Graph convolution-attention mechanism-based unmanned aerial vehicle group formation recognition method, medium and equipment

Abstract

The invention discloses an unmanned aerial vehicle group formation recognition method based on a graph convolution-attention mechanism, and belongs to the technical field of unmanned aerial vehicle group formation recognition, and the method comprises the following steps: S1, obtaining the position information of an unmanned aerial vehicle group target; s2, constructing unmanned aerial vehicle group graph node data according to the position information of the unmanned aerial vehicle group target obtained in the step S1, and extracting the unmanned aerial vehicle group graph node data according to the constructed unmanned aerial vehicle group graph node data to obtain unmanned aerial vehicle group graph node features; and S3, extracting node features of the unmanned aerial vehicle group graph obtained in the step S2 by using a graph convolution-attention mechanism network to realize formation recognition of the unmanned aerial vehicle group. According to the identification method, the influence of the unmanned aerial vehicle group member target features on the network does not need to be considered, so that the identification rate of the unmanned aerial vehicle group formation can be effectively improved under the condition that the unmanned aerial vehicle group member target texture features are not prominent.

Description

Technical Field

[0001] The present invention belongs to the technical field of drone swarm formation recognition, and specifically relates to a drone swarm formation recognition method, medium and equipment based on a graph convolution-attention mechanism. Background Art

[0002] From a military perspective, the rapid development of drone swarm technology implies a challenge to the traditional joint air defense system, and will even trigger a disruptive change in the air defense equipment system. With the continuous maturity of drone swarm combat concepts and technologies, anti-drone swarm combat has become more complex, and the development of countermeasure capabilities is urgent. For a small number of drone swarms, countermeasures such as navigation deception, physical capture and physical destruction can be taken, but it is difficult to effectively counter large-scale drone swarms. Therefore, identifying the formation of drone swarms and combining relevant information to find important communication node drones in the drone swarm will help to formulate a reasonable drone swarm countermeasure strategy and improve the countermeasure effect. At present, the field of drone swarm formation recognition is in its infancy, and related research on drone swarm formation recognition methods is still in development. From the current research results, there is no public technology to explain the drone swarm formation recognition method. When the existing image-level formation recognition algorithm is used for drone swarm formation recognition, the convolutional neural network is easy to lose the characteristics of drone swarm members in the process of extracting drone swarm formation features, and obtains very little drone swarm formation feature information, which is difficult to use to distinguish the type of drone swarm formation, so the recognition effect is poor. Summary of the invention

[0003] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a method, medium and device for drone swarm formation recognition based on graph convolution-attention mechanism.

[0004] The first object of the present invention is to provide a method for identifying drone group formation based on graph convolution-attention mechanism, comprising the following steps:

[0005] S1. Obtain the location information of the drone group target;

[0006] S2. Constructing drone swarm graph node data according to the location information of the drone swarm target obtained in step S1, and extracting the drone swarm graph node data according to the constructed drone swarm graph node data to obtain drone swarm graph node features;

[0007] S3. Use the graph convolution-attention mechanism network to extract the node features of the drone group graph obtained in step S2 to realize drone group formation recognition.

[0008] Preferably, step S1 comprises the following steps:

[0009] S11, obtaining a flight video of the drone swarm by capturing the drone swarm target, and obtaining a flight image of the drone swarm by intercepting the flight video of the drone swarm across frames;

[0010] S12: Input the drone swarm flight image obtained in step S11 into the YOLOv5 target detection algorithm, and output the location information of the drone swarm targets.

[0011] Preferably, S2 comprises the following steps:

[0012] S21, according to the location information of the drone swarm target obtained in step S1, the drone swarm graph node data is obtained through the NetworkX library;

[0013] S22, normalizing the drone swarm graph node data obtained in step S21 to obtain normalized drone swarm graph node data, that is, the drone swarm graph node data is normalized by affine transformation to ensure that the dimension of the normalized graph node is consistent with the dimension of the multi-layer perceptron, so as to facilitate the subsequent extraction of drone swarm graph node features;

[0014] S23. Input the normalized drone group graph node data obtained in step S22 into a 64-dimensional multi-layer perceptron to obtain the drone group graph node features.

[0015] Preferably, in step S3, the graph convolution-attention mechanism network consists of a four-layer network, a graph pooling layer and a Softmax function.

[0016] Preferably, each network layer consists of a graph convolution layer and a multi-layer perceptron with different convolution kernels.

[0017] Preferably, the number of neurons of the four multilayer perceptrons is 64, 128, 256, and 512, respectively.

[0018] Preferably, step S3 comprises the following steps:

[0019] S31, inputting the drone group graph node features obtained in step S2 into the four-layer network in the graph convolution-attention mechanism network to obtain the drone group graph node feature information with node association;

[0020] S32, inputting the drone group graph node feature information obtained in step S31 into the graph pooling layer in the graph convolution-attention mechanism network to obtain the feature vector of the drone group graph node;

[0021] S33. Calculate the feature vector of the drone swarm graph node obtained in step S32 through the Softmax function to obtain the probability of each drone swarm formation type. The drone swarm formation type corresponding to the maximum probability is the formation category represented by the drone swarm, thereby realizing drone swarm formation recognition.

