CN106408587B - Regard SAR image segmentation method and device more - Google Patents

Abstract

The present invention provides a multi-view SAR image segmentation method and device, wherein the method includes: reading the multi-view SAR image to be segmented; initializing the double weight w; repeatedly performing the following steps: calculating the Gamma distribution scale parameter β; calculating the above The probability of the image p(z|w); calculate the distribution function p(w) of the above double weight w; calculate the quality function L; update the above double weight w according to the gradient method; substitute the updated double weight w into the quality function L; until When |L ^(t+1) ‑L ^(t) | is less than the preset threshold ε, stop performing the above steps, and determine the category of each pixel in the above image according to the current double weight w; output segmentation according to the category of each pixel result. The anti-noise performance of the segmented image of the invention is better, the phenomenon of misclassification of the segmented result is less, and the fitting of the segmented boundary is accurate.

Description

Multi-view SAR image segmentation method and device

technical field

The present invention relates to the technical field of image segmentation, in particular to a multi-view SAR image segmentation method and device.

Background technique

Synthetic Aperture Radar (SAR) is a high-resolution imaging radar that records the shape of ground objects by receiving the electromagnetic waves scattered by the target and converting them into images. The inherent speckle noise caused by its unique imaging mechanism gives the image segmentation a Great difficulty came. Although multi-look technology can reduce part of the noise, there are still a lot of speckle noise in multi-look SAR images in practical applications. Therefore, the anti-noise and accuracy of multi-look SAR image segmentation methods have always been a hot research issue.

At present, the methods of multi-view SAR image segmentation mainly include: threshold method, boundary method, clustering method and statistical model method. Among them, the most widely used is the statistical method, which usually uses a mixture model to describe the complex distribution of images. The most commonly used mixture model is the Gaussian Mixture Model (Gaussian Mixture Model, GMM), which assumes that the gray value of the pixel in the image obeys the Gaussian distribution. However, the weight coefficient in the traditional GMM is represented by a vector and only related to clustering. Single weight, and the SAR image obeys the Gamma distribution, the above problems will cause the traditional GMM to model the SAR image inaccurately.

Aiming at the problems of poor anti-noise performance and unsatisfactory segmentation results of the multi-view SAR image segmentation method mentioned above, no effective solution has been proposed yet.

Contents of the invention

In view of this, the object of the present invention is to provide a multi-view SAR image segmentation method and device, which can improve the anti-noise performance of image segmentation and enhance the segmentation accuracy.

In order to achieve the above object, the technical solution adopted in the embodiment of the present invention is as follows:

In a first aspect, an embodiment of the present invention provides a multi-view SAR image segmentation method, including:

Read the multi-view SAR image to be segmented; the above multi-view SAR image to be segmented is represented by the characteristic field z:

z={z _i (x _i ,y _i ):i=1,…,n}

where i is the pixel index, n is the total number of pixels, z _i is the intensity of pixel i, ( _xi , y _i ) ∈ D is the grid position of pixel i, and D is the image domain;

Initialize the dual weight w:

w _i =(w _il :l=1,...,k)

Among them, w _il is the category weight of category l containing pixel i, which satisfies k is the number of classes;

Repeat the following steps: Calculate the Gamma distribution scale parameter β:

Among them, α is the Gamma distribution shape parameter;

Calculate the probability p(z|w) of the above image;

Where p(z _i |w _il ) is the probability density of z _i defined by the double-weight Gamma mixture model, as follows:

Calculate the distribution function p(w) of the above double weight w:

Among them, A is the normalization coefficient, η is the neighborhood effect coefficient, N _i is a set of 8 neighborhood pixels centered on ( _xi , y _i ), and satisfies i'∈N _i , i'≠i;

Calculating the quality function L, the above-mentioned quality function L is the logarithmic function of the joint probability distribution function of the probability p(z|w) and the distribution function p(w);

Update the above double weight w according to the gradient method:

w ^(t+1) = w ^(t) +ξΔw ^(t)

Among them, t is the number of iterations, ξ is the step size, and Δw ^(t) is the gradient, as follows:

Substitute the updated double weight w into the quality function L;

Until |L ^(t+1) -L ^(t) | is less than the preset threshold ε, stop performing the above steps, and determine the category to which each pixel in the above image belongs according to the current double weight w;

Output the segmentation result according to the class to which each pixel belongs.

With reference to the first aspect, the embodiment of the present invention provides a first possible implementation manner of the first aspect, wherein calculating the quality function L includes:

Calculate the probability p(z|w) and the distribution function p(w) joint probability distribution function p(z,w):

p(z,w)=p(z|w)p(w);

Take the logarithm of the joint probability distribution function:

L(w)=log p(z,w)=log p(z|w)+log p(w).

With reference to the first aspect, the embodiment of the present invention provides a second possible implementation manner of the first aspect, wherein determining the category to which the foregoing pixel belongs includes:

Compute the maximum of double weights:

c _i =arg max{w _il ,l=1,...,k};

Among them, c _i is the label of the category to which the i-th pixel belongs;

The above-mentioned maximum value is taken as the category to which the above-mentioned pixel belongs.

In combination with the first aspect, the embodiment of the present invention provides a third possible implementation manner of the first aspect, wherein outputting the segmentation result according to the category to which each pixel belongs includes: updating with the mean value of the intensities of all pixels in the same category in the image The intensities of all pixels in the same category are obtained as an average image; the average image is displayed by a display device.

In combination with the first aspect, the embodiment of the present invention provides a fourth possible implementation of the first aspect, which also includes: setting constants, the above-mentioned constants include: the number of classes k, the shape parameter α of Gamma distribution, the normalization constant A and the neighbor Domain effect coefficient η.

In the second aspect, the embodiment of the present invention also provides a multi-view SAR image segmentation device, including: a reading module for reading the multi-view SAR image to be segmented; the above-mentioned multi-view SAR image to be segmented is represented by a characteristic field z:

z={z _i (x _i ,y _i ):i=1,…,n}

where i is the pixel index, n is the total number of pixels, z _i is the intensity of pixel i, ( _xi , y _i ) ∈ D is the grid position of pixel i, and D is the image domain;

The double weight initialization module is used to initialize the double weight w:

w _i =(w _il :l=1,...,k)

Among them, w _il is the category weight of category l containing pixel i, which satisfies k is the number of classes;

The weight parameter iterative update module is used to repeatedly perform the following calculations: Calculate the Gamma distribution scale parameter β:

Among them, α is the Gamma distribution shape parameter;

Calculate the probability p(z|w) of the image;

Where p(z _i |w _il ) is the probability density of z _i defined by the double-weight Gamma mixture model, as follows:

Calculate the distribution function p(w) of the above double weight w:

Among them, A is the normalization coefficient, η is the neighborhood effect coefficient, N _i is a set of 8 neighborhood pixels centered on ( _xi , y _i ), and satisfies i'∈N _i , i'≠i;

Calculating the quality function L, the above-mentioned quality function L is the logarithmic function of the joint probability distribution function of the probability p(z|w) and the distribution function p(w);

Update the above double weight w according to the gradient method:

w ^(t+1) = w ^(t) +ξΔw ^(t)

Among them, t is the number of iterations, ξ is the step size, and Δw ^(t) is the gradient, as follows:

Substitute the updated double weight w into the quality function L;

Until |L ^(t+1) -L ^(t) | is less than the preset threshold ε, stop performing the above calculation;

A category determination module, configured to determine the category to which each pixel in the image belongs according to the current double weight w;

The output module is configured to segment the image according to the category to which each pixel belongs, and output the segmentation result.

In combination with the second aspect, the embodiment of the present invention provides a first possible implementation manner of the second aspect, wherein the above-mentioned weight parameter iterative update module includes: a first calculation unit for calculating the probability p(z|w) and the distribution The joint probability distribution function p(z,w) of the function p(w):

p(z,w)=p(z|w)p(w);

The second calculation unit is used to take the logarithm of the above joint probability distribution function:

L(w)=log p(z,w)=log p(z|w)+log p(w).

In combination with the second aspect, an embodiment of the present invention provides a second possible implementation manner of the second aspect, wherein the category determination module includes: a maximum value calculation unit, configured to calculate the maximum value of the double weight:

c _i =arg max{w _il ,l=1,...,k};

Among them, c _i is the label of the category to which the i-th pixel belongs;

A class determining unit, configured to use the above maximum value as the class to which the pixel belongs.

