CN104751185A - SAR image change detection method based on mean shift genetic clustering - Google Patents

Abstract

The present invention proposes a SAR image change detection method based on mean shift genetic clustering, and its realization steps are: (1) import image; (2) construct difference image; (3) mean shift filter; (4) genetic fuzzy clustering class; (5) segment the difference image; (6) output the result. The present invention can not only effectively reduce the influence of the area between the change class and the non-change class on the detection result of the synthetic aperture radar SAR image change, suppress the inherent noise in the back synthetic aperture radar SAR image, but also combine the mean shift filter, fuzzy aggregation The local optimum of the class and the global optimization ability of the genetic algorithm accelerate the convergence speed of the algorithm, reduce the missing information in the detection results, and have high change detection accuracy.

Description

SAR Image Change Detection Method Based on Mean Shift Genetic Clustering

technical field

The invention belongs to the technical field of image processing, and further relates to a synthetic aperture radar (Synthetic Aperture Radar, SAR) image change detection method based on mean shift genetic clustering in the technical field of image change detection. The invention can be applied to the fields of dynamic detection of lake water level, dynamic detection of crop growth state, urban planning, military reconnaissance, etc., to detect changes of ground objects over time.

Background technique

SAR image change detection is to analyze multiple SAR images in the same area at different times, and use the gray value of the difference image to divide the image into a changing area and an invariant area, and detect the information that the ground objects in the area change with time. Synthetic aperture radar (SAR) has the characteristics of all-weather and all-weather, is not affected by the weather, and has a certain penetration ability. It is a good source of change detection images. Research on SAR image change detection technology has a very broad application prospect.

Fuzzy C-means clustering is the most widely used cluster-based change detection method. In recent years, a series of improved methods based on fuzzy C-means clustering have been proposed. Maoguo Gong, Zhiqiang Zhou, and JingjingMa proposed an improved The FLICM (the reformulated FLICM, RFLICM) SAR image change detection method, compared with the existing change detection method based on fuzzy C-means clustering, this method more accurately solves the problem of image change detection. There is still room for improvement in terms of accuracy and computing speed. First, RFLICM randomly selects the initial clustering center, which makes the method very sensitive to the initial center point of clustering. Moreover, RFLICM uses the objective function as the reference point for clustering, so it is easy to fall into local optimum.

Xidian University disclosed a genetic kernel fuzzy clustering SAR image change detection method in its patent application "A SAR image change detection method based on genetic kernel fuzzy clustering" (patent application number: 201410497802.8, publication number: CN104268574A) Detection method. This method calculates the difference map and gray matrix of two SAR images, uses genetic fuzzy clustering to obtain the population, calculates the segmentation threshold according to the population, and completes the segmentation of the difference map according to the segmentation threshold, and obtains the result of change detection. This method combines the global search ability of the genetic algorithm and the local search ability of the kernel fuzzy clustering algorithm, accelerates the convergence speed of the algorithm, and effectively reduces the operation speed of the algorithm. However, this method still has the disadvantages that it is difficult to choose a suitable kernel function, and it cannot remove the inherent noise in SAR images well, and it is sensitive to noise points, which reduces the accuracy of change detection.

Wang Haoran disclosed a SAR image change detection method based on non-mean filtering in his patent application "SAR image change detection method based on non-local mean value" (patent application number: 201310529323.5, publication number: CN103927737A). The method includes preprocessing two SAR images acquired at different times in the same region; using the preprocessed two SAR images to construct a ratio difference image map; traversing each pixel of the ratio difference image map, and calculating the smoothness index matrix of each pixel ; Perform non-local mean filtering on the two SAR images after preprocessing, and then perform a ratio operation to obtain a non-local mean filtering ratio map; use the smoothing index as a weight, sum the ratio difference image map and the non-local mean filtering ratio image to obtain the final A difference image map; using the fuzzy local C-means clustering method to segment the final difference image map to obtain a change detection result map. Although this method effectively suppresses the noise, it can better represent the real change information and improve the accuracy of the change detection results. However, there are still shortcomings in that it takes a lot of computing time to complete the similarity calculation and search between different pixels, the time complexity is high, and it cannot effectively reduce the change of the region pair between the changed class and the non-changed class. The impact of detection results reduces the accuracy of change detection.

Contents of the invention

The purpose of the present invention is to overcome the above-mentioned deficiencies in the prior art, and propose a SAR image change detection method based on mean shift genetic clustering. Combining the mean shift filter, the local optimum of the fuzzy clustering algorithm and the global optimization ability of the genetic algorithm, the inherent noise in the synthetic aperture radar SAR image is well removed, and the missed detection information in the detection result is reduced. The change detection accuracy of the algorithm speeds up the convergence speed of the algorithm.

