SAR depth super-resolution imaging method and related device based on autoencoder

Abstract

The present application discloses a SAR deep super-resolution imaging method and related devices based on an autoencoder, and relates to the field of remote sensing imaging technology. The method comprises obtaining an initial parameter set; the initial parameter set comprises a plurality of radar operating parameter groups and a plurality of real SAR images; each radar operating parameter group is respectively combined with each real SAR image to construct a plurality of simulation scenes; all simulation scenes constitute a simulation scene data set; the observation scene is divided based on the radar operating parameter group to obtain a first scene; an initialized imaging result is obtained based on the first scene; an autoencoder is pre-trained based on the simulation scene data set to obtain an initialized network parameter; according to the initialized network parameter and the initialized imaging result, an imaging loss function is cyclically iterated to be minimized to obtain a final imaging result. The present application utilizes the autoencoder to fully utilize the redundancy of the backscattering coefficient in the SAR imaging process, and improves the sampling efficiency.

Description

Technical Field

[0001] The present application relates to the field of remote sensing imaging technology, and in particular to a SAR depth super-resolution imaging method based on an autoencoder and related devices. Background Art

[0002] The application of Compressive Sensing (CS) in Synthetic Aperture Radar (SAR) systems helps achieve super-resolution imaging. CS uses sparse prior information to accurately reconstruct signals from a small number of measured signals. In SAR imaging based on compressed sensing theory, sparse information is used to restore the imaging results of SAR echoes in the case of insufficient sampling. However, the sampling efficiency of the current SAR imaging algorithm based on compressed sensing theory is low. Summary of the invention

[0003] The purpose of this application is to provide a SAR depth super-resolution imaging method and related devices based on an autoencoder, which can improve sampling efficiency.

[0004] To achieve the above objectives, this application provides the following solutions:

[0005] In a first aspect, the present application provides a SAR depth super-resolution imaging method based on an autoencoder, comprising:

[0006] Acquire an initial parameter set; the initial parameter set includes several radar operating parameter groups and several real SAR images; the radar operating parameter group includes a scene center reference slant range, a satellite flight speed, an azimuth antenna size, a number of transmitted pulses, a modulation frequency of a transmitted signal, a time vector of a radar transmitting pulse, and an operating wavelength;

[0007] Combining each of the radar operating parameter groups with each of the real SAR images to construct a number of simulation scenarios; all simulation scenarios constitute a simulation scenario data set;

[0008] Divide the observation scene based on the radar working parameter group to obtain a first scene; obtain an initialization imaging result based on the first scene;

[0009] Pre-training the autoencoder based on the simulation scene data set to obtain initialization network parameters;

[0010] According to the initialized network parameters and the initialized imaging result, the imaging loss function is minimized through cyclic iteration to obtain the final imaging result.

[0011] In a second aspect, the present application provides a SAR depth super-resolution imaging device based on an autoencoder, comprising:

[0012] The initial parameter set acquisition module is used to: acquire an initial parameter set; the initial parameter set includes a plurality of radar operating parameter groups and a plurality of real SAR images; the radar operating parameter group includes a scene center reference slant range, a satellite flight speed, an azimuth antenna size, a number of transmitted pulses, a modulation frequency of a transmitted signal, a time vector of a radar transmitting pulse, and an operating wavelength;

[0013] A simulation scene data set construction module is used to: combine each radar working parameter group with each real SAR image to construct a number of simulation scenes; all simulation scenes constitute a simulation scene data set;

[0014] An initialization imaging result generating module is used to: divide the observation scene based on the radar working parameter group to obtain a first scene; and obtain an initialization imaging result based on the first scene;

[0015] An autoencoder pre-training module is used to: pre-train the autoencoder based on the simulation scene data set to obtain initialization network parameters;

[0016] The final imaging result generating module is used to: according to the initialization network parameters and the initialization imaging result, cyclically iteratively minimize the imaging loss function to obtain the final imaging result.

[0017] In a third aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned autoencoder-based SAR depth super-resolution imaging method.

[0018] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned autoencoder-based SAR depth super-resolution imaging method.

[0019] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0020] The present application provides a SAR depth super-resolution imaging method and related devices based on an autoencoder. Existing SAR super-resolution imaging based on compressed sensing theory focuses on forcibly converting a two-dimensional backscattering coefficient matrix into a vector, ignoring the redundancy between rows and columns, thereby destroying the internal structure and failing to fully utilize the redundancy of the backscattering coefficient matrix. In response to the above problems, the present application uses an autoencoder to fully utilize the redundancy of the backscattering coefficients in the SAR imaging process and improves the sampling efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 This is an application environment diagram of a SAR depth super-resolution imaging method based on an autoencoder in an embodiment of the present application;

[0023] Figure 2 A flowchart of a SAR depth super-resolution imaging method based on an autoencoder provided in one embodiment of the present application;

[0024] Figure 3 A schematic diagram of the principle of SAR depth super-resolution imaging based on an autoencoder provided in an embodiment of the present application;

[0025] Figure 4 A schematic diagram of functional modules of a SAR depth super-resolution imaging device based on an autoencoder provided in one embodiment of the present application;

[0026] Figure 5 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0028] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0029] The SAR depth super-resolution imaging method based on the autoencoder provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, integrated on the server 104, or placed on the cloud or other servers. The terminal 102 can send the initial parameter set to the server 104. After the server 104 receives the initial parameter set, for the initial parameter set, the server 104 combines each radar working parameter group with each real SAR image respectively to construct several simulation scenes; all simulation scenes constitute a simulation scene data set; the observation scene is divided based on the radar working parameter group to obtain the first scene; the initialization imaging result is obtained based on the first scene; the autoencoder is pre-trained based on the simulation scene data set to obtain the initialization network parameters; according to the initialization network parameters and the initialization imaging results, the imaging loss function is minimized cyclically and iteratively to obtain the final imaging result. The server 104 can feedback the final imaging result obtained to the terminal 102. In addition, in some embodiments, the SAR depth super-resolution imaging method based on the autoencoder can also be implemented independently by the server 104 or the terminal 102. For example, the terminal 102 can directly perform SAR depth super-resolution imaging processing on the initial parameter set, or the server 104 can obtain the initial parameter set from the data storage system and perform SAR depth super-resolution imaging processing on the initial parameter set.

[0030] The terminal 102 may be, but is not limited to, various desktop computers and laptop computers. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers, or may be a cloud server.

[0031] In an exemplary embodiment, Figure 2 As shown, a SAR depth super-resolution imaging method based on an autoencoder is provided. The method is executed by a computer device, and specifically can be executed by a computer device such as a terminal or a server alone, or can be executed by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used as an example to illustrate, including the following steps 201 to 205. Among them:

[0032] Step 201: Acquire an initial parameter set; the initial parameter set includes several radar operating parameter groups and several real SAR images; the radar operating parameter group includes a scene center reference slant range, satellite flight speed, azimuth antenna size, number of transmitted pulses, modulation frequency of the transmitted signal, time vector of the radar transmitting pulse, and operating wavelength.

[0033] Step 202: Combine each radar operating parameter group with each real SAR image to construct a plurality of simulation scenarios; all simulation scenarios constitute a simulation scenario data set.

[0034] Step 203: Divide the observation scene based on the radar operating parameter group to obtain a first scene; and obtain an initialization imaging result based on the first scene.

[0035] Step 204: pre-training the autoencoder based on the simulation scenario data set to obtain initialization network parameters.

[0036] Step 205: According to the initialized network parameters and the initialized imaging result, the imaging loss function is minimized cyclically and iteratively to obtain the final imaging result.

[0037] Implement the above steps 201 to 205, use the autoencoder structure to extract the latent backscatter coefficients, and use its internal redundancy. In the SAR super-resolution imaging problem, the autoencoder is used to fully utilize the redundancy of the backscatter coefficients in the SAR imaging process. In addition, since Convolutional Neural Networks (CNN) is always superior to other deep learning structures in image processing, the encoder and decoder are modeled as multi-layer CNNs, and a SAR deep super-resolution imaging network based on the autoencoder is constructed.

[0038] In step 201, the initial parameter set includes P groups of radar operating parameters and Q real SAR images.

[0039] Radar operating parameters are f1 = {R0, V e ,L a ,N a ,λ,Kr,ta}, where R0 is the reference slant distance from the center of the scene, V e is the satellite flight speed, L a is the antenna size in azimuth, N a is the number of transmitted pulses, λ is the operating wavelength, Kr is the modulation frequency of the transmitted signal, ta=[η1,η2,...,η m ,...η Na ] T is the time vector of the radar transmitting pulse, where η m represents the non-uniform sampling time when the SAR transmits the mth pulse, [·] T represents the transpose operation. Q real SAR images constitute the real SAR image set σ q represents the qth real SAR image.

[0040] In another exemplary embodiment of the present application, the above step 202 may be replaced by the following steps 301 to 303:

[0041] Step 301: combining each radar operating parameter group and each real SAR image to obtain a plurality of parameter image combinations; for each parameter image combination, constructing an observation matrix corresponding to the radar operating parameter group according to the radar operating parameter group and the slant range of all scene points in the delineated scene at each sampling time;

[0042] Step 302: Calculate the observation equation corresponding to the parameter image combination according to the observation matrix corresponding to the radar operating parameter group and the real SAR image in the parameter image combination;

[0043] Step 303: Based on the matched filtering principle, according to the observation equation corresponding to the parameter image combination and the observation matrix, obtain the simulation scene corresponding to the parameter image combination; for each parameter image combination, repeat the above steps 301 to 303 to obtain the simulation scene corresponding to each parameter image combination.

[0044] The observation matrix D corresponding to the p-th radar working parameter group is constructed according to the p-th radar working parameter group in step 201 and the slant distances of all scene points in the delineated scene at each sampling time. p , the observation matrix is ​​specifically expressed as:

[0045]

[0046] in, Denotes the crossing time as η m , slope distance is R n (η m ) is the reconstruction element of the target point, and ε(·) represents the step function.

[0047] According to the observation matrix D p and the real SAR image, and obtain the observation equation S m =D p σ q ; Among them, S m is the observation equation corresponding to the mth parameter image combination.

[0048] According to the principle of matched filtering, the simulation scene is obtained is the simulation scene corresponding to the mth parameter image combination.

[0049] According to the combination of different radar working parameters and real SAR images, steps 301 to 303 are repeated to obtain M = PQ simulation scenes, and M simulation scenes constitute a simulation scene data set.

[0050] In step 203, an initialization imaging result is obtained based on the first scene, which specifically includes the following steps 401 to 406:

[0051] Step 401: obtaining the original echo signal matrix of the echo pixel points of the SAR;

[0052] Step 402: obtaining a range-compressed time-domain echo signal based on the original echo signal matrix;

[0053] Step 403: Acquire scene point coordinates of the first scene and satellite sampling position coordinates, and obtain slant distances of all scene points in the first scene at each sampling time according to the scene point coordinates of the first scene and the satellite sampling position coordinates;

[0054] Step 404: determining a time delay based on the slant range of all scene points at each sampling time, and obtaining echo time delay signals of all scene points in the first scene according to the time delay and the distance compressed time domain echo signal at each azimuth sampling time;

[0055] Step 405: superimpose the echo delay signal obtained at each sampling moment with the azimuth compensation phase at the corresponding sampling moment to obtain an azimuth compression matrix;

[0056] Step 406: Re-divide the azimuth compression matrix to obtain an initialized imaging result.

[0057] The echo pixel points of low-oversampled SAR are a two-dimensional matrix. Based on this, the original echo signal matrix is ​​expressed as:

[0058]

[0059] Among them, S echo,m,k Indicates that the azimuth sampling time is η m , the distance sampling time is τ k The echo, that is, the echo matrix S echo The element in the mth row and kth column.

[0060] The specific process of obtaining the time domain echo signal after range compression in step 402 is as follows:

[0061] First, the original echo signal matrix is ​​transformed into a distance-oriented Fourier transform. Secondly, the echo signal obtained after the distance-oriented Fourier transform is subjected to distance compression based on matched filtering. Finally, the echo signal after distance compression is subjected to a distance-oriented inverse Fourier transform, and finally the time-domain echo signal after distance compression is obtained. Specifically:

[0062] (1) Perform a range-to-Fourier transform on the original echo signal matrix to obtain the range-frequency domain echo signal S echo,r_fft .

[0063] The Fourier transform of the embodiment of the present invention is different from the fast Fourier transform (FFT). The specific operation method is as follows: extract the original echo signal matrix S echo The first row of this matrix is r / 2 elements and N r / 2 elements are swapped, and then FFT is performed on the swapped row matrix. After the FFT operation is completed, the first half of the elements and the second half of the elements of the row matrix after FFT are swapped again, and the distance Fourier transform of the first row of echo signal data is completed. The above operation is performed on each row of data in the echo matrix in turn to complete the distance Fourier transform operation of the original echo signal matrix. The signal after the distance Fourier transform is the distance frequency domain echo signal, denoted as S echo,r_fft .

[0064] (2) The distance frequency domain echo signal S echo,r_fft , perform distance compression based on matched filtering to obtain the distance compression frequency domain echo signal S′ echo,rc_fft .

[0065] In the frequency domain, a matched filter is constructed to achieve distance compression. The expression of the matched filter is as follows:

[0066] In the formula, H rc represents a matched filter, fr represents the frequency of range sampling, j represents an imaginary unit, and exp{*} represents an exponential function.

[0067] The distance frequency domain echo signal S echo,r_fft Each row of is respectively associated with the filter H rc Multiply point by point, and record the multiplied two-dimensional matrix as S echo,rc_fft Then, for the two-dimensional matrix S echo,rc_fft The distance is padded with α times of zero at both ends of the dimension. The matrix after zero padding is Then perform the inverse Fourier transform to complete the pulse compression. The signal after the range compression is used as the range compression frequency domain echo signal, which is recorded as S′ echo,rc_fft Where Na represents the number of pulses emitted, and Nr represents the number of points sampled in the range direction.

[0068] (3) Distance compression frequency domain echo signal S′ echo,rc_fft , perform inverse Fourier transform to obtain the range compressed time domain echo signal S echo,rc .

[0069] The method of distance to Fourier inverse transform is the same as the steps of distance to Fourier transform described above, except that the FFT function used in the process is replaced by the IFFT function.

[0070] The observation scene (ie, the first scene) S1 is defined below.

[0071] Taking the scene center reference slant distance R0 as the center, take a scene that is not smaller than the observation scene size (i.e., set scene), and divide the scene into grids. Assume that there are P points along the x-axis and Q points along the y-axis. The spacing between grids should be less than or equal to half of the azimuth antenna size L. a / 2, the spacing between grids is l, l≤L a / 2, but the grid spacing l should not be too small, because too small a grid spacing cannot optimize its reconstruction performance and will further affect the efficiency of the algorithm. Assume that the center coordinates (x p ,0,0), where The coordinates of each scene point can be obtained, and the coordinates of each scene point are expressed as follows:

[0072]

[0073] Assume that the initial position of the satellite sampling is [x track ,0,H], then the satellite's sequential sampling position coordinates are {[x track +V e η1,0,H],…,[x track +V e η m ,0,H],…[x track +V e η Na ,0,H]}. Finally, the slant distance R of all scene points at each sampling time is calculated based on the scene point coordinates and the satellite sampling position coordinates. p,q (η m ).

[0074] At each azimuth sampling moment, according to the time delay 2R p,q (η m ) / c, obtain the echo delay signal S of all scene points in the above-defined scene S1 echo,rc (2R p,q (η m ) / c). The echo delay signal S obtained at each moment echo,rc (2R p,q (η m ) / c) are multiplied by the azimuth compensation phase exp{j4πR p,q (η m ) / λ} (that is, superimpose the echo delay signal obtained at each moment and the azimuth compensation phase at the corresponding moment), and put it into a 1×N 2 The empty matrix S echo,ac , where c is the speed of light. Finally, the matrix S after azimuth compression echo,acPerform N×N redivision to obtain the initial imaging result

[0075] In order to further explore the redundancy of scene information and improve sampling efficiency, such as Figure 3 As shown, the present application uses an autoencoder to fully utilize the redundancy of backscatter coefficients in the SAR imaging process. An autoencoder is a deep neural network that learns efficient data encoding, which uses an encoder network to convert input data into a lower dimensional hidden layer that describes the code used to represent the input data. In addition to the encoder, a reconstruction network, the decoder, is also learned, in which the autoencoder strives to generate a representation as close to its original input data as possible from the reduced hidden layer. If the dimension of the output of the hidden layer is lower than the input data, the output of the hidden layer can be regarded as a compressed representation of the input data.

[0076] The autoencoder includes an encoder ξ(·), a decoder ζ(·), a flattened layer Γ(·) and a fully connected module Ξ(·) connected in sequence; wherein the encoder is composed of multiple convolutional layers; the decoder is composed of multiple convolutional layers, and both the encoder and the decoder are CNN models. The fully connected module Ξ(·) consists of a real fully connected layer Ξ R () and the imaginary fully connected layer Ξ I ()constitute.

[0077] 1): The encoder consists of multiple convolutional layers. The output of the encoder is expressed as follows:

[0078] L=ξ(σ)=ψ K (ψ K-1 (…ψ 1 (σ))) (4);

[0079] Where L = ξ(σ) represents the output of the encoder, ψ k represents the kth convolutional layer. Assume A k is the output of the kth convolutional layer in the encoder, A 0 Represents initialization input, then

[0080] A k =ψ k (A k-1 )=M(H(C(A k-1 ,W k ,B k ))) (5);

[0081] Among them, Μ() represents the maximum pooling, which extracts the local hidden layer from the previous hidden layer to reduce the spatial size, number of parameters and computational cost of the CNN autoencoder; H() is a nonlinear activation function (ReLU), which aims to increase the nonlinear characteristics of the network while maintaining the receptive field of the convolutional layer; C() represents a two-dimensional convolutional layer, which is the core block of the convolutional layer. Assume that the output of C() at the jth feature map of the kth convolutional layer is represented as C() is represented as

[0082]

[0083] Where W k and B k represents the weight and bias of the kth convolutional layer; i and j are the rows and columns of the two-dimensional convolutional layer weight W, that is, is the (i, j)th element of the two-dimensional convolutional layer weight matrix of the kth convolutional layer.

[0084] In this application, the encoder is composed of three convolutional layers, wherein the filter size corresponding to the three convolutional layers is 3×3, the convolutional layer step size is 1, the padding size is 1, and the three convolutional layers correspond to 16, 8, and 4 filters respectively.

[0085] 2) The decoder consists of multiple convolutional layers. The output of the decoder is expressed as follows:

[0086]

[0087] Where k = K + 1, K + 2, ..., 2K, the encoder and decoder are composed of k convolutional layers respectively, so the autoencoder has a total of 2K convolutional layers; the CNN network layer ψ in the decoder k () is connected by the up sampling layer U(), the nonlinear activation function H() and the two-dimensional convolution layer. The output of the kth convolution layer can be expressed as follows:

[0088] A k =ψ k (A k-1 )=H(C(U(A k-1 ,W k ,B k ))) (8);

[0089] in

[0090] In this application, the decoder is composed of three convolutional layers, wherein the filter size corresponding to the three convolutional layers is 3×3, the convolutional layer step size is 1, the padding size is 1, and the three convolutional layers correspond to 4, 8, and 1 filters respectively.

[0091] 3) Flattening layer Γ() converts the decoder output Perform real and virtual separation, and the real and virtual separation process is shown in the following formula.

[0092]

[0093] in, and Decoder output The real and imaginary parts of .

[0094] 4) In the fully connected layer Ξ(·), W R = [Re(D) - Im(D)] and W I = [Im(D)-Re(D)], where W R and W I Represent the real part of the fully connected layer Ξ R () and the imaginary fully connected layer Ξ I () weight; D represents the measurement matrix of the echo in step 203 (ie, formula (1)); for the real part of the fully connected layer Ξ R () and the imaginary fully connected layer Ξ I (), no activation function is used and the bias is set to zero.

[0095] Based on simulation scene dataset Pre-train the network to obtain the initialization network parameters Θ of the autoencoder 0 During the network pre-training process, the model loss function is set and minimized. The calculation formula of the model loss function is as follows:

[0096]

[0097] Train the network to obtain the initialization network parameters Θ of the autoencoder 0 ; Θ={(W 1 ,B 1 ),(W 2 ,B 2 ),…,(W 2k ,B 2k )} represents the autoencoder parameters, including the model parameters of each convolutional layer in the encoder and the model parameters of each convolutional layer in the decoder.

[0098] Input initialization network parameters Θ 0 And initialization imaging results The imaging loss function J2 is minimized by cyclic iteration to obtain the final imaging result. Then step 205 specifically includes: substituting the initialized network parameters and the initialized imaging results into the imaging loss function, performing iterative optimization based on the gradient descent method of block coordinate descent, and stopping the iteration when the set conditions are met to obtain the final imaging result.

[0099] The imaging loss function is expressed as follows:

[0100]

[0101] Among them, J2 is the imaging loss function, Θ represents the autoencoder parameters, σ represents the input of the autoencoder; s is the observation equation, including the observation equation S of all parameter image combinations m ;Ξ R () represents the real part of the fully connected layer of the autoencoder, Ξ I () represents the imaginary fully connected layer of the autoencoder, Γ() represents the flat layer of the autoencoder, ξ() represents the encoder of the autoencoder, and ζ() represents the decoder of the autoencoder.

[0102] The loss function J2 is a non-convex optimization problem. This application is based on the gradient descent method of block coordinate descent, and solves Θ and σ simultaneously through an iterative optimization method. In the algorithm, J2 is the objective function of the optimization problem, and the number of iterations is represented by l. Since Θ and σ are negative numbers, we note that the conjugate gradient of Θ and σ gives the direction of the maximum first-order change. Therefore, Θ and σ change with the gradient direction. and Update, each image forms the estimated weight of the autoencoder. The first and second steps of the algorithm are implemented by gradient descent, and the specific implementation method is as follows:

[0103] In the lth iteration, the network weight Θ is estimated by

[0104]

[0105] In the above formula, It can be directly calculated by the back-propagation algorithm, which can calculate the gradient of the loss function with respect to the convolutional layer weights.

[0106] In the lth iteration, the backscatter coefficient σ is estimated by

[0107]

[0108] In the above formula, The back propagation algorithm can be used to calculate the gradient of the loss function relative to the convolutional layer weights. 1 =Ψ(σ,W 1 ,B 1 ) represents the output after the first convolutional layer, then

[0109]

[0110] in Representation calculation The instantaneous result.

[0111]

[0112] The (n,m)th element in the output matrix after the first convolutional layer is expressed as:

[0113]

[0114] in

[0115]

[0116] Where U and V represent matrices The number of rows and columns of θ j ′(0,m)=0,θ j ′(n,0)=0 and θ j ′(n,m)=0.

[0117] Set the conditions to in, is the imaging loss function value corresponding to the lth iteration number, is the imaging loss function value corresponding to the l-1th iteration. Stop iteration and get the final imaging result

[0118] The present application also provides an application scenario, which applies the above-mentioned SAR deep super-resolution imaging method based on autoencoder. Specifically: The SAR deep super-resolution imaging method based on autoencoder provided in this embodiment can be applied in SAR imaging scenarios. The SAR imaging scenario includes a data acquisition link and a SAR imaging processing link; the initial data set enters the SAR imaging processing link from the data acquisition link to obtain the corresponding final imaging result. The SAR deep super-resolution imaging method based on autoencoder provided in this embodiment belongs to the SAR imaging processing link. Specifically in the SAR imaging processing link process, each radar working parameter group in the initial data set can be combined with each real SAR image to construct several simulation scenes; all simulation scenes constitute a simulation scene data set; the observation scene is divided based on the radar working parameter group to obtain the first scene; the initialization imaging result is obtained based on the first scene; the autoencoder is pre-trained based on the simulation scene data set to obtain the initialization network parameters; according to the initialization network parameters and the initialization imaging results, the imaging loss function is minimized iteratively to obtain the final imaging result.

[0119] Based on the same inventive concept, the embodiment of the present application also provides a SAR depth super-resolution imaging device based on an autoencoder for implementing the above-mentioned SAR depth super-resolution imaging method based on an autoencoder. The implementation solution provided by the device to solve the problem is similar to the implementation solution recorded in the above-mentioned method, so the specific limitations in one or more embodiments of the SAR depth super-resolution imaging device based on an autoencoder provided below can refer to the limitations of the SAR depth super-resolution imaging method based on an autoencoder above, and will not be repeated here.

[0120] In an exemplary embodiment, Figure 4 As shown, a SAR depth super-resolution imaging device based on an autoencoder is provided, comprising:

[0121] The initial parameter set acquisition module T1 is used to: acquire an initial parameter set; the initial parameter set includes a plurality of radar operating parameter groups and a plurality of real SAR images; the radar operating parameter group includes a scene center reference slant range, a satellite flight speed, an azimuth antenna size, a number of transmitted pulses, a modulation frequency of a transmitted signal, a time vector of a radar transmitting pulse, and an operating wavelength;

[0122] The simulation scene data set construction module T2 is used to: combine each radar working parameter group with each real SAR image to construct a plurality of simulation scenes; all simulation scenes constitute a simulation scene data set;

[0123] The initialization imaging result generating module T3 is used to: divide the observation scene based on the radar working parameter group to obtain a first scene; and obtain the initialization imaging result based on the first scene;

[0124] The autoencoder pre-training module T4 is used to: pre-train the autoencoder based on the simulation scene data set to obtain initialization network parameters;

[0125] The final imaging result generating module T5 is used to iteratively minimize the imaging loss function according to the initialization network parameters and the initialization imaging result to obtain the final imaging result.

[0126] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 5As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store SAR depth super-resolution imaging data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a SAR depth super-resolution imaging method based on an autoencoder is implemented.

[0127] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0128] In an exemplary embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0129] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0130] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0131] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0132] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0133] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0134] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A SAR depth super-resolution imaging method based on an autoencoder, characterized in that: The SAR depth super-resolution imaging method based on the autoencoder comprises: Acquire an initial parameter set; the initial parameter set includes several radar operating parameter groups and several real SAR images; the radar operating parameter group includes a scene center reference slant range, a satellite flight speed, an azimuth antenna size, a number of transmitted pulses, a modulation frequency of a transmitted signal, a time vector of a radar transmitting pulse, and an operating wavelength; Combining each of the radar operating parameter groups with each of the real SAR images to construct a number of simulation scenarios; all simulation scenarios constitute a simulation scenario data set; Divide the observation scene based on the radar working parameter group to obtain a first scene; obtain an initialization imaging result based on the first scene; Pre-training the autoencoder based on the simulation scene data set to obtain initialization network parameters; According to the initialized network parameters and the initialized imaging result, the imaging loss function is minimized through cyclic iteration to obtain the final imaging result.

2. The SAR depth super-resolution imaging method based on the autoencoder according to claim 1 is characterized in that: Each radar operating parameter group is combined with each real SAR image to construct several simulation scenarios, including: Step 301: combining each radar operating parameter group and each real SAR image to obtain a plurality of parameter image combinations; for each parameter image combination, constructing an observation matrix corresponding to the radar operating parameter group according to the radar operating parameter group and the slant range of all scene points in the delineated scene at each sampling time; Step 302: Calculate the observation equation corresponding to the parameter image combination according to the observation matrix corresponding to the radar operating parameter group and the real SAR image in the parameter image combination; Step 303: Based on the matched filtering principle, according to the observation equation corresponding to the parameter image combination and the observation matrix, obtain the simulation scene corresponding to the parameter image combination; for each parameter image combination, repeat the above steps 301 to 303 to obtain the simulation scene corresponding to each parameter image combination.

3. The SAR depth super-resolution imaging method based on autoencoder according to claim 1, characterized in that: Obtaining an initialization imaging result based on the first scene specifically includes: Obtain the original echo signal matrix of the SAR echo pixel points; Obtaining a range-compressed time-domain echo signal based on the original echo signal matrix; Acquire scene point coordinates and satellite sampling position coordinates of the first scene, and obtain slant distances of all scene points in the first scene at each sampling time according to the scene point coordinates of the first scene and the satellite sampling position coordinates; Determine the time delay based on the slant range of all scene points at each sampling time, and obtain the echo time delay signals of all scene points in the first scene according to the time delay and the distance compressed time domain echo signal at the sampling time of each azimuth; The echo delay signal obtained at each sampling moment is superimposed with the azimuth compensation phase at the corresponding sampling moment to obtain an azimuth compression matrix; The azimuth compression matrix is ​​re-divided to obtain an initialized imaging result.

4. The SAR depth super-resolution imaging method based on autoencoder according to claim 1, characterized in that: The autoencoder includes an encoder, a decoder, a flattening layer and a fully connected module connected in sequence; wherein the encoder is composed of multiple convolutional layers; and the decoder is composed of multiple convolutional layers.

5. The SAR depth super-resolution imaging method based on autoencoder according to claim 1, characterized in that: According to the initialized network parameters and the initialized imaging result, the imaging loss function is minimized iteratively to obtain the final imaging result, specifically including: The initialized network parameters and the initialized imaging results are substituted into the imaging loss function, and an iterative optimization is performed based on the gradient descent method of block coordinate descent. When the set conditions are met, the iteration is stopped to obtain the final imaging result.

6. The SAR depth super-resolution imaging method based on autoencoder according to claim 5, characterized in that: The imaging loss function is expressed as follows: Where J2 is the imaging loss function, Θ represents the autoencoder parameter, σ represents the input of the autoencoder, s is the observation equation, and Ξ R () represents the real part of the fully connected layer of the autoencoder, Ξ I () represents the imaginary fully connected layer of the autoencoder, Γ() represents the flat layer of the autoencoder, ξ() represents the encoder of the autoencoder, and ζ() represents the decoder of the autoencoder.

7. The SAR depth super-resolution imaging method based on autoencoder according to claim 6, characterized in that: The setting conditions are in, is the imaging loss function value corresponding to the lth iteration number, is the imaging loss function value corresponding to the l-1th iteration number.

8. A SAR depth super-resolution imaging device based on an autoencoder, characterized in that: The SAR depth super-resolution imaging device based on the autoencoder comprises: The initial parameter set acquisition module is used to: acquire an initial parameter set; the initial parameter set includes a plurality of radar operating parameter groups and a plurality of real SAR images; the radar operating parameter group includes a scene center reference slant range, a satellite flight speed, an azimuth antenna size, a number of transmitted pulses, a modulation frequency of a transmitted signal, a time vector of a radar transmitting pulse, and an operating wavelength; A simulation scene data set construction module is used to: combine each radar working parameter group with each real SAR image to construct a number of simulation scenes; all simulation scenes constitute a simulation scene data set; An initialization imaging result generating module is used to: divide the observation scene based on the radar working parameter group to obtain a first scene; and obtain an initialization imaging result based on the first scene; An autoencoder pre-training module is used to: pre-train the autoencoder based on the simulation scene data set to obtain initialization network parameters; The final imaging result generating module is used to: according to the initialization network parameters and the initialization imaging result, cyclically iteratively minimize the imaging loss function to obtain the final imaging result.

9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the SAR depth super-resolution imaging method based on the autoencoder according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the autoencoder-based SAR depth super-resolution imaging method described in any one of claims 1 to 7 is implemented.

Images