CN115063320A - SAR image speckle removing method and system based on maximum posterior probability estimation - Google Patents

SAR image speckle removing method and system based on maximum posterior probability estimation Download PDF

Info

Publication number
CN115063320A
CN115063320A CN202210803867.5A CN202210803867A CN115063320A CN 115063320 A CN115063320 A CN 115063320A CN 202210803867 A CN202210803867 A CN 202210803867A CN 115063320 A CN115063320 A CN 115063320A
Authority
CN
China
Prior art keywords
sar image
function
probability estimation
image
speckle
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210803867.5A
Other languages
Chinese (zh)
Inventor
王小青
陈明瑞
朱宇霆
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Sun Yat Sen University
Original Assignee
Sun Yat Sen University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Sun Yat Sen University filed Critical Sun Yat Sen University
Priority to CN202210803867.5A priority Critical patent/CN115063320A/en
Publication of CN115063320A publication Critical patent/CN115063320A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/15—Correlation function computation including computation of convolution operations
    • G06F17/153—Multidimensional correlation or convolution
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
    • G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10032—Satellite or aerial image; Remote sensing
    • G06T2207/10044—Radar image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20076—Probabilistic image processing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Software Systems (AREA)
  • Mathematical Optimization (AREA)
  • Databases & Information Systems (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Algebra (AREA)
  • Medical Informatics (AREA)
  • Multimedia (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Operations Research (AREA)
  • Probability & Statistics with Applications (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

The invention discloses a maximum posterior probability estimation-based SAR image speckle removing method and a system, wherein the method comprises the following steps: calculating the SAR image through a maximum posterior probability estimation algorithm to construct an SAR image target function; decomposing the SAR image target function by an alternating direction multiplier method to obtain a subfunction; and solving the subfunctions and carrying out iterative optimization processing on the SAR image target function based on the updating condition rule to obtain the speckle-removing noise-free SAR image. The system comprises: the system comprises a construction module, a decomposition module and an optimization module. By using the SAR image optimization method and device, the SAR image optimization precision can be improved under the condition that the SAR image target function is not constrained. The SAR image speckle removing method and system based on the maximum posterior probability estimation can be widely applied to the field of SAR image processing.

