A sub-aperture decomposition based speckle noise suppression method for SAR images

Abstract

The application discloses a SAR image speckle noise suppression method based on sub-aperture decomposition and belongs to the field of radar image processing. The method comprises the following steps: performing sub-aperture segmentation processing and modulus quantization processing on single-view complex image data of a SAR image respectively, and sequentially obtaining sub-aperture images and a full-aperture image of the single-view complex image data; performing logarithmic transformation on an original three-dimensional tensor composed of the sub-aperture images and the full-aperture image to generate a noisy three-dimensional tensor; establishing a denoising objective function model containing a global low-rank regularization term, a non-local denoising regularization term and an edge preserving regularization term according to the noisy three-dimensional tensor, and performing iterative updating on the denoising objective function model until a preset maximum iteration number is met, and finally outputting a final denoised image with suppressed speckle noise. The application can improve the imaging quality of a SAR image after denoising processing.

Description

Technical Field

[0001] This invention relates to the field of radar image processing technology, and in particular to a method for suppressing speckle noise in SAR images based on sub-aperture decomposition. Background Technology

[0002] Synthetic Aperture Radar (SAR), as an important active microwave imaging sensor, plays a crucial role in numerous fields such as environmental monitoring, geological mapping, and ocean observation due to its all-weather, all-day operation. However, the unique coherent imaging mechanism of SAR inevitably leads to the interference of speckle noise in its images. Speckle noise is a multiplicative noise, fundamentally different from the Gaussian additive noise in optical images. It severely degrades the quality of SAR images, significantly reducing their interpretability and greatly hindering subsequent tasks such as target detection, recognition, and classification.

[0003] While existing processing methods have achieved some success in suppressing speckle noise in SAR images, they have failed to fully explore and utilize the auxiliary noise reduction function of sub-aperture images generated from SAR images. This results in low quality of the generated SAR images when suppressing speckle noise.

[0004] Therefore, there is an urgent need for a SAR image speckle noise suppression method based on sub-aperture decomposition to solve the above-mentioned technical problems. Summary of the Invention

[0005] This invention provides a SAR image speckle noise suppression method based on sub-aperture decomposition, which can improve the imaging quality of SAR images after speckle noise suppression processing. The technical solution is as follows:

[0006] On the one hand, a method for suppressing speckle noise in SAR images based on sub-aperture decomposition is provided, the method comprising:

[0007] The single-view complex image data of the SAR image are subjected to sub-aperture segmentation and modulus quantization processing respectively to obtain the sub-aperture image and full-aperture image of the single-view complex image data in sequence;

[0008] A logarithmic transformation is performed on the original three-dimensional tensor composed of the sub-aperture image and the full aperture image to generate a noisy three-dimensional tensor.

[0009] Based on the noisy 3D tensor, a denoising objective function model is established, which includes a global low-rank regularization term, a nonlocal denoising regularization term, and an edge-preserving regularization term. The denoising objective function model is iteratively updated until the preset maximum number of iterations is met, and the final denoised image with suppressed speckle noise is output.

[0010] On the other hand, a SAR image speckle noise suppression device based on sub-aperture decomposition is provided, the device comprising:

[0011] The processing module is used to perform sub-aperture segmentation and modulus quantization processing on the single-view complex image data of the SAR image, so as to obtain the sub-aperture image and full-aperture image of the single-view complex image data in sequence.

[0012] The generation module performs a logarithmic transformation on the original three-dimensional tensor composed of the sub-aperture image and the full aperture image to generate a noisy three-dimensional tensor.

[0013] The update module is used to establish a denoising objective function model containing a global low-rank regularization term, a non-local denoising regularization term, and an edge-preserving regularization term based on the noisy three-dimensional tensor, and to iteratively update the denoising objective function model until the preset maximum number of iterations is met, and output the final denoised image with suppressed speckle noise.

[0014] On the other hand, a computer device is provided, the computer device including a memory and a processor, the memory for storing a computer program, and the processor for executing the computer program stored in the memory to implement the steps of the SAR image speckle noise suppression method based on sub-aperture decomposition described above.

[0015] On the other hand, a computer-readable storage medium is provided, wherein a computer program is stored therein, and when the computer program is executed by a processor, it implements the steps of the SAR image speckle noise suppression method based on sub-aperture decomposition described above.

[0016] On the other hand, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the SAR image speckle noise suppression method based on sub-aperture decomposition described above.

[0017] The technical solution provided by this invention can bring at least the following beneficial effects: First, the single-view complex image data of the SAR image is segmented into multiple sub-aperture images through sub-aperture segmentation. Then, a full-aperture image is obtained through modulo and quantization. The full-aperture image and the multiple sub-aperture images are stacked into a set of three-dimensional tensors. Based on a preset denoising objective function model, the transformed three-dimensional tensor is iteratively updated sequentially, including global low-rank regularization, non-local denoising regularization, and edge-preserving regularization, to obtain the final denoised image. This method not only efficiently suppresses speckle noise but also better preserves image detail features, effectively improving the quality of SAR images, thereby strongly promoting the in-depth application and continuous development of SAR technology in various fields. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a SAR image speckle noise suppression method based on sub-aperture decomposition provided in an embodiment of the present invention;

[0020] Figure 2 This is a flowchart of a sub-aperture decomposition method provided in an embodiment of the present invention;

[0021] Figure 3 This is a schematic diagram of a Hamming window truncated according to an embodiment of the present invention;

[0022] Figure 4 This is a SAR image with significant speckle noise provided in an embodiment of the present invention;

[0023] Figure 5 This is a SAR image after noise reduction processing provided in an embodiment of the present invention;

[0024] Figure 6 This is a structural diagram of a SAR image speckle noise suppression device based on sub-aperture decomposition provided in an embodiment of the present invention;

[0025] Figure 7 This is a hardware architecture diagram of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0027] As mentioned earlier, while existing speckle suppression methods can achieve certain results, these methods still have various limitations and fail to fully utilize the imaging characteristics of SAR images.

[0028] Based on this, the concept of the present invention is to use the sub-aperture image obtained by sub-aperture decomposition as additional information to assist the speckle noise suppression task of SAR image. This not only effectively suppresses speckle noise, but also preserves image detail features, thereby significantly improving the quality of SAR image.

[0029] The following describes the specific implementation of the above concept.

[0030] Please refer to Figure 1 The present invention provides a SAR image speckle noise suppression method based on sub-aperture decomposition, the method comprising:

[0031] Step 100: Perform sub-aperture segmentation and modulus quantization processing on the single-view complex image data of the SAR image to obtain the sub-aperture image and full-aperture image of the single-view complex image data in sequence.

[0032] Step 102: Perform a logarithmic transformation on the original three-dimensional tensor composed of the sub-aperture image and the full aperture image to generate a noisy three-dimensional tensor;

[0033] Step 104: Based on the noisy 3D tensor, establish a denoising objective function model including a global low-rank regularization term, a non-local denoising regularization term, and an edge-preserving regularization term, and iteratively update the denoising objective function model until the preset maximum number of iterations is met, and output the final denoised image that suppresses speckle noise.

[0034] In this embodiment of the invention, the single-view complex image data is first segmented into multiple sub-aperture images through sub-aperture segmentation. A full-aperture image is then obtained through modulo and quantization. The full-aperture image and the multiple sub-aperture images are stacked into a set of three-dimensional tensors. Based on a preset denoising objective function model, the transformed three-dimensional tensor is iteratively updated, including global low-rank regularization, non-local denoising regularization, and edge-preserving regularization, to obtain the final denoised image. This method not only efficiently suppresses speckle noise but also better preserves image detail features, effectively improving SAR image quality and thus powerfully promoting the in-depth application and continuous development of SAR technology in various fields.

[0035] The following description Figure 1 The execution method for each step is shown.

[0036] First, for step 100, the single-view complex image data of the SAR image is subjected to sub-aperture segmentation and modulus quantization processing respectively, to obtain the sub-aperture image and full-aperture image of the single-view complex image data in sequence.

[0037] Considering that sub-aperture images obtained by sub-aperture decomposition of single-view complex image data of SAR images can serve as additional information to assist in speckle noise suppression of SAR images, this embodiment of the invention decomposes them using the following method: Fourier transform is performed on the single-view complex image data along the azimuth direction to obtain a frequency domain signal converted from the spatial domain to the range-Doppler domain; energy compensation processing is performed on the spectrum of the frequency domain signal according to the energy distribution in the azimuth direction of the range-Doppler domain, so that the non-edge regions of the spectrum have the same energy; the compensated spectrum is equally divided according to a preset number of sub-aperture images to obtain multiple sub-aperture bands; wherein adjacent sub-apertures overlap each other, and the overlap bandwidth is half of the sub-aperture bandwidth; sidelobe suppression processing is performed on all sub-aperture bands according to a preset Hamming window, and the processing result is subjected to inverse Fourier transform to quantize and obtain the sub-aperture image of the single-view complex image data.

[0038] Specifically, firstly, a Fourier transform is performed on the single-view complex image data along the azimuth direction to convert it to the range-Doppler domain. This transformation is performed using the Discrete Fourier Transform (DFT) formula:

[0039]

[0040] Where f(n) is the original time-domain signal, F(u) is the transformed frequency-domain signal, N is the number of azimuth points, u is the frequency index, and n is the time-domain index, realizing the conversion from the spatial domain to the range-Doppler domain.

[0041] Next, based on the energy distribution in the azimuth direction in the range-Doppler domain, energy compensation is performed on the spectrum to ensure that the energy in the non-edge regions of the azimuth spectrum remains basically consistent.

[0042] For details, please refer to the following: Figure 2 and Figure 3Based on a set number of sub-aperture images *x*, the Doppler spectrum is divided into *x* sub-bands, with each adjacent sub-band overlapping by 50% of the sub-aperture bandwidth. This configuration increases the number of sub-aperture images to acquire more information while reducing the correlation between sub-apertures. Experimental analysis shows a non-linear relationship between the number of sub-apertures *N* and denoising performance. As the number of sub-apertures increases, the relative resolution of the denoised SAR image initially increases and then decreases. Multiple comparative experiments were conducted, processing multiple SAR images using different numbers of sub-apertures. In each experiment, other parameters were kept constant, only the number of sub-apertures was changed. The denoising effect and image resolution preservation under different numbers of sub-apertures were evaluated by calculating the image resolution and equivalent number of views (ENL) of the processed images. Analysis of multiple sets of experimental data revealed that a sub-aperture number of 5 effectively suppresses speckle noise while preserving image resolution well, exhibiting the best overall performance.

[0043] Next, other frequency bands are truncated and set to zero using a Hamming window to obtain range-Doppler domain data for different sub-apertures. The Hamming window truncation operation can effectively suppress spectral sidelobes and reduce interference between sub-apertures.

[0044] Finally, the processed range-Doppler domain data is converted back to the spatial domain by inverse Fourier transform, and then several sub-aperture images are obtained by image quantization.

[0045] In this embodiment of the invention, the full aperture image is obtained by taking the modulus and quantizing the single-view complex image data. The specific process of this operation is well known to those skilled in the art and will not be described in detail here.

[0046] For step 102, a logarithmic transformation is performed on the original three-dimensional tensor composed of the sub-aperture image and the full aperture image to generate a noisy three-dimensional tensor.

[0047] In this embodiment of the invention, the original three-dimensional tensor composed of sub-aperture images and full-aperture images is first subjected to a logarithmic transformation. This transformation converts the speckle noise of the image from multiplicative noise to additive noise. The transformation formula is as follows:

[0048]

[0049] Where A is the original three-dimensional tensor, Y is the tensor after logarithmic transformation, M is the noise-free three-dimensional tensor under ideal conditions, and X is the tensor of M after logarithmic transformation. N represents noise.

[0050] For step 104, a denoising objective function model containing a global low-rank regularization term, a non-local denoising regularization term, and an edge-preserving regularization term is established based on the noisy three-dimensional tensor. The denoising objective function model is iteratively updated until the preset maximum number of iterations is met, and the final denoised image with suppressed speckle noise is output.

[0051] Based on the global low-rank characteristics of sub-aperture and full-aperture images, a denoising objective function model is established, which successively includes a global low-rank regularization term, a nonlocal denoising regularization term, and an edge-preserving regularization term:

[0052]

[0053] Where F represents the global low-rank regularization term; NL represents the nonlocal denoising regularization term; S represents the edge-preserving regularization term; and Y is the noisy three-dimensional tensor after logarithmic transformation. For the denoising and dimensionality reduction tensor, B is an orthogonal basis. and B * λ1 and λ2 are the coefficients of the nonlocal denoising regularization term and the edge-preserving regularization term, respectively, representing the optimal solution of the denoising objective function model.

[0054] For this denoising objective function model, the three sub-expressions are not calculated by addition, but each sub-expression is a constraint condition that jointly controls and adjusts the optimal solution of the model. In other words, to obtain the optimal denoising tensor, it needs to be processed by these three sub-expressions to ensure the best quality of the denoised image.

[0055] In this embodiment of the invention, the iterative update process includes the following steps: S1, performing singular value decomposition calculation on the noisy 3D tensor using modulo 3 expansion to obtain a dimensionality-reduced tensor and an orthogonal basis to satisfy the global low-rank regularization constraint after relaxation processing; S2, performing nonlocal denoising processing on the dimensionality-reduced tensor to obtain a denoised dimensionality-reduced tensor, and performing dimensionality-up processing on the denoised dimensionality-reduced tensor according to the orthogonal basis to obtain a denoised 3D tensor to satisfy the nonlocal denoising regularization constraint; S3, performing edge preservation processing on the denoised 3D tensor and the noisy 3D tensor to obtain an updated 3D tensor to satisfy the edge preservation regularization constraint; S4, repeating the iterative update of steps S1-S3 with the updated 3D tensor as the noisy 3D tensor, and quantizing the denoised 3D tensor obtained in the last round of iterative update to obtain the final denoised image.

[0056] Specifically, step S1 is used to perform global low-rank regularization on the tensor. There is strong global spatial similarity between the full-aperture and sub-aperture images of SAR. Despite the influence of speckle noise, the three-dimensional tensor they constitute exhibits low-rank characteristics overall. This low-rank property means that data in a high-dimensional space can be approximated by a lower-dimensional subspace. In the absence of noise, an ideal SAR image should have an even lower rank, but the presence of noise increases the rank.

[0057] Therefore, the objective function for the global low-rank regularization term can be established as follows:

[0058]

[0059] To optimize the objective function, and considering the reduction in noise level after iterative regularization, it is relaxed as follows:

[0060]

[0061] in ×3 This means performing matrix multiplication on the third dimension of the tensor to achieve dimensional transformation; I is the identity matrix.

[0062] Singular value decomposition is performed on the modulo-3 expansion of Y, i.e., (Y) (3) =USV T Let the dimension after dimensionality reduction be dim. Then, select the column number of dim corresponding to V to construct B, i.e., B = V(:, 1: dim). Further calculations yield... The dimension is M×N×dim, which simplifies complex optimization problems into singular value decomposition, thus simplifying the calculation.

[0063] It is worth noting that the dimensionality reduction mentioned above is relative to the dimension of the original tensor. For example, initially, five sub-aperture images and one full-aperture image are combined to form an original three-dimensional tensor Y (M×N×6). After dimensionality reduction transformation, we obtain a dimensionality-reduced tensor. And an orthogonal basis B (3×6).

[0064] Furthermore, the main function of the nonlocal denoising regularization term is to remove noise. Nonlocal denoising is performed on the dimensionality-reduced tensor and its corresponding orthogonal basis obtained after the above processing, including the following steps: dividing the dimensionality-reduced image corresponding to the dimensionality-reduced tensor into blocks to obtain multiple image blocks; calculating the similarity (e.g., Euclidean distance) between each image block and other image blocks, and assigning the image blocks to different image groups based on the calculation results, for example, grouping image blocks with high similarity into one group to obtain different image block groups; weighting the similar image blocks in each nonlocal group, and performing kernel norm minimization calculation on the weighted image block matrix to obtain the denoised dimensionality-reduced image and its corresponding denoised dimensionality-reduced tensor.

[0065] Specifically, in this embodiment, the denoising process employs the Weighted Nuclear Norm Minimization (WNNM) method. This method achieves denoising by weighting similar image patches and then minimizing the nuclear norm of the weighted image patch matrix. Its principle is based on the similarity of similar image patches in the low-rank structure, which can effectively remove noise.

[0066] Furthermore, the edge-preserving regularization term is implemented after nonlocal denoising to preserve edge information, prevent over-smoothing, and retain a certain degree of strong scattering points. Specifically, it first assigns different weight values ​​based on the distribution of neighboring pixel values ​​in the denoised image, resulting in an edge-like image w1(i,j):

[0067]

[0068] Where I0 represents the noisy 3D image, and I1 represents the denoised 3D image; (i,j) are the coordinates of the pixel in the image; w0 is a temporary variable before the edge-like image truncation and normalization process; the value of w1 is between 0 and 1, which is equivalent to a weight with the same size as the entire image; S represents a window region centered on the pixel I1(i,j). is the average pixel value within region S, h is the attenuation coefficient, and T is the truncation threshold.

[0069] Next, regularization calculations are performed on the denoised 3D image, the edge-class image, and the noisy 3D image to obtain the updated 3D tensor corresponding to the updated image.

[0070]

[0071] Therefore, after processing the above three sub-formulas, we can obtain the updated three-dimensional tensor of one iteration. Then, we use the updated three-dimensional tensor of this round as input to repeat the above processing process until the preset number of iterations is met. The default setting is three iterations. In the last iteration, the denoised and dimensionality-reduced tensor obtained by non-local denoising is output as the optimal solution. After dimensionality-upgrading and summing, we obtain the final denoised image that suppresses speckle noise.

[0072] The noise reduction effect of the method proposed in the embodiments of the present invention is as follows: Figure 4 and Figure 5 As shown, where Figure 4 This is a SAR image with significant speckle noise. Figure 5 The images are SAR images after noise reduction using this method. Comparing the two, it can be seen that the image with this method has significantly higher clarity and quality than the unprocessed image, and the detailed features of the image are not lost after processing. This demonstrates the effectiveness and practicality of this method.

[0073] Please refer to Figure 6 This invention provides a SAR image speckle noise suppression device based on sub-aperture decomposition, the device comprising:

[0074] The processing module 600 is used to perform sub-aperture segmentation and modulus quantization processing on the single-view complex image data of the SAR image, so as to obtain the sub-aperture image and the full-aperture image of the single-view complex image data in sequence.

[0075] The generation module 602 is used to perform a logarithmic transformation on the original three-dimensional tensor composed of the sub-aperture image and the full aperture image to generate a noisy three-dimensional tensor.

[0076] The update module 604 is used to establish a denoising objective function model containing a global low-rank regularization term, a non-local denoising regularization term, and an edge-preserving regularization term based on the noisy three-dimensional tensor, and to iteratively update the denoising objective function model until the preset maximum number of iterations is met, and output the final denoised image with suppressed speckle noise.

[0077] In this embodiment of the invention, when the processing module 600 performs sub-aperture segmentation processing on the single-view complex image data to obtain the sub-aperture image of the single-view complex image data, it specifically performs the following operations:

[0078] The single-view complex image data is subjected to Fourier transform along the azimuth direction to obtain a frequency domain signal converted from the spatial domain to the range-Doppler domain;

[0079] The frequency domain signal spectrum is subjected to energy compensation processing based on the energy distribution of the azimuth direction in the range-Doppler domain, so that the non-edge regions of the spectrum have the same energy.

[0080] The compensated spectrum is divided equally according to the preset number of sub-aperture images to obtain multiple sub-aperture bands; wherein, adjacent sub-apertures overlap each other, and the overlap bandwidth is half of the sub-aperture bandwidth.

[0081] Sidelobe suppression processing is performed on all the sub-aperture bands according to the preset Hamming window, and the processing result is subjected to inverse Fourier transform to quantize and obtain the sub-aperture image of the single-view complex image data.

[0082] In this embodiment of the invention, when the update module 604 iteratively updates the denoising objective function model until a preset maximum number of iterations is met, and outputs the final denoised image with suppressed speckle noise, it specifically performs the following operations:

[0083] S1. Perform singular value decomposition calculation on the noisy three-dimensional tensor with modular three expansion to obtain the dimension-reduced tensor and orthogonal basis, so as to satisfy the global low-rank regularization term constraint after relaxation.

[0084] S2. Perform nonlocal denoising on the reduced-dimensional tensor to obtain a denoised reduced-dimensional tensor, and perform dimensionality-increasing on the denoised reduced-dimensional tensor according to the orthogonal basis to obtain a denoised three-dimensional tensor, so as to satisfy the nonlocal denoising regularization term constraint.

[0085] S3. Perform edge preservation processing on the denoised 3D tensor and the noisy 3D tensor to obtain an updated 3D tensor to satisfy the edge preservation regularization constraint.

[0086] S4. Repeat steps S1-S3 iteratively update the updated 3D tensor as a noisy 3D tensor, and quantize the denoised 3D tensor obtained in the last round of iterative update to obtain the final denoised image.

[0087] In this embodiment of the invention, the noise reduction objective function model is established using the following formula:

[0088]

[0089] Where F represents the global low-rank regularization term; NL represents the nonlocal denoising regularization term; S represents the edge-preserving regularization term; and Y is the noisy three-dimensional tensor after logarithmic transformation. For the denoising and dimensionality reduction tensor, B is an orthogonal basis. and B * λ1 and λ2 are the coefficients of the nonlocal denoising regularization term and the edge-preserving regularization term, respectively, representing the optimal solution of the denoising objective function model.

[0090] In this embodiment of the invention, when the update module 604 performs nonlocal denoising processing on the dimensionality-reduced tensor to obtain a denoised dimensionality-reduced tensor, it specifically performs the following operations:

[0091] The reduced image corresponding to the reduced tensor is divided into blocks to obtain multiple image blocks;

[0092] Calculate the similarity of each image patch to other image patches, and classify the image patches into different image groups based on the calculation results;

[0093] The similar image blocks in each non-local group are weighted, and the kernel norm of the weighted image block matrix is ​​minimized to obtain the denoised and dimensionality-reduced image and its corresponding denoised and dimensionality-reduced tensor.

[0094] In this embodiment of the invention, when the update module 604 performs edge-preserving processing on the denoised 3D tensor and the noisy 3D tensor to obtain an updated 3D tensor, it specifically performs the following operations:

[0095] Obtain the distribution of neighborhood pixel values ​​of the pixels in the denoised 3D image corresponding to the denoised 3D tensor;

[0096] Based on the preset weight values ​​of each neighboring pixel value, the class edge image w1(i,j) is calculated:

[0097]

[0098] Where (i,j) are the coordinates of the pixel in the image, w0 is the temporary variable before the edge image truncation normalization process, and T is the preset truncation threshold.

[0099] Based on the denoised 3D image, the edge-class image, and the noisy 3D tensor corresponding to the noisy 3D tensor, the updated 3D tensor corresponding to the updated image is calculated.

[0100]

[0101] Where I0 represents a noisy 3D image and I1 represents a denoised 3D image.

[0102] It should be noted that the SAR image speckle noise suppression device based on sub-aperture decomposition provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the SAR image speckle noise suppression device based on sub-aperture decomposition provided in the above embodiments and the SAR image speckle noise suppression method based on sub-aperture decomposition are based on the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0103] Embodiments of this application also provide a computer device, please refer to... Figure 7 The computer device includes a processor and a memory, the memory storing at least one instruction, at least one program, code set or instruction set, the at least one instruction, at least one program, code set or instruction set being loaded and executed by the processor to implement the SAR image speckle noise suppression method based on sub-aperture decomposition provided in the above method embodiments.

[0104] Embodiments of this application also provide a computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement the SAR image speckle noise suppression method based on sub-aperture decomposition provided in the above-described method embodiments.

[0105] Embodiments of this application also provide a computer program product, which includes a computer program. A processor of a computer device reads the computer program from a computer-readable storage medium and executes the computer program, causing the computer device to perform any of the sub-aperture decomposition-based SAR image speckle noise suppression methods described in the above embodiments.

[0106] For ease of description, the above systems or devices are described separately as various modules or units based on their functions. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware components.

[0107] As can be seen from 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 platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0108] Finally, it should be noted that in this document, relational terms such as first, second, third, and fourth are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0109] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for suppressing speckle noise in SAR images based on sub-aperture decomposition, characterized in that, The method includes: The single-view complex image data of the SAR image are subjected to sub-aperture segmentation and modulus quantization processing respectively to obtain the sub-aperture image and full-aperture image of the single-view complex image data in sequence; A logarithmic transformation is performed on the original three-dimensional tensor composed of the sub-aperture image and the full aperture image to generate a noisy three-dimensional tensor. Based on the noisy 3D tensor, a denoising objective function model is established, including a global low-rank regularization term, a nonlocal denoising regularization term, and an edge-preserving regularization term. The denoising objective function model is iteratively updated until a preset maximum number of iterations is met, outputting the final denoised image with suppressed speckle noise, including: S1. Perform singular value decomposition calculation on the noisy three-dimensional tensor with modular three expansion to obtain the dimension-reduced tensor and orthogonal basis, so as to satisfy the global low-rank regularization term constraint after relaxation. S2. Perform nonlocal denoising on the reduced-dimensional tensor to obtain a denoised reduced-dimensional tensor, and perform dimensionality-increasing on the denoised reduced-dimensional tensor according to the orthogonal basis to obtain a denoised three-dimensional tensor, so as to satisfy the nonlocal denoising regularization term constraint. S3. Perform edge preservation processing on the denoised 3D tensor and the noisy 3D tensor to obtain an updated 3D tensor to satisfy the edge preservation regularization term constraint. S4. Repeat steps S1-S3 iteratively update the updated three-dimensional tensor as a noisy three-dimensional tensor, and quantize the denoised three-dimensional tensor obtained in the last round of iterative update to obtain the final denoised image. The noise reduction objective function model is established using the following formula: in, F Represents a global low-rank regularization term; NL This represents a nonlocal denoising regularization term. S This indicates a regularization term that preserves the edges; The noisy three-dimensional tensor after logarithmic transformation. To reduce noise and dimension in the tensor, It is an orthogonal basis. and This represents the optimal solution to the noise reduction objective function model. and These are the coefficients of the nonlocal noise reduction regularization term and the edge preservation regularization term, respectively.

2. The method as described in claim 1, characterized in that, Sub-aperture segmentation is performed on single-view complex image data to obtain sub-aperture images of the single-view complex image data, including: The single-view complex image data is subjected to Fourier transform along the azimuth direction to obtain a frequency domain signal converted from the spatial domain to the range-Doppler domain; The frequency domain signal spectrum is subjected to energy compensation processing based on the energy distribution of the azimuth direction in the range-Doppler domain, so that the non-edge regions of the spectrum have the same energy. The compensated spectrum is divided equally according to the preset number of sub-aperture images to obtain multiple sub-aperture bands; wherein, adjacent sub-apertures overlap each other, and the overlap bandwidth is half of the sub-aperture bandwidth. Sidelobe suppression processing is performed on all the sub-aperture bands according to the preset Hamming window, and the processing result is subjected to inverse Fourier transform to quantize and obtain the sub-aperture image of the single-view complex image data.

3. The method as described in claim 1, characterized in that, The nonlocal denoising process performed on the reduced-dimensional tensor to obtain a denoised reduced-dimensional tensor includes: The reduced image corresponding to the reduced tensor is divided into blocks to obtain multiple image blocks; Calculate the similarity of each image patch to other image patches, and classify the image patches into different image groups based on the calculation results; The similar image blocks in each non-local group are weighted, and the kernel norm of the weighted image block matrix is ​​minimized to obtain the denoised and dimensionality-reduced image and its corresponding denoised and dimensionality-reduced tensor.

4. The method as described in claim 1, characterized in that, The step of performing edge-preserving processing on the denoised 3D tensor and the noisy 3D tensor to obtain an updated 3D tensor includes: Obtain the distribution of neighborhood pixel values ​​of the pixels in the denoised 3D image corresponding to the denoised 3D tensor; Based on the preset weight value of each neighboring pixel value, the edge image is calculated. : in,( i , j () represents the coordinates of the pixel in the image. w 0 is a temporary variable before the image truncation and normalization process; T is a preset truncation threshold. Based on the denoised 3D image, the edge-class image, and the noisy 3D tensor corresponding to the noisy 3D tensor, the updated 3D tensor corresponding to the updated image is calculated. : in, Represents a noisy 3D image. This represents a denoised 3D image.

5. A SAR image speckle noise suppression device based on sub-aperture decomposition, characterized in that, The apparatus, used in the method of any one of claims 1-4, comprises: The processing module is used to perform sub-aperture segmentation and modulus quantization processing on the single-view complex image data of the SAR image, so as to obtain the sub-aperture image and full-aperture image of the single-view complex image data in sequence. The generation module performs a logarithmic transformation on the original three-dimensional tensor composed of the sub-aperture image and the full aperture image to generate a noisy three-dimensional tensor. The update module is used to establish a denoising objective function model containing a global low-rank regularization term, a non-local denoising regularization term, and an edge-preserving regularization term based on the noisy three-dimensional tensor, and to iteratively update the denoising objective function model until the preset maximum number of iterations is met, and output the final denoised image with suppressed speckle noise.

6. A computer device, characterized in that, The computer device includes a memory and a processor. The memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory to implement the steps of the method according to any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the steps of the method described in any one of claims 1-4.

8. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1-4.

Images