Depth BiSAR target structure parameter estimation method and system under motion error

Abstract

The invention provides a depth B SAR target structure parameter estimation method and system under a motion error, and relates to the technical field of radar, and the method comprises the following steps: building a Bi SAR parameter imaging network based on a parameter echo model and under the condition that the motion error exists, and pre-training a low-order Doppler compensation module; performing coarse imaging by using output data of the low-order Doppler compensation module, generating an observation matrix, and dividing the observation matrix into a plurality of local observation matrixes; constructing a combined adaptive network, inputting a local observation matrix to the combined adaptive network, and compensating a high-order space-variant error in the observation matrix; and inputting target radar echo data into the trained low-order Doppler compensation module and the combined adaptive network to obtain structural parameters of the target radar echo data. The method provided by the invention can effectively reduce the interference of the motion error on the estimation of the target structure parameter in the previous algorithm, greatly improve the calculation efficiency, reduce the calculation time consumption, and realize the efficient and high-precision target structure parameter.

Description

Technical Field

[0001] The present invention relates to the technical field of radar, and particularly to a method for estimating BiSAR target structure parameters of depth under motion errors. Background Art

[0002] Currently, in the imaging of Synthetic Aperture Radar (SAR), the imaging algorithm has very high requirements for the accuracy of the radar platform position information, which directly determines the imaging quality. However, due to limitations in aspects such as the positioning system accuracy and data refresh rate, motion errors are inevitable. Especially in a radar system like Bistatic Synthetic Aperture Radar (BiSAR) where the transmitter and receiver are distributed on different platforms, the phase accumulation error caused by motion errors will be even greater, which will significantly reduce the imaging quality.

[0003] In the existing related technologies, the algorithms for compensating motion errors generally include: Phase Gradient Autofocus (PGA) algorithm, algorithms based on Doppler information and image entropy, sparse-driven motion error compensation methods, etc. However, for the PGA algorithm, it is based on strong outliers and cannot adapt to complex targets; for the algorithms based on Doppler information and image entropy, they cannot be applied to extended targets; for the sparse-driven motion error compensation methods, they are not only sensitive to parameters, but also have low computational efficiency.

[0004] In addition, through research by the inventor, it is found that existing algorithms generally are based on point target scattering models. Such models only process the phase components brought by the change of range history, ignoring the influence of target scattering characteristics. The imaging results are prone to losing target structure information, which is not conducive to BiSAR image interpretation and target recognition. At the same time, existing algorithms also do not consider two different error categories in motion errors and do not process them separately.

[0005] That is to say, in the existing related technologies, the motion error compensation algorithms not only have limitations in models, but also cannot achieve efficient and high-precision target parameter estimation, and are limited in BiSAR imaging applications. Summary of the Invention

[0006] In order to solve the technical problems in the related technologies, the present invention provides a method for estimating BiSAR target structure parameters of depth under motion errors.

[0007] According to the first aspect of the present invention, there is provided a method for estimating BiSAR target structure parameters of depth under motion errors, including the following steps:

[0008] Step S1: Build a BiSAR parameter imaging network based on the parametric echo model with motion errors. The BiSAR parameter imaging network includes a low-order Doppler error compensation module, an adaptive high-order spatially variant error compensation module, and a deep proximal mapping module. Among them, the low-order Doppler error compensation module is configured to be able to compensate for the low-order Doppler error in the radar echo by estimating the Doppler center frequency, the chirp rate, and the third-order error coefficient. The adaptive high-order spatially variant error compensation module is configured to be able to reconstruct the observation matrix under high-order spatially variant errors through learning. The deep proximal mapping module is configured to be able to implement an approximate point operator function.

[0009] Step S2: Pre-train the low-order Doppler compensation module.

[0010] Step S3: Coarsely image using the output data of the trained low-order Doppler compensation module, generate an observation matrix in combination with the imaging result and set parameters, and divide it into multiple local observation matrices.

[0011] Step S4: Train the adaptive high-order spatially variant error compensation module and the deep proximal mapping module to obtain a combined adaptive network. Input the local observation matrices obtained in Step S3 into the obtained combined adaptive network, and obtain network parameters through deep learning to adaptively compensate for the high-order spatially variant errors in the observation matrix. Among them, the combined adaptive network is configured as a network with a multi-level cascade structure, and each layer of this network structure is a combination of an adaptive high-order spatially variant error compensation module and a deep proximal mapping module.

[0012] Step S5: Input the target radar echo data into the trained low-order Doppler compensation module obtained in Step S2 and the trained combined adaptive network obtained in Step S4 in sequence to obtain the structural parameters of the target radar echo data.

[0013] Optionally, Step S2 specifically includes:

[0014] Step S2-1: Generate simulated radar echo data s according to the motion error echo model.

[0015] Step S2-2: Separate the real and imaginary parts of the simulated radar echo data s and input them into two separate network channels respectively.

[0016] Step S2-3: Perform iterative learning and output three parameters {f dce , f dre , f dte}, and perform parameter compensation according to the following formula:

[0017]

[0018] The loss function loss me in network training is:

[0019]

[0020] Wherein, f dce is the Doppler center frequency error, f dre is the Doppler chirp frequency error, f dte is the Doppler third-order error, is the output after low-order Doppler compensation, vec[·] is the column vectorization operation, exp[·] is the exponential function, η is the azimuth slow time, re(·) is the real part extraction, im(·) is the imaginary part extraction, and ‖·‖2 is the Euclidean distance operation.

[0021] Optionally, the step S2-1 specifically includes:

[0022] Step S2-1-1: Based on the scattering models of point targets, line targets, flat targets, vertical targets, and dihedral targets, as well as the motion error model, establish a parameterized echo model S p :

[0023]

[0024] Wherein, is the full phase change, which includes the phase caused by the delay of transmission and reception low-order Doppler error phase and high-order spatially variant error phase m and n are the number of range cells and azimuth cells in the scene respectively, A p , A l , A lv , A d and A o are the amplitudes of the flat target, linear target, vertical linear target, dihedral target, and point target respectively, are the reflection coefficient change parameters of the flat target, linear target, vertical linear target, and dihedral target caused by the attitudes and range frequencies of the transmitter and receiver respectively, η is the azimuth slow time variable, f r is the range frequency domain variable, k p , k l , k lv and k d are the numbers of the flat target, linear target, vertical linear target, and dihedral target respectively, x m and y n are the range coordinate and azimuth coordinate of the corresponding cell respectively; Z(η) is the attitude of the transmitter and receiver at the azimuth time, including the azimuth angle and elevation angle;

[0025] Step S2-1-2: The parameterized echo model S pThe simulated radar echo data s rewritten in matrix form:

[0026]

[0027] Υ = [Υ p Υ l Υ lv Υ d Υ o

[0028] where Υ is the target echo envelope matrix, which is obtained by splicing the flat target echo envelope matrix Υ p , the line target echo envelope matrix Υ l , the vertical target echo envelope matrix Υ lv , the dihedral target echo envelope matrix Υ d and the point target echo envelope matrix Υ o . Φ all is the total phase change matrix, which is obtained by the Hadamard product of the transmit-receive delay phase matrix Φ f , the low-order Doppler error phase matrix Φ ef and the high-order space-variant error phase matrix Φ eh . is the Hadamard product operation, and a is the scattering coefficient vector.

[0029] Optionally, the step S3 specifically includes:

[0030] Step S3-1: Input the radar data into the low-order Doppler compensation module obtained after training in step S2 to obtain the compensated data as the output;

[0031] Step S3-2: Coarsely image the compensated data obtained in step S3-1 using the PFA algorithm, generate an observation matrix in combination with the imaging result and set parameters, and divide the observation matrix into multiple local observation matrices.

[0032] Optionally, the step S4 specifically includes:

[0033] Step S4-1: Train the adaptive high-order space-variant error compensation module and the deep proximal mapping module and obtain the combined adaptive network;

[0034] Step S4-2: Divide the local observation matrix obtained in step S3 into a real part channel and an imaginary part channel, and input them into the adaptive high-order space-variant error compensation module in the combined adaptive network obtained in step S4-1 respectively to obtain the adjusted local observation matrix H mei , and perform gradient descent operation through the following formula:

[0035]

[0036] ​where re(·) is to take the real part, and im(·) is to take the imaginary part. is the intermediate process variable of the k-th layer of the i-th local observation matrix. I is the identity matrix, and ρ k , μ k are the step sizes of the real part channel and the imaginary part channel in the k-th iteration, respectively. is the value obtained from the i-th local observation matrix in the (k - 1)-th layer. is the data corresponding to the i-th local observation matrix in the output of the trained network in step S2. V1 and V2 are adaptive matrices to adapt to the differences of different local observation matrices.

[0037] Step S4-3: Input the real part channel output and the imaginary part channel output of the adaptive high-order spatially variant error compensation module in step S4-2 into the deep proximal mapping module for processing, and successively pass through the real-imaginary interaction unit, the encoding-decoding unit, and the real-imaginary interaction unit. The output result is a k :

[0038]

[0039] where a k is the scattering coefficient vector of the output of the k-th layer. is the real-imaginary interaction unit. is the proximal mapping operator, and r k is the intermediate process variable of the k-th layer, and β k is the self-learning constraint in the k-th layer.

[0040] Step S4-4: Iterate successively until the last layer is reached.

[0041] Step S4-5: Calculate the loss function Loss in the network training using the output of the last layer:

[0042]

[0043] where a is the scattering coefficient vector. is the target scattering coefficient vector output by the last layer N.

[0044] Optionally, step S5 specifically includes:

[0045] Step S5-1: Input the target radar echo data into the trained low-order Doppler compensation module obtained in step S2 to perform low-order Doppler error compensation.

[0046] Step S5-2: Construct a local observation matrix from the output of the low-order Doppler error compensation module and input it into the trained combined adaptive network obtained in step S4 for iterative calculation.

[0047] Step S5-3: Output the structural parameters of the target radar echo data after iteration is completed.

[0048] Optionally, the step S5-3 specifically includes:

[0049] After iteration is completed, obtain the target scattering coefficient vector of the target radar echo data, and solve the target scattering coefficient vector to obtain the structural parameters of the target radar echo data:

[0050]

[0051] In the formula, is the finally estimated target scattering coefficient, G(a) is the regularization constraint, and λ is the weight coefficient;

[0052] Solving the target scattering coefficient vector is divided into two steps: gradient descent and proximal operator:

[0053] r k = a k-1 - ρH T (Ha k-1 - s)

[0054] a k = prox λ,g (r k )

[0055] In the formula, is the total observation matrix, ρ is the gradient descent step size, prox λ,g (·) is the proximal mapping operator, and g is the regularization constraint.

[0056] According to the second aspect of the present invention, there is also provided a BiSAR target structure parameter estimation system for depth under motion error, which is applied to the BiSAR target structure parameter estimation method for depth under motion error described in any one of the technical solutions in the first aspect of the present invention. The BiSAR target structure parameter estimation system for depth under motion error includes:

[0057] BiSAR parameter imaging network, including a low-order Doppler error compensation module, an adaptive high-order space-variant error compensation module, and a depth proximal mapping module;

[0058] Data processing module, configured to perform rough imaging using the output data of the trained low-order Doppler compensation module, generate an observation matrix by combining the imaging result and set parameters, and divide it into multiple local observation matrices;

[0059] Combined adaptive network, configured as a network with a multi-level cascade structure, where each layer of the network structure is a combination of an adaptive high-order space-variant error compensation module and a depth proximal mapping module, and is used to obtain network parameters through deep learning and adaptively compensate for high-order space-variant errors in the observation matrix

[0060] A data input module, configured to input target radar echo data into the trained low-order Doppler compensation module and the trained combined adaptive network;

[0061] A data output module, configured to output the structural parameters of the target radar echo data.