The idea of the present invention to achieve the above purpose is: after constructing the difference image, firstly perform mean shift filtering on the difference image to obtain the difference image after denoising, and then perform fuzzy genetic clustering on the difference image after denoising to obtain the difference image after denoising The clustering center of the image, and then use the clustering center of the difference image after denoising to obtain the segmentation threshold, segment the difference image after denoising, and obtain the change detection result map of the synthetic aperture radar SAR image.

In order to achieve the above object, the concrete realization steps of the present invention are as follows:

A SAR image change detection method based on mean shift genetic clustering, comprising the steps of:

(1) Import image:

Import two synthetic aperture radar SAR images of the same size acquired at different times in the same area;

(2) Construct difference image:

(2a) Calculate the neighborhood difference image of two synthetic aperture radar SAR images;

(2b) Calculate the neighborhood ratio image of two synthetic aperture radar SAR images;

(2c) Calculate the normalized difference image after the fusion of the neighborhood difference image of the two synthetic aperture radar SAR images and the neighborhood ratio image of the two synthetic aperture radar SAR images;

(3) Mean shift filtering:

(3a) Using the kernel density estimation method, obtain the kernel density value of each pixel in the difference image after fusion of the neighborhood difference image of the two synthetic aperture radar SAR images and the neighborhood ratio image of the two synthetic aperture radar SAR images;

(3b) Subtract the gray value of each pixel of the fused normalized difference image from the kernel density value of each pixel of the fused difference image obtained in step (3a), and take the result of the subtraction absolute value;

(3c) judging whether the absolute value is less than the set threshold, if so, execute step (3d); otherwise, execute step (3a);

(3d) Using the absolute value as the pixel gray value of the difference image to obtain the difference image after denoising;

(4) Genetic fuzzy clustering:

(4a) Initialize the first-generation population of the difference image after denoising;

(4b) According to the following formula, calculate the global division index of the first-generation population of the difference image after denoising:

$\mathrm{<span\; class="notranslate">\; J</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; c</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}^{\mathrm{<span\; class="notranslate">\; m</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>{{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ;</span>$

Among them, J(t) represents the global division index of the first generation population of the difference image after denoising, t represents the evolution times of the first generation population of the difference image after denoising, Σ represents the sum operation, and i represents the difference after denoising The number of the i-th class in the cluster center of the image, c indicates the number of clusters of the difference image after denoising, j indicates the number of the jth pixel of the difference image after denoising, n indicates the difference image after denoising The total number of pixels, μ _ij represents the degree of membership of the jth pixel of the denoised difference image belonging to the i-th class in the cluster center of the denoised difference image, and the value range of μ _ij is [0, 1] , and must satisfy Constraint conditions, m represents the fuzzy index factor, m is a positive number with a value greater than 1, d _ij ² represents the jth pixel point of the difference image after denoising to the i-th class in the cluster center of the difference image after denoising Distance, H(j) represents the gray value of the jth pixel of the fused normalized difference image;

(4c) According to the following formula, calculate the individual fitness of the first generation population of the difference image after denoising:

$\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; J</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; ;</span>$

Among them, f(t) represents the individual fitness of the first generation population of the difference image after denoising, t represents the evolution times of the first generation population of the difference image after denoising, J(t) represents the global division index;

(4d) A next-generation population of denoised difference images is obtained using genetic manipulation;

(4e) Calculate the fitness of each individual in the next generation population of the difference image after denoising according to the method in step (4c);

(4f) Determine whether the next-generation population of the difference image after denoising is stable, if so, perform step (4g); otherwise, perform step (4d);

(4g) calculating the fitness maximum value of the next generation population of the difference image after denoising;

(4h) taking the individual with the maximum fitness value of the next generation population of the difference image after denoising as the cluster center of the difference image after denoising;

(5) Segment difference image:

(5a) Calculate the elements of the membership degree matrix of the difference image after denoising according to the following formula:

${\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; =</span>{<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; c</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; (</span>\frac{{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; Y</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; V</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{{{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; kj</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; )</span>}^{\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; )</span>}^{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ;</span>$

Among them, μ _ij represents the element of the membership matrix of the difference image after denoising, i represents the serial number of the i-th class in the cluster center of the difference image after denoising, and j represents the number of the j-th pixel of the difference image after denoising Sequence number, Σ represents the summation operation, k represents the sequence number of the kth class in the clustering center of the difference image after denoising, c represents the number of clusters of the difference image after denoising, ||·|| represents the Euclidean distance operation, Y _j represents the gray value of the jth pixel of the difference image after denoising, V _i represents the ith cluster center of the difference image after denoising, d _kj ² represents the jth pixel point of the difference image after denoising to The cluster center distance of the kth class of the difference image after denoising, m represents the fuzzy index factor, and m is a positive number with a value greater than 1;

(5b) Find the number of rows where the minimum value of all elements of the degree of membership matrix of the difference image after denoising is located;

(5c) the number of rows where the minimum value obtained in step (5b) is used as the segmentation threshold of the difference image after denoising;

(5d) Determine whether the gray value of each pixel of the difference image after denoising is less than the segmentation threshold of the difference image after denoising, if so, perform step (5e); otherwise, perform step (5f);

(5e) classify the segmentation threshold pixel points smaller than the difference image after denoising into the non-change class of the difference image after denoising;

(5f) classify the segmentation threshold pixel points greater than or equal to the difference image after denoising into the change class of the difference image after denoising;

(6) Output result:

For the non-change class of the difference image after denoising and the change class of the difference image after denoising, the change detection result maps of two synthetic aperture radar SAR images are obtained.

Compared with the prior art, the present invention has the following advantages:

First, because the present invention adopts the method of fusing the neighborhood difference image of two synthetic aperture radar SAR images and the neighborhood ratio image of two synthetic aperture radar SAR images in the process of constructing the difference image, it is very large The background information is suppressed to a certain extent, and the speckle noise is suppressed, which effectively reduces the influence of the area between the change class and the non-change class on the change detection results of the synthetic aperture radar SAR image, and overcomes the existing technology's influence on the change class and the non-change class. The region between the change classes has the disadvantage of more missed detection information, so that the present invention improves the accuracy of the change detection of the synthetic aperture SAR image.

Second, because the present invention uses the kernel density estimation method to calculate the difference after the fusion of the neighborhood difference image of two synthetic aperture radar SAR images and the neighborhood ratio image of two synthetic aperture radar SAR images in the process of mean shift filtering The kernel density value of each pixel of the image suppresses the inherent noise in the synthetic aperture SAR image, overcomes the shortcoming of the prior art that is sensitive to noise points, and improves the robustness of the present invention.

Third, because the present invention uses genetic operations in the process of genetic fuzzy clustering, which speeds up the convergence speed of the algorithm and overcomes the shortcomings of the prior art that consume a large amount of computing time and have more missing information, making the present invention Improves the accuracy of change detection in synthetic aperture radar SAR images.

Description of drawings

Fig. 1 is a flow chart of the present invention;

Fig. 2 is a simulation diagram of the present invention.

detailed description

The steps of the present invention will be described in further detail below in conjunction with the accompanying drawings.

With reference to accompanying drawing 1, concrete steps of the present invention are as follows.

Step 1, import the image.

Import two synthetic aperture radar SAR images of the same size acquired at different times in the same area.

Step 2, Construct difference image.

Calculate the neighborhood difference image of two synthetic aperture radar SAR images according to the following formula:

$\mathrm{<span\; class="notranslate">\; S</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 255</span>}<span\; class="notranslate">\; -</span>\frac{<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span>}{\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; ;</span>$

Among them, S represents the neighborhood difference image of two SAR SAR images, |·| represents the absolute value operation, Σ represents the sum operation, X ₁ (i,j) represents the first SAR SAR image The set of neighborhoods of pixels corresponding to positions i and j, X ₂ (i, j) represents the set of neighborhoods of pixels corresponding to positions i and j of the second synthetic aperture radar SAR image, H represents the two synthetic aperture radar SAR images The side length of the aperture radar SAR image neighborhood, its value is 3.

Calculate the neighborhood ratio image of two synthetic aperture radar SAR images according to the following formula:

$\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 255</span>}<span\; class="notranslate">\; \&times;</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; L</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; min</span>}<span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \}</span>}{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; L</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \}</span>}<span\; class="notranslate">\; ;</span>$

Among them, R represents the neighborhood ratio image of two SAR SAR images, i represents the serial number of the i-th pixel of the two SAR SAR images, and L represents the side length of the neighborhood of the two SAR SAR images , the value of L is 3, Σ represents the summation operation, min represents the minimum value operation, N ₁ (i) represents the neighborhood set of the i-th pixel in the first synthetic aperture radar SAR image, N ₂ (i) represents The neighborhood set of the i-th pixel in the second synthetic aperture radar SAR image, and max indicates the maximum value operation.

