Radar tracking performance calculation method based on graph neural network
Through the radar tracking performance deviation model based on graph neural network, a relationship diagram affecting the parameter difference factors is constructed, which solves the problem that traditional methods are difficult to reflect complex dynamic characteristics, and achieves more accurate radar tracking performance evaluation and more efficient experimental data use.
Patent Information
- Application Number
- CN202411006988.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-25
- Publication Date
- 2024-11-12
AI Technical Summary
Traditional radar tracking performance evaluation methods are difficult to fully reflect complex dynamic characteristics and subtle changes, especially when training data is limited, it is difficult to accurately construct the mapping between influencing factors and tracking performance.
Using a graph neural network-based method, a performance deviation model is constructed, including feature embedding, graph structure learning, graph attention aggregation and multi-layer perceptron calculation. By constructing a relationship diagram between the differential factors of the influence parameter, distance measurement and vector similarity are used. The loss function of the degree metric, train the model to improve the accuracy of performance bias prediction.
It improves the accuracy of radar tracking performance evaluation and the accuracy of model calculation performance deviations, reduces test costs, and improves the utilization efficiency of historical test data.
Smart Images

Figure CN118938145A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of radar tracking, and specifically is a radar tracking performance estimation method based on graph neural network. Background Art
[0002] Radar tracking technology is one of the key technologies in modern military, aviation and other fields. It is of great significance to achieve accurate detection, positioning and tracking of targets. In radar tracking systems, tracking performance evaluation methods play a vital role. They play an irreplaceable role in measuring the performance of radar tracking systems, optimizing system parameters, and improving tracking accuracy. Traditional radar tracking performance evaluation methods mainly rely on statistical analysis and error metrics. Although these methods can provide a certain degree of performance evaluation, they often cannot fully reflect the complex dynamic characteristics and subtle changes in the tracking process. Deep learning, as a powerful machine learning method, can learn the target's motion pattern, the interference characteristics of the environment, and the performance characteristics of the radar system from a large amount of radar tracking data, so as to more accurately evaluate the tracking performance. However, how to use deep learning models to construct the mapping key between influencing factors and tracking performance, especially when training data is limited, still needs further exploration. Summary of the invention
[0003] The present invention proposes a radar tracking performance estimation method based on graph neural network, in order to make full use of the relationship between radar difference factors and performance deviation to construct a performance deviation prediction model, so as to accurately predict the performance deviation of the radar under current test conditions and historical conditions, and thereby estimate the tracking performance of the radar, thereby improving the performance evaluation accuracy.
[0004] The technical solution to realize the present invention is: a radar tracking performance estimation method based on graph neural network, the steps are as follows:
[0005] Step 1: Analyze radar technical parameters and construct input vector.
[0006] Step 2: Construct a performance deviation model based on graph neural network, including a feature embedding module, a graph structure learning module, a graph attention aggregation module, a multi-layer perceptron calculation module, and a performance deviation prediction module based on a fully connected neural network.
[0007] Step 3: Use the input vector to train the radar tracking performance deviation model based on the graph neural network to obtain the optimal radar tracking performance deviation model.
[0008] Step 4: Use the optimal radar tracking performance deviation model to estimate the radar tracking performance under given test conditions.
[0009] Compared with the prior art, the present invention has the following significant advantages:
[0010] (1) The present invention uses a deep neural network to model the relationship between the difference factor and the tracking performance difference, rather than directly modeling the relationship between the influencing factor and the tracking performance, thereby reducing the size of the mapping space of the performance evaluation problem, thereby improving the fitting ability of the model and thus improving the accuracy of the performance evaluation;
[0011] (2) The present invention constructs a relationship graph between the difference factors of radar tracking influencing parameters and uses a graph neural network to aggregate information between different difference factors, so that the difference factors between the influencing parameters contain more information, thereby further improving the accuracy of the model calculation performance deviation;
[0012] (3) The present invention realizes the training and updating of the radar tracking performance deviation network between the current test sample and the historical test sample by constructing a loss function based on the fusion of two kinds of distance measurement and vector similarity measurement, so as to improve the sensitivity of the model to the influencing parameters, thereby improving the accuracy of the model calculation performance deviation;
[0013] (4) The present invention creatively proposes a radar tracking performance estimation method. Based on the evaluation results of historical test data, the present invention models the mapping relationship between the difference factors of the historical test sample conditions corresponding to the historical test data under the current test sample conditions and the tracking performance differences, thereby realizing the calculation of the radar tracking performance deviation between the current test sample conditions and the historical test sample conditions, thereby improving the utilization efficiency of the historical test data and reducing the test cost. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] Figure 1 This is a flow chart of the radar tracking performance estimation method based on graph neural network of the present invention.
