CN103198482B - Based on the method for detecting change of remote sensing image that disparity map fuzzy membership merges - Google Patents

Abstract

The invention discloses a remote sensing image change detection method based on fusion of difference map fuzzy membership degree, which mainly solves the problem that the existing change detection method cannot effectively remove false change information and maintain edge information. The implementation process is: input two remote sensing images with different time phases, calculate the structural similarity coefficients of the corresponding pixels, and obtain a similarity difference map; make a difference between the two remote sensing images to obtain a difference image; The pixels of the value map are labeled with categories to obtain a category label map; the difference map is filtered according to the category label map to obtain a denoising difference map; the similarity difference map and the denoising difference map are fuzzy membership fusion and combined Classify the change detection results. The invention has strong noise resistance, can effectively remove false change information, and at the same time retain better edge information, has high detection result accuracy, and can be used for urban expansion monitoring, forest and vegetation change monitoring.

Description

Change Detection Method of Remote Sensing Image Based on Difference Map Fuzzy Membership Fusion

technical field

The invention belongs to the field of digital image processing, and mainly relates to remote sensing image change detection, in particular to a remote sensing image change detection method based on fusion of difference map fuzzy membership degrees, which can be used for remote sensing image analysis and processing.

Background technique

The change detection of remote sensing images is the process of identifying the state changes or phenomena changes of objects by analyzing and extracting the differences in electromagnetic spectrum characteristics or spatial structure characteristics between remote sensing images in different time periods in the same area. It has been widely used in many fields of national economy and national defense construction, such as agricultural survey, forest and vegetation change monitoring, urban expansion monitoring, military target monitoring, etc.

The general steps of common remote sensing image change detection methods are to construct a difference map first, and then select an appropriate threshold to classify the difference map into change classes and non-change classes. The construction of the difference map and the processing of the difference map are important steps in image change detection. A relatively simple difference image construction method is the difference method. This method is easy to implement, but the constructed difference image has more noise, and an effective method is needed to deal with the noise in the difference image. Removing the more severe noise in the difference image and enhancing the pixels with small gray value changes can effectively improve the quality of the difference image and make the detection results more accurate.

In order to construct a better difference map, some scholars construct a difference map by measuring the similarity of the corresponding pixel gray value of the two-temporal images, Inglada and Mercier (2007) in the article "A New Statistical Similarity Measure for Change Detection in Multitemporal SAR Images and Its Extension to MultiscaleChange Analysis, IEEE Transactions on Geoscience and Remote Sensing, 2007,45(5):1432-1445" proposed a SAR image change detection method based on statistical similarity, which uses KL divergence to measure two-time The statistical similarity of the corresponding pixel neighborhood of the phase image is used to construct the difference map, and then the threshold value is used to segment the difference map to obtain the change result. This method is to model the local histogram, but there are few pixels in the local area, so it is difficult to effectively model it. , so the detection result of this method is poor. He (2010) proposed a change based on Euclidean distance in the article "Application of EuclideanNorm in Multi-temporal Remote Sensing Image Change Detection, International Congresson Image and Signal Processing (CISP'2010), 2010,5:2111-2115" Detection method, this method constructs a difference map by calculating the Euclidean distance of two-temporal remote sensing images of multiple bands, and then performs threshold segmentation on the difference map to obtain the change result. This method can effectively reduce the influence of part of the noise but the detection result still exists. More pseudo-change information.

In order to combine the advantages of different difference maps, some scholars have fused different difference maps. Ma Guorui and other scholars (2006) wrote in the article "Remote Sensing Image Change Detection Based on Fusion and Generalized Gaussian Model", Journal of Remote Sensing, 2006, 10(6): 847-853" proposed a method of fusing the difference map and the ratio map using the product fusion strategy. This method can suppress the background and enhance the change area to a certain extent, but the method is not stable and sometimes suppresses the change area. Wang Min and Zhang Xingyue (2010) proposed in the article "Remote Sensing Image Change Detection Based on Multi-Feature Evidence Fusion. Journal of Remote Sensing, 2010, 14(2): 1-7" to use evidence theory fusion method to fuse multiple feature difference maps. , this method improves the detection accuracy of the single feature detection method, but the structural similarity measurement method adopted by this method is unstable, thus reducing the correct rate of detection results. Du et al. (2012) in the article "Fusion of Difference Images for Change Detection over Urban Areas, IEEE Journal of Selected Topics in Applied Earth Obserbations and Remote Sensing, 2012, 5(4): 1076-1086" proposes a multi-difference map A change detection method based on feature-level and decision-level fusion. This method combines the advantages of multiple difference maps and improves the accuracy of change detection. However, this method does not fuse difference maps according to the advantages and disadvantages of difference maps. Therefore, The correct rate of detection results has not improved much, and may sometimes decrease.