According to the following formula, the difference image after the fusion of the neighborhood difference image of two SAR images and the neighborhood ratio image of two SAR SAR images is calculated:

$\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\underset{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span>\frac{{<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; +</span>\frac{{\mathrm{<span\; class="notranslate">\; d</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span>\frac{{<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; +</span>\frac{{\mathrm{<span\; class="notranslate">\; d</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; ;</span>$

Among them, X(x, y) represents, x, y represents the position of the clustering center of the neighborhood difference image of two SAR images, i, j represent the neighborhood difference image of two SAR SAR images The position of the center pixel of the neighborhood set of , ∈ represents the operation of taking set elements, M _{x, y} represents the neighborhood set of the neighborhood difference image of two SAR SAR images, and the neighborhood of two SAR SAR images The size of the neighborhood side of the difference image is L, and the value of L is 7. Σ represents the sum operation, exp represents the exponential operation, |·| represents the absolute value operation, and h(i,j) represents two synthetic aperture radar images The gray value of the pixel point corresponding to the neighborhood difference image of the SAR image at position i and j, h(x, y) represents the difference between the neighborhood difference images of two synthetic aperture radar SAR images at positions x and y Corresponding pixel gray value, δ represents the adjustment parameter, and its value is 1, d represents the neighborhood of the i, j position in the neighborhood difference image of the two synthetic aperture radar SAR images to the two synthetic aperture radar SAR images The Euclidean distance between the center positions x and y of the difference image clustering, μ represents the adjustment parameter, and its value is 1, and R(i, j) represents the neighborhood ratio image of two synthetic aperture radar SAR images at positions i and j The gray value of the pixel corresponding to above.

Using the linear function method, the difference image after the fusion of the neighborhood difference image of two synthetic aperture radar SAR images and the neighborhood ratio image of two synthetic aperture radar SAR images is normalized, and the normalized value after fusion is obtained. The gray value of each pixel in the difference image.

Step 3, mean shift filtering.

The first step is to use the kernel density estimation method to obtain the kernel density value of each pixel in the fused difference image of the neighborhood difference image of two synthetic aperture radar SAR images and the neighborhood ratio image of two synthetic aperture radar SAR images .

In the second step, the gray value of each pixel of the fused normalized difference image is subtracted from the kernel density value of each pixel of the fused difference image, and the absolute value of the subtraction result is taken.

Step 3, judge whether the absolute value is smaller than the set threshold, if yes, execute step 4 of this step; otherwise, execute step 1 of this step.

In step 4, the absolute value is used as the pixel gray value of the difference image to obtain the difference image after denoising.

Step 4, genetic fuzzy clustering.

Step 1: Initialize the first-generation population of the difference image after denoising, and set the number of cluster centers of the first-generation population of the difference image after denoising, that is, the number of clusters of the difference image after denoising, to 2. The number of individuals is set to 30, the maximum number of evolutions is set to 100, and the threshold range of the termination condition is set to 10 ^-8 , the pixel gray value of the difference image after denoising is randomly selected as the initial aggregation of the difference image after denoising. class center value.

Step 2, according to the following formula, calculate the global division index of the first-generation population of the difference image after denoising:

$\mathrm{<span\; class="notranslate">\; J</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; c</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}^{\mathrm{<span\; class="notranslate">\; m</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>{{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ;</span>$

Among them, J(t) represents the global division index of the first generation population of the difference image after denoising, t represents the evolution times of the first generation population of the difference image after denoising, Σ represents the sum operation, and i represents the difference after denoising The number of the i-th class in the cluster center of the image, c indicates the number of clusters of the difference image after denoising, j indicates the number of the jth pixel of the difference image after denoising, n indicates the difference image after denoising The total number of pixels, μ _ij represents the degree of membership of the jth pixel of the denoised difference image belonging to the i-th class in the cluster center of the denoised difference image, and the value range of μ _ij is [0, 1] , and must satisfy Constraint conditions, m represents the fuzzy index factor, the value of m is 2, d _ij ² represents the distance from the jth pixel of the denoised difference image to the i-th class in the cluster center of the denoised difference image, H( j) represents the gray value of the jth pixel of the fused normalized difference image.

Step 3, according to the following formula, calculate the individual fitness of the first generation population of the difference image after denoising:

$\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; J</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; ;</span>$

Among them, f(t) represents the individual fitness of the first generation population of the difference image after denoising, t represents the evolution times of the first generation population of the difference image after denoising, J(t) represents the global Divide indicators.

Step 4, using the selection, crossover and mutation operations of the genetic algorithm to obtain the next generation population of the difference image after denoising.

In step 5, calculate the fitness of each individual in the next generation population of the difference image after denoising according to the method in step 3.

Step 6, judge whether the next-generation population of the difference image after denoising is stable, if so, execute step 7 of this step; otherwise, execute step 4 of this step.

Step 7, calculate the maximum fitness value of the next generation population of the difference image after denoising.

In step 8, the individual with the maximum fitness of the next generation population of the difference image after denoising is used as the cluster center of the difference image after denoising.

Step 5, segment the difference image.

