Remote sensing knowledge recommendation method and device, electronic equipment and storage medium

Abstract

The application provides a remote sensing knowledge recommendation method and device, electronic equipment and storage medium, the method comprises the following steps: obtaining the first similarity score and the third similarity score between the target demand information and each node in the remote sensing knowledge graph based on the target demand information and the remote sensing knowledge graph; obtaining the second similarity score between the target demand information and each node based on the first similarity score between the target demand information and each node; calculating the comprehensive similarity score between the target demand information and each node based on the second similarity score and the third similarity score between the target demand information and each node; determining the remote sensing knowledge recommendation result based on the comprehensive similarity score. The application can ensure that the recommendation result of remote sensing knowledge accurately matches the deep semantics of user demand and covers multiple associated nodes in the remote sensing knowledge graph, and can realize more accurate and comprehensive remote sensing knowledge recommendation for user demand.

Description

Technical Field

[0001] This invention relates to the field of knowledge recommendation technology, and in particular to a remote sensing knowledge recommendation method, apparatus, electronic device, and storage medium. Background Technology

[0002] Remote sensing technology, with its multi-platform, multi-band, and high-precision data acquisition capabilities, has become a core technological support for intelligent decision-making in multiple fields. The timeliness and accuracy of remote sensing knowledge acquisition determine the application effectiveness of remote sensing technology.

[0003] However, due to problems such as terminology gaps, lack of implicit semantic understanding, and knowledge fragmentation in traditional remote sensing knowledge recommendation methods, it is difficult for traditional remote sensing knowledge recommendation methods to make timely and accurate recommendations based on user needs, resulting in a situation of "data explosion, knowledge scarcity" in remote sensing knowledge recommendation.

[0004] Among these challenges, the terminology gap refers to the semantic gap between user needs described in natural language used by non-remote sensing professionals and specialized remote sensing terminology, resulting in low relevance between the remote sensing knowledge recommended by traditional methods and user needs. The lack of implicit semantic understanding means that traditional remote sensing knowledge recommendation methods rely on keyword matching and fail to understand the implicit semantics within user needs, also leading to low relevance. Knowledge fragmentation refers to the wide scope, multiple dimensions, and strong correlations of remote sensing knowledge, which is typically scattered across different sources (documents, knowledge bases, and web pages), limiting the effectiveness of traditional knowledge recommendation methods. Therefore, how to achieve more accurate and comprehensive remote sensing knowledge recommendation tailored to user needs is a pressing technical problem that needs to be solved in this field. Summary of the Invention

[0005] This invention provides a remote sensing knowledge recommendation method, apparatus, electronic device, and storage medium to address the shortcomings of traditional knowledge recommendation methods in the prior art, which have low adaptability to user needs and are limited, thereby achieving more accurate and comprehensive remote sensing knowledge recommendation tailored to user requirements.

[0006] This invention provides a remote sensing knowledge recommendation method, comprising the following steps.

[0007] The target demand information and the remote sensing knowledge graph are respectively input into the first calculation model and the third calculation model. The first similarity score between the target demand information output by the first calculation model and each node in the remote sensing knowledge graph is obtained. The third similarity score between the target demand information and each node is obtained. The first similarity score between the target demand information and each node is input into the second calculation model. The second similarity score between the target demand information and each node is obtained.

[0008] Based on the second similarity score and the third similarity score between the target requirement information and each node, a comprehensive similarity score between the target requirement information and each node is calculated.

[0009] Based on the comprehensive similarity score between the target demand information and each node, a target node is determined among the nodes in the remote sensing knowledge graph. The node information of the target node is then used as the remote sensing knowledge recommendation result for the target demand information. The node information of any node in the remote sensing knowledge graph includes a corresponding remote sensing knowledge entry, and the endpoint node of any directed edge in the remote sensing knowledge graph is a child node of the starting point of the endpoint of that directed edge. The first calculation model is constructed based on the cosine similarity algorithm; the second calculation model is constructed based on the recursive retrieval algorithm; and the third calculation model is constructed based on the diffusion algorithm.

[0010] According to a remote sensing knowledge recommendation method provided by the present invention, the specific steps of the first calculation model obtaining the cosine similarity between the target demand information and each node and a first similarity score based on the target demand information and the remote sensing knowledge graph include: generating an embedding vector corresponding to the target demand information and an embedding vector corresponding to each node; calculating a weight factor corresponding to each node based on the hierarchy of each node; calculating the cosine similarity between the target demand information and each node using a cosine similarity algorithm based on the embedding vector corresponding to the target demand information and the embedding vector corresponding to each node; and calculating the first similarity score between the target demand information and each node based on the weight factor corresponding to each node and the cosine similarity between the target demand information and each node.

[0011] According to a remote sensing knowledge recommendation method provided by the present invention, the specific steps of the second calculation module to obtain a second similarity score between the target demand information and each node based on a first similarity score between the target demand information and each node include: calculating the second similarity score between the target demand information and each node based on the second similarity score between the target demand information and the parent node of each node, the first similarity score between the target demand information and each node, and the number of child nodes of each node in the remote sensing knowledge graph.

[0012] According to a remote sensing knowledge recommendation method provided by the present invention, the third calculation model obtains the cosine similarity and third similarity score between the target demand information and each node based on the target demand information and the remote sensing knowledge graph. The specific steps include: calculating the second similarity score between the target demand information and each node based on the third similarity score corresponding to the parent node of each node and the weight factor corresponding to the directed edge between each node and the parent node of each node.

[0013] According to a remote sensing knowledge recommendation method provided by the present invention, the step of calculating a comprehensive similarity score between the target demand information and each node based on a second similarity score and a third similarity score between the target demand information and each node includes: obtaining a second weight value corresponding to the second similarity score between the target demand information and each node and a third weight value corresponding to the third similarity score between the target demand information and each node based on the demand focus of the target demand information, wherein the demand focus of the target demand information is accuracy priority or comprehensiveness priority; and calculating a comprehensive similarity score between the target demand information and each node based on the second similarity score and the third similarity score, as well as the second weight value and the third weight value.

[0014] According to a remote sensing knowledge recommendation method provided by the present invention, before inputting the target demand information and the remote sensing knowledge graph into the first calculation model and the third calculation model respectively, the method further includes: constructing a satellite ontology knowledge base, a satellite payload knowledge base, a satellite product knowledge base, and an algorithm knowledge base; the satellite ontology knowledge base includes multiple remote sensing knowledge related to the satellite ontology; any remote sensing knowledge related to the satellite ontology in the satellite ontology knowledge base includes the name of the satellite ontology and the name of the satellite payload carried on the satellite ontology; the satellite payload knowledge base includes multiple remote sensing knowledge related to the satellite payload; any remote sensing knowledge related to the satellite payload in the satellite payload knowledge base includes... The remote sensing knowledge includes the name of any satellite payload and the name of the satellite product that can be obtained based on the data collected by any satellite payload; the satellite product knowledge base includes remote sensing knowledge related to multiple satellite products; the remote sensing knowledge related to any satellite product in the satellite product knowledge base includes the name of any satellite product and the name of the algorithm required to calculate any satellite product; the remote sensing knowledge related to any algorithm in the algorithm knowledge base includes the name of any algorithm and the calculation steps; establishing each node in the remote sensing knowledge graph with each satellite body in the satellite body knowledge base, each satellite payload in the satellite payload knowledge base, and the satellite... The system establishes a one-to-one correspondence between each satellite product in the satellite product knowledge base and each algorithm in the algorithm knowledge base. It identifies the remote sensing knowledge related to each satellite body as the node information of the corresponding node in the remote sensing knowledge graph; the remote sensing knowledge related to each satellite payload as the node information of the corresponding node in the remote sensing knowledge graph; the remote sensing knowledge related to each satellite product as the node information of the corresponding node in the remote sensing knowledge graph; and the remote sensing knowledge related to each algorithm as the node information of the corresponding node in the remote sensing knowledge graph. The hierarchy of each satellite body node in the remote sensing knowledge graph is determined as Level 1, the hierarchy of each satellite payload node is determined as Level 2, the hierarchy of each satellite product node is determined as Level 3, and the hierarchy of each algorithm node is determined as Level 4. Based on the node information and hierarchy of any node in the remote sensing knowledge graph, the child nodes of any node are determined. If the child nodes of any node are not empty, the node is determined as the starting node of a directed edge, the child nodes of any node are determined as the ending nodes of the directed edge, and the directed edge connects the node and its child nodes.