Contents of the invention

The purpose of the present invention is to overcome the deficiencies of the above-mentioned transformation detection technology, and propose a remote sensing image change detection method based on the fuzzy membership degree fusion of the difference map, to reduce the influence of noise on the detection results, reduce the false change information in the detection results, and improve The accuracy of the test results.

To achieve the above object, the detection method of the present invention comprises the following steps:

(1) Input two registered remote sensing images X ₁ and X ₂ of the same area with different time phases of size I×J, and calculate the structural similarity of the corresponding pixels of the two remote sensing images X ₁ and X ₂ Coefficient SIM(m,n), get a similarity coefficient matrix SIM:

SIM＝{SIM(m,n)|1≤m≤I, 1≤n≤J}

in, $\mathrm{SIM}(m,\mathrm{no})=\mathrm{\λ}\frac{2{\mathrm{\μ}}_1(m,\mathrm{no})\&Center\; Dot;{\mathrm{\μ}}_2(m,\mathrm{no})+C}{{\mathrm{\μ}}_1^2(m,\mathrm{no})+{\mathrm{\μ}}_2^2(m,\mathrm{no})+C}+(1-\mathrm{\λ})\frac{2{\mathrm{\σ}}_1(m,\mathrm{no})\·{\mathrm{\σ}}_2(m,\mathrm{no})+C}{{\mathrm{\σ}}_1^2(m,\mathrm{no})+{\mathrm{\σ}}_2^2(m,\mathrm{no})+C},$

In the formula, m and n are the row number and column number of the image respectively, m=1,2,...,I, n=1,2,...,J, μ ₁ (m,n), σ ₁ (m,n) and (m,n) are the mean, standard deviation and variance of the pixel values in the local area of the remote sensing image X ₁ centered on the pixel (m,n) and the window size is w×w, μ ₂ (m,n), σ ₂ (m,n) and (m,n) are the mean, standard deviation and variance of the pixel values in the local area of the remote sensing image X ₂ centered on the pixel point (m,n) and the window size is w×w, and the value range of the window w is 3 ~9, C is a constant used to avoid instability when the denominator is close to zero, the value range of C is C>0, λ is the weight coefficient, and the value range of the structural similarity coefficient is 0≤SIM(m, n)≤1;

(2) Linearly map the structural similarity coefficient SIM(m,n) of the similarity coefficient matrix SIM at position (m,n) to the interval [0,255], and obtain the similarity difference map X _S at position (m,n) Pixel gray value X _S (m,n) at:

X _S (m,n)=(1-SIM(m,n))×255,

(3) Calculate the difference between the pixel gray values X ₁ (m,n) and X ₂ (m,n) at the corresponding positions (m,n) of the remote sensing images X ₁ and X ₂ to obtain the difference X _d (m,n)=|X ₁ (m,n)-X ₂ (m,n)|, according to the order from left to right and from top to bottom, calculate the corresponding position of remote sensing image X ₁ and X ₂ The difference between the gray value of the pixel, and get a difference difference map X _d :

X _d ={X _d (m,n)|1≤m≤I, 1≤n≤J},

(4) Perform median filtering with a window size of 3×3 on the pixels in the difference difference map X _d to obtain the filtered difference map X _f , and perform statistical histogram threshold segmentation on the filtered difference map X _f to obtain initial classification map X _m ;

(5) According to the gray value range of the filtered difference map X _f and the initial classification map X _m , classify the pixels in the difference difference map X _d to obtain a class label map X _b ;

(6) Filter the pixels at position (m, n) in the difference difference map X _d according to the mark at the position (m, n) in the class mark map X _b , and obtain the denoising difference map X _N ;

(7) Calculate the change class membership degree of the pixel at position (m, n) in the similarity difference map X _S (m,n) and non-changing class membership (m,n), calculate the change class membership degree of the pixel at position (m,n) in the denoising difference image X _N (m,n) and non-changing class membership (m,n);

(8) Calculate the change class membership of the corresponding pixel points (m, n) in the similarity difference map X _S and the denoising difference map X _N (m,n) and (m,n) fusion membership value H _c (m,n), and then calculate the non-changing class membership of the pixel (m,n) and Fusion membership degree H _u (m,n) of (m,n);

(9) Create a fused image X _I with the same size as the similarity difference map X _S , if H _c (m,n)>H _u (m,n), then the value of the fused image X _I at the pixel is marked as 1, otherwise it is recorded as 0, thus obtaining a change detection result image.

