Sparse ISAR translation compensation and imaging integration method based on super network
Through the integrated method of sparse ISAR translation compensation and imaging based on hypernetwork, the problem of degradation of translation compensation performance of ISAR system under low signal-to-noise ratio and echo defect conditions is solved, and high-quality focused imaging and robustness are achieved. Reduces computational complexity.
Patent Information
- Application Number
- CN202410972830.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-19
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-07-19
AI Technical Summary
The existing ISAR system has a degradation in the translational compensation performance under low signal-to-noise ratio and echo defect conditions, resulting in defocusing or inability to image. The initialization parameters are sensitive to echo signal-to-noise ratio, and require manual adjustments to limit the Practicality.
Using a sparse ISAR translational compensation and imaging integrated method based on hypernetwork, by building a finite-layer deep network, echo noise characteristics are extracted, optimal parameters are dynamically generated, time and space complexity are reduced, and robustness is improved.
It improves the translation compensation accuracy and ISAR imaging quality, realizes high-quality focused imaging, improves the robustness of different signal-to-noise ratio scenarios, and reduces the computational complexity.
Smart Images

Figure CN118915064A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of radar remote sensing technology, and further proposes an ISAR translation compensation and imaging integration method, which can be used for shipborne and airborne ISAR systems to extract important information on target shape, structure and posture. Background Art
[0002] Inverse synthetic aperture radar (ISAR) can obtain high-resolution two-dimensional images of non-cooperative targets all day and all weather, so it plays an important role in space and air target monitoring, and accurate translation compensation is the key to achieving high-resolution focused imaging of targets. However, the radar is limited by the working mode and cannot continuously observe the target, resulting in azimuth echo defects; at the same time, the electromagnetic environment is becoming increasingly complex, and the signal-to-noise ratio of the target echo is very low. Since the existing methods solve the translation compensation and imaging problems step by step, the translation compensation performance drops sharply under the above-mentioned low signal-to-noise ratio and echo defect conditions, resulting in image defocus or even failure to image. At the same time, the initialization parameters of the existing translation compensation and imaging methods are sensitive to the echo signal-to-noise ratio, and usually require multiple manual parameter tuning, which greatly limits their practicality. Therefore, it is urgent to study the integrated processing method of ISAR motion compensation and imaging under complex observation conditions such as low signal-to-noise ratio and echo defect.
[0003] The patent document with application number CN202110263357.9 discloses a "range alignment method and system for inverse synthetic aperture radar imaging". The technology is completed in four parts. First, the ISAR raw data after pulse compression is saved, and each pulse echo pulse compression is Fourier transformed to obtain the frequency domain data of each pulse; then the frequency domain data of each pulse is subjected to a recursive algorithm to obtain the recursive parameters and deduces the information outside the radar transmission signal bandwidth; then the recursive pulse frequency domain data is imaged by inverse Fourier transform to obtain a super-resolution image; finally, the super-resolution image is subjected to range alignment processing by the global minimum entropy algorithm to obtain the range alignment error of each pulse compression image. This method aims to minimize the Shannon entropy of the average range image and performs offset estimation of the range image envelope. The algorithm has a large number of iterations, resulting in a long algorithm operation time.
[0004] The patent document with application number CN2020975711.6 discloses a "SR-ISAR imaging and autofocusing method based on CV-ADMMN", which first models the sparse aperture ISAR echo signal, then constructs a CV-ADMMN forward propagation model including a reconstruction layer, a noise reduction layer, and a multiplier update layer, and embeds the autofocusing module into the forward propagation model, and finally uses CV-ADMMN to solve the L1 norm optimization problem to obtain the final imaging result. However, since this method regards different echo signal-to-noise ratio scenarios as different tasks, different imaging networks need to be trained according to different signal-to-noise ratios during training, so it lacks echo signal-to-noise ratio robustness, which greatly increases the time and space complexity. Summary of the invention
[0005] The purpose of the present invention is to address the deficiencies of the above-mentioned prior art and propose a sparse ISAR translational compensation and imaging integration method based on a hypernetwork, so as to improve the accuracy of translational compensation and the quality of ISAR imaging under low signal-to-noise ratio conditions, improve the robustness of the network to the echo signal-to-noise ratio, and reduce the time and space complexity.
[0006] The technical solution of the present invention is to expand the iterative method of integrating translation compensation and ISAR imaging into a deep network of finite layers, and introduce a super network that can extract echo noise characteristics to generate deep network parameters for echoes with different signal-to-noise ratios. The implementation steps include the following:
[0007] (1) Based on the translation matrix F(θ) that produces translation error for the echo and the sparse measurement matrix Φ a , by performing a range Fourier transform D on the ISAR image r and the azimuthal Fourier transform D a , establish the sparse observation model S under echo defect:
[0008] S=F(θ)☉(D r XD a Φ a )+Z
[0009] Where θ represents the translation parameter, X represents the ISAR image to be solved, Z represents the noise, and ⊙ represents the Hadamard product;
[0010] (2) According to the sparse observation model S under echo defect conditions, the sparse ISAR translation compensation and imaging optimization objective function J1(X,θ) based on L1 norm optimization is constructed:
[0011]
[0012] Where, E(θ)=(F(θ)) * , (·) * represents the conjugate operation, ||·||F and ||·||1 represent the matrix F norm and matrix 1 norm respectively, and λ is the weight coefficient;
[0013] (3) According to the TMC-EADMM algorithm, the optimization objective function J1(X,θ) is iteratively solved to obtain the iterative formula at the current iteration time t, that is, to update the ISAR image X (t) and estimate the translation parameter θ (t) :
[0014]
[0015] Among them, S (t) Indicates that after E(θ (t) ) The tth echo after compensation, S( t) =E(θ (t) )⊙S (t-1) , X (t) represents the t-th reconstructed ISAR image, θ (t) represents the t-th estimated translation parameter, D r and D a Respectively represent the distance dictionary and the direction dictionary, Φ a represents a sparse observation matrix;
[0016] (4) According to the iterative formula X (t) and θ (t) Constructing a sparse ISAR translation compensation and imaging integrated network based on a hypernetwork:
[0017] 4a) Take T iterations in the iterative algorithm and convert the iterative formula X( t ) and θ( t ) corresponds to an image reconstruction module and a translation compensation module;
[0018] 4b) connecting the image reconstruction module and the translation compensation module in sequence to obtain the t-th layer sub-network, and then cascading multiple sub-networks to obtain a backbone network including T-layer sub-networks;
[0019] 4c) Design a hypernetwork based on residual network, extract the features of ISAR images with different signal-to-noise ratios through the hypernetwork, and generate the optimal parameters such as regularization coefficient λ, penalty factor ρ and relaxation factor γ for each layer of the backbone network according to these features;
[0020] 4d) combining the backbone network and the corresponding super networks of each layer to form an integrated network of translation compensation and ISAR imaging;
[0021] (5) Randomly generate N b The amplitudes of the scattering points X follow a normal distribution and the phases follow a uniform distribution. Different ISAR scenarios are obtained by setting different signal-to-noise ratios, thereby generating N bEcho data in different scenarios Use it as a training set;
[0022] (6) inputting the training set into the translation compensation and ISAR imaging integrated network, training it using the gradient descent method, and obtaining a trained translation compensation and ISAR imaging integrated network;
[0023] (7) The measured Yak-42 data is input into the trained translation compensation and imaging integrated network, and the final ISAR imaging results are obtained through network forward propagation.
[0024] Compared with the prior art, the present invention has the following advantages:
[0025] First, the present invention constructs a sparse ISAR translation compensation and imaging optimization objective function based on L1 norm optimization, and proposes a sparse ISAR joint translation compensation and imaging method TMC-EADMM, thereby improving the translation compensation accuracy and ISAR imaging quality, and can achieve high-quality focused imaging that is superior to existing algorithms.
[0026] Second, the present invention constructs a sparse ISAR translational compensation and high-resolution imaging integrated network based on a hypernetwork, extracts ISAR image features with different signal-to-noise ratios through the hypernetwork, dynamically generates the optimal internal parameters of the backbone network, and improves the computational efficiency and robustness to different signal-to-noise ratios.
[0027] Third, the present invention designs a sparse ISAR translation compensation and imaging integrated loss function, thereby achieving effective network training and effectively improving the translation compensation accuracy and imaging quality in low signal-to-noise ratio and echo defects. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] Figure 1 It is a flow chart for realizing the present invention;
[0029] Figure 2 Schematic diagram of the network structure of sparse ISAR translation compensation and imaging integration based on hypernetwork in the present invention;
[0030] Figure 3 A schematic diagram of the backbone network structure constructed in the present invention;
[0031] Figure 4 A schematic diagram of the super network structure constructed in the present invention;
[0032] Figure 5 The envelope alignment comparison results are obtained by testing the present invention and the existing minimum entropy envelope alignment plus expansion network self-focusing imaging method under 50% defect conditions at echo signal-to-noise ratios of 5dB and 10dB.
[0033] Figure 6 The imaging comparison results are obtained by testing the present invention and the existing minimum entropy envelope alignment plus expansion network self-focusing imaging method under 50% defect conditions at echo signal-to-noise ratios of 5dB and 10dB. DETAILED DESCRIPTION
[0034] The embodiments and effects of the present invention are further described in detail below in conjunction with the accompanying drawings. The embodiments are only used to illustrate the implementation scheme of the present invention and do not constitute any limitation to the present invention. Obviously, for professionals in this field, after understanding the content and principles of the present invention, it is possible to make various modifications and changes in form and details without departing from the principles and structures of the present invention. However, these modifications and changes based on the ideas of the present invention are still within the protection scope of the present invention.
[0035] refer to Figure 1 , the implementation steps of this embodiment include the following:
[0036] Step 1: Establish a sparse observation model under echo defect conditions.
[0037] 1.1) Assume that the radar transmits a linear frequency modulated pulse signal, and the target is represented by H scattering points. The de-linear frequency modulated echo signal s received by the radar is obtained r (f r ,m):
[0038]
[0039] Among them, σ h is the backscattering coefficient of the hth scattering point, rect(·) is the rectangular window function, is the distance frequency, R h (t m ) is the distance from the hth scattering point in the mth echo to the radar, m = 1, 2, ... M, M represents the total number of pulses, B is the bandwidth of the transmitted signal, f c is the center frequency, r m =r R (t m )-R ref (t m ) is the target translation component, R ref (t m ) is the reference distance that changes with slow time, (x h ,y h ) is the starting coordinate of the scattering point h, is the random phase error between echoes caused by the system error, is Gaussian white noise;
[0040] 1.2) According to the echo signal s received by the radar r (fr ,m) construct the translation matrix F(θ):
[0041]
[0042] in, is the translation matrix The element in the nth row and the lth column, θ={r1,…,r L ,φ1,…,φ L} represents the translation parameter, l=1,...,L, L represents the number of echoes, 0<L<M, f n =nB / N, n=1,…,N, N represents the number of sampling points, r l represents the envelope offset of the lth column echo, represents the total phase error between the echoes in the lth column;
[0043] 1.3) Let the size of the ISAR image be U×V, and the element in the image be i h (u,v), u∈[1,U], v∈[1,V];
[0044] 1.4) Perform Fourier transform on the ISAR image in range and azimuth, and add the translation error F n,l (θ), and the echo signal s containing translation error is obtained r (n,l):
[0045]
[0046] Where d is the azimuth defect vector, which means that in the case of sparse observation, the original vector of length Q is randomly downsampled to obtain a vector of length L. l is the lth element in d;
[0047] 1.5) r (n,l) as a matrix The elements of the nth row and the lth column in , construct the sparse observation model S with translation error in the case of echo defect:
[0048] S=F(θ)⊙(D r XD a Φ a )+Z
[0049] in, is a distance dictionary, whose element in row n and column u is
[0050] is the Doppler dictionary, whose element in row v and column m is
[0051] Represents the orientation observation matrix, which is obtained by removing I M×M The specific ML column in I M×M represents the identity matrix, ⊙ represents the Hadamard product,
[0052] represents the ideal two-dimensional image corresponding to the target, For noise.
[0053] In this embodiment, the size of the ISAR image is set to but not limited to P=128 and Q=256.
[0054] Step 2: According to the sparse observation model S under echo defect conditions, a sparse ISAR translation compensation and imaging optimization objective function based on L1 norm optimization is constructed.
[0055] 2.1) Take the conjugate of the translation matrix F(θ) to obtain the phase compensation matrix E(θ) = (F(θ)) * , the phase compensation of the real echo data S is performed through this matrix, that is, E(θ)⊙S;
[0056] 2.2) The imaging result X is transformed by Fourier transform in range and azimuth to obtain the echo result, namely D r XD a Φ a ;
[0057] 2.3) By subtracting steps 2.1) from 2.2) and squaring the F-norm of the difference, we get the optimization objective function J0(X,θ):
[0058]
[0059] Among them, ||·|| F represents the norm of matrix F;
[0060] 2.4) Add the sparse constraint term ||X||1 to the optimization objective function J0(X,θ) to obtain the sparse ISAR translation compensation and imaging optimization objective function J1(X,θ):
[0061]
[0062] Among them, ||·||1 represents the l1 norm and λ is the regularization coefficient.
[0063] Step 3: According to the TMC-EADMM algorithm, the optimization objective function J1(X,θ) is iteratively solved to obtain the iterative formula at the current iteration time t, that is, to update the ISAR image X (t) and estimate the translation parameter θ (t) .
[0064] 3.1) Fix the phase compensation term E(θ) in J1(X,θ) and obtain the updated ISAR image X (t) The expression is:
[0065]
[0066] in, is the distance dictionary, is the Doppler dictionary, represents the orientation measurement matrix, ⊙ represents the Hadamard product, represents the ideal two-dimensional image corresponding to the target, θ( t-1 ) represents the translation parameter estimated at the t-1th time, S (t-1) Indicates that after E(θ (t-1) ) is the t-th echo after compensation, λ is the regularization coefficient;
[0067] 3.2) Use EADMM algorithm to calculate X (t) Perform an iterative solution:
[0068] 3.2.1) According to the input echo signal S (t-1) , the slack variable for the b-1th iteration and Update ISAR image
[0069]
[0070] in,(·) H represents conjugate transpose;
[0071] 3.2.2) According to the parameters of the b-1th iteration and the ISAR image of the bth iteration Update Lagrange multipliers
[0072]
[0073] 3.2.3) According to the Lagrange multiplier of the bth iteration ISAR images of Update auxiliary variables
[0074]
[0075] Among them, S λ / ρ (·) represents the soft threshold function with parameter λ / ρ;
[0076] 3.2.4) Parameters and Perform relaxation calculations to obtain relaxation variables and
[0077]
[0078] Among them, ρ is the initialization parameter iteration step, γ is the relaxation factor;
[0079] 3.3) The ISAR image X obtained at the tth iteration (t) Fixed, extract the terms related to the translation in J1(X,θ) to obtain the estimated translation parameter θ (t) The expression is:
[0080]
[0081] Among them, θ (t) ={r1,…,r L ,φ1,…,φ L}, E(θ)=(F(θ)) * , F(θ) is the translation matrix, (·) * represents the conjugation operation;
[0082] 3.4) Set θ (t) In Expand the term and remove θ (t) The terms in the expression that are independent of θ give θ (t) The following simplified formula:
[0083]
[0084] Among them, Re{·} represents the real part operation, E n,l (θ) is the phase compensation matrix The element in the nth row and the lth column, f n =nB / N, n=1,…,N, N represents the number of sampling points, r l is the target translation component of the lth echo, φ l represents the total phase error between the echoes in the lth column, l = 1, ..., L, where L represents the number of echoes;
[0085] 3.5) For θ (t) To solve:
[0086] 3.5.1) Taking ρ r / N S The scene is discretized into I parts for the interval, where ρ r Indicates the distance resolution, N S Represents the discrete number of copies, so that r l ≈r l (i), r l (i) = iρ r / NS , i = -I / 2, ..., I / 2-1, and solve the estimated value of the translational component by traversal search
[0087]
[0088] 3.5.2) Estimate the value based on the translational component Get an estimate of the total phase error between echoes
[0089]
[0090] Among them, phase(·) represents the extraction phase operation;
[0091] 3.5.3) According to and Get the estimated translation parameters
[0092] 3.6) According to the obtained translation parameter estimate θ (t) Construct E(θ) and compensate the echo to obtain the tth echo signal S (t) :
[0093] S (t) =E(θ (t) )⊙S (t-1) ;
[0094] 3.7) According to steps 3.2), 3.5) and 3.6), the iterative steps of the TMC-EADMM iterative algorithm at the current iteration number t are obtained:
[0095] 3.7.1) Repeat the iterations in steps 3.2.1)-3.2.4) and The iteration stops when the number of iterations b = B, and the current updated ISAR image X( t ):
[0096]
[0097] 3.7.2) According to steps 3.5.1)-3.5.2) and The expression of the current updated estimated translation parameter is obtained
[0098] 3.8) According to the above steps 3.1)-3.7), the iterative formula at the current iteration time t is obtained:
[0099] 3.8.1) Update ISAR image X( t ):
[0100] for b=1:B
[0101]
[0102] After the loop is finished, the final output result As the updated ISAR image X (t) ,Right now
[0103] 3.8.2) Estimate the translation parameter θ (t) :
[0104]
[0105] s (t) =E(θ (t) )⊙s (t-1) .
[0106] Step 4: Construct a sparse ISAR translation compensation and high-resolution imaging integrated network based on a hypernetwork.
