points of interest displayed within the display window at the display level with the largest scale. Based on the parameter, the range of the current display window, and the scales at all levels, the maximum number of points of interest allowed for thinning at each level is calculated. Divide the maximum number of points by the number of nodes in the quadtree at each level to obtain the maximum number of points of interest allowed for each unit node, which meets the requirements for the number of points of interest at large scales. At small scales, the maximum number of points of interest allowed for unit nodes is set manually.

[0025] Preferably, the rules for optimizing the thinning of points of interest are formulated:

[0026] 1) The algorithm efficiency is sufficient when thinning points of interest in the face of a large amount of data;

[0027] 2) The extracted set of points of interest has a high importance level;

[0028] 3) The distribution of the extracted set of points of interest is relatively uniform, avoiding overlapping and overcrowding.

[0029] Based on these three rules for optimizing points of interest, this application is based on the thinning of regular grids, improves the thinning method, and establishes a thinning method applicable to points of interest in ordinary government digital maps. Among them, based on the regular grid, for each display level during thinning, it is necessary to re-traverse grids with different numbers and index the points inside the grids. The algorithm is improved for efficiency issues; based on the point spacing in the regular grid and the consistency of the map density between each display level, taking the number of points on the screen as a parameter factor and the case of selecting one point in a unit grid to optimize the algorithm, and establishing a method of quaternary discretization to perform thinning.

[0030] Preferably, the interest point thinning architecture: Optimize based on the regular grid thinning method, divide the distribution range of interest points into regular grids, divide the area into blocks, the area where each interest point exists is clearly defined, establish a regular grid index for interest points, mark the interest points with the grid number when extracting and processing interest points, and reduce the traversal of unnecessary interest points; when screening interest points, use the grid as the basic unit, and perform a screening process for a single grid based on the importance of the interest point and the point distance factor of the distribution. In the whole process of grading interest points, the smallest processing unit is the smallest grid unit at each display scale, reducing the processing spatial range and the number of objects.

[0031] Preferably, grid thinning based on full quaternary discretization: Store the geospatial information in the quadtree nodes to implement the quadtree index of spatial data. Divide the determined spatial area into four equal sub-spatial areas. Each sub-space in the tree structure corresponds to a tree node. Recursively divide each sub-space into four sub-spaces again, and so on until the tree level reaches a certain depth or meets specific conditions. All the elements within the spatial range belong to a certain leaf node within a smallest area. When inserting the elements into the quadtree, calculate the node to which it belongs using the geometric attributes of the elements.

[0032] Preferably, when thinning interest point data with a wide spatial range and a large amount of data, considering the uniform distribution and processing efficiency of interest points in the entire mapping area, rasterizing the entire mapping area is the optimal solution. Combining the fact that there are multiple relationships between standard scales, using quaternary discretization, gradually divide the mapping area. Each level of the tree represents the grid situation divided at each level of the display scale; First, obtain the minimum bounding rectangle MBR of the point group, the rectangle range Xmax, Xmin, Ymax, Ymin; Establish a full quadtree with the corresponding depth according to the number n of display scale levels; Set the counterclockwise area sequence code as 1, 2, 3, 4; The root node represents the maximum bounding rectangle of the interest point distribution in the mapping area. Each node has four child nodes. The four child nodes of the root node correspond to the four areas of 0, 1, 2, 3, upper right, upper left, lower left, and lower right respectively. Each node extends until the maximum depth of the tree, that is, the maximum display scale level.

[0033] The cartographic area is hierarchically grid-divided by establishing a full quadtree. All points of interest fall into the divided grids. The grids at each level to which the points of interest belong are recursively calculated through the coordinates of the points of interest, that is, the node information to which the points of interest belong can be obtained. When inserting each point of interest, the relationship between the data of the point of interest and the node is calculated. When inserting a point, calculate the area X - Xmin, Y - Ymin and its encoding of the area to which it belongs until the depth reaches the maximum value encoding {0, 2, 1, 3, 4...}, and save the point in the leaf node to obtain the quadtree index of the point of interest.

[0034] Preferably, the points of interest are screened according to rules. In the regular quaternary discrete class grid, each display scale level corresponds to one layer of the tree. The process of grading the points of interest is completed by traversing the tree once, which has higher efficiency compared to the ordinary thinning method that needs to loop through completely according to the number of grids when thinning each display scale level.

[0035] When thinning the second display level, traverse to the node at depth 1 of the tree. Use the node as the processing unit to screen all the points of interest within the spatial area represented by the node. The screening rules are as follows: Judge the points of interest. If there are points of interest with an automatic grading attribute less than or equal to the current level, these points are the points selected in the previous level. According to the grading rules, the points selected in the small-scale are still selected in the larger scale, and the points are still selected. If the number of graded points is not enough for the number of points selected in the grid, select points according to the following method. If the node does not contain points of interest, sort all the points of interest in the node according to the importance weight, form a point list in descending order, and select the first point of interest in the point list for the assignment of the display level value; if there are points with the same weight during sorting, use the geometric position of the point of interest as the basis to calculate the distance between the point of interest and the center of the node area, and give priority to the point with a smaller distance to ensure that the distance between the grids is relatively uniform. Keep looping and selecting until the number of selected points reaches the maximum allowable number of points for the node at this depth.

