CN115269899A - Remote sensing image overall planning system based on remote sensing knowledge map - Google Patents
Remote sensing image overall planning system based on remote sensing knowledge map Download PDFInfo
- Publication number
- CN115269899A CN115269899A CN202211029192.XA CN202211029192A CN115269899A CN 115269899 A CN115269899 A CN 115269899A CN 202211029192 A CN202211029192 A CN 202211029192A CN 115269899 A CN115269899 A CN 115269899A
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- knowledge
- entity
- module
- graph
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/53—Querying
- G06F16/535—Filtering based on additional data, e.g. user or group profiles
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/5866—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using information manually generated, e.g. tags, keywords, comments, manually generated location and time information
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Computational Linguistics (AREA)
- Library & Information Science (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides a remote sensing image overall planning system based on a remote sensing knowledge graph, and relates to the technical field of knowledge graphs. According to the remote sensing knowledge graph construction method, the remote sensing image is firstly interpreted, the remote sensing knowledge graph is constructed in an auxiliary mode through the interpretation information, the information of the remote sensing image is greatly utilized, the semantic content of the remote sensing knowledge graph is enriched, accurate recommendation can be achieved when the image is recommended, and the experience and satisfaction of a user are improved.
Description
Technical Field
The invention relates to the technical field of knowledge maps, in particular to a remote sensing image overall planning system based on a remote sensing knowledge map.
Background
With the rapid development of earth observation technology, the observation capability of human beings on the earth reaches an unprecedented level. The satellite loads with different imaging modes, different wave bands and different resolutions enable the remote sensing data to be diversified day by day, the remote sensing data volume is obviously increased, the updating period is shortened, and the timeliness is stronger and stronger. The telemetric data exhibits significant "big data" characteristics.
As an important component of artificial intelligence, the knowledge graph was formally proposed by Google in 2012, 5/17, and its original purpose is to improve the capability of search engine and enhance the search quality and search experience of users. With the continuous development of artificial intelligence, knowledge maps have been widely applied in information retrieval, intelligent question-answering and recommendation systems and other scenarios.
The recommendation algorithms in the field of remote sensing mainly include recommendation algorithms based on collaborative filtering, recommendation algorithms based on content and hybrid recommendation algorithms. The collaborative filtering method builds a model by using behavior preference data of user history, but the method is often troubled by the sparse problem of behavior relation data between users and articles and the cold start problem when recommending new users or new articles. The content-based recommendation algorithm mainly performs complex feature extraction and modeling on the article, and the complex feature engineering is often involved and is not explanatory. When the current recommendation system recommends images, semantic information of the images can be ignored, accurate recommendation can not be performed through semantic description of the images, and output recommendation results are often not in line with expectations of users, so that the experience and satisfaction of the users are greatly reduced.
Disclosure of Invention
Based on the technical problems, the remote sensing image overall planning system based on the remote sensing knowledge graph provided by the invention has the advantages that the remote sensing image is firstly interpreted, the remote sensing knowledge graph is constructed in an auxiliary manner by utilizing the interpretation information, the information of the remote sensing image is greatly utilized, the semantic content of the remote sensing knowledge graph is enriched, accurate recommendation can be realized when the image is recommended, and the experience and the satisfaction of a user are improved.
In order to achieve the technical purpose, the invention provides a remote sensing image overall planning system based on a remote sensing knowledge graph, which comprises:
the remote sensing knowledge acquisition module is configured to acquire a remote sensing image set and automatically interpret the remote sensing data set to obtain interpretation information, wherein the interpretation information comprises: the method comprises the steps that object categories contained in remote sensing images and related background knowledge and attribute knowledge are contained, wherein the attribute knowledge comprises associated attributes and characteristic attributes;
the knowledge map building module is configured to build a remote sensing ontology base according to the interpretation information, identify a remote sensing entity by combining characteristic attributes, extract the entity-association and entity-background relation of the remote sensing image according to the association attributes and background knowledge in the attribute knowledge to form a plurality of triples, and further build the remote sensing knowledge map;
the system comprises a user information acquisition module, a word vector generation module and a word vector generation module, wherein the user information acquisition module is configured to acquire user query information, perform semantic analysis on the user query information, extract key words and associated words which are interested by a user and map the key words and the associated words into word vectors;
the knowledge map storage module is configured to perform vectorization representation on a remote sensing knowledge map, the remote sensing knowledge map is converted into a plurality of entity vectors and relationship vectors, and then the triple vectors are obtained through calculation;
and the personalized recommendation module is configured to calculate the similarity of the current word vector and the triple vectors, sort the triple vectors from high to low according to the calculation result of the similarity, list the first k triple vectors in a recommendation candidate table, screen the recommendation candidate table to obtain recommended remote sensing images, and output the recommended remote sensing images as recommendation results.
