An intelligent real aperture scanning radar angular super-resolution imaging method

Abstract

This invention discloses an intelligent angular super-resolution imaging method for real-aperture scanning radar, applied in the field of radar imaging technology. Existing real-aperture radar angular super-resolution methods based on Bayesian and regularized frameworks still suffer from problems such as difficulty in manually selecting parameters and slow iterative update speed under low signal-to-noise ratio (SNR) conditions. This invention first transforms the convolution inversion problem of real-aperture scanning radar into an echo autocorrelation matrix inversion problem to improve the ill-conditioned nature of the inverse matrix. Second, a learnable correction matrix is ​​introduced into the iterative adaptive solution method to combine iterative adaptive solution with a deep network. Finally, the echo autocorrelation matrix is ​​updated through iterative learning to reduce the impact of noise on the inversion results and improve the angular resolution of real-aperture radar. Simulation results show that the method of this invention maintains good angular super-resolution performance under low SNR conditions.

Description

Technical Field

[0001] This invention belongs to the field of radar imaging technology, and specifically relates to a scanning radar angle super-resolution imaging technology. Background Technology

[0002] The angular resolution of a real aperture radar is related to the antenna aperture. In practical applications, due to space constraints on the radar platform, the aperture size is limited, resulting in low angular resolution for real aperture radars, making it difficult to meet the application requirements for high-resolution imaging in fields such as autonomous landing of aircraft and terrain mapping.

[0003] The echo sequence of a scanning radar along the angular direction can be modeled as a convolution relationship between the target scattering coefficient and the antenna pattern modulation function. Therefore, its angular resolution can be improved through convolution inversion. However, when the antenna size is limited, the convolution inversion process exhibits severe ill-conditioning due to the low-pass characteristics of the antenna pattern. To suppress the ill-conditioning of the convolution inversion process, scholars at home and abroad have proposed a real aperture radar angular super-resolution method based on Bayesian and regularized frameworks by introducing constraints on the target and noise.

[0004] The above framework can effectively improve the super-resolution imaging performance of scanning radar, but under low signal-to-noise ratio conditions, there are still problems such as difficulty in manually selecting parameters and slow iteration update speed. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes an intelligent real aperture scanning radar angular super-resolution imaging method. Based on an iterative adaptive solution method and combined with the parameter learning capability of deep networks, it avoids the problems of slow parameter selection and iterative updates in traditional algorithms, and can maintain good angular super-resolution performance even under low signal-to-noise ratio conditions.

[0006] The technical solution adopted in this invention is: an intelligent real aperture scanning radar angular super-resolution imaging method, comprising:

[0007] S1. Preprocess the acquired echo data;

[0008] S2. Initialize the echo autocorrelation matrix and the correction matrix;

[0009] S3. Calculate the target scattering coefficient based on the echo autocorrelation matrix and correction matrix, as well as the preprocessing results from step S1.

[0010] S4. If the set number of iterations is reached, output the current target scattering coefficient; otherwise, update the echo autocorrelation matrix and correction matrix and return to step S3.

[0011] The beneficial effects of this invention are as follows: First, the method of this invention transforms the convolution inversion problem of real aperture scanning radar into the problem of solving the echo autocorrelation matrix inversion, thereby improving the ill-conditioned nature of the inverse matrix. Second, a learnable correction matrix is ​​introduced into the iterative adaptive solution method to achieve the combination of the iterative adaptive solution method and deep networks. Finally, the echo autocorrelation matrix is ​​updated through iterative learning to improve the angular resolution of real aperture radar. Compared with traditional IAA methods, the proposed method avoids the problems of parameter selection and slow iterative updates in traditional algorithms. Furthermore, due to the learning and fitting capabilities of deep networks, the proposed method can maintain good angular super-resolution performance under low signal-to-noise ratio conditions. Attached Figure Description

[0012] Figure 1 This is a flowchart of an intelligent real aperture scanning radar angular super-resolution imaging method according to the present invention.

