Land monitoring sample intelligent management system and method

Abstract

The application discloses a kind of land monitoring sample intelligent management system and method, and the multidimensional land classification system formed by classification system construction module comprehensively covers the natural, economic, social attribute of land, satisfies diversified demand;At the same time, the complex relationship of land entity is graphically displayed by knowledge graph construction module, the inherent relationship of attribute is excavated, and the understanding of land characteristics is deepened;Secondly, a uniform standard is provided for multi-source data by data specification definition module, and the heterogeneous data is automatically processed by multi-source data fusion module according to the uniform standard, so that the difference is eliminated to form a high-quality data set;In addition, the knowledge graph elements are converted into vectors by vector library construction module, which is convenient for computer processing;With the help of vector library, the intelligent query module receives natural language queries, quickly retrieves and deeply excavates potential information, and realizes efficient analysis and utilization of land monitoring data.

Description

Technical Field

[0001] This invention relates to the field of land monitoring sample management technology, and in particular to an intelligent management system and method for land monitoring samples. Background Technology

[0002] With the acceleration of urbanization and the increasing demand for refined land resource management, land monitoring is playing an increasingly crucial role in areas such as national land planning, resource protection, and disaster early warning. As the foundation of land monitoring work, the efficient management and intelligent analysis of land monitoring sample data are essential for accurately understanding land conditions.

[0003] Traditional land monitoring sample management models have several problems: On the one hand, land classification systems are often relatively simple and fail to fully and meticulously reflect the complex attributes and diverse characteristics of land. Different regions and projects may use different classification standards, resulting in poor data consistency and comparability, which greatly hinders data integration and sharing. On the other hand, land monitoring sample data comes from a wide range of sources and is highly heterogeneous. The data may come from various means such as satellite remote sensing, drone aerial photography, and ground surveys. These data have significant differences in format, accuracy, and semantics. Traditional manual processing methods are not only inefficient but also prone to errors, making it difficult to achieve effective fusion and in-depth mining of multi-source data. Summary of the Invention

[0004] In view of this, the present invention proposes an intelligent management system and method for land monitoring samples, which can effectively solve the shortcomings of existing technologies that make it difficult to comprehensively and meticulously reflect the complex attributes and diverse characteristics of land, and make it difficult to achieve effective integration and in-depth mining of multi-source data.

[0005] The technical solution of this invention is implemented as follows:

[0006] A smart management system for land monitoring samples, comprising:

[0007] The classification system construction module is used to construct a multidimensional land classification system;

[0008] The data specification definition module is used to define the metadata specifications for land monitoring sample data.

[0009] The multi-source data fusion module is used to acquire multi-source heterogeneous data and automatically process the multi-source heterogeneous data based on the multi-dimensional land classification system and metadata specifications to form a standardized dataset.

[0010] The knowledge graph construction module is used to extract key entities and relationships from standardized datasets to build a land monitoring knowledge graph;

[0011] The vector library construction module is used to convert entities, relations, and attribute values ​​in the knowledge graph into machine-computable vectors to build a vector library for land monitoring sample data.

[0012] The intelligent query module receives natural language query statements, retrieves the candidate entities / relationships with the highest similarity from the vector library, and returns the query results.

[0013] As a further optional solution to the aforementioned intelligent management system for land monitoring samples, the data specification definition module includes:

[0014] The image data metadata definition unit is used to define the metadata fields of image data, including spatial location, acquisition time, and sensor type.

[0015] The structured data metadata definition unit is used to define the metadata fields of structured data, including field name, type, length, and data format.

[0016] As a further optional solution to the aforementioned intelligent management system for land monitoring samples, the knowledge graph construction module includes:

[0017] The entity extraction unit is used to extract key entities from a standardized dataset;

[0018] The relationship identification unit is used to identify and establish the relationships between entities;

[0019] The knowledge disambiguation unit is used to disambiguate ambiguous entities.

[0020] The knowledge structure optimization unit is used to supplement and improve the knowledge structure of the knowledge graph.

