A fast and accurate prediction method and system for RCS sequence based on target characteristic mode

Abstract

The present invention relates to a method and system for quickly and accurately predicting an RCS sequence based on a target characteristic mode, the method comprising: constructing a prediction network based on motion parameter data of a time series, using a trained model to predict the motion parameters of a target in the future; extracting the characteristic values ​​and characteristic modes inherent to the target; using the target inherent characteristic modes to quickly and accurately obtain the electromagnetic response of the target in the time series through linear reconstruction. The present invention realizes the rapid and accurate prediction of the target RCS sequence by constructing a motion parameter prediction network based on a time series and combining it with the characteristic mode linear reconstruction.

Description

Technical Field

[0001] The invention belongs to the technical field of radar stealth, and in particular relates to a method and system for quickly and accurately predicting an RCS sequence based on a target characteristic mode. Background Art

[0002] The radar cross-section (RCS) of a target is a key parameter that characterizes the scattering characteristics of a target under radar wave irradiation. Its accurate calculation is of great significance for early warning radar systems, navigation radar systems, tracking radar systems, target detection and identification, and stealth technology evaluation.

[0003] A variety of mature and distinctive methods have been developed for the calculation of target RCS sequences, which are suitable for different application scenarios and computing requirements. The first is the high-frequency asymptotic method based on the approximate solution of the wave equation of electromagnetic waves under high-frequency conditions, which estimates the RCS sequence by tracing the ray path and calculating the reflection, diffraction and other effects, including physical optics and geometric optics. Although the physical optics and geometric optics methods have high computational efficiency, the calculation results are not accurate enough when dealing with complex targets. The second is the full-wave simulation method that relies on Maxwell's equations or their equivalent forms to obtain the scattering field of the target under electromagnetic wave irradiation. It includes differential equation method and integral equation method: the integral equation method is mainly based on the moment method, which requires the target to be discretized, and the characteristic information of the current distribution, far field and near field of the target is obtained by matrix inversion. Its calculation accuracy is very high, but the calculation complexity is high; the differential equation law includes the finite element method and the finite difference method. This method discretizes the entire calculation domain and takes the field in space as the unknown quantity, which has extremely high requirements for storage and computing power. As an emerging electromagnetic analysis method, the characteristic mode method can not only fully reflect the key characteristics of the antenna such as the resonant frequency and radiation current, but is also widely used in the design and optimization of antennas such as monopoles and metasurfaces. It can also be used to analyze the electromagnetic scattering characteristics of targets and the synthesis of radar patterns. This study linearly reconstructs the induced current and electric field through the characteristic mode inherent in the target, and realizes the rapid and accurate calculation of the RCS sequence during the target movement. Summary of the invention

[0004] The object of the present invention is to provide a method and system for quickly and accurately predicting an RCS sequence based on a target characteristic mode, which can achieve fast and accurate prediction of the target RCS sequence.

[0005] The technical solution to achieve the purpose of the present invention is: a method for quickly and accurately predicting an RCS sequence based on a target characteristic mode, and the specific steps are as follows:

[0006] Step 1, target motion parameter prediction: construct a motion parameter prediction network based on time series, model the motion trajectory of the target, consider the propagation characteristics of the radar signal and the motion state of the target, and use the trained model to predict the motion parameters of the target in the future; the motion parameters include center position coordinates, pitch angle, azimuth angle and roll angle;

[0007] Step 2, static feature extraction: Based on the structural information of the target and the equivalent principle, the scattered field of the target is equivalent to the field generated by the equivalent source on the surface of the ideal conductor in free space and the impedance matrix is ​​obtained; the weighted equation is obtained according to the generalized eigenvalue equation, and the corresponding eigenvalue and eigenmode vector are solved;

[0008] Step 3: Dynamic feature reconstruction: According to the predicted target motion parameters, extracted eigenvalues ​​and characteristic patterns, the induced current on the target surface is linearly superimposed with a finite number of characteristic currents, the characteristic electric field is calculated by the characteristic current, and the characteristic electric field is linearly superimposed as a total field to reconstruct the RCS sequence of the target.

[0009] A fast and accurate prediction system for RCS sequence based on target characteristic mode, comprising:

[0010] The motion parameter prediction module builds a motion parameter prediction network based on time series, models the motion trajectory of the target, considers the propagation characteristics of the radar signal and the motion state of the target, and uses the trained model to predict the motion parameters of the target in the future; the motion parameters include the center position coordinates, pitch angle, azimuth angle and roll angle;

[0011] The static feature extraction module uses the target's structural information and the equivalent principle to equate the target's scattered field to the field generated by an equivalent source on the surface of an ideal conductor in free space and obtain the impedance matrix; the weighted equation is obtained based on the generalized eigenvalue equation to solve the corresponding eigenvalue and eigenmode vector;

[0012] The dynamic feature reconstruction module performs linear superposition of a finite number of characteristic currents on the induced current on the target surface according to the predicted target motion parameters, extracted eigenvalues ​​and characteristic patterns, calculates the characteristic electric field through the characteristic current, and linearly superimposes the characteristic electric field into a total field to reconstruct the RCS sequence of the target.

