A method of forward-looking super-resolution imaging of a mobile platform using a combined structured sparsity and bayesian framework

Abstract

The application provides a kind of motorized platform forward-looking super-resolution imaging method of combined structured sparse and Bayesian framework, it is related to remote sensing technical field, comprising: establishing high dynamic platform forward-looking imaging signal processing model;Receive echo signal, according to echo signal and high dynamic platform forward-looking imaging signal processing model, obtain the preprocessed signal of echo signal in two-dimensional time domain.Construct phase compensation factor, according to phase compensation factor and the preprocessed signal of two-dimensional time domain, obtain multi-domain joint phase correction signal.The multi-domain joint phase correction signal is expressed as the form of overcomplete dictionary and signal scattering coefficient convolution, and the variable estimation result of the preprocessed signal of two-dimensional time domain in all distance gates is determined;According to the variable estimation result under all distance gates, generate the super-resolution forward-looking imaging of two-dimensional scene.Make the high-order motion of motorized platform be considered and the target structure with block sparse characteristics, realize the forward-looking super-resolution imaging of motorized platform to the surface target scene with structured sparse.

Description

Technical Field

[0001] This invention relates to the field of remote sensing technology, and in particular to a forward-looking super-resolution imaging method for a mobile platform that combines structured sparse and Bayesian frameworks. Background Technology

[0002] Forward-looking (FLAG) mode is an important imaging mode in remote sensing, capable of detecting and imaging the area directly in front of an aircraft's flight path at all times and from a long distance. Therefore, this imaging mode has broad application prospects in missile radar imaging equipment and aircraft radar imaging equipment. Notably, when using FLAG imaging in a radar system, it can be used to detect and identify targets of interest within the area directly in front of the user. Compared to single-pulse target detection methods, FLAG effectively improves detection and identification accuracy, thus making FLAG radar imaging a key research direction in radar imaging.

[0003] It is important to note that when an aircraft is in motion and the radar system is using forward-looking mode imaging, the aircraft's motion affects the slant range history (a parameter in radar imaging). In this case, the radar imaging results may show missing targets. Therefore, in scenarios where equipment equipped with radar imaging systems is using maneuvering trajectories, how to effectively achieve super-resolution imaging of the radar system becomes a pressing technical problem to be solved. Summary of the Invention

[0004] This invention provides a forward-looking super-resolution imaging method for a maneuvering platform that combines structured sparsity and a Bayesian framework. This method enables forward-looking super-resolution imaging of a maneuvering platform with structured sparsity surface target scenes by taking into account the higher-order motion of the maneuvering platform and the target structure with block sparsity characteristics.

[0005] To achieve the above technical objectives, the embodiments of the present invention adopt the following technical solutions:

[0006] In a first aspect, the present invention provides a method for forward-looking super-resolution imaging of a maneuvering platform that combines structured sparseness and Bayesian frameworks, the method comprising:

[0007] A forward-looking imaging signal processing model for a high-dynamic platform is established. Echo signals are received, and based on the echo signals and the high-dynamic platform forward-looking imaging signal processing model, a preprocessed signal in the two-dimensional time domain is obtained. A phase compensation factor is constructed, and based on the phase compensation factor and the preprocessed signal in the two-dimensional time domain, a multi-domain joint phase correction signal is obtained. The compensation factor is used to compensate for higher-order phase terms and range travel corrections introduced by the high-dynamic platform's maneuverability. The multi-domain joint phase correction signal is represented as a convolution of an overcomplete dictionary and signal scattering coefficients, and the variable estimation results of the preprocessed signal in the two-dimensional time domain are determined for all range gates. Based on the variable estimation results for all range gates, a super-resolution forward-looking image of the two-dimensional scene is generated.

[0008] The forward-looking super-resolution imaging method for maneuvering platforms provided in this invention embodiment considers the phase perturbation introduced by the higher-order motion of the high-dynamic platform, constructs a virtual multi-domain joint phase compensation factor, improves the complete dictionary, accurately builds a linear regression model, and thus characterizes the forward-looking imaging geometry of the high-dynamic platform.

[0009] In conjunction with the first aspect, in one possible implementation, the echo signal includes at least one range-gate signal. The above-described representation of the multi-domain joint phase-corrected signal as a convolution of an overcomplete dictionary and signal scattering coefficients, and the determination of variable estimation results for the preprocessed signal in the two-dimensional time domain across all range cells, includes: representing the multi-domain joint phase-corrected signal as a convolution of an overcomplete dictionary and signal scattering coefficients. For each range-gate signal, the following operations are performed: the convolution form is represented as a vector form, and probabilistic models are performed on the scattering coefficients and echo vectors in vector form. Parameter estimation of the scattering coefficient vectors is achieved using variational Bayesian inference and the expectation-maximization algorithm, determining the variable estimation results under the range gates. All range cells are traversed to obtain variable estimation results for all distances in the two-dimensional scene.

[0010] Understandably, since the echo signal includes multiple range gate signals, it is necessary to calculate the variable estimates for each range gate in order to obtain an ultra-high resolution two-dimensional image.

[0011] In conjunction with the first aspect, another possible implementation of the method may further include: transmitting a radar signal, wherein the direction of propagation of the radar signal is the forward-looking imaging direction.

[0012] In conjunction with the first aspect, another possible implementation involves estimating the parameters of the scattering coefficient vector using variational Bayesian inference and the expectation-maximization algorithm, and determining the variable estimation results for the range gate, including:

[0013] Based on a Bayesian framework, the forward-looking imaging problem is transformed into a Bayesian posterior probability problem. A structured sparse prior model is constructed for the forward-looking imaging scene and noise, and the variables under the range gate are initialized. Variational Bayesian inference is used to calculate the posterior functions of the observed initial variables and related latent variables. The expectation-maximization method is used to estimate the variable estimation results under the range gate and the update parameters of related latent variables.

[0014] It is worth mentioning that, based on the sparse Bayesian framework, the super-resolution forward-looking imaging problem is transformed into a maximum a posteriori probability problem. Combining the physical meaning of forward-looking imaging of the mobile platform, reasonable and interpretable prior probability models are established for the sparse scene to be solved and the observed echoes. Considering the structured sparsity of surface targets, analytical estimates of the parameters of each variable in the forward-looking scene are derived using the ideas of variational Bayesian inference and expectation maximization algorithm. This improves the sparse reconstruction of sparse Bayesian learning and obtains high-quality forward-looking remote sensing images with target detail information.

[0015] In conjunction with the first aspect, another possible implementation is to represent the multi-domain joint phase correction signal as a convolution of an overcomplete dictionary and signal scattering coefficients, as follows: If the definition of an overcomplete dictionary is: The convolutional model is represented as: y = ψσ + ξ; where the overcomplete dictionary matrix is ​​defined as ψ = P a ⊙L, P a The triangular base phase is represented by L, and the antenna pattern is represented by L.

[0016] Secondly, embodiments of the present invention also provide a forward-looking super-resolution imaging device for a mobile platform that combines structured sparse and Bayesian frameworks, including: a signal modeling module, a preprocessing module, a matrix construction module, a parameter estimation module, and a two-dimensional imaging module.

[0017] The signal modeling module is used to establish a forward-looking imaging signal processing model for a high-dynamic platform.

[0018] The preprocessing module is used to receive the echo signal and, based on the echo signal and the high dynamic range platform forward-looking imaging signal processing model, obtain the preprocessed signal of the echo signal in the two-dimensional time domain.

[0019] The matrix construction module is used to construct a phase compensation factor and obtain a multi-domain joint phase correction signal based on the phase compensation factor and the preprocessed signal in the two-dimensional time domain. The compensation factor is used to compensate for the higher-order phase terms and distance travel correction introduced by the high-dynamic platform maneuverability.

[0020] The parameter estimation module is used to represent the multi-domain joint phase-corrected signal as a convolution of an overcomplete dictionary and signal scattering coefficients, and to determine the variable estimation results of the preprocessed signal in the two-dimensional time domain at all range gates.

[0021] The 2D imaging module is used to generate super-resolution forward-looking images of a 2D scene based on the variable estimation results of all distance gates.

[0022] In conjunction with the second aspect, in one possible implementation, the echo signal is a signal that includes at least one distance gate.

[0023] Specifically, the parameter estimation module is used to represent the multi-domain joint phase correction signal as a convolution of an overcomplete dictionary and signal scattering coefficients; and performs the following operations for each range gate signal:

[0024] The convolutional form is represented as a vector form, and the scattering coefficient and echo vector in vector form are probabilistically modeled. The parameters of the scattering coefficient vector are estimated by variational Bayesian inference and expectation-maximization algorithm to determine the variable estimation results of the range gate. All range cells are traversed to obtain the variable estimation results for all distances in the two-dimensional scene.

[0025] In conjunction with the second aspect, in another possible implementation, the forward-looking imaging device may further include a transmitting module for transmitting radar signals, the radar signals propagating in the forward-looking imaging direction.

[0026] In conjunction with the second aspect, another possible implementation involves the parameter estimation module specifically used for: transforming the forward-looking imaging problem into a Bayesian posterior probability problem based on a Bayesian framework; creating a structured sparse prior model for the forward-looking imaging scene and noise; and initializing the variables under the range gate. Variational Bayesian inference is then used to calculate the posterior functions of the observed initial variables and related latent variables. Finally, the expectation-maximization method is used to estimate the variable estimation results under the range gate and the updated parameters of the related latent variables.

[0027] In conjunction with the second aspect, another possible implementation is that the matrix construction module is specifically used to: represent the multi-domain joint phase correction signal as a convolution of an overcomplete dictionary and signal scattering coefficients, as follows: If the definition of an overcomplete dictionary is: The convolutional model can be represented as: y = ψσ + ξ; where the overcomplete dictionary matrix is ​​defined as ψ = P a ⊙L, P a The triangular base phase is represented by L, and the antenna pattern is represented by L.

[0028] Understandably, the device provided in the second aspect and any of its possible implementations can be referenced to the beneficial effects of the first aspect and any of its possible design methods, which will not be repeated here. Attached Figure Description

[0029] Figure 1 A flowchart of an imaging method for a radar system provided in an embodiment of the present invention;

[0030] Figure 2 This invention provides a high dynamic platform forward-looking super-resolution imaging coordinate system.