[0021] As a further optional solution to the aforementioned intelligent management system for land monitoring samples, the knowledge disambiguation unit includes:

[0022] The ambiguity identification subunit is used to analyze the contextual information of entities, data source relationships, and business logic to identify entities that have ambiguity or conflicting expressions.

[0023] The disambiguation strategy implementation subunit is used to perform semantic parsing and normalization of ambiguous entities by combining the knowledge base in the land monitoring field and preset rules.

[0024] The verification subunit is used to verify whether the disambiguated entities conform to the logical consistency of the land monitoring scenario based on the relationships between entities in the knowledge graph and business constraints.

[0025] As a further optional solution to the aforementioned intelligent management system for land monitoring samples, the vector library construction module includes:

[0026] The model selection unit is used to select a knowledge graph embedding model for knowledge graph embedding.

[0027] The vector transformation unit is used to convert entities, relations, and attribute values ​​in the knowledge graph into machine-computable vectors.

[0028] The normalization unit is used to perform unified normalization on the transformed vectors and build a vector library.

[0029] As a further optional solution to the aforementioned intelligent management system for land monitoring samples, the vector transformation unit includes:

[0030] The rotation embedding encoding subunit is used to perform rotation embedding encoding on entities using the RotatE model, generating low-dimensional vectors that preserve relational orientation.

[0031] The projection matrix transformation subunit is used to perform projection matrix transformation on the relation using the TransR model, mapping the head entity and tail entity to the semantic space of the relation;

[0032] The normalized scaling subunit is used to extract contextual semantic vectors from textual data in attribute values ​​using the BERT model, and to convert numerical data into fixed-dimensional vectors using normalized scaling.

[0033] The fusion subunit is used to merge vectors of entities, relations, and attribute values ​​through weighted concatenation to form a comprehensive vector representation.

[0034] As a further optional solution to the aforementioned intelligent management system for land monitoring samples, the intelligent query module includes:

[0035] The query preprocessing unit is used to preprocess natural language query statements to obtain core semantics;

[0036] The query vector transformation unit is used to transform core semantics into query vectors using a natural language processing model.

[0037] The similarity calculation unit is used to calculate the similarity between the query vector and the vectors in the vector library using the cosine similarity algorithm.

[0038] The results return unit is used to return the most relevant query results based on the similarity calculation.

[0039] A method for intelligent management of land monitoring samples, specifically including:

[0040] Construct a multi-dimensional land classification system;

[0041] Define the metadata specifications for land monitoring sample data;

[0042] Acquire multi-source heterogeneous data and automatically process the multi-source heterogeneous data based on a multi-dimensional land classification system and metadata specifications to form a standardized dataset;

[0043] Key entities and relationships are extracted from standardized datasets to construct a land monitoring knowledge graph;

[0044] Transform entities, relationships, and attribute values ​​in knowledge graphs into machine-computable vectors to build a vector library of land monitoring sample data;

[0045] It receives a natural language query and retrieves the candidate entities / relationships with the highest similarity from the vector library, then returns the query results.

[0046] A computing device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the intelligent management method for land monitoring samples.

[0047] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the intelligent management method for land monitoring samples.