[0022] The second object of the present invention is to provide a computer-readable storage medium, which includes a stored computer program; wherein, when the computer program is running, it controls the device where the computer-readable storage medium is located to execute the steps of the above-mentioned drone group formation recognition method.

[0023] The third object of the present invention is to provide a terminal device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the steps of the above-mentioned drone swarm formation recognition method when executing the computer program.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] The method provided by the present invention is to apply a graph convolutional neural network to the low-altitude drone group formation recognition on the basis of detecting the position information of the drone group members. Except for the feature extraction of the drone group target involved in the detection process, the features of the drone group target in the image are not used in other stages. Therefore, the present invention does not need to consider the influence of the target features of the drone group members on the network. Therefore, the present invention can effectively improve the recognition rate of the drone group formation when the texture features of the drone group members are not prominent. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A flowchart of a method for identifying drone swarm formations based on a graph convolution-attention mechanism provided in an embodiment of the present invention;

[0027] Figure 2 It is a graph node data representation diagram of a drone group in a triangle and rectangular formation type in an embodiment of the present invention;

[0028] Figure 3 is a flow chart of step S2 in an embodiment of the present invention;

[0029] Figure 4 This is a structural diagram of the graph convolution-attention mechanism network in an embodiment of the present invention. DETAILED DESCRIPTION

[0030] The following is a description of the embodiments of the present invention. Figures 1 to 4 , the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] Example 1

[0032] like Figure 1 As shown, an embodiment of the present invention provides a method for identifying drone group formation based on graph convolution-attention mechanism, which specifically includes the following steps:

[0033] S1. Obtaining the location information of the drone group target, specifically including the following steps:

[0034] The drone swarm targets are captured by an optical camera to obtain the drone swarm flight video, and the drone swarm flight image is obtained by intercepting the drone swarm flight video across frames;

[0035] S12. Input the drone swarm flight image obtained in step S11 into the YOLOv5 target detection algorithm, and output the location information of the drone swarm target, that is, the center point location coordinates (x, y) of the drone swarm target.

[0036] S2. Based on the location information of the drone group target obtained in step S1, the drone group graph node data is constructed, such as Figure 2 As shown, according to the constructed drone group graph node data, the drone group graph node data is extracted to obtain the drone group graph node features, such as Figure 3 As shown, the following steps are included:

[0037] S21, based on the location information of the drone swarm target obtained in step S1, obtain the drone swarm graph node data by using the NetworkX library that can be used to create, operate and learn the structure, dynamics and functions of complex graph networks;

[0038] S22, normalizing the drone swarm graph node data obtained in step S21 to obtain normalized drone swarm graph node data, that is, the drone swarm graph node data is normalized by affine transformation to ensure that the dimension of the normalized graph node is consistent with the dimension of the multi-layer perceptron, so as to facilitate the subsequent extraction of drone swarm graph node features;

[0039] S23, inputting the normalized N drone swarm graph node data obtained in step S22 into a 64-dimensional multi-layer perceptron to obtain drone swarm graph node features with a dimension of N×64;

[0040] S3. Use the graph convolution-attention mechanism network to extract the node features of the drone group graph obtained in step S2 to realize drone group formation recognition.

[0041] like Figure 4As shown, the graph convolution-attention mechanism network in the embodiment of the present invention is composed of a four-layer network, a graph pooling layer and a Softmax function. Each layer of the network is composed of a graph convolution layer and a multi-layer perceptron with different convolution kernels. The number of neurons of the four multi-layer perceptrons is 64, 128, 256, and 512 respectively. The number of neurons represents the dimension of the output of each layer of the network. The specific analysis is as follows:

[0042] (1) Graph Convolutional Layer

[0043] The node feature information of the drone swarm graph does not contain the node feature information other than itself, while the drone swarm formation recognition needs to connect the feature information of other nodes. The graph convolution layer associates each node feature in the drone swarm graph node data by aggregation, thereby obtaining the drone swarm graph node feature information with node association.

[0044] (2) Graph Pooling and Softmax Function

[0045] The input of the graph pooling operation is the representation of all nodes in the drone swarm graph node data, and the output is the representation of the entire graph. This operation is used to generate a graph representation, and the characteristic attributes of the drone swarm graph nodes are calculated into a tensor representing the graph. The main idea of ​​graph pooling is that the graph pooling operation itself is insensitive to the order of the drone swarm graph nodes, and aggregates and summarizes the information of individual nodes and edges. The aggregation methods for implementing graph pooling mainly include summing all node features in the drone swarm graph data, taking the average value, and finding the maximum or minimum value of each element. The aggregation method of the embodiment of the present invention is to average all node features of each graph to obtain the drone swarm graph node feature vector:

[0046]

[0047] Among them, h UG V represents the node feature vector of the drone swarm graph; UG A node set representing the node data of the drone swarm graph; h i Represents the characteristics of node i.

[0048] The Softmax function is mainly used to calculate the probability of each type of drone group formation.