[0015] The present invention also provides a remote sensing knowledge recommendation device, comprising the following modules.

[0016] The model reasoning module is used to input target demand information and remote sensing knowledge graph into a first computing model and a third computing model, respectively, to obtain a first similarity score between the target demand information output by the first computing model and each node in the remote sensing knowledge graph, to obtain a third similarity score between the target demand information output by the third computing model and each node, to input the first similarity score between the target demand information and each node into a second computing model, and to obtain a second similarity score between the target demand information and each node output by the second computing model.

[0017] The result fusion module is used to calculate a comprehensive similarity score between the target requirement information and each node based on the second similarity score and the third similarity score between the target requirement information and each node.

[0018] The knowledge recommendation module is used to determine the target node among the nodes in the remote sensing knowledge graph based on the comprehensive similarity score between the target demand information and each node, and to determine the node information of the target node as the remote sensing knowledge recommendation result for the target demand information.

[0019] The node information of any node in the remote sensing knowledge graph includes a remote sensing knowledge correspondence, and the endpoint node of any directed edge in the remote sensing knowledge graph is a child node of the endpoint and starting point of any directed edge; the first calculation model is constructed based on the cosine similarity algorithm; the second calculation model is constructed based on the recursive retrieval algorithm; and the third calculation model is constructed based on the diffusion algorithm.

[0020] The present invention also provides an electronic device, including 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 remote sensing knowledge recommendation method as described above.

[0021] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the remote sensing knowledge recommendation method as described above.

[0022] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the remote sensing knowledge recommendation method as described above.

[0023] The remote sensing knowledge recommendation method, apparatus, electronic device, and storage medium provided by this invention utilize a first computational model based on a cosine similarity algorithm to capture the fine-grained semantic similarity between the target's information needs and nodes in the remote sensing knowledge graph based on a hierarchical structure, thereby strengthening the weight of deep-level nodes. A second computational model based on a recursive retrieval algorithm is used to optimize the similarity score of the first model, enhancing the modeling ability for complex path associations in the remote sensing knowledge graph. A third computational model based on a diffusion algorithm is used to expand the implicit associations between nodes and uncover potential semantic connections. Finally, by fusing the second similarity score output by the second computational model and the third similarity score output by the third computational model, a comprehensive similarity score is generated. This ensures that the remote sensing knowledge recommendation results accurately match the deep semantics of user needs while covering multiple associated nodes in the remote sensing knowledge graph, thus outputting more comprehensive and professionally relevant remote sensing knowledge recommendation results. This enables more accurate and comprehensive remote sensing knowledge recommendations tailored to user needs. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in this 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0025] Figure 1 This is one of the flowcharts of the remote sensing knowledge recommendation method provided by the present invention.

[0026] Figure 2 This is the second flowchart of the remote sensing knowledge recommendation method provided by the present invention.

[0027] Figure 3 This is a schematic diagram of the remote sensing knowledge recommendation device provided by the present invention.

[0028] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0030] In the description of this application, the terms "first," "second," etc., are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class, without limiting the number of objects; for example, a first object can be one or more. Furthermore, in the description of this application, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects have an "or" relationship.

[0031] It should be noted that with the development of remote sensing technology, the amount of remote sensing data has surged, but there are significant difficulties in effectively transforming this massive amount of remote sensing data into remote sensing knowledge that users can understand and utilize.

[0032] Specifically, non-remote sensing professionals (such as those in functional departments, management departments, agricultural workers, and rescue personnel) have become important users of remote sensing knowledge. Their user needs are typically expressed in natural language; for example, a user need might be the text "can see satellites in region A." However, this natural language user need has a semantic gap with remote sensing terminology, such as "Gaofen-1 satellite" or "multispectral bands." Traditional knowledge recommendation methods often rely on keyword matching or simple semantic models, which fail to identify the semantic relationship between user needs and remote sensing terminology, and struggle to grasp the implicit technical logic within user needs. This results in a low degree of fit between the remote sensing knowledge recommended by traditional methods and user requirements.

[0033] Furthermore, remote sensing knowledge involves multi-dimensional information such as satellite parameters (e.g., the revisit cycle of Gaofen-1 satellite), application scenarios (e.g., the thermal infrared band requirements for disaster monitoring), and algorithm models (e.g., NDVI vegetation index calculation). However, this knowledge is scattered across different sources such as documents, knowledge bases, and web pages. Traditional knowledge recommendation methods rely on keyword matching or single-dimensional collaborative filtering, which cannot integrate multi-modal features such as user responsibilities (e.g., the jurisdiction of agricultural departments), image attributes (e.g., cloud cover percentage), and land cover features (e.g., forest cover). They also struggle to bridge terminology gaps and establish cross-domain semantic mappings (e.g., "monitoring fires" requires associating thermal infrared bands with short-revisit satellite parameters). In addition, the heterogeneity of different data sources (e.g., differences in optical and radar image formats) and high annotation costs further exacerbate the fragmentation of remote sensing knowledge. This results in traditional knowledge recommendation methods having significant limitations when recommending remote sensing knowledge, remaining only at a superficial knowledge matching level and failing to achieve precise knowledge-driven services. Therefore, how to achieve more accurate and comprehensive remote sensing knowledge recommendation tailored to user needs is a pressing technical problem that needs to be solved in this field.

[0034] The following is combined with Figures 1-2 This invention describes the remote sensing knowledge recommendation method provided by the present invention.

[0035] Figure 1 This is one of the flowcharts illustrating the remote sensing knowledge recommendation method provided by this invention. Figure 2 This is the second flowchart illustrating the remote sensing knowledge recommendation method provided by this invention. For example... Figure 1 and Figure 2 As shown, the method includes the following steps: Step 101: Input the target demand information and the remote sensing knowledge graph into the first calculation model and the third calculation model respectively, obtain the first similarity score between the target demand information output by the first calculation model and each node in the remote sensing knowledge graph, obtain the third similarity score between the target demand information output by the third calculation model and each node, input the first similarity score between the target demand information and each node into the second calculation model, and obtain the second similarity score between the target demand information and each node output by the second calculation model.

[0036] In this system, the node information of any node in the remote sensing knowledge graph includes a remote sensing knowledge correspondence, and the endpoint node of any directed edge in the remote sensing knowledge graph is a child node of the endpoint and starting point of any directed edge; the first calculation model is constructed based on the cosine similarity algorithm; the second calculation model is constructed based on the recursive retrieval algorithm; and the third calculation model is constructed based on the diffusion algorithm.

[0037] It should be noted that the executing entity in this embodiment of the invention is a remote sensing knowledge recommendation device. This remote sensing knowledge recommendation device can be configured in electronic devices such as computers or servers.

[0038] Specifically, the target demand information is the recommendation target of the remote sensing knowledge recommendation method provided by this invention. Based on the remote sensing knowledge recommendation method provided by this invention, remote sensing knowledge with a high degree of fit with the target demand information can be retrieved from massive amounts of remote sensing knowledge and used as the remote sensing knowledge recommendation result for the target demand information.

[0039] In this embodiment of the invention, the user's needs, determined based on actual requirements, can be identified as target requirement information.

[0040] The target requirement information in this embodiment of the invention may include text information expressed in natural language, such as the text information: "Satellites in region A can be seen." The target requirement information in this embodiment of the invention may also include any one or more of image information, language information, and video information. The specific content of the target requirement information is not limited in this embodiment of the invention.

[0041] It should be noted that the target requirement information in the embodiments of the present invention can be obtained based on user input. The embodiments of the present invention do not specifically limit the user input method.

[0042] As an optional embodiment, the method further includes inputting target demand information and remote sensing knowledge graph into a first calculation model and a third calculation model, respectively, obtaining a first similarity score between the target demand information output by the first calculation model and each node in the remote sensing knowledge graph, obtaining a third similarity score between the target demand information output by the third calculation model and each node, and inputting the first similarity score between the target demand information and each node into a second calculation model to obtain a second similarity score between the target demand information and each node output by the second calculation model. Before this, the method further includes constructing a satellite ontology knowledge base, a satellite payload knowledge base, a satellite product knowledge base, and an algorithm knowledge base. The satellite ontology knowledge base includes multiple remote sensing knowledge related to satellite ontology. Any remote sensing knowledge related to any satellite ontology in the satellite ontology knowledge base includes the name of any satellite ontology and the name of the satellite payload carried on any satellite ontology, as well as the launch time and on-orbit status of any satellite ontology. The satellite payload knowledge base includes at least one of the following: duration, revisit period, abbreviation, and English abbreviation; the satellite payload knowledge base includes remote sensing knowledge related to multiple satellite payloads; the remote sensing knowledge related to any satellite payload in the satellite payload knowledge base includes the name of any satellite payload and the name of the satellite product that can be obtained based on the data collected by any satellite payload, as well as at least one of the following: spectral band information, spectral information, spatial resolution information, and temporal resolution information of any satellite payload; the satellite product knowledge base includes remote sensing knowledge related to multiple satellite products; the remote sensing knowledge related to any satellite product in the satellite product knowledge base includes the name of any satellite product and the name of the algorithm required to calculate any satellite product, as well as the type and / or application scenario of any satellite product; the algorithm knowledge base includes remote sensing knowledge related to multiple algorithms; the remote sensing knowledge related to any algorithm in the algorithm knowledge base includes the name and calculation steps of any algorithm, as well as any calculated inversion algorithm and / or verification algorithm.

[0043] Establish a one-to-one correspondence between each node in the remote sensing knowledge graph and each satellite ontology in the satellite ontology knowledge base, each satellite payload in the satellite payload knowledge base, each satellite product in the satellite product knowledge base, and each algorithm in the algorithm knowledge base. Determine the remote sensing knowledge related to each satellite ontology as the node information of the corresponding node in the remote sensing knowledge graph; determine the remote sensing knowledge related to each satellite payload as the node information of the corresponding node in the remote sensing knowledge graph; determine the remote sensing knowledge related to each satellite product as the node information of the corresponding node in the remote sensing knowledge graph; and determine the remote sensing knowledge related to each algorithm as the node information of the corresponding node in the remote sensing knowledge graph.

[0044] The hierarchy of the node corresponding to each satellite body in the remote sensing knowledge graph is determined as Level 1, the hierarchy of the node corresponding to each satellite payload is determined as Level 2, the hierarchy of the node corresponding to each satellite product is determined as Level 3, and the hierarchy of the node corresponding to each algorithm is determined as Level 4.

[0045] Based on the node information and hierarchy of any node in the remote sensing knowledge graph, the child nodes of any node are determined. If the child nodes of any node are not empty, the node is determined as the starting node of the directed edge, and the child nodes of any node are determined as the ending nodes of the directed edge. The directed edge connects any node with the child nodes of any node.

[0046] It should be noted that, before inputting the target requirement information into the first calculation model, the second calculation model, and the third calculation model respectively, a remote sensing knowledge graph has been constructed based on the data in the satellite knowledge base, the satellite payload knowledge base, the satellite product knowledge base, and the algorithm knowledge base in this embodiment of the invention.

[0047] Among them, the satellite ontology knowledge base, satellite payload knowledge base, satellite product knowledge base, and algorithm knowledge base can be generated by collecting and organizing various types of remote sensing data.

[0048] The satellite body knowledge base contains remote sensing knowledge related to multiple satellite bodies. Any remote sensing knowledge related to a satellite body may include, but is not limited to, the name of the aforementioned satellite body (e.g., Gaofen-1, optical satellite, or GF-5 satellite) and the name of the satellite payload carried on the aforementioned satellite body. The remote sensing knowledge related to the aforementioned satellite body may also include at least one of the following: launch time, on-orbit duration, revisit period, abbreviation, and English abbreviation of the aforementioned satellite body.

[0049] The satellite payload knowledge base contains remote sensing knowledge related to multiple satellite payloads. Remote sensing knowledge related to any satellite payload may include, but is not limited to, the name of the aforementioned satellite payload (e.g., a visible or short-wave infrared camera or a multispectral camera) and the name of the satellite product that can be obtained based on the data collected by the aforementioned satellite payload. The remote sensing knowledge related to the aforementioned satellite payload may also include at least one of the spectral information, spectral information, spatial resolution information, and temporal resolution information of the aforementioned satellite payload.

[0050] The satellite product knowledge base contains remote sensing knowledge related to multiple satellite products. The remote sensing knowledge related to any satellite product may include, but is not limited to, the name of the aforementioned satellite product (e.g., normalized vegetation knowledge, ratio vegetation index, or crop classification satellite product) and the name of the algorithm required to calculate the aforementioned satellite product. The remote sensing knowledge related to the aforementioned satellite product may also include the type and / or application scenario of the aforementioned satellite product.

[0051] The algorithm knowledge base contains remote sensing knowledge related to multiple algorithms. Any algorithm-related remote sensing knowledge may include the name of the algorithm (e.g., spectral unmixing algorithm) and calculation steps, as well as the inversion algorithm and / or verification algorithm of the algorithm.

[0052] In this embodiment of the invention, a remote sensing knowledge graph can be constructed based on a satellite ontology knowledge base, a satellite payload knowledge base, a satellite product knowledge base, and an algorithm knowledge base.

[0053] Specifically, in this embodiment of the invention, a correspondence can be established between any satellite ontology in the satellite ontology knowledge base and a node in the remote sensing knowledge graph. The remote sensing knowledge related to the satellite ontology is determined as the node information of the corresponding node of the satellite ontology in the remote sensing knowledge graph. The name of the node corresponding to the remote sensing knowledge related to the satellite ontology in the remote sensing knowledge graph and the type of the node corresponding to the remote sensing knowledge related to the satellite ontology in the remote sensing knowledge graph are determined as the satellite ontology.

[0054] In this embodiment of the invention, a correspondence can be established between any satellite payload in the satellite payload knowledge base and a node in the remote sensing knowledge graph. The remote sensing knowledge related to the satellite payload is determined as the node information of the corresponding node in the remote sensing knowledge graph. The name of the node corresponding to the remote sensing knowledge related to the satellite payload in the remote sensing knowledge graph and the type of the node corresponding to the remote sensing knowledge related to the satellite payload in the remote sensing knowledge graph are determined as satellite payloads.

[0055] In this embodiment of the invention, a correspondence can be established between any satellite product in the satellite product knowledge base and a node in the remote sensing knowledge graph. The remote sensing knowledge related to the satellite product is determined as the node information of the corresponding node of the satellite product in the remote sensing knowledge graph. The name of the node corresponding to the remote sensing knowledge related to the satellite product in the remote sensing knowledge graph and the type of the node corresponding to the remote sensing knowledge related to the satellite product in the remote sensing knowledge graph are determined as satellite products.

[0056] In this embodiment of the invention, a correspondence can be established between any algorithm in the algorithm knowledge base and a node in the remote sensing knowledge graph. The remote sensing knowledge related to the aforementioned algorithm is determined as the node information of the corresponding node in the remote sensing knowledge graph. The name of the node corresponding to the remote sensing knowledge related to the aforementioned algorithm in the remote sensing knowledge graph is determined, and the type of the node corresponding to the remote sensing knowledge related to the aforementioned algorithm in the remote sensing knowledge graph is determined as the algorithm.

