An unmanned aerial vehicle group target intention recognition method based on an LSTM-GCN network and an electronic device
By fusing the motion state and formation topology features of a UAV swarm through an LSTM-GCN network, a dynamic adjacency matrix is constructed, which solves the problem of insufficient fusion of temporal and spatial features in UAV swarm target intent recognition and achieves high-precision swarm collaborative intent recognition.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XIDIAN UNIV
- Filing Date
- 2026-04-22
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies cannot fully integrate temporal evolution features and spatial coordination features in the target intent recognition of UAV swarms, resulting in limited intent recognition accuracy. In particular, misjudgment and recognition lag are prone to occur in complex scenarios, which cannot meet the requirements for accurate recognition of UAV swarm combat intent.
By employing an LSTM-GCN network, dynamic adjacency matrix is constructed by acquiring the motion state features and formation topology features of the UAV. The temporal and spatial features are processed by combining LSTM and GCN networks, and the node connection weights are adaptively adjusted to achieve intent recognition.
It improves the accuracy and robustness of drone swarm intent recognition, enabling precise identification of swarm collaboration intent in multi-swarm cross-operations and complex environments.
Smart Images
Figure CN122433797A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of intent recognition, specifically relating to a method and electronic device for target intent recognition of UAV swarms based on LSTM-GCN networks. Background Technology
[0002] In the task of target intent recognition in UAV swarms, traditional deep learning algorithms often use single temporal models (such as LSTM) or spatial models (such as GCN), which can only extract partial feature information and cannot fully integrate the temporal evolution features and spatial coordination features of UAV swarms, resulting in limited intent recognition accuracy. At the same time, when processing complex scene data, single-modal models are difficult to take into account both temporal dynamics and spatial correlations. In situations such as multi-swarm cross-operations and complex environmental interference, they are prone to intention misjudgment and recognition lag, which cannot meet the actual needs of accurate UAV swarm combat intent recognition. Furthermore, existing methods have single feature extraction dimensions and insufficient swarm coordination information mining, which cannot characterize the core coordination features of swarm combat intent. They also have significant shortcomings in the recognition accuracy of intents such as attacks and reconnaissance that rely on swarm coordination. Summary of the Invention
[0003] To address the aforementioned problems in the existing technology, this invention provides a method and electronic device for identifying the target intent of a drone swarm based on an LSTM-GCN network.
[0004] The technical problem to be solved by this invention is achieved through the following technical solution: In a first aspect, the present invention provides a method for identifying the target intent of an unmanned aerial vehicle (UAV) swarm based on an LSTM-GCN network, comprising: Acquire the motion state characteristics of each UAV and the formation topology characteristics of the UAV swarm; The dynamic adjacency matrix corresponding to the current moment of the UAV swarm is determined based on the coupling relationship between the motion state features and the formation topology features. The intention of the UAV swarm is determined by processing the dynamic adjacency matrix and the motion state features using an LSTM-GCN network.
[0005] In one embodiment, determining the dynamic adjacency matrix of the UAV swarm at the current moment based on the coupling relationship between the motion state features and the formation topology features includes: The formation of the drone swarm is determined based on the motion state characteristics, the formation topology characteristics, and the preset mapping model; The static basic adjacency matrix is determined based on the formation, and the static basic adjacency matrix includes the static basic weights between UAV i and UAV j; The dynamic adjacency matrix corresponding to the current moment of the UAV swarm is calculated based on the motion state characteristics and the static basic adjacency matrix.
[0006] In one embodiment, the dynamic adjacency matrix corresponding to the current moment of the UAV swarm is calculated based on the motion state features and the static basic adjacency matrix, including: Calculate the state difference degree based on the motion state characteristics between UAV i and UAV j at time t; The static base weights between UAV i and UAV j are then corrected using the state difference degree to obtain the dynamic weights between UAV i and UAV j at time t; the state difference degree is negatively correlated with the dynamic weights. The dynamic adjacency matrix is obtained based on the dynamic weights between UAV i and UAV j at time t.
[0007] In one embodiment, the state difference degree is calculated based on the motion state characteristics between UAV i and UAV j at time t, including: Calculate the velocity difference and heading angle difference between UAV i and UAV j at time t; The state difference between UAV i and UAV j is determined based on the speed difference and the heading angle difference; wherein, the dynamic weight characterizes the cooperative strength between UAV i and UAV j.
