A method for urban block function classification integrating POI pre-classification and graph neural network

Abstract

The present invention discloses a method for classifying urban block functions by integrating POI pre-classification and graph neural network. The method comprises the following steps: S1, constructing a Delaunay triangulation network using urban block POI, S2, determining the importance ranking of POIs in the urban block according to the average side length of the Delaunay triangulation network edge connected to each POI point, S3, selecting the top three POI types in the urban block as the combined labels of the urban block functions, S4, merging the combined labels to obtain pseudo labels of urban block functions, S5, establishing a POI graph network of each urban block based on the Delaunay triangulation network, S6, training a graph neural network classification model using the pseudo labels of urban block functions, and S7, classifying urban block functions using the trained graph neural network classification model. The method of the present invention can extract the spatial structure information of POI data, enrich the semantic information of POI data, and thus improve the accuracy of urban block function classification.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an urban block function classification method integrating POI pre-classification and graph neural network. Background Art

[0002] The development of cities has formed different functional areas. These urban functional areas are important spatial carriers for realizing urban economic and social functions, meeting people's different needs for social and economic activities. Their distribution reflects the characteristics of the city. The identification and classification of urban functional areas is not only conducive to the cognition and management of the city, but also to the structural optimization and development planning of the city. POI data is an important carrier of geographic entity information in the city and is widely used for the identification and extraction of urban functional areas.

[0003] Deficiencies of existing technology:

[0004] Existing studies rarely consider the spatial structure of POI point sets when classifying urban functional areas. The connection pattern of POIs can express their co-occurrence pattern in urban blocks and is also a specific manifestation of urban block functions. This is an important basis for classifying urban functional areas. Therefore, when classifying urban block functions, the lack of spatial structure information makes the semantic information of POI data inaccurate and insufficient, which often leads to a decrease in classification accuracy. Summary of the invention

[0005] The purpose of the present invention is to provide a method for urban block function classification that integrates POI pre-classification and graph neural network to solve the problems raised in the above background technology.

[0006] A method for classifying urban block functions by integrating POI pre-classification and graph neural network, characterized in that the classification method comprises the following steps:

[0007] S1, construct Delaunay triangulation using city block POI;

[0008] S2, determining the importance ranking of the POIs in the city block according to the average side length of the Delaunay triangulation edges connected to each POI point in step S1;

[0009] S3, selecting the top three POI types in the city block according to the ranking in step S2 as the combined label of the city block function;

[0010] S4, merging the combined labels obtained in step S3 to obtain pseudo labels of urban block functions;

[0011] S5. Establish a POI map network for each city block based on the Delaunay triangulation network;

[0012] S6. Using the pseudo labels of city block functions obtained in step S4, training a graph neural network classification model;

[0013] S7. Use the graph neural network classification model trained in step S4 to perform urban block functional classification.

[0014] As a further improvement of the present invention, the average side length of the Delaunay triangulation edges connected to the POI point in step S2 of the method can represent the importance of the POI in the block, and the average side length in a certain area is proportional to the importance.

[0015] As a further improvement of the present invention, the combined label in step S3 of the method is a character string consisting of the top three POI type names in the city block.

[0016] 9. As a further improvement of the present invention, the combined label merging in step S4 of the present method includes three rounds, wherein:

[0017] In the first round of label merging, the combinations of labels with the exact same label were merged into one category, and the number of city blocks with each type of label was counted;

[0018] The second round of label merging is based on the similarity of string labels. The similarity between all labels is calculated, and the type with fewer city blocks in two string labels with similarity greater than a certain threshold is merged into the type with more city blocks.

[0019] The third round of label merging takes into account that each functional type of urban block has specific POI type characteristics. By manually identifying the string labels obtained after the second round of label merging, the urban blocks with different POI type characteristics in the labels are merged into several final urban block functional types.

[0020] As a further improvement of the present invention, step S5 of the method exports the Delaunay triangulation of each block and saves it as POI map network data that meets the subsequent model input requirements.

[0021] As a further improvement of the present invention, in step S6 of the method, a supervised classification model in a graph neural network is adopted in the training, and the model is trained together with POI graph network data and block function pseudo-label data.

[0022] As a further improvement of the present invention, in step S7 of the method, the POI map network data of the urban blocks that need to be functionally classified is input into the trained model for prediction to obtain the urban block functional classification results.

[0023] As a further improvement of the present invention, the input network data in step S7 of the method includes non-training data and training data, wherein:

[0024] The input network data is non-training data, and the classification accuracy is directly obtained by comparing the prediction results with the functional labels of urban blocks;

[0025] The input network data is used as training data, and the urban block functional types predicted by the model are empirically compared with the urban block Tiantu map images, and the comparison results are statistically analyzed to determine the accuracy of urban block functional classification.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] The present invention constructs the urban block POI data as graph network data, combines the urban block function pseudo-labels obtained by pre-classification, and uses the graph classification model in the graph neural network to perform urban block function classification. The method can extract the geographical proximity relationship of POI data and the co-occurrence pattern in the urban block, enrich the semantic information of POI data, and thus improve the accuracy of urban block function classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a flow chart of a method for urban block function classification integrating POI pre-classification and graph neural network of the present invention;

[0029] Figure 2 The present invention is a Delaunay triangulation diagram based on city block POI, which is a city block function classification method integrating POI pre-classification and graph neural network;

[0030] Figure 3 A diagram of the training process of a graph neural network classification model for a method for classifying urban block functions that integrates POI pre-classification and graph neural network according to the present invention;

[0031] Figure 4 This is a comparison and analysis diagram of the urban block function classification results of the urban block function classification method that integrates POI pre-classification and graph neural network in the present invention and the Tiandi Map image (the color of the POI point symbol in the figure indicates its type, the size indicates its area weight, and the title is the type predicted by the urban block model). DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] Example

[0034] See also Figure 1-3The present invention provides the following technical solution: a method for classifying urban block functions by integrating POI pre-classification and graph neural network, comprising the following steps:

[0035] S1. Construct a Delaunay triangulation network using POIs in urban blocks; triangulate all POI points in the block into an irregular triangulation network that meets the Delaunay criterion.

[0036] S2, determining the importance ranking of POIs in the city block based on the average length of the Delaunay triangulation edges connected to each POI point;

[0037] POIs with longer average side lengths in a certain area are more important than other POIs, while the POIs with the longest average side lengths in the entire block dominate the block and can represent the main functional type of the block.

[0038] S3, selecting the top three POI types in the city block as the combined label of the city block function;

[0039] For example, if the top three POIs in a city block are all of type "company enterprise", the combined label of the city block is represented by a string of "company enterprise enterprise enterprise enterprise";

[0040] In addition, the number of extractions is determined to be three. Too many extractions will make the initial string labels too numerous and too complex, which is not conducive to the subsequent label merging processing. If the number of extractions is less than three, the string labels obtained are not sufficient to represent the comprehensive functional areas with balanced multiple functions. Therefore, the POI types ranked in the top three in extraction weight can not only represent the main functions of urban blocks, but also are sufficient to identify urban blocks with complex functions.

[0041] S4. Merge the combined labels obtained in S3 to obtain pseudo labels of urban block functions. Take 1,000 urban blocks in the central urban area of ​​Changzhou as an example:

[0042] In the first round of merging, the blocks with the same combination labels were simply grouped into one category, and the number of city blocks of each type was counted. After the merging was completed, the initial labels of the 1,000 city blocks were merged into 483 types.

[0043] The second round of merging is based on the similarity of string labels. Python's built-in string similarity calculation method is used. This method processes strings into sets and calculates the ratio of the number of elements in the intersection of two sets to the number of elements in the union as the similarity of the two strings. After multiple experiments, the similarity threshold is set to 0.5. The types with fewer city blocks in the two string labels with a similarity greater than 0.5 are merged into the types with more city blocks. After this step, all city blocks are merged into 40 types.

[0044] Finally, considering that each functional type of urban block has specific POI type characteristics, for example, the POI string label of the residential area should contain "real estate", while the urban blocks with all labels "door address" or no POI distribution are merged into other types because they cannot be judged. After manually identifying these string labels, the urban blocks with specific POI type characteristics are classified into one of the final 9 urban block functional types, and the urban block functional pseudo labels used for subsequent model supervision training are obtained as shown in Table 1.

[0045] S5. Establish a POI map network for each city block based on the Delaunay triangulation network.

[0046] S6. Use pseudo labels of urban block functions to train the graph neural network classification model;

[0047] Since a supervised classification model is used, the POI graph network data of the block and the pseudo-label data of the block functions are used to train the graph classification model;

[0048] like Figure 2 As shown in the figure, the training of the model is mainly parameter adjustment training. After multiple rounds of training, the model obtained by training the parameter combination with the best average performance is selected for the subsequent prediction of urban block functional types.

