Time-frequency representation enhancement method and device based on structured short-term sparse representation

Through the time-frequency characterization enhancement method based on structured short-time sparse representation, a time-frequency characterization sparse observation model was established and the alternating direction multiplier method was used to solve the problem of low resolution in traditional time-frequency characterization, and the complex observation conditions were realized The efficient time-frequency characterization of the radar echo signal has been enhanced.

CN118731891BActive Publication Date: 2024-11-15NAT UNIV OF DEFENSE TECH
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411206972.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-30
Publication Date
2024-11-15
Estimated Expiration
2044-08-30

AI Technical Summary

Technical Problem

Under complex observation conditions, the traditional time-frequency characterization method has a low resolution, making it difficult to effectively extract the micro-movement characteristics in radar echo signals, especially under the radar system transmission power limit, target maneuvering flight and electromagnetic interference, resulting in Time frequency characterization resolution decreases and false points increase.

Method used

The time-frequency characterization enhancement method based on structured short-time sparse representation is adopted. By establishing a time-frequency characterization sparse observation model, the sparse and structural continuous characteristics are constrained by norms, convolutional weighting and morphological filtering, combined with the alternating direction multiplier method Iterative solution is carried out to obtain a two-dimensional time-frequency characterization of micro-movement targets with sparse and structural features enhanced.

Benefits of technology

It significantly improves the time-frequency characterization resolution of defective radar echoes, reduces false points, improves the characterization accuracy of target microDoppler features, and can effectively extract the prior information in the time frequency domain in complex observation scenarios, enhancing the target time Frequency characterization.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118731891B_ABST
    Figure CN118731891B_ABST
Patent Text Reader

Abstract

The present application relates to a method and device for enhancing time-frequency representation based on structured short-time sparse representation. A sparse observation model of time-frequency representation of radar echo data is established based on the relationship between the radar echo data model and the time-frequency representation to be solved. l 1 norm, convolution weighting and morphological filtering are used to constrain the sparse and structural continuity characteristics of the target in the sparse observation model of time-frequency representation, and a structured sparse optimization model of time-frequency representation is obtained. After the real-time radar echo data of the micro-motion target is input into the structured sparse optimization model, the alternating direction multiplier method is used to iteratively solve the structured sparse optimization model to obtain a two-dimensional time-frequency representation of the micro-motion target with jointly enhanced sparse and structural features. This method can be used to improve the resolution of the time-frequency representation of the defective radar echo by mining the prior information of the micro-motion target in the time-frequency domain in order to solve the problem of low resolution of traditional time-frequency representation and performance degradation under complex observation conditions.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of radar signal processing technology, and in particular to a method and device for enhancing time-frequency representation based on structured short-time sparse representation. Background Art

[0002] As a powerful tool for non-stationary signal processing, time-frequency representation can characterize the time-varying Doppler modulation characteristics of micro-motion signals and is widely used in the study of radar target micro-motion characteristics. Time-frequency representation is mainly divided into two categories: linear time-frequency representation and quadratic Cohen-type time-frequency representation. Linear time-frequency representation does not produce cross terms, but due to the limitation of the uncertainty principle, the time-frequency resolution is usually low. Although the quadratic Cohen-type time-frequency representation can provide higher time-frequency resolution, it is susceptible to cross-term interference when processing multi-component signals.

[0003] In actual observation scenarios, due to complex observation conditions such as the radar system's transmit power limitation, target maneuvering, and electromagnetic interference in a strong confrontation environment, the actual received radar echo signal-to-noise ratio is often low, and some echo data is unavailable and defective. Under the above complex observation conditions, the performance of traditional time-frequency characterization methods such as short-time Fourier transform is significantly reduced, and the target time-frequency characterization will be severely defocused and contain a large number of false points, resulting in reduced resolution and difficulty in extracting micro-motion features. Summary of the invention

[0004] Based on this, it is necessary to provide a time-frequency representation enhancement method and device based on structured short-time sparse representation, which can improve the time-frequency representation resolution of defective radar echoes in order to address the above technical problems.

[0005] A time-frequency representation enhancement method based on structured short-term sparse representation, the method comprising:

[0006] Based on the relationship between the radar echo data model and the time-frequency representation to be solved, a sparse observation model of time-frequency representation of radar echo data is established;

[0007] use norm, convolution weighting and morphological filtering, constraining the sparse and structural continuity characteristics of the target in the time-frequency representation sparse observation model, and obtaining a structured sparse optimization model of time-frequency representation;

[0008] Obtain real-time radar echo data of micro-moving targets;

[0009] After the real-time radar echo data is input into the structured sparse optimization model, the structured sparse optimization model is iteratively solved by using the alternating direction multiplier method to obtain a two-dimensional time-frequency representation of the micro-motion target with sparse and structural features jointly enhanced.

[0010] In one embodiment, the time-frequency representation sparse observation model is expressed as:

[0011] ;

[0012] In the above formula, represents the reconstructed time-frequency representation vector, represents the defect echo observation vector, i.e., the radar echo data model, Represents the downsampling matrix, which consists of elements 0 and 1, where the echo element sampling position is 1 and the missing position is 0. represents the inverse short-time Fourier transform matrix, where represents the sliding window matrix, , is a diagonal matrix, represents the block inverse Fourier transform matrix, represents the observation noise vector.