[0008] In one embodiment, the motion state characteristics include: the speed of the UAV, the rate of change of speed, altitude, the rate of change of altitude, acceleration, heading angle, the rate of change of heading angle, and the distance from the user's base.
[0009] In one embodiment, the formation topology features include at least one of the following: the aggregation degree, stability, and connectivity of the UAV swarm; The aggregation degree is calculated based on the position coordinates of all UAVs, and the stability and connectivity are calculated based on the dynamic adjacency matrix at time t; the connectivity includes average connectivity and global connectivity.
[0010] In one embodiment, the stability is calculated based on the following formula (1): Formula (1); in, This is the variance calculation function. Let be the dynamic adjacency matrix at time t. Indicates stability.
[0011] In one embodiment, the average connectivity is calculated based on the following formula (2): Formula (2); Where V represents the number of drones in the drone swarm. Represents the dynamic adjacency matrix at time t The element in the i-th row and j-th column, if =1 indicates that there is a communication link between drone i and drone j. If the value is 0, it means there is no communication link between drone i and drone j. Indicates average connectivity; The global connectivity rate is calculated based on the following formula (3): Formula (3); This represents the global connectivity.
[0012] Secondly, the present invention provides an electronic device, comprising: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus. Memory, used to store computer programs; When the processor executes the program stored in memory, it implements any of the steps of the above-mentioned UAV swarm target intent recognition method based on LSTM-GCN network.
[0013] Thirdly, the present invention provides a computer program product containing instructions that, when run on a computer, cause the computer to perform the steps of any of the above-described methods for identifying the target intent of a drone swarm based on an LSTM-GCN network.
[0014] This invention provides a method for identifying the target intent of a drone swarm based on an LSTM-GCN network, comprising: acquiring the motion state features of each drone and the formation topology features of the drone swarm; determining the dynamic adjacency matrix corresponding to the current moment of the drone swarm based on the coupling relationship between the motion state features and the formation topology features; and processing the dynamic adjacency matrix and the motion state features using an LSTM-GCN network to determine the intent of the drone swarm. The method of this application constructs a dynamic adjacency matrix from multi-dimensional features, thereby combining it with an LSTM-GCN network to achieve intent recognition, which can improve the accuracy of intent recognition.
[0015] The present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description
[0016] Figure 1 This is a flowchart illustrating a method for identifying the target intent of an unmanned aerial vehicle (UAV) swarm based on an LSTM-GCN network, provided in an embodiment of the present invention. Figure 2 A diagram illustrating the formation of a drone swarm; Figure 3 yes Figure 1 A flowchart illustrating step S2; Figure 4 This is a topological diagram illustrating the typical formation interaction relationships of a drone swarm; Figure 5 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation
[0017] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.
[0018] To address the aforementioned problems, this invention provides a method for identifying the target intent of a drone swarm based on an LSTM-GCN network, specifically combined with... Figure 1 ,include: Step S1: Obtain the motion state characteristics of each UAV and the formation topology characteristics of the UAV swarm.
[0019] Specifically, the motion state characteristics include at least one of the following: the UAV's speed, rate of change of speed, altitude, rate of change of altitude, acceleration, heading angle, rate of change of heading angle, and distance from friendly base.
[0020] The speed of a drone is the magnitude of its displacement per unit time, calculated from the distance at consecutive moments, as shown in the following formula: ; in, Indicates the speed of the drone. Indicates displacement. Indicates time interval, This represents the change in displacement in the x-direction. This represents the change in displacement in the y-direction.
[0021] The formula for calculating the rate of change of velocity is: ; in, Represents the rate of change of velocity. Represents the velocity at time t+1. This represents the velocity at time t.
[0022] The altitude h of the drone is equal to its distance from the ground in the z-direction, denoted as... , This represents the distance between the z-direction and the ground at time t.
[0023] height change rate It is obtained by subtracting the altitude values of the target drone at consecutive moments before and after, as shown in the following formula: ; This represents the height at time t+1. This represents the altitude at time t.
[0024] acceleration The rate of change of the velocity vector has both magnitude and direction, and is calculated from the velocities at consecutive moments in time, as shown in the following formula: ; in, This represents the velocity vector at time t+1. This represents the velocity vector at time t.
