A SAR echo signal simulation method and system based on deep learning

Abstract

The present invention discloses a SAR echo signal simulation method and system based on deep learning, which relate to the technical field of radar signal processing. The method constructs a concentric circle operator through a concentric circle algorithm, pre-processes radar parameters and scene information to obtain an initial echo signal, inputs the initial echo signal into a deep residual network and an iterative numerical optimization block, implements data consistency constraints, constructs a data set using historical echo signals simulated by traditional methods and trains an echo generation model, reconstructs echoes according to radar parameters and scene information through the echo generation model, and obtains high-precision SAR echo signals. In summary, the present invention innovatively combines the concentric circle algorithm with deep learning, and learns and optimizes echoes through a deep neural network. This combination not only improves the simulation accuracy of the echo signal, but also effectively reduces the error in the echo signal, so that the acquired echo is more accurate and reliable.

Description

Technical Field

[0001] The present invention relates to the technical field of radar signal processing, and in particular to a SAR echo signal simulation method and system based on deep learning. Background Art

[0002] Radar technology is widely used in many fields such as missile guidance, ground observation, disaster monitoring and environmental protection due to its all-weather, all-day and long-distance detection capabilities. In the development of synthetic aperture radar (SAR) systems, the rapid simulation technology of radar echo signals is particularly important. It can simulate SAR echo signals in various complex environments, thereby effectively evaluating system performance, optimizing real-time imaging processing hardware, and reducing development risks. This technology not only improves R&D efficiency and accuracy, but also expands the application potential of radar in military and civilian fields, and promotes the progress and development of related technologies.

[0003] SAR imaging relies on the simulation and processing of echo signals. Traditional SAR echo simulation methods are mainly based on time domain algorithms or frequency domain algorithms. Although these methods can theoretically obtain accurate echo signals, they have the following problems in practical applications:

[0004] 1. High computational complexity: Traditional time-domain algorithms require a lot of computing resources. Especially when the scene is large and the number of simulation points is large, the computational efficiency is extremely low and it is difficult to meet the needs of real-time applications.

[0005] 2. Low echo generation accuracy: The concentric circle echo simulation algorithm makes partial approximation in the distance direction and is more efficient than the traditional time domain echo simulation method. However, since all point targets in the scene are distributed according to integer multiples of the distance sampling interval, the resulting SAR echo is not accurate.

[0006] 3. Motion errors are difficult to embed: Methods based on azimuth frequency domain processing usually need to meet or approximately meet the assumption that the echo signal is azimuthally constant. However, in many SAR application modes with non-uniform linear trajectories and complex geometric configurations, the echo signal no longer meets the assumption of azimuth invariance, which introduces difficulties to simulation methods based on azimuth frequency domain processing. The imaging inverse processing echo generation algorithm has certain advantages in computational efficiency and can generate original echo signals corresponding to the scene, but it cannot explain the impact of the sensor trajectory deviation relative to the nominal straight line path and cannot embed motion errors. It has great limitations in application, and the accuracy of the generated echo is lower than that of the time domain algorithm.

[0007] In summary, the problems and defects of the existing technology are: how to design a more efficient algorithm to reduce the computational complexity, especially in large scenes and multi-point targets; how to flexibly embed motion errors to accurately reflect the actual trajectory of the sensor in the echo; how to ensure that the generated echo signal is as close to the actual situation as possible in amplitude and phase to meet the needs of high-precision imaging. Summary of the invention

[0008] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a SAR echo signal simulation method and system based on deep learning, aiming to solve the problems of high computational complexity and obvious motion errors in the echo signal simulation in the prior art.

[0009] A first aspect of the present invention is to provide a SAR echo signal simulation method based on deep learning, the method comprising:

[0010] Constructing a concentric circle operator according to a concentric circle echo generation algorithm, and preprocessing the currently input radar parameters and scene information by means of the concentric circle operator; wherein the preprocessing of the currently input radar parameters and scene information includes converting the radar parameters and the scene information into an initial echo signal suitable for an echo generation model;

[0011] The initial echo signal is introduced into a deep residual network, and deep feature extraction and denoising are performed on the initial echo signal through the deep residual network to obtain an intermediate echo signal, and the intermediate echo signal is introduced into an iterative numerical optimization block, and data consistency constraints are performed on the intermediate echo signal through the iterative numerical optimization block; wherein the deep residual network is a fusion network of a CNN convolutional neural network and a residual learning network, and the deep residual network and the iterative numerical optimization block are alternately operated to form the echo generation model;

[0012] Constructing a data set from historical echo signals simulated by a preset method, and importing the data set into the echo generation model for model training to adjust model parameters until the model converges;

[0013] The trained echo generation model is used to simulate and reconstruct the echo signal according to the pre-provided radar parameters and scene information to obtain the SAR echo signal.

[0014] According to one aspect of the above technical solution, the initial echo signal is introduced into a deep residual network, deep feature extraction and denoising are performed on the initial echo signal through the deep residual network to obtain an intermediate echo signal, the intermediate echo signal is introduced into an iterative numerical optimization block, and the step of performing data consistency constraints on the intermediate echo signal through the iterative numerical optimization block includes:

[0015] The initial echo signal is introduced into the deep residual network, and the initial echo signal is regularized by the deep residual network to remove noise and artifacts in the initial echo signal to obtain an intermediate echo signal;

[0016] The intermediate echo signal is introduced into the iterative numerical optimization block, and the conjugate gradient algorithm is used to perform an iterative algorithm solution on the intermediate echo signal to constrain data consistency.

[0017] According to one aspect of the above technical solution, the initial echo signal is introduced into the deep residual network, the initial echo signal is regularized by the deep residual network, the noise and artifacts in the initial echo signal are removed, and the step of obtaining the intermediate echo signal includes:

[0018] Preprocessing the initial echo signal, decomposing dual-channel data according to a real channel and an imaginary channel, including real data and imaginary data, and normalizing the real data and the imaginary data;

[0019] Inputting the normalized real data and the imaginary data into the deep residual network; wherein the deep residual network includes a plurality of residual blocks connected in series, the residual blocks connected in series to the front end all include a 3×3 convolutional layer, a batch normalization layer and an activation function, and the last residual block connected in series to the back end only includes a 3×3 convolutional layer and a batch normalization layer;

[0020] gradually extracting signal features in the real data and the imaginary data through the plurality of 3×3 convolutional layers and the activation function to suppress noise features and artifact features;

[0021] Mapping the signal features to the real channel and the imaginary channel to obtain denoised real data and imaginary data;

[0022] The denoised real data and imaginary data are denormalized by the batch normalization layer and merged into complex signals to reconstruct the intermediate echo signal in complex form.

[0023] According to one aspect of the above technical solution, the step of gradually extracting the signal features in the real data and the imaginary data through the plurality of the 3×3 convolutional layers and the activation function to suppress the noise features and the artifact features includes:

[0024] Capturing shallow features in the real data and the imaginary data through a low-level convolutional layer, including capturing noise features and artifact features;

[0025] The global context information in the real data and the imaginary data is obtained through a high-level convolutional layer in combination with the activation function, so as to separate the signal and the phase error according to the global context information and obtain the signal feature.

[0026] According to one aspect of the above technical solution, a data set is constructed from historical echo signals simulated by a preset method, and the data set is imported into the echo generation model for model training to adjust model parameters until the model converges, including:

[0027] Simulating a historical echo signal using a time domain echo simulation algorithm, and generating a label data set according to the historical echo signal;

[0028] The label data set and the input radar parameters and scene information form a data pair, and the data pair is imported into the echo generation model for model training;

[0029] The mean square error between echoes is selected as the loss function to measure the difference between the predicted value and the true value, so as to adjust the model parameters according to the difference until the model converges.

[0030] According to one aspect of the above technical solution, the step of simulating and reconstructing the echo signal according to the pre-provided radar parameters and scene information by the trained echo generation model to obtain the SAR echo signal includes:

[0031] Importing pre-provided radar parameters and scene information into the trained echo generation model;

[0032] According to the radar parameters and scene information, the echo signal is simulated and reconstructed through the echo generation model to obtain the SAR echo signal.

[0033] A second aspect of the present invention is to provide a SAR echo signal simulation system based on deep learning, which is applied to the method described in the above technical solution, and the system comprises:

[0034] A preprocessing module, configured to construct a concentric circle operator according to a concentric circle echo generation algorithm, and preprocess the currently input radar parameters and scene information by means of the concentric circle operator; wherein the preprocessing of the currently input radar parameters and scene information includes converting the radar parameters and the scene information into an initial echo signal suitable for an echo generation model;

[0035] A learning module, used for importing the initial echo signal into a deep residual network, performing deep feature extraction and denoising on the initial echo signal through the deep residual network to obtain an intermediate echo signal, importing the intermediate echo signal into an iterative numerical optimization block, and performing data consistency constraints on the intermediate echo signal through the iterative numerical optimization block; wherein the deep residual network is a fusion network of a CNN convolutional neural network and a residual learning network, and the deep residual network and the iterative numerical optimization block are alternately operated to form the echo generation model;

[0036] A training module, used to construct a data set from historical echo signals simulated by a preset method, and import the data set into the echo generation model for model training to adjust model parameters until the model converges;

[0037] The reconstruction module is used to simulate and reconstruct the echo signal according to the pre-provided radar parameters and scene information through the trained echo generation model to obtain the SAR echo signal.

[0038] According to one aspect of the above technical solution, the learning module is used for:

[0039] The initial echo signal is introduced into the deep residual network, and the initial echo signal is regularized by the deep residual network to remove noise and artifacts in the initial echo signal to obtain an intermediate echo signal;

[0040] The intermediate echo signal is introduced into the iterative numerical optimization block, and the conjugate gradient algorithm is used to perform an iterative algorithm solution on the intermediate echo signal to constrain data consistency.

[0041] A third aspect of the present invention is to provide a readable storage medium having a computer program stored thereon, which implements the method described in the above technical solution when executed by a processor.

[0042] A fourth aspect of the present invention is to provide an electronic device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the method described in the above technical solution when executing the computer program.

[0043] Compared with the prior art, the SAR echo signal simulation method and system based on deep learning shown in the present invention have the following beneficial effects:

[0044] In the present invention, a concentric circle operator is constructed by a concentric circle algorithm, and radar parameters and scene information are preprocessed according to the concentric circle operator to obtain an initial echo signal. The initial echo signal is then input into a residual learning network based on CNN, namely a deep residual network and iterative numerical optimization block, and data consistency constraints are implemented to construct an echo generation model. Then, a data set is constructed using historical echo signals simulated by traditional methods to train the echo generation model. Finally, the echo is reconstructed according to radar parameters and scene information through the echo generation model to obtain a high-precision SAR echo signal. In summary, the present invention innovatively combines the concentric circle algorithm with deep learning, and learns and optimizes the echo through a deep neural network. This combination not only improves the simulation accuracy of the echo signal, but also effectively reduces the error in the echo signal, so that the acquired echo is more accurate and reliable, and can effectively solve the computational complexity and motion error problems in the prior art. It is suitable for a variety of radar parameters and complex scene information, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0046] Figure 1 Schematic diagram of a flow chart of a SAR echo signal simulation method based on deep learning in one embodiment of the present invention;

[0047] Figure 2 This is a structural block diagram of a SAR echo signal simulation system based on deep learning in one embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings. Several embodiments of the present invention are shown in the accompanying drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0049] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.

[0050] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0051] Embodiment 1

[0052] See also Figure 1 The first embodiment of the present invention provides a SAR echo signal simulation method based on deep learning, the method comprising steps S10 to S40:

[0053] Step S10, constructing a concentric circle operator according to the concentric circle echo generation algorithm, and preprocessing the currently input radar parameters and scene information through the concentric circle operator.

[0054] The preprocessing of the currently input radar parameters and scene information includes converting the radar parameters and the scene information into initial echo signals suitable for an echo generation model.

[0055] Step S20, importing the initial echo signal into a deep residual network, performing deep feature extraction and denoising processing on the initial echo signal through the deep residual network to obtain an intermediate echo signal, importing the intermediate echo signal into an iterative numerical optimization block, and performing data consistency constraints on the intermediate echo signal through the iterative numerical optimization block.

[0056] Among them, the deep residual network is a fusion network of a CNN convolutional neural network and a residual learning network, and the deep residual network and the iterative numerical optimization block run alternately to form the echo generation model.

[0057] In this embodiment, the initial echo signal is introduced into a deep residual network, deep feature extraction and denoising are performed on the initial echo signal through the deep residual network to obtain an intermediate echo signal, the intermediate echo signal is introduced into an iterative numerical optimization block, and the step of performing data consistency constraints on the intermediate echo signal through the iterative numerical optimization block includes:

[0058] The initial echo signal is introduced into the deep residual network, and the initial echo signal is regularized by the deep residual network to remove noise and artifacts in the initial echo signal to obtain an intermediate echo signal;

[0059] The intermediate echo signal is introduced into the iterative numerical optimization block, and the conjugate gradient algorithm is used to perform an iterative algorithm solution on the intermediate echo signal to constrain data consistency.

[0060] Specifically, in this embodiment, the radar parameters and scene information are preprocessed to obtain the initial echo signal, and then the initial echo signal is sequentially introduced into the deep residual network and the iterative numerical optimization block. The purpose is to perform feature learning on the initial echo signal through the CNN-based residual learning network, that is, the deep residual network shown in this embodiment, and output the intermediate echo signal. Moreover, each time the deep residual network is processed and learned, the intermediate echo signal as the transit data is solved by an iterative algorithm through the subsequent iterative numerical optimization block, thereby iteratively processing the initial echo signal to make the data more consistent.

[0061] More specifically, in this embodiment, the initial echo signal obtained after preprocessing is introduced into the deep residual network, and the initial echo signal is regularized through various layers in the deep residual network to remove noise and artifacts in the initial echo signal to obtain an intermediate echo signal containing depth features. The intermediate echo signal containing depth features is then imported into the iterative numerical optimization block, and the conjugate gradient algorithm applied in the iterative numerical optimization block is used to iteratively solve the intermediate echo signal to complete the consistency constraint on the data.

[0062] It should also be noted that the deep residual network and the iterative numerical optimization block are an alternating network architecture to construct an echo generation model, which may include multiple alternating structures to form a model architecture for feature learning of the initial echo signal, i.e., the echo, and at the end of this embodiment, the signal will be simulated and reconstructed through the alternating network architecture.

[0063] Further, the initial echo signal is introduced into the deep residual network, the initial echo signal is regularized by the deep residual network, the noise and artifacts in the initial echo signal are removed, and the step of obtaining the intermediate echo signal includes:

[0064] Preprocessing the initial echo signal, decomposing dual-channel data according to a real channel and an imaginary channel, including real data and imaginary data, and normalizing the real data and the imaginary data;

[0065] Inputting the normalized real data and the imaginary data into the deep residual network; wherein the deep residual network includes a plurality of residual blocks connected in series, the residual blocks connected in series to the front end all include a 3×3 convolutional layer, a batch normalization layer and an activation function, and the last residual block connected in series to the back end only includes a 3×3 convolutional layer and a batch normalization layer;

[0066] gradually extracting signal features in the real data and the imaginary data through the plurality of 3×3 convolutional layers and the activation function to suppress noise features and artifact features;

[0067] Mapping the signal features to the real channel and the imaginary channel to obtain denoised real data and imaginary data;

[0068] The denoised real data and imaginary data are denormalized by the batch normalization layer and merged into complex signals to reconstruct the intermediate echo signal in complex form.

[0069] The step of gradually extracting the signal features in the real data and the imaginary data through the plurality of 3×3 convolutional layers and the activation function to suppress noise features and artifact features includes:

[0070] Capturing shallow features in the real data and the imaginary data through a low-level convolutional layer, including capturing noise features and artifact features;

[0071] The global context information in the real data and the imaginary data is obtained through a high-level convolutional layer in combination with the activation function, so as to separate the signal and the phase error according to the global context information and obtain the signal feature.

[0072] Specifically, in this embodiment, before the initial echo signal is introduced into the deep residual network, the initial echo signal will also be preprocessed, which is not equal to the preprocessing of the radar parameters and scene information. Specifically, the initial echo signal is decomposed into dual-channel data according to the real channel and the imaginary channel, including real data and imaginary data, and the real data and the imaginary data need to be normalized respectively, and then the normalized real data and imaginary data are introduced into the deep residual network, and the signal features in the real data and the imaginary data, that is, the deep features, are gradually extracted through multiple 3×3 convolutional layers and activation functions, so as to suppress the noise features and artifact features in the signal, and then the signal features are mapped back to the real channel and the imaginary channel respectively to obtain the real data and the imaginary data after denoising, and finally the denoised real data and the imaginary data are denormalized through the batch normalization layer, and the denormalized real data and the imaginary data are merged into a complex signal in complex form, thereby reconstructing the intermediate echo signal in complex form.

[0073] In the process of feature extraction and noise and artifact suppression through multiple 3×3 convolutional layers and activation functions, the low-level convolutional layers in the multiple 3×3 convolutional layers will first be used to capture the shallow features in the real data and the imaginary data respectively to capture the noise features and artifact features. Then, the high-level convolutional layers or the middle and high-level convolutional layers in the multiple 3×3 convolutional layers are combined with the activation function to obtain the global context information in the real data and the imaginary data based on the attention mechanism. Finally, the signal and motion error (such as phase error) are separated according to the global context information to obtain the signal features.

[0074] Step S30, constructing a data set from historical echo signals simulated by a preset method, and importing the data set into the echo generation model for model training to adjust model parameters until the model converges.

[0075] In this embodiment, the steps of constructing a data set using historical echo signals simulated by a preset method, importing the data set into the echo generation model for model training, and adjusting model parameters until the model converges include:

[0076] Simulating a historical echo signal using a time domain echo simulation algorithm, and generating a label data set according to the historical echo signal;

[0077] The label data set and the input radar parameters and scene information form a data pair, and the data pair is imported into the echo generation model for model training;

[0078] The mean square error between echoes is selected as the loss function to measure the difference between the predicted value and the true value, so as to adjust the model parameters according to the difference until the model converges.

[0079] Specifically, in this embodiment, the traditional time-domain echo simulation algorithm is used to simulate according to the pre-input radar parameters and scene information to obtain a historical echo signal, that is, an accurate echo, and then a label data set is generated based on the historical echo signal to form a data pair with the radar parameters and scene information. The label is used for supervised learning, and the mean square error between the echoes is selected as the loss function to measure the difference between the predicted value and the true value, so as to adjust the model parameters of the echo generation model according to the difference, and repeat it multiple times until the model converges.

[0080] Step S40: The trained echo generation model is used to simulate and reconstruct the echo signal according to the pre-provided radar parameters and scene information to obtain a SAR echo signal.

[0081] In this embodiment, the step of simulating and reconstructing the echo signal according to the pre-provided radar parameters and scene information by the trained echo generation model to obtain the SAR echo signal includes:

[0082] Importing pre-provided radar parameters and scene information into the trained echo generation model;

[0083] According to the radar parameters and scene information, the echo signal is simulated and reconstructed through the echo generation model to obtain the SAR echo signal.

[0084] Specifically, the radar parameters and scene information that have not been preprocessed are imported into the echo generation model including the deep residual network and its iterative numerical optimization block. According to the radar parameters and scene information, the echo signal is simulated and reconstructed through the echo generation model to obtain a high-precision SAR echo signal. And compared with the traditional time domain echo generation algorithm, the reconstruction quality, efficiency and effectiveness of the deep learning SAR echo signal simulation method based on the concentric circle operator in this embodiment are verified.

[0085] Compared with the prior art, the SAR echo signal simulation method based on deep learning shown in this embodiment has the following beneficial effects:

[0086] In this embodiment, a concentric circle operator is constructed by a concentric circle algorithm, and radar parameters and scene information are preprocessed according to the concentric circle operator to obtain an initial echo signal. The initial echo signal is then input into a CNN-based residual learning network, namely a deep residual network and iterative numerical optimization block, and data consistency constraints are implemented to construct an echo generation model. Then, a data set is constructed using historical echo signals simulated by traditional methods to train the echo generation model. Finally, the echo is reconstructed according to radar parameters and scene information through the echo generation model to obtain a high-precision SAR echo signal. In summary, this embodiment innovatively combines the concentric circle algorithm with deep learning, and learns and optimizes the echo through a deep neural network. This combination not only improves the simulation accuracy of the echo signal, but also effectively reduces the error in the echo signal, so that the acquired echo is more accurate and reliable, and can effectively solve the computational complexity and motion error problems in the prior art. It is suitable for a variety of radar parameters and complex scene information, and has broad application prospects.

[0087] Embodiment 2

[0088] See also Figure 2 The second embodiment of the present invention provides a SAR echo signal simulation system based on deep learning, which is applied to the method described in the above embodiment. The system includes:

[0089] A preprocessing module 10 is used to construct a concentric circle operator according to a concentric circle echo generation algorithm, and preprocess the currently input radar parameters and scene information through the concentric circle operator; wherein the preprocessing of the currently input radar parameters and scene information includes converting the radar parameters and the scene information into an initial echo signal suitable for an echo generation model;

[0090] A learning module 20 is used to import the initial echo signal into a deep residual network, perform deep feature extraction and denoising on the initial echo signal through the deep residual network to obtain an intermediate echo signal, import the intermediate echo signal into an iterative numerical optimization block, and perform data consistency constraints on the intermediate echo signal through the iterative numerical optimization block; wherein the deep residual network is a fusion network of a CNN convolutional neural network and a residual learning network, and the deep residual network and the iterative numerical optimization block are alternately operated to form the echo generation model;

[0091] A training module 30 is used to construct a data set from historical echo signals simulated by a preset method, and import the data set into the echo generation model for model training to adjust model parameters until the model converges;

[0092] The reconstruction module 40 is used to simulate and reconstruct the echo signal according to the pre-provided radar parameters and scene information through the trained echo generation model to obtain the SAR echo signal.

[0093] Wherein, the learning module 20 is used for:

[0094] The initial echo signal is introduced into the deep residual network, and the initial echo signal is regularized by the deep residual network to remove noise and artifacts in the initial echo signal to obtain an intermediate echo signal;

[0095] The intermediate echo signal is introduced into the iterative numerical optimization block, and the conjugate gradient algorithm is used to perform an iterative algorithm solution on the intermediate echo signal to constrain data consistency.

[0096] Wherein, the learning module 20 is also used for:

[0097] Preprocessing the initial echo signal, decomposing dual-channel data according to a real channel and an imaginary channel, including real data and imaginary data, and normalizing the real data and the imaginary data;

[0098] Inputting the normalized real data and the imaginary data into the deep residual network; wherein the deep residual network includes a plurality of residual blocks connected in series, the residual blocks connected in series to the front end all include a 3×3 convolutional layer, a batch normalization layer and an activation function, and the last residual block connected in series to the back end only includes a 3×3 convolutional layer and a batch normalization layer;

[0099] gradually extracting signal features in the real data and the imaginary data through the plurality of 3×3 convolutional layers and the activation function to suppress noise features and artifact features;

[0100] Mapping the signal features to the real channel and the imaginary channel to obtain denoised real data and imaginary data;

[0101] The denoised real data and imaginary data are denormalized by the batch normalization layer and merged into complex signals to reconstruct the intermediate echo signal in complex form.

[0102] Wherein, the learning module 20 is also used for:

[0103] Capturing shallow features in the real data and the imaginary data through a low-level convolutional layer, including capturing noise features and artifact features;

[0104] The global context information in the real data and the imaginary data is obtained through a high-level convolutional layer in combination with the activation function, so as to separate the signal and the phase error according to the global context information and obtain the signal feature.

[0105] Compared with the prior art, the SAR echo signal simulation system based on deep learning shown in this embodiment has the following beneficial effects:

[0106] In this embodiment, a concentric circle operator is constructed by a concentric circle algorithm, and radar parameters and scene information are preprocessed according to the concentric circle operator to obtain an initial echo signal. The initial echo signal is then input into a CNN-based residual learning network, namely a deep residual network and iterative numerical optimization block, and data consistency constraints are implemented to construct an echo generation model. Then, a data set is constructed using historical echo signals simulated by traditional methods to train the echo generation model. Finally, the echo is reconstructed according to radar parameters and scene information through the echo generation model to obtain a high-precision SAR echo signal. In summary, this embodiment innovatively combines the concentric circle algorithm with deep learning, and learns and optimizes the echo through a deep neural network. This combination not only improves the simulation accuracy of the echo signal, but also effectively reduces the error in the echo signal, so that the acquired echo is more accurate and reliable, and can effectively solve the computational complexity and motion error problems in the prior art. It is suitable for a variety of radar parameters and complex scene information, and has broad application prospects.

[0107] Embodiment 3

[0108] A third embodiment of the present invention provides a readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method described in the above embodiment is implemented.

[0109] Embodiment 4

[0110] A fourth embodiment of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the method described in the above embodiment when executing the computer program.

[0111] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples" or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0112] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for those of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A SAR echo signal simulation method based on deep learning, characterized in that: The method comprises: Constructing a concentric circle operator according to a concentric circle echo generation algorithm, and preprocessing the currently input radar parameters and scene information by means of the concentric circle operator; wherein the preprocessing of the currently input radar parameters and scene information includes converting the radar parameters and the scene information into an initial echo signal suitable for an echo generation model; The initial echo signal is introduced into a deep residual network, and deep feature extraction and denoising are performed on the initial echo signal through the deep residual network to obtain an intermediate echo signal, and the intermediate echo signal is introduced into an iterative numerical optimization block, and data consistency constraints are performed on the intermediate echo signal through the iterative numerical optimization block; wherein the deep residual network is a fusion network of a CNN convolutional neural network and a residual learning network, and the deep residual network and the iterative numerical optimization block are alternately operated to form the echo generation model; Constructing a data set from historical echo signals simulated by a preset method, and importing the data set into the echo generation model for model training to adjust model parameters until the model converges; The trained echo generation model is used to simulate and reconstruct the echo signal according to the pre-provided radar parameters and scene information to obtain the SAR echo signal.

2. The SAR echo signal simulation method based on deep learning according to claim 1, characterized in that: The steps of importing the initial echo signal into a deep residual network, performing deep feature extraction and denoising processing on the initial echo signal through the deep residual network to obtain an intermediate echo signal, importing the intermediate echo signal into an iterative numerical optimization block, and performing data consistency constraints on the intermediate echo signal through the iterative numerical optimization block include: The initial echo signal is introduced into the deep residual network, and the initial echo signal is regularized by the deep residual network to remove noise and artifacts in the initial echo signal to obtain an intermediate echo signal; The intermediate echo signal is introduced into the iterative numerical optimization block, and the conjugate gradient algorithm is used to perform an iterative algorithm solution on the intermediate echo signal to constrain data consistency.

3. The SAR echo signal simulation method based on deep learning according to claim 2, characterized in that: The step of importing the initial echo signal into the deep residual network, performing regularization processing on the initial echo signal through the deep residual network, removing noise and artifacts in the initial echo signal, and obtaining an intermediate echo signal includes: Preprocessing the initial echo signal, decomposing dual-channel data according to a real channel and an imaginary channel, including real data and imaginary data, and normalizing the real data and the imaginary data; Inputting the normalized real data and the imaginary data into the deep residual network; wherein the deep residual network includes a plurality of residual blocks connected in series, the residual blocks connected in series to the front end all include a 3×3 convolutional layer, a batch normalization layer and an activation function, and the last residual block connected in series to the back end only includes a 3×3 convolutional layer and a batch normalization layer; gradually extracting signal features in the real data and the imaginary data through the plurality of 3×3 convolutional layers and the activation function to suppress noise features and artifact features; Mapping the signal features to the real channel and the imaginary channel to obtain denoised real data and imaginary data; The denoised real data and imaginary data are denormalized by the batch normalization layer and merged into complex signals to reconstruct the intermediate echo signal in complex form.

4. The SAR echo signal simulation method based on deep learning according to claim 3, characterized in that: The step of gradually extracting signal features in the real data and the imaginary data through a plurality of the 3×3 convolutional layers and the activation function to suppress noise features and artifact features includes: Capturing shallow features in the real data and the imaginary data through a low-level convolutional layer, including capturing noise features and artifact features; The global context information in the real data and the imaginary data is obtained through a high-level convolutional layer in combination with the activation function, so as to separate the signal and the phase error according to the global context information and obtain the signal feature.

5. The SAR echo signal simulation method based on deep learning according to any one of claims 1 to 4, characterized in that: The steps of constructing a data set from historical echo signals obtained by simulation using a preset method, and importing the data set into the echo generation model for model training to adjust model parameters until the model converges include: Simulating a historical echo signal using a time domain echo simulation algorithm, and generating a label data set according to the historical echo signal; The label data set and the input radar parameters and scene information form a data pair, and the data pair is imported into the echo generation model for model training; The mean square error between echoes is selected as the loss function to measure the difference between the predicted value and the true value, so as to adjust the model parameters according to the difference until the model converges.

6. The SAR echo signal simulation method based on deep learning according to claim 5, characterized in that: The step of simulating and reconstructing the echo signal according to the pre-provided radar parameters and scene information by the trained echo generation model to obtain the SAR echo signal includes: Importing pre-provided radar parameters and scene information into the trained echo generation model; According to the radar parameters and scene information, the echo signal is simulated and reconstructed through the echo generation model to obtain the SAR echo signal.

7. A SAR echo signal simulation system based on deep learning, characterized in that: The method applied to any one of claims 1 to 6, wherein the system comprises: A preprocessing module, configured to construct a concentric circle operator according to a concentric circle echo generation algorithm, and preprocess the currently input radar parameters and scene information by means of the concentric circle operator; wherein the preprocessing of the currently input radar parameters and scene information includes converting the radar parameters and the scene information into an initial echo signal suitable for an echo generation model; A learning module, used for importing the initial echo signal into a deep residual network, performing deep feature extraction and denoising on the initial echo signal through the deep residual network to obtain an intermediate echo signal, importing the intermediate echo signal into an iterative numerical optimization block, and performing data consistency constraints on the intermediate echo signal through the iterative numerical optimization block; wherein the deep residual network is a fusion network of a CNN convolutional neural network and a residual learning network, and the deep residual network and the iterative numerical optimization block are alternately operated to form the echo generation model; A training module, used to construct a data set from historical echo signals simulated by a preset method, and import the data set into the echo generation model for model training to adjust model parameters until the model converges; The reconstruction module is used to simulate and reconstruct the echo signal according to the pre-provided radar parameters and scene information through the trained echo generation model to obtain the SAR echo signal.

8. The SAR echo signal simulation system based on deep learning according to claim 7, characterized in that: The learning modules are used to: The initial echo signal is introduced into the deep residual network, and the initial echo signal is regularized by the deep residual network to remove noise and artifacts in the initial echo signal to obtain an intermediate echo signal; The intermediate echo signal is introduced into the iterative numerical optimization block, and the conjugate gradient algorithm is used to perform an iterative algorithm solution on the intermediate echo signal to constrain data consistency.

9. A readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.

Images