[0013] In one embodiment, the structured sparse optimization model of the time-frequency representation is expressed as:

[0014] ;

[0015] In the above formula, and Respectively Norm and norm, Represents the data error term, which is used to measure the similarity between the reconstructed time-frequency representation and the original time-frequency representation. Represents the structural weighted vector, which is obtained by the structural weighted matrix Stacking along the frequency dimension gives: , , represents the modified structured sparse prior regularization term.

[0016] In one embodiment, the structural weighting matrix is ​​constructed by convolution weighting operation and morphological filtering, and is expressed as:

[0017] ;

[0018] In the above formula, Indicates size All 1 matrices, represents matrix element-wise division, represents any minimum value, represents the target structure mask obtained by convolution operation, Represents the use of morphological operators Make corrections.

[0019] In one embodiment, when solving the structured sparse optimization model using the alternating direction multiplier method:

[0020] After introducing a splitting variable to transform the structured sparse optimization model, a corresponding augmented Lagrangian function is constructed;

[0021] Performing a scale transformation on the Lagrangian multiplier in the augmented Lagrangian function using a penalty term coefficient, and introducing a dual variable to transform the augmented Lagrangian function;

[0022] According to the transformed augmented Lagrangian function, the structured sparse optimization model is decomposed into multiple sub-problems for solving.

[0023] In one embodiment, before iteratively solving multiple sub-problems, the special structure of the observation matrix is ​​used to transform the matrix-vector multiplication into the matrix element dot product, and the variables are iteratively updated in matrix form.

[0024] In one embodiment, during each iterative calculation process of performing alternating iterative solutions to multiple sub-problems:

[0025] According to the splitting variables and dual variables calculated in the previous iteration, the ridge regression problem containing data error terms and augmented Lagrangian terms is solved to update the global variables;

[0026] According to the dual variables calculated in the previous iteration and the global variables updated in the current iteration, a complex soft threshold operator is derived based on the structured sparse regularization term and the augmented Lagrangian term to update the splitting variables;

[0027] Finally, the updated global variables and split variables are minimized by combining the current iteration calculations to update the dual variables.

[0028] In one embodiment, when performing alternate iterative solutions to multiple sub-problems, the number of iterations or the error precision is set to obtain the final target time-frequency enhanced representation;

[0029] Among them, when the number of iteration rounds reaches a preset number or the adjacent iteration error reaches a preset value, the reconstructed time-frequency representation obtained by this round of iteration is to realize the structured short-time sparse time-frequency enhanced representation.

[0030] The present application also provides a time-frequency representation enhancement device based on structured short-time sparse representation, the device comprising:

[0031] A time-frequency representation sparse observation model building module is used to build a time-frequency representation sparse observation model of radar echo data based on the relationship between the radar echo data model and the time-frequency representation to be solved;

[0032] Model optimization module for utilizing norm, convolution weighting and morphological filtering, constraining the sparse and structural continuity characteristics of the target in the time-frequency representation sparse observation model, and obtaining a structured sparse optimization model of time-frequency representation;

[0033] Real-time data acquisition module, used to obtain real-time radar echo data of micro-moving targets;

[0034] The time-frequency enhancement module is used to input the real-time radar echo data into the structured sparse optimization model, and then iteratively solve the structured sparse optimization model using an alternating direction multiplier method to obtain a two-dimensional time-frequency representation of the micro-motion target with sparse and structural features jointly enhanced.

[0035] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0036] Based on the relationship between the radar echo data model and the time-frequency representation to be solved, a sparse observation model of time-frequency representation of radar echo data is established;

[0037] use norm, convolution weighting and morphological filtering, constraining the sparse and structural continuity characteristics of the target in the time-frequency representation sparse observation model, and obtaining a structured sparse optimization model of time-frequency representation;

[0038] Acquire real-time radar echo data of micro-moving targets;

[0039] After the real-time radar echo data is input into the structured sparse optimization model, the structured sparse optimization model is iteratively solved by using the alternating direction multiplier method to obtain a two-dimensional time-frequency representation of the micro-motion target with sparse and structural features jointly enhanced.

[0040] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0041] Based on the relationship between the radar echo data model and the time-frequency representation to be solved, a sparse observation model of time-frequency representation of radar echo data is established;

[0042] use norm, convolution weighting and morphological filtering, constraining the sparse and structural continuity characteristics of the target in the time-frequency representation sparse observation model, and obtaining a structured sparse optimization model of time-frequency representation;

[0043] Acquire real-time radar echo data of micro-moving targets;

[0044] After the real-time radar echo data is input into the structured sparse optimization model, the structured sparse optimization model is iteratively solved by using the alternating direction multiplier method to obtain a two-dimensional time-frequency representation of the micro-motion target with sparse and structural features jointly enhanced.

