Method and system for organizing seabed data based on spatial entity objects

Abstract

This invention provides a method and system for organizing seabed data based on spatial entity objects. It includes constructing a multi-dimensional data layer structure, comprising a spatial object layer, a storage structure layer, an unstructured data layer, and a verification and extension layer. Data in each layer is composed of nodes. Data nodes are created within the data layers, and a heterogeneous graph neural network with data generation information is used within and between layers to systematically, conveniently, efficiently, and veribly extend the data organization. The data generation information involves numerous social vessels, improving the efficiency of data generation and further data organization.

Description

Technical Field

[0001] This invention relates to methods and systems for organizing seabed data, and particularly to methods and systems for organizing seabed data based on spatial entity objects, belonging to the field of geographic data processing. Background Technology

[0002] Research on nearshore seabed areas requires the integration and analysis of multi-source heterogeneous data, including geological and geomorphological data, geophysical data, hydrological and hydrodynamic data, marine biological and ecological data, legal and political data, and related mineral, environmental, and resource data. The challenge lies in organically organizing this diverse data and addressing technical issues such as ease of retrieval, real-time updates, and scalability. This necessitates a reorganization of data processing and the logical relationships between data sources. Based on this, a unified data organization model structure should be designed to integrate multi-source heterogeneous data (such as satellite remote sensing, seabed exploration, and seismic surveys) through a hierarchical and categorized approach. The fused data will then be applied to the formulation and optimization of nearshore seabed area plans, supporting precise decision-making and international legal compliance assessments, and providing strong technical support for the protection of maritime rights and resource development. Summary of the Invention

[0003] To address the aforementioned issues, this invention is designed from the following key aspects, thereby finding a method for organizing real-time multi-source heterogeneous data: first, the classification, relationships, and storage structure design of data types, as well as verifiability; second, data fusion; and third, data update schemes.

[0004] Based on the above design considerations, the present invention provides a method for organizing seabed data based on spatial entity objects, comprising the following steps:

[0005] S1 constructs a multi-dimensional data layer structure, including a spatial object layer, a storage structure layer, an unstructured data layer, and a verification and extension layer. The data in each layer is composed of nodes.

[0006] S2 establishes heterogeneous graph neural networks H within and between layers of the data layer structure. i (V i E i ,R i ,T i D i ) and interlayer heterogeneous graph neural network H int (V int E int ,R int ,T int D int ), where V i V int E i E int ,R i Rint ,T i T int D i D int Five sets of variables represent the network node set, path set, node relationship set, data type set, and data information set of both entities, respectively. The data information set includes data content and data generation information. Let v be two nodes in the multidimensional data layer structure. m ,v n If m and n are node codes, then the path E between the two nodes is... mn =(v m ,r mn ,v n The path E between the data type set and the data information set. TD =(t m ,r m ,d m )∪(t n ,r n ,d n )∪(X m ,r′ mn ,Y n ),X m =t m ∪d m ,Y n =t n ∪d n ,t n ∈T i ,t m ∈T i This includes the path between the data type sets and data information sets of the two nodes, the path between the data type sets and data information sets of the two nodes, and the path E between the node and the data type sets and data information sets. vTD =(v k ,r vTD ,T i )∪(v k ,r vTD D i ), (E mn ∪E TD ∪E vTD )=(E i ∪E int ), node v k For T i and D i For nodes in or outside the specified layer, k represents the node v. k The encoding includes data generation information such as the method of data generation, the time of data generation, and the update history;

[0007] It should be understood that heterogeneous graph neural networks are a framework for data fusion. Once constructed, it is only necessary to store the corresponding nodes in the data layer and define a path, relationship, type, and data information to complete the fusion. The reversible mapping between nodes can be completed through the path, making retrieval convenient.

[0008] The fusion of S3 data first involves cleaning and standardizing the multidimensional data layer structure through data preprocessing; then, the data from each layer of S1 are filled into the nodes of S2, and the fusion of three-dimensional vector data, grid data, marine hydrological and meteorological data, and real-scene three-dimensional data is completed respectively.