[0049]

[0050] Among them, e hUG Represents the index of the feature vector of the drone swarm graph node; N represents the number of drone swarm graph nodes.

[0051] In the embodiment of the present invention, step S3 specifically includes the following steps:

[0052] S31, inputting the drone group graph node features obtained in step S2 into the four-layer network in the graph convolution-attention mechanism network to obtain the drone group graph node feature information with node association;

[0053] S32, inputting the drone group graph node feature information obtained in step S31 into the graph pooling layer in the graph convolution-attention mechanism network to obtain the feature vector of the drone group graph node;

[0054] S33, calculating the feature vectors of the drone group graph nodes obtained in step S32 through the Softmax function to obtain the probability of each drone group formation type, and the drone group formation type corresponding to the maximum probability is the formation category represented by the drone group, thereby realizing drone group formation recognition.

[0055] An embodiment of the present invention provides a computer-readable storage medium, which includes a stored computer program; wherein the computer program, when running, controls the device where the computer-readable storage medium is located to execute the steps of the drone group formation recognition method provided in Example 1.

[0056] Example 3

[0057] An embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When executing the computer program, the processor implements the steps of the drone swarm formation recognition method provided in Example 1.

[0058] The performance of the drone group formation recognition method based on graph convolution-attention mechanism provided in Example 1 of the present invention is studied below.

[0059] The formation recognition method provided in Example 1 of the present invention is compared with the existing image-level algorithms (formation recognition algorithm based on VGG-16, formation recognition algorithm based on ResNet-50, and formation recognition algorithm based on ResNet-101) and position-level algorithms (formation recognition algorithm based on AGCN and formation recognition algorithm based on GAT) for the recognition results of the drone group formation. The comparison results are shown in Table 1 below.

[0060] Table 1 Comparative experimental results of UAV group formation recognition algorithms

[0061]

[0062]

[0063] It can be seen from the results in Table 1 that the average classification accuracy of the formation recognition method provided in Example 1 of the present invention is 79.56%, which is significantly higher than other recognition methods, indicating that the formation recognition method provided in Example 1 of the present invention has a better drone group formation recognition effect.

[0064] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A method for drone group formation recognition based on graph convolution-attention mechanism, characterized in that: The following steps are involved: S1. Obtain the location information of the drone group target; S2. Constructing drone swarm graph node data according to the location information of the drone swarm target obtained in step S1, and extracting the drone swarm graph node data according to the constructed drone swarm graph node data to obtain drone swarm graph node features; S3. Use the graph convolution-attention mechanism network to extract the node features of the drone group graph obtained in step S2 to realize drone group formation recognition.

2. The method for identifying drone group formation based on graph convolution-attention mechanism according to claim 1 is characterized in that: Step S1 comprises the following steps: S11, obtaining a flight video of the drone swarm by capturing the drone swarm target, and obtaining a flight image of the drone swarm by intercepting the flight video of the drone swarm across frames; S12: Input the drone swarm flight image obtained in step S11 into the YOLOv5 target detection algorithm, and output the location information of the drone swarm targets.

3. The method for identifying drone group formation based on graph convolution-attention mechanism according to claim 1 is characterized in that: S2 contains the following steps: S21, according to the location information of the drone swarm target obtained in step S1, the drone swarm graph node data is obtained through the NetworkX library; S22, normalizing the drone group graph node data obtained in step S21 to obtain normalized drone group graph node data; S23. Input the normalized drone group graph node data obtained in step S22 into a 64-dimensional multi-layer perceptron to obtain the drone group graph node features.

4. The method for identifying drone group formation based on graph convolution-attention mechanism according to claim 1 is characterized in that: In step S3, the graph convolution-attention mechanism network consists of a four-layer network, a graph pooling layer and a Softmax function.

5. The method for identifying drone group formation based on graph convolution-attention mechanism according to claim 4 is characterized in that: Each layer of the network consists of a graph convolutional layer and a multi-layer perceptron with different convolution kernels.

6. The method for identifying drone swarm formations based on graph convolution-attention mechanism according to claim 5 is characterized in that: The number of neurons in the four multilayer perceptrons are 64, 128, 256, and 512 respectively.

7. The method for identifying drone group formation based on graph convolution-attention mechanism according to claim 6 is characterized in that: Step S3 includes the following steps: S31, inputting the drone group graph node features obtained in step S2 into the four-layer network in the graph convolution-attention mechanism network to obtain the drone group graph node feature information with node association; S32, inputting the drone group graph node feature information obtained in step S31 into the graph pooling layer in the graph convolution-attention mechanism network to obtain the feature vector of the drone group graph node; S33. Calculate the feature vector of the drone swarm graph node obtained in step S32 through the Softmax function to obtain the probability of each drone swarm formation type. The drone swarm formation type corresponding to the maximum probability is the formation category represented by the drone swarm, thereby realizing drone swarm formation recognition.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program; wherein, when the computer program is run, it controls the device where the computer-readable storage medium is located to execute the steps of the drone group formation recognition method as described in any one of claims 1 to 7.

9. A terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the steps of the drone group formation recognition method as described in any one of claims 1 to 7 when executing the computer program.

Images