[0045] The time-frequency representation enhancement method and device based on structured short-time sparse representation establishes a time-frequency representation sparse observation model of radar echo data based on the relationship between the radar echo data model and the time-frequency representation to be solved. l 1 norm, convolution weighting and morphological filtering are used to constrain the sparse and structural continuity characteristics of the target in the sparse observation model of time-frequency representation, and a structured sparse optimization model of time-frequency representation is obtained. After the real-time radar echo data of the micro-motion target is input into the structured sparse optimization model, the alternating direction multiplier method is used to iteratively solve the structured sparse optimization model to obtain a two-dimensional time-frequency representation of the micro-motion target with jointly enhanced sparse and structural features. This method can be used to improve the resolution of the time-frequency representation of the defective radar echo by mining the prior information of the micro-motion target in the time-frequency domain in order to solve the problem of low resolution of traditional time-frequency representation and performance degradation under complex observation conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 1 is a flow chart of a method for enhancing time-frequency representation based on structured short-term sparse representation in one embodiment;

[0047] Figure 2 A schematic flow chart of each iterative calculation process step when multiple sub-problems are alternately iterated to solve in one embodiment;

[0048] Figure 3 It is a schematic diagram of a specific iterative updating process of converting each sub-problem represented by a vector into a matrix form and then using an alternating direction multiplication method in another embodiment;

[0049] Figure 4 This is a schematic diagram of the experimental results of a simulated cone target data under different sparse sampling rates in a simulation experiment, where: Figure 4 (a) Figure 4 (c) and Figure 4 (e) Schematic diagram of the two-dimensional time-frequency representation of the target obtained by short-time Fourier transform for the simulated cone target at 70%, 50% and 30% sparseness rates; Figure 4 (b) Figure 4 (d) and Figure 4 (f) Schematic diagram of the two-dimensional time-frequency representation of the target obtained by this method for the simulated cone target at 70%, 50% and 30% sparseness rates;

[0050] Figure 5This is a schematic diagram of the experimental results of measuring metal cone data in a darkroom at different sparse sampling rates in a simulation experiment, where: Figure 5 (a) Figure 5 (c) and Figure 5 (e) Schematic diagram of the target two-dimensional time-frequency representation obtained by short-time Fourier transform when measuring the metal cone in a darkroom at sparse rates of 70%, 50% and 30%, respectively. Figure 5 (b) Figure 5 (d) and Figure 5 (f) Schematic diagram of the target two-dimensional time-frequency representation obtained by this method when measuring a metal cone in a darkroom at sparse rates of 70%, 50%, and 30% respectively;

[0051] Figure 6 is a structural block diagram of a time-frequency representation enhancement device based on structured short-time sparse representation in one embodiment;

[0052] Figure 7 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0054] In view of the problem that in actual observation scenarios, due to complex observation conditions such as radar system transmission power limitation, target maneuvering flight, and electromagnetic interference in a strong confrontation environment, the actual received radar echo signal-to-noise ratio is often low, and some echo data is unavailable due to defects, in this application,

[0055] like Figure 1 As shown, a time-frequency representation enhancement method based on structured short-term sparse representation is provided, comprising the following steps:

[0056] Step S100: establishing a sparse observation model of time-frequency representation of radar echo data based on the relationship between the radar echo data model and the time-frequency representation to be solved.

[0057] Step S110, using Norm, convolution weighting and morphological filtering are used to constrain the sparse and structural continuity characteristics of the target in the sparse observation model of time-frequency representation, and a structured sparse optimization model of time-frequency representation is obtained.

[0058] Step S120, acquiring real-time radar echo data of the micro-motion target.

[0059] Step S130, after the real-time radar echo data is input into the structured sparse optimization model, the structured sparse optimization model is iteratively solved using the alternating direction multiplier method to obtain a two-dimensional time-frequency representation of the micro-motion target with joint enhancement of sparse and structural features.

[0060] In this embodiment, a sparse observation model of time-frequency representation of radar echo data is first constructed, and then the prior information related to the time-frequency representation recovery is introduced, an optimization framework is established to solve the recovery result, and then the optimized model is alternately iterated and solved by the alternating direction multiplier method to obtain the reconstructed time-frequency enhanced representation.

[0061] Furthermore, step S100 and step S110 are preliminary optimization model construction steps, and steps S120 to S130 are the process of performing two-dimensional time-frequency enhancement characterization on real-time radar echo data using the optimization model.

[0062] In step S100, considering that the target micro-motion produces a modulation effect on the radar echo, which is mainly reflected in the micro-Doppler modulation of the echo, the radar echo of the micro-motion target can be regarded as a multi-component non-stationary signal. Since the Doppler spectrum of the target is broadened due to micro-motion, the traditional spectrum analysis method based on Fourier transform cannot accurately describe the micro-motion signal due to the lack of time and frequency positioning. As a powerful tool for non-stationary signal processing, time-frequency analysis can describe the time-varying frequency component of the signal by characterizing the energy distribution of the signal on a two-dimensional time-frequency plane, and is widely used to analyze the micro-Doppler characteristics of the target.

[0063] In this embodiment, the time-frequency analysis uses short-time Fourier transform (STFT). Since the short-time Fourier transform has the advantages of linear reversibility and no cross terms, it can be used to process micro-motion signals.

[0064] Specifically, unlike Fourier transform, which processes the entire signal duration, STFT uses a time window to divide the signal into multiple continuous short-time segments, and performs Fourier transform on each short-time segment to obtain the frequency distribution information of different segments at the corresponding time point. The short-time Fourier transform of can be expressed as:

[0065] (1)

[0066] In formula (1), represents the time-frequency representation of the micro-motion signal obtained using STFT, So A symmetric window function with centered on , such as the Hamming window and Gaussian window, is used to ensure the stability of the signal within the window.