In the first step, calculate the elements of the membership degree matrix of the difference image after denoising according to the following formula:

${\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; =</span>{<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; c</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; (</span>\frac{{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; Y</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; V</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{{{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; kj</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; )</span>}^{\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; )</span>}^{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ;</span>$

Among them, μ _ij represents the element of the membership matrix of the difference image after denoising, i represents the serial number of the i-th class in the cluster center of the difference image after denoising, and j represents the number of the j-th pixel of the difference image after denoising Sequence number _, Σ indicates the sum operation, k indicates the serial number of the kth class in the clustering center of the difference image after denoising, c indicates the number of clusters of the difference image after denoising, ||·|| indicates the Euclidean distance operation, Y _j represents the gray value of the jth pixel of the difference image after denoising, V _i represents the ith cluster center of the difference image after denoising, d _kj ² represents the jth pixel point of the difference image after denoising to The cluster center distance of the k-th class of the difference image after denoising, m represents the fuzzy index factor, and the value of m is 2.

The second step is to calculate the number of rows where the minimum value of all elements of the membership degree matrix of the difference image after denoising is located.

In the third step, the row number of the minimum value of all elements of the membership degree matrix of the post-noise difference image is used as the segmentation threshold of the post-noise difference image.

Step 4, judge whether the gray value of each pixel of the difference image after denoising is smaller than the segmentation threshold of the difference image after denoising, if so, perform step 5 of this step; otherwise, perform step 6 of this step.

Step 5: Classify the segmentation threshold pixel points smaller than the difference image after denoising into the non-change class of the difference image after denoising.

In the sixth step, the segmentation threshold pixel points greater than or equal to the difference image after denoising are classified into the change class of the difference image after denoising.

Step 6, output the result.

The non-change class of the difference image after denoising and the change class of the difference image after denoising are obtained, and then the pixels of the difference image after denoising are binarized according to the classification results to obtain the change of two synthetic aperture radar SAR images The result graph of the test.

The simulation results of the present invention will be further described below in conjunction with FIG. 2 .

1. Simulation environment:

The emulation of the present invention is that core 22.26GHZ is configured in computer, memory 1G, WINDOWS XP system and computer software are configured to carry out under MATLAB 2010 environment.

2. Simulation content:

The data used in the simulation of the present invention are two sets of synthetic aperture radar SAR image data sets. The first synthetic aperture SAR image data set is the synthetic aperture radar SAR image of the farmland area of the village of Feltwell in the United Kingdom. The size of the two images is 470×335 pixels. It is caused by factors such as the radiation characteristics of electromagnetic waves and artificially embedding some change information. The second set of data sets are synthetic aperture radar SAR images in Bern, Switzerland. The size of the two images is 301×301 pixels. The changes between the two images are caused by floods near the suburbs of Bern.

Accompanying drawing 2 (a) is the result diagram of using the fuzzy C-means change detection method for the first combined synthetic aperture radar SAR image data set. Accompanying drawing 2 (b) is the result diagram of using the FLICM change detection method for the first combined SAR SAR image data set. Accompanying drawing 2 (c) is the result diagram of using the RFLICM change detection method for the first combined synthetic aperture radar SAR image data set. Accompanying drawing 2 (d) is the result diagram of adopting the present invention for the first combined synthetic aperture radar SAR image data set. Accompanying drawing 2 (e) is the result figure of adopting fuzzy C-means change detection method in the second combined SAR image data set. Accompanying drawing 2 (f) is the result figure that adopts FLICM change detection method for the second combined aperture radar SAR image data set. Accompanying drawing 2 (g) is the result figure that adopts the RFLICM change detection method of the second combined aperture radar SAR image data set. Accompanying drawing 2 (h) is the second combined aperture radar SAR image data set adopting the result diagram of the present invention. The white areas in the eight result maps are all changing areas, and the black areas are all non-changing areas.

3. Simulation result analysis:

Observing the white area in Fig. 2(a), it can be seen that using the fuzzy C-means synthetic aperture radar SAR image change detection method of the prior art, there is more noise in the change detection result of the first combined synthetic aperture radar SAR SAR image.

Observing the white area in Figure 2(b), it can be seen that using the prior art fuzzy local information C-mean FLICM synthetic aperture radar SAR image change detection method, the change detection results of the first combination of synthetic aperture radar SAR images are better than those in the attached Figure 2(a) The fuzzy C-means synthetic aperture radar SAR image change detection method using the prior art has been improved, but there are still many false detection information.

Observing the white area in Figure 2(c), it can be seen that the change detection result of the first combined SAR SAR image using the improved fuzzy local information C-mean FLICM synthetic aperture radar SAR image change detection method of the prior art, Compared with Fig. 2(a) and Fig. 2b, the change detection results of synthetic aperture radar SAR image are improved, but the edge information of the change area is smoothed, and there are still some false detection information.