[0031] Figure 3 This is a flowchart of an imaging method for another radar system provided in an embodiment of the present invention;

[0032] Figure 4 This is a schematic diagram of distributed markers in a simulation experiment provided by an embodiment of the present invention;

[0033] Figure 5 This is a schematic diagram of the reconstruction obtained by simulation under the original scheme provided in an embodiment of the present invention;

[0034] Figure 6 This is a schematic diagram of the reconstruction obtained by a radar forward-looking imaging method provided in an embodiment of the present invention. Detailed Implementation

[0035] To facilitate understanding of the solutions in the embodiments of the present invention, the proper nouns appearing in the embodiments of the present invention will first be explained.

[0036] Dictionary matrix: This can be understood as a matrix, which is a set of numbers arranged in a rectangular array. In this embodiment of the invention, the dictionary matrix is ​​used to characterize the target scattering coefficient during radar imaging.

[0037] Synthetic Aperture Radar (SAR) is a method for imaging based on single-track synthetic aperture radar data.

[0038] Bistatic SAR: This is a method of imaging using two SAR systems with different trajectories, mainly used to overcome the limitation of traditional monostatic SAR in being unable to image in the area directly in front.

[0039] Distributed targets: When an imaging region includes multiple independently distributed individuals, these independently distributed individuals are called distributed targets.

[0040] Imaging area: During radar imaging, a pulse signal is emitted, and the area covered by the pulse signal is the imaging area of ​​the radar system.

[0041] Doppler beam sharpening (DBS) method: The working mode of airborne pulse Doppler radar that uses the Doppler effect to improve azimuth resolution through signal processing.

[0042] Sparse reconstruction theory: If a signal is compressible or sparse in a certain transform domain, then the signal can be sampled at a sampling rate much lower than the Nyquist sampling rate, and the original signal can be reconstructed with high probability by solving an optimization problem.

[0043] Sparsity Adaptive Matching Pursuit (SAMP) is a signal reconstruction algorithm for compressed sensing. The SAMP method does not require prior knowledge of the sparsity of the signal. It selects echo signals step by step by setting a fixed step size and uses a backtracking approach to filter out the echo signal that best matches the reconstructed signal from the candidate set.

[0044] The radar imaging process in the embodiments of the present invention will be described below.

[0045] As is well known, radar systems employ imaging techniques within the field of microwave remote sensing. In some forward-looking imaging modes, high range resolution can be achieved through pulse compression technology. In this approach, the high-resolution performance of radar imaging is determined by the bandwidth of the transmitted pulse signal, while the azimuth resolution is determined by the size of the antenna aperture. Therefore, limited by the target aperture size, this imaging method struggles to acquire satisfactory high-resolution images.

[0046] On the other hand, to obtain high-resolution two-dimensional microwave remote sensing images, a virtual large aperture can be synthesized along the azimuth direction. Furthermore, SAR, DBS, and bistatic SAR methods can be applied to remote sensing applications. It's important to understand that when applying monostatic SAR and DBS methods to radar systems, in forward-looking imaging mode, the radar's emitted pulse signal propagates directly in front of the device (i.e., the direction of the device's movement or the direction directly in front of the device). The received echo signals have the same Doppler history, resulting in left and right Doppler blurring, as well as a decreasing Doppler gradient. All of these factors prevent the formation of a virtual aperture, thus affecting the azimuth resolution and consequently the quality of the radar image obtained under forward-looking imaging.

[0047] Furthermore, in the application of bistatic SAR to radar systems, the transmitter and receiver are designed separately. In this case, introducing additional Doppler phase can compensate for the inability of monostatic SAR to synthesize a virtual aperture. However, due to the dual-platform architecture of such radar systems, issues such as time, frequency, and spatial synchronization, as well as communication problems between multiple platforms, limit the application scenarios of bistatic SAR. Understandably, monopulse imaging can achieve high azimuth focusing capability, but the imaging system performs well only with a single beam and a single target; its application is limited when multiple targets are involved.

[0048] In summary, among these forward-looking imaging modes, single-channel systems are simple and easy to implement. However, their azimuth resolution is affected by issues such as the actual aperture length in the azimuth direction, blurring between the left and right Doppler views, and low Doppler gradients. Furthermore, their imaging quality is problematic in multi-target imaging environments, making them unsuitable for radar imaging systems. While bistatic SAR radar systems can achieve improved azimuth resolution, their engineering applications are difficult, preventing their use in practical radar systems and equipment.

[0049] Furthermore, to address the limited lateral range resolution in single-base real-beam forward-looking imaging systems, some implementations introduce super-resolution imaging and sparse reconstruction theory into single-base scanning imaging systems. This transforms the forward-looking imaging problem into a linear regression problem, allowing the construction of an overcomplete dictionary matrix in linear regression through beam scanning, thereby achieving super-resolution imaging of the forward-looking mode. Moreover, this greedy tracking method, represented by the Sparse Adaptive Matching Pursuit (SAMP) algorithm, has attracted considerable attention in the field of forward-looking imaging due to its simple configuration and fast tracking capabilities. However, this greedy tracking method is prone to getting trapped in local optima, and in most cases, it fails to find the global optimum. In such cases, if the imaging region in the forward-looking imaging mode includes distributed targets, the imaging results will affect the reconstruction performance of distributed targets in the radar image.

[0050] It's worth mentioning that sparse Bayesian learning is a classic approach in sparse reconstruction theory. Specifically, this method, based on a Bayesian framework, equates the forward-looking imaging problem to a probabilistic model parameter estimation problem. It models the forward-looking imaging scene based on sparse Bayesian priors, giving the forward-looking reconstruction model a physically interpretable meaning. This unique approach achieves superior reconstruction performance compared to other recovery methods. However, for forward-looking imaging systems with trajectory maneuvers, the introduction of higher-order motion affects the slant range history, thereby disrupting the traditional dictionary matrix and the sparse recovery process, severely impacting reconstruction performance. Traditional sparse Bayesian learning is based on the assumption of "pixel sparsity." In the reconstruction results of blocky surface targets in the forward-looking scene, the image exhibits discontinuities and missing target details, failing to reflect true ground information and severely affecting subsequent target recognition applications.

[0051] Understandably, when a radar system performs forward-looking detection and imaging, the lateral resolution within the imaging area of ​​the forward-looking scene is related to the azimuth real aperture length of the radar antenna. During the geometric construction process in forward-looking mode, when the radar beam emission direction is the same as the movement direction of the equipment carrying the radar system, the left and right sides of the imaging scene will have the same spatial cone angle, causing left and right Doppler blurring in forward-looking mode.

[0052] In some implementations, super-resolution imaging technology is based on monostatic radar real-beam scanning. It fully utilizes the target area information contained in the scan echo sequence to invert the echo sequence from the data domain to the target domain, thereby obtaining an angular resolution that exceeds the real aperture beamwidth. In forward-looking radar imaging, super-resolution technology has become an important research direction.

[0053] Essentially, the azimuth scan echo of a moving platform can be transformed into a convolution model of the target scattering rate azimuth distribution function and the antenna pattern. Theoretically, from the perspective of imaging mechanism, it is feasible to achieve high-resolution azimuth imaging using convolution inversion methods. Specifically, the forward-looking radar imaging model can be reorganized into a linear regression model. Under the assumption of sparsity, relying on the sparse microwave imaging processing framework, super-resolution sparse reconstruction of the forward-looking image can be achieved. Furthermore, it can be well compatible with existing real-beam scanning radar systems and operating modes, and is low-cost and highly efficient.

[0054] In some implementations, pixel-sparse Bayesian learning can be used to achieve forward-looking imaging of the target area based on the carrier platform. Specifically, firstly, a scanning imaging geometric model and signal processing model of the airborne platform are established, and range processing is achieved through pulse compression. Further, the preprocessed signal is represented as a convolution of the target discrete points and the antenna pattern. For example, given the observed value y, the reconstruction problem x of the forward-looking scene is represented as a linear regression model. The convolutional form is expressed as: y = Ax + n, to complete the construction of the measurement dictionary matrix. Further, based on the Bayesian framework, a suitable prior model of the forward-looking scene is performed using a Gaussian distribution, and the parameters of relevant variables are estimated using a maximum a posteriori approach to recover the target to be reconstructed.

[0055] It should be noted that when considering forward-looking imaging of the target area of ​​an airborne platform, the influence of acceleration is not considered. The influence of 3D acceleration is not considered when constructing the Doppler overcomplete dictionary, resulting in a measurement matrix that cannot accurately describe the forward-looking imaging scene of the maneuvering platform, leading to a mismatch in the sparse reconstruction kernel. Furthermore, current forward-looking imaging methods based on sparse Bayesian learning are based on the assumption of pixel sparsity, resulting in discontinuous reconstruction results and a loss of target detail information. Therefore, in scenes containing area targets, the sparse reconstruction performance of traditional pixel-sparse Bayesian learning-based methods will significantly decrease, impacting subsequent detection applications.

[0056] Therefore, this invention provides a method for super-resolution forward-looking imaging of a maneuvering platform that combines structured sparsity and a Bayesian framework. Considering the phase perturbations introduced by the higher-order motion of the high-dynamic platform, a virtual multi-domain joint phase compensation factor is constructed, an improved and complete dictionary is used, and a linear regression model is accurately built to characterize the forward-looking imaging geometry of the high-dynamic platform. Based on the sparse Bayesian framework, the super-resolution forward-looking imaging problem is transformed into a maximum a posteriori probability problem. Combining the physical meaning of the maneuvering platform's forward-looking imaging, reasonable and interpretable prior probability models are established for the sparse scene to be solved and the observed echoes. Considering the structured sparsity characteristics of surface targets, analytical estimates of the parameters of each variable in the forward-looking scene are derived using variational Bayesian inference and the expectation-maximization algorithm. This improves the sparse reconstruction of sparse Bayesian learning, resulting in high-quality forward-looking remote sensing images with detailed target information.

[0057] The following will provide a detailed description of the forward-looking super-resolution imaging method for maneuvering platforms that combines structured sparse and Bayesian frameworks provided in the embodiments of the present invention.

[0058] Please refer to Figure 1 The flowchart below shows the forward-looking super-resolution imaging method for a maneuvering platform based on a combination of structured sparse and Bayesian frames provided in an embodiment of the present invention. Figure 1 As shown, the method includes steps 101-105.

[0059] Step 101: Establish a forward-looking imaging signal processing model for a high-dynamic platform.

[0060] Specifically, using the intermediate time and position of the high dynamic range platform during echo data acquisition as a reference, and with the projection point of the high dynamic range platform's position on the ground as the coordinate center, the horizontal equivalent velocity direction of the high dynamic range platform as the y-axis, and the upward direction as the z-axis, a forward-looking super-resolution imaging coordinate system for the high dynamic range platform is established. Please refer to [reference needed]. Figure 2 , which represents the high dynamic platform forward-looking super-resolution imaging coordinates established according to the aforementioned coordinate center and direction. For example... Figure 2 As shown, The yellow shaded area represents the forward-looking scan imaging area, and Q represents any point on the curved trajectory.G Let Q be the projection point on the ground plane, G be the intersection point of the beam center of the mobile platform and the ground-illuminated scene, and R be the projection point of Q on the ground plane. s ω represents the slant distance from the scene center at the time of imaging center, H represents the flight altitude of the maneuvering platform at the time of imaging center, and ω v Indicates the beam scanning angular velocity. This represents the three-dimensional velocity of a high-dynamic platform. Representing the three-dimensional acceleration of a high-dynamic platform, θ(t) = ω v t represents the instantaneous beam azimuth angle. R(t) represents the true slant range history between the target and the maneuvering platform, which varies with azimuth time and incorporates three-dimensional acceleration; it can be represented as a vector. The length, specifically as shown in the figure

[0061] As shown in Equation 1:

[0062]

[0063] Where |·| represents the modulo operation on the vector, This indicates that the summation operation is performed sequentially over subscripts n = 0, 1, 2, 3, 4, where n represents the summation variable and k represents the summation value. n Let R(t) represent the nth order Maclaurin expansion coefficients of the precise true slant distance history R(t) at t=0, which can be expressed by the following formula 2:

[0064]

[0065] Where n! represents the factorial from 1 to n, This represents the nth derivative of a function with respect to a variable.

[0066] To facilitate the expression of the McLaurin expansion coefficients, three spatial angle information is constructed, which is represented by the following formula 3:

[0067]

[0068] At this point, the McLaughlin expansion coefficients can be expressed by the following formula 4:

[0069]

[0070] Step 102: Receive the echo signal and obtain the preprocessed signal of the echo signal in the two-dimensional time domain based on the echo signal and the forward-looking imaging signal processing model of the high dynamic platform.

[0071] For example, assuming the radar transmits a linear frequency modulated pulse signal at a preset pulse repetition frequency, the pulse signal can be represented by the following formula 5:

[0072]

[0073] Among them, t r T represents the distance to the fast time variable. p κ represents the pulse width, f0 represents the radar carrier frequency, and κ represents the radar carrier frequency. a Let λ represent the frequency modulation slope, λ represent the wavelength, rect(·) represent the rectangular window function, exp{·} represent the complex exponential function, π represent pi, and c represent the speed of light.

[0074] Specifically, the radar receiver receives the echo signal, performs carrier frequency removal processing to generate a demodulated baseband echo signal, and then performs matched filtering in the range direction to obtain a pulse compression signal. Its two-dimensional time domain representation can be expressed by the following formula 6:

[0075]

[0076] Where l(·) represents the antenna pattern function, σ represents the backscattering coefficient, j represents the imaginary unit, θ(t) represents the instantaneous scanning angle between the target and the platform, and B represents the bandwidth.

[0077] Step 103: Construct a phase compensation factor. Based on the phase compensation factor and the preprocessed signal in the two-dimensional time domain, obtain a multi-domain joint phase correction signal.

[0078] The compensation factor is used to compensate for the higher-order phase terms and distance travel corrections introduced by the high-dynamic platform maneuverability.

[0079] First, a high-order motion phase compensation function is constructed in the two-dimensional time domain. This compensation function can be expressed using the following formula:

[0080] Equation 7 means:

[0081]

[0082] Where, k 2a ,k 3a ,k 4a This indicates that the coefficients of the McLaurin expansion include higher-order terms of acceleration, which can be specifically expressed by the following formula 8:

[0083]

[0084] At this point, after compensating for the higher-order phase terms, the phase in the distance frequency domain and the slow time domain can be expressed by the following formula:

[0085] Equation 9 is expressed as:

[0086]

[0087] Where, k 2_rest ,k 3_rest ,k 4_restBoth indicate that the coefficients in the McLaughlin expansion do not include the coefficient term for acceleration.

[0088] A distance travel compensation factor is constructed in the distance frequency domain and the slow time domain. The distance compensation factor is expressed by the following formula:

[0089] Equation 10 means:

[0090]

[0091] After high-order phase compensation and range travel correction are completed in multiple domains, the preprocessed signal in the two-dimensional time domain is expressed by the following formula 11:

[0092]

[0093] Among them, R rw (t)=R S +k 2_rest t 2 +k 3_rest t 3 +k 4_rest t 4 R pa (t)=R S +k1t+k 2_rest t 2 +k 3_rest t 3 +k 4_rest t 4 These represent the slope history in the envelope and phase terms after multi-domain joint phase compensation, respectively.

[0094] Step 104: Represent the multi-domain joint phase correction signal as a convolution of an overcomplete dictionary and signal scattering coefficients, and determine the variable estimation results of the two-dimensional time-domain preprocessed signal at all range gates.

[0095] The echo signal includes at least one range gate signal. In order to facilitate the generation of high-resolution radar images, each range gate signal is processed to obtain the corresponding variable estimation results, so as to finally generate a two-dimensional high-resolution radar image.

[0096] The multi-domain joint phase-corrected signal is represented as a convolution of an overcomplete dictionary and the signal scattering coefficients. For each range-gated signal, the following operations are performed: the convolution form is represented as a vector form, and probabilistic models are performed on the scattering coefficients and echo vectors in vector form. Parameter estimation of the scattering coefficient vectors is achieved using variational Bayesian inference and the expectation-maximization algorithm, determining the variable estimation results under the range gate. All range cells are traversed to obtain the variable estimation results for all ranges in the 2D scene.

[0097] Specifically, after multi-domain joint phase correction, the signal y pre(τ,t) can be expressed as the convolution of an overcomplete dictionary and the signal scattering coefficients, as shown in Equation 12:

[0098]

[0099] The overcomplete dictionary factor is defined as follows:

[0100] Considering the influence of noise, the above convolution model can be represented in vector form:

[0101] y=ψσ+ξ Formula 13

[0102] The overcomplete dictionary matrix is ​​defined as ψ = P a ⊙L, P a Let L represent the triangular base phase and L represent the antenna pattern. Their expressions are as follows:

[0103]

[0104] Among them, R g (θ k ,t m ) represents the extension of the slant range in the azimuth direction, l h H represents the spatial sampling point of the antenna pattern vector, H represents the effective length of each antenna pattern vector, K represents the number of scattering points in the azimuth dimension, and M represents the number of sampling points in the azimuth direction.

[0105] Specifically, for each range gate of the scattering coefficient of the forward-looking scene, a Bayesian posterior probability problem of the scattering coefficient of the forward-looking scene is constructed. Using the idea of ​​the student-t distribution, the problem of solving the Bayesian posterior probability is transformed into the problem of solving the maximum a posteriori probability based on the hierarchical student-t distribution, according to the observed values ​​of the scattering coefficient of the forward-looking scene and the overcomplete dictionary matrix. The parameters of the relevant variables are solved by using variational Bayesian inference and the idea of ​​expectation maximization, so as to achieve super-resolution imaging of the current range gate.

[0106] For example, the above-described method for estimating variables can be further divided into several implementation steps, please refer to [reference needed]. Figure 3 This is a flowchart of a forward-looking imaging method for another radar system provided in an embodiment of the present invention. Figure 3 As shown, it includes steps 301-303.

[0107] S301. Based on the Bayesian framework, the forward-looking imaging problem is transformed into a Bayesian posterior probability solution problem, and a structured sparse prior model is performed on the forward-looking imaging scene and noise, and the variables under the distance gate are initialized.

[0108] Specifically, in this embodiment of the invention, the backscattering coefficient of each range gate in the region to be reconstructed is used as the random variable to be reconstructed, the preprocessed signal is used as the observation variable to construct a Bayesian posterior probability model, and a structured sparse prior model is performed on the forward-looking imaging scene and noise, and the relevant parameters are initialized.

[0109] First, the noise sparsity is modeled as a Student-t distribution.

[0110]

[0111] Where α represents noise precision, which is the reciprocal of noise, and η represents gamma degrees of freedom. This indicates that the random variable ζ follows a gamma distribution with scale parameter a and shape parameter b. Represents the gamma function. This represents a Gaussian distribution.

[0112] At this point, the distribution function of the observed variable is expressed as:

[0113]

[0114] Where Λ=diag(λ1,…,λ m ,…λ M ) is an implicit function used to help improve the precision of each observed variable. diag(·) represents a diagonal matrix, and the precision α follows a gamma distribution, expressed as:

[0115] p(α)=Gam(α|a1,a2) Formula 19

[0116] The degrees of freedom η are also sparsely modeled as a gamma distribution:

[0117] p(η)=Gam(η|c1,c2) ​​Formula 20

[0118] The sparse modeling of the backscattering coefficients to be reconstructed is expressed as follows:

[0119]

[0120] in, This represents the k-th element in Ξ. This represents the structural factor; and the latent variable β is modeled as a gamma distribution, expressed as:

[0121]

[0122] S302. Calculate the posterior functions of the observed variables and related latent variables using variational Bayesian inference.

[0123] The posterior function of all variables is estimated by minimizing the relative entropy Kullback-Leibler (KL) divergence between the true posterior distribution and the approximate posterior function. The KL divergence is defined as:

[0124]

[0125] The posterior function is approximated by the following formula:

[0126]

[0127]

[0128] Where Δ={σ,α,β,Λ,η} represents the set of random variables and latent variables.

[0129] Calculate the posterior of α:

[0130]

[0131] in, This represents the expectation of f(ε) with respect to all variables in Δ except ζ, where ζ is one of the variables in the parameter set Δ, and const indicates that it is related to the variable. Irrelevant constants Expressing expectations

[0132] The obtained posterior probability follows a gamma distribution, and is expressed as:

[0133]

[0134] Calculate λ m Posterior:

[0135]

[0136] The obtained posterior probability follows a gamma distribution, and is expressed as:

[0137]

[0138] Calculate the posterior of η:

[0139]

[0140] The obtained posterior probability follows a gamma distribution, and is expressed as:

[0141]

[0142] S303. Use the expectation-maximization method to estimate the scattering coefficients and update parameters of related latent variables.

[0143] Calculate the posterior of σ:

[0144]

[0145] The obtained posterior probability follows a Gaussian distribution, and is expressed as:

[0146]

[0147] Where, variance Σ σ and mean μ σ They are represented as follows:

[0148] Σ σ -1 =<α>ψ T <Λ>ψ+<Ξ>,μ σ =Σ σ <α>ψ T Formula 33

[0149] Due to Ξ k It is β k With β k+j For variables with j = ±1, their posterior probabilities cannot be directly obtained using variational Bayes. Therefore, the expectation-maximization method is used to estimate the parameters. First, the E-step calculates the Q equation for β:

[0150]

[0151] In the M-step, the parameter β is updated by maximizing the Q equation; therefore, the Q equation is computed with respect to the parameter β. k First derivative:

[0152]

[0153] At this point, update β k The suboptimal solution is:

[0154]

[0155] Step 105: Generate super-resolution forward-looking imaging of the two-dimensional scene based on the variable estimation results under all distance gates.

[0156] The forward-looking imaging payload is mounted on a mobile platform whose trajectory is a curve, specifically characterized by three-dimensional acceleration in its motion parameters. The scene contains targets with block-sparse characteristics. The forward-looking imaging system operates in single-base scanning mode, and the main radar system parameters are shown in Table 1.

[0157] Table 1 Main Radar Parameters

[0158] carrier frequency X-band velocity vector [0,200,7]m / s Azimuth real aperture 0.55m acceleration vector <![CDATA[[0.3,0.7,0.5]m / s 2 ]]> Transmission bandwidth 60MHz Beam scanning angular velocity 60 degrees / s Distance sampling points 2048 Beam scanning range -10～10° Reference distance 12km Pulse repetition frequency 1000Hz

[0159] like Figure 2As shown, in the area directly in front (i.e. Figure 2 Within the yellow shaded area on the ground (as shown in the image), arrays of surface targets with block sparse characteristics are arranged along the x-axis and y-axis directions, respectively, to verify the correctness and effectiveness of the forward super-resolution imaging method combining structured sparseness and Bayesian framework mentioned in this invention.

[0160] For example, in some simulation experiments, the parameters in Table 1 are used as examples. Figure 4 The image shown is a schematic diagram of the original array. Figure 4 As can be seen, the imaging scene of the radar system includes a distributed dot matrix. In some implementation methods, it is reconstructed using the original technology, resulting in... Figure 5 The reconstruction result is shown below. Figure 5 As shown, the distributed dot matrix has low fidelity, the position of the distributed dots is not clearly visible, and the image resolution is low. When using the forward-looking imaging mode provided in this embodiment of the invention, the following can be obtained: Figure 6 The image shown is as follows: Figure 6 As shown, the structure of the distributed dot matrix can be clearly seen, and the image resolution is high.

[0161] In summary, while the original algorithm can reconstruct the foreground scene, it performs poorly in reconstructing planar targets with block sparsity, resulting in discontinuous planar target structures and missing detail information. The proposed algorithm, however, can achieve super-resolution reconstruction of planar targets with block sparsity while preserving their structural information.

[0162] This invention also provides a forward-looking super-resolution imaging device for a mobile platform that combines structured sparse and Bayesian frameworks, including: a signal modeling module, a matrix construction module, a parameter estimation module, and a two-dimensional imaging module.

[0163] The signal modeling module is used to establish a forward-looking imaging signal processing model for the high-dynamic platform. The matrix construction module is used to construct phase compensation factors and an overcomplete dictionary; these compensation factors compensate for higher-order phase terms and range travel corrections introduced by the high-dynamic platform's maneuverability. The parameter estimation module is used to perform structured sparse probabilistic modeling of the forward-looking scene based on a Bayesian framework. It estimates the posterior probability distribution of each variable using variational Bayesian derivation and the expectation-maximization algorithm to determine the sparse reconstruction of the forward-looking scene. The 2D imaging module is used to traverse all range cells, obtain the variable estimation results for the 2D scene, and achieve super-resolution forward-looking imaging of the 2D scene.

[0164] In some implementations, the parameter estimation module is also used to: transform the forward-looking imaging problem into a Bayesian posterior probability problem based on a Bayesian framework, and to perform a structured sparse prior model of the forward-looking imaging scene and noise; to calculate the posterior functions of the observed variables and related latent variables using variational Bayesian inference; and to estimate the scattering coefficients and the update parameters of the related latent variables using the expectation-maximization method.

[0165] In another possible implementation, the matrix construction module is specifically used to: after multi-domain joint phase correction, the signal is represented as a convolution of an overcomplete dictionary and the signal scattering coefficients as follows: If the definition of an overcomplete dictionary is: The convolutional model can be represented as: y = ψσ + ξ; where the overcomplete dictionary matrix is ​​defined as ψ = P a ⊙L, P a The triangular base phase is represented by L, and the antenna pattern is represented by L.

