CN103198480A - Remote sensing image change detection method based on area and Kmeans clustering - Google Patents

Abstract

The invention discloses a remote sensing image change detection method based on an area and Kmeans clustering. The remote sensing image change detection method based on the area and the Kmeans clustering mainly solves the problems that an existing detection result has isolated pixel points and cavities exist in the area. The remote sensing image change detection method based on the area and the Kmeans clustering comprises the steps of reading-in two registered images X1 and X2, wherein the time phase of the image X1 is different from the time phase of the image X2; constructing a difference image after the image X1 and the image X2 are filtered; carrying out maximum entropy threshold segmentation on the difference image to extract interested areas and confirm non-variable areas; calculating feature vectors of the interested areas and the non-variable areas; dividing all areas into two classes according to the characteristics of the confirmed non-variable areas and all the interested areas by means of the Kmeans clustering algorithm; using an area corresponding to a difference image with a high graying value as a variable area and other areas of the difference image are used as non-variable areas according to a clustering result, and finally obtaining a change detection result. According to the remote sensing image change detection method based on the area and the Kmeans clustering, the detection result can maintain the consistency of the interior of the area, the isolated pixel points are removed, and detection precision is improved. The remote sensing image change detection method based on the area and the Kmeans clustering can be used for resource monitoring and disaster evaluation.

Description

Method for detecting change of remote sensing image based on zone and Kmeans cluster

Technical field

The invention belongs to technical field of image processing, relate to remote sensing image and change detection, specifically a kind of method for detecting change of remote sensing image based on zone and Kmeans cluster is applicable to remote sensing image processing and analysis.

Background technology

Remote Sensing Imagery Change Detection is exactly the zone that changes in the phase remote sensing images when extracting two, be widely used in numerous areas economic and society, as fields such as land resource monitoring, aquatic monitoring, environmental monitoring, forestry monitoring, agricultural investigation, vegetation covering and weather monitoring, disaster monitoring and assessment, city management planning even military surveillance and battlefield estimations.

In the practical application, what the variation of the variation of variation, land deterioration and the desertification that covers as the soil, the variation of rivers and lakes, forest cover change, crops etc. presented at remote sensing images is not single or the variation of two or three pixels or piece, but the different regional change of area size shape, even if having the city management planning, disaster monitoring of the region of variation of more small size etc., the area of its region of variation also is can be to form the zonule that is communicated with.The resolution of remote sensing images improves gradually, and the connected pixel in real change zone increases thereupon, and is also more apparent important based on the method for detecting change of remote sensing image in zone.Human eye be not be by pixel read and understand the information of image, the zone carries out and be based on.Relatively obtain differential image for 2 o'clock phase images, region of variation wherein is the key area of remote Sensing Image Analysis and decipher, is visual important area.

For the change information that can effectively detect 2 o'clock phase remote sensing images with improve and change the degree of accuracy that detects, change the effective way that detection then is head it off from the angle of zone and vision attention.Ten thousand red woodss in 2012 etc. are at document " carrying out the research of SAR image change detection method in the interesting areas aspect. " (mapping journal, 2012,41 (2): 239-245.) proposed a kind ofly to carry out the SAR image change detection method in the interesting areas aspect.This method is at first extracted area-of-interest to differential image; Then each area-of-interest is regarded as a data point and calculated its feature, according to the threshold technology of threshold criteria all data points are cut apart, obtain final variation testing result.The area-of-interest that this method requires to extract must contain non-region of variation, otherwise the region of variation mistake can be divided into non-region of variation, reduces greatly and detects effect; This method is extracted region of interest with the FCM cluster and is had more omission, causes bigger transmission error and loses too much change information.Zhang Min in 2010 Xian Electronics Science and Technology University its patented claim " based on the method for detecting change of remote sensing image of conspicuousness tolerance " (number of patent application: 201210051159.7, publication number: proposed a kind of method for detecting change of remote sensing image based on Chi-square distance and conspicuousness tolerance CN102629377A).This method is at first carried out conspicuousness tolerance to differential chart and is obtained difference image, cut apart to difference image with based on the differential image of Chi-square distance respectively, utilize the segmentation result of the segmentation result correction difference image of Chi-square distance difference image, obtain final variation testing result.This method can suppress the generation of pseudo-change information, makes that by correction the location of region of variation is more accurate, but this method can not keep the edge of region of variation well owing to there is more omission.