[0107] refer to Figure 2 , the implementation of this step includes the following:
[0108] 4.1) Take T iterations of the TMC-EADMM iterative algorithm and convert the iterative formula X contained in the tth iteration into (t) and θ (t) They correspond to the image reconstruction module and the translation compensation module respectively;
[0109] The image reconstruction module comprises G layers, each layer is composed of four cascade units: an image reconstruction unit, a denoising unit, a multiplier update unit and a relaxation unit, wherein:
[0110] The image reconstruction unit is used to (t-1) , the slack variable of the g-1th level and Update ISAR image
[0111]
[0112] in, is the distance dictionary, is the Doppler dictionary, represents the azimuth observation matrix, ρ is the iteration step size;
[0113] The multiplier update unit is used to update the parameters of the g-1th layer and and the ISAR image of the g-1th level Update Lagrange multipliers
[0114]
[0115] The denoising unit is used to calculate the Lagrange multiplier of the g-th level. and the ISAR image of the gth level Update auxiliary variables
[0116]
[0117] Among them, λ is the regularization coefficient, S λ / ρ (·) represents the soft threshold function with parameter λ / ρ;
[0118] The relaxation unit is used to adjust the parameters and Perform relaxation calculations to obtain relaxation variables and
[0119]
[0120] Among them, γ is the relaxation factor;
[0121] The translation compensation module is composed of two units, a translation parameter estimation unit and a translation phase compensation unit, connected in cascade, wherein:
[0122] The translation parameter estimation unit is used to estimate the value according to the translation component and the phase error estimate Get the estimated translation parameters after the current update in:
[0123]
[0124] In the formula, N is the number of sampling points;
[0125] The translation phase compensation unit is used to estimate the translation parameter θ according to the obtained translation parameter (t) Construct the phase compensation matrix E(θ (t) ), and compensate the echo S (t) :
[0126] S (t) =E(θ (t) )⊙S (t-1)
[0127] Here, ⊙ represents the Hadamard product.
[0128] 4.2) Connect the image reconstruction module and the translation compensation module in sequence to obtain the t-th layer sub-network, and then cascade multiple sub-networks to obtain a backbone network containing T-th layer sub-networks, such as Figure 3 As shown;
[0129] 4.3) Design a super network module based on residual network:
[0130] refer to Figure 4 , the implementation of this step includes the following:
[0131] 4.3.1) Construct a range-Doppler imaging unit and use the RD algorithm to calculate the echo S output by the t-th layer backbone network. (t-1) Perform imaging, and take the imaging result as a model to obtain the input image of the hypernetwork feature extraction unit
[0132]
[0133] in, is the distance dictionary, is the Doppler dictionary, represents the orientation observation matrix, |·| represents the modulo operation;
[0134] 4.3.2) Construct feature extraction unit:
[0135] First, set up 7 convolution modules a1, a2, a3, a4, a5, b1, b2, two residual modules r1, r2, and two rectified linear unit ReLU activation layers:
[0136] The first convolution module a1 is composed of a 3×3 standard convolution layer with an input dimension of 1 and an output dimension of 32, a batch normalization layer, a ReLU activation layer, and a maximum pooling layer maxpool cascaded;
[0137] The second convolutional module a2 consists of a 3×3 standard convolutional layer with an input dimension of 32 and an output dimension of 32;
[0138] The third convolution module a3 consists of a 3×3 standard convolutional layer with an input dimension of 32 and an output dimension of 32;
[0139] The fourth convolution module a4 is composed of a 3×3 standard convolution layer with an input dimension of 32 and an output dimension of 64;
[0140] The fifth convolution module a5 consists of a 3×3 standard convolutional layer with an input dimension of 64 and an output dimension of 64;
[0141] The sixth convolutional module b1 consists of a 1×1 standard convolutional layer with an input dimension of 32 and an output dimension of 32;
[0142] The seventh convolutional module b2 consists of a 1×1 standard convolutional layer with an input dimension of 32 and an output dimension of 64;
[0143] The first residual module r1 is cascaded by the second convolution module a2, the batch normalization layer, the ReLU activation layer and the third convolution module a3;
[0144] The second residual module r2 is cascaded by the fourth convolution module a4, the batch normalization layer, the ReLU activation layer and the fifth convolution module a5;
[0145] Next, the first residual module r1 is connected in parallel with the sixth convolution module b1 to obtain the first parallel multi-scale convolution module, and the second residual module r2 is connected in parallel with the seventh convolution module b2 to obtain the second parallel multi-scale convolution module; then the first convolution module a1, the first parallel multi-scale convolution module, the first rectified linear unit ReLU activation layer, the second parallel multi-scale convolution module, and the second rectified linear unit activation layer are sequentially cascaded to form a feature extraction unit;
[0146] The feature extraction unit is used to extract the features of the input image, and the feature information output by it is expressed as:
[0147]
[0148] Among them, i is the i-th convolution module a i The output of , relu(·) represents the ReLU activation function;
[0149] 4.3.3) Construct an optimal parameter extraction unit consisting of an average pooling layer avgpool, a fully connected layer W containing 64 nodes and a cascade of sigmoid activation functions. Its input is the feature O5 output by the feature extraction unit, and the output is a parameter vector of length 3G. This parameter vector is the internal parameter of the t-th layer subnetwork in the backbone network.
[0150]
[0151] Among them, the parameter vector Including the regularization coefficient λ, penalty factor ρ and relaxation factor γ in all levels of the image reconstruction module in the t-th backbone network, that is, G represents the number of each level of the image reconstruction module, avgpool(·) represents the average pooling function, and sigmoid(·) represents the sigmoid activation function;
[0152] 4.3.4) Cascading the range-Doppler imaging unit, the feature extraction unit and the optimal parameter extraction unit in sequence to obtain a super network module;
[0153] 4.4) Combine the backbone network and the corresponding supernetworks of each layer to obtain the t-th layer sub-combination network:
[0154] 4.4.1) Pass the output of the t-1th layer backbone network through the tth layer super network to obtain the optimal parameters of the tth layer backbone network;
[0155] 4.4.2) These optimal parameters and the output of the t-1th layer backbone network are used as the input of the tth layer backbone network, and finally the output S of the tth layer backbone network is obtained. (t) ;
[0156] 4.5) Cascade multiple sub-combination networks to obtain a super-network-based sparse ISAR motion compensation and imaging integration network.
[0157] Step 5: Generate training set.
[0158] 5.1) The scattering points whose positions obey random distribution, whose amplitudes obey normal distribution and whose phases obey uniform distribution in [0,2π] are taken as image data X;
[0159] 5.2) Substitute the image X into the sparse observation model S = F (θ) ⊙ (D r XD a Φ a )+Z obtains the echo data with defective noise and translation error, and uses it as the network input data S, wherein the translation error includes the echo envelope offset error and the echo random phase error, the echo envelope offset error obeys the uniform distribution of [-20m, 20m], and the echo random phase error obeys the Gaussian distribution with parameters of (0, π / 2);
[0160] 5.3) Construct the number of samples to be N b Echo data As a training set;
[0161] In this embodiment, the training set image size is set to but not limited to 128×256, the echo defect rate is set to 50%, the echo signal-to-noise ratio range is set to but not limited to 5-15dB, and the number of training set samples N is b Assume but not limited to 100.