[0062] According to the third aspect of the present invention, there is also provided a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the computer program, it can implement the steps of the method for estimating the BiSAR target structure parameters under motion error according to any one of the technical solutions in the first aspect of the present invention.

[0063] According to the third aspect of the present invention, there is also provided a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it can implement the steps of the method for estimating the BiSAR target structure parameters under motion error according to any one of the technical solutions in the first aspect of the present invention.

[0064] Advantageous effects:

[0065] 1. Through the above technical solutions, the present invention uses the method for estimating the BiSAR target structure parameters under motion error to estimate the target parameter information. Compared with the target parameter estimation methods in the existing related technologies, the BiSAR parameter imaging network established by the present invention based on the parameter echo model and in the presence of motion error can break through the limitations of the point target model in the existing related algorithms, realize the compensation of two types of motion errors, can effectively improve the accuracy of the target parameter estimation results, improve the operation efficiency, realize efficient and high-precision target parameter estimation, and can be widely applied to BiSAR imaging.

[0066] 2. Other advantageous effects or advantages of the present invention will be described in detail in the specific implementation manners. Description of the Drawings

[0067] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention, and those skilled in the art can obtain other drawings without creative efforts based on these drawings.

[0068] Among them:

[0069] Figure 1It is a schematic diagram of the process flow of the method for estimating the BiSAR target structure parameters under motion error provided by an exemplary embodiment of the present invention;

[0070] Figure 2 It is a schematic diagram of a typical configuration of BiSAR;

[0071] Figure 3 It is a schematic diagram of the architecture process of the method for estimating the BiSAR target structure parameters under motion error of the present invention. In this figure, DECM is the low-order Doppler error compensation module, AHCM is the adaptive high-order spatially variant error compensation module, and DPMM is the depth proximal mapping module;

[0072] Figure 4 It is a schematic diagram of the processing flow of the low-order Doppler error compensation module (DECM) provided by an embodiment of the present invention. In this figure, ReLu is the rectified linear unit function, conv represents convolution, concat represents co-input, Avg Pool represents average pooling, FC represents the fully connected layer, and CoBlock and SpBlock are two sub-network structures respectively;

[0073] Figure 5 It is a schematic diagram of the processing flow of the adaptive high-order spatially variant error compensation module (AHCM) provided by an embodiment of the present invention. In this figure, ReLu is the rectified linear unit function and conv represents convolution;

[0074] Figure 6 It is a schematic diagram of the structure of the real and imaginary part interaction unit (RIIU) in the depth proximal mapping module (DPMM) provided by an embodiment of the present invention. In this figure, PReLu is the parametric rectified linear unit and conv represents the convolutional layer;

[0075] Figure 7 It is a schematic diagram of the single-layer network structure of the imaging network provided by an embodiment of the present invention. In this figure, RB is the residual network and RIIU is the real and imaginary part interaction unit;

[0076] Figure 8 It is a diagram of the estimated result of the target structure parameters verified in an exemplary embodiment of the present invention. Detailed implementation manners

[0077] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention.

[0078] Accordingly, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0079] In addition, the terms "including" and "having" and any variations thereof mentioned in the description of the present invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but optionally further includes other steps or units not listed, or optionally further includes other steps or units inherent to these processes, methods, products, or devices. It should also be noted that in the embodiments of the present invention, words such as "exemplary" or "for example" are used to represent examples, illustrations, or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the embodiments of the present invention should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Rather, the use of words such as "exemplary" or "for example" is intended to present relevant concepts in a specific manner.

[0080] For the convenience of those skilled in the relevant art to have a clearer and more accurate understanding of the technical solution of the present invention, the following will describe in more detail the problems to be solved by the present invention in combination with the existing relevant technologies.

[0081] In the existing relevant technologies, the algorithms for compensating motion errors generally include: the phase gradient autofocus algorithm (this algorithm is based on strong outliers and cannot adapt to complex targets), the algorithm based on Doppler information and image entropy (this algorithm cannot be applied to extended targets), and the sparse-driven motion error compensation method (which is not only sensitive to parameters but also has low computational efficiency).

[0082] For BiSAR echoes, they include: the phase modulation component caused by the historical change of the target distance, the envelope and phase changes caused by the target scattering characteristics, and the error phase caused by motion errors.

[0083] However, the current algorithms for compensating motion errors not only have their own limitations, but also generally are based on the point target scattering model, ignoring the influence of the target scattering characteristics. The structural information of the target in the imaging result will be missing and the two different error categories in the motion errors are not considered, ultimately resulting in a low quality of the imaging result, which is not conducive to BiSAR image interpretation and target recognition.

[0084] For example, in the literature "Azimuth Migration-Corrected Phase Gradient Autofocus for Bistatic SAR Polar Format Imaging" in IEEE Geoscience and Remote Sensing Letters, vol. 18, no. 4, pp. 697-701, April 2021, the Phase Gradient Autofocus (PGA) algorithm is used for SAR motion error estimation and the motion error is compensated. However, algorithms such as this for motion compensation and parameter estimation based on strong outliers are prone to failure under complex targets.

[0085] For another example, in the literature "A Rise-Dimensional Modeling and Estimation Method for Flight Trajectory Error in Bistatic Forward-Looking SAR" in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 10, no. 11, pp. 5001-5015, Nov. 2017, the image entropy of the imaging result is directly used for processing. Although the final image quality will be higher, since the image evaluation index has no direct relationship with the motion error, the focusing quality may be limited.

[0086] At the same time, the above methods are all carried out under the point model, and for extended complex targets, the structural information of the target may be lost.

[0087] In summary, in the existing related technologies, the motion error compensation algorithm not only has limitations in the model, but also cannot achieve efficient and high-precision target parameter estimation, and is limited in BiSAR imaging applications.

[0088] After research, the inventors believe that to accurately compensate for the motion error in BiSAR data, the key lies in two aspects. On the one hand, a BiSAR parametric echo model considering the motion error needs to be established, and on the other hand, different compensations need to be carried out for low-order Doppler errors and high-order spatially variant errors.

[0089] In view of this, the present invention provides a brand-new solution, that is, a method for estimating the structural parameters of a BiSAR target under motion error. The method of the present invention constructs a parameter estimation of depth BiSAR under motion error (DBPE-net), establishes a parameterized target echo model of BiSAR with two different motion errors according to the BiSAR structured scattering model, and compensates the first type of error through a pre-trained low-order Doppler error compensation module (Doppler Error Compensation Module, DECM). After that, the target parameters are obtained by compensating the high-order spatial-variance error through the combined module of the adaptive high-order spatial-variance error compensation module (Adaptive High-order Spatial-variance Error Compensation Module, AHCM) and the deep proximal mapping module (Deep Proximal Mapping Module, DPMM). It can break through the inherent limitations of the point target model and the attribute scattering center model in the existing related algorithms, compensate for two types of motion errors, effectively improve the accuracy of the target parameter estimation result and improve the operation efficiency.

[0090] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings.

[0091] As Figure 1 shown, according to the first aspect of the present invention, a method for estimating the structural parameters of a BiSAR target under motion error is provided, including the following steps:

[0092] Step S1: Build a BiSAR parameter imaging network based on the parameter echo model and in the presence of motion error. The BiSAR parameter imaging network includes a low-order Doppler error compensation module, an adaptive high-order spatial-variance error compensation module, and a deep proximal mapping module; wherein, the low-order Doppler error compensation module is configured to be able to compensate for the low-order Doppler error in the radar echo by estimating the Doppler center frequency, the chirp rate, and the third-order error coefficient, the adaptive high-order spatial-variance error compensation module is configured to be able to reconstruct the observation matrix under the high-order spatial-variance error by learning, and the deep proximal mapping module is configured to be able to implement an approximate point operator function;

[0093] Step S2: Pre-train the low-order Doppler compensation module;

[0094] Step S3: Use the output data of the trained low-order Doppler compensation module for rough imaging, generate an observation matrix in combination with the imaging result and the set parameters, and divide it into multiple local observation matrices;

[0095] Step S4: Train the adaptive high-order spatially variant error compensation module and the deep proximal mapping module to obtain a combined adaptive network. Input the local observation matrix obtained in Step S3 into the obtained combined adaptive network, and obtain network parameters through deep learning to adaptively compensate for the high-order spatially variant errors in the observation matrix. Among them, the combined adaptive network is configured as a network with a multi-level cascade structure, and each layer of this network structure is a combination of an adaptive high-order spatially variant error compensation module and a deep proximal mapping module;

[0096] Step S5: Input the target radar echo data into the trained low-order Doppler compensation module obtained in Step S2 and the trained combined adaptive network obtained in Step S4 in sequence to obtain the structural parameters of the target radar echo data.

[0097] Through the above technical solution, the present invention uses a method for estimating the structural parameters of a deep BiSAR target under motion errors to estimate the target parameter information. Compared with the target parameter estimation methods in the existing related technologies, the BiSAR parameter imaging network established by the present invention based on the parameter echo model and in the presence of motion errors can break through the limitations of the point target model in the existing related algorithms, realize the compensation of two types of motion errors, and can improve the operation efficiency on the basis of effectively improving the accuracy of the target parameter estimation result, realize high-efficiency and high-precision target parameter estimation, and can be widely applied to BiSAR imaging.

[0098] The following describes the method for estimating the structural parameters of a deep BiSAR target under motion errors of the present invention in conjunction with an exemplary embodiment.

[0099] It should be noted that this exemplary embodiment uses model simulation data for experimental verification, and the operation is performed by a 10th generation Intel Xeon Processor (Skylake, IBRS, 64GB RAM) and an NVIDIA V100 (32GB memory).

[0100] S0: Theoretical analysis and preparation.

[0101] S01: Model basis.

[0102] According to the requirements of the actual imaging scenario, the method for estimating the structural parameters of a deep BiSAR target under motion errors proposed by the present invention mainly targets targets composed of five typical scattering models: point targets, linear targets, vertical linear targets, flat plates, and dihedral angles.

[0103] The typical configuration of BiSAR can be referred to Figure 2 as shown.

[0104] In the BiSAR configuration, and respectively represent the elevation angle and azimuth angle of the transmitter and the receiver, where η is the slow-time variable, and the positions of the two platforms are [x T (η), y T (η), z T (η)] and [x R (η), y R (η), z R (η)].

[0105] For any point target P(x m , y n , 0) in the scene, the range history of the target can be approximated as:

[0106]

[0107] The echo expression of the target echo in the range frequency domain after two-dimensional matched filtering is:

[0108]

[0109] Where:

[0110]

[0111]

[0112] In the formula, θ T (η) represents the elevation angle of the transmitter, represents the azimuth angle of the transmitter, θ R (η) represents the elevation angle of the receiver, represents the azimuth angle of the receiver, η is the slow-time variable, f r is the range frequency domain variable, x m and y n are the two-dimensional coordinates of the point target in the scene area respectively; ΔR is the range difference from the reference point; Λ(f r , η; x m , y n ) is the scattering coefficient; is the phase caused by the propagation delay of transmission and reception.

[0113] For point targets, the scattering coefficient is a constant, but for other complex targets (such as linear targets, flat targets, etc.), the scattering coefficient varies with the observation angle and signal frequency.

[0114] Therefore, it is necessary to introduce the scattering models of flat targets, linear targets, vertical linear targets, and dihedral targets, which can be expressed as:

[0115]

[0116] In the above formula, Ap 、A l 、A lv 、A d are the amplitudes of the flat target, linear target, vertical linear target, and dihedral target, respectively; Λ p 、Λ l 、Λ lv 、Λ d are the echo envelope change parameters of the flat target, linear target, vertical linear target, and dihedral target caused by the attitudes and distance frequencies of the transmitter and receiver, respectively.

[0117] Meanwhile, the echo also contains an error phase, and its expression is as follows:

[0118]

[0119] f de (η) = f dce η + f drc η 2 + f dte η 3

[0120] In the formula, f de is the low-order Doppler error, and f dce 、f drc 、f dte The three parameters represent the Doppler center frequency error, Doppler frequency modulation slope error, and Doppler third-order error in sequence, and h is the high-order space-variant error; is the low-order Doppler error phase, is the high-order space-variant error phase.

[0121] Based on the above analysis, the parametric echo model S p :

[0122]

[0123]

[0124] In the above formula, is the full phase change, which includes the phase caused by the delay of transmission and reception low-order Doppler error phase high-order space-variant error phase η is the azimuth slow-time variable; f r is the range frequency domain variable; A p 、A l 、A lv 、A d 、A o are the amplitudes of the flat, linear target, vertical linear target, dihedral, and point target, respectively; The parameters of the reflection coefficient variation caused by the attitude and distance frequency of the transmitter and receiver for the flat plate, linear target, vertical linear target, and dihedral target, respectively; k p k l k lv k d The numbers of the flat plate, linear target, vertical linear target, and dihedral, respectively; m and n are the numbers of range cells and azimuth cells in the scene, respectively, and x m , y n are the range coordinate and azimuth coordinate of the corresponding cell; Z(η) is the attitude of the transmitter and receiver at the azimuth time, including the azimuth angle and elevation angle.

[0125] S02: Problem modeling.

[0126] The above echo model can be rewritten in matrix form s:

[0127]

[0128]

[0129] where Υ is the target echo envelope matrix, which is obtained by splicing the echo envelopes of the Υ p flat plate target, the echo envelopes of the Υ l linear target, the echo envelopes of the Υ lv vertical target, the echo envelopes of the Υ d dihedral target, and the echo envelope matrix of the Υ o point target; Φ all is the total phase change matrix, which is obtained by the Hadamard product of the Φ f transmission-reception delay phase, the Φ ef low-order Doppler error phase, and the Φ eh high-order space-variant error phase matrix (Hadamard product, ); a is the scattering coefficient vector of the target. Some parameter expressions are as follows:

[0130]

[0131]

[0132] where, is the scattering coefficient of the K p th flat plate target in the Mth range cell and the Nth azimuth cell. Similarly, is the scattering coefficient of the vertical target, is the scattering coefficient of the linear target, is the scattering coefficient of the dihedral target; Λ p is the parameter of the reflection coefficient variation of the flat plate target; f rZ is the frequency in each range direction; ηV is the moment in each azimuth direction; are the length and width of each board target; x M , y N is the same as x above m , y n ; [·] T is the matrix transpose operation. Υ l Υ lv Υ d Υ o and Φ f Φ ef Φ eh The expressions are similar and will not be elaborated here.

[0133] According to the above formula, solving the scattering coefficient vector a can solve the target parameters, that is, to solve the following problem,

[0134]

[0135] where, is the finally estimated target scattering coefficient, G(a) is the regularization constraint, and λ is its weight.

[0136] Solving this problem is divided into two steps: gradient descent and proximal mapping, which can be represented by the following formulas respectively:

[0137] r k = a k-1 - ρH T (Ha k-1 - s)

[0138] a k = prox λ,g (r k )

[0139] where, is the total observation matrix; ρ is the gradient descent step size; prox λ,g (·) is the proximal mapping, g, λ are the regularization constraint and its weight respectively; r k is the intermediate process variable.

[0140] S1: Build the network.

[0141] Build an adaptive learning network for target parameter estimation, including a local dictionary generation module and an adaptive learning module. The network input is the bistatic SAR target echo y i , and the output is the estimated scattering coefficient vector a. The target parameter estimation result is the target position and geometric parameters (A e ; x e , y e ; L e , H e)It can be directly obtained from the network output result a. For the architecture process of the BiSAR target structure parameter estimation method for depth under motion error, please refer to Figure 3 as shown below.

[0142] S2: Pretrain the low-order Doppler compensation module.

[0143] S21: Build the network structure of the low-order Doppler compensation module. Its processing flow can be referred to Figure 4 , first, divide the echo into two channels of real part and imaginary part and input them into the Resnet network structure. Among them, CoBlock increases the number of channels without downsampling, and SpBlock increases the number of channels through downsampling, and finally outputs three parameters {f dce , f dre , f dte}. Then use the following formula for parameter compensation:

[0144]

[0145] In the formula, is the output after low-order Doppler compensation, vec(·) is the column vectorization operation, exp(·) is the exponential function, and η is the azimuth slow time.

[0146] The loss function of the network is as follows:

[0147]

[0148] re(·) is to take the real part, im(·) is to take the imaginary part, and ‖·‖2 is the Euclidean distance operation.

[0149] S22: Generate simulated radar echo data s according to the motion error echo model. The parameter information used is shown in the following table:

[0150] Parameter type Value Center frequency 9.6 GHz Bandwidth 400 MHz Sampling frequency interval 10.25 MHz Transmitter elevation angle 28.02°～30.57° Transmitter azimuth angle 36.54°～41.53° Receiver elevation angle 26.23°～29.09° Receiver azimuth angle 3.95°～4.60° Number of angle sampling points 32

[0151] First, randomly determine the position and number of targets to generate echoes. Then, add low-order Doppler error and high-order space-variant error. Among them, the low-order Doppler error uses a fixed coefficient, f dce ranges from 0.8 - 1.2, f dre ranges from 0.018 - 0.022, f dte ranges from 0.018–0.022; the high-order space-variant error is represented by a randomly varying coefficient;

[0152] S23: Use the generated data to pretrain the low-order Doppler error compensation module.

[0153] S3: Use the output data of the trained low-order Doppler compensation module for rough imaging and create an observation matrix.

[0154] S31: Input the radar data into the pre-trained low-order Doppler compensation module to obtain the compensated data as the output.

[0155] S32: Coarsely image the compensated data using the PFA algorithm, and based on the imaging result, construct multiple local observation matrices.

[0156] S4: Train the combined network of the adaptive high-order spatially variant error compensation module and the deep proximal mapping module.

[0157] S41: Build the network structure. The processing flow of this network structure can be referred to as shown in Figure 5 Divide the observation matrix into the real part and the imaginary part and input them into the block network, estimate the influence of the high-order spatially variant error on the real part and the imaginary part of the observation matrix, and obtain the adjusted observation matrix. Its processing process can be expressed by the following formula:

[0158]

[0159] In the formula, re(·) is to take the real part, and im(·) is to take the imaginary part; is the original observation matrix; H me is the adjusted observation matrix; is Figure 5 the convolution, Relu activation function, and convolution operations shown in

[0160] In this step, if the original observation matrix is used, a large amount of computing and storage resources will be required, and the processing will be inefficient. Therefore, by inputting the local observation matrices divided in step S3 here, the large network can be divided into small processing units. In this way, the requirement for the computing power of the processor can be significantly reduced, the transportation efficiency can be improved, and efficient processing can be achieved without loss of accuracy.

[0161] S42: Perform gradient descent operations, and the operation formula is as follows:

[0162]

[0163] where, is the intermediate process variable of the k-th layer of the i-th local observation matrix; I is the identity matrix; ρ k μ k are the step sizes of the real part channel and the imaginary part channel in the k-th iteration respectively; is the value obtained by the i-th local observation matrix in the (k - 1)-th layer; is the data corresponding to the i-th local observation matrix in the output result of the network trained in step S2; V1 and V2 are adaptive matrices to adapt to the differences of different local observation matrices;

[0164] S43: Input the two-channel outputs of the adaptive high-order spatially variant error compensation module into the Deep Proximal Mapping Module (DPMM) for processing.

[0165] The DPMM includes an encoder-decoder structure, which can learn the ability to approximate the point operator to capture the complex relationships in the echoes under the influence of errors, breaking through the limitations of traditional manually specified regular expressions. In addition, the DPMM also includes a Real And Imaginary Part Interaction Unit (RIIU), which can handle the phase relationship between the real and imaginary parts. Its structure can be referred to Figure 7 as shown.

[0166] The input expression of the DPMM is as follows:

[0167]

[0168] where is the Real And Imaginary Part Interaction Unit, r k is the intermediate variable of the k-th layer, a k is the scattering coefficient vector of the output of the k-th layer, is the proximal mapping operator, β k is the self-learned constraint in the k-th layer;

[0169] S5: Iterative calculation.

[0170] Since the output of the last layer is the most important, the loss function of the entire network can be:

[0171]

[0172] where is the target scattering coefficient vector of the output of the last layer N.

[0173] The single-layer network structure of the imaging network can be referred to Figure 6 as shown. The processing flow of the method for estimating the structural parameters of the BiSAR target under motion error of the present invention can be referred to Figure 3 as shown.

[0174] Please refer to Figure 8 , Figure 8 The content shown is the test result graph of the generated simulation data, where each vertical line represents a target at that position. By comparing the true value and the estimated value, it can be seen that the parameter estimation result of the method of the present invention is close to the actual motion error.

[0175] The method of the present invention will be further described below in combination with a specific exemplary embodiment.

[0176] Step 1: Obtain training data.

[0177] To further verify the effectiveness of the method, the inventor used real radar data collected by himself, and the radar parameters were still the same as those of the simulation data. Among them, the transmitting station flew at an altitude of 1649 m, and the receiving station flew 400 m behind it, at a parallel distance of 1652 m and an altitude of 2153 m;

[0178] Step 2: Select the regional echo data of the target to be built and use the pre-trained low-order Doppler network for compensation;

[0179] Step 3: Coarsely image the compensated data and construct a local observation matrix;

[0180] Step 4: Input the local observation matrix into the subsequent network for iteration, and finally output the scattering coefficient vector of the target.

[0181] In summary, through the BiSAR parameter imaging motion error compensation method based on deep learning of the present invention, the limitations of the point target model and the attribute scattering center model in the existing related technologies can be broken through; moreover, the method of the present invention uses different structures to process the two motion errors respectively, and can achieve more accurate motion error compensation; at the same time, the method of the present invention can also effectively reduce the calculation amount, improve the calculation efficiency, and can achieve efficient and accurate target parameter estimation.

[0182] According to the second aspect of the present invention, there is also provided a deep BiSAR target structure parameter estimation system under motion error, which is applied to the deep BiSAR target structure parameter estimation method under motion error in any one of the technical solutions of the first aspect of the present invention. The deep BiSAR target structure parameter estimation system under motion error includes:

[0183] A BiSAR parameter imaging network, including a low-order Doppler error compensation module, an adaptive high-order space-variant error compensation module, and a deep proximal mapping module;

[0184] A data processing module, configured to coarsely image using the output data of the trained low-order Doppler compensation module, generate an observation matrix in combination with the imaging result and set parameters, and divide it into multiple local observation matrices;

[0185] A combined adaptive network, configured as a network with a multi-level cascade structure, and each layer of the network structure is a combination of an adaptive high-order space-variant error compensation module and a deep proximal mapping module, and is used to obtain network parameters through deep learning and adaptively compensate the high-order space-variant errors in the observation matrix

[0186] A data input module, configured to input target radar echo data into the trained low-order Doppler compensation module and the trained combined adaptive network;

[0187] A data output module for outputting the structural parameters of the target radar echo data.

[0188] The BiSAR target structure parameter estimation system for depth under motion error according to the present invention can break through the limitations of the point target model and the attribute scattering center model in the existing related algorithms, compensate for two types of motion errors, effectively improve the accuracy of the target parameter estimation result and improve the operation efficiency.

[0189] According to the third aspect of the present invention, there is also provided a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it can implement the steps of the method for estimating the structural parameters of the BISAR target for depth under motion error in any one of the technical solutions in the first aspect of the present invention.

[0190] It can be understood that in this embodiment, the memory may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a read-only memory, a flash memory, a hard disk, or a solid-state drive; in addition, the memory may also include a combination of the above types of memories. The present invention does not make specific limitations on this.

[0191] Similarly, the processor can implement or execute various exemplary logical steps described in connection with the disclosure of the present invention. The processor may be a central processing unit, a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field-programmable gate array, or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute various exemplary logical steps described in connection with the disclosure of the present invention. The processor may also be a combination of computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0192] According to the fourth aspect of the present invention, there is also provided a computer-readable storage medium, on which a computer program is stored, characterized in that when the computer program is executed by a processor, it can implement the steps of the method for estimating the structural parameters of the BISAR target for depth under motion error in any one of the technical solutions in the first aspect of the present invention.

[0193] In this embodiment, a computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (a non-exhaustive list) of the computer-readable storage medium include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a register, a hard disk, an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above, or any other form of computer-readable storage medium well known in the art. An exemplary storage medium is coupled to the processor such that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium may also be a component of the processor. The processor and the storage medium may be located in an application specific integrated circuit (ASIC). In the embodiments of the present invention, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0194] The above is only the specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for estimating the structural parameters of BiSAR targets at depth under motion errors, characterized in that, It includes the following steps: Step S1: Build a BiSAR parameter imaging network based on the parametric echo model with motion errors. The BiSAR parameter imaging network includes a low-order Doppler error compensation module, an adaptive high-order spatially variant error compensation module, and a deep proximal mapping module. Among them, the low-order Doppler error compensation module is configured to be able to compensate for the low-order Doppler error in the radar echo by estimating the Doppler center frequency, the chirp rate, and the third-order error coefficient. The adaptive high-order spatially variant error compensation module is configured to be able to reconstruct the observation matrix under high-order spatially variant errors through learning. The deep proximal mapping module is configured to be able to implement the proximal mapping operator function. Step S2: Pretrain the low-order Doppler compensation module. Step S3: Coarsely image using the output data of the trained low-order Doppler compensation module, generate an observation matrix by combining the imaging result and the set parameters, and divide it into multiple local observation matrices. Step S4: Train the adaptive high-order spatially variant error compensation module and the deep proximal mapping module to obtain a combined adaptive network. Input the local observation matrices obtained in Step S3 into the obtained combined adaptive network, and obtain the network parameters through deep learning to adaptively compensate for the high-order spatially variant errors in the observation matrix. Among them, the combined adaptive network is configured as a network with a multi-level cascade structure, and each layer of this network structure is a combination of an adaptive high-order spatially variant error compensation module and a deep proximal mapping module. Step S5: Input the target radar echo data into the trained low-order Doppler compensation module obtained in Step S2 and the trained combined adaptive network obtained in Step S4 in sequence to obtain the structural parameters of the target radar echo data.

2. The method for estimating the BiSAR target structure parameters of the depth under motion error according to claim 1, wherein, The specific content of Step S2 includes: Step S2-1: Generate simulated radar echo data s according to the motion error echo model. Step S2-2: Separate the real part and the imaginary part of the simulated radar echo data s and input them into two separate network channels respectively. Step S2-3: Perform iterative learning and output three parameters {f dce , f dre , f dte}, and perform parameter compensation according to the following formula: The loss function loss in network training me is as follows: where \(f\) dce is the Doppler center frequency error, \(f\) dre is the Doppler chirp frequency error, \(f\) dte is the Doppler third-order error, is the output after low-order Doppler compensation, vec[·] is the column vectorization operation, exp[·] is the exponential function, \(\eta\) is the azimuth slow time, re(·) is to take the real part, im(·) is to take the imaginary part, and \(\|\cdot\|_2\) is the Euclidean distance operation.