[0009] Optionally, the spatial object layer includes data belonging to the data type set in the layer, and the characteristics and uses of the data content belonging to the data information set for each data type are determined as shown in the table below (the names of the tables that appear later are arranged in order of Chinese characters, such as the second page, the third page, etc.):

[0010]

[0011]

[0012] The storage structure layer is the storage structure for the nodes and their relationships in the construction layer. It includes the following seven tables: metadata storage structure, geometric information storage structure, linked object storage structure, attribute storage structure, material, texture, and image data storage structure, geometric template storage structure, and extended content storage structure.

[0013]

[0014]

[0015] The source object and the target object belong to two nodes in the layer.

[0016]

[0017] By linking the description in the object storage structure The call establishes paths between spatial object nodes. and the path between field names The relationships between data types and descriptions within and between the preceding seven tables are illustrated by these relationships. Establish the path between them It also utilizes the relationships between nodes or field names in the spatial object layer and field names, data types, and descriptions in the preceding seven tables. Establish the path between them

[0018] The unstructured data layer is used to store policies and regulations, videos of biological and ecological resources, and approvals for marine protection rights. It specifically includes unstructured data and the association mechanism between the spatial object layer and the unstructured data layer, as shown in the following tables:

[0019]

[0020]

[0021] According to the description in the association mechanism The call establishes a path between the spatial node and the data types and descriptions in the unstructured data layer.

[0022] The validation and extension layer is used to validate the insertion of new data into other layers, establish new relationships between new data and other existing data, validate queries for inserting new data, and verify whether the process can be completed accurately. It includes validation nodes and extension nodes. Validation nodes establish path links with nodes in other layers to realize data insertion, querying, and updating. After validation is completed, any data in the validation node is deleted. Extension nodes include the addition of data nodes in other layers and the updating of corresponding data information in the nodes. Similarly, the addition and updating are realized through heterogeneous graph neural network path links.

[0023] It should be understood that the nodes, path sets, and node relationship sets of this invention are mathematical abstractions of data, while the data type sets and data information sets are the specific data corresponding to the abstracted nodes. This specific data is also abstracted as nodes in the neural network. That is, although the data types are indicated in the table above... describe describe Relationship type However, in heterogeneous graph neural networks, data types, descriptions, and relation types are all treated as network nodes. In other words, the data types in the table... describe describe Relationship type Equations represent the specific data to which a node belongs. When the path between nodes belongs to a path in a heterogeneous graph neural network, then the nodes at both ends of the path belong to V. int The corresponding path set, node relationship set, data type set, and data information set are all denoted as E. int R int T int D int This is to distinguish the case in layers where i is the subscript.

[0024] E mn =(v m ,rmn ,v n The nodes in the diagram can be nodes abstracted from the data characteristics, data purpose, field name, data type, and description of the nine tables mentioned above in the spatial object layer, storage structure layer, and unstructured data layer. In this case, the path between nodes is represented as E. mn Alternatively, it can be specific data information, including the data types and descriptions in the nine tables mentioned above. In this case, the path nodes between data information and the paths between data information are respectively represented as E. TD and E vTD Therefore, there is That is, when E mn When at least one node in a relation is represented by specific data information, the corresponding relation subset r m ,r n ,r′ mn ,r vTD All four are r mn A subset of. The time-related data in the above nine tables belongs to the data generation information, while the addition and updates of other data types, as well as the generation and update time of the "description" itself in the tables, all belong to the data generation time and update history in the data generation information.

[0025] The data generation method includes the following steps:

[0026] For data in the spatial object layer, the following steps are involved:

[0027] Q1. Install detectors and satellite positioning systems on social vessels, or on social vessels, survey vessels, and / or submarines.

[0028] Q2 records the detection signals periodically while the ship is sailing.

[0029] Q3 combines the detection signals and satellite positioning at the time of detection to create images of the shallow seabed;

[0030] Q4. Acquire mid- and deep-sea exploration data, create mid- and deep-sea seabed images, and stitch them together with shallow seabed images to form a stitched image. Mark the data in the spatial object layer to form three-dimensional real-scene data.

[0031] For the storage structure layer and unstructured data layer, the relevant data is obtained through computer systems installed on social vessels, or on social vessels, exploration vessels, or submarines, and then stored in the corresponding storage structure layer and unstructured data layer of the remote server.

[0032] Understandably, the participation of numerous social vessels and frequent exploration activities has increased the efficiency of shallow sea exploration data generation.

[0033] Optionally, the method for fusing 3D vector data, grid data, marine hydrological and meteorological data, and real-scene 3D data includes the following steps:

[0034] Methods for fusing 3D vector data include:

[0035] S3-1-1 stores the acquired 3D vector data containing 3D spatial coordinates (i.e., the composite data formed by mapping and associating the three types of data—data type, data adjustment, and data usage—with the 3D spatial coordinates) into the spatial object layer, and fills the data into the nodes of the corresponding metadata storage structure, geometric information storage structure, linked object storage structure, attribute storage structure, material, texture, and image data storage structure, geometric template storage structure, extended content storage structure, and unstructured data layer.

[0036] It is easy to understand that 3D vector data is actually data from a layer of spatial objects with 3D spatial coordinates that are mapped and annotated on the stitched image.

[0037] S3-1-2 uses the heterogeneous graph neural network model built through S2 to complete the path links between nodes that have been filled with data, thereby achieving data fusion; the methods for fusion of grid data include:

[0038] S3-2-1 Divide the stitched image into a fixed-resolution grid, and encode the grid units using a preset encoding rule;

[0039] For each encoded node stored in the spatial object layer, S3-2-2 uses the heterogeneous graph neural network model constructed in S2 to complete the path links between the nodes in the spatial object layer, the nodes in each storage structure and the nodes in the unstructured data layer that are filled with data, so as to achieve data fusion.

[0040] Methods for fusing marine hydrological and meteorological data include:

[0041] S3-3-1 Acquire and analyze marine environmental information field data, and group information field data with the same or similar characteristics into one field node;

[0042] S3-3-2 Obtain the distribution of field nodes in the spatial object layer, storage structure layer, and unstructured data layer;

[0043] For example, the distribution of field nodes on spatial object features (such as shallow sea areas, deep sea areas, trenches, etc.) in different grid nodes in the spatial object layer; the distribution of field nodes on geometric features on different internal surfaces and boundary lines in the storage structure layer; the distribution of field nodes on features at different times in the unstructured data layer; and the distribution of field nodes on mixed features.

[0044] For all the aforementioned distributions, each field node in all layers is treated as a node as described in S2. Based on the heterogeneous graph neural network model constructed in S2, the path links between information field data in nodes in the spatial object layer, each storage structure layer, and the unstructured data layer are also completed to achieve data fusion.

[0045] Methods for fusing real-scene 3D data include:

[0046] S3-4-1 Acquiring Large-Scale 3D Model Data: 3DTiles Specification;

[0047] S3-4-2 integrates 3D reality data under the 3DTiles specification.

[0048] Optionally, the preset encoding rules include Morton traversal order rules, and the information field includes various fields composed of temperature, density, salinity, marine luminescence, ocean current vector velocity, transparency, water color, and sea surface wind; the three-dimensional real-scene data includes at least one of osgb, gltf, and b3dm.

[0049] Another objective of this invention is to provide a seabed data organization system based on spatial entity objects, including social vessels, or social vessels, exploration vessels, and / or submarines, and a remote server. The social vessels and exploration vessels are equipped with detectors for creating three-dimensional real-scene data, and corresponding computer systems are set up to acquire relevant data and store it in the corresponding storage structure layer and unstructured data layer of the remote server. The remote server receives signals from the detectors and exploration devices, constructs a spatial object layer, processes the signals, completes the creation of three-dimensional real-scene data, and implements the seabed data organization method of spatial entity objects through the installed software system.