[0015] Figure 2 This is a structural diagram of the performance deviation model based on graph neural network of the present invention. DETAILED DESCRIPTION
[0016] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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 creative work are within the scope of protection of the present invention.
[0017] The following will further introduce the specific implementation method, as well as the technical difficulties and inventive points of this invention in combination with this design example.
[0018] In this embodiment, combined with Figure 1The radar tracking performance estimation method based on graph neural network described in the present invention is to construct a mapping model between the difference of factors affecting radar tracking performance (i.e., difference factor) and the difference of tracking performance (i.e., performance deviation model) based on graph neural network, and to model the complex relationship between factors such as the mobility of the model target, the interference of the environment, and the parameter setting of the radar system and the radar tracking performance, so as to more accurately evaluate the radar tracking performance. The specific steps are as follows:
[0019] Step 1: Analyze radar technical parameters and construct input vector.
[0020] The key factors that affect the tracking results in the radar tracking process are analyzed from the description of the experimental sample X and the historical similar sample Y, from the four perspectives of radar, target, environment and interference.
[0021] Get the key parameter set V of radar anti-interference performance of any test sample X X = {v x1 ,v x2 ,…,v xa ,…,v xA} and a set of key parameters V of radar anti-interference performance of a historically similar sample Y Y = {v y1 ,v y2 ,…,v ya ,…,v yA}, where v xa represents the ath key parameter in the test sample X, v ya Represents historical similar test samples V Y The ath key parameter in ; A represents the total number of key parameters.
[0022] In this embodiment, the key parameters obtained from the radar perspective are v x1 =v y1 =Working system, v x2 =v y2 =Working mode, v x3 =v y3 = bandwidth, v x4 =v y4 = Anti-interference method, the key parameters obtained from the target perspective are v x5 =v y5 = target type, v x6 =v y6 = target speed, v x7 =v y7 = target starting position, the key parameters obtained from the environmental perspective are v x8 =v y8 = area, v x9 =v y9 = scene type, v x10=v y10 = Weather, the key parameters obtained from the interference perspective are v x11 =v y11 = frequency, v x12 =v y12 = power, v x13 =v y13 =Interference pattern, v x14 =v y14 =Interference bandwidth; The above key influencing parameters are grouped into a set V X With V Y , so A = 14; the total number of test samples is 3000, and the number of historical similar samples is the same as the number of test samples. 80% of the data set is used for training and the remaining 20% is used for testing.
[0023] Step 2: Construct Figure 2 The performance deviation model based on graph neural network shown in the figure includes: a feature embedding module, a graph structure learning module, a graph attention aggregation module, a multi-layer perceptron calculation module and a performance deviation prediction module based on a fully connected neural network.
[0024] Step 2.1: Feature embedding module for V X and V Y Vectorized embedding is performed, and the radar parameter feature embedding matrix M of the test sample X is obtained accordingly. X The radar parameter feature embedding matrix M of the historical similar test sample Y Y ; In this example, each row of the matrix represents the encoding of each feature, and the matrix dimension is 14×14.
[0025] Step 2.2: Calculate M in the graph structure learning module X The i-th row in the encoding feature M X (i) and M Y The jth row in the encoding feature M Y (j), thereby calculating the similarity S(i,j) between the encoded features in each row and forming a feature relationship graph F,i,j∈[1,K] with a dimension of K×K.