In combination with the second aspect, the embodiment of the present invention provides a third possible implementation of the second aspect, wherein the above-mentioned output module includes: an average value unit, which is used to update the average value of the intensities of all pixels in the same category in the image The intensities of all pixels in the category are used to obtain an average image; the display unit is configured to display the average image through a display device.

In combination with the second aspect, the embodiment of the present invention provides a fourth possible implementation manner of the second aspect, which also includes: a constant setting module for setting constants, the above constants include: class number k, Gamma distribution morphological parameter α, regression The normalization constant A and the neighborhood interaction coefficient η.

The multi-view SAR image segmentation method and device provided by the embodiments of the present invention use a double-weight Gamma mixture model based on generic field representation to model the image feature field; in order to introduce neighborhood relations, a Markov random field model (Markov Random Filed, MRF), combined with the error sum of squares theory, regards the central pixel’s category weight as the mean value, and constructs the sum of squares function of the difference between it and the neighborhood pixel’s category weight to describe the difference of pixels in the neighborhood window, and uses the error sum of squares Representing the difference can make full use of the category information and enhance the segmentation accuracy of the image. At the same time, when estimating the dual weights in the Gamma mixture model, the embodiment of the present invention uses the gradient method to solve the dual weights, which has a fast iteration speed and is less prone to local optimal problems.

Description of drawings

In order to illustrate the technical solutions of the embodiments of the present invention more clearly, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention, and thus It should be regarded as a limitation on the scope, and those skilled in the art can also obtain other related drawings based on these drawings without creative work.

Fig. 1 shows a flow chart of a multi-view SAR image segmentation device provided by an embodiment of the present invention;

Fig. 2 shows a schematic structural diagram of a multi-view SAR image segmentation device provided by an embodiment of the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only It is a part of embodiments of the present invention, but not all embodiments. The components of the embodiments of the invention generally described and illustrated in the figures herein may be arranged and designed in a variety of different configurations. Accordingly, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts belong to the protection scope of the present invention.

Considering the problem of poor anti-noise performance and poor accuracy in the multi-view SAR image segmentation method in the prior art, the embodiment of the present invention provides a multi-view SAR image segmentation method and device, which can use corresponding software and hardware accomplish. The following is described by way of examples.

Example 1

Fig. 1 shows a schematic flowchart of a multi-view SAR image segmentation method provided by an embodiment of the present invention. The specific flow of the method shown in FIG. 1 will be described in detail below.

Step S110, read the multi-view SAR image to be segmented.

The multi-view SAR image to be segmented is represented by the feature field z:

z={z _i (x _i ,y _i ):i=1,…,n}

where i is the pixel index, n is the total number of pixels, z _i is the intensity of pixel i, ( _xi , y _i )∈D is the grid position of pixel i, and D is the image domain.

Step S120, initialize the double weight w.

In this embodiment, a double weight w is defined to represent the relationship between a pixel and a category, as follows:

w _i =(w _il :l=1,...,k)

Among them, w _il is the category weight of category l containing pixel i, which satisfies k is the number of classes. Among them, the initial double weight is randomly generated, the row represents the pixel, and the column represents the category. The value range of the elements in the matrix is 0-1, and the sum of each row is 1.

Step S130, calculating the Gamma distribution scale parameter β.

In this embodiment, the Gamma distribution scale parameter β _l is defined as a function of the category weight w _il , specifically as follows:

Among them, α is the shape parameter of Gamma distribution, and for multi-view SAR images, α is equal to the number of views. Based on the double weights, the probability distribution of the characteristic field is defined in combination with the Gamma mixture model, where the product of the distribution scale parameter and the distribution shape parameter in the mixture model is the mean value.

Step S140, calculating the probability p(z|w) of the above image.

Assuming that the eigenvalues of each pixel in the image are independent of each other, the above probability rate is defined as:

Where p(z _i |w _il ) is the probability density of z _i defined by the double-weight Gamma mixture model, as follows:

Step S150, calculating the distribution function p(w) of the above-mentioned double weight w.

Specifically, based on the error sum of squares theory, combined with MRF to define the difference between the central pixel and its neighboring pixels, the distribution function p(w) of the double weight w is obtained as:

Among them, A is the normalization coefficient, which controls the clustering scale; η is the neighborhood effect coefficient, which characterizes the strength of neighborhood influence; N _i is a set of 8 neighborhood pixels centered on ( _xi , y _i ), and satisfies i '∈N _i , i'≠i.

Step S160, calculating a quality function L, which is a logarithmic function of the joint probability distribution function of the probability p(z|w) and the distribution function p(w).

Among them, the calculation of the quality function L specifically includes:

Calculate the joint probability distribution function p(z,w) of the probability p(z|w) and the distribution function p(w):

p(z,w)=p(z|w)p(w);

The quality function L is defined as the logarithmic function of p(z,w), that is, take the logarithm of the above formula:

L(w)=log p(z,w)=log p(z|w)+log p(w).

Step S170, updating the above-mentioned double weight w according to the gradient method.

w ^(t+1) = w ^(t) +ξΔw ^(t)

Among them, t is the number of iterations, ξ is the step size, and Δw ^(t) is the gradient, which is defined as the derivative of the above quality function L as follows:

Step S180, substituting the updated double weight w into the quality function L.

Step S190, repeat the above steps S130-S180, until |L ^(t+1) -L ^(t) | is less than the preset threshold ε, stop performing the above steps, and determine each pixel in the image according to the current double weight w category.

Determining the category to which each of the above pixels belongs specifically includes the following steps:

(1) Calculate the maximum value of the above double weights:

c _i =arg max{w _il ,l=1,...,k};

Among them, c _i is the label of the category to which the i-th pixel belongs;

(2) Use the maximum value as the class to which the pixel belongs.

From the definition of the double weight in step S120 above, it can be seen that it is a relationship matrix between pixels and categories, and the column corresponding to the maximum value in each row is the category to which the pixel belongs.

Step S200, outputting the segmentation result according to the class to which each pixel belongs.

Specifically, after determining which category each pixel of the image to be segmented belongs to through the above steps, the intensity of all pixels in the same category (that is, the gray value of the pixel) is averaged, and the average is used as the intensity value of this category to obtain the segmentation result. updating the intensities of all pixels in the same category with the mean value of the intensities of all pixels in the same category in the above image to obtain a mean value image; displaying the mean value image through a display device.

When the method of this embodiment is actually implemented, before performing the above steps after inputting the image to be segmented, it also includes the step of setting a constant, which specifically includes: the number of classes k, the Gamma distribution morphological parameter α, the normalization constant A and Neighborhood effect coefficient η.

In the multi-view SAR image segmentation method provided by the embodiment of the present invention, the image feature field is modeled by using a double-weight Gamma mixture model based on the generic field representation; Filed, MRF), combined with the error sum of squares theory, treat the central pixel category weight as the mean value, and construct the sum of squares function of the difference between it and the neighborhood pixel category weights to describe the difference of pixels in the neighborhood window, and use the error sum of squares to represent The difference can make full use of the category information and enhance the segmentation accuracy of the image; at the same time, when estimating the double weights in the Gamma mixture model, the embodiment of the present invention uses the gradient method to solve the double weights. details as follows:

(1) Use the Gamma mixture model to describe the probability distribution of the SAR image feature field, and use the category field to represent the relationship between pixels and categories. The relationship is represented by a matrix, and the category matrix is used as the double weight of the Gamma mixture model. The two angles of clustering and clustering work together on the Gamma distribution, which can better describe the characteristic field than the single weight that is only related to clustering in the traditional mixed model;

(2) Based on the error sum of squares and Markov Random Field (MRF) theory, the sum of the squares of the difference between the central pixel and its 8 neighboring pixels’ category weights is used to describe the degree of feature difference in the neighborhood window, and then the double weight is defined The distribution function of , when the sum of squares is larger, it means that the pixel feature difference in the window is greater, that is, the probability of belonging to different categories is greater;

(3) Use Bayesian theorem to define the joint probability distribution function of eigenfield and double weight, use its logarithmic function as the quality function, and use the gradient method to solve the double weight to obtain the best estimated value corresponding to the minimized quality function.

The effects of the embodiments of the present invention can be further illustrated by the following simulation experiments:

(1) Simulation experiment conditions

In this embodiment, the simulation is realized by using MATLAB 2011a software programming on a Windows 7 ultimate system with Core(TM) i5-3470 3.20GHz CPU.

The simulation data 1 is a simulated SAR image, which contains 3 homogeneous regions, which are obtained by adding random numbers from the Gamma distribution with a shape parameter of 4 and a scale parameter of 2, 10, and 20, respectively, to the standard template image. The simulation data 2 is the real SAR image, which contains two homogeneous regions. Since the real remote sensing image has no standard template, the hand-painted template is regarded as the standard template. The image size of the above simulation data is 128×128, and the total number of pixels of the image is n=16384.

(2) Simulation results

In order to prove the effectiveness of the algorithm of this embodiment, the template image is used as a standard to generate a confusion matrix for the segmentation results of the algorithm of this embodiment and the comparison algorithm, and calculate its user accuracy, product accuracy and Kappa value, so as to quantitatively analyze the method of this embodiment (as shown in Table 1), where "-" indicates that there is no such area in the image. It can be seen from Table 1 that the total precision and Kappa value of this embodiment are higher than those of the comparison algorithm, which accurately verifies the validity of the method of this embodiment from a numerical point of view. The method of this embodiment has better anti-noise performance, less misclassification of segmentation results, and accurate segmentation boundary fitting; while the comparison algorithm cannot effectively overcome complex noise, and there are many misclassified pixels in the segmentation results, resulting in extremely poor visual effects.

Table 1

Example 2

In combination with the foregoing embodiments, this embodiment provides a multi-view SAR image segmentation device, referring to the schematic structural diagram of the multi-view SAR image segmentation device shown in FIG. 2 , the device includes: a reading module 301, a dual weight initialization module 302, Weight parameter iterative update module 303 , category determination module 304 and display module 305 .

The details are as follows:

The reading module 301 is used for reading the multi-view SAR image to be segmented.

The multi-view SAR image to be segmented is represented by the feature field z:

z={z _i (x _i ,y _i ):i=1,…,n}

where i is the pixel index, n is the total number of pixels, z _i is the intensity of pixel i, ( _xi , y _i )∈D is the grid position of pixel i, and D is the image domain.

The dual weight initialization module 302 is configured to initialize the dual weight w.

In this embodiment, a double weight w is defined to represent the relationship between a pixel and a category, as follows:

w _i =(w _il :l=1,...,k)

Among them, w _il is the category weight of category l containing pixel i, which satisfies k is the number of classes. Among them, the initial double weight is randomly generated, the row represents the pixel, and the column represents the category. The value range of the elements in the matrix is 0-1, and the sum of each row is 1.

The weight parameter iterative update module 303 is used to repeatedly perform the following calculations:

Computes the gamma distribution scale parameter β.

Define the Gamma distribution scale parameter β _l as a function of the category weight w _il , as follows:

Among them, α is the shape parameter of Gamma distribution, and for multi-view SAR images, α is equal to the number of views. Based on the double weights, the probability distribution of the characteristic field is defined in combination with the Gamma mixture model, where the product of the distribution scale parameter and the distribution shape parameter in the mixture model is the mean value.

Calculate the probability p(z|w) of the image.

Assuming that the eigenvalues of each pixel in the image are independent of each other, the above probability rate is defined as:

Where p(z _i |w _il ) is the probability density of z _i defined by the double-weight Gamma mixture model, as follows:

Calculate the distribution function p(w) of the above double weight w.

Specifically, based on the error sum of squares theory, combined with MRF to define the difference between the central pixel and its neighboring pixels, the distribution function p(w) of the double weight w is obtained as:

Among them, A is the normalization coefficient, which controls the clustering scale; η is the neighborhood effect coefficient, which characterizes the strength of neighborhood influence; N _i is a set of 8 neighborhood pixels centered on ( _xi , y _i ), and satisfies i '∈N _i , i'≠i.

Calculate the quality function L, which is the logarithmic function of the joint probability distribution function of the probability p(z|w) and the distribution function p(w).

The above weight parameter iterative update module 303 includes: a first calculation unit for calculating the joint probability distribution function p(z,w) of the probability p(z|w) and the distribution function p(w):

p(z,w)=p(z|w)p(w);

The second calculation unit is used to take the logarithm of the above formula:

L(w)=log p(z,w)=log p(z|w)+log p(w).

The above dual weight w is updated according to the gradient method.

w ^(t+1) = w ^(t) +ξΔw ^(t)

Among them, t is the number of iterations, ξ is the step size, and Δw ^(t) is the gradient, which is defined as the derivative of the above quality function L as follows:

Substitute the updated double weight w into the quality function L;

Until |L ^(t+1) -L ^(t) | is less than the preset threshold ε, stop performing the above steps.

A category determining module 304, configured to determine the category to which each pixel in the image belongs according to the current dual weight w.

The above-mentioned category determination module 304 includes: a maximum value calculation unit, which is used to calculate the maximum value of the double weight:

c _i =arg max{w _il ,l=1,...,k}

Among them, c _i is the label of the category to which the i-th pixel belongs;

A class determining unit, configured to use the above maximum value as the class to which the pixel belongs.

From the definition of the above double weight, it can be seen that it is a relationship matrix between pixels and categories, and the column corresponding to the maximum value in each row is the category to which the pixel belongs.

The output module 305 is configured to output the segmentation result according to the category to which each pixel belongs.

Specifically, the above-mentioned output module 305 includes:

Mean value unit, for the mean value of the intensity of all pixels in the same category in the image to update the intensity of all pixels in the same category, to obtain the mean value image;

The display unit is used for displaying the mean value image through a display device.

When the device of this embodiment is actually implemented, it also includes: a constant setting module, which is used to set constants, and the above constants include: the number of classes k, the Gamma distribution shape parameter α, the normalization constant A and the neighborhood interaction coefficient η.

The implementation principle and technical effects of the multi-view SAR image segmentation device provided in this embodiment are the same as those of the foregoing embodiments. For brief description, for parts not mentioned in this embodiment, reference may be made to the corresponding content in the foregoing embodiments.

If the above functions are realized in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the essence of the technical solution of the present invention or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including Several instructions are used to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program codes. .

The above is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Anyone skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present invention. Should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims (10)

1. a kind of regarding SAR image segmentation method more, which is characterized in that including：

It reads to be split mostly regarding SAR image；It is described to be split mostly to be indicated using Characteristic Field z depending on SAR image：

Z={ z_i(x_i,y_i):I=1 ..., n }

Wherein, i is pixel index, and n is total pixel number, z_iFor the intensity of pixel i, (x_i, y_i) ∈ D be pixel i lattice site, D For image area；

The double weight w of initialization：

W=(w_il:L=1 ..., k)

Wherein, w_ilInclude the generic weight of pixel i for generic l, meetsK is class number；

Repeat following step：Calculate Gamma distribution scale parameters β_l：

Wherein, α is Gamma distributional pattern parameters；

Calculate the probability p (z | w) of described image；

Wherein p (z_i|w_il) it is the z defined with double weight Gamma mixed models_iProbability density, it is as follows：

Wherein Ga (z_i|β_l) it is with z_iFor independent variable, β_lFor the Gamma distribution probability density functions of scale parameter, Γ (α) is with α For the Gamma functions of parameter；

Calculate the distribution function p (w) of double weight w：

Wherein, A is normalization coefficient, and η is neighborhood function coefficient, N_iFor with (x_i,y_i) centered on 8 neighborhood territory pixel set, and it is full Sufficient i ' ∈ N_i,i'≠i；

Calculate merit functions L, the merit functions L is the probability p (z | w) and distribution function p (w) joint probability point The logarithmic function of cloth function；

Double weight w are updated according to gradient method：

w^(t+1)=w^(t)+ξΔw^(t)

Wherein, t is iterations, and ξ is step-length, Δ w^(t)It is as follows for gradient：

Updated double weight w are substituted into the merit functions L；

Until | L^(t+1)-L^(t)| when being less than preset threshold epsilon, stop executing above-mentioned steps, it is true according to current double weight w Determine the classification belonging to each pixel in described image；

Segmentation result is exported according to the classification belonging to each pixel.

2. according to the method described in claim 1, it is characterized in that, calculating merit functions L includes：

Calculate joint probability distribution function ps (z, w) of the probability p (z | w) with distribution function p (w)：

P (z, w)=p (z | w) p (w)；

Logarithm is taken to the joint probability distribution function：

L (w)=logp (z, w)=logp (z | w)+logp (w).

3. according to the method described in claim 1, it is characterized in that, determining that the classification belonging to the pixel includes：

Calculate the maximum value of double weights：

c_i=arg max { w_il, l=1 ..., k }；

Wherein, c_iFor the label of ith pixel generic；

Using the maximum value as the classification belonging to the pixel.

4. according to the method described in claim 1, it is characterized in that, according to the classification output segmentation knot belonging to each pixel Fruit includes：All pictures in the same category are updated with the mean value of the intensity of all pixels in same category in described image The intensity of element, obtains mean value image；The mean value image is shown by display device.

5. according to the method described in claim 1, it is characterized in that, further including：

Constant is set, and the constant includes：Class number k, Gamma distributional pattern parameter alpha, normaliztion constant A and neighborhood function coefficient η。

6. a kind of regarding SAR image segmenting device more, which is characterized in that including：

Read module, it is to be split mostly regarding SAR image for reading；It is described to be split mostly to be indicated using Characteristic Field z depending on SAR image：

Z={ z_i(x_i,y_i):I=1 ..., n }

Wherein, i is pixel index, and n is total pixel number, z_iFor the intensity of pixel i, (x_i, y_i) ∈ D be pixel i lattice site, D For image area；

Double weights initialisation modules, for initializing double weight w：

W=(w_il:L=1 ..., k)

Wherein, w_ilInclude the generic weight of pixel i for generic l, meetsK is class number；

Weight parameter iteration update module, for repeating following step：Calculate Gamma distribution scale parameters β_l：

Wherein, α is Gamma distributional pattern parameters；

Calculate the probability p (z | w) of described image；

Wherein p (z_i|w_il) it is the z defined with double weight Gamma mixed models_iProbability density, it is as follows：

Wherein Ga (z_i|β_l) it is with z_iFor independent variable, β_lFor the Gamma distribution probability density functions of scale parameter, Γ (α) is with α For the Gamma functions of parameter；

Calculate the distribution function p (w) of double weight w：

Wherein, A is normalization coefficient, and η is neighborhood function coefficient, N_iFor with (x_i,y_i) centered on 8 neighborhood territory pixel set, and it is full Sufficient i ' ∈ N_i,i'≠i；

Calculate merit functions L, the merit functions L is the probability p (z | w) and distribution function p (w) joint probability point The logarithmic function of cloth function；

Double weight w are updated according to gradient method：

w^(t+1)=w^(t)+ξΔw^(t)

Wherein, t is iterations, and ξ is step-length, Δ w^(t)It is as follows for gradient：

Updated double weight w are substituted into the merit functions L；

Until | L^(t+1)-L^(t)| when being less than preset threshold epsilon, stop executing above-mentioned steps；

Category determination module, for determining the classification in described image belonging to each pixel according to current double weight w；

Output module, for exporting segmentation result according to the classification belonging to each pixel.

7. device according to claim 6, which is characterized in that the weight parameter iteration update module includes：

First computing unit, the joint probability distribution function p (z, w) for calculating probability p (z | w) and distribution function p (w)：

P (z, w)=p (z | w) p (w)；

Second computing unit, for taking logarithm to the joint probability distribution function：

L (w)=logp (z, w)=logp (z | w)+logp (w).

8. device according to claim 6, which is characterized in that the category determination module includes：

Maximum value calculation unit, the maximum value for calculating double weights：

c_i=arg max { w_il, l=1 ..., k }；

Wherein, c_iFor the label of ith pixel generic；

Classification determination unit, for using the maximum value as the classification belonging to the pixel.

9. device according to claim 6, which is characterized in that the output module includes：

The mean value of equal value cell, the intensity for all pixels in same category in described image updates in the same category All pixels intensity, obtain mean value image；

Display unit shows the mean value image for passing through display device.

10. device according to claim 6, which is characterized in that further include：

Constant setup module, for constant to be arranged, the constant includes：Class number k, Gamma distributional pattern parameter alpha, normalization are normal Number A and neighborhood function coefficient η.