Networking radar anomaly detection method and system based on graph neural network

Abstract

The invention provides a networked radar anomaly detection method and system based on a graph neural network, and belongs to the technical field of Internet of Things security, and the method comprises the steps: completing a target track route detected by a radar according to the recording of a track detected by the radar for a target, and determining a track result observed by each radar in the networked radar; calculating the attention value of the track observed by each radar node, adaptively determining an adjacent matrix between the nodes, and taking the adjacent matrix as one of the characteristics of the networking radar; predicting the node condition at the next moment through a graph neural network according to the observation result of each node in the networking radar and the adjacent matrix between the radar nodes; and comparing the node track information predicted by each radar node at the next moment with a track result observed by the radar at the next moment so as to judge the radar node in an abnormal state in the networking radar. According to the method, the sensor in the abnormal state in the Internet of Things can be rapidly detected in real time, the robustness is relatively high, and the accuracy is relatively high.

Description

Technical Field

[0001] The present invention belongs to the field of networked radar safety technology, and in particular, relates to a networked radar anomaly detection method and system based on a graph neural network. Background Art

[0002] In recent decades, the networked radar system (NRS) has received extensive attention and has been applied to more and more fields. Networked radars can obtain the multi-directional scattering coefficient of the target through cooperative detection by multiple radars in the network, which can greatly improve the detection and tracking capabilities of the target. At present, networked radar systems mostly adopt the method of detect before track (DBT), that is, the local radar node performs a single-cycle threshold detection on the received echo signal at each moment to obtain some false alarm points that may be the target, and then transmits these points or the target tracks obtained by tracking and filtering these points to the fusion center for fusion. This paper defines this method as multi-sensor fusion based on DBT (MSF-DBT). MSF-DBT has good performance under high signal-to-noise ratio, but when the signal-to-noise ratio is low, it is often difficult for the target to pass the single-cycle threshold detection, resulting in information loss of local nodes, resulting in a serious decline in algorithm performance. Based on the traditional networked radar detection method, the traditional method still has many shortcomings. First, due to the heterogeneity of networked radars, it becomes a difficult problem to uniformly represent different radar systems. Secondly, in the networked radar, the impact of the interfered nodes is not taken into account, resulting in reduced detection accuracy when there are too many nodes in the networked radar. Finally, the traditional adjacency matrix is ​​predetermined based on the Euclidean metric between nodes. This measurement method greatly reduces the robustness of the algorithm. Summary of the invention

[0003] The present invention aims to overcome the above-mentioned shortcomings of the prior art and provide a networked radar anomaly detection method and system based on graph neural network. The networked radar anomaly detection method can quickly detect the position of radar nodes in an abnormal state in the networked radar in real time and has strong robustness.

[0004] The second purpose of the present invention is to address the poor robustness of traditional measurement methods and provide a method based on an adaptive adjacency matrix to perform graph association on networked radars.

[0005] In order to achieve the above purpose, the present invention adopts the following technical solutions:

[0006] A networked radar anomaly detection method based on a graph neural network, the networked radar anomaly detection method comprising the following steps:

[0007] Step 1: Record the track detected by the radar, improve the target track route detected by the radar, and determine the track results observed by each radar in the networked radar.

[0008] Step 2: Calculate the attention value of the track observed by each radar node, and adaptively determine the adjacency matrix between nodes, so as to use the adjacency matrix as one of the features of the networked radar.

[0009] Step 3: Use the graph neural network to predict the node status at the next moment by combining the observation results of each node in the networked radar with the adjacency matrix between radar nodes.

[0010] Step 4: Compare the node track information predicted by each radar node at the next moment with the track result observed by the radar at the next moment to determine the radar node in the networked radar that is in an abnormal state.

[0011] Among them, the specific process of obtaining the track of each radar node in the networked radar by observing the target in step 1 includes:

[0012] 1.1. The trajectory of target motion in the networked radar is simulated to construct a target motion data set, wherein the target motion model includes three target motion conditions: uniform linear motion, uniformly accelerated linear motion, and variable accelerated linear motion.

[0013] 1.2. Construct a networked radar observation data set and set different interference situations in the networked radar. For example, for a six-node radar, there are 64 interference situations. The networked radar in each interference situation randomly observes different target movements to generate a radar observation data set.

[0014] The specific implementation process of step 2 includes:

[0015] 2.1. Encode the track features into vector representations suitable for input into the attention mechanism. Use a convolutional neural network to process time series data and extract key features. Standardize the track features to ensure that the range of feature values ​​is uniform, thereby avoiding computational instability caused by some feature values ​​being too large or too small in the attention calculation.

[0016] 2.2. Use the attention mechanism to calculate the attention value between nodes, that is, the similarity weight between each radar node. Use the fully connected layer to convert the features of each node into Query, Key and Value vectors.

[0017] And calculate the attention value between nodes.

[0018]

[0019] 2.3. Determine the adjacency relationship between radar nodes based on the attention value, directly generate the adjacency matrix based on the attention weight matrix, and update the weight of the generated attention value through back propagation during the training process, so that the adjacency matrix can dynamically reflect the changes in the relationship between nodes.

[0020] The specific implementation process of step 3 includes:

[0021] 3.1. At each time step, the observation features of the radar node and the corresponding adjacency matrix are input into the graph neural network.

[0022] 3.2. Through the graph convolution layer and the graph attention layer, the adjacency matrix is ​​used to transfer and fuse information between nodes and update the feature representation of each node.

[0023] 3.3. Let the network output the predicted trajectory of each node at the next moment. These prediction results are based on the relationship between nodes and the current observation information.

[0024] Wherein, step 4 specifically includes:

[0025] 4.1. Obtain the actual track information of the target that each radar node will observe at the next moment.

[0026] 4.2. Compare the predicted track information of each radar node with the actual observed track information, calculate the error between the two, and use RMSE to calculate the error.

[0027]

[0028] 4.3. According to the tolerance of the system, set an error threshold. If the prediction error of a radar node exceeds this threshold, the node may be in an abnormal state.

[0029] 4.4. Feedback the anomaly detection results to the GNN model for further optimization of the model. The ability to identify anomalies can be enhanced in subsequent training.

[0030] A second aspect of the present invention relates to a radar anomaly detection system for a networked radar based on a dynamic graph, comprising:

[0031] The acquisition module records the track detected by the radar, improves the target track route detected by the radar, and determines the track results observed by each radar in the network radar;

[0032] The adjacency matrix generation module calculates the attention value of the track observed by each radar node and adaptively determines the adjacency matrix between nodes, so as to use the adjacency matrix as one of the characteristics of the networked radar;

[0033] The prediction module uses the graph neural network to predict the node status at the next moment based on the observation results of each node in the networked radar and the adjacency matrix between radar nodes;

[0034] The abnormality detection module compares the node track information predicted by each radar node at the next moment with the track result observed by the radar at the next moment to determine the radar node in the networked radar that is in an abnormal state.

[0035] In summary, the solution proposed by the present invention has the following advantages:

[0036] The present invention observes the target trajectory at different moments by each radar node in the networked radar, realizes joint processing between multiple types of radars of different systems and modes, and has good scalability; and adaptively generates a corresponding adjacency matrix for the networked radar according to the similarity between the nodes, predicts the trajectory measurement value of the target observed by each radar node in the networked radar node at the next moment; by comparing the trajectory measurement value predicted by each radar node at the next moment with the trajectory measurement value observed by the corresponding radar node at the next moment, the radar nodes in the networked radar that are in an abnormal state are detected in real time, and the abnormal data is detected more quickly, so that processing and response can be performed more timely; the present invention utilizes a graph attention mechanism to enhance the spatial connection between radar nodes and the sensitivity to anomalies. The graph attention mechanism always pays attention to the correlation between each node. When the correlation between the nodes decreases, the abnormal node can be found, thereby reducing the importance of the node, thereby avoiding the networked radar from being further disturbed by error information, thereby performing more accurate anomaly detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 The present invention is a flow chart of a radar anomaly detection method for a networked radar based on a dynamic graph. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0040] Example 1

[0041] This embodiment provides a radar anomaly detection method for a networked radar based on a dynamic graph. Figure 1 The networked radar anomaly detection method comprises the following steps:

[0042] Step 1: Record the track detected by the radar, improve the target track route detected by the radar, and determine the track results observed by each radar in the networked radar.

[0043] By simulating the track observed by the radar node of the networked radar, the track is simulated. According to the traditional motion mode, we simulate the motion conditions of three targets: uniform linear motion, uniformly accelerated linear motion and variable accelerated linear motion. Let each node of the networked radar randomly observe the same target each time. The specific process of obtaining the track of each radar node in the networked radar when observing the target in this step includes:

[0044] 1.1. The trajectory of target motion in the networked radar is simulated to construct a target motion data set, wherein the target motion model includes three target motion conditions: uniform linear motion, uniformly accelerated linear motion, and variable accelerated linear motion.

[0045] 1.2. Construct a networked radar observation data set and set different interference situations in the networked radar. For example, for a six-node radar, there are 64 interference situations. The networked radar in each interference situation randomly observes different target movements to generate a radar observation data set.

[0046] Step 2: Calculate the attention value of the track observed by each radar node, and adaptively determine the adjacency matrix between nodes, so as to use the adjacency matrix as one of the features of the networked radar.

[0047] Through the characteristics of the radar itself, when observing the target, the observation result is the track characteristics of the target within a period of time. By inputting the aircraft characteristics into the attention mechanism, the attention coefficient between the nodes is calculated, and then the adjacency matrix of the networked radar is determined. The specific implementation process of this step includes:

[0048] 2.1. Encode the track features into vector representations suitable for input into the attention mechanism. Use a convolutional neural network (CNN) to process time series data and extract key features. Standardize the track features to ensure that the range of feature values ​​is uniform, thereby avoiding computational instability caused by some feature values ​​being too large or too small in the attention calculation.

[0049] 2.2. Use the attention mechanism to calculate the attention value between nodes, that is, the similarity weight between each radar node. Use the fully connected layer to convert the features of each node into Query, Key and Value vectors.

[0050] And calculate the attention value between nodes.

[0051]

[0052] 2.3. Determine the adjacency relationship between radar nodes based on the attention value, directly generate the adjacency matrix based on the attention weight matrix, and update the weight of the generated attention value through back propagation during the training process, so that the adjacency matrix can dynamically reflect the changes in the relationship between nodes.

[0053] Step 3: Use the graph neural network to predict the node status at the next moment by combining the observation results of each node in the networked radar with the adjacency matrix between radar nodes.

[0054] By taking advantage of the fact that graph neural networks are sensitive to adjacency matrices, graph neural networks are used to predict the target trajectory of the networked radar at the next moment. The specific implementation process of this step includes:

[0055] 3.1. At each time step, the observation features of the radar node and the corresponding adjacency matrix are input into the graph neural network.

[0056] 3.2. Through the graph convolution layer and the graph attention layer, the adjacency matrix is ​​used to transfer and fuse information between nodes and update the feature representation of each node.

[0057] 3.3. Let the network output the predicted trajectory of each node at the next moment. These prediction results are based on the relationship between nodes and the current observation information.

[0058] Step 4: Compare the node track information predicted by each radar node at the next moment with the track result observed by the radar at the next moment to determine the radar node in the networked radar that is in an abnormal state.

[0059] 4.1. Obtain the actual track information of the target that each radar node will observe at the next moment.

[0060] 4.2. Compare the predicted track information of each radar node with the actual observed track information, calculate the error between the two, and use RMSE to calculate the error.

[0061]

[0062] 4.3. According to the tolerance of the system, set an error threshold. If the prediction error of a radar node exceeds this threshold, the node may be in an abnormal state.

[0063] 4.4. Feedback the anomaly detection results to the GNN model for further optimization of the model. The ability to identify anomalies can be enhanced in subsequent training.

[0064] Example 2

[0065] This embodiment provides a networked radar anomaly detection system based on a dynamic graph for implementing the networked radar anomaly detection method based on a dynamic graph described in Embodiment 1. A networked radar anomaly detection system based on a dynamic graph includes:

[0066] The acquisition module records the track detected by the radar, improves the target track route detected by the radar, and determines the track results observed by each radar in the network radar;

[0067] The adjacency matrix generation module calculates the attention value of the track observed by each radar node and adaptively determines the adjacency matrix between nodes, so as to use the adjacency matrix as one of the characteristics of the networked radar;

[0068] The prediction module uses the graph neural network to predict the node status at the next moment based on the observation results of each node in the networked radar and the adjacency matrix between radar nodes;

[0069] The abnormality detection module compares the node track information predicted by each radar node at the next moment with the track result observed by the radar at the next moment to determine the radar node in the networked radar that is in an abnormal state.

[0070] Furthermore, the acquisition module includes:

[0071] The simulation submodule simulates the trajectory of the target movement in the networked radar and constructs a target movement data set, wherein the target movement model includes three target movement conditions: uniform linear motion, uniformly accelerated linear motion, and variable accelerated linear motion.

[0072] Furthermore, the adjacency matrix generation module includes:

[0073] The feature extraction submodule encodes the track features into vector representations suitable for input into the attention mechanism. It uses a convolutional neural network (CNN) to process time series data and extract key features, and standardizes the track features to ensure that the range of feature values ​​is uniform, thereby avoiding computational instability caused by some feature values ​​being too large or too small in the attention calculation;

[0074] The attention value calculation submodule uses the attention mechanism to calculate the attention value between nodes, that is, the similarity weight between each radar node. The fully connected layer is used to convert the features of each node into Query, Key and Value vectors. And the attention value between nodes is calculated;

[0075] Furthermore, the prediction module includes:

[0076] The network update submodule uses the graph convolution layer and the graph attention layer to transfer and fuse information between nodes using the adjacency matrix to update the feature representation of each node;

[0077] Furthermore, the anomaly detection module includes:

[0078] The loss update submodule compares the predicted track information of each radar node with the actual observed track information, calculates the error between the two, and uses RMSE to calculate the error;

[0079] The threshold determination submodule sets an error threshold according to the system tolerance. If the prediction error of a radar node exceeds this threshold, the node may be in an abnormal state.

[0080] The principles, formulas and parameter definitions involved in the above embodiments are all applicable and will not be reviewed one by one here.

[0081] Please note that the technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification. The above embodiments only express several implementation methods of the present application, and their descriptions are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, without departing from the concept of the present application, several variations and improvements can be made, which all belong to the scope of protection of the present application. Therefore, the scope of protection of the patent in this application shall be based on the attached claims.

Claims

1. A networked radar anomaly detection method based on graph neural network, characterized in that: The steps include: Step 1: Record the track of the target detected by the radar, improve the target track route detected by the radar, and determine the track result observed by each radar in the network radar; By simulating the trajectory of the target observed by the radar nodes of the networked radar, we simulate the trajectory. Based on the traditional motion mode, we simulate the motion of three targets: uniform linear motion, uniformly accelerated linear motion, and variable accelerated linear motion, so that each node of the networked radar randomly observes the same target each time; Step 2: Calculate the attention value of the track observed by each radar node, and adaptively determine the adjacency matrix between the nodes, so as to use the adjacency matrix as one of the features of the networked radar; Through the characteristics of the radar itself, when observing a target, the observation result is the track characteristics of the target within a period of time. By inputting the aircraft characteristics into the attention mechanism, the attention coefficient between the nodes is calculated, and then the adjacency matrix of the networked radar is determined; Step 3: The observation results of each node in the networked radar and the adjacency matrix between radar nodes are used to predict the node status at the next moment through the graph neural network; Taking advantage of the fact that graph neural networks are sensitive to adjacency matrices, graph neural networks are used to predict the target trajectory of the networked radar at the next moment. Step 4: Compare the node track information predicted by each radar node at the next moment with the track result observed by the radar at the next moment to determine the radar node in the networked radar that is in an abnormal state.

2. A networked radar anomaly detection method based on graph neural network as claimed in claim 1, characterized in that: Step 1: The specific process of obtaining the track of each radar node in the networked radar by observing the target includes: 1.

1. Simulate the trajectory of target motion in the networked radar to construct a target motion data set, wherein the target motion model includes three target motion conditions: uniform linear motion, uniformly accelerated linear motion, and variable accelerated linear motion; 1.

2. Construct a networked radar observation data set and set different interference situations in the networked radar. For example, for a six-node radar, there are 64 interference situations. The networked radar in each interference situation randomly observes different target movements to generate a radar observation data set.

3. The method for detecting anomalies of a networked radar based on a graph neural network according to claim 1, characterized in that: The specific implementation process of step 2 includes: 2.

1. Encode the track features into vector representations suitable for input into the attention mechanism; use convolutional neural networks to process time series data and extract key features, standardize the track features to ensure that the range of feature values ​​is unified, thereby avoiding computational instability caused by some feature values ​​being too large or too small in the attention calculation; 2.

2. Use the attention mechanism to calculate the attention value between nodes, that is, the similarity weight between each radar node; use the fully connected layer to convert the features of each node into Query, Key and Value vectors; and calculate the attention value between nodes; 2.

3. Determine the adjacency relationship between radar nodes based on the attention value, directly generate the adjacency matrix based on the attention weight matrix, and update the weight of the generated attention value through back propagation during the training process, so that the adjacency matrix can dynamically reflect the changes in the relationship between nodes.

4. The method for detecting anomalies of a networked radar based on a graph neural network according to claim 1, characterized in that: The specific implementation process of step 3 includes: 3.

1. At each time step, the observation features of the radar node and the corresponding adjacency matrix are input into the graph neural network; 3.

2. Through the graph convolution layer and the graph attention layer, the adjacency matrix is ​​used to transfer and fuse information between nodes and update the feature representation of each node; 3.

3. Let the network output the predicted trajectory of each node at the next moment. These prediction results are based on the relationship between nodes and the current observation information.

5. The method for detecting anomalies of a networked radar based on a graph neural network according to claim 1, characterized in that: Step 4 specifically includes: 4.

1. Obtain the actual track information of the target that each radar node will observe at the next moment; 4.

2. Compare the predicted track information of each radar node with the actual observed track information, calculate the error between the two, and use RMSE to calculate the error; 4.

3. According to the tolerance of the system, set an error threshold; if the prediction error of a radar node exceeds this threshold, the node may be in an abnormal state; 4.

4. Feedback the anomaly detection results to the GNN model for further optimization of the model; the ability to identify abnormal situations can be enhanced in subsequent training.

6. A radar anomaly detection system based on a dynamic graph network radar, comprising: The acquisition module records the track detected by the radar, improves the target track route detected by the radar, and determines the track results observed by each radar in the network radar; The adjacency matrix generation module calculates the attention value of the track observed by each radar node and adaptively determines the adjacency matrix between nodes, so as to use the adjacency matrix as one of the characteristics of the networked radar; The prediction module uses the graph neural network to predict the node status at the next moment based on the observation results of each node in the networked radar and the adjacency matrix between radar nodes; The abnormality detection module compares the node track information predicted by each radar node at the next moment with the track result observed by the radar at the next moment to determine the radar node in the networked radar that is in an abnormal state.

7. The radar anomaly detection system based on dynamic graph networked radar according to claim 1, characterized in that: The acquisition module comprises: The simulation submodule simulates the trajectory of the target movement in the networked radar and constructs a target movement data set, wherein the target movement model includes three target movement conditions: uniform linear motion, uniformly accelerated linear motion, and variable accelerated linear motion.

8. The radar anomaly detection system based on dynamic graph networked radar according to claim 1, characterized in that: The adjacency matrix generation module comprises: The feature extraction submodule encodes the track features into vector representations suitable for input into the attention mechanism; uses a convolutional neural network (CNN) to process time series data and extract key features, and standardizes the track features to ensure that the range of feature values ​​is uniform, thereby avoiding computational instability caused by some feature values ​​being too large or too small in the attention calculation; The attention value calculation submodule uses the attention mechanism to calculate the attention value between nodes, that is, the similarity weight between each radar node; uses the fully connected layer to convert the features of each node into Query, Key and Value vectors; and calculates the attention value between nodes.

9. The radar anomaly detection system based on dynamic graph networked radar according to claim 1, characterized in that: The prediction module comprises: The network update submodule uses the graph convolution layer and the graph attention layer to transfer and fuse information between nodes using the adjacency matrix to update the feature representation of each node.

10. The radar anomaly detection system based on dynamic graph networked radar according to claim 1, characterized in that: The anomaly detection module comprises: The loss update submodule compares the predicted track information of each radar node with the actual observed track information, calculates the error between the two, and uses RMSE to calculate the error; The threshold determination submodule sets an error threshold according to the system tolerance; if the prediction error of a radar node exceeds this threshold, the node may be in an abnormal state.

Images