[0162] Step 6: Train the sparse ISAR translation compensation and high-resolution imaging integrated network.
[0163] 6.1) Define the loss function of the integrated network:
[0164] loss=loss true +βloss constraint
[0165] Among them, β is the loss weight coefficient, losstrue is the cumulative fidelity term, loss constraint is a sparse constraint term, and its expression is as follows:
[0166]
[0167] In the formula, T represents the total number of network layers, is the ISAR image reconstructed by the tth layer of the network, S( t ) is the network layer t echo, is the distance dictionary, is the Doppler dictionary, represents the orientation observation matrix;
[0168] 6.2) Let batch be the single batch processing size, lr be the learning rate during integrated network training, and Θ0 be the network parameters during the integrated network training phase: in, represents the parameters of the convolution module with a convolution kernel size of 3×3 in the hypernetwork, n c Indicates the serial number of the convolution module, represents the parameters of the convolution module with a convolution kernel size of 1×1 in the hypernetwork, n d represents the serial number of the convolutional module, and W represents the parameters of the fully connected layer in the hypernetwork;
[0169] 6.3) Each time, select a batch of data from the training set and input it into the integrated network to calculate the loss value in each batch
[0170]
[0171] Among them, N batch is the number of samples in each batch, S represents the input echo data;
[0172] 6.4) Apply the back-propagation algorithm to calculate the gradient of the loss function with respect to any parameter in the integrated network According to the gradient of the solution Update the network parameters of the integrated network to obtain the network parameters Θ′ of the current training phase:
[0173]
[0174] 6.5) Repeat steps 6.3)-6.4) until the loss function converges to obtain a trained translation compensation and ISAR imaging integrated network.
[0175] In this embodiment, during the training process, the batch size of a single batch processing is set but not limited to 8, the learning rate lr=0.005, and the Adam optimizer is used to update the network parameters.
[0176] Step 7: Obtain ISAR target imaging results.
[0177] Input the measured air-space target echo data into the trained integrated network, and perform the following operations for different network layer numbers t:
[0178] When the number of network layers is t=1, the air-space target echo data is first input into the first layer of the super network to obtain the optimal parameters of the first layer of the backbone network; then these optimal parameters and the initial echo are used as the input of the first layer of the backbone network to obtain the output echo S of the first layer of the backbone network. (1) ;
[0179] When the number of network layers is 1<t<T, the output of the t-1th layer backbone network is first passed through the tth layer super network to obtain the optimal parameters of the tth layer backbone network; then these optimal parameters and the output of the t-1th layer backbone network are used as the input of the tth layer backbone network to obtain the output echo S of the tth layer backbone network (t) ;
[0180] When the number of network layers t = T, the output of the T-1th layer backbone network is first passed through the Tth layer super network to obtain the optimal parameters of the Tth layer backbone network; then these optimal parameters and the output of the T-1th layer backbone network are used as the input of the Tth layer backbone network to obtain the final output echo S( T ) and imaging results X( T ), at the same time, according to the output echo S( T ) to get the envelope alignment result
[0181] In this embodiment, the test is performed using the echo data of the actual Yak-42 aircraft.
[0182] The effect of the present invention can be further illustrated by the following simulation experiment:
[0183] 1. Simulation Experiment Conditions
[0184] The software platform of the simulation experiment of the present invention is Windows 10 operating system and Pytorch 3.9, and the hardware configuration is: Core i9-13900K CPU and NVIDIA GeForce RTX 4090 GPU.
[0185] The simulation experiment of the present invention uses point simulation data with random positions, normal amplitudes and uniform phase distribution of [0, 2π], and the number of images is 100.
[0186] The comparison method of the simulation experiment is to combine the envelope alignment method MERA based on minimum entropy and the sparse ISAR self-focusing imaging method AF-ADN based on CV-ADMMN to obtain the MERA+AF-ADN method to perform translation compensation and imaging on echoes containing translation errors.
[0187] 2. Simulation content and results analysis
[0188] Simulation 1: Under the above simulation conditions, the present invention and the existing MERA+AF-ADN method are used to perform translation compensation on the measured ISAR echo data containing translational components under the conditions of a defect rate of 50% and a signal-to-noise ratio of 5dB and 10dB, respectively, to obtain the corresponding envelope alignment results, as shown in FIG. Figure 5 As shown, where:
[0189] Figure 5 (a) is the envelope alignment result obtained by the prior art for the measured Yak-42 aircraft data when the echo defect rate is 50% and the echo signal-to-noise ratio is 10dB.
[0190] Figure 5 (b) is the envelope alignment result obtained by the prior art for the measured Yak-42 aircraft data when the echo defect rate is 50% and the echo signal-to-noise ratio is 5dB.
[0191] Figure 5 (c) is the envelope alignment result obtained by the present invention for the measured Yak-42 aircraft data when the echo defect rate is 50% and the echo signal-to-noise ratio is 10dB.
[0192] Figure 5 (d) is the envelope alignment result obtained by the present invention for the measured Yak-42 aircraft data when the echo defect rate is 50% and the echo signal-to-noise ratio is 5dB.
[0193] Depend on Figure 5 It can be seen that the envelope alignment result obtained by the comparison method will have envelope jumps, which will affect the subsequent imaging results. Compared with the prior art, the present invention realizes envelope alignment in different signal-to-noise ratio echo scenarios without jumping.
[0194] Simulation 2: Under the above simulation conditions, the measured ISAR echo data containing translational components are imaged using the present invention and the existing MERA+AF-ADN method under the conditions of a defect rate of 50% and a signal-to-noise ratio of 5dB and 10dB, respectively, to obtain the corresponding ISAR imaging results, as shown in FIG. Figure 6 As shown, where:
[0195] Figure 6(a) is the imaging result obtained by the prior art on the measured Yak-42 aircraft data when the echo defect rate is 50% and the echo signal-to-noise ratio is 10dB.
