A remote sensing image retrieval method and device based on knowledge graph
Through the remote sensing image retrieval method based on knowledge graphs, the neural network is used to optimize the graph edges, and the problem of slow retrieval of massive remote sensing images is solved, and efficient image retrieval is achieved, which is suitable for remote sensing image management and storage of massive data.
Patent Information
- Application Number
- CN202011572887.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2020-12-24
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2040-12-24
AI Technical Summary
When processing massive remote sensing images, the prior art cannot improve the search speed while ensuring the search accuracy, resulting in a long time spent in the search process.
The remote sensing image retrieval method based on knowledge graph is adopted to construct knowledge graphs, extract node features of remote sensing images, and use neural networks to optimize the graph edges to achieve rapid retrieval.
While maintaining the search accuracy, the time consumption during the search process is reduced, the efficiency of remote sensing image retrieval is improved, and these images can be better stored, managed and utilized.
Smart Images

Figure CN112579813B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of remote sensing information technology applications, and in particular to a remote sensing image retrieval method and device based on a knowledge graph. Background Art
[0002] With the continuous development of the aerospace field, the quality and quantity of remote sensing images have also been rapidly improved. Therefore, how to effectively store, manage and use these remote sensing images has become an important link in the aerospace field. At present, the commonly used image retrieval methods include: (1) texture-based image retrieval methods, which use the texture information of the image to search in the database; (2) content-based image retrieval methods, which extract the color, layout and other information of the image to form features as the retrieval target; (3) deep learning-based image retrieval methods, which use the deeper features of the image extracted by neural networks.
[0003] Although the above three methods can improve the image retrieval accuracy, they cannot effectively improve the image retrieval speed while ensuring the accuracy when processing massive remote sensing images.
[0004] Therefore, it is necessary to develop a new remote sensing image retrieval method to overcome the shortcomings of the existing technology. Summary of the invention
[0005] An object of the present invention is to provide a remote sensing image retrieval method based on a knowledge graph, which can solve the problem in the prior art that the retrieval process consumes a long time when retrieving remote sensing images.
[0006] To achieve the above-mentioned purpose, the present invention provides a remote sensing image retrieval method based on a knowledge graph, comprising the following steps: constructing a knowledge graph for remote sensing image retrieval, wherein the knowledge graph stores remote sensing image templates corresponding to node features; obtaining a remote sensing image to be analyzed and performing node feature extraction on the remote sensing image; and using the knowledge graph to perform retrieval according to the node features of the remote sensing image to be analyzed to obtain a matching remote sensing image template.
[0007] In a further embodiment, the step of constructing a knowledge graph for remote sensing image retrieval further includes: collecting all remote sensing images within a preset range and at a preset spatial resolution; classifying the remote sensing images according to their spatial attribute information and content, and dividing them into multiple image blocks of the same size; extracting node features of the segmented remote sensing images and performing dimensionality reduction compression, storing the dimensionality reduced and compressed remote sensing images and their spatial attribute information as feature information of remote sensing images, and forming a remote sensing image database; establishing nodes of each layer from the top layer to the bottom layer according to the feature information in the remote sensing image database, and the nodes of the lower layer have a subordinate relationship with the adjacent nodes of the upper layer; adding edges between the nodes of the bottom layer to construct the knowledge graph; and optimizing the edges of the knowledge graph through a neural network model.
[0008] In a further embodiment, the spatial attribute information includes latitude and longitude information and geographic location information of the remote sensing image.
[0009] In a further embodiment, the node feature is a rotationally invariant feature of the remote sensing image.
[0010] In a further embodiment, the step of adding edges between the nodes in the bottom layer further includes: calculating the similarity between the feature strings corresponding to any two nodes in the bottom layer, and adding an edge between two nodes whose similarity exceeds a preset value.
[0011] In a further embodiment, the step of establishing each node of each layer from the top layer to the bottom layer according to the characteristic information in the remote sensing image database further includes: establishing each node of each layer in three layers according to the characteristic information in the remote sensing image database, wherein the nodes of the first layer are category attributes, the nodes of the second layer are basic area attributes, and the nodes of the third layer are remote sensing image blocks of the area indicated by the nodes of the second layer; wherein the nodes of the second layer are subordinate objects of the nodes of the first layer, and the third layer serves as the bottom layer.
[0012] In a further embodiment, the step of optimizing the edges of the knowledge graph through a neural network model further includes: abstracting the underlying nodes of the primary knowledge graph into an adjacency matrix of the graph and an attribute matrix of the nodes; inputting the adjacency matrix of the graph and the attribute matrix of the nodes into the neural network model, and updating the node features of the underlying nodes and updating the edges of the knowledge graph through self-encoding and decoding.
[0013] In a further embodiment, the step of using the knowledge graph to search according to the node features of the remote sensing image to be analyzed to obtain a matching remote sensing image template further includes: using the knowledge graph to match the node features of the remote sensing image to be analyzed with the node features of the knowledge graph, and searching along the edge of the knowledge graph according to the similarity between the features until the node feature with the highest similarity to the node feature of the remote sensing image to be analyzed is retrieved as the target node feature; obtaining the remote sensing image template corresponding to the target node feature in the knowledge graph as the remote sensing image template that matches the remote sensing image to be analyzed.
[0014] In a further embodiment, the step of using the knowledge graph to match the node features of the remote sensing image to be analyzed with the node features of the knowledge graph further includes: from the top layer to the bottom layer, the node features of the remote sensing image to be analyzed are matched with the node features of the corresponding layer of the knowledge graph layer by layer to determine the category, the area and similar remote sensing image blocks of the remote sensing image to be analyzed.
[0015] To achieve the above-mentioned purpose, the present invention also provides a remote sensing image retrieval device based on a knowledge graph, comprising: a construction unit, which constructs a knowledge graph for remote sensing image retrieval, wherein the knowledge graph stores remote sensing image templates corresponding to node features; an acquisition unit, which is used to acquire a remote sensing image to be analyzed and extract node features thereof; and a retrieval unit, which uses the knowledge graph to perform retrieval according to the node features of the remote sensing image to be analyzed to obtain a matching remote sensing image template.
[0016] Compared with the prior art, the beneficial effect of the present invention is that: by establishing a knowledge graph and retrieving remote sensing images through the knowledge graph, the knowledge graph can better utilize the internal connections and hidden feature information between images, while maintaining the accuracy of image retrieval, reducing the time consumed in the retrieval process, and improving the efficiency of remote sensing image retrieval in large amounts of data, thereby better storing, managing and utilizing these remote sensing images. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] The technical solutions and other beneficial effects of the present invention will be made apparent by describing in detail the specific embodiments of the present invention in conjunction with the accompanying drawings.
[0018] Figure 1 A flow chart of a remote sensing image retrieval method based on a knowledge graph provided by an embodiment of the present invention;
[0019] Figure 2 A schematic diagram of the structure of a knowledge graph provided by an embodiment of the present invention;
[0020] Figure 3A schematic diagram of the structure of an autoencoding network provided by an embodiment of the present invention;
[0021] Figure 4 A schematic diagram of the structure of a remote sensing image retrieval device based on a knowledge graph provided by an embodiment of the present invention;
[0022] Figure 5 A schematic diagram of the structure of a building unit provided in one embodiment of the present invention;
[0023] Figure 6 A schematic diagram of the structure of an optimization subunit provided in one embodiment of the present invention;
[0024] Figure 7 A schematic diagram of the structure of a retrieval unit provided in one embodiment of the present invention. DETAILED DESCRIPTION
[0025] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.
[0026] Knowledge Graph is a semantic network that can make full use of the semantic relationship between entities. It connects nodes with certain common attributes through different types of edges to form a complex network that can be used to optimize the retrieval process. Knowledge graph is called knowledge domain visualization or knowledge domain mapping map in the library and information industry. It is a series of various graphics that show the development process and structural relationship of knowledge. It uses visualization technology to describe knowledge resources and their carriers, and mines, analyzes, constructs, draws and displays knowledge and their mutual connections. Taking the social relationship knowledge graph as an example, the nodes in the graph represent people, and the edges between the nodes represent the relationship between people; taking the financial knowledge graph as an example, the nodes in the graph represent accounts, and the edges between the nodes represent the transfer relationship between accounts. Compared with traditional image retrieval methods, knowledge graphs can better utilize the internal associations and hidden feature information between images. Therefore, the present invention adopts the knowledge graph method to apply to the retrieval of remote sensing images. The technical solution of the present invention is further explained below in conjunction with the accompanying drawings and embodiments.
[0027] like Figure 1 As shown, it is a flow chart of a remote sensing image retrieval method based on a knowledge graph provided by an embodiment of the present invention. This embodiment provides a remote sensing image retrieval method based on a knowledge graph, including steps 1 to 3.
[0028] Step 1: Construct a knowledge graph for remote sensing image retrieval, in which remote sensing image templates corresponding to node features are stored.
[0029] In a further embodiment, step 1 also includes steps 11 to 16.
[0030] Step 11: Collect all remote sensing images within a preset range and at a preset spatial resolution.
[0031] Specifically, all remote sensing images within a preset range and at a preset spatial resolution are collected through the global geographic information system.
[0032] Step 12: Classify the remote sensing image according to the spatial attribute information and content of the remote sensing image, and divide it into multiple image blocks of the same size.
[0033] Specifically, the spatial attribute information includes the latitude and longitude information and geographical location information of the remote sensing image.
[0034] Specifically, the image blocks are divided into multiple image blocks of the same size, each of which is image data of 512 pixels*512 pixels in size. For the convenience of processing, they are numbered according to the row and column numbers.
[0035] Step 13: extract node features of the segmented remote sensing image and perform dimensionality reduction compression, store the dimensionality reduced and compressed remote sensing image and its spatial attribute information as feature information of the remote sensing image, and form a remote sensing image database.
[0036] Since the remote sensing images have a certain rotation, the remote sensing images taken at different angles and remote sensing images of different scales are normalized to extract features with different properties such as rotation invariance and scale invariance. Taking SIFT features as an example, the extracted features are reduced in dimension by BOVW technology.
[0037] SIFT feature is a scale-invariant feature transformation. SIFT feature remains invariant to rotation, scale scaling, brightness change, etc., and is a very stable local feature.
[0038] BOVW technology is a bag of visual words model, which is a commonly used document representation method in the field of information retrieval. In information retrieval, the BOW model assumes that for a document, its word order, grammar, syntax and other elements are ignored, and it is only regarded as a collection of several words. The appearance of each word in the document is independent and does not depend on whether other words appear.
[0039] Step 14: Establish nodes of each layer from the top layer to the bottom layer according to the feature information in the remote sensing image database, and the nodes of the lower layer have a subordinate relationship with the adjacent nodes of the upper layer.
[0040] The knowledge graph includes multiple layers with affiliation. Usually, the first layer is the concept terms, the second layer is the basic attributes, and the third and subsequent layers contain gradually more sub-nodes.
[0041] like Figure 2 As shown, it is a schematic diagram of the structure of the knowledge graph provided by the present invention. The nodes of each layer are established in three layers according to the feature information in the remote sensing image database, wherein the nodes 100 of the first layer are category attributes, the nodes 200 of the second layer are basic area attributes, and the nodes 300 of the third layer are remote sensing image blocks of the areas indicated by the nodes 200 of the second layer.
[0042] The nodes of the second layer are subordinate objects of the nodes of the first layer, and the third layer serves as the bottom layer.
[0043] Specifically, the nodes 100 of the first layer include several major categories of geographical attributes such as cities, waters, forests, and mountains. The nodes 200 of the second layer are classic representatives of the categories shown by the nodes 100 of the first layer. For example, the sub-nodes of cities include classic representatives such as Beijing and Shanghai.
[0044] Step 15: Add edges between the nodes at the bottom layer to construct the knowledge graph.
[0045] Specifically, the similarity between the feature strings corresponding to any two nodes of the bottom layer is first calculated, and an edge is added between two nodes whose similarity exceeds a preset value. For example, the Euclidean distance is calculated, and the formula is as follows:
[0046]
[0047] Step 16: Optimize the edges of the knowledge graph through a neural network model.
[0048] In a further embodiment, step 16 also includes steps 161 and 162.
[0049] Step 161: Abstract the bottom-level nodes of the primary knowledge graph into an adjacency matrix of the graph and an attribute matrix of the nodes.
[0050] A∈R N*N Represents the adjacency matrix of the graph, where N is the number of nodes. If there is an edge between two nodes i and j, then A ij =1, otherwise A ij =0.
[0051] X=[x 1 ,x 2 …x n ] represents the attribute matrix of the node, where x i Represents the attribute characteristics of node i.
[0052] Step 162: Input the adjacency matrix of the graph and the attribute matrix of the nodes into the neural network model, update the node features of the underlying nodes and update the edges of the knowledge graph through self-encoding and decoding.
[0053] like Figure 3 As shown in the figure, it is a schematic diagram of the structure of the autoencoding network. The encoding process is as follows:
[0054]
[0055]
[0056]
[0057] Among them, W and are trainable weights, is the output of node i after the kth encoding, is the attention mechanism weight, N i is the set of neighbor nodes of node i, and σ is the activation function, such as the ReLU function.
[0058] The decoding process is as follows:
[0059]
[0060]
[0061]
[0062] The loss function is:
[0063]
[0064] After multiple encoding, decoding, and training parameter weights, a new node feature matrix is finally obtained. As the output of the knowledge graph autoencoder neural network.
[0065] Through the trained node feature matrix H k , calculate the node similarity between any two nodes, such as the Euclidean distance between features. If it is less than a certain threshold, an edge is established between the two nodes, otherwise it is deleted.
[0066] Step 2: Obtain a remote sensing image to be analyzed and extract node features from it.
[0067] Specifically, node features include category attributes and basic area attributes.
[0068] Step 3: Use the knowledge graph to search according to the node features of the remote sensing image to be analyzed to obtain a matching remote sensing image template.
[0069] In a further embodiment, step 3 also includes step 31-step 32.
[0070] Step 31: Using the knowledge graph, match the node features of the remote sensing image to be analyzed with the node features of the knowledge graph, and search along the edge of the knowledge graph according to the similarity between the features until the node feature with the highest similarity to the node feature of the remote sensing image to be analyzed is retrieved as the target node feature.
[0071] Specifically, from the top layer to the bottom layer, the node features of the remote sensing image to be analyzed are matched with the node features of the corresponding layer of the knowledge graph layer by layer to determine the category, the area and similar remote sensing image blocks of the remote sensing image to be analyzed.
[0072] Step 32: Obtain a remote sensing image template corresponding to the target node feature in the knowledge graph as a remote sensing image template that matches the remote sensing image to be analyzed.
[0073] Based on the same inventive concept, the embodiment of the present invention also provides a remote sensing image retrieval device based on a knowledge graph, such as Figure 4 FIG. 1 is a schematic diagram of a remote sensing image retrieval device based on a knowledge graph according to an embodiment of the present invention. The retrieval device includes a construction unit 1, an acquisition unit 2, and a retrieval unit 3.
[0074] The construction unit 1 constructs a knowledge graph for remote sensing image retrieval, and the knowledge graph stores remote sensing image templates corresponding to node features.
[0075] like Figure 5 FIG. 1 is a schematic diagram of a structure of a construction unit 1 provided in an embodiment of the present invention. In a further embodiment, the construction unit 1 includes a collection subunit 11, a segmentation subunit 12, a dimensionality reduction and compression subunit 13, a node establishment subunit 14, an edge addition subunit 15, and an optimization subunit 16.
[0076] The acquisition subunit 11 is used to acquire all remote sensing images within a preset range and with a preset spatial resolution.
[0077] Specifically, all remote sensing images within a preset range and at a preset spatial resolution are collected through the global geographic information system.
[0078] Segmentation subunit 12: classifies the remote sensing image according to the spatial attribute information and content of the remote sensing image, and divides the image into a plurality of image blocks of the same size.
[0079] Specifically, the spatial attribute information includes the latitude and longitude information and geographical location information of the remote sensing image.
[0080] Specifically, the image blocks are divided into multiple image blocks of the same size, each of which is image data of 512 pixels*512 pixels in size. For the convenience of processing, they are numbered according to the row and column numbers.
[0081] Dimensionality reduction and compression subunit 13: used for extracting node features of the segmented remote sensing image and performing dimension reduction and compression, storing the remote sensing image and its spatial attribute information after dimension reduction and compression as feature information of the remote sensing image, and forming a remote sensing image database.
[0082] Since the remote sensing images have a certain rotation, the remote sensing images taken at different angles and remote sensing images of different scales are normalized to extract features with different properties such as rotation invariance and scale invariance. Taking SIFT features as an example, the extracted features are reduced in dimension by BOVW technology.
[0083] SIFT feature is a scale-invariant feature transformation. SIFT feature remains invariant to rotation, scale scaling, brightness change, etc., and is a very stable local feature.
[0084] BOVW technology is a bag of visual words model, which is a commonly used document representation method in the field of information retrieval. In information retrieval, the BOW model assumes that for a document, its word order, grammar, syntax and other elements are ignored, and it is only regarded as a collection of several words. The appearance of each word in the document is independent and does not depend on whether other words appear.
[0085] Establishing node subunit 14: establishing nodes of each layer from the top layer to the bottom layer according to the characteristic information in the remote sensing image database, and there is a subordinate relationship between the nodes of the lower layer and the adjacent nodes of the upper layer.
[0086] The knowledge graph includes multiple layers with affiliation. Usually, the first layer is the concept terms, the second layer is the basic attributes, and the third and subsequent layers contain gradually more sub-nodes.
[0087] Specifically, the nodes of each layer are established in three layers according to the feature information in the remote sensing image database, wherein the nodes of the first layer are category attributes, the nodes of the second layer are basic area attributes, and the nodes of the third layer are remote sensing image blocks of the areas indicated by the nodes of the second layer.
[0088] The nodes of the second layer are subordinate objects of the nodes of the first layer, and the third layer serves as the bottom layer.
[0089] Specifically, the nodes of the first layer include several major categories of geographical attributes, such as cities, waters, forests, and mountains. The nodes of the second layer are classic representatives of the categories shown by the nodes of the first layer. For example, the sub-nodes of cities include classic representatives such as Beijing and Shanghai.
[0090] Add edge subunit 15: used to add edges between the nodes in the bottom layer to construct the knowledge graph.
[0091] Specifically, the similarity between the feature strings corresponding to any two nodes of the bottom layer is first calculated, and an edge is added between two nodes whose similarity exceeds a preset value. For example, the Euclidean distance is calculated, and the formula is as follows:
[0092]
[0093] Optimization subunit 16: Optimizing the edges of the knowledge graph through a neural network model.
[0094] like Figure 6 FIG. 1 is a schematic diagram of the structure of an optimization subunit 16 provided in an embodiment of the present invention. The optimization subunit 16 includes an abstraction module 161 and an update module 162 .
[0095] Abstraction module 161: abstracting the bottom-level nodes of the primary knowledge graph into an adjacency matrix of the graph and an attribute matrix of the nodes.
[0096] A∈R N*N Represents the adjacency matrix of the graph, where N is the number of nodes. If there is an edge between two nodes i and j, then A ij =1, otherwise A ij =0.
[0097] X=[x 1 ,x 2 …x n ] represents the attribute matrix of the node, where x i Represents the attribute characteristics of node i.
[0098] Update module 162: Input the adjacency matrix of the graph and the attribute matrix of the node into the neural network model, update the node features of the underlying nodes and update the edges of the knowledge graph through self-encoding and decoding.
[0099] The encoding process is as follows:
[0100]
[0101]
[0102]
[0103] Among them, W and are trainable weights, is the output of node i after the kth encoding, is the attention mechanism weight, N i is the set of neighbor nodes of node i, and σ is the activation function, such as the ReLU function.
[0104] The decoding process is as follows:
[0105]
[0106]
[0107]
[0108] The loss function is:
[0109]
[0110] After multiple encoding, decoding, and training parameter weights, a new node feature matrix is finally obtained. As the output of the knowledge graph autoencoder neural network.
[0111] Through the trained node feature matrix H k , calculate the node similarity between any two nodes, such as the Euclidean distance between features. If it is less than a certain threshold, an edge is established between the two nodes, otherwise it is deleted.
[0112] The acquisition unit 2 is used to acquire a remote sensing image to be analyzed and perform node feature extraction on it.
[0113] Specifically, node features include category attributes and basic area attributes.
[0114] The retrieval unit 3: uses the knowledge graph to perform retrieval according to the node features of the remote sensing image to be analyzed, and obtains a matching remote sensing image template.
[0115] like Figure 7 FIG. 2 is a schematic diagram of the structure of a search unit 3 provided in an embodiment of the present invention. In a further embodiment, the search unit 3 includes a matching subunit 31 and an acquisition subunit 32 .
[0116] Matching subunit 31: using the knowledge graph, matching the node features of the remote sensing image to be analyzed with the node features of the knowledge graph, searching along the edge of the knowledge graph according to the similarity between the features until the node features with the highest similarity to the node features of the remote sensing image to be analyzed are retrieved as the target node features.
[0117] Specifically, from the top layer to the bottom layer, the node features of the remote sensing image to be analyzed are matched with the node features of the corresponding layer of the knowledge graph layer by layer to determine the category, the area and similar remote sensing image blocks of the remote sensing image to be analyzed.
[0118] Acquisition subunit 32: Acquisition of a remote sensing image template corresponding to the target node feature in the knowledge graph as a remote sensing image template that matches the remote sensing image to be analyzed.
[0119] This embodiment provides a remote sensing image retrieval method and device based on a knowledge graph. By establishing a knowledge graph and retrieving remote sensing images through the knowledge graph, the knowledge graph can better utilize the internal connections and hidden feature information between images, while maintaining the accuracy of image retrieval, reducing the time consumed in the retrieval process, and improving the efficiency of remote sensing image retrieval in large amounts of data, thereby better storing, managing, and utilizing these remote sensing images.
[0120] The above is a detailed introduction to a remote sensing image retrieval method and device based on a knowledge graph provided in an embodiment of the present invention. This article uses specific examples to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the technical scheme and core ideas of the present invention. Ordinary technicians in this field should understand that they can still modify the technical schemes recorded in the aforementioned embodiments, or replace some of the technical features therein with equivalents; and these modifications or replacements do not make the essence of the corresponding technical scheme deviate from the scope of the technical schemes of the embodiments of the present invention.
Claims
1. A remote sensing image retrieval method based on knowledge graph, characterized in that: The following steps are involved: Constructing a knowledge graph for remote sensing image retrieval, wherein the knowledge graph stores remote sensing image templates corresponding to node features; Acquire a remote sensing image to be analyzed and extract node features from it; Using the knowledge graph to search according to the node features of the remote sensing image to be analyzed, to obtain a matching remote sensing image template; The step of constructing a knowledge graph for remote sensing image retrieval further includes: Collect all remote sensing images within a preset range and at a preset spatial resolution; Classifying the remote sensing image according to the spatial attribute information and content of the remote sensing image, and dividing the remote sensing image into a plurality of image blocks of the same size; Extracting node features of the segmented remote sensing image and performing dimensionality reduction compression, storing the dimensionality reduced and compressed remote sensing image and its spatial attribute information as feature information of the remote sensing image to form a remote sensing image database, wherein the spatial attribute information includes latitude and longitude information and geographic location information of the remote sensing image; The nodes of each layer are hierarchically established from the top layer to the bottom layer according to the characteristic information in the remote sensing image database, and the nodes of the lower layer have a subordinate relationship with the nodes of the adjacent upper layer; the step of hierarchically establishing the nodes of each layer from the top layer to the bottom layer according to the characteristic information in the remote sensing image database further comprises: the nodes of each layer are established in three layers according to the characteristic information in the remote sensing image database, wherein the nodes of the first layer are category attributes, the nodes of the second layer are basic area attributes, and the nodes of the third layer are remote sensing image blocks of the areas indicated by the nodes of the second layer; wherein the nodes of the second layer are subordinate objects of the nodes of the first layer, and the third layer serves as the bottom layer; Add edges between the nodes of the bottom layer to construct the knowledge graph, calculate the similarity between the feature strings corresponding to any two nodes of the bottom layer, and add edges between two nodes whose similarity exceeds a preset value; The edges of the knowledge graph are optimized through a neural network model.
2. The method according to claim 1, characterized in that The node feature is a feature of the remote sensing image that is rotationally invariant.
3. The method according to claim 1, characterized in that The step of adding edges between the nodes of the bottom layer further comprises: The similarity between the feature strings corresponding to any two nodes of the bottom layer is calculated, and an edge is added between two nodes whose similarity exceeds a preset value.
4. The method according to claim 1, characterized in that: The step of optimizing the edges of the knowledge graph through a neural network model further includes: Abstracting the bottom-level nodes of the knowledge graph into an adjacency matrix of the graph and an attribute matrix of the nodes; The adjacency matrix of the graph and the attribute matrix of the nodes are input into the neural network model, and the node features of the underlying nodes and the edges of the knowledge graph are updated by self-encoding and decoding.
5. The method according to claim 1, characterized in that The step of using the knowledge graph to search according to the node features of the remote sensing image to be analyzed to obtain a matching remote sensing image template further includes: Using the knowledge graph, matching the node features of the remote sensing image to be analyzed with the node features of the knowledge graph, searching along the edge of the knowledge graph according to the similarity between the features until the node feature with the highest similarity to the node feature of the remote sensing image to be analyzed is retrieved as the target node feature; A remote sensing image template corresponding to the target node feature in the knowledge graph is obtained as a remote sensing image template that matches the remote sensing image to be analyzed.
6. The method according to claim 5, characterized in that The step of using the knowledge graph to match the node features of the remote sensing image to be analyzed with the node features of the knowledge graph further includes: From the top layer to the bottom layer, the node features of the remote sensing image to be analyzed are matched with the node features of the corresponding layer of the knowledge graph layer by layer to determine the category, the area and similar remote sensing image blocks of the remote sensing image to be analyzed.
7. A remote sensing image retrieval device based on knowledge graph, characterized in that: include: A construction unit, constructing a knowledge graph for remote sensing image retrieval, wherein the knowledge graph stores remote sensing image templates corresponding to node features; An acquisition unit, used for acquiring a remote sensing image to be analyzed and performing node feature extraction on the image; A retrieval unit, which uses the knowledge graph to perform retrieval according to the node features of the remote sensing image to be analyzed, and obtains a matching remote sensing image template; The step of constructing a knowledge graph for remote sensing image retrieval further includes: Collect all remote sensing images within a preset range and at a preset spatial resolution; Classifying the remote sensing image according to the spatial attribute information and content of the remote sensing image, and dividing the remote sensing image into a plurality of image blocks of the same size; Extracting node features of the segmented remote sensing image and performing dimensionality reduction compression, storing the dimensionality reduced and compressed remote sensing image and its spatial attribute information as feature information of the remote sensing image to form a remote sensing image database, wherein the spatial attribute information includes latitude and longitude information and geographic location information of the remote sensing image; The nodes of each layer are hierarchically established from the top layer to the bottom layer according to the characteristic information in the remote sensing image database, and the nodes of the lower layer have a subordinate relationship with the nodes of the adjacent upper layer; the step of hierarchically establishing the nodes of each layer from the top layer to the bottom layer according to the characteristic information in the remote sensing image database further comprises: the nodes of each layer are established in three layers according to the characteristic information in the remote sensing image database, wherein the nodes of the first layer are category attributes, the nodes of the second layer are basic area attributes, and the nodes of the third layer are remote sensing image blocks of the areas indicated by the nodes of the second layer; wherein the nodes of the second layer are subordinate objects of the nodes of the first layer, and the third layer serves as the bottom layer; Add edges between the nodes of the bottom layer to construct the knowledge graph, calculate the similarity between the feature strings corresponding to any two nodes of the bottom layer, and add edges between two nodes whose similarity exceeds a preset value; The edges of the knowledge graph are optimized through a neural network model.