Summary of the invention

The objective of the invention is at the deficiency in the above-mentioned method for detecting change of remote sensing image, propose a kind of method for detecting change of remote sensing image based on zone and Kmeans cluster, to reduce omission, accurately locate the region of variation edge, improve and change the precision that detects.

Implementation of the present invention comprises the steps:

(1) reads in two width of cloth remote sensing images X ₁And X ₂, and to X ₁And X ₂Carry out neighbour average filtering, obtain filtered image Y ₁And Y ₂

(2) according to filtered image Y ₁And Y ₂, structural differences image: Y=|Y ₁-Y ₂|;

(3) utilize the maximum entropy threshold T of principle of maximum entropy calculated difference image Y;

(4) pixel that is not less than maximum entropy threshold T by gray-scale value among the differential image Y forms several connected regions, and regards each connected region as a region of interest ROI, and all area-of-interests are constituted an area-of-interest set { ROI};

(5) with gray-scale value among the differential image Y less than the pixel of α * T as non-variation pixel certainly, and with certainly non-variation pixel set NRset of all certainly non-variation pixels formations among the differential image Y, should regard a non-region of variation NR certainly as by certainly non-variation pixel set NRset, wherein, 0＜α＜1;

(6) calculate area-of-interest set { the nucleus gray average K of area grayscale average G, the area grayscale maximal value M of each area-of-interest and area-of-interest among the ROI}, and the vector [G that these three features are constituted, M, K] as the proper vector of this area-of-interest;

(7) calculate area grayscale average U, the area grayscale maximal value V of non-region of variation NR certainly and the nucleus gray average S of non-region of variation certainly, and the vector [U, V, S] that these three features are constituted is as the proper vector of non-region of variation certainly;

(8) { each area-of-interest is as regional ensemble R={ROI} ∪ NR to be sorted among the ROI}, and ∪ is for getting the union symbol will to affirm the set of non-region of variation NR and area-of-interest;

(9) proper vector that adopts the Kmeans clustering algorithm to treat All Ranges among the specification area set R is carried out cluster, and that class zone that the area grayscale average component of cluster centre vector is bigger is as final region of variation, other remaining among differential image Y zones then as final non-region of variation, are obtained final variation testing result.

The present invention compared with prior art has following advantage:

(1) the present invention is cut apart the extraction area-of-interest owing to be that level in the zone changes detection to differential image, has improved segmentation precision, has reduced the transmission error of Region Segmentation to testing result.

(2) the present invention will affirm that the cooperation of non-variation set of pixels is that a certainly non-region of variation participates in the Kmeans cluster, the All Ranges that guarantee to participate in the Kmeans cluster necessarily contains region of variation and non-region of variation, thereby avoided participating in only to contain region of variation in the zone of Kmeans cluster the situation that the region of variation mistake is divided into non-region of variation is taken place.

Simulation result proves, compares with existing method, and the present invention can take into account false-alarm and omission, can keep the edge of region of variation, has improved the precision that detects, and has solved the more problem of omission in the existing method.

Description of drawings

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

Fig. 2 is three groups of 2 o'clock phase remote sensing images using of the present invention and changes and detect reference diagram;

Fig. 3 is with the present invention and the existing method variation testing result figure to Fig. 2 remote sensing images.

Embodiment

As follows to performing step of the present invention with reference to Fig. 1:

Step 1 is read in two width of cloth remote sensing images X of the registration that areal obtains constantly in difference ₁And X ₂, this two width of cloth image X ₁And X ₂Size be the capable J of I row.