In an embodiment of the present invention, the remote sensing data set includes a remote sensing data set and a remote sensing image set, the remote sensing data set is a text data recorded in the remote sensing data set, and includes a phenology data, a geosciences data, a geospatial data, and a history data, and the remote sensing data set is interpreted, that is, the text data is analyzed and filed to form background knowledge related to the remote sensing image.
In an embodiment of the present invention, automatically interpreting the remote sensing image set includes:
preprocessing the remote sensing image set to enable the quality of the remote sensing image set to meet the requirement of interpretation;
and constructing a deep learning network model, interpreting the remote sensing image set, and combining the background knowledge to obtain interpretation information.
In an embodiment of the present invention, the user information collecting module includes:
the current user information acquisition unit is configured to acquire current user query information and perform semantic analysis on the current user query information to obtain a current word vector;
the historical user information acquisition unit is configured to acquire historical user query information and perform semantic analysis on the historical user query information to obtain a historical word vector;
the current user information acquisition unit and the historical user information acquisition unit respectively comprise semantic analysis models, and the semantic analysis models adopt natural language analysis tools.
In an embodiment of the present invention, the system further includes:
the knowledge map embedding module is configured to carry out vector representation learning on the remote sensing knowledge map by combining historical word vectors to obtain an updated and completed remote sensing knowledge map;
the knowledge graph embedding module is positioned between the user information acquisition module and the knowledge graph storage module.
In a specific embodiment of the invention, the knowledge map embedding module comprises a knowledge map embedding model, and the knowledge map embedding model is a distance-based model and is used for learning the embedded representation of the remote sensing entity and the relationship and updating the remote sensing knowledge map.
In a specific embodiment of the present invention, the knowledge graph embedding module is further configured to sort the remote sensing entities in the updated remote sensing knowledge graph, and add the historical word vectors into the prediction process to predict new triples and complete the remote sensing knowledge graph.
In an embodiment of the present invention, the constructing the remote sensing ontology library includes:
designing a remote sensing body, wherein the remote sensing body is used for carrying out structured organization on remote sensing knowledge and remote sensing data;
giving semantic association to the remote sensing body by combining the characteristic attribute and the association attribute in the interpretation information;
the characteristic attributes comprise space-time characteristic attributes of the remote sensing images, and the association attributes comprise space-time association.
In an embodiment of the present invention, the constructing the remote sensing ontology library further includes:
the remote sensing body adopts a coarse-fine structure in design, and comprises an upper layer concept and a lower layer concept;
and combining the background knowledge, and performing concept correction and completion upwards from the lower-layer concept to obtain a perfect remote sensing ontology base.
In a specific embodiment of the present invention, the constructing the remote sensing knowledge-graph comprises:
carrying out entity identification on the remote sensing ontology in the remote sensing ontology library according to the object category in the interpretation information, and carrying out entity classification by combining with the characteristic attribute;
completing relation extraction according to the associated attributes in the attribute knowledge and the background knowledge to obtain the relation between the remote sensing entities;
and carrying out graph network linkage on the triples to obtain the remote sensing knowledge graph.
Compared with the prior art, the invention has the beneficial effects that at least:
(1) The remote sensing image recommendation method based on the semantic information is used for interpreting the remote sensing image, the interpretation information is fully utilized to construct the remote sensing knowledge map, the obtained remote sensing knowledge map contains rich semantic information, and efficient and accurate image recommendation can be achieved.
(2) According to the invention, through carrying out semantic analysis on the query information of the user, carrying out remote sensing image recommendation on the analysis result, and comparing the similarity of the word vector and the triple vector, accurate recommendation can be realized, and the requirements of the user are met.
(3) The invention also utilizes the historical word vectors to assist in completing the remote sensing knowledge graph, enlarges the range of the remote sensing knowledge graph, enables the remote sensing knowledge graph to contain information which is more interesting to the user, can recommend images which accord with the user and the user when image recommendation is carried out, and improves the experience of the user.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.
Fig. 1 is a schematic structural diagram of a remote sensing image overall planning system based on a knowledge graph according to an embodiment of the present invention;
fig. 2 is a schematic structural diagram of a system for overall planning of remote sensing images based on a knowledge-graph according to another embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. It should be noted that, unless otherwise conflicting, the embodiments and features of the embodiments of the present invention may be combined with each other, and the technical solutions formed are all within the scope of the present invention.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the invention. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.
Referring to fig. 1, the present invention provides a remote sensing image overall planning system based on a remote sensing knowledge graph, including:
the remote sensing knowledge acquisition module is configured to acquire a remote sensing data set and automatically interpret the remote sensing data set to obtain interpretation information, wherein the interpretation information comprises: the remote sensing image comprises an object category, related background knowledge and attribute knowledge, wherein the attribute knowledge comprises associated attributes and characteristic attributes.
The remote sensing data set comprises a remote sensing data set and a remote sensing image set, the remote sensing data set is text data for recording the remote sensing data set, and specifically comprises physical and geographic data, geological data, historical data and the like, and the remote sensing data set is interpreted, namely the text data is analyzed and filed to form background knowledge related to the remote sensing image.
Specifically, the remote sensing knowledge acquisition module comprises an interpretation model, and the remote sensing image set is automatically interpreted by the interpretation model, wherein the interpretation model is a deep learning network model.
In an embodiment of the present invention, a process of automatically interpreting the remote sensing image set is as follows:
(1) And preprocessing the remote sensing image set to enable the quality of the remote sensing image set to meet the interpretation requirement, wherein the preprocessing comprises geometric correction, atmospheric correction, image enhancement, principal component analysis and the like.
(2) And constructing a deep learning network model, interpreting the remote sensing image set, and combining the geological data to obtain interpretation information, specifically classifying the geological objects of the remote sensing image set and obtaining the attributes of the geological objects.
(3) And establishing a geography knowledge base according to the geography data and the geography data, performing supplementary confirmation and correction on classification results in the aspect of practice and obeying certain geography rules based on symbolic logical reasoning, and performing classification updating and decision analysis by combining with a geography classification chart obtained by visual interpretation.
Specifically, the attribute of the surface feature object is attribute knowledge contained in the remote sensing image, specifically, the attribute knowledge refers to the correlation attribute between surface features in the remote sensing image and the intrinsic characteristic attribute of the surface features, the attribute knowledge is obtained by learning and extracting the surface feature attribute of the remote sensing image set according to the surface feature data, mining and expressing the surface feature attribute and combining a deep learning network model.
And the knowledge map construction module is configured to construct a remote sensing ontology base according to the interpretation information, identify the remote sensing entity by combining the characteristic attributes, extract the entity-association and entity-background relation of the remote sensing image according to the association attributes and background knowledge in the attribute knowledge to form a plurality of triples, and further construct and obtain the remote sensing knowledge map.
The knowledge graph can be logically divided into a mode layer and a data layer, wherein the data layer mainly comprises a series of facts, and the knowledge is stored by taking the facts as units. If a fact is expressed by a triple of (entity 1, relationship, entity 2), (entity, attribute value), a graph database may be selected as the storage medium. The mode layer is built on the data layer, and a series of fact expressions of the data layer are mainly specified through the ontology library. The ontology is a concept template of the structured knowledge base, and the knowledge base formed by the ontology base has a strong hierarchical structure and a small redundancy degree.