[0025] Heading angle This refers to the relative position of the target drone and our own base, specifically: ; Rate of change of heading angle The calculation is as follows: ; This represents the heading angle at time t+1. This represents the heading angle at time t.
[0026] Distance from friendly base The calculation is as follows: ; in, The coordinates of our base, ( , , ( ) represents the coordinates of the target UAV at time t.
[0027] The target formation of a drone swarm is not a static geometric arrangement, but a dynamic collaborative pattern deeply bound to combat intent. The topological structure of different formations directly serves their mission requirements. This study extracts 8-dimensional motion state features to construct a mapping model from formation topology to combat intent. It focuses on analyzing the formation generation logic corresponding to four typical intents: attack, reconnaissance, patrol, and retreat. By combining dynamic adjacency matrices to quantify the connectivity and stability of the formations, dynamic evaluation indicators are established.
[0028] Combination Figure 2 , Figure 2 The middle (a) formation is rectangular, and its core purpose is patrol or reconnaissance. Figure 2 The middle (b) formation is a wedge-shaped formation, with the core intention being attack. Figure 2 The middle (c) formation is a vertical formation, often used for retreat missions; Figure 2 The middle (d) formation is a diamond formation, often used for patrol missions; Figure 2The middle (e) is a fan-shaped formation, typically used for reconnaissance and patrol missions.
[0029] Based on the above correspondence between formations and mission scenarios, the mapping relationship between formation and combat intent can be summarized as shown in Table (1):
[0030] Table (1) The core logic behind differentiated formation designs for different operational intentions is to maximize mission execution efficiency by adjusting the cluster topology. Quantitative indicators of formation topology characteristics include aggregation degree, stability, and connectivity. Specifically, the formation topology features include at least one of the following: aggregation degree, stability, and connectivity of the UAV swarm; wherein the aggregation degree is calculated based on the position coordinates of all UAVs, and the stability and connectivity are calculated based on the dynamic adjacency matrix at time t; the connectivity includes average connectivity and global connectivity.
[0031] The aggregation degree is calculated using the standard deviation of the position coordinates of all UAVs, using the following formula: ; in, V For the number of drones, For the first i The coordinates of the drone The coordinates are the center coordinates of the cluster.
[0032] The stability is calculated based on the following formula (1): Formula (1); in, This is the variance calculation function. Let be the dynamic adjacency matrix at time t. Indicates stability.
[0033] The average connectivity is calculated based on the following formula (2): Formula (2); Where V represents the number of drones in the drone swarm. Represents the dynamic adjacency matrix at time t The element in the i-th row and j-th column, if =1 indicates that there is a communication link between drone i and drone j. If the value is 0, it means there is no communication link between drone i and drone j. Indicates average connectivity; The global connectivity rate is calculated based on the following formula (3): Formula (3); This represents the global connectivity.
[0034] Step S2: Determine the dynamic adjacency matrix corresponding to the current moment of the UAV swarm based on the coupling relationship between the motion state features and the formation topology features.
[0035] In one embodiment, combined with Figure 3 Step S2 includes: Step S21: Determine the formation of the drone swarm based on the motion state features, the formation topology features, and the preset mapping model.
[0036] Specifically, the typical formations, core topological features, and associated motion states corresponding to different combat intentions have corresponding mapping relationships, i.e., preset mapping models, as shown in Table (2) below:
[0037] Table (2) Specifically, based on the coupling relationship between the motion state characteristics (speed, speed change rate, altitude, altitude change rate, acceleration, heading angle, heading angle change rate and distance from the base of the UAV) and the formation topology characteristics (aggregation degree, stability and connectivity of the UAV swarm) calculated in step S1, combined with the preset mapping model in Table (2), the formation of the UAV swarm can be determined.
[0038] Step S22: Determine the static basic adjacency matrix based on the formation.
[0039] Within a drone swarm, drones are not independent entities, but rather tightly coupled through multi-dimensional relationships involving command, coordination, and cover, forming a dynamically evolving collaborative system. These interactions not only determine the overall operational efficiency of the swarm but also serve as the core basis for understanding its collective intent. For example, in an attack mission scenario, there is a one-way command flow between the core command unit and the execution units, a synchronized coordination between the main units, and a defensive cover relationship between the auxiliary drones and the core attack units. These interwoven and dynamically changing relationships enable the swarm to overcome the limitations of individual capabilities and complete complex swarm tasks. (See details...) Figure 4 , Figure 4 This represents the topology of typical formation interactions in a drone swarm.