Comparing the white areas in accompanying drawing 2(d) and accompanying drawing 2(a), accompanying drawing 2(b), accompanying drawing 2(c), it can be seen that because the present invention adopts the mean value shift filter, the synthetic aperture is effectively suppressed The inherent noise of the radar SAR image, combined with the local optimum of fuzzy clustering and the global optimization ability of genetic algorithm, accelerates the convergence speed of the algorithm, reduces the missed detection information in the detection results, and has the highest change detection accuracy.

Observing the white area in Figure 2(e), it can be seen that there are many false detections in the change detection results of the second combined SAR SAR image using the fuzzy C-means synthetic aperture radar SAR image change detection method of the prior art information.

Observing the white area in Fig. 2(f), it can be seen that using the fuzzy local information C-mean value FLICM synthetic aperture radar SAR image change detection method of the prior art, the change detection result of the second combined synthetic aperture radar SAR image is compared with The fuzzy C-means synthetic aperture radar SAR image change detection method in Fig. 2(e) of the prior art has been improved, but there are still more false detection information.

Observing the white area in Figure 2(g), it can be seen that the improved fuzzy local information C-mean FLICM synthetic aperture radar SAR image change detection method adopted in the prior art is comparable to the change detection result of the second combined synthetic aperture radar SAR image. Comparing with Fig. 2(e) and Fig. 2f, the SAR image change detection results are improved, but the edge information of the change area is smoothed, and there are still some missed detection information.

Comparing the white areas in accompanying drawing 2(h) and accompanying drawing 2(e), accompanying drawing 2(f), accompanying drawing 2(g), it can be seen that because the present invention adopts mean shift filtering, the synthetic aperture is effectively suppressed The inherent noise of the radar SAR image, combined with the local optimum of fuzzy clustering and the global optimization ability of genetic algorithm, accelerates the convergence speed of the algorithm, reduces the missed detection information in the detection results, and has the highest change detection accuracy.

On two sets of synthetic aperture radar SAR image data sets, the SAR image change detection method based on the mean shift genetic clustering of the present invention and the fuzzy C-means, fuzzy local information C-means FLICM, and improved fuzzy local information C The mean RFLICM synthetic aperture radar SAR image change detection method calculates the missed detection number, false detection number, total error number and calculation time of the two sets of synthetic aperture radar SAR image change detection results. Among them, the number of missed detections is the pixels that have actually changed but not detected, and the number of false detections is the pixels that have not actually changed but are detected as changes, and the total number of errors = the number of missed detections + the number of false detections. The calculation time is the time to obtain the synthetic aperture radar SAR image change detection result map by using the above four methods.

The effect of the present invention can be obtained by the number of missing detections in Table 1 obtained by the simulation experiment, the number of false detections, the total number of errors and the four indexes of calculation time to evaluate the quality of the change detection method.

Through Table 1, compare the fuzzy C-means of the prior art, the fuzzy local information C-means FLICM, the improved fuzzy local information C-means RFLICM synthetic aperture radar SAR image change detection method and the SAR image change based on the mean shift genetic clustering of the present invention detection method, it can be seen that the change detection results of the two sets of synthetic aperture radar SAR image data sets have the least number of false detections and the number of total errors, with less number of missed detections and less calculation time, that is, the highest Change detection accuracy.

Table 1 Evaluation index of two groups of synthetic aperture radar SAR image change detection results

Claims (5)

1., based on a SAR image change detection for average drifting genetic cluster, comprise the steps:

(1) image is imported:

Importing areal, the synthetic-aperture radar SAR image that the two width sizes do not obtained in the same time are identical;

(2) structural differences image:

(2a) the neighborhood error image of two width synthetic-aperture radar SAR image is calculated;

(2b) the neighborhood ratio images of two width synthetic-aperture radar SAR image is calculated;

(2c) the normalization differential image after the neighborhood error image of two width synthetic-aperture radar SAR image and the neighborhood ratio images fusion of two width synthetic-aperture radar SAR image is calculated;

(3) average drifting filtering:

(3a) utilize Density Estimator method, obtain the cuclear density value of each pixel of differential image after the neighborhood error image of two width synthetic-aperture radar SAR image and the neighborhood ratio images fusion of two width synthetic-aperture radar SAR image;

(3b) gray-scale value of each pixel of normalization differential image after fusion and step (3a) are obtained the cuclear density value subtraction of each pixel of the differential image after merging, and subtraction result is taken absolute value;

(3c) judge whether absolute value is less than the threshold value of setting, if so, then perform step (3d); Otherwise, perform step (3a);

(3d) using the pixel gray-scale value of absolute value as differential image, differential image after denoising is obtained;

(4) Genetic-fuzzy cluster:

(4a) first generation population of differential image after initialization denoising;

(4b) first generation population overall situation Classification Index of differential image after denoising according to the following formula, is calculated:

$J\left(t\right)=\underset{i=1}{\overset{c}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{\&mu;}}_{\mathrm{ij}}^m\left(t\right){d_{\mathrm{ij}}}^{2}H\left(j\right);$

Wherein, J (t) represent differential image after denoising first generation population overall situation Classification Index, t represents the evolution number of times of the first generation population of differential image after denoising, Σ represents sum operation, i represents the sequence number of i-th class in the cluster centre of differential image after denoising, and c represents the cluster number of differential image after denoising, and j represents the sequence number of a jth pixel of differential image after denoising, n represents the total pixel number of differential image after denoising, μ _ijrepresent that the jth pixel of differential image after denoising is under the jurisdiction of the degree of membership of the i-th class in the cluster centre of differential image after denoising, μ _ijspan is [0,1], and must meet constraint condition, m represents the Fuzzy Exponential factor, and m is the positive number that value is greater than 1, d _ij ²represent the distance of the i-th class in the cluster centre of differential image after a jth pixel to denoising of differential image after denoising, H (j) represents the gray-scale value of a jth pixel of the normalization differential image after fusion;

(4c) the ideal adaptation degree of the first generation population of differential image after denoising according to the following formula, is calculated:

$f\left(t\right)=\frac{1}{1+J\left(t\right)};$

Wherein, f (t) represents the ideal adaptation degree of the first generation population of differential image after denoising, and t represents the evolution number of times of the first generation population of differential image after denoising, and J (t) represents the overall Classification Index of differential image after denoising;

(4d) genetic manipulation is adopted to obtain the population of future generation of differential image after denoising;

(4e) according to the fitness of each individuality in the population of future generation of differential image after the method calculating denoising in step (4c);

(4f) after judging denoising, whether the population of future generation of differential image stablizes, and if so, then performs step (4g); Otherwise, perform step (4d);

(4g) the fitness maximal value of the population of future generation of differential image after denoising is calculated;

(4h) using the cluster centre of the individuality of the fitness maximal value of the population of future generation of differential image after denoising as differential image after denoising;

(5) differential image is split:

(5a) element of the subordinated-degree matrix of differential image after denoising according to the following formula, is calculated:

${\mathrm{\&mu;}}_{\mathrm{ij}}={\left(\underset{k=1}{\overset{c}{\mathrm{\&Sigma;}}}{\left(\frac{{\left|\right|Y_j-V_i\left|\right|}^2}{{d_{\mathrm{kj}}}^2}\right)}^{\frac{1}{m-1}}\right)}^{-1};$

Wherein, μ _ijthe element of the subordinated-degree matrix of differential image after expression denoising, i represents the sequence number of i-th class in the cluster centre of differential image after denoising, j represents the sequence number of a jth pixel of differential image after denoising, Σ represents sum operation, k represents the sequence number of a kth class in differential image cluster centre after denoising, c represents the cluster number of differential image after denoising, || || represent and ask Euclidean distance to operate, Y _ja jth pixel gray-scale value of differential image after expression denoising, V _ii-th cluster centre of differential image after expression denoising, d _kj ²represent the cluster centre distance of differential image kth class after a jth pixel to denoising of differential image after denoising, m represents the Fuzzy Exponential factor, and m is the positive number that value is greater than 1;

(5b) line number at the subordinated-degree matrix all elements minimum value place of differential image after denoising is asked;

(5c) line number at minimum value place step (5b) obtained is as the segmentation threshold of differential image after denoising;

(5d) after judging denoising, whether the gray-scale value of each pixel of differential image is less than the segmentation threshold of differential image after denoising, if so, performs step (5e); Otherwise, perform step (5f);

(5e) will the segmentation threshold pixel of differential image after denoising be less than, be classified as the non-changing class of differential image after denoising;

(5f) will the segmentation threshold pixel of differential image after denoising be more than or equal to, be classified as the change class of differential image after denoising;

(6) Output rusults:

To the change class of differential image after the non-changing class of differential image after the denoising obtained and denoising, the result figure that the change obtaining two width synthetic-aperture radar SAR image detects.