Step 2 was to first o'clock phase images X ₁Carry out neighborhood filtering, obtain filtered first o'clock phase images Y ₁

The method of neighborhood filtering is a lot, and neighbours territory mean filter, neighbours territory medium filtering, eight neighbour average filterings, eight neighborhood medium filtering and N are for example arranged ³Neighbour average filterings etc., present embodiment adopts N ³Neighbour average filtering, its filter step is as follows:

2a) at first o'clock phase images X ₁In, with pixel (i, j) centered by, choose the neighborhood territory pixel gained set N of this pixel _Ij=(i, j), (i-2, j), (i+2, j), (i-1, j-1), (i-1, j), (i-1, j+1), (i, j-2), (i, j-1), (i, j+1), (i, j+2), (i+1, j-1), (i+1, j), (i+1, j+1) }, wherein, i and j are respectively the row and column sequence number in the image, and their limited range is i=3,, I-2, j=3 ..., J-2;

2b) calculation procedure 2a) set N in _IjIn whole averages of grey scale pixel values, and with this average as filtering after first o'clock phase images Y ₁At pixel (i, gray-scale value Y j) ₁(i, j);

2c) for first o'clock phase images X ₁Middle i=3 ..., I-2 and j=3 ..., the boundary pixel point beyond the J-2 (i, j), with this gray values of pixel points X ₁(i is j) as first o'clock phase images Y after the filtering ₁At pixel (i, gray-scale value Y j) ₁(i, j);

2d) repeating step 2a) to 2c), until handling first o'clock phase images X ₁In all pixels, obtain filtered first o'clock phase images Y ₁

Step 3 was to second o'clock phase images X ₂Carry out the neighborhood filtering same with step 2, obtain filtered second o'clock phase images Y ₂

Step 4 is to image Y after the filtering ₁And Y ₂The gray-scale value of corresponding pixel points subtract each other and take absolute value, obtain a width of cloth differential image Y, Y=|Y ₁-Y ₂|, the size of differential image Y is the capable J row of I.

Step 5, by the maximum entropy threshold T of following formula calculated difference image Y:

$T=\underset{\mathrm{th}}{\arg \max }\{-\underset{q=1}{\overset{\mathrm{th}}{\mathrm{\Σ}}}\frac{p_q}{p_{\mathrm{th}}}\log \frac{p_q}{p_{\mathrm{th}}}-\underset{q=\mathrm{th}+1}{\overset{\mathrm{Gray}}{\mathrm{\Σ}}}\frac{p_q}{1-p_{\mathrm{th}}}\log \frac{p_q}{1-p_{\mathrm{th}}}\},$

Wherein, gray-scale value th is between the minimal gray level of image and the arbitrary gray level between the maximum gray scale,

The value of gray-scale value th when making objective function get maximal value, p _q=n _q/ (I * J) is the probability that gray level q occurs in the image, n _qBe that gray level is the number of the pixel of q in the image,

Be all probability sums that are not more than the gray level appearance of threshold value th in the image, Gray is the maximum gray scale of image.

Step 6, the pixel that is not less than maximum entropy threshold T by gray-scale value among the differential image Y forms several connected regions, regards each connected region as a region of interest ROI, and all area-of-interests are constituted an area-of-interest set { ROI}.

Step 7 is obtained area-of-interest set { each regional proper vector among the ROI}.

7a) calculate area grayscale average G, the area grayscale maximal value M of region of interest ROI and the nucleus gray average K of area-of-interest according to following formula:

$G=\frac{1}{\mathrm{Nroi}}\underset{(i,j)\∈\mathrm{ROI}}{\mathrm{\Σ}}Y(i,j),$

$K=\frac{1}{\mathrm{Nkp}}\underset{(i,j)\∈\mathrm{KP}}{\mathrm{\Σ}}Y(i,j),$

$M=\underset{(i,j)\∈\mathrm{ROI}}{\max }\left(Y(i,j)\right),$

Wherein, ROI be area-of-interest set area-of-interest among the ROI}, Nroi is the number of pixel in the region of interest ROI, Y (i, j) be pixel among the differential image Y (i, gray-scale value j), KP are the nucleus of region of interest ROI, KP={ (i, j) | and Y (i, j)〉G, (i, j) ∈ ROI}, Nkp are the number of pixel among the nucleus KP;

7b) by step 7a) in area grayscale average G, area grayscale maximal value M and nucleus gray average K constitute vector [G, M, K], and should vector as the proper vector of region of interest ROI;

7c) repeating step 7a) to 7b), { All Ranges among the ROI} obtains area-of-interest set { each regional proper vector among the ROI} until handling the area-of-interest set.