[0013] Figure 2 This is a geometric model diagram of an airborne real aperture radar in an embodiment of the present invention.

[0014] Figure 3 This is a structural diagram of each computational layer in this invention;

[0015] Among them, (a) is the target scattering coefficient solution layer, (b) is the autocorrelation matrix update calculation layer, and (c) is the parameter training and update module.

[0016] Figure 4 The image shows the original distribution and original echo (SNR=15dB) of the target in this embodiment of the invention.

[0017] (a) represents the original scene, and (b) represents the actual echo data at SNR=15dB.

[0018] Figure 5 This is a comparison chart of the extended target reconstruction results of various methods in the embodiments of the present invention;

[0019] Among them, (a) is the reconstruction result of the L2 norm regularization method, (b) is the reconstruction result of the L1 norm regularization method, (c) is the reconstruction result when the regularization parameter γ=0.2 of the combined traditional IAA method is used, (d) is the reconstruction result when the regularization parameter γ=0.05 of the traditional IAA method is used, and (e) is the reconstruction result of the method of the present invention. Detailed Implementation

[0020] To facilitate understanding of the technical content of this invention by those skilled in the art, the following description, in conjunction with the accompanying drawings, further illustrates the invention.

[0021] like Figure 1 The flowchart of a real aperture radar adaptive combined regularized angle super-resolution imaging method of the present invention is shown below. The specific steps are as follows:

[0022] Step 1: Echo data acquisition and preprocessing;

[0023] like Figure 2 As shown in Table 1, this invention employs an airborne real-aperture radar motion geometry model, and the specific parameter values ​​of the airborne platform system are shown in Table 1. To simulate the low signal-to-noise ratio environment of real-world conditions, a noise level of 15 dB was added to this simulation.

[0024] Table 1 Simulation Parameters

[0025]

[0026] In this embodiment, the simulated scanning detection area is set as follows: The antenna beamwidth is The scanning speed is .

[0027] The initial position of the radar platform is The relative position of target E is The target's historical distance is The expression is:

[0028] (1)

[0029] If a radar transmits a linear frequency modulated signal, the received signal can be expressed as:

[0030] (2)

[0031] Where t is the time in the range direction (faster) and τ is the time in the azimuth direction (slower). These represent distance and angle variables, respectively. Indicates in The target scattering coefficient at that location; where c is the speed of electromagnetic wave propagation, and ; express Continuous antenna pattern function; pulse width of transmitted signal , The unit of time measurement is microsecond; carrier frequency. , This indicates frequency tuning, and signal bandwidth , This is a distance-oriented time window function.

[0032] After echo preprocessing, the echo data from a real aperture scanning radar can be expressed as the convolution relationship between target scattering and the point spread function. The echo vector of the m-th range cell can be expressed in matrix form as follows:

[0033] (3)

[0034] in, This represents the echo data vector of the m-th distance cell, with the superscript T indicating transpose. This represents the target scattering coefficient vector at distance m. To represent a complex number, Let m represent the system noise vector of the m-th distance cell. For a real aperture scanning radar system matrix, The measurement matrix steering vector representing the k-th target direction, i.e., the k-th column of the system matrix H, contains scanning information and can also be denoted as... The system matrix can be represented as:

[0035] (4)

[0036] The discrete antenna pattern function can be expressed as: , Let represent the amplitude at point l in the antenna pattern function, and N, K, and L represent the angular dimensions of the echo data, the angular dimensions of the target scattering element, and the antenna pattern dimension, respectively.

[0037] Step 2: Construct the initialization module;

[0038] The algorithm initialization module primarily sets the echo autocorrelation matrix and regularization parameters. In the traditional Iterative Adaptive Approach (IAA) algorithm, the echo autocorrelation matrix needs to be initialized. And manually select the regularization parameter γ, and then through The reconstruction results converged in the next iteration.