3. The method for estimating BiSAR target structure parameters of depth under motion error according to claim 2, characterized in that The specific content of Step S2-1 includes: Step S2-1-1: Based on the scattering models of point targets, line targets, flat targets, vertical targets, and dihedral targets, as well as the motion error model, establish a parameterized echo model S under motion errors p : In the formula, is the full-phase change, which includes the phase caused by the delay of transmission and reception low-order Doppler error phase and high-order spatially variant error phase p m and n are the number of range cells and azimuth cells in the scene respectively, A l A lv A d and A o are the amplitudes of the flat target, linear target, vertical linear target, dihedral target and point target respectively, are the reflection coefficient change parameters of the flat target, linear target, vertical linear target and dihedral target caused by the attitudes and distance frequencies of the transmitter and receiver respectively, η is the azimuth slow-time variable, f r is the range frequency-domain variable, k p k l k lv and k d are the numbers of the flat target, linear target, vertical linear target and dihedral target respectively, x m and y n are the range coordinate and azimuth coordinate of the corresponding cell respectively, Z(η) is the attitude of the transmitter and receiver at the azimuth time, including the azimuth angle and elevation angle; Step S2-1-2: Rewrite the parameterized echo model S p as the simulated radar echo data s in matrix form: γ = [Υ p Υ l Υ lv Υ d Υ o ​ Where, Υ is the target echo envelope matrix, which is obtained by splicing the flat target echo envelope matrix Υ p , the line target echo envelope matrix Υ l , the vertical target echo envelope matrix Υ lv , the dihedral target echo envelope matrix Υ d and the point target echo envelope matrix Υ o . Φ all is the total phase change matrix, which is obtained by the Hadamard product of the transmit-receive delay phase matrix Φ f , the low-order Doppler error phase matrix Φ ef and the high-order space-variant error phase matrix Φ eh . is the Hadamard product operation, and a is the scattering coefficient vector.​ 4. The method for estimating the BiSAR target structure parameters of the depth under motion error according to claim 1, wherein The specific content of Step S3 includes: Step S3-1: Input the radar data into the low-order Doppler compensation module trained in Step S2 to obtain the compensated data as the output. Step S3-2: Coarsely image the compensated data obtained in Step S3-1 using the PFA algorithm, generate an observation matrix by combining the imaging result and the set parameters, and divide the observation matrix into multiple local observation matrices.

5. The method for estimating BiSAR target structure parameters of depth under motion error according to claim 3, characterized in that, The specific content of Step S4 includes: Step S4-1: Train the adaptive high-order spatially variant error compensation module and the deep proximal mapping module to obtain a combined adaptive network. Step S4-2: Divide the partial observation matrix obtained in Step S3 into a real part channel and an imaginary part channel, and input them into the adaptive high-order spatially variant error compensation module in the combined adaptive network obtained in Step S4-1 respectively to obtain the adjusted partial observation matrix H mei , and perform gradient descent operation through the following formula: where re(·) is to take the real part and im(·) is to take the imaginary part, is the intermediate process variable of the k-th layer of the i-th local observation matrix, I is the identity matrix, ρ k , μ k are the step sizes of the real part channel and the imaginary part channel in the k-th iteration respectively; is the value obtained from the (k - 1)-th layer of the i-th local observation matrix; is the data corresponding to the i-th local observation matrix in the output result of the trained network in step S2; V1 and V2 are adaptive matrices to adapt to the differences of different local observation matrices; Step S4-3: The real part channel output and the imaginary part channel output of the adaptive high-order spatially variant error compensation module in Step S4-2 are respectively input to the depth proximal mapping module for processing, and successively pass through the real-imaginary interaction unit, the encoding-decoding unit, and the real-imaginary interaction unit, and the output result is a k : where a k is the scattering coefficient vector of the output of the k-th layer, is the real and imaginary part interaction unit, is the proximal mapping operator, r k is the intermediate process variable of the k-th layer, β k is the self-learning constraint in the k-th layer; Step S4-4: Iterate successively until the last layer is reached. Step S4-5: Calculate the loss function Loss in the network training using the output of the last layer. where a is the scattering coefficient vector, is the target scattering coefficient vector output by the last layer N.

6. The method for estimating BiSAR target structure parameters of depth under motion error according to claim 5, characterized in that, The specific content of Step S5 includes: Step S5-1: Input the target radar echo data into the trained low-order Doppler compensation module obtained in Step S2 to perform low-order Doppler error compensation. Step S5-2: Construct a local observation matrix from the output of the low-order Doppler error compensation module and input it into the trained combined adaptive network obtained in Step S4 for iterative calculation. Step S5-3: Output the structural parameters of the target radar echo data after iteration is completed.

7. The method for estimating the BiSAR target structure parameters of the depth under motion error according to claim 6, characterized in that The specific steps of Step S5-3 include: After iteration is completed, obtain the target scattering coefficient vector of the target radar echo data, and solve the target scattering coefficient vector to obtain the structural parameters of the target radar echo data: In the formula, is the finally estimated target scattering coefficient, G(a) is the regularization constraint, and λ is the weight coefficient; Solving the target scattering coefficient vector is divided into two steps: gradient descent and proximal operator approximation: r k = a k-1 -ρH T (Ha k-1 - s) a k = prox λ,g (r k ) where is the total observation matrix, ρ is the gradient descent step size, and prox λ,g (·) is the proximal mapping, and g is the regularization constraint.

8. A BiSAR target structure parameter estimation system for depth under motion error, characterized in that, Applied to the BiSAR target structure parameter estimation method for depth under motion error described in any one of claims 1-7, the BiSAR target structure parameter estimation system for depth under motion error includes: A BiSAR parameter imaging network, including a low-order Doppler error compensation module, an adaptive high-order spatially variant error compensation module, and a depth proximal mapping module; A data processing module, configured to perform rough imaging using the output data of the trained low-order Doppler compensation module, generate an observation matrix by combining the imaging result and set parameters, and divide it into multiple local observation matrices; A combined adaptive network, configured as a network with a multi-level cascade structure, where each layer of the network structure is a combination of an adaptive high-order spatially variant error compensation module and a depth proximal mapping module, and is used to obtain network parameters through deep learning to adaptively compensate for high-order spatially variant errors in the observation matrix A data input module, configured to input target radar echo data to the trained low-order Doppler compensation module and the trained combined adaptive network; A data output module, configured to output the structural parameters of the target radar echo data.

9. A computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it can implement the steps of the BiSAR target structure parameter estimation method for depth under motion error described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it can implement the steps of the BiSAR target structure parameter estimation method for depth under motion error described in any one of claims 1-7.

Images