Description

SAR image speckle removing method and system based on maximum posterior probability estimation
Technical Field
The invention relates to the field of SAR image processing, in particular to an SAR image speckle removing method and system based on maximum posterior probability estimation.
Background
Synthetic Aperture Radar (SAR) remote sensing has the advantages of all-weather, high resolution and the like, and is widely applied to various fields of agriculture, forestry, environmental protection, disaster prevention and reduction, ocean monitoring, military and the like. However, due to the imaging mechanism specific to SAR, there is a severe speckle noise interference in SAR images. The existence of speckle noise can greatly affect the subsequent interpretation work of the SAR image, and in order to improve the interpretability of the SAR image, the conventional method is that before image processing and application work, a speckle removing algorithm is applied to an original SAR image for noise reduction preprocessing, for example, the conventional noise reduction method includes a SAR image speckle removing algorithm based on a spatial domain and an image speckle removing method based on learning, the SAR image is subjected to noise reduction preprocessing through a local spatial filter of a linear model, but the local spatial filter does not consider the correlation between structural textures, and only uses simple local weighted averaging to suppress noise, so that the problems of incomplete speckle removal or edge blurring, overlarge calculation amount, limitation on the scene which can be processed by the speckle removal model and the like are caused.
Disclosure of Invention
In order to solve the technical problems, the invention aims to provide a maximum posterior probability estimation-based SAR image speckle removing method and system, which can optimize SAR images in different occasions and improve the optimization precision of the SAR images.
The first technical scheme adopted by the invention is as follows: a SAR image speckle removing method based on maximum posterior probability estimation comprises the following steps:
calculating the SAR image through a maximum posterior probability estimation algorithm to construct an SAR image target function;
decomposing the SAR image target function by an alternating direction multiplier method to obtain a subfunction;
and solving the subfunctions and carrying out iterative optimization processing on the SAR image target function based on the updating condition rule to obtain the speckle-removing noise-free SAR image.
Further, the step of calculating the SAR image by the maximum a posteriori probability estimation algorithm to construct the SAR image target function specifically includes:
acquiring an observation image and carrying out logarithm processing on the observation image to obtain an observation image with additive noise;
based on a maximum posterior probability estimation algorithm, carrying out negative logarithm processing on an observation image with additive noise to obtain prior information of a noise-free image;
and substituting the prior information of the noiseless image into the observation image with the additive noise to construct an SAR image target function.
Further, the SAR image is represented as follows:
I＝R×N
in the above formula, I represents an observation image, R represents an ideal noise-free image, and N represents speckle noise.
Further, the SAR image objective function is as follows:
Figure BDA0003735678870000021
in the above formula, X represents an independent variable, L represents an equivalent view,
Figure BDA0003735678870000022
a priori information representing a noise-free image,
Figure BDA0003735678870000023
representing the denoising result, e X Representing an exponential function and Y representing the original SAR image.
Further, the step of decomposing the SAR image target function by the alternating direction multiplier method to obtain a subfunction specifically includes:
introducing an auxiliary variable into an SAR image target function, and constructing an augmented Lagrange function;
substituting an augmented Lagrange multiplier and a penalty coefficient into an augmented Lagrange function based on an alternating direction multiplier method to obtain a Lagrange function;
decomposing the Lagrange function to obtain a subfunction;
the sub-functions include a first sub-function, a second sub-function, and a third sub-function.
Further, the augmented lagrange function is expressed as follows:
Figure BDA0003735678870000024
in the above formula, a represents an augmented lagrange multiplier, ρ represents a penalty coefficient,
Figure BDA0003735678870000025
a priori information representing Z, Z representing an introduced elastic constraint variable, A T Representing the transpose of the incoming augmented lagrange multiplier matrix.
Further, the step of solving the subfunction and performing iterative optimization processing on the SAR image target function based on the update condition rule to obtain the speckle-removed noise-free SAR image specifically includes:
solving the first sub-function by a Newton method to obtain a first optimized value;
solving the second sub-function by a deep learning method to obtain a second optimized value;
updating the augmented Lagrange multiplier through a third sub-function to obtain a third optimized value;
integrating the first optimized value, the second optimized value and the third optimized value to perform iterative optimization on the SAR image target function based on the updating condition rule;
and repeating the solving step and the updating step until the iterative optimization condition reaches a preset termination condition, stopping optimization, and outputting the speckle-free SAR image.
Further, the step of solving the second sub-function by a deep learning method to obtain a second optimized value specifically includes:
performing equivalent transformation processing on the second sub-function to construct a loss function of a convolutional neural network model in the SAR image denoising problem;
adding multiplicative speckle noise to the SAR image to obtain a simulated SAR image;
inputting the simulated SAR image into a convolutional neural network model for feature fusion processing based on a loss function to obtain a high-resolution feature map;
based on the convolutional neural network model, carrying out repeated feature fusion processing and down-sampling processing on the feature information of the high-resolution feature map until an SAR image with feature mapping information is obtained, and stopping the fusion step and the down-sampling step;
the SAR image with the feature mapping information is used as a second optimized value.
Further, the update condition rule is specifically as follows:
ρ k+1 ＝μ×ρ k
in the above formula, μ represents a given constant, ρ k And representing the penalty coefficient of the k-th iteration optimization.
The second technical scheme adopted by the invention is as follows: a SAR image despeckle system based on maximum posterior probability estimation comprises:
the construction module is used for calculating the SAR image through a maximum posterior probability estimation algorithm and constructing an SAR image target function;
the decomposition module is used for decomposing the SAR image target function by an alternative direction multiplier method to obtain a subfunction;
and the optimization module is used for solving the subfunctions and carrying out iterative optimization processing on the SAR image target function based on the updating condition rule to obtain the speckle-removing noise-free SAR image.
The method and the system have the beneficial effects that: according to the SAR image optimization method, the target function which accords with the speckle noise characteristics of the SAR image is deduced by the maximum posterior probability estimation method, and the SAR image target function is decomposed and fitted and optimized by using the alternative direction multiplier method, so that the optimization method is more flexible, can improve the SAR image optimization precision, and is suitable for SAR image optimization on more different occasions.
Drawings
FIG. 1 is a flowchart illustrating the steps of a SAR image speckle reduction method based on maximum posterior probability estimation according to the present invention;
FIG. 2 is a block diagram of the SAR image speckle reduction system based on maximum posterior probability estimation according to the present invention;
Detailed Description
The invention is described in further detail below with reference to the figures and the specific embodiments. The step numbers in the following embodiments are provided only for convenience of illustration, the order between the steps is not limited at all, and the execution order of each step in the embodiments can be adapted according to the understanding of those skilled in the art.
Referring to fig. 1, the invention provides a maximum posterior probability estimation-based SAR image speckle removing method, which comprises the following steps:
s1, calculating the SAR image through a maximum posterior probability estimation algorithm, and constructing an SAR image target function;
s11, acquiring an observation image and carrying out logarithm taking processing on the observation image to obtain an observation image with additive noise;
specifically, the SAR image is represented as follows:
I＝R×N
in the above formula, I represents an observation image, R represents an ideal noiseless image, and N represents a speckle noise;
speckle noise N obeys the gamma distribution:
Figure BDA0003735678870000041
in the above formula, L represents an equivalent visual number, p N Probability density function, L, representing obedience of speckle noise L Representing the equivalent visual number L raised to the power L, Γ (L) representing a gamma function, N L-1 Denotes the power L-1 of N, e -LN Representing an exponential function.
The following expression is required for the equivalent view:
Figure BDA0003735678870000042
in the above formula, e (i) represents the image mean intensity, and var (i) represents the image variance.
Taking the logarithm of I to obtain ln (I) ═ ln (r) + ln (n), and if not denoted as Y ═ X + S, then S ═ ln (n) obeys the following probability density distribution function:
Figure BDA0003735678870000043
in the above formula, p S Probability density function representing S obedience, e -LS 、
Figure BDA0003735678870000044
Both represent exponential functions.
S12, based on the maximum posterior probability estimation algorithm, carrying out negative logarithm processing on the observed image with additive noise to obtain prior information of the noise-free image;
specifically, in order to obtain an observed image X ═ ln (r) having additive noise, it is equivalent to obtain max P (X | Y) by the maximum a posteriori probability estimation method. And is also provided with
P(X|Y)∝P(Y|X)×P(X)
In the above equation, P (X | Y) represents the probability of the event X, and P (Y | X) represents the probability of the event Y when the event X is known to occur.
Then there is
Figure BDA0003735678870000045
In the above formula, the first and second carbon atoms are,
Figure BDA0003735678870000051
means that X with the maximum probability P (Y | X) is obtained,
Figure BDA0003735678870000052
denotes X that maximizes the probability P (Y | X).
Taking negative logarithm of the above formula, the maximization problem is converted into minimization problem, there are
Figure BDA0003735678870000053
The above formula may also be expressed by the following form:
Figure BDA0003735678870000054
in the above formula, D (X, Y) — ln (P (Y | X)) represents a fidelity term for measuring the degree of similarity between X and Y,
Figure BDA0003735678870000055
Figure BDA0003735678870000056
a priori information representing X;
s13, substituting the prior information of the noiseless image into the observed image with the additive noise to construct an SAR image target function;
in particular, for the above formula
Figure BDA00037356788700000514
Generally, a constraint term is used to constrain X for denoising purposes, and for D (X, Y) — ln (P (Y | X)), it satisfies
D(X,Y)＝-ln(P Y|X (Y|X))＝-ln(P S (Y-X))∝L(e Y-X +X-Y)
Substitution of D (X, Y)
Figure BDA0003735678870000057
The SAR image objective function can be obtained, which is specifically expressed as follows:
Figure BDA0003735678870000058
in the above formula, X represents an independent variable, L represents an equivalent view,
Figure BDA0003735678870000059
a priori information representing a noise-free image,
Figure BDA00037356788700000510
representing the denoising result, e X Representing an exponential function and Y representing the original SAR image.
S2, decomposing the SAR image target function through an alternating direction multiplier method to obtain a subfunction;
s21, introducing auxiliary variables into the SAR image target function, and constructing an augmented Lagrangian function;
specifically, an auxiliary variable Z is introduced into an SAR image target function formula, namely
Figure BDA00037356788700000511
Further introducing an augmented Lagrange multiplier A and a penalty coefficient rho, the augmented Lagrange function is expressed as follows:
Figure BDA00037356788700000512
in the above formula, a represents an augmented lagrange multiplier, ρ represents a penalty coefficient,
Figure BDA00037356788700000513
a priori information representing Z, Z representing an introduced elastic constraint variable, A T Representing the transpose of the introduced augmented lagrange multiplier matrix.
S22, substituting an augmented Lagrange multiplier and a penalty coefficient into an augmented Lagrange function based on an alternating direction multiplier method to obtain the Lagrange function;
s23, decomposing the Lagrangian function to obtain a subfunction;
and S24, wherein the subfunctions comprise a first subfunction, a second subfunction and a third subfunction.
Specifically, using the alternating direction multiplier method, the optimization problem represented by the augmented lagrange function can be decomposed into three sub-optimization problems, and assuming that in the (k +1) th iteration, the three optimization sub-functions can be expressed as follows:
a first sub-function:
Figure BDA0003735678870000061
the second sub-function:
Figure BDA0003735678870000062
the third sub-function:
A k+1 ＝A k +ρ k (X k+1 -Z k+1
and S3, solving the subfunctions and carrying out iterative optimization processing on the SAR image target function based on the updating condition rule to obtain the speckle-removing noise-free SAR image.
S31, solving the first sub-function through a Newton method to obtain a first optimized value;
specifically, a newton method is used for solving, and a specific iterative formula for solving the first sub-function by using the newton method is as follows:
Figure BDA0003735678870000063
in the above formula, X ij k 、Y ij 、Z ij k And A ij k Respectively representing the ith row and jth column elements in matrix X, Y, Z, A in the kth update iteration,
Figure BDA0003735678870000064
a numerical value is indicated.
S32, solving the second sub-function through a deep learning method to obtain a second optimized value;
s321, performing equivalent transformation processing on the second sub-function, and constructing a loss function of a convolutional neural network model in the SAR image denoising problem;
in particular, for the second sub-problem represented by the second sub-function, it can be regarded as a loss function of the convolutional neural network in the image denoising problem. For ease of understanding, the second function is equivalently transformed as follows:
Figure BDA0003735678870000065
in the above formula, the first and second carbon atoms are,
Figure BDA0003735678870000066
representing Gaussian noise, Z k+1 、X k+1 、A k And ρ k The values representing Z, X, A and ρ found in the (k +1) th iteration of the loop are shown, respectively.
A Gaussian noise denoiser represented by a second sub-function equivalent transform is used to remove the variance as
Figure BDA0003735678870000071
Gaussian noise. In order to adapt the fitting optimization scheme to various scenes, the method is not suitable for
Figure BDA0003735678870000072
Applying any constraints, and fitting and solving the optimization problem represented by the second sub-function by using a convolutional neural network in a deep learning method;
s322, multiplicative speckle noise adding processing is carried out on the SAR image to obtain a simulated SAR image;
specifically, multiplicative speckle noise is added into a common gray image, namely an SAR image, to serve as a simulated SAR image, an original gray image serves as a label, the data set serves as a training set to be trained to obtain a convolutional neural network, and the convolutional neural network is used for optimizing and solving a second sub-problem represented by a second sub-function.
S323, inputting the simulated SAR image into a convolutional neural network model for feature fusion processing based on a loss function to obtain a high-resolution feature map;
specifically, the framework of the neural network is set as follows: firstly, the depth of the neural network is limited to eight layers, so that the neural network has good performance and small calculation amount; secondly, the neural network should always keep the processed image with high resolution, and in order to achieve the purpose, the neural network should treat the denoising task as a pixel-level task; the network is set to be a parallel connection multi-resolution network, and multi-resolution features are repeatedly fused, so that the designed network can keep detailed information of high resolution in the whole feature extraction process, and the SAR image denoising precision is improved.
S324, based on the convolutional neural network model, carrying out repeated feature fusion processing and down-sampling processing on feature information of the high-resolution feature map until an SAR image with feature mapping information is obtained, and stopping the fusion step and the down-sampling step;
and S325, taking the SAR image with the feature mapping information as a second optimized value.
Specifically, the first stage of neural network training includes feature extraction branches; inputting the high resolution feature map (resolution and feature channel numbers H, W and C) into a repeat bottleneck module; the downsampling operation includes two new feature maps, A (resolution and feature channel numbers H, W and C) and B (resolution and feature channel numbers H/2, W/2, and 2C).
The second stage comprises two feature extraction branches; the input to the first branch is feature map a and the input to the second branch is feature map B. After inputting the four repeated bottleneck modules, a downsampling operation step is performed to obtain feature maps A '(H, W, C), B' (H/2, W/2, 2C), D (H/2, W/2, 2C) and E (H/4, W/4, 4C). Meanwhile, the information fusion layer firstly adjusts the resolution and the channel number of the second branch feature map B ' to be the same as those of the first branch feature map A ' to obtain a feature map B ', and then adds and fuses the feature map A ' and the feature map B ' to obtain a feature map F. And adding and combining the images D and B' to obtain a characteristic image G. After the information fusion layer, three feature maps, E, F and G, are finally output in the second stage.
The third stage comprises three feature extraction branches; the inputs of the three branches correspond to E, F and G respectively, and the specific operation process of the stage is similar to that of the second stage; however, the third branch no longer performs the downsampling operation, and finally the information fusion layer iteratively combines the features of the three branches to obtain feature maps H, I and J. H is used as the feature output of the network extraction for final prediction and classification.
The network designed in the way is connected with the multi-resolution subnets in parallel, information is repeatedly fused, a large amount of detail characteristics are stored in shallow low-resolution characteristics, and the SAR image denoising precision is improved;
the above steps S321 to S325 are that the second sub-function is highly consistent with the loss function of the common gaussian denoising neural network, and the neural network can be regarded as performing optimization solution on the loss function under a certain rule, so that the second sub-function can be solved by using a neural network method, in the network design, the H image is designed as a final output image, and other images are only used as variables needed in some calculations, such as a feature image, a down-sampling image, and the like. Therefore, the output result only selects the H image, and the neural network method is designed to solve the second subproblem, so that the final output H image of the neural network can be regarded as the solution of the second subproblem;
s33, updating the augmented Lagrange multiplier through a third sub-function to obtain a third optimized value;
s34, integrating the first optimized value, the second optimized value and the third optimized value to perform iterative optimization on the SAR image target function based on the updating condition rule;
in particular, the third sub-function is used to update the augmented Lagrangian multiplier A k Updating penalty coefficient rho by updating rule k The update condition rule is specifically as follows:
ρ k+1 ＝μ×ρ k
in the above formula, μ represents a given constant, ρ k To representAnd (5) a penalty coefficient of the k-th iteration optimization.
And S35, repeating the steps S31 to S34 until the iteration optimization condition reaches a preset iteration termination condition, stopping optimization, and outputting the speckle-free SAR image.
Specifically, the process of decomposing, solving and fitting the SAR image target function is continuously repeated until the iteration optimization condition reaches a preset iteration termination condition, the optimization is stopped, and the speckle-free SAR image X is output k+1 Through reasonable mathematical calculations and experience, the updated termination conditions are as follows:
||X k+1 -X k || F ≤10 -6
||X k+1 -Z k+1 || F ≤10 -6
||Z k+1 -Z k || F ≤10 -6
in the above formula, X k+1 Representing a despeckle, noiseless, image, X k Representing the matrix X, Z obtained in the k-th cycle k+1 Representing the value of the matrix Z obtained in the (k +1) th cycle, Z k Representing the matrix Z, | · | | | non-calculation of the kth cycle F Representing the F-norm of the matrix.
Referring to fig. 2, a maximum a posteriori probability estimation based SAR image despeckle system comprises:
the construction module is used for calculating the SAR image through a maximum posterior probability estimation algorithm and constructing an SAR image target function;
the decomposition module is used for decomposing the SAR image target function by an alternative direction multiplier method to obtain a subfunction;
and the optimization module is used for solving the subfunctions and carrying out iterative optimization processing on the SAR image target function based on the updating condition rule to obtain the speckle-removing noise-free SAR image.
The contents in the method embodiments are all applicable to the system embodiments, the functions specifically implemented by the system embodiments are the same as those in the method embodiments, and the beneficial effects achieved by the system embodiments are also the same as those achieved by the method embodiments.
While the preferred embodiments of the present invention have been illustrated and described, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

Claims (10)

1. A SAR image speckle removing method based on maximum posterior probability estimation is characterized by comprising the following steps:
calculating the SAR image through a maximum posterior probability estimation algorithm to construct an SAR image target function;
decomposing the SAR image target function by an alternating direction multiplier method to obtain a subfunction;
and based on the updating condition rule, solving the subfunction and performing iterative optimization processing on the SAR image target function to obtain the speckle-removing noise-free SAR image.
2. The maximum a posteriori probability estimation based SAR image speckle removing method according to claim 1, wherein the step of calculating the SAR image by a maximum a posteriori probability estimation algorithm to construct the SAR image target function specifically comprises:
acquiring an observation image and carrying out logarithm processing on the observation image to obtain an observation image with additive noise;
based on a maximum posterior probability estimation algorithm, carrying out negative logarithm processing on an observation image with additive noise to obtain prior information of a noise-free image;
and substituting the prior information of the noiseless image into the observation image with the additive noise to construct an SAR image target function.
3. The maximum a posteriori probability estimation based SAR image speckle reduction method according to claim 2, wherein the SAR image is represented as follows:
I＝R×N
in the above formula, I represents an observation image, R represents an ideal noise-free image, and N represents speckle noise.
4. The maximum a posteriori probability estimation-based SAR image speckle reduction method according to claim 3, wherein the SAR image objective function is as follows:
Figure FDA0003735678860000011
in the above formula, X represents an independent variable, L represents an equivalent visual number,
Figure FDA0003735678860000012
a priori information representing a noise-free image,
Figure FDA0003735678860000013
representing the denoising result, e X Representing an exponential function and Y representing the original SAR image.
5. The SAR image speckle reduction method based on the maximum posterior probability estimation as claimed in claim 4, wherein the step of decomposing the SAR image objective function by the alternative direction multiplier method to obtain the sub-function specifically comprises:
introducing an auxiliary variable into an SAR image target function, and constructing an augmented Lagrange function;
substituting an augmented Lagrange multiplier and a penalty coefficient into an augmented Lagrange function based on an alternating direction multiplier method to obtain a Lagrange function;
decomposing the Lagrange function to obtain a subfunction;
the sub-functions include a first sub-function, a second sub-function, and a third sub-function.
6. The SAR image speckle reduction method based on the maximum a posteriori probability estimation is characterized in that the augmented Lagrangian function is expressed as follows:
Figure FDA0003735678860000021
in the above formula, a represents an augmented lagrange multiplier, ρ represents a penalty coefficient,
Figure FDA0003735678860000022
a priori information representing Z, Z representing an introduced elastic constraint variable, A T Representing the transpose of the introduced augmented lagrange multiplier matrix.
7. The SAR image speckle reduction method based on maximum a posteriori probability estimation according to claim 6, wherein the step of solving the subfunction and performing iterative optimization processing on the SAR image objective function based on the update condition rule to obtain the speckle reduction noise-free SAR image specifically comprises:
solving the first sub-function by a Newton method to obtain a first optimized value;
solving the second sub-function by a deep learning method to obtain a second optimized value;
updating the augmented Lagrange multiplier through a third sub-function to obtain a third optimized value;
integrating the first optimized value, the second optimized value and the third optimized value to perform iterative optimization on the SAR image target function based on the updating condition rule;
and repeating the solving step and the updating step until the iterative optimization condition reaches a preset termination condition, stopping optimization, and outputting the speckle-free SAR image.
8. The maximum a posteriori probability estimation-based SAR image speckle reduction method according to claim 7, wherein the step of solving the second sub-function by a deep learning method to obtain the second optimized value specifically comprises:
performing equivalent transformation processing on the second sub-function to construct a loss function of a convolutional neural network model in the SAR image denoising problem;
adding multiplicative speckle noise to the SAR image to obtain a simulated SAR image;
inputting the simulated SAR image into a convolutional neural network model for feature fusion processing based on a loss function to obtain a high-resolution feature map;
based on a convolutional neural network model, carrying out repeated feature fusion processing and downsampling processing on feature information of the high-resolution feature map until an SAR image with feature mapping information is obtained, and stopping the fusion step and the downsampling step;
the SAR image with the feature mapping information is used as a second optimized value.
9. The maximum a posteriori probability estimation based SAR image speckle removal method according to claim 8, wherein the update condition rule is specifically as follows:
ρ k+1 ＝μ×ρ k
in the above formula, μ represents a given constant, ρ k And representing the penalty coefficient of the k-th iteration optimization.
10. A SAR image speckle removing system based on maximum posterior probability estimation is characterized by comprising the following modules:
the construction module is used for calculating the SAR image through a maximum posterior probability estimation algorithm and constructing an SAR image target function;
the decomposition module is used for decomposing the SAR image target function by an alternative direction multiplier method to obtain a subfunction;
and the optimization module is used for solving the subfunctions and carrying out iterative optimization processing on the SAR image target function based on the updating condition rule to obtain the speckle-removing noise-free SAR image.
CN202210803867.5A 2022-07-07 2022-07-07 SAR image speckle removing method and system based on maximum posterior probability estimation Pending CN115063320A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210803867.5A CN115063320A (en) 2022-07-07 2022-07-07 SAR image speckle removing method and system based on maximum posterior probability estimation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210803867.5A CN115063320A (en) 2022-07-07 2022-07-07 SAR image speckle removing method and system based on maximum posterior probability estimation

Publications (1)

Publication Number Publication Date
CN115063320A true CN115063320A (en) 2022-09-16

Family

ID=83205527

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210803867.5A Pending CN115063320A (en) 2022-07-07 2022-07-07 SAR image speckle removing method and system based on maximum posterior probability estimation

Country Status (1)

Country Link
CN (1) CN115063320A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116385319A (en) * 2023-05-29 2023-07-04 中国人民解放军国防科技大学 Radar image speckle filtering method and device based on scene cognition

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116385319A (en) * 2023-05-29 2023-07-04 中国人民解放军国防科技大学 Radar image speckle filtering method and device based on scene cognition
CN116385319B (en) * 2023-05-29 2023-08-15 中国人民解放军国防科技大学 Radar image speckle filtering method and device based on scene cognition

Similar Documents

Publication Publication Date Title
CN109859147B (en) Real image denoising method based on generation of antagonistic network noise modeling
CN106920220B (en) The turbulent flow method for blindly restoring image optimized based on dark primary and alternating direction multipliers method
CN112184577B (en) Single image defogging method based on multiscale self-attention generation countermeasure network
CN112233026A (en) SAR image denoising method based on multi-scale residual attention network
CN114019467B (en) Radar signal identification and positioning method based on MobileNet model transfer learning
CN112233046B (en) Image restoration method under Cauchy noise and application thereof
Jiao et al. Guided-Pix2Pix: End-to-end inference and refinement network for image dehazing
CN112330569A (en) Model training method, text denoising method, device, equipment and storage medium
CN115063320A (en) SAR image speckle removing method and system based on maximum posterior probability estimation
Febin et al. A retinex-based variational model for enhancement and restoration of low-contrast remote-sensed images corrupted by shot noise
CN107292855A (en) A kind of image de-noising method of the non local sample of combining adaptive and low-rank
CN113096032A (en) Non-uniform blur removing method based on image area division
CN109615590B (en) Sonar image enhancement method based on fuzzy algorithm and fractional differential algorithm
Zhou et al. Multi-scale and attention residual network for single image dehazing
CN115660980A (en) Unsupervised underwater image enhancement method and related equipment
CN115937302A (en) Hyperspectral image sub-pixel positioning method combined with edge preservation
Cao et al. Remote sensing image recovery and enhancement by joint blind denoising and dehazing
CN110517196B (en) SAR image denoising method and system
CN109727200A (en) Similar block based on Bayes's tensor resolution piles up Denoising method of images and system
CN110648291B (en) Unmanned aerial vehicle motion blurred image restoration method based on deep learning
Li et al. Image dehazing using multi-scale recursive networks
CN114723614B (en) Image restoration method based on wavelet domain ADMM depth network
Lu Analysis of Image Restoration Based on EM Algorithm
Zeng et al. Hyperspectral Image Restoration via Global Total Variation Regularized Local Nonconvex Low-Rank Matrix Approximation
Ye et al. Comparative and Quantitative Study of FundamentalApproaches on Digital Aerial Image Deblurring

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination