A multi-band ISAR high-resolution imaging method, device and equipment

Abstract

This invention provides a multi-band ISAR high-resolution imaging method, apparatus, and device. The method includes: performing sparse coarse estimation in the discrete domain using a sparse Bayesian learning method to obtain initial discrete coordinate coarse estimation vectors and initial complex amplitude coarse estimation vectors; constructing a nonlinear least squares cost function using a variable projection method based on a multi-band joint echo theory model and the initial complex amplitude coarse estimation vectors; optimizing the discrete coordinate vectors in the nonlinear least squares cost function within the corresponding neighborhood using the initial discrete coordinate coarse estimation vectors to obtain target coordinate estimation vectors; updating the continuous manifold dictionary in the multi-band joint echo theory model using the target coordinate estimation vectors to calculate the updated complex scattering coefficient vectors; and generating a final high-resolution ISAR image using the target coordinate estimation vectors and the updated complex scattering coefficient vectors. This achieves high-resolution multi-band ISAR imaging.

Description

Technical Field

[0001] This invention relates to the field of inverse synthetic aperture radar imaging technology, specifically to a multi-band ISAR high-resolution imaging method, apparatus, and equipment. Background Technology

[0002] In the field of Inverse Synthetic Aperture Radar (ISAR) imaging technology, acquiring high-resolution images of space targets typically requires transmitting wide-bandwidth signals. However, limited by the hardware and spectrum resources of a single radar, its instantaneous bandwidth often falls short of the requirements for detecting the fine structure of the target. Therefore, multi-band fusion technology, which transmits multiple narrow-band signals with different center frequencies and discrete spectra, and then coherently synthesizes them in the frequency domain to form a large equivalent bandwidth, has become an important approach to overcoming the resolution bottleneck. However, achieving high-precision coherent fusion of cross-band signals and accurately reconstructing the target image remains the core signal processing challenge currently facing this technology.

[0003] To achieve high-resolution imaging from incomplete frequency domain sampling data, sparse reconstruction methods based on compressed sensing have been introduced into this field. Among them, the gridded sparse recovery method based on a discretized dictionary constructs a sensing matrix through a pre-defined distance grid and reconstructs the target using its sparsity, exhibiting high computational efficiency and strong robustness to noise. On the other hand, to fundamentally avoid errors caused by grid discretization, a gridless sparse recovery method has been proposed. This method directly models and optimizes the target parameters in a continuous parameter space, theoretically possessing the potential to achieve super-resolution.

[0004] However, the aforementioned existing technologies still have significant shortcomings when applied to practical multi-band, large-span imaging scenarios. Mesh-based methods heavily rely on pre-defined discrete grids, and when the target's true position deviates from the grid center, an inherent off-grid error occurs. Under multi-band, large-span synthesis conditions, this minute geometric positional deviation translates into severe nonlinear phase error in the frequency domain, disrupting the coherence between signals from different frequency bands and resulting in defocusing and false targets in the fused image. Attempting to suppress this error by significantly increasing the grid density leads to excessively high computational complexity, making it difficult to meet real-time processing requirements. While meshless methods can avoid off-grid errors, their solutions typically involve high-dimensional convex optimization, resulting in huge memory consumption and slow convergence speed, making them unsuitable for handling high-dimensional observation data. Therefore, existing technologies struggle to balance effectively correcting cross-band phase mismatch to maintain signal coherence with controlling the computational complexity of the algorithm. Summary of the Invention

[0005] To address the aforementioned problems in the prior art, this invention provides a multi-band ISAR high-resolution imaging method, apparatus, and device.

[0006] The technical problem to be solved by this invention is achieved through the following technical solution: In a first aspect, the present invention provides a multi-band ISAR high-resolution imaging method, comprising: Acquire echo signals and corresponding frequency sampling points of the ISAR system in multiple non-overlapping frequency bands; The echo signal and frequency sampling points are vertically stitched together to obtain the global observation vector and global frequency vector. A multi-band joint echo theoretical model is constructed based on global frequency vector and point scattering model; The discrete sensing dictionary matrix is ​​obtained by using the global frequency vector to perform grid discretization in the distance space. Based on the discrete sensing dictionary matrix and the global observation vector, the discrete domain sparse coarse estimation is performed using the sparse Bayesian learning method to obtain the initial discrete coordinate coarse estimation vector and the initial complex amplitude coarse estimation vector. Based on the multi-band joint echo theoretical model and the initial complex amplitude coarse estimation vector, a nonlinear least squares cost function is constructed using the variable projection method. Using the initial discrete coordinate coarse estimation vector as the initial value, the discrete coordinate vector in the nonlinear least squares cost function is optimized and solved within the corresponding neighborhood range to obtain the target coordinate estimation vector; The dictionary of continuous manifolds in the multi-band joint echo theoretical model is updated using the target coordinate estimation vector, and the updated complex scattering coefficient vector is calculated. The final high-resolution ISAR image is generated using the target coordinate estimation vector and the updated complex scattering coefficient vector.

[0007] In a second aspect, the present invention provides a multi-band ISAR high-resolution imaging device, which includes: an acquisition unit, a stitching processing unit, a model building unit, a sparse estimation unit, an optimization solution unit, an update unit, and an output unit. The acquisition unit is used to acquire echo signals of the ISAR system in multiple non-overlapping frequency bands and the corresponding frequency sampling points; The splicing processing unit is used to perform vertical splicing processing on the echo signal and frequency sampling points respectively, to obtain the global observation vector and the global frequency vector. The model building unit is used to construct a multi-band joint echo theoretical model based on the global frequency vector and the point scattering model. The sparse estimation unit is used to perform grid discretization processing in the distance space using the global frequency vector to obtain a discrete sensing dictionary matrix. Based on the discrete sensing dictionary matrix and the global observation vector, a sparse Bayesian learning method is used to perform sparse coarse estimation processing in the discrete domain to obtain the initial discrete coordinate coarse estimation vector and the initial complex amplitude coarse estimation vector. The model building unit is also used to construct a nonlinear least squares cost function by using the variable projection method through a multi-band joint echo theory model and an initial complex amplitude coarse estimation vector. The optimization unit is used to optimize the discrete coordinate vector in the nonlinear least squares cost function within the corresponding neighborhood range, using the initial discrete coordinate coarse estimation vector as the initial value, to obtain the target coordinate estimation vector; The update unit is used to update the continuous manifold dictionary in the multi-band joint echo theoretical model using the target coordinate estimation vector, and calculate the updated complex scattering coefficient vector. The output unit is used to generate the final high-resolution ISAR image using the target coordinate estimation vector and the updated complex scattering coefficient vector.

[0008] Thirdly, the present invention provides a multi-band ISAR high-resolution imaging device, comprising: a processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the multi-band ISAR high-resolution imaging device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the multi-band ISAR high-resolution imaging method as described in the first aspect above.

[0009] This invention provides a multi-band ISAR high-resolution imaging method, apparatus, and device. The multi-band ISAR high-resolution imaging method includes: acquiring echo signals and corresponding frequency sampling points of an ISAR system in multiple non-overlapping frequency bands; vertically stitching the echo signals and frequency sampling points to obtain a global observation vector and a global frequency vector; constructing a multi-band joint echo theoretical model based on the global frequency vector and a point scattering model; performing grid discretization processing on the global frequency vector in the range space to obtain a discrete sensing dictionary matrix; and performing sparse coarse estimation processing in the discrete domain using a sparse Bayesian learning method based on the discrete sensing dictionary matrix and the global observation vector to obtain an initial discrete... The invention employs a coarse estimation vector for the target coordinates and an initial coarse estimation vector for the complex amplitude. Based on a multi-band joint echo theory model and the initial coarse estimation vector for the complex amplitude, a nonlinear least squares cost function is constructed using the variable projection method. Using the initial discrete coarse estimation vector for the target coordinates as the initial value, the discrete coordinate vectors in the nonlinear least squares cost function are optimized within the corresponding neighborhood to obtain the target coordinate estimation vector. The continuous manifold dictionary in the multi-band joint echo theory model is updated using the target coordinate estimation vector, and the updated complex scattering coefficient vector is calculated. The final high-resolution ISAR image is generated using the target coordinate estimation vector and the updated complex scattering coefficient vector. In this invention, firstly, sparse Bayesian learning is used for discrete-domain sparse coarse estimation, obtaining the initial target position and amplitude estimates with lower computational complexity, providing a good starting point for subsequent optimization, and solving the problem of excessively high computational complexity caused by a significant increase in grid density in gridding methods. Then, based on the initial estimate, a nonlinear least squares cost function is constructed using the variable projection method, and continuous optimization is performed on the discrete coordinates within the neighborhood. This avoids the inherent off-grid error of the pre-set grid and solves the problems of nonlinear phase error and signal coherence degradation caused by geometric position deviations in multi-band applications. Finally, the optimized coordinates are used to update the continuous manifold dictionary and calculate the complex scattering coefficients to generate a high-resolution ISAR image. This integrates multi-band information and solves the challenges of huge memory consumption and slow convergence speed caused by high-dimensional convex optimization in meshless methods. In summary, this invention effectively corrects cross-band phase mismatch to maintain signal coherence while controlling the computational complexity of the algorithm, achieving high-resolution imaging of multi-band ISAR.

[0010] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0011] Figure 1 A flowchart illustrating a multi-band ISAR high-resolution imaging method provided in an embodiment of the present invention; Figure 2 An exemplary schematic diagram comparing the multi-band fusion imaging method of the present invention with the results of traditional single-band imaging is shown; Figure 3An exemplary schematic diagram illustrates the accuracy and efficiency of the method of the present invention in estimating the distance to five targets in three independent runs; Figure 4 This is a schematic diagram of the structure of a multi-band ISAR high-resolution imaging device provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of a multi-band ISAR high-resolution imaging device provided in an embodiment of the present invention. Detailed Implementation

[0012] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0013] To achieve high-resolution imaging of multi-band ISAR, this invention provides a method for high-resolution imaging of multi-band ISAR. Figure 1 This is a flowchart illustrating a multi-band ISAR high-resolution imaging method provided in an embodiment of the present invention, as shown below. Figure 1 As shown, it includes: S101. Obtain the echo signals of the ISAR system in multiple non-overlapping frequency bands and the corresponding frequency sampling points.

[0014] S102. Vertically stitch the echo signal and frequency sampling points to obtain the global observation vector and global frequency vector.

[0015] Assuming the radar operates at To leverage the high-resolution advantages of full-bandwidth synthesis across multiple non-overlapping discrete frequency bands, the data from each band must be vectorized and integrated. echo signals in each frequency band With the corresponding frequency sampling points Construct global observation vectors by vertical concatenation and global frequency vector : ; in It is the total number of sampling points across the entire frequency band. Indicates the first The echo signal corresponding to each discrete frequency band. Indicates the first Frequency sampling points corresponding to each discrete frequency band Represents the set of complex numbers. It represents the set of real numbers.

[0016] S103. Construct a multi-band joint echo theoretical model based on global frequency vector and point scattering model.

[0017] Alternatively, the multi-band joint echo theoretical model can be expressed as: ; ; in, This represents the vector of the multi-band joint echo theoretical model. Indicates the first One target (scattering center), Indicates the total number of targets. Indicates the first The true target coordinates of each target This represents multi-band combined additive white Gaussian noise. Indicates the first The complex scattering coefficients of a target, Represents a vector consisting of the coordinates of all real targets. express The corresponding dictionary of continuous manifolds, Represents the imaginary unit. This represents a column vector consisting of the complex magnitudes of all targets. Represents the global frequency vector. This represents the initial reference distance when discretizing the distance space. It represents the speed of light.

[0018] S104. The global frequency vector is used to perform grid discretization in the distance space to obtain the discrete sensing dictionary matrix. Based on the discrete sensing dictionary matrix and the global observation vector, the sparse Bayesian learning method is used to perform sparse coarse estimation in the discrete domain to obtain the initial discrete coordinate coarse estimation vector and the initial complex amplitude coarse estimation vector.

[0019] Optionally, S104 includes: determining the theoretical resolution of the distance space based on the spectral range covered by the global frequency vector, and uniformly discretizing the distance space with the theoretical resolution as the step size to generate a discrete distance grid point vector; Construct a discrete sensing dictionary matrix based on the global frequency vector and the discrete distance grid point vector; A sparse linear model is established using global observation vectors and discrete sensing dictionary matrices; Based on a sparse linear model and discrete distance grid point vectors, iterative optimization is performed using a sparse Bayesian learning method and an expectation-maximization algorithm to obtain an initial discrete coordinate coarse estimation vector and an initial complex amplitude coarse estimation vector.

[0020] Optionally, based on a sparse linear model and discrete distance grid point vectors, iterative optimization is performed using a sparse Bayesian learning method and an expectation-maximization algorithm to obtain an initial coarse estimation vector of discrete coordinates and an initial coarse estimation vector of complex magnitudes, including: A sparse Bayesian learning method is used to set a sparse prior distribution for the scattering coefficients at each position of the discrete distance grid point vector. Based on the global observation vector and the sparse prior distribution, the posterior distribution of the scattering coefficients is obtained by using Bayes' theorem. The posterior distribution of the scattering coefficients and the hyperparameters of the sparse linear model are iteratively updated using the expectation-maximization algorithm until the first iteration convergence condition is met. The amplitude threshold is set using the statistic of the posterior distribution of the scattering coefficients when the first iteration convergence condition is met. From all scattering coefficients corresponding to discrete distance grid point vectors, scattering coefficients with posterior amplitudes exceeding an amplitude threshold are selected as significant scattering centers. The positions in the discrete distance grid point vectors corresponding to the significant scattering centers and the corresponding posterior mean values ​​in the posterior distribution of the scattering coefficients are used as the initial discrete coordinate coarse estimation vector and the initial complex amplitude coarse estimation vector, respectively.

[0021] Optionally, the convergence condition for the first iteration is that the change in the posterior distribution of the scattering coefficients or the hyperparameters of the sparse linear model is continuously less than a preset threshold.

[0022] In this embodiment, a one-dimensional imaging observation window (the theoretical resolution of the distance space) is set with a fixed spatial step size. Divide the distance space into A vector of discrete distance grid points is formed by uniformly distributed discrete distance grid points. , Discrete distance grid point vectors Substitute into the guide vector generation function In the process, a discrete sensing dictionary matrix is ​​constructed. : ; At this point, by discretizing the continuous manifold dictionary on a preset grid, the global observation vector... The mapping relationship between the discrete target scattering coefficients and the scattering coefficients can be established as the following sparse linear model: ; In the formula, This represents the sparse scattering coefficient vector on a discrete grid. Due to the actual number of scattering centers (targets)... Much smaller than the total number of grid cells , It exhibits strong sparsity.

[0023] To solve the equation Strongly sparse vectors in This invention employs a sparse Bayesian learning framework. This framework introduces an automatic correlation determination mechanism, using the measured global observation vectors... This is for condition-driven parameter iterative updates.

[0024] Suppose we want to find a strongly sparse vector Each element follows a complex Gaussian prior distribution with zero mean and independent variances: ; In the formula: Represents the prior variance matrix and strongly sparse vectors The corresponding probability distribution It is the first Discrete distance grid points The prior variance of the upscattering coefficient is used as a hyperparameter to control the sparsity at that point. Let represent the prior variance matrix (hyperparameter vector). It is a diagonal matrix composed of all hyperparameters, defined as follows: , Indicates the first The strongly sparse values ​​corresponding to discrete-distance grid points, and the hyperparameters corresponding to background grid points where no target exists when the algorithm converges iteratively. This will approach zero, thus statistically achieving the forced zeroing of the coefficients of the targetless grid, i.e., sparsity constraint.

[0025] In the algorithm's... In the next iteration, with For each input, alternately perform the following two core steps until the convergence condition is met: First, update the posterior statistics: fix the hyperparameter matrix of the current iteration. Calculate the grid scattering coefficient vector posterior covariance matrix With the posterior mean vector : ; ; In the formula, The variance parameter of the multi-band joint additive complex white Gaussian noise; It is the conjugate transpose of the discrete perceptron dictionary matrix. Represents in The most likely complex amplitude estimate at each discrete grid point under the constraint.

[0026] Then, hyperparameter updates are performed: using the posterior statistic calculated in the previous step, the hyperparameters controlling sparsity are updated by maximizing the marginal likelihood function of the data. ; In the formula, posterior mean vector The One element; For the posterior covariance matrix The The main diagonal elements represent the uncertainty of the magnitude estimation at that grid point. It should be noted that, in this embodiment, In , In and discrete distance grid points It can be understood as an index value, and the index values ​​are consistent.

[0027] Once the iterative process meets the preset threshold, the posterior mean vector is detected by setting an adaptive threshold. Extract the set of grid indices corresponding to the significantly non-zero elements in the matrix, and denote it as the effective support set. Assume the final number of detected targets is... Therefore, the final output of this stage is: Initial discrete coordinate coarse estimation vector Based on the index set From grid collection The corresponding discrete spatial coordinates extracted from them, i.e. .

[0028] Initial complex amplitude coarse estimation vector : The posterior mean at the corresponding grid location, i.e.: , Indicates the first Candidate target locations (i.e., grid points) The posterior mean estimate of the scattering coefficient at the corresponding location.

[0029] Due to the discretization effect of the preset grid, the output coordinate coarse estimation vector There is a grid quantization error, also known as off-grid error. Under full-band, wide-bandwidth observation conditions, this error can lead to severe cross-band phase mismatch and amplitude attenuation. Therefore, It can only be used as a high-confidence initial point for the target location. It needs to be substituted into the nonlinear least squares cost function of the next stage as the initial iteration value for refinement in the continuous domain.

[0030] S105. Based on the multi-band joint echo theoretical model and the initial complex amplitude coarse estimation vector, a nonlinear least squares cost function is constructed using the variable projection method.

[0031] Alternatively, the nonlinear least squares cost function can be expressed as: ; ; in, express The corresponding nonlinear least squares cost function, Represents a vector consisting of the coordinates of all real targets. Represents the global observation vector. express The corresponding dictionary of continuous manifolds, express The corresponding complex scattering coefficient vector estimation, Denotes the square of the L2 norm. express The corresponding Moore-Penrose pseudoinverse, This indicates the conjugate transpose.

[0032] S106. Using the initial discrete coordinate coarse estimation vector as the initial value, optimize the discrete coordinate vector in the nonlinear least squares cost function within the corresponding neighborhood range to obtain the target coordinate estimation vector.

[0033] Optionally, S106 includes: Obtain the neighborhood range corresponding to the initial discrete coordinate coarse estimation vector; The initial discrete coordinate coarse estimation vector is used as the initial value of the nonlinear least squares cost function, and the trust region reflection algorithm is used for optimization within the neighborhood until the second iteration convergence condition is met. The discrete coordinate vector that satisfies the convergence condition of the second iteration is used as the target coordinate estimation vector.

[0034] Extracted in the second stage This serves as the initial input to the optimization algorithm. To ensure the algorithm's robustness and prevent phase unwrapping ambiguity between multi-band gaps, constraint boundaries are introduced, constructing a constrained optimization problem: ; In the formula, To search for the lower bound; The upper bound of the search; The slack variable parameter is typically taken as the grid step size. One-half of.

[0035] The above formulas are solved iteratively using either Trust-Region-Reflective or Levenberg-Marquardt algorithms based on Jacobian matrices.

[0036] When the descent of the objective function or the parameter update meets the preset tolerance (i.e., the convergence condition of the second iteration is met), the algorithm terminates and outputs the target coordinate estimation vector with sub-grid accuracy. .

[0037] S107. Update the continuous manifold dictionary in the multi-band joint echo theoretical model using the target coordinate estimation vector, and calculate the updated complex scattering coefficient vector.

[0038] The precise coordinate vector obtained through the above process Substituting back into the continuous parameterized model of the first stage, the dictionary matrix of the continuous manifold is updated as follows: .

[0039] Then, using the formula The updated complex scattering coefficient vector is calculated. : .

[0040] S108. The final high-resolution ISAR image is generated using the target coordinate estimation vector and the updated complex scattering coefficient vector.

[0041] Optionally, S108 includes: Construct an ideal full-band manifold dictionary based on the target coordinate estimation vector; The ideal full-band signal is calculated using the ideal full-band manifold dictionary and the updated complex scattering coefficient vector. The initial ISAR image is obtained by performing a two-dimensional inverse Fourier transform on the ideal full-band signal; The initial ISAR image is normalized to obtain the final high-resolution ISAR image.

[0042] It should be noted that the final high-resolution ISAR image output in this step It achieved coherent energy accumulation at the limit of physical bandwidth, eliminated the high-intensity grating lobes and image defocusing caused by discrete spectrum, and completed high-quality super-resolution fusion imaging.

[0043] This invention proposes a hierarchical dual-band high-resolution fusion imaging method that progresses from coarse to fine. By organically combining discrete-domain global search with continuous-domain local fine-tuning, it systematically addresses the shortcomings of existing sparse reconstruction techniques. Specifically, this invention first fully leverages the advantages of sparse Bayesian learning—its strong global search capability and insensitivity to initial values—to rapidly lock the sparse topological skeleton of the target on a low-dimensional coarse grid. This perfectly avoids the non-convex trap of easily getting into local minima when directly using nonlinear optimization to process complex cross-coupling terms, ensuring that the algorithm can robustly converge to the global optimum under complex conditions. Second, this invention uses the extracted topological skeleton as a high-confidence prior initial value input to the nonlinear optimization module, performing continuous-domain dimensionality reduction iterations and micrometer-level ultra-high-precision positioning only on a few activated non-zero scattering points. This completely eliminates the off-grid phenomenon caused by grid quantization at the physical level. The error is precisely compensated for the amplified nonlinear phase rotation in multi-band large-span synthesis, ensuring perfect coherent accumulation of dual-band signals and eliminating image defocus and false sidelobes. Finally, with the above-mentioned "qualitative first, quantitative later" collaborative mechanism, this invention completely breaks through the pressure of semi-definite programming inversion faced by traditional meshless methods, as well as the computational power shackles of conventional discrete reconstruction relying on extremely high-density fine grids. With an extremely low computational cost far less than global fine grid search, it achieves physical-level super-resolution imaging that surpasses traditional theories, greatly reducing the amount of computation and memory usage, and has great engineering potential for real-time operation on resource-constrained computing platforms such as spaceborne or airborne platforms.

[0044] In addition, other algorithms that achieve global discrete sparse reconstruction and initial topological skeleton extraction can also be used in this invention, such as the Orthogonal Matching Pursuit (OMP) algorithm, based on... Convex optimization algorithms that minimize norm (such as basis pursuit backpropagation) or fast adaptive iterative shrinking threshold algorithm (FASTA) replace the sparse Bayesian learning (SBL) algorithm in this paper.

[0045] Other algorithms that achieve subgrid continuous domain refinement and off-grid error correction, such as the RELAX (Relax) algorithm and its variants, and the Alternating Projection algorithm, are used to replace the nonlinear least squares (NLLS) algorithm in this paper.

[0046] Furthermore, to verify the effectiveness of the multi-band ISAR high-resolution imaging method provided by this invention, simulation experiments were conducted. Specifically, comparative experiments were performed using the method of this invention and a traditional single-band imaging method under the same hardware platform and experimental scenario. Figure 2 An exemplary schematic diagram comparing the results of the multi-band fusion imaging method of the present invention with those of traditional single-band imaging is shown. Figure 2As shown, the main lobe width of the one-dimensional range image obtained by using only low-frequency (200MHz) data is significantly large, resulting in insufficient resolution. However, the reconstructed curve after fusing dual-frequency data using the method of this invention (SBL+NLLS) is highly consistent with the theoretical ideal full-bandwidth signal result curve in terms of amplitude and position. This indicates that the present invention effectively achieves coherent synthesis of cross-frequency band signals, suppresses grating lobes caused by spectral gaps, and achieves a limit resolution comparable to that of a full-bandwidth signal.

[0047] Figure 3 An exemplary diagram illustrates the accuracy and efficiency of the method of the present invention in estimating the distance to five targets in three independent runs. Wherein, Figure 3 Figure (a) shows a schematic diagram comparing the coarse and fine distance estimates with the actual distance obtained from the first run, which took 0.0850 seconds. Figure 3 Figure (b) shows a schematic diagram illustrating the comparison results of the second run taking 0.0759 seconds. Figure 3 Figure (c) illustrates the comparison results for the third run, which took 0.0805 seconds. From... Figure 3 It can be seen that the results of the three runs are highly consistent and stable: the coarse distance estimate based on the discrete grid has a significant error due to grid quantization (typical error of about 0.01 meters), while the fine distance estimate after continuous domain nonlinear refinement is within 10 meters of the true distance. -7 The results are perfectly matched at the meter level, achieving sub-millimeter level ultra-high precision positioning; at the same time, the time for a single complete processing is less than 0.1 seconds, verifying that the method of the present invention has efficient and stable computing performance without sacrificing accuracy.

[0048] This invention provides a multi-band ISAR high-resolution imaging method. First, sparse Bayesian learning is used for discrete-domain sparse coarse estimation, obtaining initial target position and amplitude estimates with low computational complexity, providing a good starting point for subsequent optimization and solving the problem of excessive computational complexity caused by significantly increasing grid density in gridded methods. Then, based on the initial estimate, a nonlinear least-squares cost function is constructed using variable projection, and continuous optimization is performed on the discrete coordinates within the neighborhood, avoiding the inherent off-grid error of the pre-set grid and solving the problems of nonlinear phase error and signal coherence degradation caused by geometric position deviations in multi-band applications. Finally, the optimized coordinates are used to update the continuous manifold dictionary and calculate the complex scattering coefficients to generate a high-resolution ISAR image, integrating multi-band information and overcoming the challenges of huge memory consumption and slow convergence speed caused by high-dimensional convex optimization in gridless methods. In summary, this invention effectively corrects cross-band phase mismatch to maintain signal coherence while controlling the algorithm's computational complexity, achieving high-resolution multi-band ISAR imaging.

[0049] The method provided in this embodiment of the invention can be applied to electronic devices. Specifically, the electronic device can be a desktop computer, a portable computer, a smart mobile terminal, a server, etc., and this embodiment of the invention does not limit the application to such devices.

[0050] Based on the same inventive concept, embodiments of the present invention also provide a multi-band ISAR high-resolution imaging device. Figure 4 This is a schematic diagram of the structure of a multi-band ISAR high-resolution imaging device provided in an embodiment of the present invention, as shown below. Figure 4 As shown, it includes: an acquisition unit 401, a splicing processing unit 402, a model building unit 403, a sparse estimation unit 404, an optimization solution unit 405, an update unit 406, and an output unit 407. Acquisition unit 401 is used to acquire echo signals of the ISAR system in multiple non-overlapping frequency bands and the corresponding frequency sampling points; The splicing processing unit 402 is used to perform vertical splicing processing on the echo signal and the frequency sampling points respectively, so as to obtain the global observation vector and the global frequency vector. Model building unit 403 is used to build a multi-band joint echo theoretical model based on the global frequency vector and point scattering model; The sparse estimation unit 404 is used to perform grid discretization processing in the distance space using the global frequency vector to obtain a discrete sensing dictionary matrix. Based on the discrete sensing dictionary matrix and the global observation vector, a sparse Bayesian learning method is used to perform sparse coarse estimation processing in the discrete domain to obtain the initial discrete coordinate coarse estimation vector and the initial complex amplitude coarse estimation vector. The model building unit 403 is also used to construct a nonlinear least squares cost function by using the variable projection method through a multi-band joint echo theory model and an initial complex amplitude coarse estimation vector. The optimization unit 405 is used to optimize the discrete coordinate vector in the nonlinear least squares cost function within the corresponding neighborhood range, using the initial discrete coordinate coarse estimation vector as the initial value, to obtain the target coordinate estimation vector. Update unit 406 is used to update the continuous manifold dictionary in the multi-band joint echo theoretical model using the target coordinate estimation vector, and calculate the updated complex scattering coefficient vector. Output unit 407 is used to generate the final high-resolution ISAR image using the target coordinate estimation vector and the updated complex scattering coefficient vector.

[0051] Figure 5This is a schematic diagram of a multi-band ISAR high-resolution imaging device provided in an embodiment of the present invention. It includes a processor 710, a storage medium 720, and a bus 730. The storage medium 720 stores machine-readable instructions executable by the processor 710. When the multi-band ISAR high-resolution imaging device is running, the processor 710 communicates with the storage medium 720 via the bus 730. The processor 710 executes the machine-readable instructions to perform the steps of the above-described method embodiment. Specific implementation methods and technical effects are similar and will not be described in detail here.

[0052] The storage medium may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the storage medium may also be at least one storage device located remotely from the aforementioned processor.

[0053] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0054] It should be noted that the terms "first," "second," etc., are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention.

[0055] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings and the disclosure, will understand and implement other variations of the disclosed embodiments in carrying out the claimed invention. In this description, the word "comprising" does not exclude other components or steps, "a" or "an" does not exclude a plurality, and "a plurality" means two or more, unless otherwise explicitly specified. Furthermore, while different embodiments may describe certain measures, this does not mean that these measures cannot be combined to produce good results.

[0056] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the inventive concept, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A multi-band ISAR high-resolution imaging method, characterized in that, include: Acquire echo signals and corresponding frequency sampling points of the ISAR system in multiple non-overlapping frequency bands; The echo signal and the frequency sampling points are vertically spliced ​​together to obtain the global observation vector and the global frequency vector. A multi-band joint echo theoretical model is constructed based on the global frequency vector and the point scattering model. The global frequency vector is used to perform grid discretization in the distance space to obtain a discrete sensing dictionary matrix. Based on the discrete sensing dictionary matrix and the global observation vector, a sparse Bayesian learning method is used to perform sparse coarse estimation in the discrete domain to obtain an initial discrete coordinate coarse estimation vector and an initial complex amplitude coarse estimation vector. Based on the multi-band joint echo theoretical model and the initial complex amplitude coarse estimation vector, a nonlinear least squares cost function is constructed using the variable projection method. Using the initial discrete coordinate coarse estimation vector as the initial value, the discrete coordinate vector in the nonlinear least squares cost function is optimized and solved within the corresponding neighborhood range to obtain the target coordinate estimation vector; The dictionary of continuous manifolds in the multi-band joint echo theory model is updated using the target coordinate estimation vector, and the updated complex scattering coefficient vector is calculated. The target coordinate estimation vector and the updated complex scattering coefficient vector are used to generate the final high-resolution ISAR image.

2. The multi-band ISAR high-resolution imaging method according to claim 1, characterized in that, The multi-band joint echo theoretical model is expressed as follows: ； ； in, This represents the vector of the multi-band joint echo theoretical model. Indicates the first One goal, Indicates the total number of targets. Indicates the first The true target coordinates of each target This represents multi-band combined additive white Gaussian noise. Indicates the first The complex scattering coefficients of a target, Represents a vector consisting of the coordinates of all real targets. express The corresponding dictionary of continuous manifolds, Represents the imaginary unit. This represents a column vector consisting of the complex magnitudes of all targets. Represents the global frequency vector. This represents the initial reference distance when discretizing the distance space. It represents the speed of light.

3. The multi-band ISAR high-resolution imaging method according to claim 1, characterized in that, The process involves discretizing the global frequency vector in the distance space using a grid to obtain a discrete sensing dictionary matrix. Based on the discrete sensing dictionary matrix and the global observation vector, a sparse Bayesian learning method is used to perform sparse coarse estimation in the discrete domain, resulting in an initial discrete coordinate coarse estimation vector and an initial complex amplitude coarse estimation vector, including: Based on the spectral range covered by the global frequency vector, the theoretical resolution of the distance space is determined, and the distance space is uniformly discretized with the theoretical resolution as the step size to generate a discrete distance grid point vector. The discrete sensing dictionary matrix is ​​constructed based on the global frequency vector and the discrete distance grid point vector; A sparse linear model is established using the global observation vector and the discrete sensing dictionary matrix; Based on the sparse linear model and the discrete distance grid point vector, iterative optimization is performed using the sparse Bayesian learning method and the expectation-maximization algorithm to obtain the initial discrete coordinate coarse estimation vector and the initial complex amplitude coarse estimation vector.

4. The multi-band ISAR high-resolution imaging method according to claim 3, characterized in that, The iterative optimization process based on the sparse linear model and the discrete distance grid point vectors, combined with the sparse Bayesian learning method and the expectation-maximization algorithm, yields the initial discrete coordinate coarse estimation vector and the initial complex magnitude coarse estimation vector, including: A sparse Bayesian learning method is used to set a sparse prior distribution for the scattering coefficients at each position of the discrete distance grid point vector, and the posterior distribution of the scattering coefficients is obtained by Bayes' theorem based on the global observation vector and the sparse prior distribution. The posterior distribution of the scattering coefficients and the hyperparameters of the sparse linear model are iteratively updated using the expectation-maximization algorithm until the first iteration convergence condition is met. An amplitude threshold is set using the statistics of the posterior distribution of the scattering coefficients corresponding to the first iteration convergence condition. From all the scattering coefficients corresponding to the discrete distance grid point vectors, scattering coefficients whose posterior amplitude exceeds the amplitude threshold are selected as significant scattering centers; The position of the discrete distance grid point vector corresponding to the significant scattering center and the corresponding posterior mean in the posterior distribution of the scattering coefficient are used as the initial discrete coordinate coarse estimation vector and the initial complex amplitude coarse estimation vector, respectively.

5. The multi-band ISAR high-resolution imaging method according to claim 4, characterized in that, The first iteration convergence condition is that the posterior distribution of the scattering coefficients or the change in the hyperparameters of the sparse linear model is continuously less than a preset threshold.

6. The multi-band ISAR high-resolution imaging method according to claim 1, characterized in that, The nonlinear least squares cost function is expressed as follows: ； ； in, express The corresponding nonlinear least squares cost function, Represents a vector consisting of the coordinates of all real targets. Represents the global observation vector, express The corresponding dictionary of continuous manifolds, express The corresponding complex scattering coefficient vector estimation, Denotes the square of the L2 norm. express The corresponding Moore-Penrose pseudoinverse, This indicates the conjugate transpose.

7. The multi-band ISAR high-resolution imaging method according to claim 1, characterized in that, The step of using the initial discrete coordinate coarse estimation vector as the initial value, and optimizing the discrete coordinate vector in the nonlinear least squares cost function within the corresponding neighborhood range to obtain the target coordinate estimation vector includes: Obtain the neighborhood range corresponding to the initial discrete coordinate coarse estimation vector; The initial discrete coordinate coarse estimation vector is used as the initial value of the nonlinear least squares cost function, and the trust region reflection algorithm is used to optimize the solution within the neighborhood until the second iteration convergence condition is met. The discrete coordinate vector that satisfies the second iteration convergence condition is used as the target coordinate estimation vector.

8. The multi-band ISAR high-resolution imaging method according to claim 1, characterized in that, The step of generating the final high-resolution ISAR image using the target coordinate estimation vector and the updated complex scattering coefficient vector includes: Based on the target coordinate estimation vector, construct an ideal full-band manifold dictionary; The ideal full-band signal is calculated using the ideal full-band manifold dictionary and the updated complex scattering coefficient vector. Perform a two-dimensional inverse Fourier transform on the ideal full-band signal to obtain the initial ISAR image; The initial ISAR image is normalized to obtain the final high-resolution ISAR image.

9. A multi-band ISAR high-resolution imaging device, characterized in that, The multi-band ISAR high-resolution imaging device includes: an acquisition unit, a stitching processing unit, a model building unit, a sparse estimation unit, an optimization solution unit, an update unit, and an output unit; The acquisition unit is used to acquire the echo signals of the ISAR system in multiple non-overlapping frequency bands and the corresponding frequency sampling points; The splicing processing unit is used to perform vertical splicing processing on the echo signal and the frequency sampling points respectively, to obtain the global observation vector and the global frequency vector. The model building unit is used to construct a multi-band joint echo theoretical model based on the global frequency vector and the point scattering model. The sparse estimation unit is used to perform grid discretization processing on the distance space using the global frequency vector to obtain a discrete sensing dictionary matrix, and based on the discrete sensing dictionary matrix and the global observation vector, to perform discrete domain sparse coarse estimation processing using the sparse Bayesian learning method to obtain an initial discrete coordinate coarse estimation vector and an initial complex amplitude coarse estimation vector. The model building unit is also used to construct a nonlinear least squares cost function using the variable projection method through the multi-band joint echo theory model and the initial complex amplitude coarse estimation vector. The optimization unit is used to optimize the discrete coordinate vector in the nonlinear least squares cost function within the corresponding neighborhood range, using the initial discrete coordinate coarse estimation vector as the initial value, to obtain the target coordinate estimation vector. The updating unit is used to update the continuous manifold dictionary in the multi-band joint echo theory model using the target coordinate estimation vector, and calculate the updated complex scattering coefficient vector. The output unit is used to generate a final high-resolution ISAR image using the target coordinate estimation vector and the updated complex scattering coefficient vector.

10. A multi-band ISAR high-resolution imaging device, characterized in that, include: The device includes a processor, a storage medium, and a bus. The storage medium stores machine-readable instructions executable by the processor. When the multi-band ISAR high-resolution imaging device is running, the processor communicates with the storage medium via the bus. The processor executes the machine-readable instructions to perform the steps of the multi-band ISAR high-resolution imaging method as described in any one of claims 1-8.

Images