[0013] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the steps of the above method are implemented when the processor executes the program.

[0014] A computer-readable storage medium stores a computer program, which implements the steps of the above method when executed by a processor.

[0015] Compared with the prior art, the significant advantages of the present invention are as follows: the present invention provides a method and system for quickly and accurately predicting the target RCS sequence based on the target characteristic mode, and the prediction network can accurately predict the motion parameter values ​​of the target during flight. Due to the orthogonality of the target characteristic mode, the use of orthogonal characteristic modes when calculating the target RCS sequence can avoid unnecessary repeated calculations in the calculation process, and the error of one characteristic mode will not affect the calculation of other characteristic modes, thereby improving the efficiency and accuracy of the target RCS prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present invention.

[0017] Figure 1 It is a schematic diagram of the overall process of a device for quickly and accurately predicting a target RCS sequence based on a target characteristic mode provided in an embodiment of the present invention.

[0018] Figure 2 This is a comparison chart of the results of bidirectional LSTM predicting the target trajectory.

[0019] Figure 3 It is a comparison chart of the predicted target RCS sequence and the true target RCS sequence. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions 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 part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in 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.

[0021] The specific implementation of the above concept is described below.

[0022] Please refer to Figure 1 In an embodiment of the present invention, a method for quickly and accurately predicting an RCS sequence based on a target characteristic mode is provided, the method comprising:

[0023] Step 1: Target motion parameter prediction: construct a prediction network based on time series motion parameters (center position: x, y, z; pitch angle, azimuth angle, roll angle), model the target's motion trajectory, consider the propagation characteristics of the radar signal and the target's motion state, and use the trained model to predict the target's motion parameters in the future, thereby providing the necessary input for the subsequent calculation of the electromagnetic scattering field;

[0024] The motion parameter prediction described above first constructs a prediction model, which includes a standardization module to preprocess the input data so that all data are within a fixed range; the data is input into a three-layer bidirectional LSTM module with 12 hidden layer nodes, and an additional dropout regularization in the middle layer to avoid overfitting of the model, and finally the final prediction data is output through the fully connected layer. The motion parameters of the target at the historical moment are input into the model, and the motion parameters of the target at the future moment are output through the prediction model. The sigmoid activation function is used to convert the input linear combination x into an output in the range of (0,1) to prevent gradient explosion in deep networks, so that the network can learn complex data and nonlinear complex function mapping between input and output, which is crucial for capturing complex patterns and trends in time series.

[0025] The training of the prediction network includes the following steps:

[0026] The acquired target motion parameter sample data is divided into a training data set and a test data set;

[0027] All motion parameter data are normalized to the maximum and minimum values ​​so that their range is fixed between 0 and 1;

[0028] ;

[0029] Where R i is the motion parameter value of target i at time; R i * is the normalized result; R x and R m are the maximum and minimum values ​​of each motion parameter data respectively.

[0030] The target motion parameters in the training data set are input into a pre-built prediction network, a pre-built bidirectional LSTM prediction network, and the bidirectional LSTM network integrates the past and future information of the target motion parameter sequence data in an innovative way to achieve a bidirectional insight into the time series data. The prediction results are compared with the corresponding labels of each sample, and the network parameters of the prediction network are trained and adjusted until a prediction model that meets the requirements is obtained;

[0031] The normalized test data set is input into the trained prediction network to predict the target motion parameters at future moments and verify its accuracy.

[0032] Step 2: static feature extraction step. Based on the structural information of the target and the equivalent principle, the scattered field of the target is equivalent to the field generated by the equivalent source on the surface of the ideal conductor in free space and the impedance matrix is ​​obtained. The weighted equation is obtained according to the generalized eigenvalue equation, and the corresponding eigenvalue and eigenmode vector are solved;

[0033] The statically extracted characteristic mode first requires the structural information of the target. Based on the equivalent principle, the scattered field of the target is equivalent to the field generated by the equivalent source on the surface of the ideal conductor in free space, thereby obtaining the scattered electric field E s , and the impedance matrix Z is obtained by the moment method:

[0034]

[0035] in, Indicates the location of the field point, represents the scattered electric field, represents the imaginary unit, and denote the vector magnetic potential and scalar electric potential, respectively. and are respectively expressed as the magnetic permeability and dielectric constant of free space, Represents the angular frequency. represents the characteristic current, represents the impedance matrix, and are all real symmetric matrices, and the subscript n represents the nth eigenmode.