In an embodiment of the present invention, the construction process of the remote sensing ontology library includes:
the remote sensing method comprises the steps of firstly designing a remote sensing body, wherein the remote sensing body is also a remote sensing concept and represents abstract description, the remote sensing body is used for carrying out structural organization on remote sensing knowledge and remote sensing data, and when the remote sensing body is designed, semantic association is given to the remote sensing body by combining characteristic attributes and association attributes in interpretation information.
Preferably, after the key remote sensing concept is determined, concept expansion can be carried out on the basis, and the remote sensing ontology is completed step by step. The remote sensing body is designed in a coarse-to-fine structure, namely, the body on the upper layer is a remote sensing concept with a larger range, the body on the lower layer is a remote sensing concept with a smaller range, concept correction and completion are carried out from the lower layer upwards, in the process, the filed background knowledge can be introduced to assist in concept correction and completion, the background knowledge comprises various recorded data of remote sensing images, and related remote sensing concepts are inquired and extracted from the data so as to perfect the remote sensing body.
In one embodiment of the invention, the process of constructing the remote sensing knowledge graph comprises the following steps:
after obtaining the remote sensing body, carrying out entity labeling, namely entity identification, according to the object categories, and carrying out entity classification by combining characteristic attributes, wherein the characteristic attributes specifically refer to characteristics such as texture characteristics, spectrum characteristics, invariant characteristics and the like of an object represented in a remote sensing image, the characteristics are remote sensing characteristics of the remote sensing image, each category has specific texture characteristics, spectrum characteristics, invariant characteristics and the like after the remote sensing image set is interpreted, the classification of the remote sensing entities is assisted after the characteristic attributes are introduced, and the obtained remote sensing entities all contain respective characteristic attributes.
And then, extracting relationships between the entities, which mainly includes extracting relationships between the entities and the associations and between the entities and the contexts, in this embodiment, extracting the relationships according to the association attributes and the context knowledge in the attribute knowledge, specifically, extracting the entities and the associations according to the association attributes, and extracting the entities and the contexts according to the context knowledge to obtain the relationships between the entities.
Further, the association attribute comprises space-time association, the remote sensing entities are subjected to space-time association relation extraction, the space-time characteristics of each remote sensing entity are extracted, and the associated space-time characteristics are divided to obtain an entity-association-entity relation path.
Further, the background knowledge comprises scenes, ground objects and the like, the relation between the remote sensing entities and the scene characteristics is extracted to form an entity-scene relation, the relation between the remote sensing entities and the characteristics of the ground objects is extracted, and preferably, the detailed extraction can be carried out according to the label classification of the ground objects to form the entity-ground object relation. And further obtaining the relation paths of the entity-scene-entity and the entity-ground object-entity.
And integrating all the relation paths to obtain entity-relation-entity triples, and carrying out graph network linkage on the triples to obtain the remote sensing knowledge graph.
The user information acquisition module is configured to acquire user query information including historical user query information and current user query information, perform semantic analysis on the user query information, extract keywords and associated words which are interested by a user, and map the keywords and the associated words into word vectors, wherein the word vectors include current word vectors and historical word vectors.
Referring to fig. 2, in an embodiment of the present invention, the user information collecting module is divided into two units, namely a current information collecting unit and a historical information collecting unit, the current information collecting unit is configured to collect current user query information, and the historical information collecting unit is configured to collect historical user query information. The two units have the same structure, and only the difference is that the information collected and processed is different. Taking a current information acquisition unit as an example for explanation, the current information acquisition unit includes:
the input part is used for receiving the query information input by the user, and the query information is generally a natural query statement.
And the semantic analysis part comprises a semantic analysis model, wherein the semantic analysis model is input into a natural query sentence, the output of the semantic analysis model is a current word vector, the semantic analysis model adopts a natural language analysis tool, the semantic analysis model is trained by utilizing a corpus, and the corpus is obtained by summarizing and sorting the information according to the remote sensing entity, the remote sensing ontology base, the background knowledge of the remote sensing image and the like. Specifically, the training process of the semantic analysis model is as follows:
firstly, the word segmentation is carried out on the corpus based on a statistical method, the word segmentation is that a sentence is split, the corpus is converted into a dictionary, and the dictionary is provided with a plurality of independent words.
Then, training an initial semantic analysis model by using a dictionary, wherein the semantic analysis model adopts a bidirectional self-attention network and comprises a plurality of self-attention coding layers, each word in the dictionary is coded before the semantic analysis model is trained, and the coding process comprises 3 embedding operations: word embedding, position embedding and segment embedding, wherein three embedded vectors are obtained as the input of a semantic analysis model according to the coding result. Then, executing two training tasks when training the semantic analysis model, wherein the first training task is to mask off the marks of a part of words in the dictionary, the training semantic analysis model predicts the marks of the words, and the training tasks can lead the model to learn the semantics of the distributed context of the masked marks; the second training task is to divide the dictionary into a positive sample and a negative sample, take the word related to the current word A as the positive sample, form the word randomly extracted from the dictionary into the negative sample, the number of the positive sample and the negative sample is the same, train the model to predict all the words related to the current word, the training task is to make the model learn the relationship prediction ability of one word and other related words, and make the model able to perform language reasoning.
After the semantic analysis model is trained, semantic analysis is carried out on the current user query information by using a natural language analysis tool and the trained semantic analysis model to obtain keywords and associated words, and the keywords and the associated words are mapped into current word vectors according to the positions and the upper and lower relations of the keywords and the associated words in sentences. The dimension of the current word vector is n.
The execution process of the historical information acquisition unit is consistent with that of the current information acquisition unit, the execution result is a historical word vector, and the dimension of the historical word vector is also n.
Referring to fig. 2, in an embodiment of the present invention, the system further includes a knowledge map embedding module, configured to perform vector representation learning on the remote sensing knowledge map by combining historical word vectors, to obtain an updated and completed remote sensing knowledge map, perform vectorization representation on the remote sensing knowledge map, convert the remote sensing knowledge map into a plurality of entity vectors and relationship vectors, calculate vector inner products of the entity vectors and the relationship vectors of each triplet, and then compress the vector to obtain a triplet vector.
In an embodiment of the present invention, the knowledge-graph embedding module includes a knowledge-graph embedding model, and the designing of the knowledge-graph embedding model includes: 1) Defining representations of entities and relationships; 2) Defining a scoring function for measuring the rationality of the triples; 3) An embedded representation of learning entities and relationships is trained. The higher the scoring function value, the more plausible, i.e. more likely to be correct, the representative triplet. In training the embedded representation of learning entities and relationships, the optimization goal is to make existing triples in the knowledge-graph score as high as possible over non-occurring triples. Knowledge-graph embedded models can be broadly divided into distance-based models, bilinear models, and neural network models, depending on the defined form of the scoring function.
In this embodiment, a distance-based model is adopted, which is specifically as follows:
the model uses a gaussian distribution to represent an entity and a relationship, the mean of the gaussian distribution represents the center position of the entity or relationship in the semantic space, and the covariance of the gaussian distribution represents the uncertainty of the entity or relationship.
(1) Modeling a gaussian distributed entity:
where h is the head entity in the triplet,
is the mean vector of the head entity and,
is the variance of the distribution of the head entity, I is the identity matrix,
represents a gaussian distribution; similarly, t is the tail entity in the triplet,
is the mean vector of the tail entity,
is the variance of the tail entity distribution.
(2) Describing the relationship between head and tail entities based on a Gaussian mixture model, wherein one relationship corresponds to a plurality of semantics, and each semantic is characterized by a Gaussian distribution:
wherein r is the relationship between the head entity and the tail entity in the triple, pi refers to a mixing factor, i represents the ith semantic,
is the weight of the ith semantic meaning,
is the embedding of the ith semantic.
The model uses similarity-based scoring functions to measure the likelihood that a triple fact holds by matching entities and the underlying semantics of the relationships in the embedded vector space. The scoring function is as follows:
wherein, a rule is set as r ≈ t-h, and the scoring function is specifically obtained by using the square of 2-norm. The probability of a triplet is determined according to a scoring function, which is formulated as follows:
the semantic number of a relationship can be learned automatically from the data by the CRP. CRP refers to the dirichlet process, which is a stochastic process.