[0048] The beneficial effects of this invention are as follows: The multi-dimensional land classification system constructed by the classification system construction module classifies land from multiple dimensions. Compared to traditional single classification systems, it can more comprehensively cover various attributes and characteristics of land. Whether it's the natural attributes of land (such as soil type and topography), economic attributes (such as land use value and output benefits), or social attributes (such as land ownership and policy planning impact), all can be reflected in this multi-dimensional system. This makes the description of land more detailed and accurate, meeting the diverse needs for land information in different application scenarios. Simultaneously, the knowledge graph construction module extracts key entities and relationships from standardized datasets to construct a land monitoring knowledge graph. The knowledge graph graphically and intuitively displays the complex relationships between land-related entities, such as the relationship between land and its surrounding environment, and the time series relationship of land use changes. Through this structured knowledge representation, it is possible to deeply explore the intrinsic connections between land attributes, further enriching the understanding of the complex characteristics of land. Secondly, the data specification definition module defines… The metadata specification for land monitoring sample data defines a unified standard framework for multi-source data. Based on this specification, the multi-source data fusion module can automatically process heterogeneous data from multiple sources such as satellite remote sensing, UAV aerial photography, and ground surveys. Through automated processing, it standardizes and transforms data from different sources and formats, eliminating format and semantic differences between data, and achieving effective fusion of multi-source data to form a high-quality standardized dataset. In addition, the vector library construction module transforms entities, relationships, and attribute values ​​in the knowledge graph into machine-computable vectors to build a vector library for land monitoring sample data. This vector representation enables computers to better understand and process land monitoring data. The intelligent query module utilizes the vector library to receive natural language queries and quickly retrieve the candidate entities / relationships with the highest similarity by calculating vector similarity. This not only improves query efficiency but, more importantly, enables in-depth mining of potential information related to query needs from the fused multi-source data, achieving in-depth analysis and utilization of land monitoring data. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a schematic diagram illustrating the composition of an intelligent management system for land monitoring samples according to the present invention;

[0051] Figure 2This is a flowchart illustrating an intelligent management method for land monitoring samples according to the present invention.

[0052] Figure 3 This is a schematic diagram of the composition of a computing device according to the present invention;

[0053] Figure 4 This is a schematic diagram of the processing flow of the intelligent query module of the present invention;

[0054] Figure 5 This is a schematic diagram of the processing flow of the query preprocessing unit of the present invention. Detailed Implementation

[0055] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. 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.

[0056] refer to Figures 1 to 5 A land monitoring sample intelligent management system includes a classification system construction module, a data specification definition module, a multi-source data fusion module, a knowledge graph construction module, a vector library construction module, and an intelligent query module, wherein:

[0057] The classification system construction module is used to build a multidimensional land classification system. Based on the "Land Use Status Classification" (GB / T21010-2017) standard and the classification of the Third National Land Survey, and taking into account the characteristics of village and town land, the construction land part is simplified to form a classification system with 13 primary categories and 53 secondary categories, including cultivated land, forest land, and transportation land. The classification system emphasizes the identifiable features of images to ensure the universality of samples. For example, the collected samples need to cover the appearance of paddy fields at different periods to avoid the problem of "different spectrums for the same object". The system also establishes a mapping relationship between the classification system and other classification systems.

[0058] Specifically, the construction land portion is simplified based on the characteristics of village and town land, and the identifiable features of the images are emphasized, making the classification system more in line with the actual land use situation. By collecting sample images of paddy fields at different times, the problem of "different spectra for the same object" is effectively avoided, and the accuracy of land classification based on images is improved. This provides a more accurate classification basis for subsequent land monitoring work and helps to accurately grasp the distribution and utilization of land resources.

[0059] Establishing a mapping relationship between classification systems and other classification systems breaks down the barriers between different classification systems. In land monitoring projects, it may be necessary to compare and analyze data with data from other regions or different periods, or to share data with other relevant departments. The existence of this mapping relationship enables data to be converted and matched between different systems, improves the universality and operability of data, and provides convenience for cross-regional and cross-time land research and management.

[0060] A data specification definition module is used to define the metadata specifications for land monitoring sample data; in some embodiments, the data specification definition module includes:

[0061] The image data metadata definition unit is used to define the metadata fields of image data, including spatial location, acquisition time, and sensor type.

[0062] The structured data metadata definition unit is used to define the metadata fields of structured data, including field name, type, length, and data format.

[0063] Specifically, the image data metadata definition unit defines detailed metadata fields for various remote sensing image data used in the project. For example, for satellite remote sensing imagery, its spatial location information must be accurate to the specific geographic coordinate range, the acquisition time must be accurate to the year, month, day, hour, and minute, and the sensor type must be recorded in detail, such as the Landsat 8 OLI sensor. For UAV aerial imagery, the spatial location is defined as the coordinates of the area covered by the flight path, the acquisition time is defined as the specific time of takeoff and landing, and the sensor type is defined as the specific camera model carried by the UAV.