[0166] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for forward-looking super-resolution imaging of a maneuvering platform combining structured sparse and Bayesian frameworks, characterized in that, include: Establish a forward-looking imaging signal processing model for a high-dynamic platform; The echo signal is received, and a preprocessed signal of the echo signal in the two-dimensional time domain is obtained based on the echo signal and the forward-looking imaging signal processing model of the high dynamic platform. A phase compensation factor is constructed, and a multi-domain joint phase correction signal is obtained based on the phase compensation factor and the preprocessed signal in the two-dimensional time domain; wherein, the compensation factor is used to compensate for the higher-order phase terms and distance travel correction introduced by the maneuverability of the high-dynamic platform; The multi-domain joint phase correction signal is represented as an overcomplete dictionary convolved with the signal scattering coefficients, and the variable estimation results of the two-dimensional time-domain preprocessed signal at all range gates are determined. Based on the variable estimation results of all the distance gates, a super-resolution forward-looking image of the two-dimensional scene is generated; The echo signal includes at least one distance gate signal; The step of representing the multi-domain joint phase-corrected signal as a convolution of an overcomplete dictionary and signal scattering coefficients, and determining the variable estimation results of the two-dimensional time-domain preprocessed signal under all range cells, includes: The multi-domain joint phase-corrected signal is represented as a convolution of an overcomplete dictionary and the signal scattering coefficients. Perform the following operations for each distance gate signal: The convolutional form is represented as a vector form, and the scattering coefficients and echo vectors in vector form are probabilistically modeled. The parameters of the scattering coefficient vector are estimated by variational Bayesian inference and expectation-maximization algorithm, and the variable estimation results under the range gate are determined. Iterate through all distance cells to obtain variable estimation results for all distances in the 2D scene; The parameter estimation of the scattering coefficient vector through variational Bayesian inference and expectation-maximization algorithm, and the variable estimation results for determining the range gate, include: Based on the Bayesian framework, the forward-looking imaging problem is transformed into a Bayesian posterior probability problem, and a structured sparse prior model is constructed for the forward-looking imaging scene and noise. The variables under the distance gate are initialized, and the structured sparse prior model includes a structured factor. The posterior functions of the initial variables and related latent variables are calculated using variational Bayesian inference. The expected maximum method is used to estimate the variable estimation results under the distance gate and the update parameters of the relevant latent variables; The multi-domain joint phase correction signal is represented as a convolution of an overcomplete dictionary and signal scattering coefficients as follows: ; If the overcomplete dictionary is defined as: The convolutional model is represented as follows: ; The overcomplete dictionary matrix is ​​defined as follows: , Indicates the triangular basis phase. This indicates the antenna radiation pattern. The backscattering coefficient distribution of the target. For antenna pattern function, This is the slant range history after multi-domain joint phase compensation. For wavelength, The backscattering coefficient, For noise vectors, For Hadamard products.

2. The method according to claim 1, characterized in that, The method further includes: The radar signal is transmitted in the forward-looking imaging direction.

3. A forward-looking super-resolution imaging device for a mobile platform that combines structured sparse and Bayesian frameworks, characterized in that, include: Signal modeling module, preprocessing module, matrix construction module, parameter estimation module, and two-dimensional imaging module; The signal modeling module is used to establish a high-dynamic platform forward-looking imaging signal processing model; The preprocessing module is used to receive the echo signal and, based on the echo signal and the high dynamic platform forward imaging signal processing model, obtain the preprocessed signal of the echo signal in the two-dimensional time domain. The matrix construction module is used to construct a phase compensation factor and obtain a multi-domain joint phase correction signal based on the phase compensation factor and the preprocessed signal in the two-dimensional time domain; wherein, the compensation factor is used to compensate for the higher-order phase terms and distance travel correction introduced by the high-dynamic platform maneuverability; The parameter estimation module is used to represent the multi-domain joint phase correction signal as a convolution of an overcomplete dictionary and signal scattering coefficients, and to determine the variable estimation results of the two-dimensional time-domain preprocessed signal at all range gates. The two-dimensional imaging module is used to generate a super-resolution forward-looking image of a two-dimensional scene based on the variable estimation results of all the range gates; the echo signal is a signal that includes at least one range gate. The parameter estimation module is specifically used to represent the multi-domain joint phase correction signal as a convolution of an overcomplete dictionary and signal scattering coefficients. Perform the following operations for each distance gate signal: The convolutional form is represented as a vector form, and the scattering coefficient and echo vector in vector form are probabilistically modeled. The parameters of the scattering coefficient vector are estimated by variational Bayesian inference and expectation-maximization algorithm, and the variable estimation results of the range gate are determined. All range cells are traversed to obtain the variable estimation results for all distances in the two-dimensional scene. The method of estimating the parameters of the scattering coefficient vector using variational Bayesian inference and expectation-maximization algorithm, and determining the variable estimation results of the range gate, includes: transforming the forward-looking imaging problem into a Bayesian posterior probability problem based on a Bayesian framework, constructing a structured sparse prior model for the forward-looking imaging scene and noise, initializing the variables under the range gate, wherein the structured sparse prior model includes a structured factor; calculating the posterior function of the initialized variables and related latent variables using variational Bayesian inference; estimating the variable estimation results under the range gate and the update parameters of the related latent variables using the expectation-maximization method; the matrix construction module is specifically used for: The multi-domain joint phase correction signal can be represented as a convolution of an overcomplete dictionary and signal scattering coefficients as follows: ; If the overcomplete dictionary is defined as: ; The convolutional model is represented as follows: ; The overcomplete dictionary matrix is ​​defined as follows: , Indicates the triangular basis phase. This indicates the antenna radiation pattern. The backscattering coefficient distribution of the target. For antenna pattern function, This is the slant range history after multi-domain joint phase compensation. For wavelength, The backscattering coefficient, For noise vectors, For Hadamard products.

Images