The model considers different semantics of each relation r, forms Gaussian distribution for each semantic, distinguishes correct entities and wrong entities through evaluation of a scoring function, and updates the remote sensing knowledge graph.
(3) And performing link prediction on the updated remote sensing knowledge graph by combining the historical word vectors to complete the remote sensing knowledge graph.
After the differentiation of entities is performed, the entities are ranked or predicted to predict a new relationship between two given entities. The method specifically comprises three prediction tasks, namely predicting a head entity of a new triple, predicting a tail entity of the new triple and predicting the relationship of the new triple.
Taking prediction of a head entity in a new triplet as an example, given a tail entity t, a history word vector is introduced, which is a vector set, and in order to predict the head entity, all entities can be used as candidate head entities h * And the historical word vectors are also used as candidate head entities to participate in prediction, as the entities in the remote sensing knowledge map comprise space and time characteristics, wherein the space characteristics are embodied by space coordinates, in order to enable the historical word vectors to participate in prediction, the dimensions of the historical word vectors are converted to be the same as those of the entities, and then the scores of the candidate head entities are calculated. The score of each candidate head entity is calculated according to the formula (4), the candidate head entities are sorted according to the sequence from large to small according to the score, a sorted list is obtained, the candidate head entities with the scores reaching the preset requirements are used as the head entities of the new triples, and the new triples are formed.
The prediction mode of the tail entity and the relation is the same as the above. And after a plurality of new triples are formed, completing the remote sensing knowledge graph.
When updating and completing the remote sensing knowledge graph, both the remote sensing entity and the relation can be represented by vectors, so that the remote sensing knowledge graph can be mapped to be represented in a vector form, including entity vectors and relation vectors, the vector inner product is obtained by the entity vectors and the relation vectors in the triples on the basis of each triplet, the result of the inner product is compressed, the obtained vector is still a vector, and the vector is used as a triplet vector.
And the personalized recommendation module is configured to calculate the similarity of the current word vector and the triple vectors, sort the triple vectors from high to low according to the calculation result of the similarity, list the first k triple vectors in a recommendation candidate table, screen the remote sensing image according to the recommendation candidate table, and output the remote sensing image as a recommendation result.
In an embodiment of the present invention, the dimension of the current word vector is first converted into the same as the triplet vector, let the dimension be n, and the similarity between the current word vector and the triplet vector is calculated, which may specifically use an n-dimensional euclidean distance:
wherein the content of the first and second substances,
is the current word vector (x) 11 ,x 12 ,...,x 1n ) And k is the current calculation dimension,
is a triplet vector (x) 21 ,x 22 ,...,x 2n )。
The smaller the euclidean distance, the higher the similarity between the two vectors.
The method comprises the steps of calculating the similarity of a plurality of current word vectors, respectively carrying out similarity calculation on all triple vectors and each current word vector, sorting the similarity calculation results of each current word vector from high to low according to the similarity, sorting the Euclidean distance values from small to large if the Euclidean distance is calculated according to the embodiment, selecting the first k results of each similarity sorting result, and listing the first k results into a candidate list, namely, taking the current word vector as a head row and taking the first k similarity results as corresponding columns to form the candidate list.
And all the triple vectors corresponding to all the similarity results in the candidate list are linked to the corresponding similarity results, the triple vectors are used as screening bases, remote sensing images containing the triples are screened in the system, and the remote sensing images are output as recommendation results.
The above description is only for the specific embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (10)
1. A remote sensing image overall planning system based on a remote sensing knowledge graph is characterized by comprising:
the remote sensing knowledge acquisition module is configured to acquire a remote sensing image set and automatically interpret the remote sensing data set to obtain interpretation information, wherein the interpretation information comprises: the method comprises the steps that object categories contained in remote sensing images and related background knowledge and attribute knowledge are contained, wherein the attribute knowledge comprises associated attributes and characteristic attributes;
the knowledge map building module is configured to build a remote sensing ontology base according to the interpretation information, identify a remote sensing entity by combining characteristic attributes, extract the entity-association and entity-background relation of the remote sensing image according to the association attributes and background knowledge in the attribute knowledge to form a plurality of triples, and further build the remote sensing knowledge map;
the system comprises a user information acquisition module, a word vector generation module and a word vector generation module, wherein the user information acquisition module is configured to acquire user query information, perform semantic analysis on the user query information, extract key words and associated words which are interested by a user and map the key words and the associated words into word vectors;
the knowledge map storage module is configured to perform vectorization representation on a remote sensing knowledge map, the remote sensing knowledge map is converted into a plurality of entity vectors and relationship vectors, and then the triple vectors are obtained through calculation;
and the personalized recommendation module is configured to calculate the similarity of the current word vector and the triple vectors, sort the triple vectors from high to low according to the calculation result of the similarity, list the first k triple vectors in a recommendation candidate table, screen the recommendation candidate table to obtain recommended remote sensing images, and output the recommended remote sensing images as recommendation results.
2. The system of claim 1, wherein the remote sensing data set comprises a remote sensing data set and a remote sensing image set, the remote sensing data set is text data for recording the remote sensing data set, and the remote sensing data set comprises phenology data, geoscience data and historical data, and the remote sensing data set is interpreted, i.e. the text data is analyzed and filed to form background knowledge related to the remote sensing image.
3. The system of claim 2, wherein automatically interpreting the set of remote sensing images comprises:
preprocessing the remote sensing image set to enable the quality of the remote sensing image set to meet the requirement of interpretation;
and constructing a deep learning network model, interpreting the remote sensing image set, and combining the background knowledge to obtain interpretation information.
4. The system of claim 1, wherein the user information collection module comprises:
the current user information acquisition unit is configured to acquire current user query information and perform semantic analysis on the current user query information to obtain a current word vector;
the historical user information acquisition unit is configured to acquire historical user query information and perform semantic analysis on the historical user query information to obtain a historical word vector;
the current user information acquisition unit and the historical user information acquisition unit respectively comprise semantic analysis models, and the semantic analysis models adopt natural language analysis tools.
5. The system of claim 4, further comprising:
the knowledge map embedding module is configured to carry out vector representation learning on the remote sensing knowledge map by combining historical word vectors to obtain an updated and completed remote sensing knowledge map;
the knowledge graph embedding module is positioned between the user information acquisition module and the knowledge graph storage module.
6. The system of claim 5, wherein the knowledge-graph embedding module comprises a knowledge-graph embedding model, wherein the knowledge-graph embedding model is a distance-based model used for learning embedded representations of remote sensing entities and relationships to update the remote sensing knowledge graph.
7. The system of claim 5, wherein the knowledge graph embedding module is further configured to rank the remote sensing entities in the updated remote sensing knowledge graph and add historical word vectors to the prediction process to predict new triples to complement the remote sensing knowledge graph.
8. The system of claim 1, wherein constructing the remote sensing ontology library comprises:
designing a remote sensing body, wherein the remote sensing body is used for carrying out structured organization on remote sensing knowledge and remote sensing data;
giving semantic association to the remote sensing body by combining the characteristic attribute and the association attribute in the interpretation information;
the characteristic attributes comprise space-time characteristic attributes of the remote sensing images, and the association attributes comprise space-time association.
9. The system of claim 8, wherein constructing the remote ontology library further comprises:
the remote sensing body adopts a coarse-fine structure in design, and comprises an upper layer concept and a lower layer concept;
and combining the background knowledge, and performing concept correction and completion upwards from the lower-layer concept to obtain a perfect remote sensing ontology base.