[0067] According to formula (1), the signal The short-term segment of is defined as:

[0068] (2)

[0069] In formula (2), Indicates that the signal is A short fragment of time, Represents a time window function Length.

[0070] The time and frequency dimensions of the time-frequency representation in formula (1) are sampled respectively, and the discrete signal model corresponding to the short-time Fourier transform can be obtained after gridding the two-dimensional time-frequency representation:

[0071] (3)

[0072] In formula (3), is the echo time sampling interval, N Indicates the number of signal sampling points, M represents the number of discrete Fourier transform points, , is the echo frequency sampling interval. To facilitate subsequent derivation, in this embodiment, .

[0073] Furthermore, formula (3) is rewritten into a matrix form, namely:

[0074] (4)

[0075] In formula (4), and Represent the complete echo observation vector and the two-dimensional time-frequency representation matrix respectively, N Indicates the number of sampling points of a complete echo, is the short-time Fourier transform matrix.

[0076] However, formula (4) gives the target echo signal model under ideal conditions. In actual situations, due to the influence of the multi-function radar system and the target maneuvering flight, the radar observation data of a single target is usually incomplete. At the same time, in actual confrontation environments, due to environmental influences such as electromagnetic interference and noise pollution, some echo data will be defective and unusable. Therefore, in actual scenarios, only part of the echo observation can be used to obtain the time-frequency representation of the target. According to formula (4), the defective echo observation, that is, the sparse observation model of the time-frequency representation of the target echo, can be expressed as:

[0077] (5)

[0078] In formula (5), Represents the reconstructed time-frequency representation vector, which is composed of the time-frequency representation matrix Stacking along the frequency dimension gives: , is a vectorized operator. represents the defect echo observation vector, i.e., the radar echo data model, , Represents the downsampling matrix, which consists of elements 0 and 1, where the echo element sampling position is 1 and the missing position is 0. If the valid echo sampling index is recorded as , ,but . represents the inverse short-time Fourier transform matrix, where represents the sliding window matrix, , is a diagonal matrix, . represents the block inverse Fourier transform matrix. Further, ,in Indicates size The identity matrix of represents the Kronecker product, is the inverse Fourier transform matrix, . represents the observation noise vector.

[0079] Furthermore, according to the time-frequency representation sparse observation model expressed by formula (5), the time-frequency enhancement representation problem can be abstracted as Recovering 2D time-frequency representation However, since the observation matrix is an underdetermined matrix, , the solution of formula (5) is an ill-posed problem, and there are multiple solutions. At the same time, affected by the observation noise, if we directly calculate This will lead to serious errors in the reconstructed time-frequency representation.

[0080] In view of this, in this embodiment, prior information related to the restoration of time-frequency representation is introduced, and an optimization framework is established to solve the restoration result.

[0081] According to the equivalent scattering center theory, the scattering of the target in the microwave frequency band can be approximated as the superposition of strong scattering centers. Since the number of strong scatterers on the micro-motion target is usually limited, they appear as a few continuous curves (time-frequency ridges) on the time-frequency plane, and the rest are background areas. Therefore, the micro-motion target echo is sparse on the time-frequency plane, and a sparse prior can be introduced to restore the two-dimensional time-frequency representation. Compared with traditional greedy and Bayesian algorithms, convex optimization algorithms can effectively balance calculation accuracy and efficiency. According to the Least Absolute Shrinkage and Seleted Operator (LASSO) model, the convex optimization algorithm can be introduced. The sparse regularized prior of the norm representation target can realize the sparse feature enhancement of the two-dimensional time-frequency representation. Therefore, the time-frequency representation in formula (5) The recovery problem can be modeled as Norm regularization problem, that is, transforming formula (5) into:

[0082] (6)

[0083] In formula (6), and Respectively Norm and Norm. Represents the data error term, which is used to measure the similarity between the restored time-frequency representation and the original time-frequency representation. is a sparse regularization term, which is a priori representation of the sparse features of the time-frequency surface. Represents the regularization term coefficient, which is used to control the sparsity of sparse features and has good noise suppression performance.

[0084] Considering that another important feature of the time-frequency representation of micro-motion targets is that the curve on the time-frequency surface has structural continuity, that is, the time-frequency curve is continuously distributed on the two-dimensional time-frequency surface and is not isolated, and the micro-Doppler changes of the target at adjacent moments are continuous. However, although in formula (6), The sparse prior introduced by the norm can achieve sparse feature enhancement more easily than other norms. However, while highlighting the strong scattering areas, it will cause the weaker scattering areas to be eliminated, losing some important information. At the same time, it also fails to capture the relationship between different scattering areas, thereby destroying the structural continuity of the target time-frequency representation, which appears as an intermittent and discontinuous curve on the time-frequency plane.

[0085] In view of the above problems, in this embodiment, the structural continuity prior of the target time-frequency representation is further introduced to achieve the joint enhancement of the target sparse features and structural features. The norm regularization problem is corrected. On the basis of formula (6), the structural continuity of the time-frequency curve is introduced into In the norm minimization, the final structured sparse optimization model for solving the time-frequency representation is expressed as:

[0086] (7)

[0087] In formula (7), Represents the structural weighted vector, which is obtained by the structural weighted matrix Stacking along the frequency dimension gives: , , Represents the modified structured sparse prior regularization term. The modified structured sparse prior regularization term can be used to achieve the joint extraction of structural features and sparse features on the target time-frequency surface. represents the Hadamard product.