[0040] To accurately depict the collaborative relationships of UAV swarms evolving over time during combat, the method in this application starts from the static formation structure of the UAV swarm and dynamically adjusts the connection weights between nodes based on real-time motion states (relative position, speed, etc.) to generate a temporal topology structure containing physical feature information, providing reliable topology input for subsequent GCN branches.
[0041] It should be noted that the static basic adjacency matrix includes the static basic weights between UAV i and UAV j.
[0042] A fixed interaction topology is provided for different formations, defining the basic weights for three types of links: command, coordination, and cover, forming the basic framework of a dynamic adjacency matrix. Differentiated link distributions are defined for five typical formations, categorized as unidirectional and bidirectional, as follows: 1) Command and control link: core node → subordinate node (unidirectional, weight 2.0), ensuring the transmission of instructions; 2) Collaborative Link: Peer-to-peer interaction between nodes (bidirectional, weight 1.5) to ensure coordinated actions; 3) Covert link: Subordinate node → Core node (unidirectional, weight 1.0), ensuring core security; The topology is quantified using a static basic adjacency matrix. Diagonal elements are 0, indicating no self-cooperative relationships. Non-zero elements represent the basic weights of the corresponding links. The resulting static basic adjacency matrix is as follows: .
[0043] Step S23: Calculate the dynamic adjacency matrix corresponding to the current moment of the UAV swarm based on the motion state features and the static basic adjacency matrix.
[0044] In one embodiment, the state difference degree is calculated based on the motion state characteristics between UAV i and UAV j at time t; and the static basic weights between UAV i and UAV j are corrected using the state difference degree to obtain the dynamic weights between UAV i and UAV j at time t; the state difference degree is negatively correlated with the dynamic weights.
[0045] In one embodiment, the velocity difference and heading angle difference between UAV i and UAV j at time t are calculated. Specifically, based on the real-time motion state of the UAVs, the cooperative weights are adaptively changed over time; the more similar the motion states, the closer the cooperative relationship, i.e., the higher the weight.
[0046] For example, the two features that have the greatest impact on the interaction relationship are selected, including speed and heading angle. Speed reflects the synchronization of UAV maneuvers, and heading angle reflects the consistency of UAV's direction of movement. The state difference is calculated and normalized to eliminate the influence of dimensions and ensure the comparability of the difference. The specific calculation formula is as follows: ; in, They represent the times respectively. t drones i With drones j The absolute values of the speed difference and heading angle difference are taken to ensure they are non-negative. This represents the velocity of drone i at time t. This represents the velocity of drone j at time t. This represents the heading angle of UAV i at time t. The heading angle of UAV j at time t is indicated.
[0047] Further normalization of the speed difference and heading angle difference is expressed as follows: ; These represent the normalized velocity difference and heading angle difference at time t, respectively.
[0048] The state difference between UAV i and UAV j is determined based on the speed difference and the heading angle difference. Specifically: .
[0049] The static base weights between UAV i and UAV j are then corrected using the state difference degree to obtain the dynamic weights between UAV i and UAV j at time t; the state difference degree is negatively correlated with the dynamic weights.
[0050] In one embodiment, an exponential decay function is used to calculate the dynamic weights to ensure their smoothness. Furthermore, the smaller the state variability, the higher the dynamic weight; that is, the state variability is negatively correlated with the dynamic weight. The dynamic weights are calculated as follows: ; in, Indicates at time t drones i and j The dynamic weights between them represent the cooperative strength between UAV i and UAV j. Indicates at time t drones i and j The static base weights between them This is the attenuation coefficient, ensuring the weight is within... Internal fluctuations This is an exponentially decaying term.
[0051] Through the above process, the dynamic weights between UAV i and UAV j are obtained, and the dynamic adjacency matrix is obtained based on the dynamic weights between UAV i and UAV j at time t.
[0052] Specifically, a zero-matrix of dimension (T, V1, V2) is created, where T represents time, V1 represents the number of drones in the column dimension, and V2 represents the number of drones in the row dimension. For each time step, all drone node pairs (i, j) are iterated over, and dynamic weights are applied. Assigning values to the all-zero matrix, the elements that are 0 in the static basic adjacency matrix (i.e., have no interaction relationship) are kept at their corresponding positions as 0, so as not to generate invalid links. Finally, a complete dynamic adjacency matrix is generated with dimensions (T, V1, V2).
[0053] Step S3: Use an LSTM-GCN network to process the dynamic adjacency matrix and the motion state features to determine the intent of each UAV swarm.
[0054] It should be noted that the LSTM-GCN network is an architecture of dual-branch parallel extraction and weighted fusion, which is suitable for the deep fusion of temporal and spatial features. The specific design is as follows: (a) An LSTM-GCN model was trained and built using the PyTorch framework for application to intent recognition of UAV swarm targets. Repeated experiments were conducted with different model parameters to determine the optimal parameter settings for each layer. The proposed LSTM-GCN network combination model mainly consists of two parts: a Graph Convolutional Neural Network (GCN) and a Long Short-Term Memory (LSTM) network. The GCN processes the dynamic adjacency matrix and node feature matrix of the UAV swarm at each time step, while the LSTM part receives the temporal spatial feature sequence extracted by the GCN and captures long-term temporal dependencies through memory units. The model adopts a dual-branch parallel feature extraction and attention-weighted fusion architecture, consisting of four parts: a feature embedding layer, an LSTM temporal branch, a GCN topology branch, and a weighted fusion layer.
[0055] (b) Core Parameter Settings: The LSTM-GCN UAV swarm intent recognition model contains many key parameters, including the number of GCN layers, the number of LSTM layers, the hidden layer dimension, the Dropout rate, the optimizer learning rate, and the weight decay coefficient. This application selects the ReLU function as the model's activation function, uses cross-entropy loss as the model's loss function, employs the AdamW optimization algorithm to update the model's parameters, and introduces a cosine annealing learning rate scheduler to improve training stability. To prevent overfitting, a Dropout method is introduced between layers, setting 10% of neurons to be randomly deactivated.
[0056] Before using the model, it needs to be trained. During model training, the model input consists of two types of tensor data: the time-series node features of the drone swarm. Dynamic adjacency matrix of drone swarm: ,in, B For batch size, T For time step, V For the number of drones, F For single-node feature dimensions, H To hide the spatial dimension, the model output is the probability of intention prediction.
[0057] The dynamic adjacency matrix is subjected to linear mapping, bidirectional temporal encoding, and feature fusion processing to obtain a temporal feature sequence, which represents the intent of the UAV swarm. Specifically, the feature extraction process is as follows: 1. Input data preparation: The time-series feature matrix X of the UAV swarm, with dimensions [missing information]. .
[0058] 2. Feature Dimension Regularization and Aggregation: For each time step T ,Will V The feature vectors of the drones are concatenated and fused to form a global feature vector containing information about all drones in the cluster. The data is then reconstructed into an LSTM with the standard input format as follows. This ensures that the input at each time step fully represents the overall motion state of the drone swarm at that moment.
[0059] 3. High-dimensional feature embedding: Through a linear mapping layer, the feature dimension of a single node is increased from... F Upgraded to H The feature dimension is updated to .
[0060] 4. Bidirectional temporal coding and context capture: The forward LSTM captures the forward motion evolution trend of the drone swarm from the beginning to the end of the column, while the backward LSTM captures the backward dependencies (such as the dependence of the final intention on previous actions) from the end to the beginning of the sequence. The bidirectional output features are normalized and processed by Dropout to prevent overfitting.
[0061] 5. Bidirectional Feature Fusion and Output: The features output by the bidirectional LSTM are fused and dimensionality reduced, resulting in a final output dimension of [dimensionality value missing]. The temporal characteristic sequence.
[0062] The training process of the LSTM-GCN intent prediction model includes the following key steps: 1. Data preprocessing and loading: The time series features and dynamic adjacency matrix of the UAV are standardized and divided into training set, validation set and test set in a ratio of 7:2:1. A dataset and data loader adapted to the model input format are constructed.
[0063] 2. Initialize the model and training components: Initialize the network structure and all trainable parameters of the LSTM-GCN dual-branch model, and set appropriate hidden layer dimensions and node numbers. Select the cross-entropy loss function and AdamW optimizer, and configure the initial learning rate. Simultaneously, introduce a learning rate scheduler to improve convergence stability. Complete the training environment configuration, laying the foundation for subsequent parameter iteration and optimization.
[0064] 3. Forward Propagation Computation: Batch feature data and the adjacency matrix are input into the model. The LSTM branch extracts temporal trajectory features, and the GCN branch mines topological interaction information. A feature fusion layer weights and fuses the two types of features, fully utilizing the complementary information of temporal and topological data. The predicted probabilities of various combat intentions are output after passing through a fully connected layer and a Softmax activation function.
[0065] 4. Loss Calculation: Using the model's predicted probabilities and the true intent labels as inputs, the batch loss is calculated using a multi-class cross-entropy loss function. To address the imbalanced sample problem, the loss is appropriately weighted to improve the model's ability to recognize minority intents. The loss value is used as the basis for backpropagation and parameter updates, guiding the direction of model optimization.
[0066] 5. Backpropagation and Gradient Calculation: Based on the loss value, the gradients of the parameters in each layer of the model are calculated using the backpropagation algorithm to obtain the parameter update direction. To avoid the gradient explosion problem caused by long sequences and graph structures, a gradient pruning strategy is used to constrain the gradient range. This ensures the stability of gradient values and prevents oscillations or non-convergence during training.
[0067] 6. Weight Update: The AdamW optimizer iteratively updates the model weights based on gradient information, gradually reducing prediction error. After each training round, model performance is evaluated on the validation set, and the learning rate is dynamically adjusted based on validation set metrics. Training stops when validation set performance shows no improvement for several consecutive rounds, and the optimal model weights are retained.
[0068] 7. Results Output and Evaluation: Load the optimal model onto an independent test set to generate UAV swarm combat intent recognition results. Calculate multi-dimensional metrics such as precision, recall, and F1 score to comprehensively evaluate model performance. Verify the effectiveness and superiority of the LSTM-GCN model in intent recognition tasks by comparing it with various benchmark models.
[0069] The method proposed in this application includes a dual-branch deep fusion model architecture: An LSTM-GCN dual-branch network structure is proposed, using LSTM to capture temporal evolution features and GCN to mine spatial collaborative features, combined with an attention-weighted fusion mechanism to achieve deep complementarity of dual-modal information; based on static formation topology and real-time motion states (velocity, heading angle), node connection weights are adaptively adjusted through an exponential decay function to generate a temporal dynamic adjacency matrix, accurately characterizing the dynamic changes in cluster collaborative relationships; and an 8-dimensional motion state feature coupled with 5 types of formation topology is established to quantify the differentiated features of different combat intentions and enhance the ability to distinguish similar intentions.
[0070] The method provided in this invention can be applied to electronic devices. Specifically, the electronic device can be a desktop computer, a portable computer, a smart mobile terminal, a server, etc. No limitation is made herein; any electronic device that can implement this invention falls within the protection scope of this invention.
[0071] Based on the same inventive concept, embodiments of the present invention also provide an electronic device. Embodiments of the present invention also provide an electronic device, such as... Figure 5 As shown, it includes a processor 601, a communication interface 602, a memory 603, and a communication bus 604, wherein the processor 601, the communication interface 602, and the memory 603 communicate with each other through the communication bus 604. Memory 603 is used to store computer programs; When the processor 601 executes the program stored in the memory 603, it implements the steps of the above-described method for identifying the target intent of a UAV swarm based on an LSTM-GCN network.
[0072] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus.
[0073] The communication interface is used for communication between the aforementioned electronic devices and other devices.
[0074] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.
[0075] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0076] The present invention also provides a computer-readable storage medium. A computer program is stored in the computer-readable storage medium, and when executed by a processor, the computer program implements the steps of the above-described method for identifying the target intent of a UAV swarm based on an LSTM-GCN network.
[0077] Optionally, the computer-readable storage medium may be non-volatile memory (NVM), such as at least one disk storage device.
[0078] Optionally, the computer-readable storage medium may also be at least one storage device located remotely from the aforementioned processor.
[0079] In another embodiment of the present invention, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to perform the steps of the above-described method for identifying the target intent of a drone swarm based on an LSTM-GCN network.
[0080] It should be noted that, for the embodiments of the device / electronic device / storage medium / computer program, since they are basically similar to the method embodiments, the description is relatively simple. For relevant parts, please refer to the description of the method embodiments. All embodiments of the above-described UAV swarm target intent recognition method based on LSTM-GCN network are applicable to the device, electronic device and storage medium, and can achieve the same or similar beneficial effects.
[0081] It should be noted that the terms "first," "second," etc., are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention.
[0082] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.
[0083] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings and the disclosure in carrying out the claimed invention. In the description of the invention, the word "comprising" does not exclude other components or steps, "a" or "an" does not exclude a plurality, and "a plurality" means two or more, unless otherwise explicitly specified. Furthermore, while different embodiments may describe certain measures, this does not mean that these measures cannot be combined to produce good results.
[0084] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus (devices), or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects, all of which are collectively referred to herein as "modules" or "systems." Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The computer program may be stored / distributed in a suitable medium, provided with or as part of other hardware, or may take other distribution forms, such as via the Internet or other wired or wireless telecommunications systems.
[0085] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0086] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0087] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0088] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.
Claims
1. A method for identifying the target intent of an unmanned aerial vehicle (UAV) swarm based on an LSTM-GCN network, characterized in that, include: Acquire the motion state characteristics of each UAV and the formation topology characteristics of the UAV swarm; The dynamic adjacency matrix of the UAV swarm at the current moment is determined based on the coupling relationship between the motion state features and the formation topology features. The intention of the UAV swarm is determined by processing the dynamic adjacency matrix and the motion state features using an LSTM-GCN network.
2. The method according to claim 1, characterized in that, The dynamic adjacency matrix of the UAV swarm at the current moment is determined based on the coupling relationship between the motion state features and the formation topology features, including: The formation of the drone swarm is determined based on the motion state characteristics, the formation topology characteristics, and the preset mapping model; The static basic adjacency matrix is determined based on the formation, and the static basic adjacency matrix includes the static basic weights between UAV i and UAV j; The dynamic adjacency matrix corresponding to the current moment of the UAV swarm is calculated based on the motion state characteristics and the static basic adjacency matrix.
3. The method according to claim 2, characterized in that, The dynamic adjacency matrix of the UAV swarm at the current moment is calculated based on the motion state features and the static basic adjacency matrix, including: Calculate the state difference degree based on the motion state characteristics between UAV i and UAV j at time t; The static base weights between UAV i and UAV j are then corrected using the state difference degree to obtain the dynamic weights between UAV i and UAV j at time t; the state difference degree is negatively correlated with the dynamic weights. The dynamic adjacency matrix is obtained based on the dynamic weights between UAV i and UAV j at time t.
4. The method according to claim 3, characterized in that, The state difference degree is calculated based on the motion state characteristics between UAV i and UAV j at time t, including: Calculate the velocity difference and heading angle difference between UAV i and UAV j at time t; The state difference between UAV i and UAV j is determined based on the speed difference and the heading angle difference; wherein, the dynamic weight characterizes the cooperative strength between UAV i and UAV j.
5. The method according to claim 1, characterized in that, The motion characteristics include: the UAV's speed, rate of change of speed, altitude, rate of change of altitude, acceleration, heading angle, rate of change of heading angle, and distance from friendly base.
6. The method according to claim 1, characterized in that the formation topology features include: At least one of the following: aggregation degree, stability degree, and connectivity degree of the drone swarm; The aggregation degree is calculated based on the position coordinates of all UAVs, and the stability and connectivity are calculated based on the dynamic adjacency matrix at time t; the connectivity includes average connectivity and global connectivity.
7. The method according to claim 6, characterized in that the stability is calculated based on the following formula (1): Official (1); in, This is the variance calculation function. Let be the dynamic adjacency matrix at time t. Indicates stability.
8. The method according to claim 6, characterized in that, The average connectivity is calculated based on the following formula (2): Official (2); Where V represents the number of drones in the drone swarm. Represents the dynamic adjacency matrix at time t The element in the i-th row and j-th column, if =1 indicates that there is a communication link between drone i and drone j. If the value is 0, it means there is no communication link between drone i and drone j. Indicates average connectivity; The global connectivity rate is calculated based on the following formula (3): Official (3); This represents the global connectivity.
9. An electronic device, characterized in that, include: The electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus. Memory, used to store computer programs; When the processor executes the program stored in the memory, it implements the steps of the UAV swarm target intent recognition method based on LSTM-GCN network according to any one of claims 1 to 8.
10. A computer program product containing instructions, characterized in that, When it is run on a computer, it causes the computer to perform the steps of the UAV swarm target intent recognition method based on LSTM-GCN network according to any one of claims 1 to 8.