[0057] It is understandable that satellite payloads need to be mounted on the satellite itself, and different satellite products can be obtained based on the data collected by the satellite payloads. These different satellite products can be obtained based on different algorithms.

[0058] Therefore, in this embodiment of the invention, based on the progressive pattern of remote sensing knowledge (satellite body - satellite payload - satellite product - algorithm), the level of remote sensing knowledge nodes related to satellite body in the remote sensing knowledge graph is determined as Level 1, the level of remote sensing knowledge nodes related to satellite payload in the remote sensing knowledge graph is determined as Level 2, the level of remote sensing knowledge nodes related to satellite product in the remote sensing knowledge graph is determined as Level 3, and the level of remote sensing knowledge nodes related to algorithm in the remote sensing knowledge graph is determined as Level 4.

[0059] In this embodiment of the invention, the name of the satellite payload carried on the satellite body can be determined based on remote sensing knowledge related to any satellite body in the satellite body knowledge base. Then, the node corresponding to the satellite payload carried on the satellite body in the remote sensing knowledge graph can be determined as a child node of the node corresponding to the satellite body. After determining the child node of the node corresponding to the satellite body in the remote sensing knowledge graph, the node corresponding to the satellite body can be determined as the starting node of a directed edge, and the child node of the node corresponding to the satellite body can be determined as the starting node of a directed edge. Then, a directed edge can be used to connect the node corresponding to the satellite body and its child nodes.

[0060] For example, if it is determined from remote sensing knowledge related to satellite body B that satellite body B carries satellite payload C, then node B (first-level node) corresponding to satellite body B and node C (second-level node) corresponding to satellite payload C in the remote sensing knowledge graph are connected by a directed edge, and node B is the starting node of the aforementioned directed edge.

[0061] In this embodiment of the invention, the name of the satellite product obtainable from the data collected by the satellite payload can be determined based on remote sensing knowledge related to any satellite payload in the satellite payload knowledge base. Then, the node corresponding to the satellite product obtainable from the data collected by the satellite payload in the remote sensing knowledge graph can be determined as a child node of the node corresponding to the satellite payload. After determining the child nodes of the node corresponding to the satellite payload in the remote sensing knowledge graph, the node corresponding to the satellite payload can be determined as the starting node of a directed edge, and the child nodes of the node corresponding to the satellite payload can be determined as the starting nodes of directed edges. Then, directed edges can be used to connect the node corresponding to the satellite payload and its child nodes.

[0062] For example, if it is determined from remote sensing knowledge related to satellite payload C that satellite product D can be obtained from data collected based on satellite payload C, then node C (secondary node) corresponding to satellite payload C and node D (tertiary node) corresponding to satellite product D in the remote sensing knowledge graph are connected by directed edges, and node C is the starting node of the above directed edge, and node D is the ending node of the above directed edge.

[0063] In this embodiment of the invention, the name of the algorithm required to calculate the satellite product can be determined based on remote sensing knowledge related to any satellite product in the satellite product knowledge base. Then, the node corresponding to the algorithm in the remote sensing knowledge graph can be designated as a child node of the node corresponding to the satellite product. After determining the child nodes of the node corresponding to the satellite product in the remote sensing knowledge graph, the node corresponding to the satellite product can be designated as the starting node of a directed edge, and the child nodes of the node corresponding to the satellite product can be designated as the starting nodes of directed edges. A directed edge can then connect the node corresponding to the satellite product and its child nodes.

[0064] For example, if the algorithm required to calculate satellite product D is determined to be algorithm E based on remote sensing knowledge related to satellite product D, then node D (third-level node) corresponding to satellite payload D in the remote sensing knowledge graph and node E (fourth-level node) corresponding to algorithm E are connected by a directed edge, and node D is the starting node of the directed edge and node E is the ending node of the directed edge.

[0065] It should be noted that, in the remote sensing knowledge graph of this invention, the level difference between any pair of parent and child nodes connected by directed edges is 1.

[0066] It should be noted that, in the embodiments of the present invention, the higher the level of a node in the remote sensing knowledge graph, the deeper the level of the node.

[0067] This invention constructs a satellite ontology knowledge base, a satellite payload knowledge base, a satellite product knowledge base, and an algorithm knowledge base. Based on these knowledge bases, it builds a hierarchical remote sensing knowledge graph. This unifies remote sensing knowledge related to satellite ontology, satellite payload, satellite products, and algorithms into standardized node information. Based on the hierarchical relationships and directed edge connections between nodes in the constructed remote sensing knowledge graph, it deeply integrates multi-source heterogeneous remote sensing knowledge, thereby enabling forward deduction and reverse tracing of remote sensing knowledge. This provides a more accurate and comprehensive data foundation for recommending remote sensing knowledge based on target demand information.

[0068] In the embodiments of the present invention, it can be used Represents a remote sensing knowledge graph. , Representing remote sensing knowledge graphs The set of nodes in the middle, Representing remote sensing knowledge graphs The set of directed edges.

[0069] In the embodiments of the present invention, it can be used Representing remote sensing knowledge graphs Any node in, ,use This indicates the target requirement information.

[0070] After obtaining the target requirement information, the target requirement information can be input into the first calculation model, the second calculation model, and the third calculation model, respectively.

[0071] The first computational model can calculate target requirement information based on an improved cosine similarity algorithm. Nodes in remote sensing knowledge graphs The first similarity score between them.

[0072] It should be noted that cosine similarity is a metric that measures the degree of similarity between two vectors in a given direction. Traditional cosine similarity algorithms calculate the similarity score by taking the cosine of the angle between the two vectors. The traditional cosine similarity algorithm can be expressed by the following formula:

[0073] (1)

[0074] in, Indicates target requirement information Nodes in remote sensing knowledge graphs Cosine similarity between them; Representing nodes in a remote sensing knowledge graph The corresponding embedding vector; Indicates target requirement information The corresponding embedding vector.

[0075] In this embodiment of the invention, the traditional cosine similarity algorithm is improved based on the progressive law of remote sensing knowledge, resulting in the aforementioned improved cosine similarity algorithm. This improved cosine similarity algorithm introduces a weighting factor into the traditional algorithm. Therefore, when calculating the first similarity score between the target demand information and any node in the remote sensing knowledge graph based on the improved cosine similarity algorithm, different weighting factors are assigned to the nodes according to their hierarchical level. This improves the retrieval priority of these nodes when searching for remote sensing knowledge with a high degree of matching with the target demand information.

[0076] The second computational model can be based on an improved recursive retrieval algorithm to calculate a second similarity score between the target demand information and each node in the remote sensing knowledge graph.

[0077] It should be noted that traditional recursive retrieval algorithms are algorithms that use recursion to solve retrieval problems. The core of traditional recursive retrieval algorithms lies in decomposing a large problem into structurally similar but smaller subproblems. The original problem is ultimately solved by solving these subproblems, and the decomposition process requires retrieving specific information or satisfying specific conditions. Traditional recursive retrieval algorithms can be represented by the following formula:

[0078] (2)

[0079] (3)

[0080] in, Indicates target requirement information Nodes in remote sensing knowledge graphs Recursive scoring between them; Representing nodes in a remote sensing knowledge graph The hierarchy ; Representing nodes in a remote sensing knowledge graph The parent node; in In this case, it represents a node in the remote sensing knowledge graph. hierarchy ,node The parent node is empty; Indicates target requirement information Nodes in remote sensing knowledge graphs Recursive scoring between them; Representing nodes in a remote sensing knowledge graph The set of parent nodes; Representing nodes in a remote sensing knowledge graph The number of child nodes.