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

1. The structural similarity measurement method proposed by the present invention can stably and effectively measure the similarity between the pixels of the two-temporal images, and the constructed similarity difference map can suppress background noise and enhance the contrast between the target and the background.

2. The present invention classifies the pixels in the difference map, which can effectively suppress the severe noise in the difference map while maintaining the edge information.

3. The present invention fuses the similarity difference map and the denoising difference map, effectively combining the advantages of the two difference maps, not only reducing the influence of noise on the detection results, but also maintaining the edge of the change area, improving the change detection results correct rate.

Description of drawings

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

Figure 2 is the first set of remote sensing images used in the experiment and the corresponding change reference images;

Fig. 3 is the second group of remote sensing images used in the experiment and the corresponding change reference images;

Fig. 4 is the change detection result figure that the present invention and comparative method simulate the first group of remote sensing images to obtain;

Fig. 5 is a diagram of change detection results obtained by simulating the second group of remote sensing images according to the present invention and the comparison method.

detailed description

With reference to Fig. 1, the realization steps of the present invention are as follows:

Step 1, input two registered remote sensing images X ₁ and X ₂ of the same area with different time phases of size I×J, as shown in Fig. 2(a) and Fig. 2(b), and calculate the two Structural similarity coefficients SIM(m,n) of corresponding pixels of remote sensing images X ₁ and X ₂ are obtained to obtain a similarity coefficient matrix SIM.

1.1) Calculate the mean value μ ₁ (m,n) of the local area centered on the pixel point (m,n) and the window size is w×w pixels in the phase remote sensing images X ₁ and X ₂ of phase 1 and phase 2 respectively ) and μ ₂ (m,n), standard deviation σ ₁ (m,n) and σ ₂ (m,n) and variance (m,n) and (m,n), m and n are the row number and column number of the image respectively, m=1,2,...,I, n=1,2,...,J, in order to compare the positioning accuracy and noise resistance Among them, the value range of the window w is 3-9, and w=5 in the example of the present invention;

1.2) Calculate the structural similarity coefficient SIM(m,n) at the corresponding pixel point (m,n) of the remote sensing images X ₁ and X ₂ in phase 1 and phase 2:

$\mathrm{<span\; class="notranslate">\; SIM</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}\frac{\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; C</span>}}{{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; C</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; )</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; C</span>}}{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; C</span>}}$

In the formula, C is the constant that is used to avoid the unstable phenomenon when the denominator is close to zero, and the value range of C is C>0, which is taken as 1 in the example of the present invention, and λ is a weight coefficient, and λ=0.95 in the example of the present invention , the calculated range of the structural similarity coefficient is 0≤SIM(m,n)≤1, and the closer SIM(m,n) is to 0, it indicates that the remote sensing images X ₁ and X ₂ are less similar at the pixel point (m,n) , the greater the possibility of belonging to the change class;

1.3) Calculate the structural similarity coefficients of all pixels in the image according to step (1.1) and step (1.2), and obtain a similarity coefficient matrix SIM={SIM(m,n)|1≤m≤I, 1≤n≤J }.

Step 2, linearly map the structural similarity coefficient SIM(m,n) of the similarity coefficient matrix SIM at position (m,n) to the interval [0,255], and obtain the similarity difference map X _S at position (m,n) Pixel gray value X _S (m,n) at:

X _S (m,n)=(1-SIM(m,n))×255.

Step 3, the pixel gray value X ₁ (m,n) and X ₂ (m,n) at the corresponding positions (m,n) of the remote sensing images X ₁ and X ₂ in the phase 1 and phase 2 are calculated Calculate the difference to get the difference X _d (m,n)=|X ₁ (m,n)-X ₂ (m,n)|, and calculate the remote sensing image X in order from left to right and from top to bottom ₁ and X ₂ The difference between the gray value of the pixel at the corresponding position in the space to obtain a difference difference map X _d :

X _d ={X _d (m,n)|1≤m≤I, 1≤n≤J}.

Step 4: Perform median filtering with a window size of 3×3 on the difference difference map X _d to obtain the filtered difference map X _f , perform statistical histogram threshold segmentation on the filtered difference map X _f to obtain the initial classification map X _m .