[0050] Beneficial effects

[0051] By structuring the data architecture within a multi-dimensional data layer structure, and considering the specific data structures of each layer, a data node-based approach is adopted. Based on these data nodes, a heterogeneous graph neural network carrying data generation information is used within and between layers to systematically, conveniently, veribly, and scalably organize the data. Furthermore, the participation of numerous social vessels in seabed data exploration improves the efficiency of data generation and further data organization. Attached Figure Description

[0052] Figure 1 This is the overall data organization diagram constructed by the seabed data organization method based on spatial entity objects in Embodiment 1 of the present invention;

[0053] Figure 2 for Figure 1 A detailed overview of the data organization structure as constructed;

[0054] Figure 3 A schematic diagram of the program interface for verifying data insertion;

[0055] Figure 4 A schematic diagram of the program interface for verifying data query;

[0056] Figure 5 A schematic diagram of the program interface for verifying data updates;

[0057] Figure 6a This is a schematic diagram of the ships used in the data generation method of the spatial object layer in Embodiment 1 of the present invention, which navigate and explore in different near-shore seabed areas.

[0058] Figure 6b Flowchart for creating 3D reality data;

[0059] Figure 6c This is a flowchart of the grid data partitioning and fusion method in Embodiment 2 of the present invention;

[0060] Figure 6d This is a flowchart of the marine hydrological and meteorological data fusion method in Embodiment 2 of the present invention;

[0061] Figure 7 This is a flowchart of the three-dimensional vector data fusion method in Embodiment 2 of the present invention. Detailed Implementation

[0062] Example 1

[0063] This embodiment will describe the method for organizing seabed data based on spatial entity objects, such as... Figure 1 As shown, it includes the following steps:

[0064] S1 constructs a multi-dimensional data layer structure, including a spatial object layer, a storage structure layer, an unstructured data layer, and a verification and extension layer. The data in each layer is composed of nodes.

[0065] S2 constructs a heterogeneous graph neural network within and between layers, forming reversible paths between nodes in each layer. The heterogeneous graph neural network within and between layers is represented by H. i (V i E i ,R i ,T i D i ) and H int (V int E int ,R int ,T int D int ).

[0066] First, the relationships and paths between nodes are explained, revolving around node sets, relationship sets, data type sets, and data-generated information sets, and are divided into three main categories, such as... Figure 2 As shown.

[0067] Since the paths between validation nodes and extension layer nodes in the validation and extension layers are described in the same way as those between other layers in a heterogeneous graph neural network, for ease of description, Figure 2 The text only lists the spatial object layer, storage structure layer, and unstructured layer to illustrate the construction of heterogeneous graph neural networks between each layer.

[0068] The spatial object layer includes three types of data as described in the first table: data type, data characteristics, and data purpose. The layer dataset to which it belongs can be found in the first table. The storage structure layer has seven types of data, corresponding to the seven tables mentioned above. The layer dataset to which it belongs can be found in the second to seventh tables, respectively. Each type of table contains three types of nodes: field name, data type, and description. The heterogeneous graph neural network built between nodes within a layer is called the intra-layer heterogeneous graph neural network, while the one built between layers is called the inter-layer heterogeneous graph neural network.

[0069] For the three types of data in the spatial object layer, the path representation between any two nodes is E. mn The proper subset of the path, the relation set of the nodes at both ends of the path is r. mn The corresponding heterogeneous graph neural network belongs to the layer. Taking the metadata storage structure in the storage structure layer as an example, the heterogeneous graph neural network built between the field name node and the node of the spatial object layer belongs to the inter-layer, while the relationship between the field name and the field name node of the material, texture, and image data storage structure belongs to the layer.