[0036] The weighted characteristic equation is obtained through the generalized eigenvalue equation:

[0037]

[0038] in, Represents the eigenvalue.

[0039] The eigenvalues ​​and eigenmode vectors corresponding to the coefficient matrix of the target are solved according to the equation.

[0040] Step 3: Dynamic feature reconstruction step: According to the predicted target motion parameters, extracted eigenvalues ​​and characteristic patterns, the induced current on the target surface is linearly superimposed with a finite number of characteristic currents, the characteristic electric field is calculated by the characteristic current, and the characteristic electric field is linearly superimposed as a total field to reconstruct the RCS sequence of the target.

[0041] To dynamically reconstruct the RCS sequence of the target, it is first necessary to approximate the induced current on the target surface as a finite term characteristic current superposition based on the statically extracted eigenvalues ​​and characteristic patterns and the predicted motion parameters:

[0042]

[0043] in, represents the induced current on the target surface, Indicates the number of truncated items, Represents the mode weight coefficient corresponding to each feature mode.

[0044] The mode weight coefficient is obtained by taking the inner product of the incident field tangent equation and the characteristic current:

[0045]

[0046] in, represents the tangential component of the incident electric field.

[0047] The characteristic electric field generated by the characteristic current is:

[0048]

[0049] in, represents the source location, represents the wave number, and represent polar angle and azimuth respectively.

[0050] Consistent with the characteristic current superposition principle, the target electric field is approximately expressed as a finite term characteristic electric field superposition:

[0051]

[0052] The RCS sequence value of the target is obtained according to the reconstructed electric field.

[0053] Based on the same inventive concept, the present invention also provides a system for quickly and accurately predicting an RCS sequence based on a target characteristic mode, which is used to implement the above prediction method. The system comprises:

[0054] The target motion parameter prediction module builds a prediction network for motion parameters based on time series. By modeling the motion trajectory of the target, considering the propagation characteristics of the radar signal and the motion state of the target, the trained model is used to predict the motion parameters of the target in the future, thereby providing the necessary input for the subsequent calculation of the electromagnetic scattering field.

[0055] The static feature extraction module uses the target's structural information and the equivalent principle to equate the target's scattered field to the field generated by an equivalent source on the surface of an ideal conductor in free space and obtain the impedance matrix. The weighted equation is obtained based on the generalized eigenvalue equation, and the corresponding eigenvalue and eigenmode vector are solved;

[0056] The dynamic feature reconstruction module performs linear superposition of a finite number of characteristic currents on the induced current on the target surface according to the predicted target motion parameters, extracted eigenvalues ​​and characteristic patterns, calculates the characteristic electric field through the characteristic current, and linearly superimposes the characteristic electric field into a total field to reconstruct the RCS sequence of the target.

[0057] The specific implementation methods of the above modules are the same as the aforementioned RCS sequence fast and accurate prediction method, which will not be repeated here.

[0058] The present invention is described in detail below in conjunction with embodiments and drawings.

[0059] Example

[0060] This embodiment sets up a model to predict the RCS sequence of three targets in motion at a frequency of 300MHz. There are 150 motion parameter samples in total, each sample has 200 motion points at different moments. The position and posture of the target are collected over time and have time series. The motion data of the first 80% of the historical moments are used to predict the motion data of the last 20% of the moments. Among them, 110 samples are used for training, and 40 samples are used to test the accuracy and generalization of the model. The final prediction results of the motion parameters of the three targets are as follows: Figure 2 As shown, the blue dotted line is the true motion parameter curve, and the red solid line is the predicted motion parameter curve.

[0061] Then, the predicted motion parameters are used to quickly and accurately calculate the RCS sequence during the motion process by dynamically reconstructing the induced current and electric field using the characteristic mode, and compared with the RCS sequence calculated by FEKO using the actual motion parameters. Since the distance between the targets during the motion process is relatively far, the influence of coupling can be ignored. Figure 3 The graph shows a comparison of three target actual RCS sequences and predicted RCS sequences, the blue solid line is the real RCS sequence value, the red dotted line is the predicted RCS sequence value, and the root mean square error between the two is 2.6145 dBsm. Table 1 shows the efficiency of RCS prediction in this embodiment.