Step 8, with gray-scale value among the differential image Y less than the pixel of α * T as non-variation pixel certainly, and with certainly non-variation pixel set NRset of all certainly non-variation pixels formations among the differential image Y, should regard a non-region of variation NR certainly as by certainly non-variation pixel set NRset, wherein, T is the maximum entropy threshold of differential image Y, and contraction coefficient 0＜α＜1 is chosen α=0.5 in embodiments of the present invention.

Step 9 is calculated the proper vector of non-region of variation NR certainly.

9a) be calculated as follows area grayscale average U, the area grayscale maximal value V of non-region of variation NR certainly and the nucleus gray average S of non-region of variation certainly:

$U=\frac{1}{\mathrm{Nnr}}\underset{(i,j)\∈\mathrm{NR}}{\mathrm{\Σ}}Y(i,j),$

$S=\frac{1}{\mathrm{Nnrkp}}\underset{(i,j)\∈\mathrm{KPNR}}{\mathrm{\Σ}}Y(i,j),$

$V=\underset{(i,j)\∈\mathrm{NR}}{\max }\left(Y(i,j)\right),$

Wherein, NR is for affirming non-region of variation, and Nnr is for affirming the number of pixel among the non-region of variation NR, Y (i, j) be that (KPNR is for affirming the nucleus of non-region of variation NR for i, gray-scale value j) for pixel among the differential image Y, KPNR={ (i, j) | and Y (i, j)〉U, (i, j) ∈ NR}, Nnrkp are the number of pixel among the nucleus KPNR;

9b) by step 9a) in area grayscale average U, area grayscale maximal value V and nucleus gray average S constitute vector [U, V, S], and should vector as the proper vector of non-region of variation NR certainly.

Step 10, { ROI} merges into a regional ensemble to be sorted: R={ROI} ∪ NR, wherein, ∪ is for getting the union symbol will to affirm the set of non-region of variation NR and area-of-interest.

Step 11, the proper vector that adopts clustering algorithm to treat All Ranges among the specification area set R is carried out cluster, and that class zone that the area grayscale average component of cluster centre vector is bigger is as final region of variation, other remaining among differential image Y zones then as final non-region of variation, are obtained final variation testing result.

Clustering algorithm has a lot, Kmeans clustering algorithm and FCM clustering algorithm etc. are for example arranged, present embodiment adopts document " Some Methods for Classification and Analysis of Multivariate Observations. " (In L.M.LeCam and J.Neyman, editors, Proceedings of Fifth Berkeley Symposium on Mathematical Statistics and Probability, volume1, pages281-297, Berkeley, CA, 1967.University of California Press.) the Kmeans clustering algorithm carries out cluster in.

Effect of the present invention can further specify by following experimental result and analysis:

1. experimental data and evaluation index

The used data of emulation experiment of the present invention are three groups of true remotely-sensed data collection.First group of two width of cloth Landsat-5 satellite TM the 4th band spectrum image that true remotely-sensed data collection is Italian Sardinia, two width of cloth image sizes are 300 * 412 pixels, the variation that takes place between them is by due to the rising of lake middle water level, comprises that 7626 change pixels and 115974 non-variation pixels; Its 2 o'clock phase original images detect reference diagram respectively shown in Fig. 2 (a1), Fig. 2 (a2), Fig. 2 (a3) with changing.Second group of two width of cloth Landsat-7EM+ the 4th band spectrum image that true remotely-sensed data collection is the Mexico countryside, the size of two width of cloth images is 512 * 512 pixels, the variation that takes place between them is to have been destroyed due to the large-area local vegetation by fire, comprises that 25589 change pixels and 236555 non-variation pixels; Its 2 o'clock phase original images detect reference diagram respectively shown in Fig. 2 (b1), Fig. 2 (b2), Fig. 2 (b3) with changing.The 3rd group of true remote sensing image data collection original image is respectively at forming at two width of cloth Landsat-5 satellite TM in west area, Italian Elba island the 4th wave band multispectral image in August, 1994 and in September, 1994, two width of cloth image sizes are 326 * 414 pixels, variation between them is because forest fire has destroyed due to a large amount of vegetation, change and comprise 2415 variation pixels and 132549 non-variation pixels in the reference diagram, its 2 o'clock phase original images detect reference diagram respectively as Fig. 2 (c1) with changing, Fig. 2 (c2), (change the white pixel region representation region of variation that detects in the reference diagram, black-pixel region is represented non-region of variation) shown in Fig. 2 (c3).

The objective evaluation index of weighing change detection algorithm among the present invention adopts false-alarm number, omission number, total wrong number and accuracy.

2. the method for contrast of the present invention's use is as described below:

Control methods 1 is Wan Honglin etc. " to be carried out the research of SAR image change detection method in the interesting areas aspect " and has proposed a kind ofly carrying out the SAR image change detection method in the interesting areas aspect at document, is designated as the ROI_FCM_THR method.

Control methods 2, be that Xian Electronics Science and Technology University is at its patented claim " based on the method for detecting change of remote sensing image of conspicuousness tolerance " (number of patent application: 201210051159.7, publication number: proposed a kind of method for detecting change of remote sensing image based on Chi-square distance and conspicuousness tolerance CN102629377A), be designated as the SMCD method.

3. experiment content and analysis

In order to verify that the inventive method changes detection at regional level and can improve Region Segmentation precision and accuracy of detection, the inventive method and ROI_FCM_THR method are compared; The region of variation of differentiation effectively and non-region of variation in order to verify three kinds of provincial characteristicss that the inventive method adopts compare the inventive method and SMCD method.

Emulation 1, to three groups of true remotely-sensed data collection in the accompanying drawing 2, change detection with existing ROI_FCM_THR method, change the evaluation index of testing result shown in first row, fourth line and the 7th row of table 1, it changes testing result figure as shown in Figure 3, wherein Fig. 3 (a1) is that the ROI_FCM_THR method is to the variation testing result figure of first group of true remotely-sensed data collection, Fig. 3 (b1) be the ROI_FCM_THR method to the variation testing result figure of second group of true remotely-sensed data collection, Fig. 3 (c1) is that the ROI_FCM_THR method is to the variation testing result figure of the 3rd group of true remotely-sensed data collection.

Emulation 2, to three groups of true remotely-sensed data collection in the accompanying drawing 2, change detection with existing SMCD method, change the evaluation index of testing result shown in second row, fifth line and the 8th row of table 1, it changes testing result figure as shown in Figure 3, wherein Fig. 3 (a2) is that the SMCD method is to the variation testing result figure of first group of true remotely-sensed data collection, Fig. 3 (b2) be the SMCD method to the variation testing result figure of second group of true remotely-sensed data collection, Fig. 3 (c2) is that the SMCD method is to the variation testing result figure of the 3rd group of true remotely-sensed data collection.

Emulation 3, to three groups of true remotely-sensed data collection in the accompanying drawing 2, change detection with the inventive method, change the evaluation index of testing result as the third line of table 1, the 6th row with shown in the 9th, it changes testing result figure as shown in Figure 3, wherein Fig. 3 (a3) is that the inventive method is to the variation testing result figure of first group of true remotely-sensed data collection, Fig. 3 (b3) be the inventive method to the variation testing result figure of second group of true remotely-sensed data collection, Fig. 3 (c3) is that the inventive method is to the variation testing result figure of the 3rd group of true remotely-sensed data collection.

