A multi-source geospatial big data knowledge graph construction method and system for smart city planning

Abstract

The application discloses a multi-source geospatial big data knowledge graph construction method and system for smart city planning, and the method comprises the following steps: collecting original data of a vector layer, a grid layer, a text layer and an image layer, and performing pretreatment to obtain vector data, grid data, text data and image data respectively; a knowledge graph ontology layer is constructed; according to the set and alignment mode constructed in the knowledge graph ontology layer, the DeepLabV3+ semantic segmentation model, the NLP model and the GIS platform based on the software platform are used to extract the respective entity set, the relationship set, the attribute set and the alignment mode in the vector layer, the grid layer, the text layer and the image layer in a target city, and a multi-source geospatial big data knowledge graph is formed. The efficient integration and accurate alignment of heterogeneous data are realized through unified reference entities, and the shortcomings of traditional methods in data integration efficiency, entity alignment accuracy and intelligent analysis capability are overcome.

Description

Technical Field

[0001] This invention belongs to the field of knowledge graph and smart city technology, and in particular relates to a method and system for constructing a multi-source geospatial big data knowledge graph for smart city planning. Background Technology

[0002] With the rapid advancement of global urbanization, the construction of smart cities places higher demands on urban planning. However, traditional urban planning tools and methods, such as Geographic Information Systems (GIS) and remote sensing technologies, face significant challenges in integrating multi-source heterogeneous geographic data. These data sources (such as remote sensing imagery, sensor networks, and socioeconomic data) differ significantly in spatial resolution, timeliness, and data type, making them difficult to apply effectively in smart city planning where real-time performance and accuracy are crucial. Furthermore, current multi-source geographic data processing still suffers from low efficiency and high complexity, which to some extent limits its practical effectiveness in dynamic decision-making and urban planning. Therefore, how to achieve efficient integration, accurate analysis, and intelligent application of multi-source heterogeneous geographic data has become a critical technical bottleneck that urgently needs to be addressed in smart city planning.

[0003] To address the aforementioned issues, Geographic Knowledge Graph (GeoKG), as an emerging technology, transforms multi-source geographic information into a structured knowledge format, enabling deep data integration and intelligent analysis. GeoKG effectively integrates various data types, including vector, raster, text, and image data, and utilizes semantic modeling techniques to accurately describe geographic entities and their relationships, providing a novel data support framework for smart city planning. However, GeoKG still faces numerous challenges in practical applications, particularly regarding data accuracy, entity alignment, and seamless integration of multimodal data. Overcoming these challenges through efficient modeling and analysis methods is a key direction for advancing the intelligent development of smart city planning. Summary of the Invention

[0004] The purpose of this invention is to address the problems existing in the prior art by providing a method and system for constructing a multi-source geospatial big data knowledge graph for smart city planning. This method overcomes the shortcomings of traditional methods in terms of data integration efficiency, entity alignment accuracy, and intelligent analysis capabilities by integrating heterogeneous data in a clever and precise manner.

[0005] According to one aspect of the present invention, a method for constructing a multi-source geospatial big data knowledge graph for smart city planning is provided, comprising:

[0006] The raw data from the vector layer, raster layer, text layer, and image layer are collected and preprocessed to obtain vector data. V Raster Data RText data T and image data I ;

[0007] Constructing a knowledge graph ontology layer, the construction of the knowledge graph ontology layer includes: based on vector data Data V Construct a vector entity set E v Vector attribute set A v and the vector relation set R v According to raster data R Construct a grid entity set E r Raster attribute set A r and the set of raster relationships R r According to the text data Data T Construct a collection of text entities E t Text attribute set A t and text relation set R t According to image data Data I Construct image entity set E i Image attribute set A i and the image relationship set R i Using street names as reference entities, the vector entity set E v Grid entity set E r Text entity set E t and the image entity set E i Alignment is performed to obtain the entity set E={E v E r E t E i}, determine the attribute set A and the relation set R based on the entity elements in the obtained entity set E, the positional relationships between the entity elements, and the attributes of the entity elements;

[0008] Based on the sets and alignment methods constructed in the ontology layer of the knowledge graph, the DeepLabV3+ semantic segmentation model, NLP model, and GIS platform are used to extract the entity sets, relation sets, attribute sets, and alignment methods of the vector layer, raster layer, text layer, and image layer of the target city, forming a multi-source geospatial big data knowledge graph.

[0009] The above-mentioned technical solution overcomes the shortcomings of traditional methods in terms of data integration efficiency, entity alignment accuracy, and intelligent analysis capabilities by efficiently integrating and accurately aligning heterogeneous data through a unified reference entity. This method ensures the automation and efficiency of data processing by combining the DeepLabV3+ semantic segmentation model, natural language processing technology, and GIS spatial analysis tools. Utilizing the powerful storage and query capabilities of graph databases, it provides a brand-new data management and analysis framework for smart city planning, demonstrating significant application effects in urban road planning, green coverage analysis, and environmental quality assessment, and providing strong technical support for the scientific decision-making of smart cities.

[0010] As a further technical means, the vector data... V The data is derived from Open Street Map and other open-source datasets in the study area, following vector data extraction principles, which include including at least all data on roads and administrative districts in the study area.

[0011] As a further technical means, constructing the vector entity set, vector attribute set, and vector relationship set includes:

[0012] Extract road names, OSM identifiers, and administrative divisions to form a vector entity set E. v ;

[0013] It's important to note that OSM stands for OpenStreetMap, a global open-source mapping project where users can freely edit and use the geographic data. OSM's data structure is designed for flexible and efficient representation of geographic information, primarily consisting of three basic elements: nodes, ways, and relations. This data can be used in various applications, including but not limited to urban planning, traffic navigation, and geographic information systems. OSM data includes various types of data such as road and rail networks, buildings, water bodies, land use, points of interest, and administrative boundaries.

[0014] Calculate the vector entity set E v The topological and geometric attributes of each entity element are used to form a vector attribute set A. v The topological attributes include at least adjacent roads and road lengths, and the geometric attributes include at least latitude and longitude.

[0015] Establish a vector entity set E v The category, length, nearest road, Euclidean distance, and spatial and attribute relationships of each entity element in the vector relation set R are used to form a vector relation set R. v .

[0016] As a further technical means, the raster data... R The data is derived from remote sensing data extracted from the study area according to the raster data extraction principles, which include: the needs of actual application scenarios and the existing data on the platform.

[0017] As a further technical means, constructing the grid entity set, grid attribute set, and grid relationship set includes:

[0018] Based on the sampling point IDs generated along the roads within the study area, and combined with the time field, a raster entity set E is formed. r ;

[0019] Based on the principles of raster data extraction, the corresponding data types are extracted from the remote sensing data within the study area to form a raster attribute set A. r ;

[0020] Based on the principles of raster data extraction, the corresponding data types in the remote sensing data within the study area are extracted, and a raster entity set E is established. r The geographical relationships between various entity elements form a grid relationship set R. r .

[0021] As a further technical means, the text data... T The data is derived from store reviews within the study area and extracted according to text data extraction rules, which include: the data must be segmented into latitude and longitude using natural language processing technology.

[0022] As a further technical means, constructing the text entity set, text relation set, and text attribute set includes:

[0023] From text data Data T Extract the store names within the study area to form a text entity set E. t ;

[0024] From text data Data T Extract the text entity set E t The text attributes of each text entity element form a text attribute set A. t ;

[0025] According to text data Data T Each shop name is used to construct a text relationship set R, which establishes the geographical relationships between the shops and the nearest road, the administrative division, and the latitude and longitude. t .

[0026] As a further technical means, the image data... IData is derived from street-side sampling points within the study area, including green view rate, building view rate, and vehicle view rate extracted from 360° panoramic images of the streets.

[0027] As a further technical means, constructing the image entity set, image attribute set, and image relationship set includes:

[0028] Based on the sampling point IDs along the roads within the study area, geographic entities in the street view images are labeled to form an image entity set E. i ;

[0029] The image attribute set A is formed by extracting green view rate, building view rate, and vehicle view rate within the study area using a deep learning model (DeepLabV3+). i ;

[0030] Establish an image relationship set R by relating the sampling points along the street to their geographical locations, including the roads, administrative divisions, and latitude and longitude. i .

[0031] According to one aspect of the present invention, a multi-source geospatial big data knowledge graph construction system for smart city planning is provided, comprising:

[0032] The data collection and processing module is used to collect raw data from the vector layer, raster layer, text layer, and image layer, and to perform preprocessing.

[0033] The ontology layer construction module is used to construct a vector entity set E based on the preprocessed vector layer. v Vector relation set A v and vector attribute set R v Construct a raster entity set E based on the preprocessed raster layers. r Grid relation set A r and the raster attribute set R r A text entity set E is constructed based on the preprocessed text layer. t Text relation set A t and text attribute set R t The image entity set E is constructed based on the preprocessed image layers. i Image Relationship Set A i and the image attribute set R i Using RoadName as the reference entity, the vector entity set E v Grid entity set E r Text entity set E t and the image entity set E i Alignment is performed using a unified reference entity to obtain the entity set E={E vE r E t E i Based on the obtained entity set E, the relation set A and attribute set R are obtained, forming the knowledge graph ontology layer;

[0034] The knowledge graph construction module is used to extract entity sets, relation sets, attribute sets, and alignment methods from various data layers in the target city based on the set and alignment of the knowledge graph ontology layer, using the DeepLabV3+ semantic segmentation model, NLP model, and GIS platform based on ArcGIS Pro, Google Earth Engine, and PyCharm platform, to form a multi-source geospatial big data knowledge graph.

[0035] Compared with the prior art, the beneficial effects of the present invention are:

[0036] This method achieves efficient integration and precise alignment of heterogeneous data, overcoming the shortcomings of traditional methods in data integration efficiency, entity alignment accuracy, and intelligent analysis capabilities. By combining the DeepLabV3+ semantic segmentation model, natural language processing technology, and GIS spatial analysis tools, it ensures automated and efficient data processing. Leveraging the powerful storage and query capabilities of the Neo4j graph database, it provides a novel data management and analysis framework for smart city planning. Through a real-world case study in Wuhan, this invention demonstrates significant application effects in urban road planning, green coverage analysis, and environmental quality assessment, providing strong technical support for scientific decision-making in smart cities. Attached Figure Description

[0037] Figure 1 This is a flowchart of a method according to an embodiment of the present invention;

[0038] Figure 2 A study area map of test data for invention examples;

[0039] Figure 3 This is the data preprocessing flow for the first step of an example of the present invention;

[0040] Figure 4 The second step of this invention involves constructing a collection of entities, relationships, and attributes by combining the characteristics of each data layer, and the alignment process for each data layer.

[0041] Figure 5 Screenshot of the knowledge graph construction for this invention example;

[0042] Figure 6 Figure a shows the effect of applying the knowledge graph constructed in this invention to the urban heat island effect.

[0043] Figure 7Figure b illustrates the effect of applying a knowledge graph constructed in this invention to the urban heat island effect.

[0044] Figure 8 Figure c illustrates the effect of applying a knowledge graph constructed in this invention to the urban heat island effect.

[0045] Figure 9 A diagram illustrating the effect of applying a knowledge graph constructed in this invention to the analysis of urban green view rate and building density.

[0046] Figure 10 The diagram illustrates the effect of constructing a knowledge graph for urban travel recommendation scenarios in this invention.

[0047] Figure 11 The diagram shows the effect of applying the knowledge graph constructed in this invention to road network and traffic flow analysis.

[0048] Figure 12 The diagram illustrates the effect of applying the knowledge graph constructed in this invention to urban land use change detection. Detailed Implementation

[0049] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] like Figure 1 As shown, a method for constructing a multi-source geospatial big data knowledge graph for smart city planning includes:

[0051] Step 1: Collect the raw data from the Vector Layer, Raster Layer, Text Layer, and Image Layer, and preprocess them to obtain vector data. V Raster Data R Text data T and image data I The processing procedure is similar to Figure 3 As shown;

[0052] like Figure 4-5 As shown, step two: Constructing the knowledge graph ontology layer. Constructing the knowledge graph ontology layer includes: based on the vector data Data... V Construct a vector entity set E v Vector attribute set A v and the vector relation set R vAccording to raster data R Construct a grid entity set E r Raster attribute set A r and the set of raster relationships R r According to the text data Data T Construct a collection of text entities E t Text attribute set A t and text relation set R t According to image data Data I Construct image entity set E i Image attribute set A i and the image relationship set R i Using RoadName as the reference entity, the vector entity set E v Grid entity set E r Text entity set E t and the image entity set E i Alignment is performed to obtain the entity set E={E v E r E t E i}, determine the attribute set A and the relation set R based on the entity elements in the obtained entity set E, the positional relationships between the entity elements, and the attributes of the entity elements;

[0053] The alignment is constructed as follows: ,

[0054] In this embodiment, heterogeneous data is efficiently integrated and accurately aligned through a unified reference entity, overcoming the shortcomings of traditional methods in terms of data integration efficiency, entity alignment accuracy, and intelligent analysis capabilities. This method combines the DeepLabV3+ semantic segmentation model, natural language processing technology, and GIS spatial analysis tools to ensure the automation and efficiency of data processing. Utilizing the powerful storage and query capabilities of graph databases, it provides a brand-new data management and analysis framework for smart city planning, demonstrating significant application effects in urban road planning, green coverage analysis, and environmental quality assessment, providing strong technical support for scientific decision-making in smart cities.

[0055] In this embodiment, vector data Data V The data is derived from Open Street Map and other open-source datasets in the study area, following vector data extraction principles, which include including at least all data on roads and administrative districts in the study area.

[0056] Raster Data RIt originates from remote sensing data extracted from the study area according to the raster data extraction principles, which include: the needs of actual application scenarios and the existing data on the platform.

[0057] Specifically, raster data R是 The data is extracted based on Google Earth Engine and includes Normalized Difference Vegetation Index (NDVE), land use data, radiation data, surface temperature data, and rainfall data. The specific product to be selected depends on the project requirements and the product offered by GEE.

[0058] Text data T The data is derived from store reviews within the study area and extracted according to text data extraction rules, which include: the data must be segmented into latitude and longitude using natural language processing technology.

[0059] Image Data I Data is derived from street sampling points within the study area, including green view rate, building view rate, and vehicle view rate extracted from 360° panoramic images of the streets.

[0060] In this embodiment, the street sampling point data is obtained from 360° panoramic images of sampling points captured by Baidu Street View API.

[0061] like Figure 5 As shown in this embodiment, when constructing the knowledge graph ontology layer in step two, the first step is the vector data Data. V Construction of a collection of entities, relations, and attributes:

[0062] In this embodiment, constructing the vector entity set, vector attribute set, and vector relationship set includes:

[0063] a) Entity Extraction: Extract the road name, OSM identifier (ID), and administrative division (District) to form a vector entity set E. v ;

[0064] In this embodiment, the vector entity set E v The expression is as follows:

[0065] ;

[0066] b) Attribute extraction: Calculate the vector entity set E v The topological and geometric attributes of each entity element are used to form a vector attribute set A. v The topological attributes include at least adjacent roads and road lengths, and the geometric attributes include at least latitude and longitude.

[0067] In this embodiment, the vector attribute set A v The expression is as follows:

[0068] ;

[0069] c) Relationship Construction: Establish spatial and attribute relationships for geographic entities, such as class, length, nearest road, near distance, and administrative division, forming a vector relationship set R. v .

[0070] In this embodiment, the vector relation set R v The expression is as follows:

[0071] ={ , , , , , , };

[0072] The second step is raster data. R Construction of a collection of entities, relations, and attributes:

[0073] a) Entity extraction: Based on the sampling point IDs generated along the roads within the study area, and combined with the time field (T), a raster entity set E is formed. r ;

[0074] In this embodiment, the grid entity set E r The expression is as follows:

[0075] ;

[0076] b) Attribute Extraction: Following the principles of raster data extraction, raster attribute values ​​such as land use type, digital elevation model (DEM), land surface temperature (LST), precipitation, and albedo are extracted from the remote sensing data within the study area to form a raster attribute set A. r ;

[0077] In this embodiment, the raster attribute set A r The expression is as follows:

[0078] ;

[0079] c) Relationship Construction: Following the principles of raster data extraction, extract the corresponding data types from the remote sensing data within the study area, such as land use type (LandUse), road location (InRoad), digital elevation model (DEM), land surface temperature (LST), and administrative division location (LocatedIn), to establish a raster entity set E. r The geographical relationships between various entity elements form a grid relationship set R. r .

[0080] In this embodiment, the grid relationship set R r The expression is as follows:

[0081] ={ , , , , , , , , , };

[0082] The third step is text data. T Construction of Chinese text entity sets, text relation sets, and text attribute sets:

[0083] a) Entity extraction: from text data Data T Extract the store names within the study area to form a text entity set E. t Specifically, the method involves extracting shop names and related information from text using Natural Language Processing (NLP) technology, forming a text-layer entity set Et.

[0084] In this embodiment, the text entity set E t The expression is as follows:

[0085] ;

[0086] b) Attribute extraction: from text data Data T Extract the text entity set E t The text attributes of each text entity element, such as detailed location, class, and sentiment analysis result, form a text attribute set A. t ;

[0087] In this embodiment, text attribute set A t The expression is as follows:

[0088] ;

[0089] c) Relationship building: Based on text data Data T For each shop name, construct a text relationship set R by establishing geographical relationships between the shop and the nearest road, the administrative division, and the latitude and longitude. t .

[0090] In this embodiment, the text relation set R t The expression is as follows:

[0091] ={ , , , , , , , , };

[0092] The fourth step is image data. I Construction of image entity sets, image attribute sets, and image relation sets:

[0093] a) Entity extraction: Based on the sampling point IDs along the roads within the study area, geographic entities in the street view image are labeled to form an image entity set E. i ;

[0094] In this embodiment, the text entity set E i The expression is as follows:

[0095] ;

[0096] b) Attribute Extraction: Using a deep learning model (DeepLabV3+), green visibility (GVI), building visibility (BVI), and vehicle visibility (VVI) within the study area are extracted to form an image attribute set A. i ;

[0097] In this embodiment, image attribute set A i The expression is as follows:

[0098] ;

[0099] c) Relationship Construction: Establish the geographical relationships between sampling points along the street and their respective roads (InRoad), administrative divisions (LocatedIn), and latitude and longitude (Longitude, Latitude) to form an image relationship set R. i .

[0100] In this embodiment, the image relationship set R i The expression is as follows:

[0101] ={ , , , , , , }

[0102] The fifth step is entity alignment:

[0103] All data layer entity collections ( Alignment is achieved by using a unified reference entity (RoadName) to generate a comprehensive entity set E, thus realizing spatial consistency between data layers. This alignment lays the foundation for the comprehensive analysis of multi-source heterogeneous data and supports subsequent expansion of relationships and attributes in the knowledge graph.

[0104] Step 3: Based on the sets and alignment methods constructed in the ontology layer of the knowledge graph, the DeepLabV3+ semantic segmentation model, NLP model, and GIS platform based on ArcGIS Pro, Google Earth Engine, and PyCharm are used to extract the entity sets, relation sets, attribute sets, and alignment methods of the vector layer, raster layer, text layer, and image layer of the target city, forming a multi-source geospatial big data knowledge graph, and then realizing the storage management of the multi-source geospatial big data knowledge graph based on Neo4j;

[0105] In this embodiment, taking Wuhan City as an example, the extraction of entities, relationships, and attributes from different data layers of a multi-source geospatial big data knowledge graph, as well as entity alignment between different data layers, are carried out. The results of triple extraction are divided into...<E, R, A> Extraction results and<E, R, E> Extraction Results. The next step is to use Python to connect to the Neo4j interface, Py2Neo, based on the extracted triples to visualize the triples for each data layer. After extracting triples for each data layer, alignment between multi-source data entities is performed in the Neo4j graph database based on RoadName.

[0106] This implementation, guided by the knowledge graph ontology layer constructed in step 2, is based on actual data from Wuhan. , , The construction of triples, combined with the py2neo interface<Entity ,Relation, Attribute> Extraction results and<Entity, Relation, Entity> The extracted results are stored in the Neo4j graph database, and alignment between multi-source data entities is performed in the Neo4j graph database based on RoadName. See the appendix for details of some of the graphs. Figure 3 Extracted<E ,R, A> as well as<E, R, E> (See Tables 1 and 2 for details)

[0107] Table 1: Knowledge Graph<E,R,A> Example table

[0108]

[0109] Table 2: Knowledge Graph<E,R,E> Example

[0110]

[0111] The specific steps are as follows:

[0112] 1.1 Vector layer entity, attribute extraction, and relation construction

[0113] The extraction of entities, attributes, and relationships in the vector layer is based on the ArcGIS Pro platform (other GIS platforms, such as QGIS, can also be used here).

[0114] (1) Entity extraction

[0115] In the vector layer, three main types of entities are involved: RoadName (road name), ID (sampling point), and District (administrative region).

[0116] a. Road Vector: Comes with the RoadName attribute and can be extracted directly.

[0117] b. Sampling point vector: Use the line-based sampling point generation tool in ArcGIS Pro to set a fixed sampling point interval based on the study area (500m in this example) and extract entity IDs.

[0118] c. Administrative district boundary vector: It contains the name of the administrative district and is extracted as a District entity.

[0119] (2) Attribute extraction

[0120] a. Road vector: Extract the Length attribute using computational geometry tools in ArcGIS Pro; road classification attributes can be obtained directly from the attribute table of the original data.

[0121] b. Sampling point vector: Use computational geometry tools to extract the Latitude and Longitude attributes.

[0122] (3)Relationship construction

[0123] Relationship construction includes the attribute relationships and spatial relationships of entities, namely <entity, relationship, attribute> and <entity, relationship, entity>:

[0124] a. Entity-attribute relationship (<entity, relationship, attribute>): According to the attribute content, extract the corresponding relationship names, such as Length, Latitude, Longitude, Class, etc. For example:

[0125] <RoadName, Length, 12.5km>

[0126] <ID, Latitude, 32.589>

[0127] <ID, Longitude, 122.65>

[0128] <RoadName, Class, Provincial Road>

[0129] Entity-entity relationship (<entity, relationship, entity>): Based on spatial analysis (such as spatial join, proximity analysis), determine the spatial relationship between entities. For example:

[0130] <ID, Located IN, District> (the administrative district to which the sampling point belongs)

[0131] <ID, Located IN, RoadName> (the road near the sampling point)

[0132] Finally, all entities, attributes, and relationships form an attribute table in ArcGIS Pro and are exported through Table to Excel. The table header is the entity (head node), Relationship 1, Relationship 2, Relationship 3..., and the table values are the attributes (tail nodes) and entities (tail nodes), forming a preliminary triple table.

[0133] 1.2 Entity, attribute extraction and relationship construction of raster layer

[0134] The extraction of entities, attributes, and relationships of the raster layer is based on the ArcGIS Pro platform and the Google Earth Engine platform.

[0135] (1)Entity extraction

[0136] In the raster layer, the following entities are mainly extracted:

[0137] a. Sampling point entity (ID): This step has been completed in the vector layer and can be reused directly.

[0138] b. Time Entities: Set the time nodes as 2000, 2010, and 2020 (the time span can be adjusted according to project needs).

[0139] (2) Attribute extraction

[0140] The attribute extraction process for raster data is consistent, and the main steps are as follows:

[0141] a. Upload the vector boundary of the study area in GEE. Based on project requirements, determine the necessary raster data, which in this example includes: Land Use; Land Surface Temperature (LST); Precipitation; Temperature; and Albedo.

[0142] b. Write JavaScript code in GEE to extract or invert raster data within the study area and download it to the local machine according to the time scales of 2000, 2010, and 2020.

[0143] c. In ArcGIS Pro, use the Extract Values ​​to Points tool to associate raster data values ​​with the sampling point ID entity.

[0144] (3) Relation extraction

[0145] The raster layer relationship extraction includes the attribute relationships of entities and the spatiotemporal relationships of entities, namely <entity, relation, attribute> and <entity, relation, entity>:

[0146] a. Entity-Attribute Relationship (<entity, relation, attribute>): Extract the corresponding relation name based on the attribute content, such as LST, LandUse, Temperature, Precipitation, etc. For example:

[0147] <ID, Length, 12.5km>

[0148] <ID, Temperature, 22.5℃>

[0149] <ID, Precipitation,856.22mm>

[0150] Entity-Entity Relationships (<entity, relationship, entity>): Determines spatial relationships between entities based on spatial analysis (such as spatial connectivity and nearest neighbor analysis). For example, it constructs temporal relationships between attributes and entities based on the year of raster data.

[0151] <ID, Located IN, District> (The administrative region to which the sampling point is located)

[0152] <ID, InRoad, RoadName> (Roads near the sampling point)

[0153] <22.5℃, Year, 2000) (Average temperature in 2000)

[0154] Finally, all entities, attributes, and relationships are generated into an attribute table in ArcGIS Pro and exported via a table to Excel converter. The table header contains entities (header nodes), relationships 1, 2, 3, etc., and the table values ​​contain attributes (tail nodes) and entities (tail nodes), forming a preliminary triplet table.

[0155] 1.3 Text-level entity, attribute extraction, and relation construction

[0156] The extraction of entities, attributes, and relationships at the text layer is based on the ArcGIS Pro and PyCharm platforms, and automated recognition is achieved using Natural Language Processing (NLP) technology. In this example, the BiLSTM-CRF and BERT models are primarily used to parse and extract text information.

[0157] (1) Entity extraction

[0158] The extraction of entities from the text layer mainly involves the store name (StoreName), and the process is as follows:

[0159] a. Data Preprocessing: The stored comment or description text is labeled and vectorized into words. A pre-trained BERT model is used to encode the text, generating a contextual representation for each word, enhancing the model's ability to understand entities.

[0160] b. Model Training: The BiLSTM-CRF model is trained using a labeled training set to learn to recognize store names. The CRF layer further optimizes the overall consistency of the label sequence, improving the model's accuracy in recognizing store names.

[0161] c. Entity Recognition: After training, use the BiLSTM-CRF model to predict the store names in the new text. For each input sentence, the model assigns a label to each word, and the words labeled as "StoreName" are recognized as store entities. (2) Attribute Extraction

[0162] The attribute extraction at the text layer mainly involves store category, sentiment tendency, location information, etc. Among them, the BERT model can be used for sentiment analysis, and BiLSTM-CRF is used for location information extraction.

[0163] The process of sentiment analysis (ReviewType) is as follows:

[0164] a. Text preprocessing, perform word segmentation (use the WordPiece word segmentation tool for Chinese).

[0165] b. Input the segmented text into the BERT model for encoding.

[0166] c. BERT performs sentiment classification based on the pre-training results and outputs categories such as negative reviews (Negative), positive reviews (Positive), and neutral reviews (Neutral).

[0167] Store attributes extract the specific store address (DetailLocation), store category (Class), and store longitude and latitude coordinates (Longitude, Latitude) based on BiLSTM-CRF.

[0168] (3) Relationship Construction

[0169] The relationship extraction at the text layer includes the attribute relationship and spatial relationship of entities, namely <entity, relationship, attribute> and <entity, relationship, entity>:

[0170] a. Entity-attribute relationship (<entity, relationship, attribute>): According to the attribute content, extract the corresponding relationship name, such as Class, ReviewType, Detail Location, etc. For example:

[0171] <StoreNameA, ReviewType, Negative Review>

[0172] <StoreNameA, Class, Chinese Food>

[0173] Entity-entity relationship (<entity, relationship, entity>): Based on spatial analysis (such as spatial connection, proximity analysis), determine the spatial relationship between entities. For example:

[0174] <StoreNameA, Located IN, District> (Store's District)

[0175] <StoreNameA, InRoad, RoadName> (Roads near the sampling point)

[0176] Finally, all entities, attributes, and relationships are generated into an attribute table in ArcGIS Pro and exported via a table to Excel converter. The table header contains entities (header nodes), relationships 1, 2, 3, etc., and the table values ​​contain attributes (tail nodes) and entities (tail nodes), forming a preliminary triplet table.

[0177] 1.4 Image layer entity, attribute extraction, and relation construction

[0178] Entities, attributes, and relationships in the image layer are extracted using ArcGIS Pro and PyCharm platforms, with automated recognition performed using the DeepLabV3+ semantic segmentation model.

[0179] (1) Entity extraction

[0180] In the image layer, the following entities are mainly extracted:

[0181] a. Sampling point entity (ID): This step has been completed in the vector layer and can be reused directly.

[0182] (2) Attribute extraction

[0183] The attribute extraction of the image layer is mainly determined based on the results of semantic segmentation and the project requirements. In this example, GVI (Green Visibility), BVI (Building Visibility), and VVI (Vehicle Visibility) are used as the specific attributes of the image layer. The specific extraction process is as follows:

[0184] a. Use sampling point data to crawl 360° panoramic images from recent years (the timeframe depends on project requirements and street view image data) based on the Baidu Street View platform;

[0185] b. Download the DeepLabV3+ model to your local machine, perform semantic segmentation on the crawled street view images of the study area, and output the semantically segmented images;

[0186] c. Calculate the percentage of different pixel values ​​in the entire image using PyCharm to obtain the GVI, BVI, and VVI of each sampling point ID.

[0187] (3) Relationship building

[0188] Image layer relation extraction includes entity attribute relations and entity spatial relations, namely <entity, relation, attribute> and <entity, relation, entity>:

[0189] a. Entity-Attribute Relationship (<entity, relation, attribute>): Extracts the corresponding relation name based on the attribute content, such as GVI, BVI, VVI, etc. For example:

[0190] <ID, GVI, 22.1>

[0191] <ID, VVI, 12.5>

[0192] Entity-Entity Relationship (<entity, relationship, entity>): Determines the spatial relationships between entities based on spatial analysis (such as spatial connectivity, nearest neighbor analysis). For example:

[0193] <ID, Located in, District> (The administrative district to which the sampling point is located)

[0194] <ID, InRoad, RoadName> (The road where the sampling point is located)

[0195] Finally, all entities, attributes, and relationships are generated into an attribute table in ArcGIS Pro and exported via a table to Excel converter. The table header contains entities (header nodes), relationships 1, 2, 3, etc., and the table values ​​contain attributes (tail nodes) and entities (tail nodes), forming a preliminary triplet table.

[0196] 1.5 Data Layer Entity Alignment

[0197] In the four data layers (vector, raster, text, and relational), each entity is associated with RoadName. Therefore, RoadName is used as the core entity to align the data across the four data layers on the PyCharm platform, ultimately constructing a knowledge representation in triple format. The core of data alignment is ensuring that entities in all data layers can be associated through RoadName to build a unified structured table. Since the core entities (ID, StoreName, etc.) of the vector, raster, and text layers are all associated with RoadName, RoadName is used as the primary index for data alignment.

[0198] During the data integration process, RoadName serves as the core entity, aligning the text, image, vector, and raster layers to construct a consistent relational dataset, which is ultimately converted into a triplet format. The specific data alignment steps are as follows:

[0199] (1): Create RoadName as the primary index

[0200] Since all entities in the four data layers can be associated with RoadName, RoadName is used as the primary index for alignment.

[0201] To ensure consistency in road names across different data sources and prevent matching failures due to spelling differences, a unified RoadName naming format is used. Therefore, all RoadNames are extracted from the vector layer to form a baseline index table, and information from other data layers is linked to this table.

[0202] (2): Text Layer Data Alignment

[0203] The address is used to match the corresponding RoadName. If the store has latitude and longitude information, spatial nearest neighbor analysis is used to associate it with the nearest RoadName.

[0204] (3): Image Layer Data Alignment

[0205] Align RoadName with geographic coordinates by linking the coordinates of the street view image to the nearest RoadName.

[0206] (4): Vector Layer Data Alignment

[0207] Vector data itself contains RoadName information, which can be directly used as the primary index for alignment. Spatial join is used to determine information such as sampling points (IDs) and administrative districts associated with the RoadName.

[0208] (5): Raster Layer Data Alignment

[0209] In the raster layer, raster data is typically bound to sample points (IDs), and these sample points (IDs) are aligned with the RoadName through the vector layer.

[0210] (6): Unify storage format and output, forming a unified data table based on RoadName. That is, using RoadName as the main index, the attributes of the four data layers are stored in the same table, as shown in the table below:

[0211] Attribute table

[0212]

[0213] Converting to triple format (<entity, relation, attribute> or <entity, relation, entity>) using PyCharm is as follows:

[0214] <StoreName, located_in, RoadName>

[0215] <ID, located_in, RoadName>

[0216] <ID, near, RoadName>

[0217] <RoadName, Temperature, 25.6℃>

[0218] This will ultimately lead to the formation of a multi-source geospatial big data knowledge graph.

[0219] Step 4: Based on the obtained multi-source geospatial big data knowledge graph, use the Neo4j graph database and Py2neo interface to perform data query, calculation and visualization applications.

[0220] Specifically: Based on Neo4j + Py2neo, it enables querying, calculation, and visualization of multi-source geospatial data. The specific steps are as follows:

[0221] A. Data Query

[0222] The Cypher query language was used to perform data retrieval in the Neo4j graph database to extract information such as roads, administrative districts, shops, transportation, environment, and land use.

[0223] The search methods include:

[0224] Location-based queries (e.g., distribution of high-temperature roads and shops within a certain area);

[0225] Time-based queries (e.g., land use trends over the past 20 years).

[0226] Queries based on multi-source data association (e.g., the relationship between temperature change and green coverage).

[0227] B. Data Calculation

[0228] Aggregate calculations (such as calculating the average temperature in a certain area or the density of shops in a certain block);

[0229] Time series analysis (e.g., analyzing land use change based on three time points: 2000, 2010, and 2020);

[0230] Spatial analysis (such as calculating the greening index and building visibility index of a street).

[0231] C. Data Visualization

[0232] By combining Python data visualization libraries (such as Matplotlib and Seaborn) and GIS visualization tools (such as Leaflet and ArcGIS), query results can be presented intuitively, generating heat maps, time-series change maps, distribution maps, etc.

[0233] It should be noted that the data query, calculation and visualization applications based on the Neo4j graph database and Py2neo interface are divided into 5 application scenarios, namely urban heat island effect analysis, urban green view rate and building density analysis, urban travel recommendation, road network and traffic flow analysis, and urban land use change detection.

[0234] like Figure 2 As shown, this embodiment selects Wuhan City for experimental verification based on data availability considerations. The spatial range of the example data is as follows. Figure 2 As shown. In this embodiment, DataV, DataR, DataT, and DataI are all unified within the research area of ​​Wuhan City.

[0235] This embodiment pre-defines five smart city applications: urban heat island effect analysis, urban green view rate and building density analysis, urban travel recommendation, road network and traffic flow analysis, and urban land use change detection.

[0236] in:

[0237] The main application of urban heat island effect analysis is to analyze urban heat island regions based on LST data from the Raster Layer. The query approach involves using Python and a multi-source geospatial big data knowledge graph to query temperature nodes from different years (2000, 2010, 2020) to identify high-temperature areas and provide cooling strategies for urban planning.

[0238] Query example (retrieving the LST value of a specific road in 2020):

[0239] MATCH (road)-[r:LST2020]->(temp)

[0240] WHERE temp.value IS NOT NULL

[0241] RETURN road.name AS RoadName, temp.value AS Temperature

[0242] Visualization:

[0243] Road temperature distribution map (showing the highest and lowest temperatures of the 10 roads); heat island trend map (analyzing the expansion of the heat island in different years), obtained by querying the graph database using py2neo nodes. Figure 6-8 .

[0244] The main application of urban green view rate and building density analysis is to analyze the relationship between green view rate and building view rate in different areas of the city based on GVI and BVI data from the Image Layer, facilitating subsequent optimization of urban layout. The query approach involves using Python and a multi-source geospatial big data knowledge graph to query the GVI and BVI values ​​of sampled points and obtain visualized results.

[0245] Query example (to obtain the GVI / BVI and geographical location of each road):

[0246] MATCH (road)-[:GVI]->(gvi_node)

[0247] MATCH (road)-[:BVI]->(bvi_node)

[0248] MATCH (road)-[:Longitude]->(lon_node)

[0249] MATCH (road)-[:Latitude]->(lat_node)

[0250] RETURN road.name AS RoadName, gvi_node.value AS GVI, bvi_node.valueAS BVI, lon_node.value AS Longitude, lat_node.value AS Latitude

[0251] Visualization:

[0252] Spatial distribution map of GVI / BVI; statistical chart of average GVI / BVI for different administrative regions, see example. Figure 9 .

[0253] The primary application of urban travel recommendation scenarios involves analyzing the distribution of positive reviews for stores within a city based on sentiment analysis data from the Text Layer, thus laying the foundation for downstream travel recommendations. The query approach utilizes Python and a multi-source geospatial big data knowledge graph to query RiverType relationships and create a basic distribution map. The returned data is then subjected to positive review rate analysis at the road scale and by store category.

[0254] Example query (search for rating categories of nearby shops):

[0255] MATCH (store)-[:NearRoad]->(road)

[0256] MATCH (store)-[:Class]->(class)

[0257] MATCH (store)-[:ReviewType]->(review)

[0258] RETURN road.name AS RoadName, class.name AS Category, review.value ASReviewType

[0259] Visualization:

[0260] A heatmap of city shop reviews, showing the percentage of positive reviews by street and shop type. The results are as follows: Figure 10 As shown.

[0261] The primary application of road network and traffic flow analysis is to analyze traffic flow on urban streets based on Image Layer VVI data, facilitating the optimization of traffic management strategies. The query approach involves using Python and multi-source geospatial big data knowledge graphs to query VVI relationships and create a basic distribution map, then performing road-scale traffic flow analysis on the returned data.

[0262] Query example (to retrieve the VVI values ​​and geographic information of each road):

[0263] MATCH (road)-[:VVI]->(vvi_node)

[0264] MATCH (road)-[:Longitude]->(lon_node)

[0265] MATCH (road)-[:Latitude]->(lat_node)

[0266] RETURN road.name AS RoadName, vvi_node.value AS VVI, lon_node.valueAS Longitude, lat_node.value AS Latitude

[0267] Visualization:

[0268] Urban traffic flow heat map (showing high and low flow areas); traffic flow trend map (displaying traffic flow trends in different areas), results as follows: Figure 11 As shown.

[0269] The primary application of urban land use change monitoring is to analyze the inflow and outflow of land use in various districts over time based on time-series land use data from the Raster Layer, providing data support for land resource management. The query approach involves using Python and a multi-source geospatial big data knowledge graph to query land use relationships to obtain the number of sampling points for each type of land in each district, and then using Python to generate land use inflow and outflow maps for different years.

[0270] Query example: MATCH (road)-[:Landuse2000]->(landuse2000)

[0271] MATCH (road)-[:Landuse2010]->(landuse2010)

[0272] MATCH (road)-[:Landuse2020]->(landuse2020)

[0273] RETURN road.name AS RoadName, landuse2000.value AS LandUse2000,landuse2010.value AS LandUse2010, landuse2020.value AS LandUse2020

[0274] Visualization:

[0275] Land use change heat map (showing land use changes from 2000 to 2020); land use conversion statistics for different administrative regions, with the final results as follows: Figure 12 As shown.

[0276] It is worth noting that the application of multi-source geospatial big data knowledge graphs in smart cities is not limited to the five types mentioned above, and the specific implementation can be determined according to the task requirements and the situation of the research area.

[0277] Based on the same technical concept as the above embodiments, the present invention provides a multi-source geospatial big data knowledge graph construction system for smart city planning, comprising:

[0278] The data collection and processing module is used to collect raw data from the vector layer, raster layer, text layer, and image layer, and to perform preprocessing.

[0279] The ontology layer construction module is used to construct a vector entity set E based on the preprocessed vector layer. v Vector relation set A v and vector attribute set R v Construct a raster entity set E based on the preprocessed raster layers. r Grid relation set A r and the raster attribute set R r A text entity set E is constructed based on the preprocessed text layer. t Text relation set A t and text attribute set R t The image entity set E is constructed based on the preprocessed image layers. i Image Relationship Set A i and the image attribute set R i Using RoadName as the reference entity, the vector entity set E v Grid entity set E r Text entity set E t and the image entity set E i Alignment is performed using a unified reference entity to obtain the entity set E={E v E r E t E i Based on the obtained entity set E, the relation set A and attribute set R are obtained, forming the knowledge graph ontology layer;

[0280] The knowledge graph construction module is used to extract entity sets, relation sets, attribute sets, and alignment methods from various data layers in the target city based on the set and alignment of the knowledge graph ontology layer, using the DeepLabV3+ semantic segmentation model, NLP model, and GIS platform based on ArcGIS Pro, Google Earth Engine, and PyCharm platform, to form a multi-source geospatial big data knowledge graph.

[0281] It should be noted that this system also includes an application module, which is used to perform data querying, calculation and visualization applications based on the obtained multi-source geospatial big data knowledge graph, using graph databases and interfaces.

[0282] In practical applications, taking Wuhan as the target city, the vector layer, raster layer, text layer, and image layer of Wuhan are imported into the data collection and processing module. After processing by this module, the vector data, raster data, text data, and image data of Wuhan are obtained.

[0283] Then, using RoadName as the reference entity, the vector data, raster data, text data, and image data of Wuhan City are integrated and aligned using the DeepLabV3+ semantic segmentation model, NLP model, and GIS platform based on ArcGIS Pro, Google Earth Engine, and PyCharm to obtain a multi-source geospatial big data knowledge graph based on Wuhan City.

[0284] Finally, based on the requirements, the application module retrieves the necessary data from the multi-source geospatial big data knowledge graph through the Neo4j graph database and Py2neo interface for data querying, calculation, and visualization.