10. The system of claim 1, wherein constructing the remote sensing knowledge-graph comprises:
carrying out entity identification on the remote sensing ontology in the remote sensing ontology library according to the object category in the interpretation information, and carrying out entity classification by combining with the characteristic attribute;
completing relation extraction according to the associated attributes in the attribute knowledge and the background knowledge to obtain the relation between the remote sensing entities;
and carrying out graph network linkage on the triples to obtain the remote sensing knowledge graph.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211029192.XA CN115269899A (en) | 2022-08-26 | 2022-08-26 | Remote sensing image overall planning system based on remote sensing knowledge map |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211029192.XA CN115269899A (en) | 2022-08-26 | 2022-08-26 | Remote sensing image overall planning system based on remote sensing knowledge map |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115269899A true CN115269899A (en) | 2022-11-01 |
Family
ID=83755908
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211029192.XA Pending CN115269899A (en) | 2022-08-26 | 2022-08-26 | Remote sensing image overall planning system based on remote sensing knowledge map |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115269899A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115858840A (en) * | 2023-02-28 | 2023-03-28 | 北京数慧时空信息技术有限公司 | Scene-based remote sensing image mosaic method |
| CN116629356A (en) * | 2023-05-09 | 2023-08-22 | 华中师范大学 | Encoder and Gaussian mixture model-based small-sample knowledge graph completion method |
| CN116740578A (en) * | 2023-08-14 | 2023-09-12 | 北京数慧时空信息技术有限公司 | Remote sensing image recommendation method based on user selection |
-
2022
- 2022-08-26 CN CN202211029192.XA patent/CN115269899A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115858840A (en) * | 2023-02-28 | 2023-03-28 | 北京数慧时空信息技术有限公司 | Scene-based remote sensing image mosaic method |
| CN116629356A (en) * | 2023-05-09 | 2023-08-22 | 华中师范大学 | Encoder and Gaussian mixture model-based small-sample knowledge graph completion method |
| CN116629356B (en) * | 2023-05-09 | 2024-01-26 | 华中师范大学 | Encoder and Gaussian mixture model-based small-sample knowledge graph completion method |
| CN116740578A (en) * | 2023-08-14 | 2023-09-12 | 北京数慧时空信息技术有限公司 | Remote sensing image recommendation method based on user selection |
| CN116740578B (en) * | 2023-08-14 | 2023-10-27 | 北京数慧时空信息技术有限公司 | Remote sensing image recommendation method based on user selection |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111753060B (en) | Information retrieval method, apparatus, device and computer readable storage medium | |
| CN109271505B (en) | Question-answering system implementation method based on question-answer pairs | |
| CN110222160B (en) | Intelligent semantic document recommendation method and device and computer readable storage medium | |
| CN115269899A (en) | Remote sensing image overall planning system based on remote sensing knowledge map | |
| CN108595636A (en) | The image search method of cartographical sketching based on depth cross-module state correlation study | |
| CN111026842A (en) | Natural language processing method, natural language processing device and intelligent question-answering system | |
| CN111159485B (en) | Tail entity linking method, device, server and storage medium | |
| CN106708929B (en) | Video program searching method and device | |
| CN111858940B (en) | Multi-head attention-based legal case similarity calculation method and system | |
| CN111324765A (en) | Fine-grained sketch image retrieval method based on depth cascade cross-modal correlation | |
| CN115934990B (en) | Remote sensing image recommendation method based on content understanding | |
| CN114332680A (en) | Image processing method, video searching method, image processing device, video searching device, computer equipment and storage medium | |
| CN109271486A (en) | A kind of similitude reservation cross-module state Hash search method | |
| CN114996488A (en) | Skynet big data decision-level fusion method | |
| CN110956044A (en) | Attention mechanism-based case input recognition and classification method for judicial scenes | |
| CN114048354B (en) | Test question retrieval method, device and medium based on multi-element characterization and metric learning | |
| CN114461890A (en) | Hierarchical multi-modal intellectual property search engine method and system | |
| Ghozzi et al. | Interval Type-2 beta fuzzy near sets approach to content-based image retrieval | |
| CN106570196B (en) | Video program searching method and device | |
| CN115687760A (en) | User learning interest label prediction method based on graph neural network | |
| CN113076476B (en) | User portrait construction method of microblog heterogeneous information | |
| CN111104492B (en) | Civil aviation field automatic question and answer method based on layering Attention mechanism | |
| CN117648984A (en) | Intelligent question-answering method and system based on domain knowledge graph | |
| CN117151222A (en) | Domain knowledge guided emergency case entity attribute and relation extraction method thereof, electronic equipment and storage medium | |
| CN115797795A (en) | Remote sensing image question-answering type retrieval system and method based on reinforcement learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |