A method for coherent speckle suppression of synthetic aperture radar image based on diffusion sampling

Abstract

The application discloses a synthetic aperture radar image speckle suppression method based on diffusion sampling, and relates to the field of synthetic aperture radar image processing. The method of the application is based on a sampling scheme of an approximate score function, reverses a transfer distribution, and applies a multiplicative noise model, thereby solving the problem of multiplicative noise degradation in a synthetic aperture radar image and avoiding supervised training and logarithmic transformation of a SAR image.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of synthetic aperture radar image processing, and particularly relates to a synthetic aperture radar image speckle suppression method based on diffusion sampling. BACKGROUND

[0002] The processing and interpretation of a synthetic aperture radar (SAR) image is an ill-posed inverse problem, and when the SAR image is processed, appropriate prior conditions need to be added to the clear SAR image. The diffusion model is widely used in image processing inverse problems such as image deblurring, image restoration and image super-resolution due to its powerful deep generative prior. However, the image degradation solved by these inverse problems is additive in nature, and the use of SAR image speckle suppression such as multiplicative noise degradation is still in its infancy.

[0003] SAR-DDPM (SAR-Denoising Diffusion Probabilistic Model) uses optical images and synthetic speckle images to train a noise predictor network, but since the speckle noise is diverse, the noise predictor trained initially cannot fully simulate the diverse speckle noise. In addition, when the intensity or type of speckle noise changes slightly, SAR-DDPM needs to be retrained with supervision. In order to solve these problems, subsequent methods use a logarithmic transformation on the speckle noise image and use a Gaussian distribution instead of the speckle noise distribution in the logarithmic domain. Such transformations and approximations are relatively rough. SUMMARY

[0004] The purpose of the present application is to provide a synthetic aperture radar image speckle suppression method based on diffusion sampling, to solve the problem of multiplicative noise degradation in a synthetic aperture radar image, and to avoid supervised training and logarithmic transformation of the SAR image.

[0005] To achieve the above purpose, the present application provides the following solutions.

[0006] In a first aspect, the present application provides a synthetic aperture radar image speckle suppression method based on diffusion sampling, comprising:

[0007] initializing a diffusion noise image at the Tth time step of the diffusion inverse process, and initializing the time step t=T, T being the total time step of the diffusion inverse process;

[0008] According to the diffusion noise image at the tth time step, an unconditional deep neural network is used to calculate the prior state estimation at the t-1th time step;

[0009] According to the observation image and the diffusion noise image of the t-th time step, a likelihood score function value of the t-th time step is estimated by using a multiplicative noise model;

[0010] According to the prior state estimation of the t-1-th time step and the likelihood score function value of the t-th time step, a diffusion noise image of the t-1-th time step is calculated;

[0011] Let the value of t be reduced by 1, return to the step of "calculating the prior state estimation of the t-1-th time step according to the diffusion noise image of the t-th time step by using the unconditional deep neural network", until the value of t is 0, and output the diffusion noise image of the 0-th time step as a clear image obtained by performing speckle suppression on the observation image.

[0012] In a second aspect, the present application provides a diffusion sampling based synthetic aperture radar image speckle suppression device, which applies the diffusion sampling based synthetic aperture radar image speckle suppression method described above, and comprises:

[0013] An initialization module is configured to initialize a diffusion noise image of a T-th time step in a diffusion inversion process, and the value of time step t is T, and T is the total time step of the diffusion inversion process;

[0014] A prior state estimation calculation module is configured to calculate a prior state estimation of a t-1-th time step according to a diffusion noise image of a t-th time step by using an unconditional deep neural network;

[0015] A likelihood score function value estimation module is configured to estimate a likelihood score function value of a t-th time step according to an observation image and the diffusion noise image of the t-th time step by using a multiplicative noise model;

[0016] A diffusion noise image calculation module is configured to calculate a diffusion noise image of a t-1-th time step according to a prior state estimation of the t-1-th time step and a likelihood score function value of the t-th time step;

[0017] A return module is configured to let the value of t be reduced by 1, return to the prior state estimation calculation module, until the value of t is 0, and output the diffusion noise image of the 0-th time step as a clear image obtained by performing speckle suppression on the observation image.

[0018] In a third aspect, the present application provides a computer device, which comprises a memory, a processor, and a computer program stored in the memory and capable of running on the processor, and the processor executes the computer program to realize the diffusion sampling based synthetic aperture radar image speckle suppression method described above.

[0019] In a fourth aspect, the present application provides a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the method for suppressing speckle of synthetic aperture radar image based on diffusion sampling.

[0020] In a fifth aspect, the present application provides a computer program product, which comprises a computer program. The computer program is executed by a processor to implement the method for suppressing speckle of synthetic aperture radar image based on diffusion sampling.

[0021] According to the embodiments provided in the present application, the present application has the following technical effects.

[0022] The present application provides a method for suppressing speckle of synthetic aperture radar image based on diffusion sampling. The method of the present application is based on a sampling scheme of approximate score function, inverse shift distribution, and application of multiplicative noise model, solves the problem of multiplicative noise degradation in synthetic aperture radar image, and avoids supervised training and logarithmic transformation of SAR image. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0024] Figure 1 A flowchart of a method for suppressing speckle of synthetic aperture radar image based on diffusion sampling is provided for an embodiment of the present application.

[0025] Figure 2 An example of an observation image is provided for an embodiment of the present application.

[0026] Figure 3 An example of an observation image is provided for an embodiment of the present application. Figure 2 A schematic diagram of suppression results of each observation image in the method is provided for an embodiment of the present application.

[0027] Figure 4 An effect comparison diagram is provided for an embodiment of the present application.

[0028] Figure 5 A structural schematic diagram of a computer device is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0029] With reference to the drawings, the technical solutions in the embodiments of the present application will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of the present application.

[0030] The above-mentioned purposes, features and advantages of the present application will be more apparent and understandable, and the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0031] In an exemplary embodiment, as shown in Figure 1 a diffusion sampling based synthetic aperture radar image speckle suppression method is provided, including the following steps 101 to 105.

[0032] Step 101, initialize the diffusion noise image at the Tth time step in the diffusion inverse process, and initialize the time step t = T, T being the total time step of the diffusion inverse process.

[0033] Step 102, according to the diffusion noise image at the tth time step, calculate the prior state estimation at the (t-1)th time step by using an unconditional deep neural network.

[0034] Step 103, according to the observation image and the diffusion noise image at the tth time step, estimate the likelihood score function value at the tth time step by using a multiplicative noise model.

[0035] Step 104, according to the prior state estimation at the (t-1)th time step and the likelihood score function value at the tth time step, calculate the diffusion noise image at the (t-1)th time step.

[0036] Step 105, let the value of t decrease by 1, return to the step of "according to the diffusion noise image at the tth time step, calculate the prior state estimation at the (t-1)th time step by using an unconditional deep neural network", until the value of t is 0, and output the diffusion noise image at the 0th time step as a clear image obtained by performing speckle suppression on the observation image.

[0037] Implementing the above-mentioned steps 101 to 105 can solve the problem of multiplicative noise degradation in synthetic aperture radar images, and avoid supervised training and logarithmic transformation of SAR images.

[0038] In another exemplary embodiment, in the above-mentioned step 101, the observed SAR image affected by speckle noise, i.e. the observation image y, is further input, given the number of views L, the diffusion positive process single-step noise variance sequence the weight coefficient γ and the total noise variance sequence Initialization parameters: Take

[0039] In another exemplary embodiment, the steps 102-105 above constitute a diffusion sampling scheme based on a score function, which utilizes a diffusion model to generate a strong prior of depth, and utilizes a multiplicative noise model to perform multiplicative noise suppression by estimating inverse transition distribution. Experiments on synthetic and real SAR images show the superiority of the synthetic aperture radar image speckle suppression method based on diffusion sampling of the present application, and the specific principles are as follows.

[0040] S1, SAR image speckle suppression model based on diffusion model.

[0041] For the task of SAR image speckle suppression, using a diffusion model to suppress the speckle of a SAR image is equivalent to implementing a SAR image guided inverse image denoising process under the influence of speckle noise, in which a stochastic differential equation representing an inverse time process can be expressed as a SAR image speckle suppression model based on a diffusion model:

[0042]

[0043] where y is a SAR image under the influence of speckle noise, also known as an observed image, x t is the diffusion noise image at the t-th time step in the diffusion inverse process, when t = 0, x0 is the clear image to be solved, is the posterior conditional score function, is a Wiener process in inverse time, and β t represents the variance of the Gaussian noise added from x t-1 to x t .

[0044] The specific form of the posterior conditional score function is:

[0045]

[0046] The first term in the posterior conditional score function is the score function of the unconditional diffusion model, which is estimated by a pre-trained noise estimation network (i.e. an unconditional deep neural network) , and the second term likelihood score function is the key to the SAR image speckle suppression model based on the diffusion model.

[0047] S2, estimation of inverse transition distribution.

[0048] In order to solve the unknown term of the SAR image speckle suppression model based on the diffusion model in step S1, the inverse transition distribution is estimated in this step.

[0049] First, S1 remains unknown terms Can be written as:

[0050] p(y∣x t )=∫p(y∣x0)p(x0∣x t )dx0(2)

[0051] Where p(x0∣x t ) is the inverse transition distribution, p(y∣x0) can be obtained by SAR image subject to coherent speckle degradation. This step is to estimate the inverse transition distribution first. p(x0∣x t )dx0 is regarded as a probability measure, then:

[0052]

[0053] Let R , and Where Is estimated by the unconditional deep neural network Further, the single-point distribution is used to estimate the inverse transition distribution, that is So as to reduce the computational complexity.

[0054] S3, the estimation of the likelihood score function.

[0055] Further to solve the likelihood score function, that is To express p(y∣x0) is:

[0056]

[0057] Where C0=(L L / Γ(L)) d , SAR image coherent speckle obeys Γ distribution Γ(L,L / α), respectively, the number of views and the basic reflectivity, Γ(·) is the Γ function, Q(·) can be expressed as Therefore, combined with S1, S2, we can get:

[0058]

[0059] Let R=logp(y∣x t ), the likelihood score function can be written as:

[0060]

[0061] Where ⊙ is the element-wise multiplication, I is the unit matrix, Is the Jacobian operator acting on the vector function.

[0062] Based on the above principle, the following formula is obtained:

[0063] 1. The formula for calculating the prior state estimation of the t-1 time step is:

[0064]

[0065] wherein, is the prior state estimation of the t-1 time step, α t is a variable related to the noise variance β t of the t time step, α t = 1-β t , x t is the diffusion noise image of the t time step, is an unconditional deep neural network for outputting the value of the score function of the unconditional diffusion model , is the gradient symbol, p t (x t ) is the distribution of the diffusion noise image of the t time step, is the total noise variance of the t time step, η t is the random Gaussian noise of the t time step, is a standard normal distribution.

[0066] 2. The formula for estimating the likelihood score function value of the t time step is:

[0067]

[0068] wherein, is the likelihood score function, is the gradient symbol, p(y∣x t ) is the positive process distribution of the observation image obtained from the diffusion noise image of the t time step, y is the observation image, x t is the diffusion noise image of the t time step, is the first intermediate variable, is the value obtained by multiplying α t from 1 to t, is the score function of the unconditional diffusion model, p t (x t ) is the distribution of the diffusion noise image of the t time step, R is the second intermediate variable, R = log p(y∣x t );

[0069] L is a given view, ⊙ is element-wise multiplication, I is an identity matrix, is the Jacobian operator acting on the vector function, is an unconditional deep neural network, which is used to output the score function of the unconditional diffusion model The value of .

[0070] 3. The formula for calculating the diffuse noise image at the t-1 time step is:

[0071]

[0072] Among them, x t-1 is the diffuse noise image at time step t, is the prior state estimate of the t-1 time step, γ is the weight coefficient, α t is the noise variance β at time step t t The relevant variable, α t =1-β t , is the likelihood score function value at the tth time step.

[0073] In another exemplary embodiment, the following experiment was conducted to verify the effect of the above method.

[0074] The method of the above embodiment is used to Figure 2 The SAR image (i.e., the observed image) shown in FIG is subjected to speckle suppression. Figure 2 SAR1, SAR2, SAR3 and SAR4 are SAR images with view count L of 1, 2, 3 and 4, respectively. The suppression results are as follows: Figure 3 As shown, the contrast effect is as follows Figure 4 As shown, Figure 4 The comparison methods in this paper are SAR-BM3D and SAR-DDPM. To quantitatively evaluate the performance of these methods, this application example selected two metrics: peak value and signal-to-noise ratio (PSNR). Higher values ​​of these metrics indicate higher SAR image quality. The results are shown in Table 1. The results demonstrate that the proposed method outperforms the comparison methods in both image quality improvement and image fidelity.

[0075] Table 1 Quantitative results of speckle suppression

[0076]

[0077] Based on the same inventive concept, the application further provides a diffusion sampling based synthetic aperture radar image speckle suppression device for implementing the diffusion sampling based synthetic aperture radar image speckle suppression method described above. The device provides a solution implementation similar to the implementation described in the above method, and therefore the specific limitations in one or more diffusion sampling based synthetic aperture radar image speckle suppression device embodiments provided below can refer to the limitations of the diffusion sampling based synthetic aperture radar image speckle suppression method described above, which will not be repeated here.

[0078] In an exemplary embodiment, a diffusion sampling based synthetic aperture radar image speckle suppression device is provided, comprising:

[0079] An initialization module is configured to initialize the diffusion noise image at the Tth time step in the diffusion inversion process, and the time step t is T, and T is the total time step of the diffusion inversion process.

[0080] An a priori state estimation calculation module is configured to calculate the a priori state estimation at the (t-1)th time step by using an unconditional deep neural network according to the diffusion noise image at the tth time step.

[0081] A likelihood score function value estimation module is configured to estimate the likelihood score function value at the tth time step by using a multiplicative noise model according to the observation image and the diffusion noise image at the tth time step.

[0082] A diffusion noise image calculation module is configured to calculate the diffusion noise image at the (t-1)th time step according to the a priori state estimation at the (t-1)th time step and the likelihood score function value at the tth time step.

[0083] A return module is configured to decrease the value of t by 1 and return to the a priori state estimation calculation module until the value of t is 0, and output the diffusion noise image at the 0th time step as a clear image obtained by performing speckle suppression on the observation image.

[0084] In an exemplary embodiment, a computer device is provided, which can be a server or a terminal, and the internal structure diagram thereof can be as shown in Figure 5As shown in the figure. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through the system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capability. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and the external device. The communication interface of the computer device is used to communicate with the terminal outside through the network connection. The computer program is executed by the processor to implement a synthetic aperture radar image speckle suppression method based on diffusion sampling.

[0085] Those skilled in the art can understand that, Figure 5 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement. In one exemplary embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps in each of the above method embodiments.

[0086] In one exemplary embodiment, a computer readable storage medium is provided, storing a computer program, which is executed by a processor to implement the steps in each of the above method embodiments.

[0087] In one exemplary embodiment, a computer program product is provided, including a computer program, which is executed by a processor to implement the steps in each of the above method embodiments.

[0088] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties.

[0089] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, databases or other media used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0090] The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a blockchain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0091] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope of the present application.

[0092] The principles and implementation modes of the present application are described by applying specific examples herein, and the above-mentioned embodiments are only used to help understand the method and its core idea of the present application; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range can be changed. In conclusion, the content of the present application should not be understood as a limitation.

Claims

1. A method for suppressing coherent speckle in synthetic aperture radar images based on diffuse sampling, characterized in that: include: Initialize the diffusion noise image at the Tth time step in the diffusion inversion process, and initialize the time step t=T, where T is the total time step of the diffusion inversion process; Based on the diffuse noise image at time step t, the prior state estimate at time step t-1 is calculated using an unconditional deep neural network. According to the observed image and the diffuse noise image at the t-th time step, the likelihood score function value at the t-th time step is estimated using the multiplicative noise model; Calculate the diffuse noise image at time step t-1 based on the prior state estimate at time step t-1 and the likelihood score function value at time step t; Subtract 1 from the value of t, and return to the step of "calculating the prior state estimate at the t-1th time step using an unconditional deep neural network based on the diffuse noise image at the tth time step" until the value of t reaches 0, and outputting the diffuse noise image at the 0th time step as the clear image obtained by performing speckle suppression on the observed image; The formula for calculating the prior state estimate at the t-1 time step is: ； in, is the prior state estimate at the t-1 time step, is the noise variance at time step t The relevant variables, , is the diffuse noise image at time step t, is an unconditional deep neural network, which is used to output the score function of the unconditional diffusion model The value of is the gradient symbol, is the distribution of the diffuse noise image at the t-th time step, is the total noise variance at the t-th time step, is the random Gaussian noise at the t-th time step, , is a standard normal distribution; The formula for estimating the likelihood score function value at time step t is: ； in, is the likelihood score function, is the gradient symbol, is the positive process distribution of the observed image obtained from the diffuse noise image at the t-th time step, To observe the image, is the diffuse noise image at time step t, is the first intermediate variable, , From 1 to t The value obtained by multiplication is is the score function of the unconditional diffusion model, is the distribution of the diffuse noise image at the t-th time step, is the second intermediate variable, ; , , For a given number of views, is element-wise multiplication, is the identity matrix, is the Jacobi operator acting on the vector function, is an unconditional deep neural network, which is used to output the score function of the unconditional diffusion model The value of The formula for calculating the diffuse noise image at time step t-1 is: ； in, is the diffuse noise image at time step t, is the prior state estimate at the t-1 time step, is the weight coefficient, is the noise variance at time step t The relevant variables, , is the likelihood score function value at the tth time step.

2. A synthetic aperture radar image speckle suppression device based on diffusion sampling, characterized in that: The synthetic aperture radar image coherent speckle suppression device based on diffuse sampling applies the synthetic aperture radar image coherent speckle suppression method based on diffuse sampling according to claim 1, and the synthetic aperture radar image coherent speckle suppression device based on diffuse sampling comprises: Initialization module, used to initialize the diffusion noise image at the Tth time step in the diffusion inversion process, the initialization time step t=T, T is the total time step of the diffusion inversion process; A priori state estimation calculation module is used to calculate the priori state estimation at the t-1 time step based on the diffuse noise image at the t time step using an unconditional deep neural network; A likelihood score function value estimation module is used to estimate the likelihood score function value at the t-th time step based on the observed image and the diffuse noise image at the t-th time step using a multiplicative noise model; a diffusion noise image calculation module, for calculating the diffusion noise image at the t-1th time step based on the prior state estimate at the t-1th time step and the likelihood score function value at the tth time step; The return module is used to reduce the value of t by 1 and return to the prior state estimation calculation module until the value of t is 0, and output the diffuse noise image at the 0th time step as the clear image obtained by performing coherent speckle suppression on the observed image.

3. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for synthetic aperture radar image coherent speckle suppression based on diffuse sampling according to claim 1.

4. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for suppressing coherent speckles in synthetic aperture radar images based on diffusion sampling according to claim 1 is implemented.

5. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method for suppressing coherent speckles in synthetic aperture radar images based on diffusion sampling according to claim 1 is implemented.

Images