CN103020978A - SAR (synthetic aperture radar) image change detection method combining multi-threshold segmentation with fuzzy clustering - Google Patents

Abstract

The invention discloses an SAR (synthetic aperture radar) image change detection method combining multi-threshold segmentation with fuzzy clustering. The method mainly aims at overcoming the defect of existing fuzzy clustering algorithms and is used for SAR image change detection by combining multi-threshold segmentation with fuzzy clustering. The implementation steps of the method include: (1), subjecting two SAR images to median filtering; (2), calculating to obtain a logarithmic ratio differential image after normalization; (3), adopting the Otsu method based on standard particle swarm optimization to perform multi-threshold segmentation to the logarithmic ratio differential image after normalization; (4), initializing membership matrixes U0 and U1; (5), adopting the FLICM (fuzzy local information C-means) algorithm to perform fuzzy clustering to pixels which cannot be determined whether changes occur or not after multi-threshold segmentation; (6), deblurring; and (7), outputting change detection results. The multi-threshold segmentation and fuzzy clustering are combined for SAR image change detection, so that change detection time is reduced, and change detection accuracy is improved.

Description

SAR image change detection method in conjunction with many Threshold segmentations and fuzzy clustering

Technical field

The invention belongs to field of computer technology, further relate to the SAR image change detection method in conjunction with many Threshold segmentations and fuzzy clustering in the technical field of image processing.The present invention by two width of cloth not simultaneously the SAR image of phase obtain differential image, again differential image is carried out many Threshold segmentations and fuzzy clustering, realize the SAR Image Change Detection, can be used for atural object covering and utilization, Natural calamity monitoring and assessment, city planning, the fields such as map renewal.

Background technology

Synthetic-aperture radar (Synthetic Aperture Radar, SAR) technology is more and more widely used in recent years.The SAR image is compared with the ordinary optical remote sensing images, has the advantages that round-the-clockly to obtain, and along with the SAR image resolution ratio improves constantly, also more and more based on its image processing techniques.The change detection techniques of image refers to obtain the landforms change information of this area in these two periods by to the contrast and analysis of areal at two width of cloth remote sensing images of different time.At present the change detection techniques based on the SAR image has obtained many application, such as at aspects such as Forest cover change, Natural calamity monitoring and assessments.

Roughly there are two kinds of methods in zone for how to find out variation from two width of cloth SAR images: a kind of is relative method after the classification, is exactly first different SAR images to be classified separately, and then the result behind the match stop, finds out the zone of variation; Another method then is the differential image that obtains first them from different SAR images, differential image is analyzed again, thereby is found region of variation, and rear a kind of method is used morely at present.Processing reality to differential image will be carried out to it a kind of division of two classifications exactly, so clustering algorithm can be with the classification problem that solves differential image.

Fuzzy C-mean algorithm (Fuzzy C-Means, FCM) algorithm is one of most popular clustering algorithm, it adopts fuzzy theory to set up the uncertainty description of sample class, obtains sample to the degree of membership of all cluster centres by the optimization aim function, and then determines the classification of each sample.But because FCM does not consider the spatial information of sample, thus responsive to noise ratio, affected the effect of cluster.Fuzzy local message C average (Fuzzy Local Information C-Means, FLICM) algorithm is a kind of newer clustering algorithm, it has done improvement for FCM to the more sensitive shortcoming of noise ratio, in the objective function of cluster, introduced the local spatial information of sample, obtained the better Clustering Effect than FCM, but owing to all samples have all been increased the calculating of neighborhood information, compare with FCM, its operand is larger.

Patent " based on the SAR image change detection method of the quantum-inspired immune clone " (number of patent application 201010230980.6 of Xian Electronics Science and Technology University's application, publication number CN101908213A), the method is by quantum bit definitions cluster centre, the cluster centre that search is optimum also obtains global threshold, but the deficiency that the method exists is, only used the objective function of FCM to construct antibody affinity degree function, not the neighborhood information of considered pixel.

Summary of the invention

The object of the invention is to overcome the deficiency that above-mentioned prior art exists, many Threshold segmentations and fuzzy clustering are combined, be used for the SAR Image Change Detection, change detection time to reduce, and raising changes the precision that detects.

For achieving the above object, the present invention at first utilizes many Threshold segmentations to determine the classification of " simply " pixel when changing detection, leaves all the other " difficulty " pixels for the FLICM fuzzy clustering algorithm again and differentiates.

The present invention includes following steps:

(1) medium filtering

Choose 3 * 3 median filters commonly used two SAR images to be detected are carried out pre-service, obtain two width of cloth images behind the medium filtering;

(2) obtain logarithm ratio differential image after the normalization

2a) adopt logarithm ratio difference formula, by two width of cloth images behind the medium filtering, obtain the logarithm ratio differential image;

2b) adopt the normalization formula that the logarithm ratio differential image is carried out obtaining after the normalized logarithm ratio differential image after the normalization;

(3) many Threshold segmentations

Large Tianjin method that employing is optimized based on the standard particle group, logarithm ratio differential image after the normalization is carried out many Threshold segmentations, the pixel of cutting apart in the rear image is divided into three classes: the pixel that does not change, the pixel that changes has occured and can not determine whether the pixel that changes has occured;

(4) initialization degree of membership matrix

4a) initialization degree of membership matrix U ₀, U ₀The degree of membership that is under the jurisdiction of non-variation class of storage pixel, degree of membership represent with a real number in interval [0,1] upper value, belongs to the pixel that do not change behind many Threshold segmentations at U ₀In degree of membership get 1, belong to the pixel that occured to change behind many Threshold segmentations at U ₀In degree of membership get 0, belong to behind many Threshold segmentations and can not determine that the pixel that whether has occured to change is at U ₀In degree of membership then generate at random;

4b) initialization degree of membership matrix U ₁, U ₁Storage pixel be under the jurisdiction of the degree of membership that changes class, degree of membership represents with a real number in interval [0,1] upper value, belongs to the pixel that do not change behind many Threshold segmentations at U ₁In degree of membership get 0, belong to the pixel that occured to change behind many Threshold segmentations at U ₁In degree of membership get 1, belong to behind many Threshold segmentations and can not determine that the pixel that whether has occured to change is at U ₁In degree of membership then equal 1 and deduct this pixel at U ₀In degree of membership;

(5) fuzzy clustering

Belong to after adopting the FLICM algorithm to many Threshold segmentations and can not determine that the pixel that whether has occured to change carries out fuzzy clustering, iteration is upgraded until reach predetermined end condition, namely before and after the iteration maximum change amount of degree of membership less than 0.00001;

(6) deblurring

The degree of membership value that is under the jurisdiction of non-variation class and variation class according to each pixel in the logarithm ratio differential image after the normalization, pixel is judged to the ownership of that larger class of degree of membership value, thereby all pixels in the logarithm ratio differential image after the normalization have been divided into two classes: non-variation class and variation class;

(7) exporting change testing result.

The present invention has the following advantages compared with prior art:

1. the present invention at first utilizes many Threshold segmentations to determine the classification of " simply " pixel, leaving all the other " difficulty " pixels for the FLICM algorithm again differentiates, directly pixel is carried out cluster with existing FLICM algorithm and compare with differentiation, the present invention has reduced and has changed operand and the working time of detecting.

2. the present invention is when using the FLICM algorithm to carry out fuzzy clustering, local window N _iIn the classification of some non-central pixel determine in many Threshold segmentations stage, these neighborhood informations of determining have improved the effect that prior art is carried out fuzzy clustering, change the precision that detects so that the present invention has further improved.

Description of drawings

Fig. 1 is process flow diagram of the present invention;

Fig. 2 is the result who in the emulation experiment Ottawa area floods SAR image is changed detection;

Fig. 3 is the result who in the emulation experiment Mexico's area fire SAR image is changed detection;

Fig. 4 is the result who in the emulation experiment Bern Urban flood SAR image is changed detection.

Embodiment

With reference to Fig. 1, the SAR image change detection method in conjunction with many Threshold segmentations and fuzzy clustering among the present invention comprises the steps:

(1) medium filtering

Choose 3 * 3 median filters commonly used two SAR images to be detected are carried out pre-service, obtain two width of cloth images behind the medium filtering;

(2) obtain logarithm ratio differential image after the normalization

2a) adopt following logarithm ratio difference formula, by two width of cloth images behind the medium filtering, obtain the logarithm ratio differential image:

I ₃＝|log(I ₁+1)-log(I ₂+1)|

Wherein, I ₃The grey scale pixel value of expression logarithm ratio differential image, I ₁And I ₂The grey scale pixel value that represents respectively two width of cloth images behind the medium filtering;

2b) adopt following normalization formula that the logarithm ratio differential image is carried out obtaining after the normalized logarithm ratio differential image after the normalization:

$I_D=\frac{255\×(I_3-I_{\min })}{(I_{\max }-I_{\min })}$

Wherein, I _DThe grey scale pixel value of the logarithm ratio differential image after the expression normalization, I ₃The grey scale pixel value of expression logarithm ratio differential image, I _MaxExpression I ₃The gray-scale value of middle maximum, I _MinExpression I ₃The gray-scale value of middle minimum;

(3) many Threshold segmentations

The logarithm ratio differential image of large Tianjin method that employing is optimized based on the standard particle group after to normalization carries out many Threshold segmentations, and the pixel of cutting apart in the rear image is divided into three classes: the pixel that does not change, the pixel that changes has occured and can not determine whether the pixel that changes has occured;

The concrete steps of many Threshold segmentations are as follows:

Be a n dimension particle during the standard particle group optimizes being used for n threshold coding of split image 3a);

3b) Application standard particle group optimizing search is so that n threshold value of large Tianjin method objective function maximum, and Tianjin method objective function is as follows greatly:

Maximize?σ ²＝ω ₀(μ ₀-μ _T) ²+ω ₁(μ ₁-μ _T) ²+...+ω _n(μ _n-μ _T) ²

Logarithm ratio differential image after the normalization can be divided into n+1 part by n threshold value, in following formula, and σ ²Represent this n+1 the variance between the part, ω ₀, ω ₁..., ω _nRepresent respectively the probability that various piece occurs, μ ₀, μ ₁..., μ _nThe average gray that represents respectively various piece, μ _TThe average gray that represents whole image;

The more new formula of particle d dimension flying speed is in the standard particle group optimizes:

$v_i^d=\mathrm{\&omega;}{v}_i^d+{\mathrm{<figure-callout\; id="1"\; label="c"\; filenames="HSA00000839307100022.png"\; state="\{\{state\}\}">c</figure-callout>}}_1{\mathrm{rand}}_1^d({\mathrm{pBest}}_i^d-x_i^d)+c_2{\mathrm{rand}}_2^d({\mathrm{gBest}}^d-x_i^d)$

Wherein,

The d dimension flying speed of i particle in the expression population, ω represents inertia weight, c ₁And c ₂The expression study factor,

The d dimension position of i particle in the expression population, With

Two separate random numbers that are uniformly distributed between 0 and 1, pBest _iBe the historical optimal location of i particle, gBest is the historical optimal location of whole population;

The more new formula of particle d dimension position is in the standard particle group optimizes:

$x_i^d=x_i^d+v_i^d$

Wherein,

The d dimension position of i particle in the expression population,

The d dimension flying speed of i particle in the expression population;

The search of 3c) optimizing by the standard particle group is found so that n threshold value of large Tianjin method objective function maximum, and the logarithm ratio differential image of this n threshold value after normalization is divided into n+1 part, is designated as according to gray-scale value order from small to large: C ₀, C ₁..., C _nC ₀In pixel be classified as the pixel that does not change, C _nIn pixel be classified as the pixel that changes occured, and C ₁To C _N-1In pixel be classified as to can not determine whether the pixel that changes has occured, thereby the pixel in the logarithm ratio differential image after the normalization has been divided into three classes;

(4) initialization degree of membership matrix

4a) initialization degree of membership matrix U ₀, U ₀The degree of membership that is under the jurisdiction of non-variation class of storage pixel, degree of membership represent with a real number in interval [0,1] upper value, belongs to the pixel that do not change behind many Threshold segmentations at U ₀In degree of membership get 1, belong to the pixel that occured to change behind many Threshold segmentations at U ₀In degree of membership get 0, belong to behind many Threshold segmentations and can not determine that the pixel that whether has occured to change is at U ₀In degree of membership then generate at random;

4b) initialization degree of membership matrix U ₁, U ₁Storage pixel be under the jurisdiction of the degree of membership that changes class, degree of membership represents with a real number in interval [0,1] upper value, belongs to the pixel that do not change behind many Threshold segmentations at U ₁In degree of membership get 0, belong to the pixel that occured to change behind many Threshold segmentations at U ₁In degree of membership get 1, belong to behind many Threshold segmentations and can not determine that the pixel that whether has occured to change is at U ₁In degree of membership then equal 1 and deduct this pixel at U ₀In degree of membership;

(5) fuzzy clustering

Belong to after adopting the FLICM algorithm to many Threshold segmentations and can not determine that the pixel that whether has occured to change carries out fuzzy clustering, iteration is upgraded until reach predetermined end condition, namely before and after the iteration maximum change amount of degree of membership less than 0.00001;

5a) the FLICM algorithm uses degree of membership to represent that pixel may belong to the degree of certain classification, and it has considered gradation of image information and spatial neighborhood information simultaneously in cluster process, and the objective function of FLICM is as follows:

$J=\underset{i\&Element;N}{\mathrm{\&Sigma;}}\underset{k=1}{\overset{C}{\mathrm{\&Sigma;}}}[u_{\mathrm{ki}}^m{d}^2(x_i,v_k)+G_{\mathrm{ki}}]$

Wherein, J represents target function value, and N represents to belong to set that can not determine the pixel composition whether variation has occured behind many Threshold segmentations, and C is that value is 2 class number, u _KiRepresent the degree of membership of i pixel on the k class, x _iThe gray-scale value that represents i pixel, v _kThe cluster centre that represents the k class, d ²(x _i, v _k) representing i pixel to the distance metric of the cluster centre of k class, m represents that value is 2 FUZZY WEIGHTED index, G _KiRepresent the fuzzy coefficient between the cluster centre of i pixel and k class, shown in it is defined as follows:

$G_{\mathrm{ki}}=\underset{\underset{i\&NotEqual;j}{j\&Element;N_i}}{\mathrm{\&Sigma;}}\frac{1}{e_{\mathrm{ij}}+1}{(1-u_{\mathrm{kj}})}^m{d}^2(x_j,v_k)$

Wherein, G _KiRepresent the fuzzy coefficient between the cluster centre of i pixel and k class, N _iThe local window of the size of expression centered by i pixel as 3 * 3, x _jThe gray-scale value and this pixel that represent j pixel are to fall into local window N _iIn non-central pixel, v _kThe cluster centre that represents the k class, u _KjRepresent the degree of membership of j pixel on the k class, e _IjRepresent the space Euclidean distance between i pixel and j the pixel, d ²(x _j, v _k) representing j pixel to the distance metric of the cluster centre of k class, m represents that value is 2 FUZZY WEIGHTED index;

5b) the FLICM algorithm adopts the mode of iterative computation to obtain so that one group of degree of membership value of its objective function minimum, and the iterative computation formula of degree of membership is as follows in the FLICM algorithm:

$u_{\mathrm{ki}}=\frac{1}{\underset{t=1}{\overset{C}{\mathrm{\&Sigma;}}}{\left(\frac{d^2(x_i,v_k)+G_{\mathrm{ki}}}{d^2(x_i,v_t)+G_{\mathrm{ti}}}\right)}^{\frac{1}{m-1}}}$

Wherein, u _KiRepresent the degree of membership of i pixel on the k class, C is that value is 2 class number, x _iThe gray-scale value that represents i pixel, v _kThe cluster centre that represents the k class, v _tThe cluster centre that represents the t class, G _KiRepresent the fuzzy coefficient between the cluster centre of i pixel and k class, G _TiRepresent the fuzzy coefficient between the cluster centre of i pixel and t class, d ²(x _i, v _k) represent that i pixel is to the distance metric of the cluster centre of k class, d ²(x _i, v _t) representing i pixel to the distance metric of the cluster centre of t class, m represents that value is 2 FUZZY WEIGHTED index;

5c) the iterative computation formula of cluster centre is as follows in the FLICM algorithm:

$v_k=\frac{\underset{i\&Element;N}{\mathrm{\&Sigma;}}{u}_{\mathrm{ki}}^m{x}_i}{\underset{i\&Element;N}{\mathrm{\&Sigma;}}{u}_{\mathrm{ki}}^m}$

Wherein, v _kThe cluster centre that represents the k class, N represent to belong to set that can not determine the pixel composition whether variation has occured, u behind many Threshold segmentations _KiRepresent the degree of membership of i pixel on the k class, x _iThe gray-scale value that represents i pixel, m represent that value is 2 FUZZY WEIGHTED index;

(6) deblurring

The degree of membership value that is under the jurisdiction of non-variation class and variation class according to each pixel in the logarithm ratio differential image after the normalization, pixel is judged to the ownership of that larger class of degree of membership value, thereby all pixels in the logarithm ratio differential image after the normalization have been divided into two classes: non-variation class and variation class;

(7) exporting change testing result.

Effect of the present invention can further specify by following emulation experiment:

1. emulation experiment condition

The emulation experiment environment: operating system is Windows XP, and CPU is AMD Athlon1.60GHz, in save as 1.75GB, programming platform is Visual C++6.0.

Emulation experiment one is the SAR Image Change Detection of Ottawa area floods, the image size is 290 * 350, Fig. 2 (a) is the SAR image in May, 1997, Fig. 2 (b) is the SAR image in August, 1997, Fig. 2 (c) is that corresponding the variation detects reference diagram, Fig. 2 (d) is the variation testing result of FCM algorithm, and Fig. 2 (e) is the variation testing result of FLICM algorithm, and Fig. 2 (f) is variation testing result of the present invention.

Emulation experiment two is SAR Image Change Detection of Mexico's area fire, the image size is 512 * 512, Fig. 3 (a) is the SAR image in April, 2000, Fig. 3 (b) is the SAR image in May, 2002, Fig. 3 (c) changes to detect reference diagram, Fig. 3 (d) is the variation testing result of FCM algorithm, and Fig. 3 (e) is the variation testing result of FLICM algorithm, and Fig. 3 (f) is variation testing result of the present invention.

Emulation experiment three is SAR Image Change Detection of Bern Urban flood, the image size is 301 * 301, Fig. 4 (a) is the SAR image in April, 1999, Fig. 4 (b) is the SAR image in May, 1999, Fig. 4 (c) changes to detect reference diagram, Fig. 4 (d) is the variation testing result of FCM algorithm, and Fig. 4 (e) is the variation testing result of FLICM algorithm, and Fig. 4 (f) is variation testing result of the present invention.

2. emulation experiment content and result

Parameter in the emulation experiment arranges as follows: the threshold number n of many Threshold segmentations is set to 3 (foundation of selection is analyzed in the emulation experiment in the threshold number parameter of back and provided) among the present invention; Two study factors of Plays particle group optimizing of the present invention all are made as 2.0, and inertia weight is made as along with search procedure from 0.9 to 0.4 linear the reduction, and the end condition that the standard particle group optimizes is that objective function evaluation number of times reaches 5000 times; The end condition of the fuzzy clustering process in the emulation experiment in three kinds of methods all is made as the maximum change amount of degree of membership before and after the iteration less than 0.00001.Data in the following emulation experiment are the average result behind the independent operating 20 times.

One: three kind of method of emulation experiment shown in Fig. 2 (d) to (f), sees Table 1 to the concrete comparative analysis of testing result to the variation testing result of this problem.As can be seen from Figure 2, testing result of the present invention and reference diagram are the most approaching.As can be seen from Table 1, the undetected pixel count of the present invention has lacked 1246 and 1309 than FCM and FLICM respectively, and total erroneous pixel number has also lacked 1003 and 372 than both respectively.Therefore from working time, can find out simultaneously, although the present invention has increased the process of many Threshold segmentations, also reduce the task amount that fuzzy clustering is carried out in the back, so that carry out cluster than direct use FLICM algorithm working time is still fast a lot.

Table 1Ottawa area floods change testing result

Two: three kinds of methods of emulation experiment shown in Fig. 3 (d) to (f), see Table 2 to the concrete comparative analysis of testing result to the variation testing result of this problem.As can be seen from Table 2, the undetected pixel count of the present invention has lacked 623 and 530 than FCM and FLICM respectively, and total erroneous pixel number has also lacked 149 and 383 than both respectively.The present invention will only consider that many Threshold segmentations of half-tone information and the FLICM algorithm of emphasis consideration neighborhood information organically combine, and promote noiseproof feature, greatly reduce undetected number, be better than two kinds of methods that contrast so that totally detect performance.Be about as can be seen from Table 2 simultaneously 30% of FLICM algorithm working time of the present invention.

Table 2 Mexico area fire changes testing result

Three: three kinds of methods of emulation experiment shown in Fig. 4 (d) to (f), see Table 3 to the concrete comparative analysis of testing result to the variation testing result of this problem.As can be seen from Table 3, change test problems for this, the present invention still has apparent in view advantage undetected aspect several, thereby total erroneous pixel number is also less, only has 112, and false detection rate only has 0.12%, and working time is also very short, less than 25% of FLICM algorithm.

Table 3Bern Urban flood changes testing result

Use following formula can calculate the present invention and on average reduced about 70% than FLICM algorithm the working time in above-mentioned three groups of emulation experiments:

$T=\frac{T_2-T_1}{T_2}$

Wherein, T represents that the present invention compares the minimizing ratio of working time, T with the FLICM algorithm ₁Represent working time of the present invention, T ₂The working time of expression FLICM algorithm.

3. the parameter of threshold number is analyzed emulation experiment

The selection of threshold number n has considerable influence to changing the effect that detects when carrying out many Threshold segmentations among the present invention.In order reasonably to select the size of threshold number n, threshold number n has been carried out the parameter analysis, 10 the inventive method of independent operating under different threshold number, the testing result after average is as shown in table 4.As can be seen from Table 4, the too small meeting of threshold number n obviously increases and changes the false detection rate that detects; The excessive increase that then can cause working time of threshold number n.After considering, it is more reasonably to select that threshold number n gets 3.

The parameter of threshold number n among table 4 the present invention is analyzed the simulation experiment result

Claims (5)

1. the SAR image change detection method in conjunction with many Threshold segmentations and fuzzy clustering comprises the steps:

(1) medium filtering

Choose 3 * 3 median filters commonly used two SAR images to be detected are carried out pre-service, obtain two width of cloth images behind the medium filtering;

(2) obtain logarithm ratio differential image after the normalization

2a) adopt logarithm ratio difference formula, by two width of cloth images behind the medium filtering, obtain the logarithm ratio differential image;

2b) adopt the normalization formula that the logarithm ratio differential image is carried out obtaining after the normalized logarithm ratio differential image after the normalization;

(3) many Threshold segmentations

Large Tianjin method that employing is optimized based on the standard particle group, logarithm ratio differential image after the normalization is carried out many Threshold segmentations, the pixel of cutting apart in the rear image is divided into three classes: the pixel that does not change, the pixel that changes has occured and can not determine whether the pixel that changes has occured;

(4) initialization degree of membership matrix

4a) initialization degree of membership matrix U ₀, U ₀The degree of membership that is under the jurisdiction of non-variation class of storage pixel, degree of membership represent with a real number in interval [0,1] upper value, belongs to the pixel that do not change behind many Threshold segmentations at U ₀In degree of membership get 1, belong to the pixel that occured to change behind many Threshold segmentations at U ₀In degree of membership get 0, belong to behind many Threshold segmentations and can not determine that the pixel that whether has occured to change is at U ₀In degree of membership then generate at random;

4b) initialization degree of membership matrix U ₁, U ₁Storage pixel be under the jurisdiction of the degree of membership that changes class, degree of membership represents with a real number in interval [0,1] upper value, belongs to the pixel that do not change behind many Threshold segmentations at U ₁In degree of membership get 0, belong to the pixel that occured to change behind many Threshold segmentations at U ₁In degree of membership get 1, belong to behind many Threshold segmentations and can not determine that the pixel that whether has occured to change is at U ₁In degree of membership then equal 1 and deduct this pixel at U ₀In degree of membership;

(5) fuzzy clustering

Belong to after adopting the FLICM algorithm to many Threshold segmentations and can not determine that the pixel that whether has occured to change carries out fuzzy clustering, iteration is upgraded until reach predetermined end condition, namely before and after the iteration maximum change amount of degree of membership less than 0.00001;

(6) deblurring

The degree of membership value that is under the jurisdiction of non-variation class and variation class according to each pixel in the logarithm ratio differential image after the normalization, pixel is judged to the ownership of that larger class of degree of membership value, thereby all pixels in the logarithm ratio differential image after the normalization have been divided into two classes: non-variation class and variation class;

(7) exporting change testing result.

2. the SAR image change detection method in conjunction with many Threshold segmentations and fuzzy clustering according to claim 1 is characterized in that step 2a) described in logarithm ratio difference formula as follows:

I ₃＝|log(I ₁+1)-log(I ₂+1)|

Wherein, I ₃The grey scale pixel value of expression logarithm ratio differential image, I ₁And I ₂The grey scale pixel value that represents respectively two width of cloth images behind the medium filtering.

3. the SAR image change detection method in conjunction with many Threshold segmentations and fuzzy clustering according to claim 1 is characterized in that step 2b) described in the normalization formula as follows:

$I_D=\frac{255\&times;(I_3-I_{\min })}{(I_{\max }-I_{\min })}$

Wherein, I _DThe grey scale pixel value of the logarithm ratio differential image after the expression normalization, I ₃The grey scale pixel value of expression logarithm ratio differential image, I _MaxExpression I ₃The gray-scale value of middle maximum, I _MinExpression I ₃The gray-scale value of middle minimum.

4. the SAR image change detection method in conjunction with many Threshold segmentations and fuzzy clustering according to claim 1, it is characterized in that, the concrete steps that the logarithm ratio differential image of the large Tianjin method optimized based on the standard particle group described in the step (3) after to normalization carries out many Threshold segmentations are as follows:

The first step is a n dimension particle during the standard particle group optimizes being used for n threshold coding of split image;

Second step, Application standard particle group optimizing search are so that n threshold value of large Tianjin method objective function maximum, and Tianjin method objective function is as follows greatly:

Maximize?σ ²＝ω ₀(μ ₀-μ _T) ²+ω ₁(μ ₁-μ _T) ²+...+ω _n(μ _n-μ _T) ²

Logarithm ratio differential image after the normalization can be divided into n+1 part by n threshold value, in following formula, and σ ²Represent this n+1 the variance between the part, ω ₀, ω ₁..., ω _nRepresent respectively the probability that various piece occurs, μ ₀, μ ₁..., μ _nThe average gray that represents respectively various piece, μ _TThe average gray that represents whole image;

The more new formula of particle d dimension flying speed is in the standard particle group optimizes:

$v_i^d=\mathrm{\&omega;}{v}_i^d+c_1{\mathrm{rand}}_1^d({\mathrm{pBest}}_i^d-x_i^d)+c_2{\mathrm{rand}}_2^d({\mathrm{gBest}}^d-x_i^d)$

Wherein,

The d dimension flying speed of i particle in the expression population, ω represents inertia weight, c ₁And c ₂The expression study factor,

The d dimension position of i particle in the expression population,

With

Two separate random numbers that are uniformly distributed between 0 and 1, pBest _iBe the historical optimal location of i particle, gBest is the historical optimal location of whole population;

The more new formula of particle d dimension position is in the standard particle group optimizes:

$x_i^d=x_i^d+v_i^d$

Wherein,

The d dimension position of i particle in the expression population, The d dimension flying speed of i particle in the expression population;

The 3rd step, by the search that the standard particle group optimizes, find so that n threshold value of large Tianjin method objective function maximum, the logarithm ratio differential image of this n threshold value after normalization is divided into n+1 part, is designated as according to gray-scale value order from small to large: C ₀, C ₁..., C _nC ₀In pixel be classified as the pixel that does not change, C _nIn pixel be classified as the pixel that changes occured, and C ₁To C _N-1In pixel be classified as to can not determine whether the pixel that changes has occured, thereby the pixel in the logarithm ratio differential image after the normalization has been divided into three classes.

5. the SAR image change detection method in conjunction with many Threshold segmentations and fuzzy clustering according to claim 1 is characterized in that, being implemented as follows of the FLICM fuzzy clustering algorithm described in the step (5):

The FLICM algorithm uses degree of membership to represent that pixel may belong to the degree of certain classification, and it has considered gradation of image information and spatial neighborhood information simultaneously in cluster process, and the objective function of FLICM is as follows:

$J=\underset{i\&Element;N}{\mathrm{\&Sigma;}}\underset{k=1}{\overset{C}{\mathrm{\&Sigma;}}}[u_{\mathrm{ki}}^m{d}^2(x_i,v_k)+G_{\mathrm{ki}}]$

Wherein, J represents target function value, and N represents to belong to set that can not determine the pixel composition whether variation has occured behind many Threshold segmentations, and C is that value is 2 class number, u _KiRepresent the degree of membership of i pixel on the k class, x _iThe gray-scale value that represents i pixel, v _kThe cluster centre that represents the k class, d ²(x _i, v _k) representing i pixel to the distance metric of the cluster centre of k class, m represents that value is 2 FUZZY WEIGHTED index, G _KiRepresent the fuzzy coefficient between the cluster centre of i pixel and k class, shown in it is defined as follows:

$G_{\mathrm{ki}}=\underset{\underset{i\&NotEqual;j}{j\&Element;N_i}}{\mathrm{\&Sigma;}}\frac{1}{e_{\mathrm{ij}}+1}{(1-u_{\mathrm{kj}})}^m{d}^2(x_j,v_k)$

Wherein, G _KiRepresent the fuzzy coefficient between the cluster centre of i pixel and k class, N _iThe local window of the size of expression centered by i pixel as 3 * 3, x _jThe gray-scale value and this pixel that represent j pixel are to fall into local window N _iIn non-central pixel, v _kThe cluster centre that represents the k class, u _KjRepresent the degree of membership of j pixel on the k class, e _IjRepresent the space Euclidean distance between i pixel and j the pixel, d ²(x _j, v _k) representing j pixel to the distance metric of the cluster centre of k class, m represents that value is 2 FUZZY WEIGHTED index;

The FLICM algorithm adopts the mode of iterative computation to obtain so that one group of degree of membership value of its objective function minimum, and the iterative computation formula of degree of membership is as follows in the FLICM algorithm:

$u_{\mathrm{ki}}=\frac{1}{\underset{t=1}{\overset{C}{\mathrm{\&Sigma;}}}{\left(\frac{d^2(x_i,v_k)+G_{\mathrm{ki}}}{d^2(x_i,v_t)+G_{\mathrm{ti}}}\right)}^{\frac{1}{m-1}}}$

Wherein, u _KiRepresent the degree of membership of i pixel on the k class, C is that value is 2 class number, x _iThe gray-scale value that represents i pixel, v _kThe cluster centre that represents the k class, v _tThe cluster centre that represents the t class, G _KiRepresent the fuzzy coefficient between the cluster centre of i pixel and k class, G _TiRepresent the fuzzy coefficient between the cluster centre of i pixel and t class, d ²(x _i, v _k) represent that i pixel is to the distance metric of the cluster centre of k class, d ²(x _i, v _t) representing i pixel to the distance metric of the cluster centre of t class, m represents that value is 2 FUZZY WEIGHTED index;

The iterative computation formula of cluster centre is as follows in the FLICM algorithm:

$v_k=\frac{\underset{i\&Element;N}{\mathrm{\&Sigma;}}{u}_{\mathrm{ki}}^m{x}_i}{\underset{i\&Element;N}{\mathrm{\&Sigma;}}{u}_{\mathrm{ki}}^m}$

Wherein, v _kThe cluster centre that represents the k class, N represent to belong to set that can not determine the pixel composition whether variation has occured, u behind many Threshold segmentations _KiRepresent the degree of membership of i pixel on the k class, x _iThe gray-scale value that represents i pixel, m represent that value is 2 FUZZY WEIGHTED index;

Belong to after using the FLICM algorithm to many Threshold segmentations and can not determine that the pixel that whether has occured to change carries out fuzzy clustering, until reach predetermined end condition, namely before and after the iteration maximum change amount of degree of membership less than 0.00001.

Images