[0062] Table 1 Efficiency statistics of predicting RCS

[0063]

[0064] from Figure 3 It can be seen that the present embodiment can accurately predict the RCS sequence of the target, and the root mean square error of the RCS is 2.6145 dBsm. According to the comparison in Table 1, it can be seen that the present embodiment has a certain improvement in efficiency in terms of time and memory overhead, and can quickly predict the RCS sequence of the target.

[0065] The above are only preferred embodiments of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications should also be considered as the protection scope of the present invention. All components not specified in this embodiment can be implemented using existing technologies.

Claims

1. A method for quickly and accurately predicting an RCS sequence based on a target characteristic mode, characterized in that: include: Step 1, target motion parameter prediction: construct a motion parameter prediction network based on time series, model the motion trajectory of the target, consider the propagation characteristics of the radar signal and the motion state of the target, and use the trained model to predict the motion parameters of the target in the future; the motion parameters include center position coordinates, pitch angle, azimuth angle and roll angle; Step 2, static feature extraction: Based on the structural information of the target and the equivalent principle, the scattered field of the target is equivalent to the field generated by the equivalent source on the surface of the ideal conductor in free space and the impedance matrix is ​​obtained; the weighted equation is obtained according to the generalized eigenvalue equation, and the corresponding eigenvalue and eigenmode vector are solved; Step 3: Dynamic feature reconstruction: According to the predicted target motion parameters, extracted eigenvalues ​​and characteristic patterns, the induced current on the target surface is linearly superimposed with a finite number of characteristic currents, and the characteristic electric field is calculated through the characteristic currents. The characteristic electric fields are linearly superimposed as the total field to reconstruct the RCS sequence of the target.

2. The method for rapid and accurate prediction of RCS sequence based on target characteristic mode according to claim 1, characterized in that: The motion parameter prediction described in step 1 first constructs a prediction model, which includes a standardization module to preprocess the input data so that all data are within a fixed range; the data is input into a three-layer bidirectional LSTM module with 12 hidden layer nodes, and the middle layer has additional dropout regularization to avoid overfitting of the model, and finally the final prediction data is output through the fully connected layer; the motion parameters of the target at the historical moment are input into the model, and the motion parameters of the target at the future moment are output through the prediction model; the output of each layer of the network is weighted and summed and a bias term is added, and the sigmoid activation function is used to convert the input linear combination x into an output in the range of (0,1); The training of the prediction network includes the following steps: Divide the acquired target motion parameter sample data into a training data set and a test data set; All motion parameter data are normalized to the maximum and minimum values ​​so that their range is fixed between 0 and 1; ； Where R i is the motion parameter value of target i at time; R i * is the normalized result; R x and R m are the maximum and minimum values ​​of each motion parameter data respectively; The target motion parameters in the training data set are input into a pre-built bidirectional LSTM prediction network, the prediction results are compared with the corresponding labels of each sample, and the network parameters of the prediction network are trained and adjusted until a prediction model that meets the requirements is obtained; The normalized test data set is input into the trained prediction network to predict the target motion parameters at future moments and verify its accuracy.

3. The method for rapid and accurate prediction of RCS sequence based on target characteristic mode according to claim 1, characterized in that: In step 2, the static feature extraction mode first requires the structural information of the target. Based on the equivalence principle, the scattered field of the target is equivalent to the field generated by the equivalent source on the surface of the ideal conductor in free space, thereby obtaining the scattered electric field E s , and the impedance matrix Z is obtained by the moment method: ； in, Indicates the location of the field point, represents the scattered electric field, represents the imaginary unit, and denote the vector magnetic potential and scalar electric potential, respectively. and are respectively expressed as the magnetic permeability and dielectric constant of free space, represents the angular frequency; represents the characteristic current, represents the impedance matrix, and are all real symmetric matrices, and the subscript n represents the nth eigenmode; The weighted characteristic equation is obtained through the generalized eigenvalue equation: ； in, represents the eigenvalue; The eigenvalues ​​and eigenmode vectors corresponding to the coefficient matrix of the target are solved according to the equation.

4. The method for rapid and accurate prediction of RCS sequence based on target characteristic mode according to claim 1, characterized in that: In step 3, the RCS sequence of the target is linearly reconstructed. First, the induced current on the target surface needs to be approximately expressed as a finite characteristic current superposition based on the statically extracted eigenvalues ​​and characteristic patterns and the predicted motion parameters: ； in, represents the induced current on the target surface, Indicates the number of truncated items, Indicates the mode weight coefficient corresponding to each characteristic mode; The mode weight coefficient is obtained by taking the inner product of the incident field tangent equation and the characteristic current: ； in, represents the tangential component of the incident electric field; The characteristic electric field generated by the characteristic current is: ； in, represents the source location, represents the wave number, and denote polar angle and azimuth respectively; Consistent with the characteristic current superposition principle, the target electric field is approximately expressed as a finite term characteristic electric field superposition: ； The RCS sequence value of the target is obtained according to the reconstructed electric field.