4.1) Perform median filtering with a window size of 3×3 on the difference difference map X _d to obtain the filtered difference map X _f ;

4.2) For the filtered difference map X _f , the condition will be satisfied The minimum gray level in the gray level is set to the low threshold T _l , where L represents the maximum pixel gray level for the accumulation of the number of gray level pixels, and its value ranges from 0 to 255, and N _x represents the gray level Level is the total number of pixels of x, λ is a proportional constant, and the value range is 0.5～0.6, and λ=0.5 in the example of the present invention;

4.3) For the filtered difference image X _f , set the minimum gray level among the gray levels greater than the low threshold T _l and satisfy the condition N _x <(1-λ)×I×J/(255-T _l ) Statistical histogram threshold

4.4) Statistical histogram threshold for the filtered difference map X _f Carry out segmentation to obtain the initial classification map X _m :

${\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span><span\; class="notranslate">\; =</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; T</span>}}\\ \mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; T</span>}}\end{array}\right.<span\; class="notranslate">\; ,</span>$

Among them, X _m (m, n) is the pixel value at (m, n) in the initial classification map X _m , and X _f (m, n) is the gray value at (m, n) in the filtered difference map X _f Degree value, mark 1 means change class, mark 0 means non-change class.

Step 5, according to the gray value range of the filtered difference map X _f and the initial classification map X _m , classify the pixels in the difference difference map X _d to obtain a class label map X _b .

5.1) Calculate the area of each area formed by the change class pixels in the initial classification map _Xm , if the area is less than 70, assign the pixel mark in this area in the initial classification map _Xm to 2, otherwise the pixel mark remains unchanged;

5.2) Calculate the maximum gray value N _max of the filtered difference image X _f , if N _max is greater than the threshold T, the initial classification image X _m is the category label image X _b ; otherwise, go to step (5.3), where the gray value Level threshold T is a constant, and the value range is 100～150, and T=100 in the example of the present invention;

5.3) For the area formed by the pixels marked as 1 in the initial classification map X _m , perform mathematical morphological expansion of a square window with a structural element of 3×3 to obtain a small extended image X _m1 , and the midpoint of the small extended image X _m1 ( The pixel value of m,n) is denoted as X _m1 (m,n);

5.4) For the area formed by the pixels marked as 1 in the initial classification map X _m , perform mathematical morphological expansion of a square window with a structural element of 7×7 to obtain a large extended image X _m2 , and the midpoint of the large extended image X _m2 ( The pixel value of m,n) is recorded as X _m2 (m,n);

5.5) Calculate the difference between the pixel gray value at the corresponding position (m, n) of the large extended image X _m2 and the small extended image X _m1 space, and obtain the difference X _m3 (m, n)=X _m2 (m, n)-X _m1 (m,n), thus obtaining an extended difference image X _m3 ={X _m3 (m,n)}, where X _m3 (m,n) is the midpoint of the extended difference image X _m3 Pixel value at (m,n);

5.6) Create a class label map X _b of the same size as the initial classification image, and assign the pixel value X _b (m,n) at the point (m,n) in the image according to the following five situations:

For a pixel point (m, n) that satisfies the condition X _m (m, n)=2, mark the value X _b (m, n) of the pixel point in the category label map X _b as 2;

For a pixel point (m,n) that satisfies the condition X _m (m,n)≠2 and X _m3 (m,n)=1, the value X _b (m,n) of the pixel point in the category label map X _b marked as 3;

For a pixel point (m,n) that satisfies the condition X _m (m,n)≠2 and X _m1 (m,n)=1, the value X _b (m,n) of the pixel point in the category label map X _b mark as 1;

For a pixel point (m, n) that satisfies the condition X _m (m, n)=0, mark the value X _b (m, n) of the pixel point in the category label map X _b as 0;

For a pixel point (m, n) that does not meet the above four conditions, the value X _b (m, n) of the pixel point in the category label map X _b is marked as 0.

Step 6, according to the pixel value at the position (m, n) in the class label map X _b , filter the pixel at the position (m, n) in the difference difference map X _d , and obtain a denoising difference map X _N.

The filtering operation is performed according to the following rules:

For a pixel that satisfies the condition X _b (m, n)=0, calculate the median value of all pixels in the 9×9 window of the pixel point in the difference difference map X _d , and assign the median value to the denoising difference Pixel value X _N (m,n) at point (m,n) in value map X _N ;

For a pixel point that satisfies the condition X _b (m, n)=1, assign the value X _d (m, n) of the pixel point in the difference difference map X _d to the point (m, n) in the denoising difference map X _N The pixel value X _N (m,n) at n);

For a pixel point that satisfies the condition X _b (m, n)=2, calculate the median value of all pixels in the 11×11 window of the pixel point in the difference difference map X _d , and assign the median value to the denoising difference Pixel value X _N (m,n) at point (m,n) in value map X _N ;

If there is a pixel that satisfies the condition X _b (m,n)=3, then calculate the value of the Gaussian kernel function whose Gaussian scale is 3×3 for the pixel in the difference difference map X _d , and assign this value to the denoising The pixel value X _N (m, n) at the point (m, n) in the difference map X _N ;

The values X _N (m, n) of all pixel points (m, n) constitute the denoising difference map X _N ={X _N (m,n)}.

Step 7, calculate the change class membership degree of the pixel point at position (m, n) in the similarity difference map X _S (m,n) and non-changing class membership (m,n), and then calculate the change class membership of the denoising difference map X _N at the position (m,n) (m,n) and non-changing class membership (m,n).

7.1) Calculate the change class membership degree of the pixel at position (m, n) in the similarity difference map X _S (m,n) and non-changing class membership (m,n):

${\mathrm{<span\; class="notranslate">\; h</span>}}_{{\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; S</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; S</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}\\ \frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \&times;</span>{<span\; class="notranslate">\; (</span>\frac{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; S</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; \&le;</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; S</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\\ \mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \&times;</span>{<span\; class="notranslate">\; (</span>\frac{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; S</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}}{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \&le;</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; S</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}\\ \mathrm{<span\; class="notranslate">\; 1</span>}& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; S</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}\end{array}\right.<span\; class="notranslate">\; ,</span>$

${\mathrm{<span\; class="notranslate">\; h</span>}}_{{\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; S</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{{\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; S</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$

Among them, T ₁ , T ₂ and T ₃ are three different thresholds, and T ₁ <T ₂ <T ₃ , T ₂ =(T ₁ +T ₃ )/2, where, The optimal threshold for segmenting the similarity difference graph X _S for the maximum entropy threshold segmentation method;

7.2) Calculate the change class membership degree of the pixel point of the denoising difference image X _N at position (m,n) (m,n) and non-changing class membership (m,n):

${\mathrm{<span\; class="notranslate">\; h</span>}}_{{\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}\\ \frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \&times;</span>{<span\; class="notranslate">\; (</span>\frac{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; \&le;</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\\ \mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \&times;</span>{<span\; class="notranslate">\; (</span>\frac{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}}{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \&le;</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}\\ \mathrm{<span\; class="notranslate">\; 1</span>}& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}\end{array}\right.<span\; class="notranslate">\; ,</span>$

${\mathrm{<span\; class="notranslate">\; h</span>}}_{{\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{{\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$

Wherein, X _N (m, n) is the pixel value at (m, n) in the denoising difference image X _N.

Step 8, calculate the change class membership of the corresponding pixel points (m, n) in the similarity difference map X _S and the denoising difference map X _N (m,n) and (m,n) fusion membership value H _c (m,n), and then calculate the non-changing class membership of the pixel (m,n) and (m,n) fusion membership _Hu (m,n).

8.1) Calculate the change class membership of the corresponding pixel points (m, n) of the similarity difference map X _S and the denoising difference map X _N (m,n) and The fusion membership value H _c (m,n) of the change class of (m,n):

${\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}<span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{{\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; S</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{{\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$

Among them, β is the change class membership of the similarity difference map (m,n) and non-changing class membership The fusion weight of (m, n), β=0.4 in the example of the present invention.

8.2) Calculate the non-change class membership of the similarity difference map X _S of the pixel point (m, n) (m,n) and non-variant class membership of the denoised difference map X _N Fusion membership degree H _u (m,n) of (m,n):

${\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}<span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{{\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; S</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{{\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; .</span>$

Step 9, create a fused image X _I with the same size as the similarity difference map X _S , if H _c (m,n)>H _u (m,n), then the pixel point (m,n) The value X _I (m,n) of the fused image X _I is assigned 1, otherwise it is assigned 0; thus a change detection result image X _I ={X _I (m,n)} is obtained.

Effect of the present invention can be further illustrated by following experimental results and analysis:

1. Experimental data

Figure 2(a) and Figure 2(b) are the first set of remote sensing data. This set of real remote sensing data is from the fourth band of Landsat7ETM+ (Enhanced Thematic Mapper Plus) in the suburbs of Mexico in April 2000 and May 2002. composition of remote sensing images. Figure 2(a) and Figure 2(b) both have a size of 512×512, 256 gray levels. The change area is mainly caused by the fire destroying a large area of local vegetation. Figure 2(c) is the corresponding change reference map , the white area in Figure 2(c) represents the changed area, where the number of changed pixels is 25599 and the number of non-changed pixels is 236545.