[0196] Figure 6 (b) is the imaging result obtained by the prior art on the measured Yak-42 aircraft data when the echo defect rate is 50% and the echo signal-to-noise ratio is 5dB.
[0197] Figure 6 (c) is the imaging result obtained by the present invention from the measured Yak-42 aircraft data when the echo defect rate is 50% and the echo signal-to-noise ratio is 10dB.
[0198] Figure 6 (d) is the imaging result obtained by the present invention from the measured Yak-42 aircraft data when the echo defect rate is 50% and the echo signal-to-noise ratio is 5dB.
[0199] Depend on Figure 6 It can be seen that the ISAR image obtained by the comparison method is seriously defocused in azimuth, while the present invention obtains ISAR images with good focusing effect in different signal-to-noise ratio echo scenes, with complete target contours and clean background.
[0200] Yes Figure 6 The three imaging evaluation indicators of reconstruction mean square error, image entropy and running time are calculated for the imaging results in Table 1.
[0201] Table 1 Measured data algorithm performance indicators
[0202]
[0203] The evaluation indicators reconstruction mean square error and image entropy in Table 1 are specifically defined as follows:
[0204] The reconstruction mean square error is a method to measure the reconstruction effect of ISAR image after translation compensation. The smaller the value, the better the image quality. It is defined as:
[0205]
[0206] Where N represents the number of samples, |·| represents the modulo operation, is the imaging result obtained by the network, X n is the label image, M is the number of complete echoes, and L is the number of echoes after defects.
[0207] The image entropy is a method to measure the average amount of information in an image. The smaller the value, the better the image quality. It is defined as:
[0208]
[0209] in, Imaging results obtained for the network The element at row u and column v in Represents the total energy of the image.
[0210] It can be seen from Table 1 that compared with the existing technology, after completing the training, the translation compensation and imaging integrated network obtained by the present invention has smaller reconstruction mean square error and image entropy under different echo signal-to-noise ratios and shorter running time, indicating that the imaging quality of the present invention is better.
[0211] In summary, the envelope alignment, imaging results and imaging evaluation indicators of the present invention under different echo signal-to-noise ratios are better than those of the existing solutions, which verifies that the present invention can improve the accuracy of network translation compensation and the quality of ISAR imaging under low signal-to-noise ratio conditions, while improving the robustness to the echo signal-to-noise ratio and reducing the advantages of time and space complexity.
[0212] It should be noted that the serial numbers of the above steps are for more clearly describing the implementation scheme of the present invention, and the order of the serial numbers is not limited.
Claims
1. A hypernetwork-based sparse ISAR translation compensation and imaging integration method, characterized in that: The steps include: (1) Based on the translation matrix F(θ) that produces translation error for the echo and the sparse measurement matrix Φ a , by performing a range Fourier transform D on the ISAR image r and the azimuthal Fourier transform D a , establish the sparse observation model S under echo defect: S=F(θ)⊙(D r XD a Φ a )+Z Where θ represents the translation parameter, X represents the ISAR image to be solved, Z represents the noise, and ⊙ represents the Hadamard product; (2) According to the sparse observation model S under echo defect conditions, the sparse ISAR translation compensation and imaging optimization objective function J1(X,θ) based on L1 norm optimization is constructed: Where, E(θ(=(F(θ)) * , (·) * represents the conjugate operation, ||·|| F and ||·||1 represent the matrix F norm and matrix 1 norm respectively, and λ is the weight coefficient; (3) According to the TMC-EADMM algorithm, the optimization objective function J1(X,θ) is iteratively solved to obtain the iterative formula at the current iteration time t, that is, to update the ISAR image X (t) and estimate the translation parameter θ (t) : Among them, S (t) Indicates that after E(θ (t) ) The tth echo after compensation, S (t) =E(θ (t) )⊙S (t-1) , X (t) represents the t-th reconstructed ISAR image, θ (t) represents the t-th estimated translation parameter, D r and D a Respectively represent the distance dictionary and the direction dictionary, Φ a represents a sparse observation matrix; (4) According to the iterative formula X (t) and θ (t) Constructing a sparse ISAR translation compensation and imaging integrated network based on a hypernetwork: 4a) Take T iterations in the iterative algorithm and convert the iterative formula X contained in the tth iteration into (t) and θ (t) The corresponding modules are image reconstruction module and translation compensation module; 4b) connecting the image reconstruction module and the translation compensation module in sequence to obtain the t-th layer sub-network, and then cascading multiple sub-networks to obtain a backbone network including T-layer sub-networks; 4c) Design a hypernetwork based on residual network, extract the features of ISAR images with different signal-to-noise ratios through the hypernetwork, and generate the optimal parameters such as regularization coefficient λ, penalty factor ρ and relaxation factor γ for each layer of the backbone network according to these features; 4d) combining the backbone network and the corresponding super networks of each layer to form an integrated network of translation compensation and ISAR imaging; (5) Randomly generate N b The amplitudes of the scattering points X follow a normal distribution and the phases follow a uniform distribution. Different ISAR scenarios are obtained by setting different signal-to-noise ratios, thereby generating N b Echo data in different scenarios N b ≥100, use it as training set; (6) inputting the training set into the translation compensation and ISAR imaging integrated network, training it using the gradient descent method, and obtaining a trained translation compensation and ISAR imaging integrated network; (7) The measured Yak-42 data is input into the trained translation compensation and imaging integrated network, and the final ISAR imaging results are obtained through network forward propagation.