As can be seen from Table 1, in the variation testing result of three kinds of methods, the inventive method is best to the overall evaluation of three groups of true remotely-sensed data collection.The inventive method is lacked 310 pixels to the total wrong number of first group of true remotely-sensed data collection than the total wrong number of ROI_FCM_THR method, lacks 97 pixels than the total wrong number of SMCD method; The total wrong number of second group of true remotely-sensed data collection is lacked 4034 pixels than the total wrong number of ROI_FCM_THR method, lack 870 pixels than the total wrong number of SMCD method; The total wrong number of the 3rd group of true remotely-sensed data collection is lacked 10341 pixels than the total wrong number of ROI_FCM_THR method, lack 170 pixels than the total wrong number of SMCD method; The total wrong number of ROI_FCM_THR method is too much, and the total wrong number of the inventive method is minimum in the existing method.The omission number of first group of remotely-sensed data collection is lacked 669 pixels respectively than the omission number of ROI_FCM_THR method, lack 337 pixels than the omission number of SMCD method; The omission number of second group of remotely-sensed data collection is lacked 5612 pixels respectively than the omission number of ROI_FCM_THR method, lack 1205 pixels than the omission number of SMCD method, the omission number of the 3rd group of remotely-sensed data collection is lacked 138 pixels than the omission number of SMCD method; The ROI_FCM_THR method is too much to the omission number of first and second groups of remotely-sensed data collection, is 2 times of the inventive method at least; The omission number of the inventive method is minimum in the existing method.This shows that the inventive method can detect change information comparatively comprehensively, exactly, reduce the omission number, take into account omission and false-alarm preferably, have the higher detection precision.As can be seen, compare with the SMCD method with the ROI_FCM_THR method from the design sketch 3 of three groups of experimental data collection, the inventive method can keep the marginal information of region of variation preferably, and the isolated pixel point of flase drop also is less.

Three groups of true remote sensing image data centralized procurements of table 1 change the performance evaluation of testing result with distinct methods

Claims (5)

1. the method for detecting change of remote sensing image based on zone and Kmeans cluster comprises the steps:

(1) reads in two width of cloth remote sensing images X ₁And X ₂, and to X ₁And X ₂Carry out neighbour average filtering, obtain filtered image Y ₁And Y ₂

(2) according to filtered image Y ₁And Y ₂, structural differences image: Y=|Y ₁-Y ₂|;

(3) utilize the maximum entropy threshold T of principle of maximum entropy calculated difference image Y;

(4) pixel that is not less than maximum entropy threshold T by gray-scale value among the differential image Y forms several connected regions, and regards each connected region as a region of interest ROI, and all area-of-interests are constituted an area-of-interest set { ROI};

(5) with gray-scale value among the differential image Y less than the pixel of α * T as non-variation pixel certainly, by all affirm that non-variation pixel constitutes a certainly non-variation pixel set NRset among the differential image Y, to affirm that non-variation pixel set NRset regards a non-region of variation NR certainly as, wherein, 0＜α＜1;

(6) calculate area-of-interest set { the nucleus gray average K of area grayscale average G, the area grayscale maximal value M of each area-of-interest and area-of-interest among the ROI}, and the vector [G that these three features are constituted, M, K] as the proper vector of this area-of-interest;

(7) calculate area grayscale average U, the area grayscale maximal value V of non-region of variation NR certainly and the nucleus gray average S of non-region of variation certainly, and the vector [U, V, S] that these three features are constituted is as the proper vector of non-region of variation certainly;

(8) { each area-of-interest is as regional ensemble R={ROI} ∪ NR to be sorted among the ROI}, and ∪ is for getting the union symbol will to affirm the set of non-region of variation NR and area-of-interest;

(9) proper vector that adopts the Kmeans clustering algorithm to treat All Ranges among the specification area set R is carried out cluster, and that class zone that the area grayscale average component of cluster centre vector is bigger is as final region of variation, other remaining among differential image Y zones then as final non-region of variation, are obtained final variation testing result.

2. according to claim 1 based on zone and the method for detecting change of remote sensing image of Kmeans cluster, wherein step (1) is described to X ₁And X ₂Carry out neighbour average filtering, adopt N ³Neighbour average filtering, its filter step is as follows:

1a) at image X ₁In, with pixel (i, j) centered by, choose the neighborhood territory pixel gained set of this pixel: N _Ij=(i, j), (i-2, j), (i+2, j), (i-1, j-1), (i-1, j), (i-1, j+1), (i, j-2), (i, j-1), (i, j+1), (i, j+2), (i+1, j-1), (i+1, j), (i+1, j+1) }, wherein, i and j are respectively the row and column sequence number in the image, and their limited range is i=3 ... I-2, j=3 ..., J-2, I are image X ₁The maximum row sequence number, J is image X ₁The maximum column sequence number;

