Synthetic Aperture Radar Image Reconstruction Methods, Devices, Storage Media, and Electronic Equipment
By using adaptive block division and Schrödinger bridge super-resolution model to process SAR echo data, the problem of information loss in post-imaging processing was solved, and the resolution of SAR images was improved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BANKER FUTURE TECH (BEIJING) CO LTD
- Filing Date
- 2026-02-26
- Publication Date
- 2026-05-26
AI Technical Summary
Existing technologies lose information from the original echo data when processing SAR images after imaging, and cannot effectively improve the resolution of SAR images.
An imaging entropy minimization algorithm is used to adaptively segment the original synthetic aperture radar echo data, and a pre-defined Schrödinger bridge super-resolution model is used to perform super-resolution processing on the low-resolution echo data blocks. Then, the data is stitched together and reconstructed to make full use of the information in the original echo data.
By directly processing the raw echo data, information loss during the imaging process is avoided, significantly improving the resolution of SAR images.
Smart Images
Figure CN121767477B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of information technology, and in particular to a synthetic aperture radar image reconstruction method, apparatus, storage medium, and electronic device. Background Technology
[0002] Synthetic Aperture Radar (SAR) is an active microwave remote sensing system with all-weather, 24 / 7 observation capabilities. It has been widely applied in topographic mapping, disaster monitoring, environmental remote sensing, and national defense. The resolution of a SAR system is a key performance indicator, directly determining the level of detail in its Earth observations. However, due to physical limitations such as radar system antenna size, transmission bandwidth, pulse repetition frequency, and platform orbital altitude, directly acquiring ultra-high-resolution SAR images is costly and technically challenging. Therefore, improving SAR image resolution through signal processing techniques is a pressing technical problem that needs to be solved.
[0003] Currently, most existing methods perform resolution processing on the SAR image after imaging. However, the raw SAR echo data contains richer information than the SAR image, which is crucial for achieving higher resolution. The current method of imaging first and then processing loses information from the raw echo data during the imaging process, and cannot effectively improve the resolution of the SAR image. Summary of the Invention
[0004] In view of this, this application provides a synthetic aperture radar image reconstruction method, apparatus, storage medium and electronic device, which mainly improves the resolution of SAR images.
[0005] According to a first aspect of this application, a synthetic aperture radar image reconstruction method is provided, the method comprising:
[0006] Acquire raw synthetic aperture radar echo data of the target area;
[0007] An imaging entropy minimization algorithm is used to adaptively divide the original synthetic aperture radar echo data into multiple blocks, in which there is an overlapping area between two adjacent original echo data blocks.
[0008] The multiple original echo data blocks are respectively input into a preset Schrödinger bridge super-resolution model for super-resolution processing to obtain multiple processed echo data blocks, wherein the resolution corresponding to the processed echo data blocks is higher than the resolution corresponding to the original echo data blocks.
[0009] The processed echo data blocks are stitched together and reconstructed to obtain a synthetic aperture radar image.
[0010] According to a second aspect of this application, a synthetic aperture radar image reconstruction apparatus is provided, the apparatus comprising:
[0011] The acquisition unit is used to acquire raw synthetic aperture radar echo data of the target area.
[0012] The block unit is used to adaptively block the original synthetic aperture radar echo data using an imaging entropy minimization algorithm to obtain multiple original echo data blocks, wherein there is an overlapping area between two adjacent original echo data blocks.
[0013] The super-resolution unit is used to input the multiple original echo data blocks into a preset Schrödinger bridge super-resolution model for super-resolution processing to obtain multiple processed echo data blocks, wherein the resolution corresponding to the processed echo data blocks is higher than the resolution corresponding to the original echo data blocks.
[0014] The reconstruction unit is used to stitch together and reconstruct the multiple processed echo data blocks to obtain a synthetic aperture radar image.
[0015] According to a third aspect of this application, a storage medium is provided that stores a computer program thereon, which, when executed by a processor, implements the above-described synthetic aperture radar image reconstruction method.
[0016] According to a fourth aspect of this application, an electronic device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the program to implement the above-described synthetic aperture radar image reconstruction method.