[0036] Preferably, the grids of each level of scale are divided according to the full quaternary discrete class, and the grid relationship between levels is fixed, which is only applicable to the case where there is an even multiple relationship with the display scale. This solves the problem of rapid indexing of points under the grids of each region and avoids repeated calculation of indexing when thinning multiple levels. In addition, when using the node area as the processing unit to calculate the point spacing and screen the optimal position, it ensures that the relative density of the entire cartographic area is uniform. The maximum number of selected points for the node at each depth is obtained by taking the average value according to the maximum allowable number of points of interest at the level. The calculation formula is as follows:

[0037]

[0038] The area S of the representative region of each level of the screen 屏Actual area S, maximum number of points m allowed on the screen, current display level n, where the maximum number of points m is determined by user-set parameters.

[0039] Preferably, the quaternary discrete grid thinning process:

[0040] Step 1: Load the vector point feature point of interest data at the most detailed scale in the format of Shp or Gdb in the system;

[0041] Step 2: Determine to execute the point of interest grading process, require the user to input parameters, the scale of each display level determines the depth of the quadtree and the relationship of the parent-child node layers between the scales of each level; the screen displays the number of points of interest that can be adjusted to calculate the maximum number of points of interest allowed within the unit grid at each level, with a default of 30; the minimum distance on the map ensures that the display effect of the points of interest at each level scale meets the human visual resolution ability;

[0042] Step 3: Read the number and relationship of the display scales set by the user, read the minimum bounding rectangle calculated from the loaded points of interest, and create a full quadtree with the specified depth;

[0043] Step 4: Insert the read points of interest in a loop, calculate the spatial index of each point of interest, and thus obtain the information of the nodes at each level where each point of interest is located;

[0044] Step 5: Traverse each node starting from the root node according to the tree structure, screen the points of interest within the node unit, and the screening condition is to first judge whether there are points of interest with a grading value less than the depth in the node. If so, select and judge whether the maximum allowed number of points is reached. Otherwise, select points according to the sorting of the importance weights of the points of interest. The selected points are all judged for the minimum distance from the previously selected points, and the selected points are automatically assigned grading attributes;

[0045] Step 6: After traversing all nodes, the attributes of the points of interest include the automatically calculated grading value. Write the attributes into the attribute table of the original Shp or Gdb file to complete the grading of the points of interest.

[0046] Preferably, the algorithm class is established as follows: Design the QTPOIntLayer C++ class to implement the algorithm module for interest point grading. For the interest point feature structure Feature, it includes the importance weight attribute, the ID of the interest point, the automatic grading attribute for saving the grading result, and the original grading attribute value. For the grid object, an abstract design structure MapRect is defined, which includes the range of the grid, and two methods: Contains to determine whether an interest point is inside the grid, and the Spilt method to perform a four-way split of the grid. The direct object structure quadtree_t includes nodes QuadNode, where a node is a grid. A node includes a depth attribute, its four child nodes, the grid MapRect represented by the node, and the interest points contained in the node. In the class, the denominator of the display scale parameter is stored in an integer array, and the number of levels for building the quadtree is obtained through this array. The attribute Limit is the maximum number of points allowed for each node, with a default value of 1. Other methods include the function CreatQuadTree to create a full quadtree with a depth corresponding to the number of display scale levels, the method InsertQuad to insert interest points to build the quadtree index, and finally the function TravelQuadTree to traverse the quadtree to implement the filtering and thinning process to complete the assignment of the automatic grading attribute AutoLevel of Feature.

[0047] Compared with the prior art, the innovation points and advantages of this application are as follows:

[0048] (1) This application ensures that the distribution characteristics and attribute information of interest points can be fully expressed at each scale. Taking interest point data as the object and using digital maps as the expression platform, it summarizes the classification and characteristics of interest points in digital maps, analyzes the requirements and practical problems of multi-scale expression of interest points in digital maps, establishes necessary grading rules, proposes a grading algorithm with an optimized scheme, and designs an algorithm for adding grading to the interest point data itself in combination with the expression characteristics of interest point data, realizing the automatic grading scheme for interest points. Applying the automatic grading method of interest points to the map compilation platform solves the problem of consuming a large amount of resources in the production process of multi-scale interest point mapping data and improves the data usage efficiency.

[0049] (2) For the current large-scale network digital map display method, it is to adopt the method of establishing a multi-scale spatial database MRDB. According to the scale of the display scale, the data of points of interest at each scale are pre-synthesized manually or with the assistance of algorithms. Since the number of levels of the digital map display scale is usually 20 levels, this method takes a lot of time. Due to the need for multi-scale display of a large amount of point-of-interest data, this application builds a spatial index for the point-of-interest data and batch-processes the point-of-interest data in a specific way, realizing the automatic classification of points of interest. This method can improve the classification efficiency of points of interest under a large amount of data to a certain extent, reduce a large amount of resource costs in the process of digital map data production, and improve the utilization efficiency of point-of-interest data. At the same time, taking the point-of-interest weight as an important classification factor also ensures the effect of multi-scale display of digital map points of interest, which has practical significance and role for the application and multi-scale display of a large amount of vector data.

[0050] (3) Based on the rule grid thinning, this application improves the thinning method to establish a thinning method applicable to the points of interest of ordinary government digital maps. For the rule grid, re-traversing different numbers of grids and indexing the points inside the grids are required for thinning at each display level, which has an efficiency disadvantage. To address the efficiency problem, the algorithm is improved. Considering the point spacing in the rule grid and the consistency of the map density between each display level, the number of screen points is taken as a parameter factor, and the case of selecting one point in the unit grid is used to optimize the algorithm, and a quaternary discrete class method is established for thinning. When thinning the points of interest under a large amount of data, the algorithm has sufficient efficiency. The extracted set of points of interest has a high importance level, and the distribution of the extracted set of points of interest is relatively uniform, avoiding capping, overcrowding. The application of a large amount of vector data in multi-scale display has a good effect, high speed, and comprehensive expression. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic diagram of the number of points allowed to be displayed on the screen at the maximum scale.

[0052] Figure 2 It is a schematic diagram of the nodes of a full quadtree and its corresponding divided grid.

[0053] Figure 3 It is a schematic diagram of the screening rules within the node grid.

[0054] Figure 4 It is a schematic diagram of the logical process of screening points of interest within each node.

[0055] Figure 5 It is a schematic diagram of the class structure of the full quadtree division.

[0056] Figure 6 It is a schematic diagram of the distribution of unclassified points of interest at 1:577791.

[0057] Figure 7 It is a schematic diagram of the distribution of points of interest after the 1:577791 regular grid classification.

[0058] Figure 8 It is a schematic diagram of the distribution of points of interest after the 1:577791 full quadtree classification.

[0059] Figure 9 It is a specific numerical comparison chart of the number of points of interest displayed at each performance level. Detailed implementation manners

[0060] The following further describes the technical solution of the method for improving the hierarchical high-load expression of points of interest in the big data map provided by this application with reference to the accompanying drawings, so that those skilled in the art can better understand this application and can implement it.

[0061] Internet digital maps have become essential demand products in people's daily lives. In digital maps, geographical entities represented by points of interest, such as landmarks, tourism, and dining, are the most concerned basic information. The efficient and reasonable display of points of interest is a necessary condition for the usability of digital maps. At present, in the big data environment, the data of points of interest is updated rapidly and the sources are rich, and the challenges faced in the process of digital map data production are even greater. How to ensure the rapid selection of points of interest data that should be expressed at multiple scales is a hot issue.

[0062] Based on the classification of points of interest in digital maps, this application first establishes a multi-scale expression method for digital maps from the fundamental source of classification requirements, improves the thinning algorithm for points of interest in the design specification grid, uses the quadtree to divide the mapping area into grids to establish a quadtree index for points of interest, develops a classification method class using an object-oriented programming language, and verifies the practicality of the improved algorithm through a classification experiment on the data of points of interest in the Beijing government version digital map.

[0063] Experiments show that the improved method has a significant improvement in classification efficiency. In terms of the classification display effect, it not only meets the requirements of the map load, but also has better rationality for the points of interest displayed at different scales. The improved method for classifying points of interest has good practical value in the rapid classification display of digital maps and the production of multi-scale databases.

[0064] I. Formulating the optimization of the point of interest thinning rule

[0065] 1) The algorithm efficiency is sufficient when thinning points of interest under a large amount of data;

[0066] 2) The extracted set of points of interest has a high importance level;

[0067] 3) The distribution of the extracted set of points of interest is relatively uniform, avoiding capping and overcrowding.

[0068] This application is optimized based on these three rules of points of interest. Based on the thinning of regular grids, the thinning method is improved to establish a thinning method applicable to points of interest in ordinary government digital maps. For the thinning at each display level in the regular grid, it is necessary to traverse different numbers of grids and index the points inside the grids again, which has an efficiency disadvantage. To address the efficiency issue, the algorithm is improved. Regarding the point spacing in the regular grid and the consistency of the map density between different display levels, the number of screen points is used as a parameter factor, and the case of selecting one point in the unit grid is used to optimize the algorithm, so as to establish a method of quaternary discretization for thinning.

[0069] II. POI Thinning Architecture

[0070] For the large amount of point-of-interest data in digital maps and the wide distribution space range, it is very challenging to directly thin the entire point group in the mapping area to ensure uniform density and distribution and to consider the importance difference of points of interest. Because when indexing, comparing, and screening hundreds of thousands or millions of point-of-interest data, the algorithm efficiency is often poor. The ability of the automatic grading operation efficiency to be acceptable to users during the actual grading process is an essential condition for the algorithm. The regular grid algorithm is a method for processing such large-scale and large-volume data, which divides the mapping area into regular grid cells. Screening processing is performed within each cell. When the unit grid is divided finely enough, the non-uniform distribution of points of interest inside the unit has little impact on the entire mapping area from a macroscopic perspective. Therefore, the density accuracy of the entire map surface is guaranteed.

[0071] This application is optimized based on the regular grid thinning method. The distribution range of points of interest is divided into regular grids, and the area is divided into blocks. The area where each point of interest exists is clearly defined, and a regular grid index is established for points of interest. When extracting and processing points of interest, the points of interest are marked with the grid numbers, reducing the traversal of unnecessary points of interest and improving the efficient management of a large number of points of interest. When screening points of interest, the grid is used as the basic unit, and a screening process is performed on a single grid based on the importance of the point of interest and the point distance factor of the distribution. During the entire process of grading points of interest, the smallest processing unit is the smallest grid unit at each display scale, reducing the processing space range and the number of objects, and accelerating the algorithm execution efficiency.