2. the SAR image change detection based on average drifting genetic cluster according to claim 1, is characterized in that: the computing formula of the neighborhood error image of two width synthetic-aperture radar SAR image described in step (2a) is as follows:

$S=255-\frac{|\mathrm{\&Sigma;}{X}_1(i,j)-\mathrm{\&Sigma;}{X}_2(i,j)|}{H\&times;H};$

Wherein, S represents the neighborhood error image of two width synthetic-aperture radar SAR image, || represent and ask absolute value operation, Σ represents sum operation, X ₁(i, j) represents the neighborhood of pixel points set that the first width synthetic-aperture radar SAR image is corresponding on i, j position, X ₂(i, j) represents the neighborhood of pixel points set that the second width synthetic-aperture radar SAR image is corresponding on i, j position, and H represents the length of side of two width synthetic-aperture radar SAR image neighborhoods, and its span is H ∈ { 3,5}.

3. the SAR image change detection based on average drifting genetic cluster according to claim 1, is characterized in that: the computing formula of the neighborhood ratio images of two width synthetic-aperture radar SAR image described in step (2b) is as follows:

$R=255\&times;\frac{\underset{i=1}{\overset{L\&times;L}{\mathrm{\&Sigma;}}}\min \{N_i\left(i\right),N_2\left(i\right)\}}{\underset{i=1}{\overset{L\&times;L}{\mathrm{\&Sigma;}}}\max \{N_1\left(i\right),N_2\left(i\right)\}};$

Wherein, R represents the neighborhood ratio images of two width synthetic-aperture radar SAR image, i represents that two width synthetic-aperture radar SAR image are in the sequence number of i-th pixel together, L represents the length of side of two width synthetic-aperture radar SAR image neighborhoods, { 3,5}, Σ represent sum operation to L ∈, min represents operation of minimizing, N ₁i () represents the Neighbourhood set of the first width synthetic-aperture radar SAR image i-th pixel, N ₂i () represents the Neighbourhood set of the second width synthetic-aperture radar SAR image i-th pixel, max represents that maximizing operates.

4. the SAR image change detection based on average drifting genetic cluster according to claim 1, is characterized in that: the concrete steps of the normalization differential image after the neighborhood error image of the calculating two width synthetic-aperture radar SAR image described in step (2c) and the neighborhood ratio images of two width synthetic-aperture radar SAR image merge are as follows:

The first step, according to the following formula, calculates the differential image after the neighborhood error image of two width synthetic-aperture radar SAR image and the neighborhood ratio images fusion of two width synthetic-aperture radar SAR image:

$X(x,y)=\frac{\underset{(i,j)\&Element;M_{x,y}}{\mathrm{\&Sigma;}}\exp (\frac{{|h(i,j)-h(x,y)|}^2}{2{\mathrm{\&delta;}}^2}+\frac{d^2}{2{\mathrm{\&mu;}}^2})R(i,j)}{\exp (\frac{{|h(i,j)-h(x,y)|}^2}{2{\mathrm{\&delta;}}^2}+\frac{d^2}{2{\mathrm{\&mu;}}^2})};$

Wherein, X (x, y) represents, x, y represent the position of the neighborhood error image cluster centre of two width synthetic-aperture radar SAR image, i, j represent the central pixel point position of the Neighbourhood set of the neighborhood error image of two width synthetic-aperture radar SAR image, and ∈ represents that getting set element operates, M _x,yrepresent the Neighbourhood set of the neighborhood error image of two width synthetic-aperture radar SAR image, the neighborhood length of side size of the neighborhood error image of two width synthetic-aperture radar SAR image is L, L ∈ { 7, 11}, Σ represents sum operation, exp represents index operation, || represent and ask absolute value operation, h (i, j) represent that the neighborhood error image of two width synthetic-aperture radar SAR image is at i, pixel gray-scale value corresponding on j position, h (x, y) represent that the neighborhood error image of two width synthetic-aperture radar SAR image is at x, pixel gray-scale value corresponding on y position, δ represents adjustment parameter, its span is (0, 100], d represents i in the neighborhood error image of two width synthetic-aperture radar SAR image, j position is to the neighborhood error image cluster centre position x of two width synthetic-aperture radar SAR image, the Euclidean distance of y, μ represents adjustment parameter, its span is (0, 100], R (i, j) represent that the neighborhood ratio images of two width synthetic-aperture radar SAR image is at i, pixel gray-scale value corresponding on j position,

Second step, utilize linear function method, differential image after the neighborhood error image of two width synthetic-aperture radar SAR image and the neighborhood ratio images of two width synthetic-aperture radar SAR image merge is normalized, obtains the gray-scale value of each pixel of normalization differential image after merging.

5. the SAR image change detection based on average drifting genetic cluster according to claim 1, it is characterized in that: the initialization described in step (4a) refers to, the cluster centre number of the first generation population of differential image after denoising is set as 2, population at individual number is set as 30, maximum evolution number of times is set as 100, end condition threshold range is set as 10 ^-8, by the pixel gray-scale value of differential image after Stochastic choice denoising, as the initial cluster center value of differential image after denoising.