[0017] By employing the above technical solutions, this application provides a synthetic aperture radar (SAR) image reconstruction method, apparatus, storage medium, and electronic device. Compared with the traditional method of imaging first and then processing, this method utilizes a preset Schrödinger bridge super-resolution model to directly perform super-resolution processing on the low-resolution original SAR echo data. This converts the low-resolution original SAR echo data into high-resolution SAR echo data before image reconstruction, fully utilizing the information in the original SAR echo data and avoiding information loss during the imaging process, thereby fundamentally improving SAR image resolution. Furthermore, due to the large size of the original SAR echo data, in order to input it into the preset Schrödinger bridge super-resolution model for processing, this application uses an imaging entropy minimization algorithm to adaptively segment the original SAR echo data, ensuring that the segmented echo data not only meets the model's input conditions but also satisfies the final imaging accuracy requirements.
[0018] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description
[0019] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:
[0020] Figure 1 A schematic flowchart of a synthetic aperture radar image reconstruction method provided in an embodiment of this application is shown;
[0021] Figure 2 A flowchart illustrating the pre-defined Schrödinger bridge super-resolution model training method provided in an embodiment of this application is shown.
[0022] Figure 3 A schematic diagram of a synthetic aperture radar image reconstruction device provided in an embodiment of this application is shown. Detailed Implementation
[0023] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.
[0024] The existing technology, which uses imaging first and then processing, loses information from the original echo data during imaging and cannot effectively improve the resolution of SAR images.
[0025] To address the aforementioned problems, embodiments of the present invention provide a synthetic aperture radar image reconstruction method, such as... Figure 1 As shown, the method includes:
[0026] Step 101: Obtain the raw synthetic aperture radar echo data of the target area.
[0027] The target area refers to the monitoring area, such as urban areas, oceans, or large areas of farmland. The raw synthetic aperture radar echo data is a two-dimensional complex matrix, including azimuth and range dimensions. These two dimensions store echo sampling data, specifically the amplitude and phase information of the target area.
[0028] The embodiments of the present invention are mainly applicable to SAR super-resolution processing. The execution subject of the embodiments of the present invention is a device or equipment capable of image reconstruction of synthetic aperture radar echo data, which can be set on the server side.
[0029] In this embodiment of the invention, low-resolution raw synthetic aperture radar echo data that needs to be super-resolution processed can be obtained from the onboard data link or the ground station.
[0030] Step 102: Adaptively divide the original synthetic aperture radar echo data into multiple blocks using the imaging entropy minimization algorithm.
[0031] There is an overlapping area between two adjacent raw echo data blocks.
[0032] In this embodiment of the invention, due to the enormous size of the original synthetic aperture radar echo data, it is usually impossible to input it into the model all at once. Therefore, it is necessary to divide the large echo matrix into small data blocks of fixed size with overlapping regions. The overlapping regions are used to eliminate edge effects during subsequent stitching. The block size should be determined comprehensively based on hardware computing power (such as GPU memory, CPU memory, etc.), model requirements (such as model receptive field, algorithm characteristics, etc.), and computational overhead. For SAR echo data, the range axis is generally divided into blocks according to the pixel scale after pulse compression, and the azimuth axis is divided into blocks according to the pulse sequence. The data block size should cover at least 4 times the maximum size of the target and not be smaller than the model receptive field. A common block size is range_block = 1024 samples and azimuth_block = 512 pulses, but this size needs to be adjusted according to the actual sampling frequency and the size of the target area. For example, in cases where the target area scene changes rapidly (such as urban areas, area boundaries, etc.), the block size may need to be reduced to range_block = 64 samples and azimuth_block = 64 pulses. For defining the overlapping area, the commonly used overlap ratio range is 25% to 50%. Generally, a 25% overlap ratio is used for situations where the target area scene changes slowly (such as open ocean or large farmland) to improve computational efficiency; while a 50% overlap ratio is used for situations where the target area scene changes rapidly to eliminate edge effects.
[0033] In some embodiments, to ensure that the segmented echo data not only meets the input conditions of the model but also the final imaging accuracy requirements, an imaging entropy minimization algorithm can be used to adaptively segment the original synthetic aperture radar echo data. Based on this, step 102 specifically includes: constructing an image entropy function, as well as memory constraints, non-aliasing constraints, and continuity constraints, and determining the data block size and the overlap size of adjacent data blocks as parameters to be optimized; based on the image entropy function, using a coarse-grid golden section search algorithm to search for the data block size, obtaining multiple candidate values; based on the multiple candidate values and the image entropy function, using a fine-search simplex algorithm to search for the target data block size and the target adjacent data block overlap size that minimize the image entropy while satisfying the memory constraints, the non-aliasing constraints, and the continuity constraints; and adaptively segmenting the original synthetic aperture radar echo data according to the target data block size and the target adjacent data block overlap size to obtain multiple original echo data blocks.
[0034] When searching for data block size using the coarse grid golden section search algorithm, the overlap size of adjacent data blocks is fixed, and multiple estimated values are extracted based on the selectable range of data block size; fast image entropy estimation is performed on the multiple estimated values to obtain the first image entropy corresponding to the multiple estimated values; based on the first image entropy corresponding to the multiple estimated values, the multiple candidate values are selected.
[0035] When searching for the target data block size and the overlap size of adjacent target data blocks using the fine-search simplex algorithm, a simplex is constructed centered on the multiple candidate values; imaging is performed based on the simplex to obtain the image information corresponding to the simplex; based on the image entropy function, as well as the memory constraint, the non-aliasing constraint, and the continuity constraint, the second image entropy corresponding to the image information is calculated; the simplex is updated using the fine-search simplex algorithm, and the simplex update process is repeated until the relative change of the second image entropy for a preset number of consecutive times is less than a preset threshold, at which point the target data block size and the overlap size of adjacent target data blocks are output.
[0036] Specifically, find a quadruple ( m , n , oa , or This minimizes the image entropy of the stitched image after segmented imaging, as minimum image entropy represents the weakest stitching artifacts. For the above quadruple, m and n These represent the number of sampling points in the azimuth and range directions for a single data block, respectively. m and n The size of the data block is determined. oa and orThese represent the number of overlapping sampling points in the azimuth and range directions of two adjacent data blocks, respectively. oa and or This determines the overlap size of adjacent data blocks. The constructed image entropy function is as follows:
[0037]
[0038] in, Represents image entropy, This represents the normalized power of the entire image after segmented imaging. i and j These represent the pixel indices in the orientation and distance directions of the image, respectively. The optimization objective is to minimize image entropy.
[0039] The specific memory constraints for the construction are as follows:
[0040]
[0041] Where M represents the maximum block memory allowed by the hardware.
[0042] The specific non-aliasing constraints are as follows:
[0043]
[0044] in, Q The number of samples corresponding to the synthetic aperture is specifically equal to the ground speed of the SAR sensor flying along the track direction multiplied by the synthetic aperture time and then multiplied by the azimuth sampling frequency. R This represents the number of chirp samples in the range direction, specifically equal to the range-direction frequency modulation bandwidth multiplied by the chirp pulse width and then multiplied by the range-direction sampling frequency.
[0045] The specific continuity constraints are as follows:
[0046]
[0047] Furthermore, a coarse-grid golden section search algorithm is used to search the data block size, i.e., only searching... m and n , oa =0.15 m , or =0.10 n . m The selectable range is [ Q , mmax ], mmax represent m Maximum value m Within the selectable range, one estimate is drawn at intervals of 1.618, such as a total of 8 estimates. n The selectable range is [ R , nmax ], nmax represent n The maximum value, similarly for n Extract several estimated values. Then, for... m and n For each estimated value, 1% of the complex sample points are randomly selected, and a short FFT is used to approximate the image for fast entropy estimation. This yields the first image entropy for each estimated value. Then, based on the first image entropy, the three groups with the smallest entropy are retained. m and n , and use it as a candidate value.
[0048] Furthermore, the fine-search simplex algorithm is used to search for the target data block size and the overlap size of adjacent target data blocks, i.e., using the three retained sets... m and n Centered on a simplex, an image is constructed. During each iteration, an image is formed on the current simplex, and the second image entropy is calculated using the image entropy function. Simultaneously, memory constraints, non-aliasing constraints, and continuity constraints are applied. The simplex is then updated. When the relative change of the second image entropy for a preset number of consecutive iterations is less than a preset threshold, the iteration stops, and the optimal solution is output, which is the target data block size and the overlap size of the target adjacent data blocks. The preset number of iterations can be set according to actual business needs, such as 3 times. The preset threshold is the allowable entropy increment threshold, which is set to 0.3% by default.
[0049] Therefore, the target data block size and the overlap size of adjacent target data blocks can be accurately calculated using the above algorithm. Based on the target data block size and the overlap size of adjacent target data blocks, the original synthetic aperture radar echo data can be adaptively segmented.
[0050] Step 103: Input the multiple original echo data blocks into the preset Schrödinger bridge super-resolution model for super-resolution processing to obtain multiple processed echo data blocks.
[0051] The resolution of the processed echo data block is higher than that of the original echo data block. The preset Schrödinger bridge super-resolution model includes a preset forward correction network and a preset backward correction network.
[0052] In this embodiment of the invention, a forward time step and a backward time step are determined respectively. Then, based on the forward time step and the preset forward correction network, the multiple original echo data blocks are subjected to forward correction processing to obtain multiple intermediate state echo data blocks. Then, based on the backward time step and the preset backward correction network, the multiple intermediate state echo data blocks are subjected to backward correction processing to obtain the multiple processed echo data blocks.
[0053] The Schrödinger Bridge super-resolution model is assumed to be a pre-trained model. Based on the total number of time steps during training, the forward time steps and backward time steps can be determined separately. For example, if 100 time steps are used during training, the forward time steps are 20 steps and the backward time steps are 80 steps.
[0054] Based on the forward stochastic differential equation of the pre-defined Schrödinger bridge super-resolution model, after running forward a certain number of steps (e.g., 20 steps), the forward stochastic differential equation of the pre-defined Schrödinger bridge super-resolution model is as follows:
[0055]
[0056] in, It is standard Brownian motion. The diffusion coefficient is a scalar. t When representing
[0057] Step by step, Represents intermediate state echo data. For drift term, To correct the network forward.
[0058] Then, starting from the intermediate state echo data, stepwise denoising is performed using the backward stochastic differential equation of the preset Schrödinger bridge super-resolution model. The backward stochastic differential equation of the preset Schrödinger bridge super-resolution model is as follows:
[0059]
[0060] in, To correct the network backwards, It is a reverse Brownian motion. The final processed echo data has a higher resolution than the original echo data block.
[0061] This invention utilizes a preset Schrödinger bridge super-resolution model to directly perform super-resolution processing on low-resolution raw synthetic aperture radar echo data, converting the low-resolution raw synthetic aperture radar echo data into high-resolution synthetic aperture radar echo data. This fully utilizes the information in the raw synthetic aperture radar echo data, avoids information loss during the imaging process, and fundamentally improves the resolution of SAR images.
[0062] Step 104: The processed echo data blocks are stitched together and reconstructed to obtain a synthetic aperture radar image.
[0063] In this embodiment of the invention, the generated high-resolution echo data blocks can be spliced and reconstructed according to their original positions. For overlapping areas, a weighted average method is used for smooth fusion to eliminate splicing seams.
[0064] To further improve the seam elimination effect, this embodiment of the invention no longer performs weighted averaging on the overlapping area, but instead performs phase error compensation to eliminate the relative phase jump between two data blocks. Specifically, step 104 includes: determining the overlapping area data between any two adjacent data blocks in the plurality of processed echo data blocks, wherein the arbitrary two adjacent data blocks include a first data block and a second data block; calculating the linear Doppler slope between the overlapping area data of the first data block and the overlapping area data of the second data block; determining the overlapping area image based on the linear Doppler slope, and the first and second data blocks; determining the non-overlapping area image based on the non-overlapping area data in the plurality of processed echo data blocks; and determining the synthetic aperture radar image based on the overlapping area image and the non-overlapping area image.
[0065] When specifically determining the overlapping region image, a phase correction factor is calculated based on the linear Doppler slope; based on the phase correction factor, phase correction is performed on the overlapping region data of the first data block to obtain the corrected overlapping region data corresponding to the first data block; based on the corrected overlapping region data corresponding to the first data block and the overlapping region data corresponding to the second data block, image reconstruction is performed to obtain the overlapping region image.
[0066] Specifically, the first data block and the second data block are two adjacent data blocks, and the overlapping area of the first data block is... Sb The overlapping area data of the second data block is Sa ,estimate Sa and Sb The linear phase error between them can be calculated using a one-dimensional correlation to determine the cross-spectrum, as shown in the following formula.
[0067]
[0068] in, They are cross-spectral vectors. The relative phase difference k This represents the azimuth number. After calculating the relative phase difference... Then, a straight line is fitted using least squares. To obtain the slope That is, the linear Doppler slope.
[0069] Next, a phase correction factor is generated based on the linear Doppler slope. Furthermore, based on the phase correction factor, for Sb Phase correction is performed using the following formula.
[0070]
[0071] in, i Represents the distance sequence number. For the corrected overlapping region data, compare it with... Sa The images are directly stitched along the azimuth direction, and then a short azimuth FFT is performed to obtain the overlapping region images. Finally, the non-overlapping region images and the overlapping region images are combined to form a synthetic aperture radar image. Since the phases are aligned, there are neither amplitude steps nor phase jumps at the stitching points, thus effectively eliminating the stitching seams.
[0072] Furthermore, embodiments of the present invention also provide a training method for a pre-defined Schrödinger bridge super-resolution model, such as... Figure 2 As shown, it includes:
[0073] Step 105: Construct the initial forward correction network and the initial backward correction network.
[0074] In this embodiment of the invention, initial parameters for the forward correction network and the backward correction network are first given.
[0075] Step 106: Based on the initial forward correction network, generate sample intermediate state echo data, and use the sample intermediate state echo data to train the initial backward correction network to obtain the trained backward correction network.
[0076] The embodiments of the present invention employ an alternating iterative training method, first using the generated sample intermediate state echo data to train and correct the parameters in the network.
[0077] Step 107: Using the trained backward correction network, generate backward state echo data, and train the initial forward correction network based on the backward state echo data.
[0078] In this embodiment of the invention, when training the forward correction network and the backward correction network, a loss function is constructed based on the predicted synthetic aperture radar echo data (high resolution) and the real synthetic aperture radar echo data (high resolution) finally output by the backward correction network. Specifically, the mean square error between the predicted image corresponding to the predicted synthetic aperture radar echo data and the real image corresponding to the real synthetic aperture radar echo data is calculated, and the loss function is constructed based on this mean square error.
[0079] To further improve training accuracy, this embodiment of the invention also introduces an auxiliary loss, namely, performing short-time Fourier transforms on the predicted synthetic aperture radar echo data and the actual synthetic aperture radar echo data respectively, to obtain the first complex matrix corresponding to the predicted synthetic aperture radar echo data. And the second complex matrix corresponding to the actual synthetic aperture radar echo data. The specific definition is as follows:
[0080]
[0081] Then based on the first complex matrix Second complex matrix Calculate the spectral convergence loss The specific formula is as follows:
[0082]
[0083] in, This represents the Frobenius norm (the square root of the sum of the squares of all elements of a matrix).
[0084] At the same time, based on the first complex matrix Second complex matrix Calculate the assignment loss of the logarithmic short-time Fourier transform. The specific formula is as follows:
[0085]
[0086] Based on the first complex matrix Second complex matrix Calculate phase loss The specific formula is as follows:
[0087]
[0088] Ultimately, the phase loss Logarithmic short-time Fourier transform assignment loss Spectral convergence loss The mean squared errors between the predicted image and the real image are summed to obtain the loss function. Based on this loss function, the forward correction network and the backward correction network are trained.
[0089] Step 108: Repeat the alternating iterative training process until the preset conditions are met, and then output the preset forward correction network and the preset backward correction network.
[0090] Among them, the preset conditions can be set according to actual business needs, such as reaching a certain number of iterations.
[0091] In this embodiment of the invention, the forward correction network and the backward correction network are continuously trained iteratively until a certain number of iterations are reached, at which point the preset forward correction network and the preset backward correction network are output, thus completing the training of the preset Schrödinger bridge super-resolution model.
[0092] This invention provides a synthetic aperture radar (SAR) image reconstruction method that utilizes a pre-defined Schrödinger bridge super-resolution model to directly perform super-resolution processing on low-resolution raw SAR echo data. This transforms the low-resolution raw SAR echo data into high-resolution SAR echo data before image reconstruction. This fully utilizes the information in the raw SAR echo data, avoiding information loss during the imaging process, thereby fundamentally improving SAR image resolution. Furthermore, due to the large size of the raw SAR echo data, this invention employs an imaging entropy minimization algorithm to adaptively segment the raw SAR echo data into blocks to ensure that the segmented echo data not only meets the model's input conditions but also satisfies the final imaging accuracy requirements.
[0093] Furthermore, as Figure 1 and Figure 2 The specific implementation of the method shown in this embodiment provides a synthetic aperture radar image reconstruction device, such as... Figure 3 As shown, the device includes: an acquisition unit 31, a segmentation unit 32, a super-resolution unit 33, and a reconstruction unit 34.
[0094] The acquisition unit 31 can be used to acquire the raw synthetic aperture radar echo data of the target area.
[0095] The block unit 32 can be used to adaptively block the original synthetic aperture radar echo data using an imaging entropy minimization algorithm to obtain multiple original echo data blocks, wherein there is an overlapping area between two adjacent original echo data blocks.
[0096] The super-resolution unit 33 can be used to input the multiple original echo data blocks into a preset Schrödinger bridge super-resolution model for super-resolution processing to obtain multiple processed echo data blocks, wherein the resolution corresponding to the processed echo data blocks is higher than the resolution corresponding to the original echo data blocks.
[0097] The reconstruction unit 34 can be used to stitch together and reconstruct the multiple processed echo data blocks to obtain a synthetic aperture radar image.
[0098] In some embodiments, the super-resolution unit 33 may be specifically used to determine the forward time step and the backward time step respectively; based on the forward time step and the preset forward correction network, perform forward correction processing on the plurality of original echo data blocks respectively to obtain a plurality of intermediate state echo data blocks; based on the backward time step and the preset backward correction network, perform backward correction processing on the plurality of intermediate state echo data blocks respectively to obtain the plurality of processed echo data blocks.
[0099] In some embodiments, the segmentation unit 32 includes: a construction module, a search module, and a segmentation module.
[0100] The construction module can be used to construct the image entropy function, as well as memory constraints, non-aliasing constraints, and continuity constraints, and to determine the data block size and the overlap size of adjacent data blocks as parameters to be optimized.
[0101] The search module can be used to search for data block sizes based on the image entropy function using a coarse grid golden section search algorithm, and obtain multiple candidate values.
[0102] The search module can also be used to search, based on the multiple candidate values and the image entropy function, using a fine-search simplex algorithm to find the target data block size and the overlap size of the target adjacent data blocks that minimize the image entropy while satisfying the memory constraint, the non-aliasing constraint, and the continuity constraint.
[0103] The block segmentation module can be used to adaptively segment the original synthetic aperture radar echo data according to the target data block size and the overlap size of adjacent target data blocks to obtain multiple original echo data blocks.
[0104] In some embodiments, the search module may be specifically used to fix the overlap size of the adjacent data blocks, extract multiple estimated values based on the selectable range of the data block size; perform fast image entropy estimation on the multiple estimated values respectively to obtain the first image entropy corresponding to the multiple estimated values respectively; and select the multiple candidate values based on the first image entropy corresponding to the multiple estimated values respectively.
[0105] In some embodiments, the search module may further be used to construct a simplex centered on the plurality of candidate values; perform imaging based on the simplex to obtain image information corresponding to the simplex; calculate a second image entropy corresponding to the image information based on the image entropy function, the memory constraint, the non-aliasing constraint, and the continuity constraint; update the simplex using a fine-search simplex algorithm, repeat the simplex update process until the relative change of the second image entropy for a preset number of consecutive times is less than a preset threshold, and output the target data block size and the overlap size of the target adjacent data blocks.
[0106] In some embodiments, the reconstruction unit 34 includes a determination module and a calculation module.
[0107] The determining module can be used to determine the overlapping area data between any two adjacent data blocks in the plurality of processed echo data blocks, wherein the any two adjacent data blocks include a first data block and a second data block.
[0108] The calculation module can be used to calculate the linear Doppler slope between the overlapping region data of the first data block and the overlapping region data of the second data block.
[0109] The determining module can also be used to determine the overlapping region image based on the linear Doppler slope, the first data block, and the second data block;
[0110] The determining module can also be used to determine the non-overlapping region image based on the non-overlapping region data in the plurality of processed echo data blocks.
[0111] The determining module can also be used to determine the synthetic aperture radar image based on the overlapping region image and the non-overlapping region image.
[0112] In some embodiments, the determining module may be specifically used to calculate a phase correction factor based on the linear Doppler slope; perform phase correction on the overlapping region data of the first data block based on the phase correction factor to obtain the corrected overlapping region data corresponding to the first data block; and perform image reconstruction based on the corrected overlapping region data corresponding to the first data block and the overlapping region data corresponding to the second data block to obtain an overlapping region image.
[0113] In some embodiments, the apparatus further includes a construction unit.
[0114] The construction unit can be used to construct an initial forward correction network and an initial backward correction network; based on the initial forward correction network, generate sample intermediate state echo data forward; use the sample intermediate state echo data to train the initial backward correction network to obtain the trained backward correction network; use the trained backward correction network to generate reverse state echo data backward; based on the reverse state echo data, train the initial forward correction network; repeat the alternating iterative training process until a preset condition is met, and then output the preset forward correction network and the preset backward correction network.
[0115] It should be noted that other corresponding descriptions of the functional units involved in the synthetic aperture radar image reconstruction device provided in this embodiment can be found in [reference needed]. Figure 1 and Figure 2 The corresponding description in [the document] will not be repeated here.
[0116] Based on the above, Figure 1 Accordingly, this embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the above-described method. Figure 1 and Figure 2 The synthetic aperture radar image reconstruction method is shown.
[0117] Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause an electronic device (such as personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of this application.
[0118] Based on the above, Figure 1 and Figure 2 The method shown, and Figure 3 To achieve the above objectives, the present application also provides an electronic device, specifically a personal computer, tablet computer, server, or other network device, as shown in the virtual device embodiment. This device includes a storage medium and a processor; the storage medium stores a computer program; the processor executes the computer program to achieve the above-described objectives. Figure 1 and Figure 2 The synthetic aperture radar image reconstruction method is shown.
[0119] Optionally, the aforementioned physical devices may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.
[0120] Those skilled in the art will understand that the physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.
[0121] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.
[0122] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platform, or it can be implemented by hardware.
[0123] This invention utilizes a pre-defined Schrödinger bridge super-resolution model to directly perform super-resolution processing on low-resolution raw synthetic aperture radar (SAR) echo data. This transforms the low-resolution raw SAR echo data into high-resolution SAR echo data before image reconstruction. This fully utilizes the information in the raw SAR echo data, avoiding information loss during the imaging process, thereby fundamentally improving SAR image resolution. Furthermore, due to the large size of the raw SAR echo data, this invention employs an imaging entropy minimization algorithm to adaptively segment the raw SAR echo data into blocks in order to input it into the pre-defined Schrödinger bridge super-resolution model. This ensures that the segmented echo data not only meets the model's input conditions but also satisfies the final imaging accuracy requirements.
[0124] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application. Those skilled in the art will understand that the modules in the apparatus of the embodiment can be distributed within the apparatus of the embodiment as described, or can be modified to be located in one or more apparatuses different from this embodiment. The modules of the above-described embodiment can be combined into one module, or further divided into multiple sub-modules.
[0125] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of any particular implementation scenario. The above disclosures are merely a few specific implementation scenarios of this application; however, this application is not limited thereto, and any variations conceived by those skilled in the art should fall within the protection scope of this application.
Claims
1. A synthetic aperture radar image reconstruction method, characterized by, include: Acquire raw synthetic aperture radar echo data of the target area; An imaging entropy minimization algorithm is used to adaptively divide the original synthetic aperture radar echo data into multiple blocks, in which there is an overlapping area between two adjacent original echo data blocks. The multiple original echo data blocks are respectively input into a preset Schrödinger bridge super-resolution model for super-resolution processing to obtain multiple processed echo data blocks, wherein the resolution corresponding to the processed echo data blocks is higher than the resolution corresponding to the original echo data blocks. The processed echo data blocks are stitched together and reconstructed to obtain a synthetic aperture radar image; The preset Schrödinger bridge super-resolution model includes a preset forward correction network and a preset backward correction network. The multiple original echo data blocks are input into the preset Schrödinger bridge super-resolution model for super-resolution processing to obtain multiple processed echo data blocks, including: Determine the forward time step and the backward time step respectively; Based on the forward time step and the preset forward correction network, the multiple original echo data blocks are respectively subjected to forward correction processing to obtain multiple intermediate state echo data blocks; Based on the backward time step and the preset backward correction network, the multiple intermediate state echo data blocks are respectively subjected to backward correction processing to obtain the multiple processed echo data blocks; The original synthetic aperture radar echo data is adaptively segmented using an imaging entropy minimization algorithm to obtain multiple original echo data blocks, including: Construct an image entropy function, as well as memory constraints, non-aliasing constraints, and continuity constraints, and determine the data block size and the overlap size of adjacent data blocks as parameters to be optimized; Based on the image entropy function, the data block size is searched using the coarse grid golden section search algorithm to obtain multiple candidate values; Based on the multiple candidate values and the image entropy function, the fine search simplex algorithm is used to search for the target data block size and the target adjacent data block overlap size that minimize the image entropy while satisfying the memory constraint, the non-aliasing constraint, and the continuity constraint. Based on the target data block size and the overlap size of adjacent target data blocks, the original synthetic aperture radar echo data is adaptively segmented to obtain multiple original echo data blocks.
2. The method of claim 1, wherein, Based on the image entropy function, the coarse grid golden section search algorithm is used to search for the data block size, resulting in multiple candidate values, including: With the overlap size of the adjacent data blocks fixed, multiple estimated values are extracted based on the selectable range of the data block size; Fast image entropy estimation is performed on the multiple estimated values to obtain the first image entropy corresponding to each of the multiple estimated values; Based on the first image entropy corresponding to the multiple estimated values, the multiple candidate values are selected.
3. The method according to claim 1, characterized in that, The step of searching for the target data block size and the overlap size of adjacent target data blocks that minimize image entropy while satisfying the memory constraint, the no-aliasing constraint, and the continuity constraint, based on the multiple candidate values and the image entropy function, using the fine-search simplex algorithm, includes: Construct a simplex around the multiple candidate values; Imaging is performed based on the simplex to obtain image information corresponding to the simplex; Based on the image entropy function, the memory constraints, the non-aliasing constraints, and the continuity constraints, calculate the second image entropy corresponding to the image information; The simplex is updated using a fine-search simplex algorithm. The simplex update process is repeated until the relative change of the second image entropy for a preset number of consecutive times is less than a preset threshold. Then, the target data block size and the overlap size of the target adjacent data blocks are output.
4. The method according to claim 1, characterized in that, The process of stitching and reconstructing the multiple processed echo data blocks to obtain a synthetic aperture radar image includes: The overlapping region data between any two adjacent data blocks in the plurality of processed echo data blocks is determined, wherein the any two adjacent data blocks include a first data block and a second data block; Calculate the linear Doppler slope between the overlapping region data of the first data block and the overlapping region data of the second data block; Based on the linear Doppler slope, and the first data block and the second data block, the overlapping region image is determined; The non-overlapping region image is determined based on the non-overlapping region data in the plurality of processed echo data blocks; The synthetic aperture radar image is determined based on the overlapping region image and the non-overlapping region image.
5. The method according to claim 4, characterized in that, Determining the overlapping region image based on the linear Doppler slope, the first data block, and the second data block includes: Calculate the phase correction factor based on the linear Doppler slope; Based on the phase correction factor, phase correction is performed on the overlapping region data of the first data block to obtain the corrected overlapping region data corresponding to the first data block. Based on the corrected overlapping region data corresponding to the first data block and the overlapping region data corresponding to the second data block, image reconstruction is performed to obtain the overlapping region image.
6. The method according to claim 1, characterized in that, Before inputting the plurality of original echo data blocks into a preset Schrödinger bridge super-resolution model for super-resolution processing to obtain a plurality of processed echo data blocks, the method further includes: Construct the initial forward correction network and the initial backward correction network; Based on the initial forward correction network, intermediate state echo data of the samples are generated forward. Using the intermediate state echo data of the sample, the initial backward correction network is trained to obtain the trained backward correction network; Using the trained backward correction network, reverse state echo data is generated backward; The initial forward correction network is trained based on the reverse state echo data; Repeat the alternating iterative training process until the preset conditions are met, and then output the preset forward correction network and the preset backward correction network.
7. A synthetic aperture radar image reconstruction device, characterized in that, include: The acquisition unit is used to acquire raw synthetic aperture radar echo data of the target area. The block unit is used to adaptively block the original synthetic aperture radar echo data using an imaging entropy minimization algorithm to obtain multiple original echo data blocks, wherein there is an overlapping area between two adjacent original echo data blocks. The super-resolution unit is used to input the multiple original echo data blocks into a preset Schrödinger bridge super-resolution model for super-resolution processing to obtain multiple processed echo data blocks. The resolution of the processed echo data blocks is higher than the resolution of the original echo data blocks. The preset Schrödinger bridge super-resolution model includes a preset forward correction network and a preset backward correction network. The reconstruction unit is used to stitch together and reconstruct the multiple processed echo data blocks to obtain a synthetic aperture radar image. The super-resolution unit is specifically used to determine the forward time step and the backward time step respectively; based on the forward time step and the preset forward correction network, the multiple original echo data blocks are subjected to forward correction processing to obtain multiple intermediate state echo data blocks; based on the backward time step and the preset backward correction network, the multiple intermediate state echo data blocks are subjected to backward correction processing to obtain multiple processed echo data blocks. The segmentation unit is specifically used to construct an image entropy function, as well as memory constraints, non-aliasing constraints, and continuity constraints, and to determine the data block size and the overlap size of adjacent data blocks as parameters to be optimized. Based on the image entropy function, a coarse-grid golden section search algorithm is used to search for the data block size, obtaining multiple candidate values. Based on the multiple candidate values and the image entropy function, a fine-search simplex algorithm is used to search for the target data block size and the target adjacent data block overlap size that minimize the image entropy while satisfying the memory constraints, the non-aliasing constraints, and the continuity constraints. According to the target data block size and the target adjacent data block overlap size, the original synthetic aperture radar echo data is adaptively segmented to obtain multiple original echo data blocks.
8. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.
9. An electronic device, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 6.