[0064] The structured data metadata definition unit standardizes the structured data involved in the project. For example, for soil sample data obtained from ground surveys, the defined field names include sample number, sampling location, soil type, organic matter content, etc.; the type is clearly specified as character type, numeric type, etc.; the length is set according to the actual situation of the data, such as the sample number being 10 characters; and the data format is specified as a specific text format or numerical precision requirement.

[0065] Thus, by defining the metadata of data from different sources using image data metadata definition units and structured data metadata definition units, a unified standard is achieved for multi-source data such as satellite remote sensing, UAV aerial photography, and ground surveys at the metadata level. This ensures consistency in data storage, transmission, and processing, enabling better compatibility and interaction between different data types and avoiding data confusion and errors caused by differences in data format or semantics. The standardized metadata provides clear guidance for subsequent multi-source data fusion modules. When automating the processing of heterogeneous multi-source data, the system can quickly identify and parse data based on the metadata specifications, eliminating the need for complex data format conversions and semantic understanding, thereby significantly improving data processing efficiency and shortening project cycles. Clear metadata definitions make the meaning and purpose of data immediately apparent. For project participants—whether data collectors, processors, or analysts—it's easy to quickly understand the background and characteristics of the data, facilitating data sharing and use. Furthermore, during long-term data storage and maintenance, the metadata specifications also facilitate data updates, retrieval, and management, ensuring the sustainable use of data.

[0066] The multi-source data fusion module is used to acquire multi-source heterogeneous data and automatically process it based on a multi-dimensional land classification system and metadata specifications to form a standardized dataset. Using API interfaces or batch import tools, the module processes the aforementioned multi-source heterogeneous data through an automated process based on the multi-dimensional land classification system and metadata specifications. Typical data with highly consistent land types in the collection area, a land type purity of over 90%, and verification by field photographs are selected as sample data. Interpretive information, supporting photographs, vector plots, 3D models, land type attributes, and management attributes for the area are extracted from relevant systems. The mapping relationship between the original classification system and the classification system of this method is used to unify and map land types. Samples with duplicate or missing data are removed. Standardization work, including coordinate transformation, format conversion, and terminology unification, is performed on the data to form a standardized dataset.

[0067] Specifically, the multi-source data fusion module first collects previous sample data, and then uses a dedicated batch import tool to import a large amount of high-definition image data taken by the drone team and structured data recorded by the ground survey team into the system.

[0068] Based on the previously established multidimensional land classification system and metadata specifications, the system initiates an automated processing flow. When selecting sample data, considering the complex land types in coastal areas, including beaches, wetlands, and construction land, to ensure the typicality of the samples, areas with highly consistent land types, a land type purity of over 95% (higher than the required 90%), and field photos for verification are selected as the source of sample data. For example, when selecting beach samples, it is ensured that the selected area is almost entirely pure sand, with no other land types mixed in, and that there are photos taken on-site as supporting evidence.

[0069] Interpretive information, evidence photos, vector plots, 3D models, land use attributes (such as sand type of beaches, vegetation type of wetlands, etc.), and management attributes (such as planned land use and protection level) of the region are extracted from relevant geographic information systems and land management systems. Then, the mapping relationship between the original classification system and the classification system of this method is used to unify and map the land use information from different sources of data. For example, if the classification of wetlands in a certain set of data is different from the classification system of this system, it can be accurately classified into the corresponding wetland secondary category in this system through the mapping relationship.

[0070] During processing, samples with duplicate or missing data are removed. Only one copy of the duplicate sample data is retained; samples lacking key information (such as missing land use attributes or coordinate information) are discarded. Finally, the retained data undergoes coordinate transformation to unify it to the same geographic coordinate system; format conversion is performed to ensure that both image data and structured data conform to the system's specified format standards; and terminology standardization is implemented to unify different descriptions of the same land use type or attribute from different data sources into the system's defined terminology, forming a standardized dataset.