[0070] For data types and data information as nodes, the network node path representation they construct is E. TD The proper subset of the path, the relation set of the nodes at both ends of the path is r. m ,r n ,r′ mn , is a relation set r mn A proper subset of. The "description" type data in the linked object storage structure describes the set of relationships between nodes in the spatial object layer; it is also a set of relationships r. mn A proper subset of .

[0071] Taking the relationships between field names, data types, and descriptions of unstructured data in the unstructured data layer; the relationships between data usage and descriptions of unstructured data in the spatial object layer; and the relationships between data types and data types in the association mechanism within the spatial object layer as examples, the constructed heterogeneous graph neural network belongs to the inter-layer category, and the path representation between any two nodes is E. vTD The proper subset of the path, the relation set of the nodes at both ends of the path is r.vTD It is also a relation set r mn A proper subset of .

[0072] Therefore, in heterogeneous graph neural networks, nodes involving both data types and descriptions from non-spatial object layers are represented by their path relation sets with subscripts ending in TD. Furthermore, if a node with a field name is also a node on the path, its subscript is uniformly represented with vTD (note that data types and descriptions do not need to be fully included; since this invention defines a path with only two nodes, it is still represented by a path set with vTD. For multi-node paths, it is considered as a concatenation of two or more paths). The relation set subscript for the corresponding path node is also vTD. If a node does not belong to either the data type or description category from the non-spatial object layer, then the corresponding path set and relation set subscript representation is mn, resulting in three main categories.

[0073] The heterogeneous graph neural network constructed in this way facilitates subsequent data fusion, data querying, updating, and verification. It can also be used for other research work based on the data access history.

[0074] Secondly, the paper introduces data verification, testing the integrity and performance of the storage structure using real-world data (such as seabed mining areas and mining boundaries). Figure 3 A program was developed to insert metadata, geometric information, and attribute data for a seabed mining area. For example... Figure 4 As shown, the metadata, geometric information, and attribute data hierarchy of mining area A were queried. Figure 5 A program to update the resource reserve attributes of mining area A.

[0075] To verify the effectiveness of the unstructured data linking mechanism, it is first necessary to ensure that unstructured data (such as policies and regulations, videos of biological and ecological resources, and approvals for marine protection rights) can be accurately associated with corresponding spatial objects (such as seabed mining areas and mining boundaries). By inserting test data and establishing relationships, it is necessary to check whether the association table (such as unstructured_data_links) can correctly record the relationship between unstructured data and spatial objects, and to verify whether query operations can efficiently retrieve unstructured data associated with specific spatial objects.

[0076] Testing the compatibility of the storage structure with new data types (such as newly added marine ecological data) primarily verifies whether the database can flexibly support data types and formats that may be added in the future. The scalability of the storage structure is evaluated by inserting new types of data (such as marine ecological monitoring data) into extension content tables (e.g., `extension_content`) and checking whether its storage, query, and update operations function correctly. For example, a section of marine ecological monitoring records, including data on species distribution and ecological environment changes, can be inserted, and it can be verified whether this data can be correctly stored and associated with relevant spatial objects. Furthermore, it is necessary to test the compatibility of new data types with existing data to ensure that the addition of data does not affect the integrity and performance of existing data.

[0077] Finally, the information on data generation in heterogeneous graph neural networks includes the data generation method, the time of data generation, and the update history, reflecting the real-time nature of data organization. The data generation method includes the following steps:

[0078] Combination Figures 6a to 6d As shown, the data in the spatial object layer includes the following steps:

[0079] Q1 Figure 6a The study defines three nearshore seabed zones: shallow, mid-water, and deep. Detectors and satellite positioning systems (not shown in the diagram) are installed on civilian vessels, research vessels, and two groups of submarines. The civilian vessels navigate in the shallow waters, the research vessels navigate in the mid-waters, one group of submarines navigates underwater in the mid-waters, and the other group of submarines navigates underwater in the deep waters.

[0080] The following is as follows Figure 6b As shown in Figure Q2, the detection signals are recorded periodically while various types of ships are navigating.

[0081] Q3 combines the detection signals and satellite positioning at the time of detection to create images of the shallow seabed;

[0082] Q4. Acquire mid- and deep-sea exploration data, create mid- and deep-sea seabed images, and stitch them together with shallow seabed images to form a stitched image. Mark the data in the spatial object layer to form three-dimensional real-scene data.

[0083] For the storage structure layer and the unstructured data layer, the corresponding data is acquired by social vessels, or by installing corresponding detection devices on social vessels, exploration vessels or submarines, and then stored in the corresponding storage structure layer and unstructured data layer.

[0084] The fusion of S3 data first involves cleaning and standardizing the multidimensional data layer structure through data preprocessing; then, the data from each layer of S1 are filled into the nodes of S2, and the fusion of three-dimensional vector data, grid data, marine hydrological and meteorological data, and real-scene three-dimensional data is completed respectively.

[0085] Example 2

[0086] This embodiment will specifically explain the data fusion in Embodiment 1.

[0087] Methods for fusing 3D vector data, such as Figure 7 As shown, it specifically includes:

[0088] S3-1-1 stores the acquired 3D vector data containing 3D spatial coordinates into the spatial object layer, and fills the data into the nodes in the corresponding metadata storage structure, geometric information storage structure, linked object storage structure, attribute storage structure, material, texture, and image data storage structure, geometric template storage structure, extended content storage structure (seven types of storage structures), and unstructured data layer.

[0089] S3-1-2 uses the heterogeneous graph neural network model (including in-layer and inter-layer models) built by S2 to complete the path links between nodes that have been filled with data, so as to achieve data fusion.

[0090] Methods for fusing grid data include, for example Figure 6c As shown:

[0091] S3-2-1 Divide the stitched image into a grid composed of rectangular grid units with a fixed resolution. The grid units are encoded using the Morton traversal order rule.

[0092] Fixed-resolution gridding is a standard method for generating point datasets. It involves setting a resolution or grid spacing and then calculating the value of each grid based on the single attribute of the point group contained in each cell.

[0093] Grid cell size is variable; areas with significant variations in marine environmental elements are represented by small grid cells, while areas with less variation are represented by large grid cells. This ensures that the grid mosaicking partitioning still covers bounded areas, and the traversal method must be able to organize the cells in an ordered manner. Each cell carries several attribute values. Neighboring cells can be aggregated to greatly reduce the amount of data. Small cells only exist in areas where attributes change rapidly between cells; shoals, coastlines, and obstacles will result in some small cells, while relatively stable or flat areas, such as the bottom of a strait, will result in some aggregated cells.

[0094] Merton sequential encoding traverses from left to right and from bottom to top, cell by cell, regardless of cell size. It first adds the X coordinate, then the Y coordinate. This can also be extended to multidimensional cases, adding the X coordinate first, then the Y coordinate, then the Z coordinate, and so on to more dimensions.

[0095] For each encoded node stored in the spatial object layer, S3-2-2 uses the heterogeneous graph neural network model constructed in S2 to complete the path links between the nodes in the spatial object layer, the nodes in each storage structure and the nodes in the unstructured data layer that are filled with data, so as to achieve data fusion.

[0096] Methods for fusing marine hydrological and meteorological data include, for example Figure 6d As shown:

[0097] S3-3-1 Acquires and analyzes marine environmental information field data, including various fields such as temperature, density, salinity composition, sea luminescence, ocean current vector velocity, transparency, water color, and sea surface wind, and groups information field data with the same characteristics into one field node;

[0098] S3-3-2 Obtain the distribution of field nodes in the spatial object layer, storage structure layer, and unstructured data layer;

[0099] For all the aforementioned distributions, each field node in all layers is treated as a node as described in S2. Based on the heterogeneous graph neural network model constructed in S2, the path links between information field data in nodes in the spatial object layer, each storage structure layer, and the unstructured data layer are also completed to achieve data fusion.

[0100] Methods for fusing real-scene 3D data include:

[0101] S3-4-1 Acquiring Large-Scale 3D Model Data: 3DTiles Specification;

[0102] S3-4-2 fuses 3D reality data OSGB under the 3DTiles specification.

[0103] Example 3

[0104] This embodiment provides a seabed data organization system based on spatial entity objects, including, for example... Figure 6a The diagram shows a social vessel, a probe vessel, two sets of submersibles, and a remote server (not shown). The social vessel, probe vessel, and two sets of submersibles are equipped with detectors for creating 3D reality data and corresponding computer systems to acquire relevant data and store it in the corresponding storage structure layer and unstructured data layer of the remote server. The remote server receives signals from the detectors and probe devices, constructs a spatial object layer, processes the signals, completes the creation of 3D reality data, and implements a method for organizing seabed data of spatial entities through the installed software system.

Claims

1. A method for organizing seabed data based on spatial entity objects, characterized in that, Includes the following steps: S1 constructs a multi-dimensional data layer structure, including a spatial object layer, a storage structure layer, an unstructured data layer, and a verification and extension layer. The data in each layer is composed of nodes. S2 establishes heterogeneous graph neural networks within and between layers of the data layer structure. and interlayer heterogeneous graph neural networks ,in Five sets of variables represent the network node set, path set, node relationship set, data type set, and data information set of both entities, respectively. The data information set includes data content and data generation information. Consider two nodes in the multidimensional data layer structure. and Encoding nodes, then the path between two nodes Path between data type sets and data information sets This includes the paths between the data type sets and data information sets of the two nodes, as well as the paths between the data type sets and data information sets of the two nodes, and the paths between a node and its data type set and data information set. , , , ,node To and Nodes in or outside the specified layer, For nodes The encoding includes data generation information such as the method of data generation, the time of data generation, and the update history; The data generation method includes the following steps: For spatial object layer data: Q1 Install detectors and satellite positioning systems on social vessels, or on social vessels, exploration vessels and / or submarines; Q2 records the detection signals periodically while the ship is sailing; Q3 combines the detection signals and satellite positioning at the time of detection to create shallow seabed images; Q4. Acquire mid- and deep-sea exploration data, create mid- and deep-sea seabed images, and stitch them together with shallow seabed images to form a stitched image. Mark the data in the spatial object layer to form three-dimensional real-scene data. For the storage structure layer and unstructured data layer, the corresponding data is obtained through social vessels, or computer systems set up on social vessels, exploration vessels or submarines, for storage in the corresponding storage structure layer and unstructured data layer of remote servers. The fusion of S3 data first involves cleaning and standardizing the multidimensional data layer structure through data preprocessing; then, the data from each layer of S1 are filled into the nodes of S2, and the fusion of three-dimensional vector data, grid data, marine hydrological and meteorological data, and real-scene three-dimensional data is completed respectively. The method for fusing 3D vector data, grid data, marine hydrological and meteorological data, and real-scene 3D data includes the following steps: Methods for fusing 3D vector data include: S3-1-1 stores the acquired 3D vector data containing 3D spatial coordinates into the spatial object layer, and fills the data into the nodes of the corresponding metadata storage structure, geometric information storage structure, linked object storage structure, attribute storage structure, material, texture and image data storage structure, geometric template storage structure, extended content storage structure, and unstructured data layer. S3-1-2 uses the heterogeneous graph neural network model built through S2 to complete the path links between the nodes that are filled with data, so as to achieve data fusion; Methods for fusing grid data include: S3-2-1 Divide the stitched image into a fixed-resolution grid, and encode the grid units using a preset encoding rule; For each encoded node stored in the spatial object layer, S3-2-2 uses the heterogeneous graph neural network model constructed in S2 to complete the path links between the nodes in the spatial object layer, the nodes in each storage structure and the nodes in the unstructured data layer that are filled with data, so as to achieve data fusion. Methods for fusing marine hydrological and meteorological data include: S3-3-1 Acquire and analyze marine environmental information field data, and group information field data with the same or similar characteristics into one field node; S3-3-2 Obtain the distribution of field nodes in the spatial object layer, storage structure layer, and unstructured data layer; For all the aforementioned distributions, each field node in all layers is treated as a node as described in S2. Based on the heterogeneous graph neural network model constructed in S2, the path links between information field data in nodes in the spatial object layer, each storage structure layer, and the unstructured data layer are also completed to achieve data fusion. Methods for fusing real-scene 3D data include: S3-4-1 Acquiring Large-Scale 3D Model Data: 3DTiles Specification; S3-4-2 integrates 3D reality data under the 3DTiles specification.

2. The method according to claim 1, characterized in that, The spatial object layer includes data belonging to the data type set within the layer, and the characteristics and uses of the data content belonging to the data information set for each data type are defined in the following table: ， The storage structure layer is the storage structure for the nodes and their relationships in the construction layer. It includes the metadata storage structure, geometric information storage structure, linked object storage structure, attribute storage structure, material, texture, and image data storage structure, geometric template storage structure, and extended content storage structure, which are represented by the following seven tables: ， ， ， The source object and the target object belong to two nodes in the layer. ， ， ， ， By linking the description in the object storage structure The call establishes paths between spatial object nodes. and the path between field names Based on the relationships between data types and descriptions within and between the previous seven tables. Establish the path between them It also utilizes the relationships between nodes or field names in the spatial object layer and field names, data types, and descriptions in the preceding seven tables. Establish the path between them ; The unstructured data layer is used to store policies and regulations, videos of biological and ecological resources, and approvals for marine protection rights. It specifically includes unstructured data and the association mechanism between the spatial object layer and the unstructured data layer, as shown in the following tables: ， ， According to the description in the association mechanism The call establishes a path between the spatial node and the data types and descriptions in the unstructured data layer. ; The validation and extension layer is used to validate the insertion of new data into other layers, establish new relationships between new data and other existing data, validate queries for inserting new data, and verify whether the process can be completed accurately. It includes validation nodes and extension nodes. Validation nodes establish path links with nodes in other layers to realize data insertion, querying, and updating. After validation is completed, any data in the validation node is deleted. Extension nodes include the addition of data nodes in other layers and the updating of corresponding data information in the nodes. Similarly, the addition and updating are realized through heterogeneous graph neural network path links.

3. The method according to claim 1, characterized in that, The preset encoding rules include the Morton traversal order rules, and the information fields include various fields composed of temperature, density, salinity, marine luminescence, ocean current vector velocity, transparency, water color, and sea surface wind; the three-dimensional real-scene data includes at least one of osgb, gltf, and b3dm.

4. The method according to claim 1, characterized in that, Fixed-resolution gridding is a standard method for generating point datasets. It involves setting a resolution or grid spacing and then calculating the value of each grid cell based on the single attribute of the point group contained in each cell. The size of the grid cells is variable. Areas with large variations in marine environmental elements are represented by small grid cells, while areas with smaller variations are represented by large grid cells.

5. A seabed data organization system based on spatial entity objects, characterized in that, It includes social vessels and a remote server, or includes social vessels, exploration vessels and / or submarines, and a remote server; the social vessels and exploration vessels are equipped with detectors for creating three-dimensional real-scene data, and are equipped with corresponding computer systems for acquiring relevant data and storing it in the corresponding storage structure layer and unstructured data layer of the remote server; the remote server receives signals from the detectors and exploration devices, constructs a spatial object layer, processes the signals, completes the creation of three-dimensional real-scene data, and implements the seabed data organization method for spatial entity objects as described in any one of claims 1-4 through the installed software system.

Images