[0039] To ensure the adaptability of the regularization parameters to targets in different orientations, the manually selected constants are initialized as random vectors. , Represents a diagonal loading matrix The diagonal elements. The introduced regularization weighting parameters can be iteratively updated during training, improving the adaptability of the regularization parameters to the imaging scene during angular super-resolution imaging. Diagonal loading matrix. It can be represented as

[0040] (5)

[0041] These are the diagonal loading matrices corresponding to the specific echo data angles. The diagonal elements.

[0042] Step 3: Construct the target scattering coefficient solution layer;

[0043] The calculation process for the target scattering coefficient is expressed as follows:

[0044] (6)

[0045] The superscript * indicates the conjugate of the matrix.

[0046] Based on the calculation process of the target scattering coefficient, a target scattering coefficient solution layer is constructed. In the initial iteration, the solution layer is initialized... and initialize the correction matrix. In order to solve for the target scattering coefficients in different directions Furthermore, through the iterative module, each computational layer and update layer solves for the... Scattering coefficient of the next iteration The implementation process of this module is as follows: Figure 3 As shown in (a).

[0047] in Calculate the numerator of the target scattering coefficient in equation (6). The denominator of the target scattering coefficient in equation (6) is used to calculate the corrected echo autocorrelation matrix. go through The computational layer can solve for the target scattering coefficient. .

[0048] Step 4: Construct the autocorrelation matrix and update the computation layer;

[0049] The calculation process of the autocorrelation matrix is ​​expressed as follows:

[0050] (7)

[0051] Where I represents the identity matrix, This represents the noise power matrix.

[0052] Based on the calculation process of the autocorrelation matrix, an autocorrelation matrix update calculation layer is constructed. In the echo autocorrelation matrix calculation layer, the first... Target scattering coefficient in the next iteration and current The correction matrix required for the next iteration The updated echo autocorrelation matrix is ​​calculated according to equation (7). The implementation process diagram of this module is as follows: Figure 3 As shown in (b). To solve for the target power in equation (7), To calculate the autocorrelation matrix in equation (7), by... By correcting the echo autocorrelation matrix with the matrix, we can obtain... In this computational layer, the correction matrix needs to be adjusted. We learn and update the diagonal elements in the code.

[0053] Step 5: Construct the parameter training and update module;

[0054] The parameter training and update module mainly performs the calculation of the loss function (L calculation layer) and the correction matrix. Update ( (Update layer), the Loss function is calculated as follows:

[0055] (8)

[0056] in, This represents the 2-norm (Euclidean norm). This represents the learnable parameter set of the IAA-Net network, i.e., the regularization parameter correction matrix. The diagonal vector. This represents the angular super-resolution reconstruction result output by IAA-Net.

[0057] In this module, the quality of the target reconstruction results during a single training iteration is evaluated by calculating the loss cost function between the reconstructed target scattering coefficients and the reference scene at different distance units. Furthermore, the regularization parameter correction matrix is ​​updated accordingly. This allows for adaptive learning of the parameter.

[0058] In the Loss function computation layer, the input is the super-resolution result from a single training iteration. Compared to real-world reference scenarios The output is the cost function between the reconstructed result and the real reference scene. This is achieved by calculating the super-resolution results. Compared to real-world reference scenarios The loss cost between different points is used to evaluate the performance of the reconstruction results. The implementation process of this module is as follows: Figure 3 As shown in (c).

[0059] In the correction matrix update layer, the correction matrix during the iteration process is updated through back propagation (BP). The input to the correction matrix update layer is the loss cost function value, and the output is the initial echo autocorrelation matrix for the next training iteration. The update process is shown in equation (9). Here, ADAM is the adaptive moment estimation function, which adaptively adjusts the learning rate based on the first-order and second-order moment estimates of the parameters. The operator for solving partial derivatives is used. The target scattering coefficients for the next training iteration can be obtained by adaptively updating the regularization parameter correction matrix of this layer. .

[0060] (9)

[0061] The implementation structure diagram of this module is as follows: Figure 3 As shown in (d). Wherein This module is for gradient calculation. It calculates the loss function. For the correction matrix The gradient information is obtained by calculating the gradient, and then the parameters are updated by the ADAM optimizer. By repeatedly executing the forward propagation, gradient calculation and parameter update process, the parameters of the neural network are continuously adjusted to reduce the value of the loss function and obtain optimized learning parameters.