[0081] It should be noted that the embodiments of the present invention define Nodes can be found in remote sensing knowledge graphs. hierarchy In this context, we can improve the accuracy of remote sensing knowledge.

[0082] In this embodiment of the invention, the traditional recursive retrieval algorithm is improved based on the progressive pattern of remote sensing knowledge, resulting in an improved recursive retrieval algorithm. The improved recursive retrieval algorithm modifies the non-directional (bidirectional) propagation of nodes in the traditional recursive retrieval algorithm to propagate only along the direction from parent node to child node, avoiding semantic confusion caused by backpropagation and conforming to the hierarchical characteristics of remote sensing knowledge.

[0083] The third calculation module can calculate the third similarity score between the target demand information and each node in the remote sensing knowledge graph based on the improved diffusion algorithm.

[0084] It should be noted that traditional diffusion algorithms are based on the diffusion phenomenon in physics to solve retrieval problems. The core of traditional diffusion algorithms lies in the selective control of propagation intensity based on structural characteristics. Traditional diffusion algorithms can be expressed by the following formula:

[0085] (4)

[0086] in, Indicates target requirement information Nodes in remote sensing knowledge graphs Diffusion fraction between; Indicates target requirement information Nodes in remote sensing knowledge graphs Diffusion fraction between; Representing nodes in a remote sensing knowledge graph The child nodes, in In this case, It is a null value; Indicates target requirement information Nodes in remote sensing knowledge graphs The diffusion fraction between them, In the case of null value, ; Representing nodes in a remote sensing knowledge graph The set of child nodes.

[0087] In this embodiment of the invention, the traditional diffusion algorithm is improved based on the progressive law of remote sensing knowledge, resulting in an improved diffusion algorithm. This improved diffusion algorithm encodes the characteristics of the "satellite body - satellite payload - satellite product" technology chain in remote sensing knowledge through a weight matrix, attenuation balance, and diffusion coefficient, enabling precise and controllable diffusion and propagation.

[0088] As an optional embodiment, the first computational model, based on target demand information and remote sensing knowledge graph, obtains the cosine similarity between target demand information and each node, as well as the first similarity score. The specific steps include: generating the embedding vector corresponding to the target demand information and the embedding vector corresponding to each node.

[0089] Specifically, the first computational model in this embodiment of the invention can generate nodes in the remote sensing knowledge graph based on graph embedding algorithms (such as the Node2Vec algorithm). corresponding embedding vector .

[0090] The first computational model in this embodiment of the invention can generate target requirement information based on a text embedding algorithm (such as Word2Vec or BERT) or a graph node mapping method. corresponding embedding vector .

[0091] The weight factor corresponding to each node is calculated based on the hierarchical calculation of each node. Based on the embedding vector corresponding to the target demand information and the embedding vector corresponding to each node, the cosine similarity between the target demand information and each node is calculated using the cosine similarity algorithm.

[0092] It should be noted that in fuzzy retrieval of remote sensing knowledge graphs, deep nodes in the remote sensing knowledge graph usually contain more specific professional information (such as nodes corresponding to satellite products or algorithms). Compared with shallow nodes (such as nodes corresponding to the satellite itself), they should be given higher initial weights to improve their retrieval priority.

[0093] Therefore, the first computational model in this embodiment of the invention can be based on nodes in the remote sensing knowledge graph. The hierarchy of nodes in the remote sensing knowledge graph is calculated using the following formula. Corresponding weighting factors:

[0094] (5)

[0095] in, Representing nodes in a remote sensing knowledge graph The corresponding weighting factor; This represents the nodes in a remote sensing knowledge graph. The hierarchy ; Indicates the overflow coefficient. ; Indicates taking and The minimum value in.

[0096] The first computational model in this embodiment of the invention uses nodes in a remote sensing knowledge graph. The corresponding weighting factor calculation introduces , can In cases where the target requirement information is large, prevent the calculation from being compared with the nodes in the remote sensing knowledge graph. The first similarity score between them overflowed.

[0097] It should be noted that the first calculation model can calculate the target requirement information based on formula (1). Nodes in remote sensing knowledge graphs Cosine similarity between .

[0098] Based on the weight factor corresponding to each node and the cosine similarity between the target requirement information and each node, the first similarity score between the target requirement information and each node is calculated.

[0099] Specifically, the first calculation model calculates the target requirement information. Nodes in remote sensing knowledge graphs Cosine similarity between Nodes in remote sensing knowledge graphs Corresponding weighting factors The target requirement information can then be calculated using the following formula. Nodes in remote sensing knowledge graphs First similarity score between them:

[0100] (6)

[0101] in, Indicates target requirement information Nodes in remote sensing knowledge graphs The first similarity score between them.

[0102] To facilitate the explanation of the specific steps of obtaining the first similarity score between the target demand information and each node based on the target demand information and remote sensing knowledge graph in the first calculation model of the present invention, the following example illustrates the first calculation model in the present invention.

[0103] The target requirement information in this example is "hyperspectral agricultural detection satellite". The feature information of each node in the remote sensing knowledge graph is shown in Table 1.

[0104] Table 1. Feature information of each node in the remote sensing knowledge graph.

[0105]

[0106] The target requirement information and the remote sensing knowledge graph mentioned above are input into the first calculation model. The first similarity score between the target requirement information and each node in the remote sensing knowledge graph output by the first calculation model is shown in Table 2.

[0107] Table 2 Analysis of Results Output from the First Computational Model

[0108]

[0109] The specific calculation process is as follows:

[0110] Optical satellites: ;

[0111] GF-5 satellite: ;

[0112] Visible shortwave infrared camera: ;

[0113] Crop classification products: ;

[0114] Spectral unmixing algorithm: .

[0115] This invention, through a depth-weighted cosine similarity algorithm, can strengthen the initial weights of deep nodes in a remote sensing knowledge graph according to user needs, better adapt to the hierarchical characteristics of remote sensing knowledge, and thus provide more accurate hierarchical semantic matching capabilities for remote sensing knowledge recommendation.

[0116] This invention effectively solves the problem of deep semantic information attenuation in traditional cosine similarity algorithms by configuring dynamic weight factors for nodes at different levels in remote sensing knowledge graphs. It avoids the problem of deep key details being easily submerged by shallow general concepts when performing semantic matching in remote sensing knowledge graphs with rich hierarchical structures, making remote sensing knowledge recommendation results more in line with users' deep needs for professional domain knowledge and more sensitive to capturing the professional intentions and detailed requirements contained in users' needs.

[0117] As an optional embodiment, the second calculation module obtains a second similarity score between the target demand information and each node based on a first similarity score between the target demand information and each node. The specific steps include: calculating the second similarity score between the target demand information and each node based on the second similarity score between the target demand information and the parent node of each node, the first similarity score between the target demand information and each node, and the number of child nodes of each node in the remote sensing knowledge graph.

[0118] It should be noted that the traditional recursive retrieval algorithm shown in formula (2) represents the target demand information. Nodes in remote sensing knowledge graphs Recursive scores between This is target demand information. With nodes parent node The average of the recursive scores is calculated without taking into account the influence of other factors.

[0119] In this embodiment of the invention, the non-directional propagation (bidirectional propagation) of nodes in the traditional recursive retrieval algorithm is modified to propagate only along the direction from the parent node to the child node, avoiding semantic confusion caused by back propagation. This conforms to the hierarchical characteristics of remote sensing knowledge and also conforms to the actual design principle in the field of remote sensing technology that "the satellite itself determines the payload capability".

[0120] The second calculation model can obtain the target requirement information using the following formula. Nodes in remote sensing knowledge graphs Second similarity score between them:

[0121] (7)

[0122] (8)

[0123] in, Indicates target requirement information Nodes in remote sensing knowledge graphs The second similarity score between them; Representing nodes in a remote sensing knowledge graph The hierarchy ; Representing nodes in a remote sensing knowledge graph The parent node; in In this case, it represents a node in the remote sensing knowledge graph. hierarchy ,node The parent node is empty; Indicates target requirement information Nodes in remote sensing knowledge graphs The second similarity score between them; Representing nodes in a remote sensing knowledge graph The set of parent nodes; Representing nodes in a remote sensing knowledge graph The number of child nodes; Indicates target requirement information Nodes in remote sensing knowledge graphs Cosine similarity between them; This represents the recursive decay coefficient.

[0124] It should be noted that the embodiments of the present invention define Nodes can be found in remote sensing knowledge graphs. hierarchy In this context, we can improve the accuracy of remote sensing knowledge.

[0125] It should be noted that in formula (8) Representing nodes in a remote sensing knowledge graph parent node The degree of contribution, if If it stops spreading, then the transmission will cease.

[0126] It should be noted that the hierarchical relationships between nodes in a remote sensing knowledge graph are relatively clear, and higher-level nodes have a strong guiding effect on lower-level nodes. Therefore, the recursive decay coefficient... It can be set relatively large, for example This allows the second computational model to place greater emphasis on the contributions of neighboring nodes, enabling recursive retrieval to delve deeper into relevant nodes, while the general domain recursive decay coefficient... The recursive decay coefficient is typically set to 0.7. Domain optimization can preserve more levels of authority.

[0127] This invention, through deep fusion of domain knowledge, directional constraints, and attenuation optimization, hierarchically encodes nodes in the remote sensing knowledge graph into a traditional recursive retrieval algorithm, significantly improving the accuracy of remote sensing knowledge recommendation. Dynamic termination and dilution control reduce invalid computations, significantly improving the efficiency of remote sensing knowledge recommendation. An explicit propagation path (satellite body → satellite payload → satellite product → algorithm) supports traceability of remote sensing knowledge recommendation results, assists expert verification, and thus enhances user trust.

[0128] As an optional embodiment, the third calculation model, based on target demand information and remote sensing knowledge graph, obtains the cosine similarity between target demand information and each node and the third similarity score through the following specific steps: calculating the second similarity score between target demand information and each node based on the third similarity score corresponding to the parent node of each node and the weight factor corresponding to the directed edge between each node and its parent node.

[0129] It should be noted that the traditional diffusion algorithm shown in formula (4) is undirected diffusion, where all nodes connected by edges in the knowledge graph propagate equally. Furthermore, the traditional diffusion algorithm shown in formula (4) ignores the weight factors corresponding to the edges in the knowledge graph, and cannot distinguish the importance of the relationships between different nodes. The traditional diffusion algorithm in formula (4) lacks a decay mechanism, which may lead to the infinite diffusion of information.

[0130] Therefore, in this embodiment of the invention, by using a weight matrix, attenuation balance, and diffusion coefficient, the characteristics of the "satellite body - satellite payload - satellite product" technology chain in remote sensing knowledge are encoded, enabling precise and controllable diffusion and propagation.

[0131] The third calculation model can obtain the target requirement information using the following formula. Nodes in remote sensing knowledge graphs Third similarity score between them:

[0132]

[0133] (9)

[0134] in, Representing nodes in a remote sensing knowledge graph With nodes parent node The weight factor corresponding to the directed edges between nodes. With nodes parent node Weight factors corresponding to the directed edges between them The initial value is 1.0, and then the nodes in the remote sensing knowledge graph are weighted based on the domain weight matrix (1.0 / 0.8 / 0.7 / 0.5). With nodes parent node Weight factors corresponding to the directed edges between them Updates will be made to quantify the strength of technical connections between different relationships in satellite engineering.

[0135] This represents the diffusion coefficient. Because the semantic relationships between nodes in a remote sensing knowledge graph are highly specialized, a higher value indicates a greater likelihood of diffusion. The value can emphasize the information transmission to directed neighbor nodes, improving diffusion efficiency. The diffusion coefficient in this embodiment of the invention... You can take 0.9.

[0136] Because traditional expansion algorithms lack a decay mechanism, the initial accumulation can lead to artificially inflated scores for nodes at higher levels. Therefore, in this embodiment of the invention... It can be used to control the information ratio.

[0137] Step 102: Calculate the comprehensive similarity score between the target requirement information and each node based on the second and third similarity scores between the target requirement information and each node.

[0138] Specifically, acquiring target requirement information and nodes in the remote sensing knowledge graph. Second similarity score between Second similarity score Subsequently, the target requirement information and nodes can be calculated using numerical methods. Overall similarity score between .

[0139] As an optional embodiment, a comprehensive similarity score between the target demand information and each node is calculated based on the second similarity score and the third similarity score between the target demand information and each node. This includes: obtaining a second weight value corresponding to the second similarity score between the target demand information and each node and a third weight value corresponding to the third similarity score between the target demand information and each node, based on the demand focus of the target demand information, where the demand focus of the target demand information is either accuracy priority or comprehensiveness priority.

[0140] It should be noted that when the focus of the target demand information is on accuracy, the second weight value can be adjusted. If set to 0.6, then the third weight value is... When the focus of the target demand information is on comprehensiveness, the second weight value can be... If set to 0.4, then the third weight value is... .For example This can increase the weight of the second similarity score between the target requirement information and each node. This can increase the weight of the third similarity score between the target requirement information and each node.

[0141] Based on the second and third similarity scores, as well as the second and third weight values, between the target requirement information and each node, a comprehensive similarity score between the target requirement information and each node is calculated.

[0142] Specifically, target demand information Nodes in remote sensing knowledge graphs The overall similarity score between them can be calculated using the following formula:

[0143]

[0144] Step 103: Based on the comprehensive similarity score between the target demand information and each node, determine the target node among all nodes in the remote sensing knowledge graph, and determine the node information of the target node as the remote sensing knowledge recommendation result for the target demand information.

[0145] Specifically, after obtaining the comprehensive similarity score between the target demand information and each node in the remote sensing knowledge graph, the nodes in the remote sensing knowledge graph can be sorted from high to low according to the comprehensive similarity score to obtain the node sequence of the remote sensing knowledge graph.

[0146] After obtaining the node sequence of the remote sensing knowledge graph, the top-ranked number of nodes in the sequence can be identified as target nodes. Then, the node information of the target nodes can be used to determine the remote sensing knowledge recommendation results based on the target needs.

[0147] This invention utilizes a first computational model based on a cosine similarity algorithm to capture the fine-grained semantic similarity between target information needs and nodes in a remote sensing knowledge graph based on hierarchical structure, strengthening the weight of deep-level nodes. A second computational model based on a recursive retrieval algorithm optimizes the similarity score of the first model, enhancing the modeling ability for complex path associations in the remote sensing knowledge graph. A third computational model based on a diffusion algorithm expands the implicit associations between nodes, uncovering potential semantic connections. Finally, by fusing the second similarity score output by the second computational model and the third similarity score output by the third computational model, a comprehensive similarity score is generated. This ensures that the remote sensing knowledge recommendation results accurately match the deep semantics of user needs while covering multiple associated nodes in the remote sensing knowledge graph, thus outputting more comprehensive and professionally relevant remote sensing knowledge recommendation results. This enables more accurate and comprehensive remote sensing knowledge recommendations tailored to user needs.

[0148] The remote sensing knowledge recommendation method provided by this invention achieves a balance between accuracy and scalability in remote sensing knowledge graph retrieval through rigorous mathematical modeling and domain parameter calibration, providing an interpretable and verifiable algorithmic foundation for the application of remote sensing technology.

[0149] In knowledge graph fuzzy retrieval, the depth-weighted initial similarity algorithm adds a depth-weighting coefficient to ensure that deeper nodes contain more specific professional information (such as satellite parameters, product algorithms, etc.).

[0150] The revised recursive retrieval algorithm makes neighbor nodes non-directional (bidirectional propagation), only propagating along the parent-child direction, avoiding semantic confusion caused by back propagation. It conforms to the hierarchical design of aerospace engineering, allowing only top-down propagation. The technical hierarchy of "platform → payload → product → algorithm" in the satellite field is consistent with the actual design principle of "platform determines payload capability" in aerospace engineering.

[0151] The revised diffusion algorithm encodes the characteristics of the "platform-payload-product" technology chain in satellite engineering through weight matrix, attenuation balance, and diffusion coefficient, achieving precise and controllable propagation.

[0152] Dynamic weight optimization automatically adjusts weight values ​​based on target requirement information (terminology / scenario), achieving a balance between accuracy and scalability.

[0153] The remote sensing knowledge recommendation method provided by this invention utilizes a depth-weighted initial similarity algorithm, combining the advantages of a modified recursive retrieval algorithm and a modified diffusion algorithm. It introduces dynamic weights to balance the contributions of recursive retrieval scores and directional diffusion scores, enabling users to accurately and comprehensively search for professional information belonging to a satellite system after fuzzy input of a space mission.

[0154] Figure 3This is a schematic diagram of the remote sensing knowledge recommendation device provided by the present invention. The following is in conjunction with… Figure 3 The remote sensing knowledge recommendation device provided by this invention will be described below. The remote sensing knowledge recommendation device described below can be referred to in correspondence with the remote sensing knowledge recommendation method provided by this invention described above. For example... Figure 3 As shown, the device includes: a model reasoning module 301, a result fusion module 302, and a knowledge recommendation module 303.

[0155] The model reasoning module 301 is used to input the target demand information and the remote sensing knowledge graph into the first calculation model and the third calculation model respectively, obtain the first similarity score between the target demand information output by the first calculation model and each node in the remote sensing knowledge graph, obtain the third similarity score between the target demand information output by the third calculation model and each node, input the first similarity score between the target demand information and each node into the second calculation model, and obtain the second similarity score between the target demand information and each node output by the second calculation model.

[0156] The result fusion module 302 is used to calculate the comprehensive similarity score between the target requirement information and each node based on the second similarity score and the third similarity score between the target requirement information and each node.

[0157] The knowledge recommendation module 303 is used to determine the target node among the nodes in the remote sensing knowledge graph based on the comprehensive similarity score between the target demand information and each node, and to determine the node information of the target node as the remote sensing knowledge recommendation result for the target demand information.

[0158] In this system, the node information of any node in the remote sensing knowledge graph includes a remote sensing knowledge correspondence, and the endpoint node of any directed edge in the remote sensing knowledge graph is a child node of the endpoint and starting point of any directed edge; the first calculation model is constructed based on the cosine similarity algorithm; the second calculation model is constructed based on the recursive retrieval algorithm; and the third calculation model is constructed based on the diffusion algorithm.

[0159] Specifically, the model reasoning module 301, the result fusion module 302, and the knowledge recommendation module 303 are electrically connected.

[0160] The remote sensing knowledge recommendation device in this embodiment of the invention utilizes a first computational model based on a cosine similarity algorithm to capture the fine-grained semantic similarity between the target's information needs and nodes in the remote sensing knowledge graph based on a hierarchical structure, thereby strengthening the weight of deep-level nodes. A second computational model based on a recursive retrieval algorithm is used to optimize the similarity score of the first model, enhancing the modeling ability for complex path associations in the remote sensing knowledge graph. A third computational model based on a diffusion algorithm is used to expand the implicit associations between nodes and uncover potential semantic connections. Finally, by fusing the second similarity score output by the second computational model and the third similarity score output by the third computational model, a comprehensive similarity score is generated. This ensures that the remote sensing knowledge recommendation results accurately match the deep semantics of the user's needs while covering multiple associated nodes in the remote sensing knowledge graph, thus outputting more comprehensive and professionally relevant remote sensing knowledge recommendation results. This enables more accurate and comprehensive remote sensing knowledge recommendations tailored to user needs.

[0161] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4 As shown, the electronic device may include: a processor 410, a communications interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communications interface 420, and the memory 430 communicate with each other through the communication bus 440. The processor 410 can call logical instructions in the memory 430 to execute a remote sensing knowledge recommendation method. This method includes: inputting target demand information and a remote sensing knowledge graph into a first computational model and a third computational model respectively; obtaining a first similarity score between the target demand information output by the first computational model and each node in the remote sensing knowledge graph; obtaining a third similarity score between the target demand information output by the third computational model and each node; inputting the first similarity score between the target demand information and each node into a second computational model; obtaining a second similarity score between the target demand information and each node output by the second computational model; and based on the second similarity score between the target demand information and each node... The system calculates a comprehensive similarity score between the target demand information and each node using both a similarity score and a third similarity score. Based on this comprehensive similarity score, the system identifies the target node from among all nodes in the remote sensing knowledge graph. The node information of the target node is then used as the remote sensing knowledge recommendation result for the target demand information. Each node in the remote sensing knowledge graph contains a corresponding remote sensing knowledge entry, and the endpoint node of any directed edge in the graph is a child node of the starting point of that directed edge. The first calculation model is based on the cosine similarity algorithm; the second calculation model is based on the recursive retrieval algorithm; and the third calculation model is based on the diffusion algorithm.

[0162] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0163] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the remote sensing knowledge recommendation method provided by the above methods. This method includes: inputting target demand information and a remote sensing knowledge graph into a first calculation model and a third calculation model respectively; obtaining a first similarity score between the target demand information output by the first calculation model and each node in the remote sensing knowledge graph; obtaining a third similarity score between the target demand information output by the third calculation model and each node; inputting the first similarity score between the target demand information and each node into a second calculation model; and obtaining a third similarity score between the target demand information and each node output by the second calculation model. The system calculates a second similarity score between each node; based on the second and third similarity scores between the target demand information and each node, it calculates a comprehensive similarity score between the target demand information and each node; based on the comprehensive similarity score between the target demand information and each node, it identifies the target node among all nodes in the remote sensing knowledge graph, and determines the node information of the target node as the remote sensing knowledge recommendation result for the target demand information; wherein, the node information of any node in the remote sensing knowledge graph includes a remote sensing knowledge correspondence, and the endpoint node of any directed edge in the remote sensing knowledge graph is a child node of the endpoint and starting point of any directed edge; the first calculation model is constructed based on the cosine similarity algorithm; the second calculation model is constructed based on the recursive retrieval algorithm; and the third calculation model is constructed based on the diffusion algorithm.

[0164] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the remote sensing knowledge recommendation method provided by the methods described above. This method includes: inputting target demand information and a remote sensing knowledge graph into a first computational model and a third computational model, respectively; obtaining a first similarity score between the target demand information output by the first computational model and each node in the remote sensing knowledge graph; obtaining a third similarity score between the target demand information output by the third computational model and each node; inputting the first similarity score between the target demand information and each node into a second computational model; and obtaining a second similarity score between the target demand information and each node output by the second computational model. Based on the second and third similarity scores between the target demand information and each node, a comprehensive similarity score between the target demand information and each node is calculated. Based on the comprehensive similarity score between the target demand information and each node, the target node is determined among all nodes in the remote sensing knowledge graph, and the node information of the target node is determined as the remote sensing knowledge recommendation result for the target demand information. Among them, the node information of any node in the remote sensing knowledge graph includes a remote sensing knowledge correspondence, and the endpoint node of any directed edge in the remote sensing knowledge graph is the child node of the endpoint and starting point of any directed edge. The first calculation model is constructed based on the cosine similarity algorithm; the second calculation model is constructed based on the recursive retrieval algorithm; and the third calculation model is constructed based on the diffusion algorithm.

[0165] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0166] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0167] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A remote sensing knowledge recommendation method, characterized in that, include: The target demand information and the remote sensing knowledge graph are respectively input into the first calculation model and the third calculation model. The first similarity score between the target demand information output by the first calculation model and each node in the remote sensing knowledge graph is obtained. The third similarity score between the target demand information and each node is obtained. The first similarity score between the target demand information and each node is input into the second calculation model. The second similarity score between the target demand information and each node is obtained. Based on the second similarity score and the third similarity score between the target demand information and each node, a comprehensive similarity score between the target demand information and each node is calculated; Based on the comprehensive similarity score between the target demand information and each node, a target node is determined among the nodes in the remote sensing knowledge graph, and the node information of the target node is determined as the remote sensing knowledge recommendation result for the target demand information. Wherein, the node information of any node in the remote sensing knowledge graph includes a remote sensing knowledge correspondence, and the endpoint node of any directed edge in the remote sensing knowledge graph is a child node of the starting node of any directed edge; the first calculation model is constructed based on the cosine similarity algorithm; the second calculation model is constructed based on the recursive retrieval algorithm; and the third calculation model is constructed based on the diffusion algorithm. The first calculation model, based on the target demand information and the remote sensing knowledge graph, obtains the cosine similarity and a first similarity score between the target demand information and each node through the following specific steps: generating an embedding vector corresponding to the target demand information and an embedding vector corresponding to each node; calculating a weight factor corresponding to each node based on the hierarchy of each node; calculating the cosine similarity between the target demand information and each node using a cosine similarity algorithm based on the embedding vector corresponding to the target demand information and the embedding vector corresponding to each node; and calculating a first similarity score between the target demand information and each node based on the weight factor corresponding to each node and the cosine similarity between the target demand information and each node. The second calculation model obtains a second similarity score between the target demand information and each node based on a first similarity score between the target demand information and each node. The specific steps include: calculating the second similarity score between the target demand information and each node based on the second similarity score between the target demand information and the parent node of each node, the first similarity score between the target demand information and each node, and the number of child nodes of each node in the remote sensing knowledge graph. The third calculation model obtains the third similarity score between the target demand information and each node based on the target demand information and the remote sensing knowledge graph. The specific steps include: calculating the third similarity score between the target demand information and each node based on the third similarity score corresponding to the parent node of each node and the weight factor corresponding to the directed edge between each node and the parent node of each node.

2. The remote sensing knowledge recommendation method according to claim 1, characterized in that, The calculation of a comprehensive similarity score between the target requirement information and each node, based on the second and third similarity scores between the target requirement information and each node, includes: Based on the demand focus of the target demand information, a second weight value corresponding to the second similarity score between the target demand information and each node and a third weight value corresponding to the third similarity score between the target demand information and each node are obtained, wherein the demand focus of the target demand information is either accuracy priority or comprehensiveness priority. Based on the second similarity score and the third similarity score between the target requirement information and each node, as well as the second weight value and the third weight value, a comprehensive similarity score between the target requirement information and each node is calculated.

3. The remote sensing knowledge recommendation method according to claim 1 or 2, characterized in that, Before inputting the target demand information and remote sensing knowledge graph into the first and third computational models respectively, the method further includes: A satellite ontology knowledge base, a satellite payload knowledge base, a satellite product knowledge base, and an algorithm knowledge base are constructed. The satellite ontology knowledge base includes multiple remote sensing knowledge related to satellites. Each satellite-related remote sensing knowledge in the satellite ontology knowledge base includes the name of the satellite and the name of the satellite payload carried by that satellite. The satellite payload knowledge base includes multiple remote sensing knowledge related to satellite payloads. Each satellite payload-related remote sensing knowledge in the satellite payload knowledge base includes the name of the satellite payload and the name of the satellite product that can be obtained based on the data collected by that satellite payload. The satellite product knowledge base includes multiple remote sensing knowledge related to satellite products. Each satellite product-related remote sensing knowledge in the satellite product knowledge base includes the name of the satellite product and the name of the algorithm required to calculate that satellite product. The algorithm knowledge base includes the name of the algorithm and the calculation steps for each algorithm. Establish a one-to-one correspondence between each node in the remote sensing knowledge graph and each satellite ontology in the satellite ontology knowledge base, each satellite payload in the satellite payload knowledge base, each satellite product in the satellite product knowledge base, and each algorithm in the algorithm knowledge base. Determine the remote sensing knowledge related to each satellite ontology as the node information of the corresponding node in the remote sensing knowledge graph; determine the remote sensing knowledge related to each satellite payload as the node information of the corresponding node in the remote sensing knowledge graph; determine the remote sensing knowledge related to each satellite product as the node information of the corresponding node in the remote sensing knowledge graph; and determine the remote sensing knowledge related to each algorithm as the node information of the corresponding node in the remote sensing knowledge graph. The hierarchy of the node corresponding to each satellite body in the remote sensing knowledge graph is determined as Level 1, the hierarchy of the node corresponding to each satellite payload is determined as Level 2, the hierarchy of the node corresponding to each satellite product is determined as Level 3, and the hierarchy of the node corresponding to each algorithm is determined as Level 4. Based on the node information and level of any node in the remote sensing knowledge graph, the child nodes of any node are determined. If the child nodes of any node are not empty, the node is determined as the starting node of the directed edge, the child nodes of any node are determined as the ending nodes of the directed edge, and the directed edge is used to connect the node and the child nodes of the node.

4. A remote sensing knowledge recommendation device, characterized in that, include: The model reasoning module is used to input target demand information and remote sensing knowledge graph into a first calculation model and a third calculation model, respectively, to obtain a first similarity score between the target demand information output by the first calculation model and each node in the remote sensing knowledge graph, to obtain a third similarity score between the target demand information output by the third calculation model and each node, to input the first similarity score between the target demand information and each node into a second calculation model, and to obtain a second similarity score between the target demand information and each node output by the second calculation model. The first calculation model, based on the target demand information and the remote sensing knowledge graph, obtains the cosine similarity and a first similarity score between the target demand information and each node through the following specific steps: generating an embedding vector corresponding to the target demand information and an embedding vector corresponding to each node; calculating a weight factor corresponding to each node based on the hierarchy of each node; calculating the cosine similarity between the target demand information and each node using a cosine similarity algorithm based on the embedding vector corresponding to the target demand information and the embedding vector corresponding to each node; and calculating a first similarity score between the target demand information and each node based on the weight factor corresponding to each node and the cosine similarity between the target demand information and each node. The second calculation model obtains a second similarity score between the target demand information and each node based on a first similarity score between the target demand information and each node. The specific steps include: calculating the second similarity score between the target demand information and each node based on the second similarity score between the target demand information and the parent node of each node, the first similarity score between the target demand information and each node, and the number of child nodes of each node in the remote sensing knowledge graph. The specific steps of the third calculation model to obtain the third similarity score between the target demand information and each node based on the target demand information and the remote sensing knowledge graph include: calculating the third similarity score between the target demand information and each node based on the third similarity score corresponding to the parent node of each node and the weight factor corresponding to the directed edge between each node and the parent node of each node. The result fusion module is used to calculate the comprehensive similarity score between the target requirement information and each node based on the second similarity score and the third similarity score between the target requirement information and each node. The knowledge recommendation module is used to determine the target node among the nodes in the remote sensing knowledge graph based on the comprehensive similarity score between the target demand information and each node, and to determine the node information of the target node as the remote sensing knowledge recommendation result for the target demand information. The node information of any node in the remote sensing knowledge graph includes a remote sensing knowledge correspondence, and the endpoint node of any directed edge in the remote sensing knowledge graph is a child node of the starting node of any directed edge; the first calculation model is constructed based on the cosine similarity algorithm; the second calculation model is constructed based on the recursive retrieval algorithm; and the third calculation model is constructed based on the diffusion algorithm.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the remote sensing knowledge recommendation method as described in any one of claims 1 to 3.

6. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the remote sensing knowledge recommendation method as described in any one of claims 1 to 3.

7. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the remote sensing knowledge recommendation method as described in any one of claims 1 to 3.

Images