Figure 3(a) and Figure 3(b) are the second set of remote sensing data. This set of real remote sensing data was obtained from the 4th band of the Landsat-5 satellite TM in two phases in August 1994 and September 1994 in the western area of Elba Island, Italy. Spectral image composition. The size of Figure 3(a) and Figure 3(b) is 326×414, 256 gray levels, the change area is caused by forest fires, Figure 3(c) is the corresponding change reference map, in Figure 3(c) The white area represents the changed area, where the number of changed pixels is 2415 and the number of non-changed pixels is 99985.

2. Comparative test

In order to illustrate the effectiveness of the present invention, the present invention is compared with the following three comparative methods.

Compared with method 1, it is a method of fuzzy membership degree fusion of the difference difference map and the similarity difference map. The comparison between the present invention and the comparison method 1 is to verify the effectiveness of the present invention in processing the difference map according to the signature map.

Comparison method 2 is to replace the method of calculating the Euclidean distance of pixels corresponding to multiple bands of two-temporal remote sensing images proposed by He (2010) in the article "Application of Euclidean Norm in Multi-temporal Remote Sensing Image Change Detection" to calculating two-temporal A method for constructing a difference map from the Euclidean distance of a single-band pixel neighborhood. The comparison between the present invention and the comparison method 2 is to verify the validity of the similarity difference map constructed in the present invention.

Comparative method 3 is a change detection method for feature-level fusion of multiple difference images proposed by Du et al. (2012) in the article "Fusion of Difference Images for Change Detection over Urban Areas", in which the feature-level fusion method also uses fuzzy membership fusion. The purpose of comparing the present invention with the comparison method 3 is to verify the effectiveness of the present invention in fusion of the similarity difference map and the denoising difference map for change detection.

3. Experimental content and analysis

Experiment 1, using the present invention and comparison method 1, comparison method 2 and comparison method 3 to carry out change detection on the first group of remote sensing images, the results are shown in Figure 4, wherein:

Fig. 4(a) is a diagram of the change detection results of the comparison method 1;

Figure 4(b) is a diagram of the change detection results of the comparison method 2;

Fig. 4(c) is a diagram of the change detection result of the comparison method 3;

Fig. 4(d) is a diagram of the change detection result of the present invention.

As can be seen from Figure 4, in addition to the change area in the detection result figures of the three comparison methods, the whole picture is also distributed with very small miscellaneous points; compared with the three comparison methods, the shape of the change area in the result figure of the present invention is Keep it well, retain relatively complete edge information, and have little pseudo-change information such as noise.

Experiment 2, using the present invention and comparison method 1, comparison method 2 and comparison method 3 to detect changes in the first group of remote sensing images, the results are shown in Figure 5, wherein:

Figure 5(a) is a diagram of the change detection results compared with Method 1;

Figure 5(b) is a diagram of the change detection results of the comparison method 2;

Figure 5(c) is a diagram of the change detection results of the comparison method 3;

Fig. 5(d) is a diagram of the change detection result of the present invention.

It can be seen from Figure 5 that only a part of the change area is detected in the detection result diagram of the comparison method 1, and there are many missed detections; the overall shape of the change area in the detection result diagram of the comparison method 2 remains good, but There are some false change information in a small area; the change area detected by the comparison method 3 is a white area, which almost covers the entire image, and the detection result cannot maintain the basic shape of the change area, and the detection result is very poor; the present invention is compared with the three Compared with the method, the shape of the change area in the detection result map is well maintained, and there is no false change information such as noise points.

Table 1 lists the quantitative results of the present invention and three comparison methods simulating the first group of remote sensing images and the second group of remote sensing images.

Table 1

As can be seen from Table 1, for the first group of remote sensing images, the number of missed detections of the present invention is less than that of comparison method 1, the number of false detections is slightly more than it, and the total number of errors is less than that of comparison method 1; Compared method 2 is more, but the number of false detections is much less than it, and the total number of errors is less than that of comparative method 2; The number is less than that of comparative method 3. For the second group of remote sensing images, the total number of errors in the present invention is very small, and the correct rate is the highest.

Therefore, on the whole, the present invention can obtain more accurate change information, has strong noise resistance, can effectively remove noise points, and can maintain better edge information of the change area; whether it is visual effects or performance indicators, This method is better.

Claims (4)

1. A remote sensing image change detection method based on difference map fuzzy membership fusion, characterized in that: comprising the steps: (1) Input two registered remote sensing images X ₁ and X ₂ of the same area with different time phases of size I×J, and calculate the structural similarity of the corresponding pixels of the two remote sensing images X ₁ and X ₂ Coefficient SIM(m,n), get a similarity coefficient matrix SIM: SIM＝{SIM(m,n)|1≤m≤I, 1≤n≤J} in, $\mathrm{SIM}(m,\mathrm{no})=\mathrm{\&lambda;}\frac{2{\mathrm{\&mu;}}_1(m,\mathrm{no})\&Center\; Dot;{\mathrm{\&mu;}}_2(m,\mathrm{no})+C}{{\mathrm{\&mu;}}_1^2(m,\mathrm{no})+{\mathrm{\&mu;}}_2^2(m,\mathrm{no})+C}+(1-\mathrm{\&lambda;})\frac{2{\mathrm{\&sigma;}}_1(m,\mathrm{no})\&CenterDot;{\mathrm{\&sigma;}}_2(m,\mathrm{no})+C}{{\mathrm{\&sigma;}}_1^2(m,\mathrm{no})+{\mathrm{\&sigma;}}_2^2(m,\mathrm{no})+C},$ In the formula, m and n are the row number and column number of the image respectively, m=1,2,...,I, n=1,2,...,J, μ ₁ (m,n), σ ₁ (m,n) and are the mean, standard deviation and variance of the pixel values in the remote sensing image X ₁ centered on the pixel point (m,n) and the window size is w×w, μ ₂ (m,n), σ ₂ (m, n) and are the mean, standard deviation and variance of the pixel values in the local area of the remote sensing image X ₂ centered on the pixel point (m, n) and the window size is w×w, the value range of the window w is 3-9, and C is A constant used to avoid instability when the denominator is close to zero. The value range of C is C>0, λ is the weight coefficient, and the value range of the structural similarity coefficient is 0≤SIM(m,n)≤1; (2) Linearly map the structural similarity coefficient SIM(m,n) of the similarity coefficient matrix SIM at position (m,n) to the interval [0,255], and obtain the similarity difference map X _S at position (m,n) Pixel gray value X _S (m,n) at: X _S (m,n)=(1-SIM(m,n))×255 (3) Calculate the difference between the pixel gray values X ₁ (m,n) and X ₂ (m,n) at the corresponding positions (m,n) of the remote sensing images X ₁ and X ₂ to obtain the difference X _d (m,n)=|X ₁ (m,n)-X ₂ (m,n)|, according to the order from left to right and from top to bottom, calculate the corresponding position of remote sensing image X ₁ and X ₂ The difference between the gray value of the pixel, and get a difference difference map X _d : X _d ＝{X _d (m,n)|1≤m≤I, 1≤n≤J} (4) Perform median filtering with a window size of 3×3 on the pixels in the difference difference map X _d to obtain the filtered difference map X _f , and perform statistical histogram threshold segmentation on the filtered difference map X _f to obtain initial classification map X _m ; (5) According to the gray value range of the filtered difference map X _f and the initial classification map X _m , classify the pixels in the difference difference map X _d to obtain a class label map X _b ; (6) According to the mark at the position (m, n) place in the category mark map X _b , filter the pixel at the position (m, n) place in the difference difference map X _d , obtain the denoising difference map X _N ; (7) Calculate the change class membership degree of the pixel at position (m, n) in the similarity difference map X _S and invariant class membership Calculate the change class membership degree of the pixel at position (m,n) in the denoising difference image X _N and invariant class membership (8) Calculate the change class membership of the corresponding pixel points (m, n) in the similarity difference map X _S and the denoising difference map X _N and The fusion membership value H _c (m,n) of the pixel, and then calculate the non-changing class membership of the pixel and Fusion membership degree H _u (m,n); (9) Establish a fused image X _I with the same size as the similarity difference map X _S , if H _c (m,n)>H _u (m,n), then the value of the fused image X _I at the pixel is marked as 1, otherwise it is recorded as 0, thus obtaining a change detection result image. 2. the remote sensing image change detection method according to claim 1, is characterized in that: step (4) is carried out statistical histogram threshold value segmentation to difference map X _f after filtering, obtains initial classification map X _m , as follows Steps to proceed: (4a) Carrying out a median filter with a window size of 3×3 to the difference difference map X _d to obtain the filtered difference map X _f ; (4b) For the filtered difference map X _f , the condition will be satisfied The minimum gray level in the gray level is set to the low threshold T _l , where L represents the maximum gray level for the accumulation of the number of gray level pixels, and its value ranges from 0 to 255, and N _x represents the gray level The total number of pixels whose level is x, λ is a proportional constant with a value range of 0.5 to 0.6, and I and J represent the number of rows and columns of the image, respectively; (4c) For the filtered difference image X _f , the minimum gray level among the gray levels that will be greater than the low threshold T _l and satisfy the condition N _x <(1-λ)×I×J/(255-T _l ) Set as statistical histogram threshold (4d) Statistical histogram threshold for the filtered difference map X _f Carry out segmentation to obtain the initial classification map X _m : ${\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span><span\; class="notranslate">\; =</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; T</span>}}\\ \mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; T</span>}}\end{array}\right.$ Among them, X _m (m, n) is the pixel value at (m, n) in the initial classification map X _m , and X _f (m, n) is the gray value at (m, n) in the filtered difference map X _f Degree value, mark 1 means change class, mark 0 means non-change class. 3. The remote sensing image change detection method according to claim 1, characterized in that: according to the gray value range and the initial classification map X _m of the difference map _Xf after the filter described in step (5), to the difference difference The pixels in the graph X _d are labeled with categories to obtain a category labeled image X _b , and the steps are as follows: (5a) Calculate the area of each area formed by the change class pixels in the initial classification map _Xm , if the area is less than 70, then assign the pixel mark in this area in the initial classification map _Xm to 2, otherwise the pixel mark remains unchanged; (5b) Calculate the maximum grayscale value _Nmax of the filtered difference image _Xf , if _Nmax is greater than the grayscale threshold T, the initial classification image _Xm is the category label image _Xb , and end; otherwise, go to (5c ), where the gray level threshold T is a constant with a range of 100 to 150; (5c) For the area formed by the pixels marked 1 in the initial classification map X _m , perform mathematical morphology expansion of a square window with a structural element of 3×3 to obtain a small extended image X _m1 , and add the small extended image X _m1 The pixel value of point (m,n) is recorded as X _m1 (m,n); (5d) For the area formed by the pixels marked 1 in the initial classification map X _m , perform mathematical morphology expansion of a square window with a structural element of 7×7 to obtain a large extended image X _m2 , and convert the large extended image X _m2 into The pixel value of point (m,n) is recorded as X _m2 (m,n); (5e) Calculate the difference between the pixel gray value at the corresponding position (m, n) of the large extended image X _m2 and the small extended image X _m1 space, and obtain the difference X _m3 (m, n) = X _m2 (m ,n)-X _m1 (m,n), thus obtaining an extended difference image X _m3 ={X _m3 (m,n)}, where X _m3 (m,n) is the extended difference image X _m3 Pixel value at point (m,n); (5f) Create a category label map X _b of the same size as the initial classification image, and assign the pixel value X _b (m,n) at the point (m,n) in the image according to the following five situations: For a pixel point (m, n) that satisfies the condition X _m (m, n) = 2, mark the value X _b (m, n) of the pixel point in the category label map X _b as 2; For a pixel point (m,n) that satisfies the condition X _m (m,n)≠2 and X _m3 (m,n)=1, the value X _b (m,n) of the pixel point in the category label map X _b marked as 3; For a pixel point (m,n) that satisfies the condition X _m (m,n)≠2 and X _m1 (m,n)=1, the value X _b (m,n) of the pixel point in the category label map X _b mark as 1; For a pixel point (m, n) that satisfies the condition X _m (m, n) = 0, mark the value X _b (m, n) of the pixel point in the category label map X _b as 0; For a pixel point (m, n) that does not meet the above four conditions, the value X _b (m, n) of the pixel point in the category label map X _b is marked as 0. 4. The remote sensing image change detection method according to claim 1, characterized in that: according to the pixel value at the position (m, n) in the class mark map X _b described in step (6), to the difference value difference map X The pixel at the position (m, n) in _d is filtered to obtain the denoising difference image X _N , according to the following rules: For a pixel point that satisfies the condition X _b (m, n) = 0, calculate the median value of all pixels in the 9×9 window of the pixel point in the difference difference map X _d , and assign the median value to the denoising difference the pixel value X _N (m,n) at the point (m,n) in the value map X _N ; For a pixel point that satisfies the condition X _b (m, n) = 1, assign the value X _d (m, n) of the pixel point in the difference difference map X _d to the point (m, n) in the denoising difference map X _N The pixel value X _N (m,n) at n); For a pixel that satisfies the condition X _b (m, n) = 2, calculate the median value of all pixels in the 11×11 window of the pixel point in the difference difference map X _d , and assign the median value to the denoising difference the pixel value X _N (m,n) at the point (m,n) in the value map X _N ; If there is a pixel that satisfies the condition X _b (m, n) = 3, then calculate the value of the Gaussian kernel function whose Gaussian scale is 3×3 for the pixel in the difference difference map X _d , and assign this value to the denoising The pixel value X _N (m, n) at the point (m, n) in the difference map X _N ; The values X _N (m, n) of all pixel points (m, n) constitute a denoising difference image X _N ={X _N (m, n)}.