[0026] Step 2.3: The graph attention aggregation module extracts the feature representation h of the kth neighbor node of the i-th node ik , and use formula (1) to aggregate the feature representation h of the kth neighbor node of the i-th node in the feature relationship graph F ik , get the updated feature representation h' of the kth neighbor node of the i-th node ik , thus obtaining the updated feature representation h' of all neighbor nodes of the i-th node in the feature relationship graph F i , and then obtain the updated feature representations of all neighbor nodes of all nodes in the feature relationship graph F and form the feature difference representation matrix H:
[0027]
[0028] In formula (1), N i represents the set of neighbor node numbers of the i-th node. In this example, the number of sets is 3000. ij is the attention weight between the i-th node and its k-th neighbor node, σ is the activation function, and W is the weight matrix in the graph attention aggregation module.
[0029] Step 2.4: After H is normalized, it is input into the multi-layer perceptron computing module for processing to obtain the feature difference representation value D.
[0030] Step 2.5: The performance deviation prediction module based on the fully connected neural network processes D using formula (2) and outputs the predicted performance deviation Pre:
[0031] Pre=W pred ·D+b pred (2)
[0032] In formula (2), W pred is the weight matrix in the performance deviation prediction module, b pred is the bias vector in the performance deviation prediction module; in this embodiment, the number of neural network layers is 10.
[0033] Step 3: Use the input vector to train the radar tracking performance deviation model based on the graph neural network to obtain the optimal radar tracking performance deviation model, as follows:
[0034] Step 3.1, establish a loss function L1 based on distance measurement, a loss function L2 based on vector similarity measurement, and a fusion loss function L;
[0035]
[0036] L=αL1+(1-α)L2 (5)
[0037] In formula (3), N is the number of samples. In this example, the number of samples is 3000. is the performance deviation of the i-th node predicted by the model, y i is the actual performance deviation of the ith node. is the predicted performance deviation value, y is the actual performance deviation value; in formula (5), α is the weight value.
[0038] The present invention needs to define the output of the model and the actual performance deviation. According to step 2, the output of the model is the predicted performance deviation after being processed by the fully connected neural network, and the actual performance deviation can be obtained from the test sample data.
[0039] Step 3.2: Use the gradient descent method to train the radar tracking performance deviation model based on the graph neural network, and calculate the fusion loss function L to update the network parameters until the fusion loss function L converges, thereby obtaining the optimal radar tracking performance deviation model;
[0040] Step 4: Use the optimal radar tracking performance deviation model to estimate the radar tracking performance under given test conditions, as follows:
[0041] Step 4.1: Obtain the radar anti-interference performance key parameter set V of the test sample Z to be predicted Z The key parameter set V' of radar anti-interference performance of similar historical samples Z And input into the optimal radar tracking performance deviation model for processing to obtain the predicted performance deviation X p .
[0042] Step 4.2: Calculate the radar tracking performance under a given test sample using formula (4):
[0043] F z =F' z +X p (6)
[0044] In formula (6), F' z is the radar tracking performance of historical similar samples of the test sample Z.
Claims
1. A radar tracking performance estimation method based on graph neural network, It is characterized in that Here are the steps: Step 1: Analyze radar technical parameters and construct input vector; Step 2: Construct a performance deviation model based on graph neural network, including a feature embedding module, a graph structure learning module, a graph attention aggregation module, a multi-layer perceptron calculation module, and a performance deviation prediction module based on a fully connected neural network; Step 3: Use the input vector to train the radar tracking performance deviation model based on the graph neural network to obtain the optimal radar tracking performance deviation model; Step 4: Use the optimal radar tracking performance deviation model to estimate the radar tracking performance under given test conditions.
2. According to the radar tracking performance estimation method based on graph neural network according to claim 1, It is characterized in that In step 1, the radar technical parameters are analyzed and the input vector is constructed; Get the key parameter set V of radar anti-interference performance of any test sample X X = {v x1 ,v x2 ,…,v xa ,…,v xA } and a set of key parameters V of radar anti-interference performance of a historically similar sample Y Y = {v y1 ,v y2 ,…,v ya ,…,v yA }, where v xa represents the ath key parameter in the test sample X, v ya Represents historical similar test samples V Y The ath key parameter in ; A represents the total number of key parameters.
3. According to the radar tracking performance estimation method based on graph neural network according to claim 2, It is characterized in that In step 2, a performance deviation model based on graph neural network is constructed, including a feature embedding module, a graph structure learning module, a graph attention aggregation module, a multi-layer perceptron calculation module, and a performance deviation prediction module based on a fully connected neural network, as follows: Step 2.1: The feature embedding module performs X and V Y Vectorized embedding is performed, and the radar parameter feature embedding matrix M of the test sample X is obtained accordingly. X The radar parameter feature embedding matrix M of the historical similar test sample Y Y ; Step 2.2: Calculate M in the graph structure learning module X When M X (i) is the encoding feature of the i-th row of the radar parameter feature embedding matrix, M Y (j) is the radar parameter feature embedding matrix M of historical similar test samples Y The j-th row encoding feature is calculated through the graph structure learning module M X (i) and M Y (j), so as to construct a feature relationship graph F, i, j ∈ [1, K] with a dimension of K × K; Step 2.3: In the graph attention aggregation module, h ik is the feature representation of the kth neighbor node of the i-th node in the feature relationship graph F, and the updated feature representation h' of the kth neighbor node of the i-th node is obtained using formula (1): ik : In formula (1), N i Represents the neighbor node number set of the i-th node in the feature relationship graph F, αi j is the attention weight between the i-th node and its k-th neighbor node in the feature relationship graph F, σ is the activation function, and W is the weight matrix in the graph attention aggregation module; Thus, the updated feature representations h'i of all neighboring nodes of the i-th node in the feature relationship graph F are obtained, and then the updated feature representations of all neighboring nodes of all nodes in the feature relationship graph are obtained to form a feature difference representation matrix H; Step 2.4, after normalizing H, input it into the multi-layer perceptron computing module for processing to obtain the feature difference representation value D; Step 2.5: The performance deviation prediction module processes D using formula (2) to obtain the predicted performance deviation Pre, that is, the predicted value of the radar tracking performance deviation under the current test sample conditions and the historical test sample conditions: Pre=Wpred·D+bpred(2) In formula (2), Wpred is the weight matrix in the performance deviation prediction module, and bpred is the bias vector in the performance deviation prediction module.
4. According to the radar tracking performance estimation method based on graph neural network according to claim 3, It is characterized in that In step 3, the input vector is used to train the radar tracking performance deviation model based on the graph neural network to obtain the optimal radar tracking performance deviation model: Step 3.1, establish the loss function L1 based on distance measurement, the loss function L2 based on vector similarity measurement, and the fusion loss function L: L=αL1+(1-α)L2(5) in, is the predicted performance deviation of the ith node in the predicted performance deviation Pre, yi is the actual performance deviation of the ith node, and N is the number of nodes; is the predicted performance deviation value, y is the actual performance deviation value; α is the weight value; Step 3.2: Use the gradient descent method to train the radar tracking performance deviation model based on the graph neural network, and calculate the fusion loss function L to update the radar tracking performance deviation model parameters until the fusion loss function L converges, thereby obtaining the optimal radar tracking performance deviation model.
5. According to the radar tracking performance estimation method based on graph neural network according to claim 4, It is characterized in that In step 4, the optimal radar tracking performance deviation model is used to estimate the radar tracking performance under given test conditions: Step 4.1: Obtain the radar anti-interference performance key parameter set V of the test sample Z to be predicted Z The key parameter set V' of radar anti-interference performance of similar historical samples Z , input them into the optimal radar tracking performance deviation model to obtain the predicted performance deviation Xp; Step 4.2: Calculate the radar tracking performance F of the test sample Z using formula (6): z : Fz=F'z+Xp(6) Wherein, F'z is the radar tracking performance of the historical similar samples of the test sample Z, and F'z is calculated and evaluated by the test data corresponding to the test sample Z.
Citation Information
Cited By
Radar working mode identification method and system based on graph neural network
CN120522643A