5. A fast and accurate prediction system for RCS sequence based on target characteristic mode, characterized in that: include: The target motion parameter prediction module builds a motion parameter prediction network based on time series, models the motion trajectory of the target, considers the propagation characteristics of the radar signal and the motion state of the target, and uses the trained model to predict the motion parameters of the target in the future; the motion parameters include the center position coordinates, pitch angle, azimuth angle and roll angle; The static feature extraction module uses the target's structural information and the equivalent principle to equate the target's scattered field to the field generated by an equivalent source on the surface of an ideal conductor in free space and obtain the impedance matrix; the weighted equation is obtained based on the generalized eigenvalue equation to solve the corresponding eigenvalue and eigenmode vector; The dynamic feature reconstruction module performs linear superposition of finite characteristic currents on the target surface based on the predicted target motion parameters, extracted eigenvalues ​​and characteristic patterns, and calculates the characteristic electric field through the characteristic current. The characteristic electric fields are linearly superimposed as the total field to reconstruct the RCS sequence of the target.

6. The system for rapid and accurate prediction of RCS sequence based on target characteristic mode according to claim 5, characterized in that: The motion parameter prediction described above first constructs a prediction model, which includes a standardization module to preprocess the input data so that all data are within a fixed range; the data is input into a three-layer bidirectional LSTM module, the number of hidden layer nodes is 12, and the middle layer has an additional dropout regularization to avoid overfitting of the model, and finally the final prediction data is output through a fully connected layer; the motion parameters of the target at the historical moment are input into the model, and the motion parameters of the target at the future moment are output through the prediction model; the output of each layer of the network is weighted and summed and a bias term is added, and the sigmoid activation function is used to convert the input linear combination x into an output in the range of (0,1) to prevent gradient explosion in the deep network; The training of the prediction network includes the following steps: Divide the acquired target motion parameter sample data into a training data set and a test data set; All motion parameter data are normalized to the maximum and minimum values ​​so that their range is fixed between 0 and 1; ； Where R i is the motion parameter value of target i at time; R i * is the normalized result; R x and R m are the maximum and minimum values ​​of each motion parameter data respectively; The target motion parameters in the training data set are input into a pre-built bidirectional LSTM prediction network, the prediction results are compared with the corresponding labels of each sample, and the network parameters of the prediction network are trained and adjusted until a prediction model that meets the requirements is obtained; The normalized test data set is input into the trained prediction network to predict the target motion parameters at future moments and verify its accuracy.

7. The system for rapid and accurate prediction of RCS sequence based on target characteristic mode according to claim 5, characterized in that: The static feature extraction first requires the structural information of the target. Based on the equivalent principle, the scattered field of the target is equivalent to the field generated by the equivalent source on the surface of the ideal conductor in free space, thereby obtaining the scattered electric field E s , and the impedance matrix Z is obtained by the moment method: ； in, Indicates the location of the field point, represents the scattered electric field, represents the imaginary unit, and denote the vector magnetic potential and scalar electric potential, respectively. and are respectively expressed as the magnetic permeability and dielectric constant of free space, represents the angular frequency; represents the characteristic current, represents the impedance matrix, and are all real symmetric matrices, and the subscript n represents the nth eigenmode; The weighted characteristic equation is obtained through the generalized eigenvalue equation: ； in, represents the eigenvalue; According to this equation, the eigenvalues ​​and eigenmode vectors corresponding to the coefficient matrix of the target are solved.

8. The system for rapid and accurate prediction of RCS sequence based on target characteristic mode according to claim 5, characterized in that: To dynamically reconstruct the RCS sequence of the target, it is first necessary to approximate the induced current on the target surface as a finite term characteristic current superposition based on the statically extracted eigenvalues ​​and characteristic patterns and the predicted motion parameters: ； in, represents the induced current on the target surface, Indicates the number of truncated items, Indicates the mode weight coefficient corresponding to each characteristic mode; The mode weight coefficient is obtained by taking the inner product of the incident field tangent equation and the characteristic current: ； in, represents the tangential component of the incident electric field; The characteristic electric field generated by the characteristic current is: ； in, represents the source location, represents the wave number, and denote polar angle and azimuth respectively; Consistent with the characteristic current superposition principle, the target electric field is approximately expressed as a finite term characteristic electric field superposition: ； The RCS sequence value of the target is obtained according to the reconstructed electric field.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method according to any one of claims 1 to 4 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images