2. The method according to claim 1, characterized in that In step (1), a sparse observation model S is established under echo defect conditions, which is implemented as follows: 1a) Assume that the radar transmits a linear frequency modulated pulse signal and the target is represented by H scattering points. The de-linear frequency modulated echo signal s received by the radar is obtained r (f r ,m): Among them, σ h is the backscattering coefficient of the hth scattering point, rect(·) is the rectangular window function, is the distance frequency, R h (t m ) is the distance from the hth scattering point in the mth echo to the radar, m = 1, 2, ... M, M represents the total number of pulses, B is the bandwidth of the transmitted signal, f c is the center frequency, r m =r R (t m )-R ref (t m ) is the target translation component, R ref (t m ) is the reference distance that changes with slow time, (x h ,y h ) is the starting coordinate of the scattering point h, is the random phase error between echoes caused by the system error, is Gaussian white noise; 1b) According to the echo signal s received by the radar r (f r ,m) construct the translation matrix F(θ): in, is the translation matrix The element in the nth row and the lth column, θ={r1,…,r L ,φ1,…,φ L } represents the translation parameter, l=1,...,L, L represents the number of echoes, 0<L<M, f n =nB / N, n=1,...,N, N represents the number of sampling points, r l represents the envelope offset of the lth column echo, represents the total phase error between the echoes in the lth column; 1c) Let the size of the ISAR image be U×V, and the element in the image be i h (u,v), u∈[1,U], v∈[1,V]; 1d) Perform Fourier transform on the ISAR image in range and azimuth, and add the translation error F n,l (θ), and the echo signal s containing translation error is obtained r (n,l): Where d is the azimuth defect vector, which means that in the case of sparse observation, the original vector of length Q is randomly downsampled to obtain a vector of length L. l is the lth element in d; 1e) s r (n,l) as a matrix The elements of the nth row and the lth column in , construct the sparse observation model S with translation error in the case of echo defect: S=F(θ)⊙(D r XD a Φ a )+Z in, is a distance dictionary, whose element in row n and column u is is the Doppler dictionary, whose element in row v and column m is represents the orientation observation matrix, which is obtained by removing I M×M The specific ML column in I M×M represents the identity matrix, ⊙ represents the Hadamard product; represents the ideal two-dimensional image corresponding to the target, For noise.
3. The method according to claim 1, characterized in that In step (3), the optimization objective function J1(X,θ) is iteratively solved according to the TMC-EADMM algorithm, which is implemented as follows: 3a) Fix the phase compensation term E(θ) in J1(X,θ) and obtain the ISAR image optimization objective function X (t) : in, is the distance dictionary, is the Doppler dictionary, represents the orientation measurement matrix, ⊙ represents the Hadamard product, represents the ideal two-dimensional image corresponding to the target, θ (t-1) represents the translation parameter estimated at the t-1th time, S (t-1) Indicates that after E(θ (t-1) ) is the t-th echo after compensation, λ is the regularization coefficient; 3b) Use the EADMM algorithm to (t) Perform an iterative solution: 3b1) According to the input echo signal S (t-1) , the slack variable for the b-1th iteration and Update ISAR image 3b2) According to the parameters of the b-1th iteration and the ISAR image of the bth iteration Update Lagrange multipliers 3b3) According to the Lagrange multiplier of the bth iteration ISAR images of Update auxiliary variables 3b4) For parameters and Perform relaxation calculations to obtain relaxation variables and Among them, ρ is the initialization parameter iteration step, γ is the relaxation factor, S λ / ρ (·) represents the soft threshold function with parameter λ / ρ; 3c) The ISAR image X obtained at the tth iteration (t) Fixed, extract the terms related to the translation in J1(X,θ), and obtain the sparse ISAR translation parameter optimization objective function θ (t) : Among them, θ (t) ={r1,…,r L ,φ1,…,φ L },E( θ )=(F(θ)) * , F(θ) is the translation matrix, (·) * represents the conjugation operation; 3d) The objective function θ (t) In Expand the term and remove θ (t) The terms in the expression that are independent of θ give θ (t) The following simplified formula: Among them, Re{·} represents the real part operation, f n =nB / N, n=1,...,N, N represents the number of sampling points, r l is the target translation component of the lth echo, φ l represents the total phase error between the echoes in the lth column, l = 1, ..., L, where L represents the number of echoes; 3e) For θ (t) To solve: 3e1) with ρ r / N S The scene is discretized into I parts for the interval, where ρ r Indicates the distance resolution, N S Represents the discrete number of copies, so that r l ≈r l (i), r l (i) = iρ r / N S , i = -I / 2, ..., I / 2-1, and solve the estimated value of the translational component by traversal search 3e2) Estimate the value based on the translational component Get an estimate of the total phase error between echoes Among them, phase(·) represents the extraction phase operation; 3e3) According to and Get the estimated translation parameters 3f) According to the obtained translation parameter estimate θ (t) Construct E(θ) and compensate the echo to obtain the tth echo signal S (t) : 3g) According to 3b), 3e) and 3f), the iterative formula at the current iteration number t is obtained: 3g1) Repeat the iteration of the EADMM algorithm until the iteration number b=B, and then stop the iteration to obtain the current updated ISAR image: Among them, the ISAR image of the bth iteration in the EADMM algorithm is 3g2) According to and The expression of the current updated estimated translation parameter is obtained 4. The method according to claim 1, characterized in that: The image reconstruction module and the translation compensation module in step 4a) have the following structures: The image reconstruction module comprises G layers, each layer is composed of four cascade units: an image reconstruction unit, a denoising unit, a multiplier update unit and a relaxation unit, wherein: The image reconstruction unit is used to (t-1) , the slack variable of the g-1th level and Update ISAR image in, is the distance dictionary, is the Doppler dictionary, represents the azimuth observation matrix, ρ is the iteration step size; The multiplier update unit is used to update the parameters of the g-1th layer and and the ISAR image of the g-1th level Update Lagrange multipliers The denoising unit is used to calculate the Lagrange multiplier of the g-th level. and the ISAR image of the gth level Update auxiliary variables Among them, λ is the regularization coefficient, S λ / ρ (·) represents the soft threshold function with parameter λ / ρ; The relaxation unit is used to adjust the parameters and Perform relaxation calculations to obtain relaxation variables and Among them, γ is the relaxation factor; The translation compensation module is composed of two units, a translation parameter estimation unit and an echo compensation unit, connected in cascade, wherein: The translation parameter estimation unit is used to estimate the value according to the translation component and the phase error estimate Get the estimated translation parameters after the current update in: In the formula, f n =nB / N, n=1,...,N, N represents the number of sampling points; The translation compensation unit is used to estimate the translation parameter θ according to the obtained translation parameter (t) Construct the phase compensation matrix E(θ (t) ), and compensate the echo S (t) : S (t) =E(θ (t) )⊙S (t-1) Here, ⊙ represents the Hadamard product.
5. The method according to claim 1, characterized in that In step 4c), a super network based on a residual network is designed and implemented as follows: 4c1) Construct a range-Doppler imaging unit and use the RD algorithm to calculate the echo S output by the t-th layer network. (t-1) Perform imaging, and take the imaging result as a model to obtain the input image of the hypernetwork feature extraction unit in, is the distance dictionary, is the Doppler dictionary, represents the orientation observation matrix, |·| represents the modulo operation; 4c2) Construct a feature extraction unit consisting of a convolution layer a1, a pooling layer, a parallel residual block r1, a convolution layer b1, a first rectified linear unit (ReLU) activation layer, a parallel residual block r2, a convolution layer b2, and a second rectified linear unit activation layer cascaded, where: The first convolution module a1 is composed of a 3×3 standard convolution layer with an input dimension of 1 and an output dimension of 32, a batch normalization layer, a ReLU activation layer, and a maximum pooling layer maxpool cascaded; The second convolutional module a6 consists of a 1×1 standard convolutional layer with an input dimension of 32 and an output dimension of 32; The third convolutional module a7 consists of a 1×1 standard convolutional layer with an input dimension of 32 and an output dimension of 64; The first residual block r1 is composed of a 3×3 standard convolutional layer a2 with an input dimension of 32 and an output dimension of 32, a batch normalization layer, a ReLU activation layer, and a 3×3 standard convolutional layer a3 with an input dimension of 32 and an output dimension of 32. The second residual block r2 is composed of a 3×3 standard convolution layer a4 with an input dimension of 32 and an output dimension of 32, a batch normalization layer, a ReLU activation layer, and a 3×3 standard convolution layer a5 with an input dimension of 32 and an output dimension of 32. The feature extraction unit is used to extract the features of the input image. The feature information output by the feature extraction unit is expressed as: Among them, i is the i-th convolution module a i The output of , relu(·) represents the ReLU activation function.
6. The method according to claim 1, characterized in that In step 4c), the regularization coefficient λ, penalty factor ρ and relaxation factor γ are generated for each layer of the backbone network according to the features of the ISAR images with different signal-to-noise ratios, and the optimal parameters are achieved as follows: Construct an optimal parameter extraction unit consisting of an average pooling layer avgpool, a fully connected layer W containing 64 nodes and a cascade of sigmoid activation functions. Its input is the feature O5 output by the feature extraction unit, and the output parameter vector with a length of 3G is used as the internal parameter of the t-th layer sub-network in the backbone network. Among them, G represents the number of each level of the image reconstruction module, and the parameter vector Including the regularization coefficient λ, penalty factor ρ and relaxation factor γ in all levels of the image reconstruction module in the t-th backbone network, that is, avgpool(·) represents the average pooling function, and sigmoid(·) represents the sigmoid activation function.
7. The method according to claim 1, characterized in that In step 4d), the backbone network and the corresponding super networks of each layer are combined to achieve the following: First, the output of the t-1th layer backbone network passes through the tth layer super network to obtain the optimal parameters of the tth layer backbone network; then these optimal parameters and the output of the t-1th layer backbone network are used as the input of the tth layer backbone network, and finally the output S of the tth layer backbone network is obtained. (t) .
8. The method according to claim 1, characterized in that Step (6) inputs the training set into the translation compensation and ISAR imaging integrated network and trains it using the gradient descent method, which is implemented as follows: 6a) Construct the loss function of sparse ISAR translation compensation and imaging integration: loss=loss true +βloss constraint Among them, β is the loss weight coefficient, loss true is the cumulative fidelity term, loss constraint is a sparse constraint term, and its expression is as follows: In the formula, T represents the total number of network layers, is the ISAR image reconstructed by the tth layer of the network, S (t) is the network layer t echo, is the distance dictionary, is the Doppler dictionary, represents the orientation observation matrix; 6b) Each time, a batch of data is selected from the training set and input into the translation compensation and ISAR imaging integrated network to calculate the loss value in each batch. Among them, N batch is the number of samples in each batch, S represents the input echo data; 6c) Apply the back propagation algorithm to calculate the gradient of the loss function with respect to any parameter in the sparse ISAR high-resolution imaging network According to the gradient of the solution Update the network parameters of the translation compensation and ISAR imaging integrated network to obtain the network parameters Θ′ of the current training stage: Among them, Θ0 is the network parameter of the sparse ISAR high-resolution imaging network training phase, and lr represents the learning rate during training; 6d) Repeat steps 6b)-6c) until the loss function converges, and obtain the final trained network parameters, thereby obtaining a trained translation compensation and ISAR imaging integrated network.
9. The method according to claim 1, characterized in that: In step (7), the measured Yak-42 data is input into the trained translation compensation and imaging integrated network for network forward propagation, which is achieved as follows: When the number of network layers is t=1, the initial echo is first input into the first layer of the super network to obtain the optimal parameters of the first layer of the backbone network. Then these optimal parameters and the initial echo are used as the input of the first layer of the backbone network to obtain the output echo S of the first layer of the backbone network. (1) ; When the number of network layers is 1<t<T, the output of the t-1th layer backbone network is first passed through the tth layer super network to obtain the optimal parameters of the tth layer backbone network, and then these optimal parameters and the output of the t-1th layer backbone network are used as the input of the tth layer backbone network to obtain the output echo S of the tth layer backbone network (t) ; When the number of network layers t = T, the output of the T-1th layer backbone network is first passed through the Tth layer super network to obtain the optimal parameters of the Tth layer backbone network, and then these optimal parameters and the output of the T-1th layer backbone network are used as the input of the Tth layer backbone network to obtain the final imaging result X (T) .
Citation Information
Patent Citations
Method for inverse synthetic aperture radar imaging based on deep ADMM network
CN109683161A
Moving target three-dimensional imaging method based on vector modeling optimization inversion
CN113985412A
Bistatic ISAR sparse aperture maneuvering target imaging method and system
CN114910906A
Sparse aperture ISAR translation compensation method based on L1 norm optimization
CN115128604A
Sparse ISAR high-resolution imaging method based on depth expansion
CN117192548A
Cited By
Maneuvering target sparse ISAR imaging network construction method based on Unfolding technology
CN120294710A
ISAR coherent translation compensation and imaging method based on motion parameter estimation network
CN120847799A
