Natural resource spatio-temporal knowledge graph construction method combining AI (Artificial Intelligence) and GIS (Geographic Information System)

Abstract

The invention discloses an AI and GIS combined natural resource space-time mapping knowledge domain construction method, which is suitable for intelligent mapping and dynamic management of natural resource space units. GIS data are acquired, multi-scale spatial range automatic identification is carried out, and map nodes are constructed; then, introducing crowd LBS behavior data, and establishing an edge relationship between dynamic nodes; a spatial topological structure between nodes is automatically learned by adopting a graph convolutional network model, and intelligent classification and correlation prediction between spatial units are realized; extracting node attribute codes, service type codes and state label codes, constructing a two-dimensional code data string, generating a standard two-dimensional code image through error correction codes and formatted information, and embedding the standard two-dimensional code image into node data; and a terminal user can access the atlas data by scanning a code and trigger permission verification and call log records. According to the method, automatic space structure identification, dynamic map updating and lightweight calling are realized, and the intelligence, integration and safety of natural resource data management are improved.

Description

Technical Field

[0001] The present invention belongs to the interdisciplinary technical field of natural resource management, geographic information system (GIS), artificial intelligence image recognition and knowledge graph construction, and specifically relates to a method for constructing a spatiotemporal knowledge graph of natural resources by combining AI and GIS. Background Art

[0002] With the ongoing digital transformation of natural resource management, national land space planning, and ecological protection, intelligent modeling and knowledge integration based on spatial data have become key technological paths. Traditional natural resource data is primarily stored in the form of static vector data, business tables, images, or documents across multiple business systems. This lack of unified data representation and efficient association mechanisms makes it difficult to integrate and analyze information across regions, time periods, and business types.

[0003] While geographic information systems (GIS) provide spatial data visualization and basic query capabilities, they lack support for dynamic evolutionary behavior, spatiotemporal topological relationships, and semantic expression, making it difficult to meet the needs of scenarios requiring intelligent identification, associated invocation, and automated reasoning. Traditional knowledge graph construction methods, while relatively mature in the areas of natural language and entity extraction, lack deep integration with geographic spatial data. This is particularly true for the automated generation of graph structures for natural resource spatial units, dynamic attribute updates, and multi-scale modeling, which present significant technical gaps.

[0004] Some existing research attempts to model the spatiotemporal attributes of spatial units such as plots and blocks based on business spatial databases. These methods, for example, employ buffer zone analysis, regional overlays, and spatial clustering to determine inter-regional associations. However, most of these methods rely on manual rules and lack automatic learning capabilities, making them difficult to adapt to large-scale, heterogeneous, multi-source, and dynamic natural resource data scenarios. Furthermore, a unified technical framework that integrates AI models, GIS spatial data, and knowledge graph representation structures has yet to be established, encompassing solutions from graph node identification and edge relationship reasoning to dynamic updates, rapid invocation, and visual access. Summary of the Invention

[0005] An embodiment of the present invention solves the technical problem of how to use artificial intelligence methods to automatically identify spatial units of different scales in large-scale natural resource spatial data, and establish spatiotemporal graph nodes and their structural association relationships with semantic and behavioral relevance.

[0006] Another embodiment of the present invention solves how to realize automatic classification and spatial association judgment of natural resource spatial units based on image visual features and graph neural networks, avoiding reliance on manual labels or static rules.

[0007] Another embodiment of the present invention solves how to encapsulate and bind the data of the natural resource spatiotemporal knowledge graph nodes in the form of QR codes, supports rapid retrieval of graph content on the terminal, and implements permission control and access behavior recording.

[0008] The present invention provides the following technical solution: a method for constructing a spatiotemporal knowledge graph of natural resources by combining AI and GIS, comprising the following steps:

[0009] Acquire natural resource business spatial data and attribute data, and build a business spatial database;

[0010] Extracting vector outlines of the spatial data based on a geographic information system platform and spatially matching them with administrative boundaries; automatically identifying and associating multi-scale spatial ranges of the natural resource spatial data using an AI image recognition algorithm; and using the multi-scale spatial data and business attribute data as machine learning training features;

[0011] Construct and train a spatial association recognition model based on graph convolutional neural networks to automatically identify the spatial association relationships between natural resource spatial units of different scales to form the spatial topological structure of nodes and edges of the spatiotemporal knowledge graph.

[0012] Furthermore, the method further comprises:

[0013] Obtain real-time crowd location information within natural resource spatial units, extract each unit's volume ratio, greening rate, average building height, maximum building height, and building density as machine learning input features, and establish real-time crowd spatial behavior patterns;

[0014] Through machine learning models, the spatiotemporal function relationship between real-time crowd location information and spatial unit attributes is analyzed, and the temporal labels of the spatiotemporal knowledge graph nodes are dynamically adjusted to achieve dynamic updating of the knowledge graph.

[0015] Furthermore, the automatic identification of spatial associations between natural resource spatial units of different scales to form a spatial topological structure of nodes and edges of a spatiotemporal knowledge graph includes:

[0016] Based on regional-scale buffer analysis, natural resource spatial units with spatial correlation relationships are automatically determined, and when the buffer overlap rate exceeds 50%, the regional attributes of the spatial units are used as machine learning training features;

[0017] Construct a temporal association model based on a temporal convolutional network, combine the spatial association recognition results with the temporal model, realize the automatic association call of business information between natural resource spatial units at different time scales, and form a cross-scale temporal topological relationship of the spatiotemporal knowledge graph.

[0018] Furthermore, the analysis method of the real-time crowd spatial behavior pattern is: when the real-time stay in the target spatial unit is more than 10 minutes and the number of people is not less than 100, it is automatically determined that there is a spatiotemporal correlation between the spatial unit and the adjacent spatial units, forming a real-time spatiotemporal knowledge graph node of the spatial unit.

[0019] Furthermore, the buffer zone analysis uses a 500-meter buffer zone outside the boundary, and automatically determines other spatial units that have regional associations with the target spatial unit through image recognition algorithms and GIS spatial overlay analysis, and generates a spatial association set.

[0020] Furthermore, the construction process of the spatial association recognition model is as follows: randomly extracting natural resource spatial unit image data, establishing a training set, a validation set, and a test set, using a graph convolutional network to extract the image's contour, color, shadow, and topological relationship features and train the model, and automatically classifying and determining the spatial unit association relationship.

[0021] Furthermore, the data of the nodes and edges of the spatiotemporal knowledge graph are integrated and embedded in the natural resource space unit attribute data in the form of QR codes. After scanning the QR code, the spatiotemporal knowledge graph data can be directly called and leave a trace of permission call.

[0022] Furthermore, the method for generating the QR code data is: the spatial scale code, business type code and real-time time series label code of each natural resource spatial unit are binary converted and connected, the data code is generated after adding the padding code and error correction code, and then the positioning pattern, formatted data and version information are added to draw together to generate the QR code.

[0023] The beneficial effects of the present invention are:

[0024] 1. This invention combines GIS contour matching with AI image recognition algorithms to achieve automatic contour recognition and scale unification of natural resource spatial units, reducing manual dependence and improving mapping efficiency and accuracy.

[0025] 2. The present invention introduces crowd residence behavior as a criterion to judge the activity correlation between spatial units in real time, so that the knowledge graph nodes have dynamic response capabilities and are more in line with real scenarios.

[0026] 3. Through the graph convolution model, the semantic and functional connections between spatial units are mined to achieve automatic learning and structural optimization of spatial relationships, thereby enhancing the intelligent expression ability of the graph.

[0027] 4. Use graph convolutional networks to perform end-to-end learning of image features such as contours, shadows, and topology, to achieve automatic classification of spatial units and association with graph structures, thereby improving recognition accuracy and generalization capabilities.

[0028] 5. Encapsulate graph node information into a QR code to enable terminal scanning and access control; it has error correction and formatting capabilities to ensure the stability and traceability of data calls, facilitating deployment and integration. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The present invention will be described in more detail below based on embodiments and with reference to the accompanying drawings, wherein:

[0030] Figure 1 A schematic flow chart of the method provided by the present invention;

[0031] Figure 2 Schematic diagram of the target block, associated blocks, and adjacent blocks in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0033] like Figure 1 The flow chart of a method for constructing a spatiotemporal knowledge graph of natural resources by combining AI and GIS is shown. The method includes the following steps:

[0034] Obtain natural resource business spatial data and attribute data (specifically, obtain them in the GIS platform) and build a business spatial database;

[0035] Extracting vector outlines of the spatial data based on a geographic information system (GIS) platform and spatially matching them with administrative boundaries, automatically identifying and associating the multi-scale spatial ranges of the natural resource spatial data using an AI image recognition algorithm, and using the multi-scale spatial data and business attribute data as machine learning training features;

[0036] A spatial association recognition model based on graph convolutional neural network (GCN) is constructed and trained to automatically identify the spatial association relationship between natural resource spatial units of different scales to form the spatial topological structure of nodes and edges of the spatiotemporal knowledge graph.

[0037] As an example of the present invention, after obtaining natural resource business spatial data and attribute data and building a business spatial database, the vector outline of the spatial data is extracted based on a geographic information system (GIS) platform and spatially matched with administrative boundaries. The multi-scale spatial range of the natural resource spatial data is automatically identified and associated through an AI image recognition algorithm. The multi-scale spatial data and business attribute data are used together as machine learning training features, specifically including:

[0038] Step 1: GIS vector contour data extraction

[0039] 1.1. Import natural resource spatial business data into the GIS platform, including remote sensing images, drone photography, and surveying and mapping vector data (such as building outlines and land use boundaries).

[0040] / 1.2. Use GIS platform vectorization tools (such as Spatial Analyst in ArcGIS or the Vector Outline Extraction plug-in in QGIS) to extract the boundary outlines of spatial data.

[0041] Use boundary extraction algorithm to extract boundaries:

[0042]

[0043] Among them, Boundary(i,j) is the labeling result of whether the image or vector data at the spatial position (i,j) is a boundary; f(i,j) is the classification or attribute value at the spatial position (such as land use type or administrative area label).

[0044] Step 2: Spatial matching of vector outlines and administrative boundaries

[0045] 2.1. Obtain administrative boundary data at various scales (national, provincial, municipal, district, county, and town scales) and import them into the GIS platform;

[0046] 2.2. Based on vector topological spatial overlay analysis, match the contour data with the administrative boundaries.

[0047] Topological space matching is to define the contour space unit S x Administrative Boundary Unit A y The spatial superposition matching degree Overlap(S x ,A y ):

[0048]

[0049] Among them, Area(S x ∩A y ) is the intersection area of the contour space unit and the administrative boundary; Area(S x) contour space unit area.

[0050] If Overlap(S x ,A y )≥0.5, then the contour space unit S x Belongs to administrative boundary unit A y

[0051] Step 3: AI image recognition algorithm automatically identifies and associates multi-scale spatial ranges

[0052] 3.1 The vector contour data obtained in the above steps and the matched administrative division space units are processed into images (converted into images of uniform scale);

[0053] 3.2 Training Convolutional Neural Network (CNN) for image feature recognition:

[0054] Assume that the input image feature mapping function is F CNN , the input image I is mapped by convolution and activation function to obtain the feature vector V of spatial image data feat :

[0055] V feat =F CNN (I; θ CNN );θ CNN is the CNN network weight parameter.

[0056] Step 4: Multi-scale spatial range machine learning feature construction

[0057] 4.1 Extract the multi-scale spatial features and business attribute features of each spatial unit and generate a fusion feature vector

[0058]

[0059] Specifically, It is also the machine learning input feature vector of the i-th spatial unit; A attr is the business attribute feature vector (such as volume ratio, building density); Scale id Encodes the scale of the spatial unit.

[0060] Step 5: Automatic identification and association of spatial ranges between spatial units

[0061] Based on the fusion feature vector obtained in step 4 Construct a graph convolutional network (GCN) to automatically identify the association between different spatial units. Based on the feature representation of the nodes in the first layer, it automatically identifies and associates the feature representation of the l+1 layer:

[0062]

[0063] in, A is the original adjacency matrix, which is an N×N matrix (N is the number of nodes in the graph). ij =1 means that nodes i and j are connected by an edge, A ij = 0 means that the relationship between the two has not changed. The matrix A reflects the original connection structure of the graph. N It is an N×N unit matrix with all diagonal elements being 1 and all other position elements being 0. The purpose of adding this unit matrix is to allow each node to retain its own characteristics during information propagation. The adjacency matrix with self-loops added indicates that each node in the original graph has an edge pointing to itself, i.e., "original graph + self-connection". In the automatic recognition method of graph convolutional networks, this allows each node to not only receive the features of its neighbors during convolution but also retain its own features, thus avoiding information loss. is the node degree matrix, W (l) is the weight matrix of the lth layer; σ is the activation function (such as ReLU function).

[0064] By continuously iterating and stacking multiple layers in step 5, the feature representation result of automatic recognition and association is finally output, Z = H (L) .

[0065] The final classification result C(i,j) of automatic spatial association recognition is:

[0066]

[0067] Among them, Z i 、Z j are the feature representations of nodes i and j after GCN output; cos(Z i ,Z j ) is the cosine similarity between the output feature vectors of two nodes, Threshold is the cosine similarity threshold for determining spatial association, preferably, Threshold = 0.8.

[0068] In the above-mentioned automatic identification of spatial associations between multi-scale spatial units using a graph convolutional neural network (GCN), the specific identification method between spatial units or the method for determining the scale range are not clearly defined. Therefore, as a further limitation of the present invention, the method further includes:

[0069] Obtain real-time crowd location information within natural resource spatial units, extract each unit's volume ratio, greening rate, average building height, maximum building height, and building density as machine learning input features, and establish real-time crowd spatial behavior patterns;

[0070] Through machine learning models, the spatiotemporal function relationship between real-time crowd location information and spatial unit attributes (volume ratio, greening rate, average building height, maximum building height and building density of each unit) is analyzed, and the temporal labels of the spatiotemporal knowledge graph nodes are dynamically adjusted to achieve dynamic updating of the knowledge graph.

[0071] Specifically include:

[0072] Step 1: Construct a set of spatial unit attribute feature vectors and a set of crowd activity data

[0073] F s (i) = [FAR i ,LR i ,H avg,i ,H max,i ,ρ i ]; among them, F s (i) is the attribute feature vector set of the i-th spatial unit, FAR i is the ratio of the total building area in the i-th spatial unit to the total land area of the unit; LR i is the ratio of green area to total land area in the spatial unit, H avg,i 、H max,i are the average height of all buildings in the i-th spatial unit and the height of the tallest building, ρ i It is the ratio of the sum of the base areas of all buildings to the land area (reflecting the density of buildings per unit area).

[0074] Crowd activity data set H t (i)

[0075] H t (i)=[N i (t),D i (t)]; N i (t) is the number of people in the ith spatial unit, D i (t) is the average length of stay per person in the i-th spatial unit.

[0076] Step 2: Establish a spatiotemporal function relationship model to predict node state label changes

[0077] T i (t+Δt)=f(F s (),H t (i)); T i (t+Δt) is the time series label of the i-th space at time t+Δt (including activity level, functional status and / or traffic flow level), f(F s (i),H t (i)) is the set of spatial unit attribute feature vectors F obtained through training s(i) With the crowd activity data set H t (i) mapping function between;

[0078] Step 3: Dynamically update the time series labels to the knowledge graph nodes

[0079] 3.1 Set each node v i The time series label T at time t+Δt i (t+Δt) is written into the graph node attributes.

[0080] 3.2 If the state changes significantly (such as exceeding a certain threshold ∈), the topology or label update of the graph structure is triggered:

[0081] Through the above steps, the spatiotemporal coupling relationship between real-time dynamic crowd behavior data and static spatial physical properties is utilized to drive the intelligent evolution of graph node states, forming a natural resource spatiotemporal knowledge graph with time-series self-adaptation capabilities.

[0082] As a further limitation of the present invention, the automatic identification of spatial associations between natural resource spatial units of different scales to form a spatial topological structure of nodes and edges of a spatiotemporal knowledge graph includes:

[0083] Based on regional-scale buffer analysis, spatially related natural resource units are automatically identified. Specifically, vector boundary data for each natural resource unit (e.g., block, plot, or district) is imported into a GIS platform. For each target unit, a buffer zone is generated by extending its boundary outward by a fixed distance (e.g., 500 meters). Based on this, all other units that overlap with the buffer zone are identified, and the geometric overlap ratio between these units and the buffer zone is calculated.

[0084] And on the basis that the buffer overlap rate exceeds 50%, that is, for those spatial units whose buffer overlap rate with the target unit exceeds 50%, the system automatically determines that they have a spatial association relationship with the target unit. This judgment method ensures that only adjacent units with actual spatial influence or functional interaction possibility are retained, while non-related units with accidental contact are excluded. The regional attributes of spatial units are used as machine learning training features. Specifically, after identifying a set of spatial units with spatial association relationships, the system further extracts the regional attribute information of these units (including volume ratio, building density, greening rate and / or building height), and constructs these attributes into standardized feature vectors for use as input to subsequent machine learning models.

[0085] The purpose of building a temporal association model based on a temporal convolutional network (TCN) is to further explore the behavioral evolution trends and business associations of these spatial units across different time dimensions. The system introduces a temporal convolutional network model. Based on the data structure of time series, the TCN model can capture the attribute changes of the same spatial unit at different times (such as daily, weekly, and monthly scales), or the evolution of the same attribute across multiple spatial units.

[0086] In actual operation, the spatial association recognition results are combined with the time series model to realize the automatic association call of business information between natural resource spatial units at different time scales. Specifically, the system uses the above-mentioned identified spatial association relationship as the connecting edge of the input graph structure, thereby forming a cross-scale time series topological relationship of the spatiotemporal knowledge graph.

[0087] The time series of regional attributes is used as input node features and fed into the TCN model for training and inference. The model can learn the delayed response relationship, covariation relationship, and trend prediction capabilities between spatial units in the temporal dimension.

[0088] Ultimately, the TCN output is used to construct a dynamic, cross-scale spatiotemporal knowledge graph topology. Each node in the graph represents a natural resource spatial unit, and each edge indicates a significant connection in space or time. This graph structure allows users to automatically access historical or neighboring business information related to the target area at any point in time, enabling efficient and automated natural resource planning, supervision, and assessment.

[0089] Regional-scale buffer analysis (500-meter buffer area, overlap rate greater than 50%) was used to determine which spatial units were spatially correlated, and the geographic spatial scale and specific threshold conditions for automatically identifying the correlation between spatial units were clearly given.

[0090] In addition, a temporal convolutional network (TCN) is used to mine the correlation information between spatial units in the time dimension, and a clear cross-scale temporal correlation model is constructed.

[0091] In the process of constructing the spatiotemporal knowledge graph of natural resources, relying solely on geographic adjacency or static attributes may not accurately reflect the interaction intensity between actual spatial units. The introduction of the real-time crowd behavior judgment standard of "staying time ≥ 10 minutes and number of people ≥ 100 people in the target spatial unit" is to dynamically identify spatial units with potential functional coupling or behavioral connections. This judgment condition can effectively exclude random passing behaviors and regard high-density crowd residence areas as active nodes with significant spatiotemporal effects, thereby supporting the timeliness and authenticity of graph node updates. This mechanism ensures that the graph structure has real-time response capabilities in rapidly changing urban spaces, which helps to improve the dynamic accuracy of graph modeling. Therefore, as a further limitation of the present invention, the analysis method of the real-time crowd spatial behavior pattern is:

[0092] When the real-time stay time in the target spatial unit is more than 10 minutes and the number of people is not less than 100, it is automatically determined that there is a spatiotemporal correlation between the spatial unit and the adjacent spatial units, forming a real-time spatiotemporal knowledge graph node of the spatial unit.

[0093] By setting the crowd residence criteria of "more than 10 minutes and more than 100 people", the system can effectively identify key nodes that exhibit high-intensity spatial behavior within a specific time period and automatically incorporate them into the spatiotemporal knowledge graph. This mechanism not only improves the behavioral representativeness of the nodes in the graph, but also strengthens the graph structure's ability to perceive dynamic changes in the crowd. The dynamic update of graph nodes can be used to support scenarios such as emergency response, regional function identification, and migrant population management, significantly improving the level of intelligence in urban governance and natural resource allocation. At the same time, this mechanism can reduce dependence on static data, realize the transformation of the knowledge graph from "graphic geography" to "behavioral geography", and enhance the practicality and predictive ability of the model.

[0094] In the modeling of spatiotemporal relationships of natural resources, the traditional way of judging spatial associations by "whether they are directly adjacent" is too rough and difficult to capture potential functional connections. The introduction of a buffer zone analysis of 500 meters outside the boundary can expand the judgment range in spatial geometry, and combine image recognition with GIS overlay analysis to more accurately identify spatial units that are adjacent but not connected but have potential for interaction. This step is an important prerequisite for constructing a real spatial interaction map. Therefore, as a further limitation of the present invention, the buffer zone analysis uses a 500-meter buffer area outside the boundary, and through image recognition algorithms and GIS spatial overlay analysis, automatically determines other spatial units that have regional associations with the target spatial unit, and generates a spatial association set.

[0095] Natural resource spatial units usually have problems of diverse shapes, different scales, and complex structures. Traditional methods that rely on manual rules or geometric adjacency to judge association relationships are difficult to adapt to large-scale, heterogeneous data scenarios. By randomly extracting image samples and constructing training sets, validation sets, and test sets, introducing a graph convolutional network to automatically extract and learn image features such as contours, colors, shadows, and topology, it is a key step in realizing intelligent recognition and association classification of spatial units, and has high adaptability and scalability. Therefore, as a further limitation of the present invention, natural resource spatial unit image data is randomly extracted, training sets, validation sets, and test sets are established, and a graph convolutional network is used to extract image contours, colors, shadows, and topological relationship features and train the model to automatically classify and determine the association relationship of spatial units.

[0096] This method automatically identifies the visual and structural features of spatial units using a graph convolutional network. It can accurately extract similarities and associations in complex terrain and heterogeneous regions, significantly improving the intelligent identification of associations between spatial units. After model training, it enables efficient batch processing, significantly reducing labor costs and error rates. The resulting classification output provides structured support for building high-quality spatial knowledge graphs, enhancing their expressiveness and generalization capabilities. It is suitable for a variety of natural resource scenarios, including planning, management, and monitoring.

[0097] As a further limitation of the present invention, the data of the nodes and edges of the spatiotemporal knowledge graph are integrated and embedded in the natural resource spatial unit attribute data in the form of QR codes. After scanning the QR code, the spatiotemporal knowledge graph data can be directly called and leave a trace of permission call.

[0098] The method for generating the QR code data is as follows: the spatial scale code, business type code and real-time time series label code of each natural resource spatial unit are binary converted and then connected, a padding code and an error correction code are added to generate a data code, and then a positioning pattern, formatted data and version information are added to draw together to generate the QR code.

[0099] The specific steps include:

[0100] Step 1: Code item conversion and data code splicing

[0101] 1.1. For each spatial unit, extract the following three core coding items:

[0102] Spatial scale coding C s , Business Type Code C b and temporal label encoding C t ; Spatial scale codes include provincial, municipal, county and / or township administrative division codes; business type codes include land use, planning approval and / or ecological restoration codes; time series label codes include statuses such as "congested", "under construction" and / or "approved".

[0103] 1.2. Convert each set of the above codes into a binary string of fixed length;

[0104] The obtained spatial scale binary string B s , Business type binary string B b and the time series label binary string B t :

[0105] B s =bin(C s ), B b =bin(C b ), B t =bin(C t ); where bin() is a binary string encoding conversion function.

[0106] 1.3. Splice into a complete data code string in a fixed order:

[0107] B data =B s ||B b ||B t , where || indicates the concatenation operation of binary strings;

[0108] Step 2: Add padding code B to the data code that is insufficient in length according to the QR code standard requirements. pad , and generate the error-correcting code B through the Reed–Solomon error correction algorithm ecc , improve the fault tolerance of QR code in scanning:

[0109] B pad =PadToFullByte(B data );

[0110] B ecc =RS_Encode(B pad ,k);

[0111] Among them, PadToFullByte() is a padding function that ensures that the length of binary data is a multiple of 8; RS_Encode() is a Reed-Solomon encoding function used to generate error-corrected data; k is the error correction level, k = L, M, Q, H, representing different degrees of fault tolerance respectively).

[0112] Step 3: Adding structure and format coding, specifically adding QR code structural information, including positioning pattern, format information and / or version information.

[0113] Step 4: Integrate the data code, error correction code and structural information to generate a final QR code graphic. The final QR code graphic is embedded in the electronic archive, attribute data or database record of the natural resource spatial unit.

[0114] Step 5. When a user accesses graph node data by scanning a code, the system records information such as user identity, scanning time, and access content to implement call permission management and tracing.

[0115] In a specific embodiment of the present invention, Hangzhou is used as an example area to construct a spatiotemporal map of natural resource spatial units in the blocks within the city. Figure 2 As shown in Figure 1, the system first randomly selects a target block DA from the blocks in the entire city. A 500-meter buffer zone is constructed with the block as the center. The system extracts all the blocks in the buffer zone that are spatially adjacent to it, denoted as D1~D n The data on the number of people staying in each block on a given weekday is extracted from high-precision crowd behavior location data (LBS data). If the same person is found to have stayed in two blocks for more than ten minutes continuously, and the number of people is no less than 200, it can be determined that there is actual spatial behavior correlation between the two blocks.

[0116] Based on this spatial behavior correlation criterion, the system selects the associated blocks DX1~DX n , numbered them, and further extracted their five major attribute information, including volume ratio, greening ratio, average building height, maximum building height, and building density. After linking these data through block numbers, a dataset MA of the target block (DA) and its spatially related blocks was generated and stored in the spatial database under the ArcGIS platform in the .shp file format.

[0117] Subsequently, the system uses the above method to extract at least 50 blocks across the city and generates 50 corresponding spatial data sets M1~M2. 50 Using a workstation equipped with a GTX1070 GPU, a graph convolutional neural network (GCN) model was constructed and trained, with the target block as the center, related blocks as nodes, and five attributes as input features. The model input is the attribute vector of each node, and the output is a prediction of whether the block is spatially associated with the target block. This graph neural network model automatically learns the mapping relationship between spatial attributes and spatial behavior, thereby achieving automatic recognition and classification of spatial structure and building a recognition module.

[0118] In subsequent operations, this module can be applied to the automatic matching and classification identification of new blocks, realize the intelligent extraction of the correlation relationship between spatial units, and further be used to dynamically construct the spatial structure layer of the natural resource spatiotemporal knowledge map.

[0119] After the graph neural network model is trained and the structural graph relationship between spatial units is formed, the system further encodes the graph node information of each target block DA and forms a QR code package to support lightweight terminal calls and permission management. Specifically, the system generates a unique data structure code for each target block node, including the following three information fields:

[0120] Spatial scale coding (e.g., zoning level number, denoted as C s );

[0121] Business type code (such as land use, planning approval, ecological restoration, recorded as C b );

[0122] Real-time status label code (such as "approved", "under construction", "warning", etc.), recorded as C t ). The system converts the above three codes into fixed-length binary strings B s ,B b ,B t , and splice them in order to get the complete data string B data =B s ||B b ||B t Then, the system performs byte padding on the data string to obtain the padding code B pad , and based on the selected error correction level (such as Q level, with 25% fault tolerance), the error correction code B is generated by the Reed-Solomon coding algorithm ecc The system further adds QR code structural elements, including the format information field B format and version information field B version , splicing to get the final QR code data string B final , and draw it as a standard QR Code graphic. Each bit in the QR code is arranged in a module, and the output image is embedded into the block DA attribute data file, map node attribute page or database index.

[0123] Based on the combination of GIS and graph convolutional neural network (GCN), this invention automatically identifies the spatial correlation between multi-scale natural resource spatial units, realizes the automatic identification of spatial correlation and the construction of knowledge graph nodes, breaks through the limitations of traditional manual definition of spatial correlation, and greatly improves the intelligence and accuracy of spatial correlation identification.