[0285] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for constructing a multi-source geospatial big data knowledge graph for smart city planning, characterized in that, include: The raw data of the vector layer, raster layer, text layer, and image layer are collected and preprocessed to obtain vector data, raster data, text data, and image data, respectively. The vector data includes at least road data and administrative region data in the study area. The raster data includes remote sensing data in the study area required by the actual application scenario and remote sensing data in the study area already available on the platform. Constructing a knowledge graph ontology layer includes: constructing a vector entity set, a vector attribute set, and a vector relationship set based on vector data; constructing a raster entity set, a raster attribute set, and a raster relationship set based on raster data; constructing a text entity set, a text attribute set, and a text relationship set based on text data; constructing an image entity set, an image attribute set, and an image relationship set based on image data; aligning the vector entity set, raster entity set, text entity set, and image entity set with street names as reference entities to obtain an entity set; and determining the attribute set and relationship set based on the entity elements in the obtained entity set, their positional relationships, and their attributes. The construction of the image entity set, image attribute set, and image relationship set includes: marking geographic entities in the image data based on the sampling point IDs along the road within the study area to form an image entity set; using the deep learning model DeepLabV3+ to extract the green visibility rate, building visibility rate, and vehicle visibility rate in the image data to form an image attribute set; and establishing the geographical location relationship between the sampling points along the street and their respective roads, administrative divisions, and latitude and longitude to form an image relationship set. Based on the sets and alignment methods constructed in the ontology layer of the knowledge graph, the DeepLabV3+ semantic segmentation model, the Natural Language Processing (NLP) technology model, and the GIS platform are used to extract the entity sets, relation sets, attribute sets, and alignment methods from the vector layer, raster layer, text layer, and image layer of the target city, forming a multi-source geospatial big data knowledge graph. The GIS platform serves as a spatial analysis tool for extracting entities, attributes, and relations from the vector layer and the raster layer. The NLP technology model is used for extracting entities, attributes, and relations from the text layer.

2. The method for constructing a multi-source geospatial big data knowledge graph for smart city planning according to claim 1, characterized in that: Constructing the vector entity set, vector attribute set, and vector relation set includes: Extract road names, OSM identifiers, and administrative divisions from vector data to form a vector entity set; Calculate the topological and geometric attributes of each entity element in the vector entity set to form a vector attribute set. The topological attributes include at least adjacent roads and road lengths, and the geometric attributes include at least latitude and longitude. Establish spatial and attribute relationships for each entity element in the vector entity set, including category, length, nearest road, Euclidean distance, and administrative division, to form a vector relationship set.

3. The method for constructing a multi-source geospatial big data knowledge graph for smart city planning according to claim 1, characterized in that: Constructing the grid entity set, grid attribute set, and grid relationship set includes: Based on the sampling point IDs generated along the roads within the study area, and combined with the time field, a raster entity set is formed; The raster attribute values ​​of each raster entity element in the raster entity set of remote sensing data within the study area are extracted to form a raster attribute set; Extract the geographical location of each raster entity element in the raster entity set from the remote sensing data within the study area, and establish the geographical location relationship of each raster entity element in the raster entity set to form a raster relationship set.

4. The method for constructing a multi-source geospatial big data knowledge graph for smart city planning according to claim 1, characterized in that: The text data includes store review data that can be segmented into latitude and longitude using natural language processing techniques.

5. The method for constructing a multi-source geospatial big data knowledge graph for smart city planning according to claim 4, characterized in that: Constructing the text entity set, text relation set, and text attribute set includes: The names of shops within the research area are extracted from the text data to form a set of text entities; The text attributes of each text entity element in the text entity set are extracted from the text data to form a text attribute set A. t ; Based on the store names in the text data, construct a set of text relationships by establishing the geographical relationships between the stores and the nearest roads, the administrative divisions they are located in, and their latitude and longitude.

6. The method for constructing a multi-source geospatial big data knowledge graph for smart city planning according to claim 1, characterized in that: The image data includes image data from the study area from which green visibility, building visibility, and vehicle visibility can be extracted.

7. A multi-source geospatial big data knowledge graph construction system for smart city planning, characterized in that, include: The data collection and processing module is used to collect raw data from the vector layer, raster layer, text layer, and image layer, and preprocess them to obtain vector data, raster data, text data, and image data, respectively. The vector data includes at least road data and administrative region data in the study area. The raster data includes remote sensing data in the study area required by the actual application scenario and remote sensing data in the study area already available on the platform. The ontology layer construction module is used to construct vector entity sets, vector attribute sets, and vector relationship sets based on vector data; raster entity sets, raster attribute sets, and raster relationship sets based on raster data; text entity sets, text attribute sets, and text relationship sets based on text data; and image entity sets, image attribute sets, and image relationship sets based on image data. Using street names as reference entities, the vector entity sets, raster entity sets, text entity sets, and image entity sets are aligned to obtain entity sets. Based on the entity elements in the obtained entity sets, their positional relationships, and attributes, attribute sets and relationship sets are determined, thus constructing the knowledge graph ontology layer. The construction of the image entity sets, image attribute sets, and image relationship sets includes: labeling geographic entities in the image data based on the IDs of roadside sampling points within the study area to form an image entity set; extracting green visibility, building visibility, and vehicle visibility from the image data using the DeepLabV3+ deep learning model to form an image attribute set; and establishing the geographical relationships between streetside sampling points and their respective roads, administrative divisions, and latitude and longitude to form an image relationship set. The knowledge graph construction module is used to extract entity sets, relation sets, attribute sets, and alignment methods from the vector layer, raster layer, text layer, and image layer of the target city based on the sets and alignment methods constructed in the ontology layer of the knowledge graph. This is achieved using the DeepLabV3+ semantic segmentation model, the natural language processing (NLP) technology model, and the GIS platform on a software platform, forming a multi-source geospatial big data knowledge graph. The GIS platform serves as a spatial analysis tool for extracting entities, attributes, and relations from the vector layer and the raster layer. The NLP technology model is used for extracting entities, attributes, and relations from the text layer.

Images