Sensor topology network generation method based on semantic recognition
By adopting a sensor topology network generation method based on semantic recognition, the problems of slow topology network construction speed and insufficient effectiveness in UAV swarms are solved. This method achieves fast and accurate topology network generation, reduces resource consumption, and adapts to dynamic UAV scenarios.
Patent Information
- Application Number
- CN202511750732.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-26
- Publication Date
- 2026-02-24
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing technologies cannot meet the speed and effectiveness requirements for topology network construction in drone swarms, especially in the case of drones with high dynamic characteristics, where the topology network establishment speed is slow and the effectiveness is insufficient.
A sensor topology network generation method based on semantic recognition is adopted. This method generates a sensor topology network for UAV swarms by establishing a semantic database, selecting starting sensor nodes, broadcasting traversal messages, filtering effective nodes, constructing node spacing and semantic weight matrices, and performing clustering and merging.
It improves the generation speed, reduces redundant forwarding by 30%-50%, improves processing efficiency by K times, has a response time of ≤100ms, an identification accuracy of ≥98%, a topology network connectivity of ≥95%, and reduces resource consumption by 25%-40%.
Smart Images
Figure CN121568142A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of wireless sensor network technology, and in particular to a method for generating sensor topology networks based on semantic recognition. Background Technology
[0002] Wireless sensors are components embedded in various intelligent electronic devices. These devices can sense information in the surrounding environment, thereby providing important data support for various control and data acquisition tasks, enabling intelligent electronic devices to be put into commercial or industrial applications.
[0003] In the control of drone swarms, it is typically necessary to establish a topology network within the swarm for internal data transmission and sharing. Because the states of drones, including location, mission status, and network status, are highly dynamic, the topology network needs to be established as quickly as possible, and the established topology network must also enable efficient information transmission. Therefore, both the speed and effectiveness of topology network establishment are crucial.
[0004] However, there is currently no technology that can meet the above requirements for constructing topology networks in drone swarms. Summary of the Invention
[0005] This application provides a sensor topology network generation method based on semantic recognition, which addresses the limitations of existing technologies in meeting the speed and efficiency requirements of sensor topology network construction for unmanned aerial vehicle (UAV) swarms.
[0006] This application provides a method for generating sensor topology networks based on semantic recognition, including: A semantic library is established based on the node type, mission status, location area, and data requirements of the UAV. The semantic library stores the semantic vectors of all UAV nodes. Select the starting sensor node; Establish a traversal message, which includes a node traversal order path, semantic enhancement fields, and control fields. The node traversal order path starts from the starting sensor node, and the semantic enhancement fields include the semantic vector of the starting sensor node and the target semantic label. Broadcast traversal messages and filter nodes based on the cosine similarity between the semantic vector of the node receiving the traversal message and the target semantic vector corresponding to the target semantic label to obtain valid nodes; The traversal message is updated according to the information of the valid nodes, and the updated traversal message is forwarded by the valid nodes. Construct a node spacing matrix and a semantic weight matrix, and fuse the node spacing matrix and the semantic weight matrix to obtain a fusion matrix; Clustering of UAV nodes based on the fusion matrix yields multiple node sets; Extract the sub-fusion matrix for each node set, and build the sub-topology based on the sub-fusion matrix; Merge any two sub-topologies based on semantic center similarity to obtain the complete topology; Update the complete topology after traversal stops.
[0007] The sensor topology network generation method based on semantic recognition in this application has the following advantages: 1. Improved generation speed. Semantic filtering reduces redundant forwarding by 30%-50%, and parallel clustering and splitting improves processing efficiency by a factor of K, where K is the number of clusters. This allows dynamic updates to require only local traversal, with a response time of ≤100ms.
[0008] 2. Enhanced effectiveness. The accuracy of semantically related node identification is ≥98%, invalid connected edges are avoided, the connectivity of the topology network is ≥95%, and it is suitable for dynamic scenarios of drone swarms.
[0009] 3. Reduced resource overhead. Semantic filtering reduces message forwarding volume, lightweight semantic computing adapts to the limited computing power of drones, and communication bandwidth usage is reduced by 25%-40%. Attached Figure Description
[0010] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0011] Figure 1 A flowchart of a sensor topology network generation method based on semantic recognition provided in an embodiment of this application. Detailed Implementation
[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.
[0013] Figure 1 A flowchart illustrating a sensor topology network generation method based on semantic recognition provided in this application embodiment. This application embodiment provides a sensor topology network generation method based on semantic recognition, including: S100: A semantic library is established based on the node type, mission status, location area, and data requirements of the UAV. The semantic library stores the semantic vectors of all UAV nodes.
[0014] For example, each drone node is assigned a unique combination of semantic tags, which includes node type tags, task status tags, location area tags, and data requirement tags; The semantic labels in the semantic label combination are converted into numerical features to form the semantic vector; The semantic library is established using the semantic vectors of all drone nodes.
[0015] Specifically, each drone node N i , i =1,2,..., n , n Assign a unique semantic tag combination to the total number of drone nodes in the cluster. L i =( l i1 , l i2 , l i3 , l i4 ),in: l i1 For node type labels, the range of values is... T ={ T 1, T 2, T 3}, T 1 = reconnaissance aircraft T 2 = Communication device, T 3 = Execution machine. l i2 This is a task status label, with a value range of... S ={ S 1, S 2, S 3}, S 1 = Free time S 2 = Busy S 3 = Fault. l i3 This is a location area label, with a value range of... R ={ R 1, R 2,..., R k}, k Divide the cluster work area into a number, such as 3 areas. k =3. l i4For data requirement tags, the range of values is... D ={ D 1, D 2, D 3}, D 1 = Image data, D 2 = Location data, D 3 = Control commands.
[0016] Semantic labels are converted into numerical features, and a quantization mapping table is constructed as shown in Table 1.
[0017] Table 1 Quantization Mapping Table
[0018] Nodes are constructed based on quantized values. i semantic vectors V i ,Right now: V i =[ q ( l i1 ), q ( l i2 ), q ( l i3 ), q ( l i4 )] In the formula, q ( l ik ) represents a node i No. k The quantified value of a semantic tag, such as a reconnaissance aircraft ( T 1) Idle time ( S 1) R 1. Region, required image data ( D 1) The node semantic vector is V i =[1.0,1.0,1.0,1.0].
[0019] Establish a semantic library for cluster nodes Lib ={ V 1, V 2,..., V n It stores the semantic vectors of all nodes and supports real-time querying and updating.
[0020] S110, Select the starting sensor node.
[0021] For example, the drone node with the largest communication coverage in the drone cluster and whose task status is idle is selected as the starting sensor node to ensure that the traversal message can spread quickly.
[0022] S120, establish a traversal message, the traversal message includes a node traversal order path, a semantic enhancement field, and a control field. The node traversal order path starts from the starting sensor node, and the semantic enhancement field includes the semantic vector of the starting sensor node and the target semantic label.
[0023] For example, the traversal message M contains three parts: path information, semantic information, and control information, with the following structure: M ={ P , Sem , Ctrl} in, P The node traversal order path, initially set to { N 0}, in the format [ n 0, n 1,..., n t ], n t For the first t The node ID of the next forwarding; Sem This is a semantic enhancement field, containing the semantic vector of the starting node. V 0. Target semantic tags L target For example, "reconnaissance aircraft in all R1 areas"; Ctrl This is a control field that contains the number of forwards. OD (Initial value) OD max semantic relevance threshold θ (Initial value 0.6).
[0024] S130, broadcast the traversal message, and filter the nodes according to the cosine similarity between the semantic vector of the node receiving the traversal message and the target semantic vector corresponding to the target semantic label to obtain valid nodes.
[0025] For example, methods for filtering nodes include: Extract the target semantic tags from the traversed messages and generate the corresponding target semantic vectors; Calculate the cosine similarity between the semantic vector of the node receiving the traversal message and the target semantic vector; If the cosine similarity of a node is greater than or equal to the semantic relevance threshold, the node is considered a valid node.
[0026] Specifically, when nodeN i Receive traversal message M First, a semantic filtering process is executed to filter invalid messages: Extracting target semantic tags from messages L target Generate target semantic vector V target ; compute nodes N i semantic vectors V i and V target cosine similarity S ( i , target ):
[0027] In the formula, v ik For nodes i Semantic vector number k Dimensional components; v tk For the target semantic vector of the th k Dimensional components; d =4, which represents the number of semantic dimensions; , They are respectively V i and V target The L2 norm.
[0028] like S ( i , target )≥ θ Then the node N i If a node is valid, it performs message updates and forwarding; otherwise, the message is discarded and not traversed, reducing redundant forwarding.
[0029] S140, the traversal message is updated according to the information of the valid node, and the updated traversal message is forwarded by the valid node.
[0030] For example, the method for updating the traversal message includes: adding the ID of the valid node to the node traversal order path of the traversal message; and adding the semantic vector of the valid node to the semantic enhancement field to form a semantic trajectory.
[0031] Specifically, effective nodes N i After updating the message, optimize the forwarding path based on semantic similarity: Path update: update the node N i Add the ID to the traversal path P A new path is obtained. P ′= P ∪{ N i}; Semantic update: In Sem Add nodes to fields N i semantic vectors V i Forming semantic trajectories Sem trace =[ V 0, V i ].
[0032] In the embodiments of this application, after updating the traversal message, the effective node queries the set of neighboring nodes within the communication range, calculates the semantic similarity between the effective node and each neighboring node, arranges the neighboring nodes in descending order of semantic similarity, and selects the multiple neighboring nodes with the highest semantic similarity as downstream forwarding nodes to realize the forwarding of the updated traversal message.
[0033] Specifically, nodes N i Query the set of neighboring nodes within its own communication range. Nei ( i ),calculate N i With each neighbor node N j ∈ Nei ( i semantic similarity S ( i , j ):
[0034] according to S ( i , j Sort neighboring nodes in descending order and select the first one. k indivual( k =3 (can be adjusted according to communication bandwidth) The node with the highest semantic similarity is used as the downstream forwarding node to avoid blind broadcasting.
[0035] Number of reposts OD ′= OD -1, if OD If '≤0, stop forwarding and output the current path. P ′.
[0036] S150, construct a node spacing matrix and a semantic weight matrix, and fuse the node spacing matrix and the semantic weight matrix to obtain a fusion matrix.
[0037] For example, based on the traversal path P ′=[ n 0, n 1,..., n t Define nodes n a and n b ( a < b ) node spacing cod ab The difference in forwarding steps between the two in the path: cod ab = b - a For example, the path [ N 0, N 1, N 3, N In 5], cod 0,3 =3, cod 1,5 =4.
[0038] Build n Semantic weight matrix W =[ w ij ],in w ij Represents a node N i and N j The semantic weight is determined by semantic similarity. S ( i , j Normalization yields:
[0039] In the formula, max k=1..n S ( i , k ) represents a node N i Maximum semantic similarity with all other nodes, ensuring w ij ∈[0,1].
[0040] The interval matrix COD With semantic weight matrix W Fusion, resulting in a fusion matrixM fusion Highlight node pairs with strong semantic connections and short paths:
[0041] In the formula, α For balance coefficient, α ∈[0,1], adjusted according to the scene, for example, taking the value of [0,1] for highly dynamic scenes. α =0.4, focusing on semantic association; static scene takes α =0.6, emphasizing path length; It is the minimum value. =10 -6 ,avoid cod ij When the denominator is 0, the fraction is equal to 0.
[0042] S160, Cluster the UAV nodes based on the fusion matrix to obtain multiple node sets.
[0043] For example, when clustering UAV nodes based on the fusion matrix, the node clustering objective function is to maximize the average fusion value of nodes within a cluster and minimize the average fusion value between clusters, wherein the average fusion value is calculated based on the fusion matrix.
[0044] Specifically, to improve parallel processing efficiency, based on the fusion matrix M fusion Perform node clustering and splitting: Define the node clustering objective function as maximizing the average fusion value of nodes within a cluster and minimizing the average fusion value between clusters:
[0045] In the formula, K The number of clusters, 2≤ K ≤5, based on the total number of nodes n Adjustment; C c For the first c Each cluster is a set of nodes; | C c | for the first c The number of nodes in a cluster.
[0046] The objective function is solved using the K-means clustering algorithm, and the results are obtained. K A set of nodes S 1, S 2,..., S K ,satisfy:
[0047] For each set of nodesS c Parallel execution of sub-topology generation: Extracting sets S c Corresponding sub-fusion matrix Set subtopology connectivity threshold γ =0.5, if Then the node N i and N j Connect within the subtopology; Constructing sub-topology based on connectivity T c =( V c , E c ),in V c = S c , .
[0048] S170, Extract the sub-fusion matrix for each set of nodes, and establish a sub-topology based on the sub-fusion matrix.
[0049] For example, semantic similarity can be used to quickly locate related nodes between sub-topologies, replacing the traditional blind traversal matching: For any two subtopologies T c and Extract their respective semantic center vectors and :
[0050] Calculate the semantic center similarity between two sub-topologies S center ( c , c ′):
[0051] like S center ( c , c ′)≥ θ merge merging threshold θ merge =0.7, then select the node pair with the highest semantic similarity in the two sub-topologies. i , j ), i ∈ C c ,j ∈ As an associated node.
[0052] S180, merge any two of the sub-topologies based on semantic center similarity to obtain the complete topology.
[0053] For example, all sub-topologies can be merged into a complete topology based on associated nodes: Initialize complete topology T = T 1; In sequence T 2,..., T K and T Merge: If the associated node is ( i , j ), then in T Add edges ( i , j ), and integrate T c All nodes and edges; After merging and removing duplicate edges, the final drone swarm sensor topology network is obtained. T =( V , E ),in V ={ N 1,..., N n}, E Let be the set of all connected edges.
[0054] S190, after the traversal stops, update the complete topology.
[0055] For example, the traversal is stopped based on a forwarding count threshold and a semantic coverage threshold. The forwarding count threshold is used to determine whether the number of message forwardings has decreased to 0, and the semantic coverage threshold is used to determine whether the semantic coverage of the traversed nodes has reached the coverage threshold, as calculated in real time. If both the forwarding count threshold and the semantic coverage threshold are met, the traversal is stopped.
[0056] Furthermore, after updating the complete topology, the semantic vectors of the nodes are monitored in real time. If the change in the semantic vectors of the nodes exceeds the change threshold, a local update of the updated complete topology is initiated.
[0057] Node status monitoring: Each node updates its own semantic vector in real time. V i Update if the location changes. l i3 Update when the task status changes. l i2; Semantic change detection: If node N i semantic vector change , change threshold Δ θ If the value is 0.3, then a local traversal is triggered; Local topology update: based on changing nodes N i Starting with the node, generate a partial traversal message, with the target semantic label being its own updated value. Repeat the above steps, only updating the associated sub-topologies, without needing to regenerate globally.
[0058] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.
[0059] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.
Claims
1. A sensor topology network generation method based on semantic recognition, characterized in that, include: A semantic library is established based on the node type, mission status, location area, and data requirements of the UAV. The semantic library stores the semantic vectors of all UAV nodes. Select the starting sensor node; A traversal message is established, which includes a node traversal order path, a semantic enhancement field, and a control field. The node traversal order path starts from the starting sensor node, and the semantic enhancement field includes the semantic vector of the starting sensor node and the target semantic label. Broadcast the traversal message, and filter the nodes according to the cosine similarity between the semantic vector of the node receiving the traversal message and the target semantic vector corresponding to the target semantic label to obtain valid nodes; The traversal message is updated according to the information of the valid nodes, and the updated traversal message is forwarded by the valid nodes. Construct a node spacing matrix and a semantic weight matrix, and fuse the node spacing matrix and the semantic weight matrix to obtain a fusion matrix; Based on the fusion matrix, the UAV nodes are clustered to obtain multiple node sets; Extract the sub-fusion matrix for each set of nodes, and establish a sub-topology based on the sub-fusion matrix; Merge any two sub-topologies based on semantic center similarity to obtain the complete topology; After the traversal stops, update the complete topology.
2. The sensor topology network generation method based on semantic recognition according to claim 1, characterized in that, The methods for establishing the semantic library include: Each drone node is assigned a unique semantic tag combination, which includes node type tag, task status tag, location area tag, and data requirement tag. The semantic labels in the semantic label combination are converted into numerical features to form the semantic vector; The semantic library is established using the semantic vectors of all drone nodes.
3. The sensor topology network generation method based on semantic recognition according to claim 1, characterized in that, The drone node with the largest communication coverage in the drone cluster and whose task status is idle is selected as the starting sensor node.
4. The sensor topology network generation method based on semantic recognition according to claim 1, characterized in that, Methods for filtering nodes include: Extract the target semantic tags from the traversed messages and generate the corresponding target semantic vectors; Calculate the cosine similarity between the semantic vector of the node receiving the traversal message and the target semantic vector; If the cosine similarity of a node is greater than or equal to the semantic relevance threshold, the node is considered a valid node.
5. The sensor topology network generation method based on semantic recognition according to claim 1, characterized in that, The method for updating the traversal message includes: adding the ID of the valid node to the node traversal order path of the traversal message; and adding the semantic vector of the valid node to the semantic enhancement field to form a semantic trajectory.
6. The sensor topology network generation method based on semantic recognition according to claim 1, characterized in that, After updating the traversal message, the effective node queries the set of neighboring nodes within the communication range, calculates the semantic similarity between the effective node and each neighboring node, arranges the neighboring nodes in descending order of semantic similarity, and selects the multiple neighboring nodes with the highest semantic similarity as downstream forwarding nodes to achieve forwarding of the updated traversal message.
7. The sensor topology network generation method based on semantic recognition according to claim 1, characterized in that, When clustering UAV nodes based on the fusion matrix, the node clustering objective function is to maximize the average fusion value of nodes within a cluster and minimize the average fusion value between clusters. The average fusion value is calculated based on the fusion matrix.
8. The sensor topology network generation method based on semantic recognition according to claim 1, characterized in that, The traversal is stopped based on a forwarding count threshold and a semantic coverage threshold. The forwarding count threshold is used to determine whether the number of message forwardings has decreased to 0, and the semantic coverage threshold is used to determine whether the semantic coverage of the traversed nodes has reached the coverage threshold, which is calculated in real time. If both the forwarding count threshold and the semantic coverage threshold are met, the traversal is stopped.
9. The sensor topology network generation method based on semantic recognition according to claim 1, characterized in that, After updating the complete topology, the semantic vectors of the nodes are monitored in real time. If the change in the semantic vectors of the nodes exceeds the change threshold, a local update of the updated complete topology is initiated.