[0072] III. Improved Grading Expression of Points of Interest

[0073] In the regular grid method, the method of dividing the grid is to fix the grid size at 2cm * 2cm, and calculate the number of grids at each level through the scale of each performance level. The setting of the grid size directly determines the number of grids. Moreover, due to the different resolutions of different screens, there are also differences in the actual corresponding distances for a grid size of 2cm. In addition, the user needs to determine the maximum allowable number of interest points within the retained unit grid, and this parameter directly determines the number of points retained after thinning. These parameters are difficult for users to directly understand, and the filled values have a significant impact on the results. Moreover, before selecting points within the unit grid, it is necessary to determine whether the grid contains interest points of the previous level. Therefore, too many points increase the algorithm complexity, and the best classification result can only be obtained through multiple comparison experiments by running the algorithm.

[0074] This application proposes a high-load improvement algorithm for interest point classification of big data maps aiming at two deficiencies in the regular grid thinning method:

[0075] 1) Spatial indexing method: The indexing method of interest points in the regular grid is through the row and column numbers of the grid where the mark is located. To ensure the rationality of this method, the number of grids is also very large at large scales, and the number of grids between different levels is different. Each interest point has an index number with little correlation at different levels. When screening interest points, points are also selected according to the grid units at each level. Since the number of grids at different levels is different, the grids where each point is located at different levels do not necessarily completely contain the corresponding ones. Under the rule that the next level must contain the ones already selected at the previous level, the density corresponding to each area is prone to be relatively uneven. Moreover, to determine a grid index for each performance level, screening points at each level requires re-traversing all the grids at that level, and the number of loops is very large, resulting in low efficiency. This application is based on the GIS quadtree indexing method, and divides the grid of the range where the interest points are distributed into a full quadtree. The quantitative relationship of the grids at each level is certain. Each node represents each grid, and the depth of the tree represents the corresponding performance level. For each grid and the grids it contains, the tree structure is used to find the child nodes and parent nodes of this node for judgment. The variables at different levels are transformed into the maximum number of points allowed in the grid at the current level. When indexing points and determining whether the current grid contains interest points already selected at the previous level (directly searching whether the interest point is included in the selected queue of its parent node), the efficiency is improved.

[0076] 2) Method for determining the screening points of each performance level: In the thinning method of regular grids, the number of points to be thinned at each performance level is calculated by multiplying the number of grids at that level by the maximum allowable number of points per unit grid set by the user. It is difficult for users to fill in parameters. According to the common expression of points of interest at different scales of actual digital maps and the fact that the quantity of digital maps for different purposes is different, this application establishes a method for changing and determining the number of points. The input parameter provided to the user is the number of points of interest displayed within the display window at the performance level with the largest scale, such as Figure 1 , calculate the maximum allowable number of points of interest for thinning at each level through the parameter, the range of the current display window, and the scales at all levels. Divide the maximum number of points by the number of nodes in the quadtree at each level to obtain the maximum allowable number of points of interest for each unit node. In this case, it meets the requirements for the number of points of interest at large scales. However, the screening is often skilled at small scales. Therefore, the maximum allowable number of points of interest for unit nodes can be manually assisted and set at small scales.

[0077] IV. Grid Thinning Based on Full Quaternion Discrete Classes

[0078] Store the geospatial information in the quadtree nodes to implement the quadtree index of spatial data. Divide the spatial area within the determined range into four equal sub-spatial areas. Each sub-space corresponds to a node in the tree structure. Recursively divide each sub-space into four sub-spaces again, and so on until the depth of the tree reaches a certain level or meets specific conditions. All the elements within the spatial range are within a minimum area, that is, they belong to a certain leaf node. When inserting the elements into the quadtree, calculate the node to which they belong using the geometric attributes of the elements.

[0079] When thinning the points of interest data with a wide spatial range and a large amount of data, considering the uniform distribution of points of interest and the processing efficiency in the entire mapping area, rasterizing the entire mapping area is the optimal solution. Combining the fact that there are multiple relationships between standard scales, using the quaternion discrete class, gradually divide the mapping area. Each level of the tree represents the grid situation divided at the display scales of each level. As Figure 2 shown, first obtain the minimum bounding rectangle MBR of the point group, the rectangle range Xmax, Xmin, Ymax, Ymin; establish a full quadtree with the corresponding depth according to the number of display scale levels n; set the counterclockwise area sequence code as 1, 2, 3, 4; the root node represents the maximum bounding rectangle of the distribution of points of interest in the mapping area. Each node has four child nodes. The four child nodes of the root node correspond to the four regions of 0, 1, 2, 3, upper right, upper left, lower left, and lower right respectively. Each node extends until the maximum depth of the tree, that is, the largest display scale level.

[0080] By establishing a full quadtree, the mapping area is divided into hierarchical grids, and all points of interest fall into the divided grids. The coordinates of the interest points are recursively calculated to obtain the grids of each level to which the interest points belong, that is, the node information to which the interest points belong can be obtained. When each point of interest is inserted, the relationship between the point of interest data and the node is calculated. The inserted point calculates its area X-Xmin, Y-Ymin and its encoding until the depth reaches the maximum value encoding {0, 2, 1, 3, 4...}, and the point is saved in the leaf node to obtain the quadtree index of the point of interest.

[0081] Filter interest points according to the rules. In the regular four-element discrete grid, each display scale level corresponds to a layer of the tree. The process of grading interest points is completed by traversing the tree once. Compared with the ordinary thinning method, it has higher efficiency when thinning each display scale level by looping according to the number of grids.