[0049] S7, using the model trained in step 6 to perform functional classification of urban blocks; inputting the urban block POI map network data that needs to be functionally classified into the trained model for prediction to obtain the urban block functional classification results;

[0050] If the input data is not training data, the classification accuracy is directly obtained by comparing the prediction results with the city block function labels;

[0051] If the input data is training data, then Figure 3 As shown in the figure, the urban block functional types predicted by the model are empirically compared with the urban block Tiantu map images, and the comparison results are statistically analyzed to determine the accuracy of urban block functional classification.

[0052] Results table of urban block function pre-classification

[0053]

[0054] Table 1

[0055] Combined with Table 1, Figure 2 and Figure 3It can be seen that the present invention constructs the urban block POI data as graph network data, combines the urban block function pseudo-labels obtained by pre-classification, and uses the graph classification model in the graph neural network to perform urban block function classification. This method can extract the geographical proximity relationship of POI data and the co-occurrence pattern in the urban block, enrich the semantic information of POI data, and thus improve the accuracy of urban block function classification.

[0056] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0057] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for urban block function classification integrating POI pre-classification and graph neural network, characterized by: This classification method includes the following steps: S1, construct Delaunay triangulation using city block POI; S2, determining the importance ranking of the POIs in the city block according to the average side length of the Delaunay triangulation edges connected to each POI point in step S1; S3, selecting the top three POI types in the city block according to the ranking in step S2 as the combined label of the city block function; S4, merging the combined labels obtained in step S3 to obtain pseudo labels of urban block functions; S5. Establish a POI map network for each city block based on the Delaunay triangulation network; S6. Using the pseudo labels of city block functions obtained in step S4, training a graph neural network classification model; S7, using the graph neural network classification model trained in step S4 to perform urban block function classification; The S4 combined label merging includes three rounds, among which: In the first round of label merging, the combinations of labels with the exact same label were merged into one category, and the number of city blocks with each type of label was counted; The second round of label merging is based on the similarity of string labels. Python's built-in string similarity calculation method is used. This method processes strings into sets, calculates the ratio of the number of elements in the intersection of two sets to the number of elements in the union as the similarity of the two strings, sets the similarity threshold to 0.5, and merges the types with fewer city blocks in the two string labels with a similarity greater than 0.5 into the types with more city blocks. All city blocks are merged into 40 types. In the third round, each functional type of urban block is considered to have specific POI type characteristics. The POI string label of the residential area should contain "real estate", while the urban blocks with all labels "door address" or no POI distribution are merged into other types because they cannot be judged. After manually identifying these string labels, the urban blocks with specific POI type characteristics are classified into one of the final 9 urban block functional types, and the urban block functional pseudo-labels for subsequent model supervision training are obtained, which include central business district, residential area, commercial area, transportation and warehousing area, comprehensive service area, cultural and educational area, administrative area, scenic area and others.

2. The method for urban block function classification integrating POI pre-classification and graph neural network according to claim 1 is characterized in that: The average length of the Delaunay triangulation edges connected to the POI point in step S2 of the method can represent the importance of the POI in the block. The average length of the edge in a certain area is proportional to the importance.

3. The method for urban block function classification integrating POI pre-classification and graph neural network according to claim 1 is characterized in that: In step S3 of the method, the combined label is a character string consisting of the top three POI type names in the city block.

4. The method for urban block function classification integrating POI pre-classification and graph neural network according to claim 1 is characterized in that: In step S5 of the method, the Delaunay triangulation of each block is exported and saved as POI map network data that meets the subsequent model input requirements.

5. The method for urban block function classification integrating POI pre-classification and graph neural network according to claim 1 is characterized in that: In step S6 of this method, a supervised classification model in a graph neural network is used for training, and the model is trained jointly using POI graph network data and block function pseudo-label data.

6. The method for urban block function classification integrating POI pre-classification and graph neural network according to claim 1 is characterized in that: In step S7 of the method, the POI map network data of the urban blocks that need to be functionally classified is input into the trained model for prediction to obtain the urban block functional classification results.

7. The method for urban block function classification integrating POI pre-classification and graph neural network according to claim 6 is characterized in that: In this method, the input network data in step S7 includes non-training data and training data, wherein: The input network data is non-training data, and the classification accuracy is directly obtained by comparing the prediction results with the functional labels of urban blocks; The input network data is used as training data, and the urban block functional types predicted by the model are empirically compared with the urban block Tiantu map images, and the comparison results are statistically analyzed to determine the accuracy of urban block functional classification.

Images