[0071] A knowledge graph construction module is used to extract key entities and relationships from a standardized dataset to construct a land monitoring knowledge graph; in some embodiments, the knowledge graph construction module includes:

[0072] The entity extraction unit is used to extract key entities from a standardized dataset;

[0073] The relationship identification unit is used to identify and establish the relationships between entities;

[0074] The knowledge disambiguation unit is used to disambiguate ambiguous entities.

[0075] The knowledge structure optimization unit is used to supplement and improve the knowledge structure of the knowledge graph.

[0076] Specifically, the entity extraction unit first scans and analyzes the standardized dataset processed by the multi-source data fusion module. The dataset contains rich land-related information, such as land types (arable land, forest land, grassland, etc.), soil properties (pH, organic matter content, etc.), topography (mountains, plains, hills, etc.), biological resources (species and quantities of flora and fauna, etc.), and human activity information (land use planning, development projects, etc.). The entity extraction unit accurately extracts key entities from this data using natural language processing technology and specific algorithm models, such as "acidic red soil," "a forest ecosystem in a certain mountainous area," and "a commercial land development project in a certain city."

[0077] The relationship recognition unit then identifies and establishes the relationships between the extracted entities. For example, there is a "habitat-organism" relationship between "a mountain forest ecosystem" and "local wild animals". By deeply analyzing the semantic and logical connections in the data, a complex relationship network between entities is constructed.

[0078] The knowledge disambiguation unit processes ambiguous entities. In the project, it was found that the entity "wetland" has different definitions and scopes in different data sources. Some data include coastal mudflats in the category of wetlands, while others do not. The knowledge disambiguation unit combines knowledge from the land monitoring field, expert knowledge, and data context information to unify and standardize the entity "wetland", clarify its definition and boundaries in the knowledge graph of this project, and eliminate ambiguity.

[0079] The knowledge structure optimization unit is used to improve the knowledge structure and enhance reasoning ability. Based on the differences between natural language and database attribute representations, it sorts out and analyzes the initial knowledge graph, and supplements and improves the knowledge graph using domain knowledge, expert knowledge, and general knowledge from land monitoring. It adds factual descriptions of time, space, land monitoring scenarios, land use concepts, and related relationships in the context of common sense, forming a more complete, structured, and reasonable knowledge architecture. For example, the time entity adds knowledge about spring, summer, autumn, and winter, and the four seasons. On this basis, reasoning rules are formulated, a knowledge graph is formed, and imported into the database, enabling the retrieval, reasoning, and description of land monitoring samples from multiple angles and dimensions through the knowledge graph.

[0080] In some embodiments, the knowledge disambiguation unit includes:

[0081] The ambiguity identification subunit is used to analyze the contextual information of entities, data source relationships, and business logic to identify entities that have ambiguity or conflicting expressions.

[0082] The disambiguation strategy implementation subunit is used to perform semantic parsing and normalization of ambiguous entities by combining the knowledge base in the land monitoring field and preset rules.

[0083] The verification subunit is used to verify whether the disambiguated entities conform to the logical consistency of the land monitoring scenario based on the relationships between entities in the knowledge graph and business constraints.

[0084] Specifically, when processing land monitoring data, the ambiguity identification subunit analyzes entities in the standardized dataset formed after the fusion of multi-source data. For example, the entity "grassland" can be ambiguous in different data sources. In some northern regions, "grassland" may mainly refer to natural pastures; while in some southern regions, "grassland" may include urban lawns and small patches of natural grassland. The ambiguity identification subunit accurately identifies ambiguous or conflicting "grassland" entities by analyzing the entity's contextual information, such as surrounding land use patterns and vegetation cover characteristics; data source relationships, such as whether the data comes from satellite remote sensing interpretation or ground field surveys; and business logic, such as the different definition requirements for grassland in land quality assessment.

[0085] The disambiguation strategy implementation subunit combines a knowledge base in the land monitoring field, which includes various land-related professional definitions, classification standards, and preset rules. For example, it sets classification rules for grasslands based on the geographical and climatic characteristics of different regions, performs semantic analysis and normalization on the identified ambiguous "grassland" entities, and uniformly classifies "grassland" of the northern natural pasture type as "natural pasture grassland"; and classifies urban lawns and small patches of natural grassland in the south into more detailed categories such as "urban lawn" and "southern natural grassland" according to their specific characteristics.

[0086] The verification subunit is based on the relationships between entities in the knowledge graph, such as the relationship between "grassland" and surrounding entities such as "water source", "soil type" and "vegetation coverage", as well as the constraints in land quality monitoring operations, such as the quality assessment index requirements for different types of grassland. It verifies whether the various "grassland" entities after disambiguation conform to the logical consistency of the land monitoring scenario, checks whether the water source around "natural pasture grassland" meets the water demand of the pasture, and whether its soil type is suitable for grass growth, etc., to ensure the accuracy and rationality of the disambiguation results.

[0087] A vector library construction module is used to convert entities, relations, and attribute values ​​in a knowledge graph into machine-computable vectors to build a vector library of land monitoring sample data; in some embodiments, the vector library construction module includes:

[0088] The model selection unit is used to select a knowledge graph embedding model for knowledge graph embedding.

[0089] The vector transformation unit is used to convert entities, relations, and attribute values ​​in the knowledge graph into machine-computable vectors.

[0090] The normalization unit is used to perform unified normalization on the transformed vectors and build a vector library.

[0091] Specifically, the model selection unit plays its role first, taking into account the complexity of land monitoring data and the characteristics of knowledge graphs, comparing various knowledge graph embedding models, such as TransE, RotatE, and BERT (adapted and adjusted here according to the characteristics of knowledge graphs), so that the model can better fit the characteristics of the relationships between entities in the land monitoring knowledge graph.

[0092] The vector transformation unit transforms entities, relations, and attribute values ​​in the knowledge graph according to the selected model. For example, for the entity "Yangtze River Basin Wetland" in the knowledge graph, the model will transform its related features, such as geographical location, area, ecological function, and other attributes, as well as the relationship with other entities (such as surrounding cities and biological species that depend on the wetland) into a high-dimensional, machine-computable vector through complex mathematical operations. Relationships such as "wetland is adjacent to city" and "biological species depend on wetland for survival" will also be represented in a specific vector form. Attribute values ​​are also transformed into corresponding vector elements according to the model rules.

[0093] The normalization unit performs unified processing on the transformed vectors. Since the transformed vectors of different entities, relations and attribute values ​​may differ in numerical range and units, the L2 normalization method is adopted to unify the magnitude of all vectors to 1 in order to facilitate subsequent vector operations and similarity calculations. For example, after the vector of the entity "Yangtze River Basin Wetland" is transformed, the values ​​of each dimension are adjusted according to certain rules to make the overall magnitude of the vector become 1, while retaining its original directional characteristics, thereby constructing a unified vector library.

[0094] In some embodiments, the vector transformation unit includes:

[0095] The rotation embedding encoding subunit is used to perform rotation embedding encoding on entities using the RotatE model, generating low-dimensional vectors that preserve relational orientation.

[0096] The projection matrix transformation subunit is used to perform projection matrix transformation on the relation using the TransR model, mapping the head entity and tail entity to the semantic space of the relation;

[0097] The normalized scaling subunit is used to extract contextual semantic vectors from textual data in attribute values ​​using the BERT model, and to convert numerical data into fixed-dimensional vectors using normalized scaling.

[0098] The fusion subunit is used to merge vectors of entities, relations, and attribute values ​​through weighted concatenation to form a comprehensive vector representation.

[0099] Specifically, when processing entities in a knowledge graph, the Rotation Embedding Encoding subunit takes the entity "Urban Commercial Land" as an example and uses the RotatE model to perform rotation embedding encoding. Based on the relational structure of the entity in the knowledge graph, the model generates a low-dimensional vector that preserves the relational direction of the "Urban Commercial Land" entity through rotation operations. For example, considering that "Urban Commercial Land" has a "proximity" relationship with "surrounding transportation facilities" and an "association" relationship with "commercial activity type", these relational directional information are preserved in the generated low-dimensional vector during the rotation embedding encoding process, so that the vector can accurately reflect the relational characteristics of the entity in the knowledge graph.

[0100] For relations in a knowledge graph, such as the relation "land use type change", the projection matrix transformation sub-unit uses the TransR model to process them, mapping the head entity (such as "arable land") and tail entity (such as "construction land") involved in the relation to the semantic space specific to the relation "land use type change". Through the projection matrix transformation, the head entity and tail entity can better reflect their relevance to the relation "land use type change" in the semantic space, highlighting the semantic features of the relation itself.

[0101] When processing attribute values, the standardization scaling subunit uses a standardization scaling method to convert numerical attribute values ​​such as "soil organic matter content" into fixed-dimensional vectors. For example, the values ​​of soil organic matter content in different regions are standardized according to certain rules to eliminate the influence of differences in units and numerical ranges, and are converted into vectors with uniform dimensions. For textual attribute values ​​such as "land use policy description", the BERT model is used to extract contextual semantic vectors. The BERT model can deeply understand the semantic information of the text and transform the "land use policy description" text into a vector rich in semantic features.

[0102] The fusion subunit takes the vectors of entities, relationships, and attribute values ​​processed above and weights them according to their importance and relevance in land monitoring. For example, different weights are assigned to the entity vector of "urban commercial land", the relationship vector of "land use type change", the attribute value vectors of "soil organic matter content" and "land use policy description", and then they are spliced ​​together to form a comprehensive vector representation that fully covers all kinds of information related to land use.

[0103] An intelligent query module is used to receive natural language query statements, retrieve the candidate entities / relationships with the highest similarity from a vector library, and return the query results; in some embodiments, the intelligent query module includes:

[0104] The query preprocessing unit is used to preprocess natural language query statements to obtain core semantics;

[0105] The query vector transformation unit is used to transform core semantics into query vectors using a natural language processing model.

[0106] The similarity calculation unit is used to calculate the similarity between the query vector and the vectors in the vector library using the cosine similarity algorithm, retrieve the top-K candidate entities / relationships with the highest similarity, narrow down the scope, construct graph query statements, verify the structured relationships of the candidate results in the knowledge base, filter invalid results, sort the query results that meet the conditions from high to low according to the matching degree, and prioritize the display of the most relevant samples.

[0107] The results return unit is used to return the most relevant query results based on the similarity calculation.

[0108] A method for intelligent management of land monitoring samples, specifically including:

[0109] Construct a multi-dimensional land classification system;

[0110] Define metadata specifications for land monitoring sample data;

[0111] Acquire multi-source heterogeneous data and automatically process the multi-source heterogeneous data based on a multi-dimensional land classification system and metadata specifications to form a standardized dataset;

[0112] Key entities and relationships are extracted from standardized datasets to construct a land monitoring knowledge graph;

[0113] Transform entities, relationships, and attribute values ​​in knowledge graphs into machine-computable vectors to build a vector library of land monitoring sample data;

[0114] It receives a natural language query and retrieves the candidate entities / relationships with the highest similarity from the vector library, then returns the query results.

[0115] It should be noted that the method also includes verifying the query results based on the land monitoring knowledge graph.

[0116] A computing device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the intelligent management method for land monitoring samples.

[0117] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the intelligent management method for land monitoring samples.

[0118] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A land monitoring sample intelligent management system, characterized in that, include: The classification system construction module is used to construct a multidimensional land classification system; The data specification definition module is used to define the metadata specifications for land monitoring sample data. The multi-source data fusion module is used to acquire multi-source heterogeneous data and automatically process the multi-source heterogeneous data based on the multi-dimensional land classification system and metadata specifications to form a standardized dataset. The knowledge graph construction module is used to extract key entities and relationships from standardized datasets to build a land monitoring knowledge graph; The vector library construction module is used to convert entities, relations, and attribute values ​​in the knowledge graph into machine-computable vectors to build a vector library for land monitoring sample data. The intelligent query module receives natural language query statements, retrieves the candidate entities / relationships with the highest similarity from the vector library, and returns the query results.

2. The intelligent management system for land monitoring samples according to claim 1, characterized in that, The data specification definition module includes: The image data metadata definition unit is used to define the metadata fields of image data, including spatial location, acquisition time, and sensor type. The structured data metadata definition unit is used to define the metadata fields of structured data, including field name, type, length, and data format.

3. The intelligent management system for land monitoring samples according to claim 2, characterized in that, The knowledge graph construction module includes: The entity extraction unit is used to extract key entities from a standardized dataset; The relationship identification unit is used to identify and establish the relationships between entities; The knowledge disambiguation unit is used to disambiguate ambiguous entities. The knowledge structure optimization unit is used to supplement and improve the knowledge structure of the knowledge graph.

4. The intelligent management system for land monitoring samples according to claim 3, characterized in that, The knowledge disambiguation unit includes: The ambiguity identification subunit is used to analyze the contextual information of entities, data source relationships, and business logic to identify entities that have ambiguity or conflicting expressions. The disambiguation strategy implementation subunit is used to perform semantic parsing and normalization of ambiguous entities by combining the knowledge base in the land monitoring field and preset rules. The verification subunit is used to verify whether the disambiguated entities conform to the logical consistency of the land monitoring scenario based on the relationships between entities in the knowledge graph and business constraints.

5. The intelligent management system for land monitoring samples according to claim 4, characterized in that, The vector library construction module includes: The model selection unit is used to select a knowledge graph embedding model for knowledge graph embedding. The vector transformation unit is used to convert entities, relations, and attribute values ​​in the knowledge graph into machine-computable vectors. The normalization unit is used to perform unified normalization on the transformed vectors and build a vector library.

6. The intelligent management system for land monitoring samples according to claim 5, characterized in that, The vector transformation unit includes: The rotation embedding encoding subunit is used to perform rotation embedding encoding on entities using the RotatE model, generating low-dimensional vectors that preserve relational orientation. The projection matrix transformation subunit is used to perform projection matrix transformation on the relation using the TransR model, mapping the head entity and tail entity to the semantic space of the relation; The normalized scaling subunit is used to extract contextual semantic vectors from textual data in attribute values ​​using the BERT model, and to convert numerical data into fixed-dimensional vectors using normalized scaling. The fusion subunit is used to merge vectors of entities, relations, and attribute values ​​through weighted concatenation to form a comprehensive vector representation.

7. The intelligent management system for land monitoring samples according to claim 6, characterized in that, The intelligent query module includes: The query preprocessing unit is used to preprocess natural language query statements to obtain core semantics; The query vector transformation unit is used to transform core semantics into query vectors using a natural language processing model. The similarity calculation unit is used to calculate the similarity between the query vector and the vectors in the vector library using the cosine similarity algorithm. The results return unit is used to return the most relevant query results based on the similarity calculation.

8. A method for intelligent management of land monitoring samples, characterized in that, Specifically, it includes: Construct a multi-dimensional land classification system; Define the metadata specifications for land monitoring sample data; Acquire multi-source heterogeneous data and automatically process the multi-source heterogeneous data based on a multi-dimensional land classification system and metadata specifications to form a standardized dataset; Key entities and relationships are extracted from standardized datasets to construct a land monitoring knowledge graph; Transform entities, relationships, and attribute values ​​in knowledge graphs into machine-computable vectors to build a vector library of land monitoring sample data; It receives a natural language query and retrieves the candidate entities / relationships with the highest similarity from the vector library, then returns the query results.

9. A computing device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the intelligent management method for land monitoring samples as described in claim 8.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the steps of the intelligent management method for land monitoring samples as described in claim 8.

Images