[0088] In this embodiment, the structural weighting matrix is ​​constructed by convolution weighting operation and morphological filtering. First, the time-frequency representation obtained by the previous iteration is subjected to convolution operation. Weighted, it can be expressed as:

[0089] (8)

[0090] In formula (8), Represents the convolution kernel parameters, which are used to adjust the weight assignment. represents a two-dimensional convolution operation, Represents the modulus operator.

[0091] Since the convolution operation utilizes the neighborhood information of the scattering points, it can adaptively capture the time-frequency curve area and effectively enhance the intensity of the target scattering area. However, due to the lack of observed echo data and the actual scene noise, the time-frequency curve structure mask obtained by the convolution operation still has discontinuities.

[0092] Considering that morphological filtering technology plays an important role in improving the quality of time-frequency representation, especially in target structure extraction and noise suppression, in this embodiment, morphology is introduced to solve the problem that the time-frequency curve structure mask obtained by the convolution operation still has discontinuous phenomena. The core idea is to use the set structural elements to traverse the time-frequency representation of the target to obtain the neighborhood information of different scattering areas (discontinuous time-frequency curves), so as to effectively detect the time-frequency curve structure and realize the fitting of the target continuous structure.

[0093] In this embodiment, the structural element is defined , is the structure element radius, and the dilation and erosion operations can be expressed as:

[0094] (9)

[0095] (10)

[0096] In formulas (9) and (10), and represents the dilation operator, and represents the corrosion operator, represent The relative offset of Represents structural elements domain.

[0097] Furthermore, based on the dilation and erosion operators, the morphological opening and closing operations can be derived as follows:

[0098] (11)

[0099] (12)

[0100] Therefore, according to the morphological operators shown in formulas (9) and (11), the target structure mask obtained by convolution operation in formula (8) is modified. Using morphological filtering, the structural weighting matrix of the target on the time-frequency plane can be expressed as:

[0101] (13)

[0102] In formula (13), Indicates size All 1 matrices, represents matrix element-wise division, represents any minimum value, represents the target structure mask obtained by convolution operation, Represents the use of morphological operators Make corrections.

[0103] It can be seen from formula (13) that the structural weighted mask is obtained through the dilation operator and the opening operation. The former mainly expands the discontinuous structural area in the time-frequency representation and connects the isolated areas on the time-frequency surface. The latter is mainly used to filter out the background noise in the mask and retain the main mask of the target.

[0104] In this embodiment, formula (7) The structural prior of the target is introduced on the basis of the norm, that is, the time-frequency representation is weighted by convolution operation and morphological filtering, so as to introduce the structural continuity of the target into the regularization term and realize the structured sparse prior modeling of the target time-frequency representation.

[0105] In step S120, after receiving the real-time radar echo data of the micro-motion target, in step S130, the structured sparse optimization model of time-frequency representation that has been constructed, namely, formula (7) and formula (13), is directly used to enhance the time-frequency representation of the target, wherein the alternating direction multiplier method with a good balance between calculation accuracy and operation efficiency is used to perform alternating iterations to solve the structured sparse optimization model of time-frequency representation.

[0106] In this embodiment, when the alternating direction multiplier method is used to solve the structured sparse optimization model: firstly, a splitting variable is introduced to transform the structured sparse optimization model, and then Indicates the introduction of split variables, and let , then transform formula (13) into:

[0107] (14)

[0108] Furthermore, the corresponding augmented Lagrangian function is constructed according to formula (14), which is expressed as:

[0109] (15)

[0110] In formula (15), represents the inner product of two vectors, and denote the Lagrange multiplier and penalty term coefficient respectively, ,By introducing the Lagrange multiplier term, the target solution can be restricted to the corresponding feasible domain, which effectively improves the robustness of the algorithm.

[0111] Next, the Lagrange multiplier in the augmented Lagrangian function is Using the penalty coefficient Perform a scale transformation, , introduce the dual variable Transforming the augmented Lagrangian function, we get:

[0112] (16)

[0113] Therefore, according to the transformed augmented Lagrangian function (Formula (16)), the structured sparse optimization model is decomposed into multiple sub-problems for solution, and the final multiple solutions are obtained:

[0114] (17)

[0115] In formula (17), the superscript Represents the variable Iterations.

[0116] In this embodiment, when formula (17) is iteratively solved, in each iterative calculation process, the original variable y and the split variable z can be jointly minimized by using the above-mentioned alternating iterative operation, and the dual variable (Lagrange multiplier) v is updated by variables y and z, further accelerating the convergence speed of the algorithm. Among them, variable y is the enhanced time-frequency representation that is continuously reconstructed in multiple iterations.

[0117] Specifically, Figure 2 As shown, according to the splitting variable calculated in the previous iteration And the dual variable , solve the ridge regression problem containing data error terms and augmented Lagrangian terms to update the global variable , based on the dual variable calculated in the previous iteration And the global variables updated after the current iteration calculation , a complex soft threshold operator is derived based on the structured sparse regularization term and the augmented Lagrangian term to update the splitting variable , and finally calculate the updated global variables in conjunction with the current iteration and split variable minimization , to update the dual variable .

[0118] Furthermore, first, in the update of the original variable y, substitute formula (16) into the first formula in formula (17), omit the terms irrelevant to the variable y, and the minimization of the variable y can be equivalent to calculating a ridge regression problem, that is:

[0119] (18)

[0120] In formula (18), represents a constant term that is independent of the variable y. From formula (18), it can be seen that minimizing the variable y k The corresponding objective function is a convex function. Further taking the partial derivative of the objective function and setting the partial derivative to zero, we can get the optimal value of the original variable y, which is expressed as:

[0121] (19)

[0122] In formula (19), represents the conjugate transpose operator, Indicates size The identity matrix of .

[0123] It can be seen from formula (19) that the solution of the original variable y involves large-scale matrix inversion, which has a heavy computational burden and is time-consuming, making it difficult to apply in actual engineering. In order to reduce the amount of algorithm calculation, firstly, according to the Matrix Inversion Lemma (MIL), formula (19) is further simplified to:

[0124] (20)

[0125] Furthermore, we can use the observation matrix The special structure of avoids matrix inversion, , according to the downsampling matrix shown in formula (5) and sliding window matrix , it can be deduced that , In addition, combined with the unitary matrix properties, for the block inverse Fourier transform matrix , it can be deduced that Therefore, it can be found that the observation matrix It has row orthogonality, that is . Therefore, the update of the original variable y can be expressed as:

[0126] (twenty one)

[0127] In this embodiment, when updating the split variable z, that is, solving the structured sparse feature, formula (16) is substituted into the second formula in formula (17), and the irrelevant terms of the split variable z are omitted, and the following is obtained:

[0128] (twenty two)

[0129] According to formula (22), Introducing structured masks and The norm represents the structured sparse prior of the target, and the optimization problem of the split variable z can be regarded as Solving the norm regularization problem. Similar to the solution of the original variable y, using gradient descent, the update of the split variable z can be expressed as The neighboring operator problem of the norm can be solved by using the soft threshold operator:

[0130] (twenty three)

[0131] In formula (23), represents the complex soft threshold operator. For any complex variable and set real number threshold , The introduction of the complex soft threshold operator can keep the data's amplitude angle unchanged and has phase preservation. At the same time, since the threshold vector m has a lower threshold in the time-frequency curve area and a higher threshold in the background area, the continuous structure of the curve on the time-frequency surface can be retained, and the structural features and sparse features of the target time-frequency representation can be jointly enhanced.

[0132] In this embodiment, the original variable and the split variable are finally combined to update the dual variable, that is, the Lagrange multiplier v is updated, and the update formula is:

[0133] (twenty four)

[0134] By introducing the matrix inversion lemma and combining the special structure of the observation matrix, formula (21) avoids large-scale matrix inversion. In this case, the computational burden of updating the original variable y mainly comes from matrix and vector multiplication.

[0135] In another embodiment, in order to further reduce the computational complexity, before iteratively solving the multiple sub-problems, each sub-problem represented by a vector is converted into a matrix form.

[0136] Specifically, considering that the downsampling matrix G is mainly composed of 0 and 1 elements, the sliding window matrix subarray The block inverse Fourier transform matrix P has a diagonal structure, which can further transform the matrix-vector multiplication into the matrix element point multiplication, and the original variable y is quickly updated and solved in matrix form, which can be specifically expressed as:

[0137] (25)

[0138] (26)

[0139] (27)

[0140] This turns matrix-vector calculations into matrix element operations:

[0141] (28)

[0142] In formula (25) to formula (28), , and Corresponding vectors , and In matrix form, Restore the original matrix from the stacked vectors. Represents the rearranged form of the sliding window matrix, consisting of each sub-window matrix The diagonal elements of are arranged in rows, express The complex conjugate of . represents the residual signal matrix, Indicates size A vector of all 1s, is the Hadamard product, It means summing the second dimension of the matrix, that is, summing the matrix elements row by row. is the inverse Fourier transform matrix, express To further improve the computational efficiency, and It can be implemented by FFT and IFFT operations.

[0143] Furthermore, the splitting variable z and the Lagrange multiplier v are iteratively updated in matrix form, which can be specifically expressed as:

[0144] (29)

[0145] (30)

[0146] Therefore, after converting each sub-problem represented by a vector into a matrix form, the Gauss-Seidel idea is used to alternately iterate formula (29) to formula (30) to solve the enhanced time-frequency representation.

[0147] In this embodiment, when multiple sub-problems are solved alternately and iteratively, the final target time-frequency representation is obtained by setting the number of iterations or the error accuracy; wherein, when the number of iteration rounds reaches a preset number or the adjacent iteration error reaches a preset value, the reconstructed time-frequency representation obtained by this round of iteration is to realize the structured short-time sparse time-frequency enhanced representation.

[0148] Specifically, when performing alternating iterative solutions, set the iterative convergence condition, that is, reach the maximum number of iterations Or adjacent iteration error ,in is the iteration error threshold. When the convergence condition is met, the iteration ends and the time-frequency enhanced representation of the target is output. .

[0149] like Figure 3 As shown in FIG. 1 , each subproblem represented by a vector is converted into a matrix form, and then the alternating direction multiplication method is used to perform a specific iterative update process.

[0150] In this paper, the effectiveness of this method is also demonstrated through experimental simulation.

[0151] like Figure 4 As shown in Figure 2, the time-frequency representation enhancement results of simulated cone targets under different sparse sampling rates are provided. Figure 4It can be found that with the decrease of sparse sampling rate, the energy concentration of time-frequency representation obtained by STFT gradually deteriorates. When the sparse sampling rate drops to 30%, the time-frequency representation corresponding to STFT shows serious defocusing phenomenon, which damages the structure of time-frequency curve and generates many false points. In contrast, the proposed method can effectively suppress false points and enhance sparse features when the sparse sampling rate is low. At the same time, the proposed method better retains the structure of time-frequency curve while reducing background noise, showing better focusing performance.

[0152] like Figure 5 As shown in Figure 1, the enhanced time-frequency characterization results of metal cones measured in darkroom at different sparse sampling rates are provided. Figure 5 It can be seen that with the decrease of sparse sampling rate, the structure of the target time-frequency curve obtained by STFT becomes less and less clear. In contrast, the proposed method still retains the structural characteristics of the target time-frequency curve well, and the curve structure is continuous and obvious.

[0153] According to the above experimental results, it can be effectively proved that the method proposed in this paper has significant advantages in enhancing sparse features and structural features, and can effectively enhance the time-frequency representation of the target in the absence of echo data, and has high engineering application value.

[0154] In the above-mentioned time-frequency representation enhancement method based on structured short-time sparse representation, a two-dimensional time-frequency enhancement representation method based on structured short-time sparse representation is proposed to solve the problem of low resolution and performance degradation of traditional time-frequency representation methods under complex observation conditions. By establishing a sparse observation model of time-frequency representation, the sparse observation model of time-frequency representation is introduced. The sparse features of the target are characterized by the convolution norm, and on this basis, the structural continuous prior of the target is characterized by combining convolution weighting and morphological filtering, thereby realizing the structured sparse prior modeling of time-frequency representation. Furthermore, the optimization solution of structured short-time sparse enhanced representation is realized based on the ADMM optimization framework. The proposed method avoids matrix inversion by utilizing the special structure of the observation matrix, and uses low-dimensional matrix element point multiplication instead of high-dimensional matrix multiplication, which effectively reduces the computational complexity. Experimental results show that compared with traditional time-frequency representation, the proposed method achieves the synergistic enhancement of target structural features and sparse features under complex observation conditions, has better time-frequency representation resolution, and improves the representation accuracy of target micro-Doppler features under complex observation scenarios.

[0155] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0156] In one embodiment, Figure 6 As shown, a time-frequency representation enhancement device based on structured short-time sparse representation is provided, comprising: a time-frequency representation sparse observation model construction module 200, a model optimization module 210, a real-time data acquisition module 220 and a time-frequency enhancement module 230, wherein:

[0157] The time-frequency representation sparse observation model building module 200 is used to establish a time-frequency representation sparse observation model of radar echo data based on the relationship between the radar echo data model and the time-frequency representation to be solved;

[0158] Model optimization module 210, for utilizing norm, convolution weighting and morphological filtering, constraining the sparse and structural continuity characteristics of the target in the time-frequency representation sparse observation model, and obtaining a structured sparse optimization model of time-frequency representation;

[0159] A real-time data acquisition module 220 is used to acquire real-time radar echo data of micro-moving targets;

[0160] The time-frequency enhancement module 230 is used to input the real-time radar echo data into the structured sparse optimization model, and then iteratively solve the structured sparse optimization model using an alternating direction multiplier method to obtain a two-dimensional time-frequency representation of the micro-motion target with sparse and structural features jointly enhanced.

[0161] For the specific limitations of the time-frequency characterization enhancement device based on structured short-time sparse representation, please refer to the limitations of the time-frequency characterization enhancement method based on structured short-time sparse representation above, which will not be repeated here. Each module in the above-mentioned time-frequency characterization enhancement device based on structured short-time sparse representation can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0162] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 7As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a time-frequency representation enhancement method based on structured short-time sparse representation is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0163] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0164] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0165] Based on the relationship between the radar echo data model and the time-frequency representation to be solved, a sparse observation model of time-frequency representation of radar echo data is established;

[0166] use norm, convolution weighting and morphological filtering, constraining the sparse and structural continuity characteristics of the target in the time-frequency representation sparse observation model, and obtaining a structured sparse optimization model of time-frequency representation;

[0167] Acquire real-time radar echo data of micro-moving targets;

[0168] After the real-time radar echo data is input into the structured sparse optimization model, the structured sparse optimization model is iteratively solved by using the alternating direction multiplier method to obtain a two-dimensional time-frequency representation of the micro-motion target with sparse and structural features jointly enhanced.

[0169] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0170] Based on the relationship between the radar echo data model and the time-frequency representation to be solved, a sparse observation model of time-frequency representation of radar echo data is established;

[0171] use norm, convolution weighting and morphological filtering, constraining the sparse and structural continuity characteristics of the target in the time-frequency representation sparse observation model, and obtaining a structured sparse optimization model of time-frequency representation;

[0172] Obtain real-time radar echo data of micro-moving targets;

[0173] After the real-time radar echo data is input into the structured sparse optimization model, the structured sparse optimization model is iteratively solved by using the alternating direction multiplier method to obtain a two-dimensional time-frequency representation of the micro-motion target with sparse and structural features jointly enhanced.