[0124] Through AI real-time analysis of crowd spatial behavior patterns, dynamic updating of the spatiotemporal attributes of natural resource spatial units, and intelligent dynamic updating of knowledge graph time series labels, the problem that existing technologies are difficult to reflect changes in spatial business status in real time is solved.

[0125] By combining the regional scale buffer with the temporal convolutional network (TCN) model, spatial business information of different time nodes can be automatically associated and called across scales, breaking through the complexity of traditional spatial data calling and fusion, and realizing the automated, multi-scale, and cross-temporal and spatial association calling of natural resource spatial information.

[0126] Through the in-depth integration of AI and GIS, this application effectively improves the management efficiency of natural resource spatial information and the intelligence level of automatic information retrieval, and significantly promotes the scientificity and efficiency of natural resource planning, management and decision-making.

[0127] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for constructing a spatiotemporal knowledge graph of natural resources by combining AI and GIS, characterized in that: The following steps are involved: Acquire natural resource business spatial data and attribute data, and build a business spatial database; Extracting vector outlines of the spatial data based on a geographic information system platform and spatially matching them with administrative boundaries; automatically identifying and associating multi-scale spatial ranges of the natural resource spatial data using an AI image recognition algorithm; and using the multi-scale spatial data and business attribute data as machine learning training features; Construct and train a spatial association recognition model based on graph convolutional neural networks to automatically identify the spatial association relationships between natural resource spatial units of different scales to form the spatial topological structure of nodes and edges of the spatiotemporal knowledge graph.

2. The method according to claim 1, characterized in that The method further comprises: Obtain real-time crowd location information within natural resource spatial units, extract each unit's volume ratio, greening rate, average building height, maximum building height, and building density as machine learning input features, and establish real-time crowd spatial behavior patterns; Through machine learning models, the spatiotemporal function relationship between real-time crowd location information and spatial unit attributes is analyzed, and the temporal labels of the spatiotemporal knowledge graph nodes are dynamically adjusted to achieve dynamic updating of the knowledge graph.

3. The method according to claim 1, characterized in that The automatic identification of spatial associations between natural resource spatial units of different scales to form a spatial topological structure of nodes and edges of a spatiotemporal knowledge graph includes: Based on regional-scale buffer analysis, natural resource spatial units with spatial correlation relationships are automatically determined, and when the buffer overlap rate exceeds 50%, the regional attributes of the spatial units are used as machine learning training features; Construct a temporal association model based on a temporal convolutional network, combine the spatial association recognition results with the temporal model, realize the automatic association call of business information between natural resource spatial units at different time scales, and form a cross-scale temporal topological relationship of the spatiotemporal knowledge graph.

4. The method according to claim 2, characterized in that The analysis method of the real-time crowd spatial behavior pattern is as follows: when the real-time stay in the target spatial unit is more than 10 minutes and the number of people is not less than 100, it is automatically determined that there is a spatiotemporal correlation between the spatial unit and the adjacent spatial units, forming a real-time spatiotemporal knowledge graph node of the spatial unit.

5. The method according to claim 3, characterized in that The buffer zone analysis uses a 500-meter buffer zone outside the boundary, and automatically determines other spatial units that have regional associations with the target spatial unit through image recognition algorithms and GIS spatial overlay analysis, and generates a spatial association set.

6. The method according to claim 1, characterized in that The construction process of the spatial association recognition model is as follows: randomly extracting natural resource spatial unit image data, establishing a training set, a validation set and a test set, using a graph convolutional network to extract the image's contour, color, shadow, and topological relationship features and train the model, and automatically classifying and determining the spatial unit association relationship.

7. The method according to any one of claims 1 to 6, characterized in that: The data of the nodes and edges of the spatiotemporal knowledge graph are integrated and embedded in the natural resource space unit attribute data in the form of QR codes. After scanning the QR code, the spatiotemporal knowledge graph data can be directly called and leave a trace of permission call.

8. The method according to claim 7, characterized in that The method for generating the QR code data is as follows: the spatial scale code, business type code and real-time time series label code of each natural resource spatial unit are binary converted and then connected, a padding code and an error correction code are added to generate a data code, and then a positioning pattern, formatted data and version information are added to draw together to generate the QR code.

Images