[0082] When thinning the second representation level, traverse to the node with depth 1 of the tree, and use the node as the processing unit to filter all the points of interest within the spatial area represented by the node. The filtering rules are as follows: Figure 3 : Judge the points of interest. If there are points of interest with automatic classification attributes that are less than or equal to the current level, these points are the points selected in the previous level. According to the classification rules, the points selected and displayed in the small scale are still selected in the larger scale. The points are still selected. If the classified points are not enough for the number of grid selection points, select points according to the following method. If the node does not contain any points of interest, all the points of interest in the node are sorted according to the importance weight, and a point column is formed in descending order. The first point of interest in the point column is selected to assign the performance level value; if there are points with the same weight in the sorting, the geometric position of the point of interest is used as the basis for calculating the distance between the point of interest and the center of the node area. The point with a smaller distance is selected first to ensure that the distance between the grids is relatively uniform. The selection is cyclic until the number of selected points reaches the maximum allowed number of points for the depth node.

[0083] According to the full quaternion discrete class, the grids of each level are divided into grids, and the grid relationship between each level is fixed. It is only applicable to the case where there is an even-number multiple relationship with the display scale, so as to solve the problem of fast indexing of points under the grids of each area, avoid repeated calculation of indexes when thinning multiple levels, and speed up the efficiency of traversal and screening. In addition, the node area is used as the processing unit to calculate the point spacing and the optimal screening position to ensure the relative uniformity of the entire mapping area. The maximum number of selected nodes at each depth is obtained by taking the average value of the maximum number of interest points allowed by the level. The calculation formula is as follows:

[0084]

[0085] Each level of screen represents the area S 屏The actual area S, the maximum number of points m allowed on the screen, and the current display level n, where the maximum number of points m is determined by user-set parameters.

[0086] V. The thinning process of the full quaternary discrete grid

[0087] Step 1: Load the vector point feature point of interest data at the most detailed scale in the format of Shp or Gdb in the system;

[0088] Step 2: Determine to execute the point of interest grading process, require the user to input parameters, the scale of each display level determines the depth of the quadtree and the relationship of the number of parent-child node layers between the scales of each level; the screen displays the number of points of interest and can adjust to calculate the maximum number of points of interest allowed within the unit grid (each node) at each level, with the default being 30; the minimum distance on the map ensures that the display effect of the points of interest at each level scale meets the human visual resolution ability;

[0089] Step 3: Read the number and relationship of the display scales set by the user, read the loaded points of interest to calculate the minimum bounding rectangle, and create a full quadtree with the specified depth;

[0090] Step 4: Loop to insert the read points of interest, calculate the spatial index of each point of interest, and thus obtain the information of the nodes at each level where each point of interest is located;

[0091] Step 5: Traverse each node starting from the root node according to the tree structure, screen the points of interest within the node unit, and the screening condition is to first judge whether there are points of interest with a grading value less than the depth in the node. If so, select and judge whether the maximum allowable number of points is reached. Otherwise, select points according to the sorting of the importance weight of the points of interest, and judge the minimum distance between the selected points and the previously selected points, and assign automatic grading attributes to the selected points;

[0092] Step 6: After traversing all nodes, the attributes of the points of interest include the calculated automatic grading value, and write the attributes into the attribute table of the original Shp or Gdb file to complete the grading of the points of interest;

[0093] In the whole process of grading the points of interest, the smallest processing unit is the node, the smallest grid unit under each display scale, which reduces the processing space range and the number of objects, and speeds up the execution efficiency of the algorithm. The logical process of screening the points of interest in each node is as Figure 4 .

[0094] VI. Establishment of the algorithm class

[0095] The idea of the algorithm is determined. Using a programming language to complete the writing of the algorithm code is the last process to implement the POI grading. Since the focus is on the execution effect of the algorithm, in the initial experiment, the C# language was used with the secondary development technology based on ArcGIS, and the ArcGIS platform was used in the form of an ArcGlS addin plugin to test the effect. In this way, relevant interfaces of ArcGIS can be directly called during the test to obtain layer data and its relevant attribute fields, greatly improving the experimental efficiency. To improve the portability of the method, this method is used in the system of the subsequent project example. This application encapsulates the improved grading method into two separate C++ classes, which is convenient for direct reuse when integrated into the system.

[0096] To make the algorithm more scalable and portable, this application designs the QTPOIntLayer C++ class to implement the algorithm module for POI grading. Among them, for the POI feature structure Feature, it includes the importance weight attribute, the ID of the POI, the automatic grading attribute for saving the grading result, and the original grading attribute value; for the grid object abstract design structure MapRect, it includes the range of the grid, and there are also two methods, Contains to judge whether the POI is inside the grid, and the Spilt method to perform a four-way split of the grid; the direct object structure quadtree_t includes the node QuadNode, and the node is a grid, and the node includes the depth attribute, its four child nodes, the grid MapRect represented by the node, and the POIs contained in the node; in the class, the parameter display scale denominator is stored in an integer array, and the number of levels for building the quadtree is obtained through this array. The attribute Limit is the maximum number of points allowed for each node, and the default value is 1. Other methods include the function CreatQuadTree to create a full quadtree with the depth corresponding to the number of levels of the display scale, the method InsertQuad to insert POIs to build the quadtree index, and finally the function TravelQuadTree to traverse the quadtree to implement the filtering and thinning process to complete the assignment of the automatic grading attribute AutoLevel of Feature. The entity structure diagram of the class and the attributes and methods included in each object structure are as Figure 5 .