[0174] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0175] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are 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.

[0176] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof 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 a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A time-frequency representation enhancement method based on structured short-term sparse representation, characterized in that: The method comprises: Based on the relationship between the radar echo data model and the time-frequency representation to be solved, a sparse observation model of time-frequency representation of radar echo data is established; use norm, convolution weighting and morphological filtering, constraining the sparse and structural continuity characteristics of the target in the time-frequency representation sparse observation model, and obtaining a structured sparse optimization model of time-frequency representation; Acquire real-time radar echo data of micro-moving targets; After the real-time radar echo data is input into the structured sparse optimization model, the structured sparse optimization model is iteratively solved by using the alternating direction multiplier method to obtain a two-dimensional time-frequency representation of the micro-motion target with sparse and structural features jointly enhanced.

2. The time-frequency representation enhancement method based on structured short-term sparse representation according to claim 1, characterized in that: The time-frequency representation sparse observation model is expressed as: ; In the above formula, represents the reconstructed time-frequency representation vector, represents the defect echo observation vector, i.e., the radar echo data model, Represents the downsampling matrix, which consists of elements 0 and 1, where the echo element sampling position is 1 and the missing position is 0. represents the inverse short-time Fourier transform matrix, where represents the sliding window matrix, , is a diagonal matrix, represents the block inverse Fourier transform matrix, represents the observation noise vector.

3. The time-frequency representation enhancement method based on structured short-term sparse representation according to claim 2 is characterized in that: The structured sparse optimization model of the time-frequency representation is expressed as: ; In the above formula, and Respectively Norm and norm, Represents the data error term, which is used to measure the similarity between the reconstructed time-frequency representation and the original time-frequency representation. Represents the structural weighted vector, which is obtained by the structural weighted matrix Stacking along the frequency dimension gives: , , represents the modified structured sparse prior regularization term.

4. The time-frequency representation enhancement method based on structured short-term sparse representation according to claim 3 is characterized in that: The structural weighted matrix is ​​constructed by convolution weighted operation and morphological filtering, and is expressed as: ; In the above formula, Indicates size All 1 matrices, represents matrix element-wise division, represents any minimum value, represents the target structure mask obtained by convolution operation, Represents the use of morphological operators Make corrections.

5. The time-frequency representation enhancement method based on structured short-term sparse representation according to any one of claims 1 to 4, characterized in that: When solving a structured sparse optimization model using the alternating direction multiplier method: After introducing a splitting variable to transform the structured sparse optimization model, a corresponding augmented Lagrangian function is constructed; Performing a scale transformation on the Lagrangian multiplier in the augmented Lagrangian function using a penalty term coefficient, and introducing a dual variable to transform the augmented Lagrangian function; According to the transformed augmented Lagrangian function, the structured sparse optimization model is decomposed into multiple sub-problems for solving.

6. The time-frequency representation enhancement method based on structured short-term sparse representation according to claim 5, characterized in that: Before iteratively solving multiple sub-problems, the special structure of the observation matrix is ​​used to transform the matrix-vector multiplication into the matrix element dot product, and the variables are iteratively updated in matrix form.

7. The time-frequency representation enhancement method based on structured short-term sparse representation according to claim 6, characterized in that: Each iterative calculation process when solving multiple sub-problems in alternating iterations: According to the splitting variables and dual variables calculated in the previous iteration, the ridge regression problem including data error terms and augmented Lagrangian terms is solved to update the global variables; According to the dual variables calculated in the previous iteration and the global variables updated in the current iteration, a complex soft threshold operator is derived based on the structured sparse regularization term and the augmented Lagrangian term to update the splitting variables; Finally, the updated global variables and split variables are minimized by combining the current iteration calculations to update the dual variables.

8. The time-frequency representation enhancement method based on structured short-term sparse representation according to claim 7, characterized in that: When solving multiple sub-problems in alternating iterations, the final target time-frequency enhanced representation is obtained by setting the number of iterations or the error precision; Among them, when the number of iteration rounds reaches a preset number or the adjacent iteration error reaches a preset value, the reconstructed time-frequency representation obtained by this round of iteration is to realize the structured short-time sparse time-frequency enhanced representation.

9. A time-frequency representation enhancement device based on structured short-term sparse representation, characterized in that: The device comprises: A time-frequency representation sparse observation model building module is used to build a time-frequency representation sparse observation model of radar echo data based on the relationship between the radar echo data model and the time-frequency representation to be solved; Model optimization module for utilizing norm, convolution weighting and morphological filtering, constraining the sparse and structural continuity characteristics of the target in the time-frequency representation sparse observation model, and obtaining a structured sparse optimization model of time-frequency representation; Real-time data acquisition module, used to obtain real-time radar echo data of micro-moving targets; The time-frequency enhancement module is used to input the real-time radar echo data into the structured sparse optimization model, and then iteratively solve the structured sparse optimization model using an alternating direction multiplier method to obtain a two-dimensional time-frequency representation of the micro-motion target with sparse and structural features jointly enhanced.

Citation Information

Patent Citations

  • Micro-motion target time-frequency graph reconstruction method based on weighted ADMM

    CN113093140A

  • Sparse aperture target structured ISAR imaging method and device with micro-motion component

    CN117890908A

  • Image processing method, system, device and storage medium

    US20240029204A1