1b) calculation procedure 1a) set N in _IjIn whole averages of grey scale pixel values, and with this average as filtering after image Y ₁Gray-scale value Y at this point ₁(i, j);

1c) for image X ₁Middle i=3 ..., I-2 and j=3 ..., the boundary pixel point beyond the J-2 (i, j), with this gray values of pixel points as filtering after image Y ₁Gray-scale value Y at this point ₁(i, j);

1d) repeating step 1a) to 1c), until handling image X ₁Whole pixels obtains image Y after the filtering ₁

1e) according to step 1a) to 1d) handle image X ₂In all pixels, obtain image Y after the filtering ₂

3. according to claim 1 based on zone and the method for detecting change of remote sensing image of Kmeans cluster, the described maximum entropy threshold T that utilizes principle of maximum entropy calculated difference image Y of step (3) wherein, undertaken by following formula:

$T=\underset{\mathrm{th}}{\arg \max }\{-\underset{q=1}{\overset{\mathrm{th}}{\mathrm{\Σ}}}\frac{p_q}{p_{\mathrm{th}}}\log \frac{p_q}{p_{\mathrm{th}}}-\underset{q=\mathrm{th}+1}{\overset{\mathrm{Gray}}{\mathrm{\Σ}}}\frac{p_q}{1-p_{\mathrm{th}}}\log \frac{p_q}{1-p_{\mathrm{th}}}\},$

Wherein, gray-scale value th is between the minimal gray level of image and the arbitrary gray level between the maximum gray scale, The value of gray-scale value th when making objective function get maximal value, p _q=n _q/ (I * J) is the probability that gray level q occurs in the image, n _qBe that gray level is the number of the pixel of q in the image,

Be all probability sums that are not more than the gray level appearance of threshold value th in the image, Gray is the maximum gray scale of image.

4. according to claim 1 based on zone and the method for detecting change of remote sensing image of Kmeans cluster, the wherein described calculating area-of-interest set of step (6) the nucleus gray average K of area grayscale average G, the area grayscale maximal value M of each area-of-interest and area-of-interest among the ROI} is calculated as follows:

$G=\frac{1}{\mathrm{Nroi}}\underset{(i,j)\∈\mathrm{ROI}}{\mathrm{\Σ}}Y(i,j),$

$K=\frac{1}{\mathrm{Nkp}}\underset{(i,j)\∈\mathrm{KP}}{\mathrm{\Σ}}Y(i,j),$

$M=\underset{(i,j)\∈\mathrm{ROI}}{\max }\left(Y(i,j)\right),$

Wherein, ROI be area-of-interest set area-of-interest among the ROI}, Nroi is the number of pixel in the region of interest ROI, Y (i, j) be pixel among the differential image Y (i, gray-scale value j), KP are the nucleus of region of interest ROI, KP={ (i, j) | and Y (i, j)〉G, (i, j) ∈ ROI}, Nkp are the number of pixel among the nucleus KP.

5. according to claim 1 based on zone and the method for detecting change of remote sensing image of Kmeans cluster, wherein area grayscale average U, the area grayscale maximal value V of the sure non-region of variation NR of the described calculating of step (7) and the nucleus gray average S of non-region of variation certainly are calculated as follows:

$U=\frac{1}{\mathrm{Nnr}}\underset{(i,j)\∈\mathrm{NR}}{\mathrm{\Σ}}Y(i,j),$

$S=\frac{1}{\mathrm{Nnrkp}}\underset{(i,j)\∈\mathrm{KPNR}}{\mathrm{\Σ}}Y(i,j),$

$V=\underset{(i,j)\∈\mathrm{NR}}{\max }\left(Y(i,j)\right),$

Wherein, NR affirms non-region of variation, and Nnr is for affirming the number of pixel among the non-region of variation NR, and KPNR is for affirming the nucleus of non-region of variation NR, KPNR={ (i, j) | Y (i, j)〉U, (i, j) ∈ NR}, Nnrkp are the number of pixel among the nucleus KPNR.

Images