[0097] VII. Digital Map Application Example

[0098] To verify the feasibility of the algorithm, the algorithm is implemented in combination with the POI grading module in the example of the Beijing Digital Map Compilation and Release System. By visualizing the grading effects of different POI thinning algorithms, comparing and analyzing the POI distribution of different algorithms at the same scale level, and counting the efficiency and thinning quantity results of the algorithm, the application types of the algorithm are summarized.

[0099] The test data used in the experiment is the point-of-interest data of Beijing, in the form of a GDB database file, which contains a total of 331,475 point features.

[0100] The classification of the point-of-interest data follows the basic rules of the ERSI data framework. The attribute fields of the point-of-interest data include name, annotation, data source, point location address, classification level, remarks, collection year, and operator, etc. Among them, the "classification level" field is a reserved field for manually classifying the point-of-interest data, which is convenient for manual auxiliary adjustment and modification of the data classification. The "importance weight" field is to add the importance weight of the point of interest in batches using the system calculation tool before the data test according to the above classification of the point-of-interest data.

[0101] (1) Experimental process

[0102] The data coverage of the experiment is the whole city of Beijing. When setting to 1:1,155,583 in the standard scale template, all the point-of-interest data is displayed full screen. Open the client of the Beijing Digital Map Compilation and Publishing System, and complete the test process of point-of-interest classification in the following steps in sequence:

[0103] Step 1: Click to import map data and load all point-of-interest data of the Gdb type into the system;

[0104] Step 2: Click on the point-of-interest classification under the data editing module of the menu bar to start the point-of-interest classification processing command;

[0105] Step 3: Set parameters in the pop-up window. The general parameters include selecting the point-of-interest data layer to participate in the processing, adding the display scales of each level, and for the detailed parameters, select the thinning method to be used. After selecting the thinning method, the tab page below will automatically provide the parameters required by the corresponding algorithm, such as the minimum point spacing, the number of grid cells, and the screen density;

[0106] 4) After processing, click to execute the multi-scale classification display toolbar to display the final point-of-interest display effect;

[0107] The specific parameter settings are as follows:

[0108] The display scale levels are selected as 10 levels suitable for Beijing from simple to detailed in the standard scale.

[0109] When selecting the thinning method of full quaternary discrete class, the maximum number of points allowed on the screen at the largest scale is 30, and the minimum point spacing is 40 pixels on the screen; for the regular thinning method, the unit grid is set to 2cm * 2cm, and the maximum number of points allowed in the unit grid is 1.

[0110] Combined with the actual display requirements, the number of points actually displayed at the first level with the smallest scale is very small, basically being the names of administrative regions at the highest level. Therefore, an artificial auxiliary setting is made for the allowable number of points of interest for the nodes at the first level, which is set to 1.

[0111] (2) Experimental Results

[0112] The scales shown are at level 10. To illustrate the effectiveness of the improved thinning algorithm of this application, the display results of points of interest in the same area at three quite different scales of 1:577791, 1:36111, and 1:2257 are selected for comparison. The three actual effects are as Figures 6 to 8 .

[0113] The running times of the two classification methods and the number of points of interest displayed at each display scale are counted in order to evaluate and judge from two aspects of time efficiency and display effect. The specific values of the number of points of interest displayed at each performance level are as Figure 9 .

[0114] (3) Result Analysis

[0115] As the display scale increases, the curve of the number of points of interest displayed shows an S-shaped trend. The results obtained by these two classification methods based on grid division for index establishment basically conform to the load rule in cartography. Next, the differences between these two classification methods in terms of display effect and running efficiency will be compared and analyzed.

[0116] 1) Ordinary Regular Grid Method

[0117] For the point-of-interest classification obtained by the ordinary regular grid thinning method, at small scales, because the grid side length is too small, some points of interest that are not necessary to be displayed are selected at lower levels. Since this method uses a unified grid size on the drawing surface to screen points of interest, the grid division is too dense at small scales, and only the points of interest are selected in the order of weights within the cell grid. Therefore, there is a phenomenon of uneven distribution of some points of interest between the grids. In terms of the execution efficiency of the algorithm, due to the different numbers of grids at each level, all points of interest need to calculate the grid indexes at each level and traverse the grids. Therefore, the execution efficiency of the algorithm is low, and when facing the data volume of more than 300,000 points of interest in Beijing, the execution time is too long.

[0118] 2) Full Quaternary Discrete Class Method

[0119] As can be seen from the figure, at 1:577791, fewer interest points are selected by the thinning method. Since the level of the display scale is related to the depth of the quadtree, the number of grids divided is small under a small scale, and fewer interest points are selected. Since points with more important weights are usually displayed under a small scale, manual assistance can be used for addition. Under a large scale, the number of grids increases, and the unit grids are associated with the grids of the upper level by the node structure of the quadtree. Therefore, the inclusion relationship between the levels of interest points is complete, and the distribution density is very uniform under a large scale. The operation efficiency is increased by 68.4% compared with before.

[0120] Through the above comparative analysis, the following conclusions can be obtained:

[0121] Ordinary regular grid: The classification result shows that the spatial distribution of interest points is good. However, under a small scale, the grid division is too dense, resulting in an insufficient importance level of the selected interest points. It is difficult to set the grid size of user parameters, and the running time is long. The classification of experimental data usually takes about 20 minutes to complete.

[0122] Full quadtree segmentation: It has a good display effect on interest points under a large scale level, with uniform distribution and correct expression of attribute information. However, the usage conditions are limited by the integral multiple relationship between the display scales. Using the quadtree method to divide grids has a higher execution efficiency under the same conditions. The classification time of experimental data is shortened to about 6 minutes compared with the ordinary regular grid.

[0123] Taking the Beijing Digital Map Compilation System project as an example, this application integrates the grid classification algorithm based on the full quadtree into the classification module of the project example, and uses the interest point data of Beijing for classification testing. The practicability of the two classification methods before and after improvement is experimentally compared, and the effectiveness of the method proposed in this application is verified.

Claims

1. An improved hierarchical high-load expression method for points of interest on big data maps, characterized in that: Based on the classification of digital map points of interest, we first establish a multi-scale expression method for digital maps based on the source of classification requirements, improve the algorithm for thinning out points of interest in the design specification grid, use quadtrees to grid the mapping area to establish a quadtree index for points of interest, and use object-oriented programming languages ​​to develop a classification method class. Aiming at the shortcomings of the regular grid thinning method, we establish a high-load improved algorithm for the classification of points of interest on big data maps: 1) Spatial indexing method: Based on the GIS quadtree indexing method, the distribution range of the interest points is divided into grids with full quadtrees. The number of grids at each level is constant, and each node represents each grid. The depth of the tree represents the corresponding level of expression. Each grid and the grids it contains use the tree structure to find the child nodes and parent nodes of the node for judgment. The variables at different levels are converted into the maximum number of points allowed in the grid at the current level. When indexing the point, it is determined whether the current grid contains the interest points selected at the previous level, and the interest point is directly searched to see if it is included in the selected queue of its parent node; 2) Method for determining the number of points to be screened at each presentation level: In the thinning method of the regular grid, the number of points that need to be thinned out at each presentation level is calculated by multiplying the number of grids at that level by the maximum number of points allowed by the unit grid set by the user. A method for changing and determining the number of points is established, and the input parameter provided to the user is the number of points of interest displayed within the display window range at the presentation level with the largest scale. The maximum number of points of interest allowed for thinning out at each level is calculated using the parameters, the range of the current display window, and the scales at each level. The maximum number of points is divided by the number of nodes in the quadtree at each level to obtain the maximum number of points of interest allowed for unit nodes at each level, which meets the requirements for the number of points of interest at large scales. At small scales, the number of points of interest allowed for unit nodes is set manually with assistance.

2. According to claim 1, the improved hierarchical high-load expression method for big data map points of interest is characterized in that: Formulate optimization of interest point thinning rules: 1) The algorithm is efficient enough when thinning out points of interest under large amounts of data; 2) The extracted interest point set has a high importance level; 3) The distribution of the extracted interest points is relatively uniform to avoid overlapping and overcrowding. This application is based on the optimization of these three points of interest rules, and takes regular grid thinning as the basis to improve the thinning method and establish a thinning method suitable for points of interest in ordinary government digital maps. In which, based on the fact that thinning at each representation level in the regular grid requires re-traversing a different number of grids and indexing the points inside the grid, the algorithm is improved for efficiency issues; based on the point spacing in the regular grid and the consistency of the surface density between each representation level, the number of screen points is used as a parameter factor, and the algorithm is optimized by selecting a point in the unit grid to establish a four-element discrete class method for thinning.

3. According to claim 1, the improved hierarchical high-load expression method for big data map points of interest is characterized in that: Point of interest thinning architecture: Based on the regular grid thinning method, it is optimized, the distribution range of the points of interest is divided into regular grids, the area is divided into blocks, the area where each point of interest exists is clearly defined, and a regular grid index is established for the points of interest. When extracting and processing the points of interest, the points of interest are marked with grid numbers to reduce the traversal of unnecessary points of interest; when screening points of interest, the grid is used as the basic unit, and a single grid is screened based on the importance of the point of interest and the distance factor of the distributed points. In the entire process of grading the points of interest, the smallest processing unit is the smallest grid unit at each display scale, which reduces the spatial scope and number of objects processed.

4. According to claim 1, the improved hierarchical high-load expression method for big data map points of interest is characterized in that: Grid thinning based on full quaternion discrete classes: store geographic spatial information in quadtree nodes, implement quadtree indexing of spatial data, divide the spatial area of ​​a certain range into four equal subspace areas, each subspace in the tree structure corresponds to a tree node, and each subspace is recursively divided into four subspaces again, and so on until the tree level reaches a certain depth or meets specific conditions. All elements within the spatial range are in a minimum area, that is, they belong to a leaf node. When inserting elements into the quadtree, the geometric attributes of the elements are used to calculate the node to which they belong.

5. According to claim 4, the improved hierarchical high-load expression method for big data map points of interest is characterized in that: When thinning out the data of interest points with a wide spatial range and a large amount of data, considering the uniform distribution of interest points in the entire mapping area and the processing efficiency, rasterizing the entire mapping area is the best solution. In combination with the multiple relationship between standard scales, the mapping area is gradually divided using the quaternion discrete class. Each level of the tree represents the grid situation of the division under each level of display scale. First, the minimum bounding rectangle MBR of the point group is obtained, and the rectangle range is Xmax, Xmin, Ymax, and Ymin. A full quadtree of the corresponding depth is established according to the number of display scale levels n. Set the counterclockwise area sequence to 1, 2, 3, 4; the root node represents the maximum circumscribed rectangle of the distribution of interest points in the mapping area, and each node has four child nodes. The four child nodes of the root node correspond to the four areas of 0, 1, 2, 3 upper right, upper left, lower left, and lower right respectively. Each node extends until the maximum depth of the tree, that is, the maximum display scale level; By establishing a full quadtree, the mapping area is divided into hierarchical grids, and all points of interest fall into the divided grids. The coordinates of the interest points are recursively calculated to obtain the grids of each level to which the interest points belong, that is, the node information to which the interest points belong can be obtained. When each point of interest is inserted, the relationship between the point of interest data and the node is calculated. The inserted point calculates its area X-Xmin, Y-Ymin and its encoding until the depth reaches the maximum value encoding {0, 2, 1, 3, 4...}, and the point is saved in the leaf node to obtain the quadtree index of the point of interest.

6. According to claim 5, the improved hierarchical high-load expression method for big data map points of interest is characterized in that: Filter points of interest according to the rules. In the regular four-element discrete grid, each display scale level corresponds to a layer of the tree. The process of grading points of interest is completed by traversing the tree once. Compared with the common thinning method, when thinning each display scale level, it must be traversed completely according to the number of grids, which has higher efficiency. When thinning out the second expression level, traverse to the node of depth 1 of the tree, and use the node as the processing unit to screen all the interest points within the spatial area represented by the node. The screening rules are as follows: judge the interest points. If there are interest points with automatic classification attributes that are less than or equal to the current level, these points are the points selected in the previous level. According to the classification rules, the points selected and displayed in the small scale are still selected in the larger scale, and the points are still selected. If the classified points are not enough for the number of grid selection points, select points according to the following method. If the node does not contain interest points, all interest points in the node are sorted according to the importance weight, and a point list is formed in order from large to small. The first interest point in the point list is selected to assign the expression level value; if there are points with the same weight in the sorting, the geometric position of the interest point is used as the basis to calculate the distance between the interest point and the center of the node area. The smaller distance is given priority to ensure that the distance between the grids is relatively uniform, and the selection is cyclic until the number of selected points reaches the maximum allowed number of points for the node of this depth.

7. The improved hierarchical high-load expression method for big data map points of interest according to claim 6 is characterized in that: According to the full quaternion discrete class, the grids of each level are divided into grids, and the grid relationship between each level is fixed. It is only applicable to the case where there is an even-number multiple relationship with the display scale, so as to solve the problem of fast indexing of points under the grids of each area and avoid repeated calculation of indexes when thinning multiple levels. In addition, the node area is used as the processing unit to calculate the point spacing and filter the optimal position to ensure the relative uniformity of the entire mapping area. The maximum number of selected nodes at each depth is obtained by taking the average value of the maximum number of interest points allowed by the level. The calculation formula is as follows: Each level of screen represents the area S 屏 The actual area is S, the maximum number of points allowed by the screen is m, and the current performance level is n, where the maximum number of points m is determined by the user-set parameters.

8. The improved hierarchical high-load expression method for big data map points of interest according to claim 1 is characterized in that: The process of thinning out the full quaternion discrete grid: Step 1: Load the most detailed scale vector point feature interest point data in Shp or Gdb format into the system; Step 2: Determine the execution of the interest point classification process, requiring the user to input parameters. The scale of each level of expression determines the depth of the quadtree and the parent-child node layer relationship between each level of scale; the number of interest points displayed on the screen can be adjusted to calculate the maximum number of interest points allowed in the unit grid at each level, the default is 30; the minimum spacing on the map ensures that the display effect of interest points at each level of scale is consistent with human visual resolution; Step 3: Read the number and relationship of display scales set by the user, read the loaded points of interest to calculate the minimum enclosing rectangle, and create a full quadtree of the specified depth; Step 4: Loop and insert the read points of interest, calculate the spatial index of each point of interest, and then get the node information of each level where each point of interest is located; Step 5: Traverse each node from the root node according to the tree structure, and filter the points of interest in the node unit. The screening condition is to first determine whether the node contains points of interest with a classification value less than the depth. If so, select and determine whether the maximum allowed number of points is reached. Otherwise, select points according to the point order of interest importance weight. The selected points are all judged by the minimum distance with the previously selected points, and the selected points are automatically assigned classification attributes; Step 6: After traversing all nodes, the attributes of the interest points include the calculated automatic classification values, and the attributes are written into the attribute table of the Shp or Gdb file to complete the classification of the interest points.

9. The improved hierarchical high-load expression method for big data map points of interest according to claim 1 is characterized in that: Algorithm class establishment: Design QTPOIntLayer C++ class to implement the algorithm module for POI classification, in which the POI feature structure Feature includes the importance weight attribute, the POI ID, the automatic classification attribute saved by the classification result, and the original classification attribute value; for the grid object abstract design structure MapRect, including the range of the grid, there are two methods Contains to determine whether the POI is inside the grid, and the Spilt method performs quad-division of the grid; the direct object structure quadtree_t includes the node QuadNode, which is a grid. The node contains the depth attribute, its four child nodes, and the grid Ma represented by the node. pRect and the points of interest contained in the node; in the class, the parameter display scale denominator is stored in an integer array, through which the number of levels of quadtree establishment is obtained, the attribute Limit is the maximum number of points allowed for each node, and the default value is 1; other methods include the function CreatQuadTree to create a full quadtree corresponding to the depth of the display scale level, the method InsertQuad to insert points of interest to establish a quadtree index, and finally the function TravelQuadTree to traverse the quadtree to implement screening and thinning processing to complete the assignment of the automatic grading attribute AutoLevel of Feature.

Images