[0062] Step 6: Solve for the final target scattering coefficient;

[0063] Repeat steps three, four, and five, and then... After L iterations and L training iterations, the updated correction matrix is ​​obtained. By iterating through all range cells of the echo matrix, the azimuth super-resolution imaging results for the entire echo matrix are obtained.

[0064] In this example, the number of iterations Set it to 10, and set the number of training iterations L to 30.

[0065] like Figure 3 As shown, Figure 4 (a) represents the original scene. Figure 4 (b) represents the actual echo data at SNR=15dB. Figure 4 This represents the "V-shaped" scene target reconstruction results of various methods. Figure 5 (a) shows the reconstruction result using the L2 norm regularization method. Figure 5 (b) shows the reconstruction result using the L1 norm regularization method. Figure 5 (c) shows the reconstruction results when the regularization parameter γ = 0.2 of the traditional IAA method is combined. Figure 5 (d) Reconstruction results of the traditional IAA method with regularization parameter γ=0.05. Figure 5 (e) represents the reconstruction result of the method of the present invention; from Figure 5 As can be seen, the reconstruction results of the method of the present invention are superior to those of other traditional methods in terms of resolution performance, and the reconstruction results are more robust.

[0066] In summary, the method of this invention addresses the manual parameter selection problem inherent in traditional IAA methods. By constructing the IAA-Net method based on deep networks, the parameter selection problem in the iterative process of traditional IAA methods is transformed into a parameter learning problem based on deep networks, achieving adaptive selection of regularization weighted parameters. Results show that, compared to traditional IAA methods, the proposed IAA-Net method can adaptively optimize regularization parameters from the data, avoiding manual parameter selection, and maintaining good angular super-resolution capability even at low signal-to-noise ratios.

[0067] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Various modifications and variations can be made to the invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the scope of the claims of the invention.

Claims

1. An intelligent real aperture scanning radar angular super-resolution imaging method, characterized in that, include: S1. Preprocess the acquired echo data; S2. Initialize the echo autocorrelation matrix and the correction matrix; The correction matrix is ​​represented as: ； Where N represents the angular dimension of the echo data, These are the correction matrices corresponding to the specific echo data angles. The diagonal elements; S3. Calculate the target scattering coefficient based on the echo autocorrelation matrix and correction matrix, as well as the preprocessing results from step S1. S4. If the set number of iterations is reached, output the current target scattering coefficients; otherwise, update the echo autocorrelation matrix and correction matrix, and return to step S3. The update expression for the echo autocorrelation matrix is: ； in, H represents the target power, and H represents the real aperture scanning radar system matrix. Indicates the first The target scattering coefficients of the next iteration, where I represents the identity matrix. Represents the noise power matrix. Indicates the first The correction matrix required for the next iteration.

2. The intelligent real aperture scanning radar angular super-resolution imaging method according to claim 1, characterized in that, The formula for calculating the target scattering coefficient is: ； The superscript * indicates matrix conjugation. Indicates the direction of the k-th target. The measurement matrix guide vector at the location, Indicates the first The echo autocorrelation matrix corresponding to the next iteration This indicates the result of echo data preprocessing. Indicates the direction of the k-th target. The scattering coefficient of the next iteration.

3. The intelligent real aperture scanning radar angular super-resolution imaging method according to claim 2, characterized in that, The correction matrix is ​​updated via backpropagation.

4. The intelligent real aperture scanning radar angular super-resolution imaging method according to claim 3, characterized in that, The expression for the Loss function used in the correction matrix update process is as follows: ； in, Describing the 2-norm, This represents the set of learnable parameters, i.e., the regularization parameter correction matrix. The diagonal vector, Indicates the first Angular super-resolution reconstruction results of the next iteration This represents the target scattering coefficient of the real reference scene.

Images