CN106485693A - Card side converts the multi-temporal remote sensing image change detecting method with reference to MRF model - Google Patents

Abstract

The invention discloses multi-temporal remote sensing image change detecting method of card side's conversion with reference to MRF model, step is followed successively by, input multi-temporal remote sensing image, Image registration, radiation normalization correction, calculating multidate difference image, iteration CST change detection, input MRF model, the modulus value based on difference image obtain final change testing result.Instant invention overcomes prior art is difficult to solve the problems, such as that high spatial resolution remote sense image background information complexity, noise jamming are serious.

Description

Card side converts the multi-temporal remote sensing image change detecting method with reference to MRF model

Technical field

The invention belongs to Remote Sensing Image Processing Technology field, more particularly to multidate of card side's conversion with reference to MRF model Remote sensing image variation detection method.

Background technology

With the continuous accumulation of multidate high-definition remote sensing data and the successive foundation of spatial database, how from this Extract in a little remotely-sensed datas and detect that change information has become the important subject of remote sensing science and Geographical Information Sciences.According to The remote sensing image of the same area difference phase, can extract the information of the dynamic changes such as city, environment, be resource management and rule Draw, the department such as environmental protection provides the foundation of science decision.China " 12 " will increase expansion enforcement Eleventh Five-Year Plan and have been turned on The high-resolution earth observation engineering of enforcement, concern include high-definition remote sensing target and space environment signature analysis and highly reliable Property basic theory and the key technology research such as automatic interpretation, become solution national security and the great demand of socio-economic development Research focus.

The change detection of remote sensing image is exactly in remotely-sensed data never of the same period, quantitatively analyzes and determine earth's surface change Feature and process.Scholars propose many effective detection algorithms from different angles with application study, such as change arrow Measure analytic approach (Change Vector Analysis, CVA), be based on clustering method of Fuzzy C-means (FCM) etc..Wherein, Traditional multidate optical remote sensing change detection based on card side's conversion (Chi-Squared Transform, CST), first calculates The average of difference image and variance matrix, are then based on confidence level again, determine the threshold value of change detection, and then obtain change inspection Survey result.In such technology, the use of the deficiency of CST is spectral information only using multidate high-resolution difference image, does not have Utilization space information.In addition, in the average for calculating difference image and variance matrix, the precision of estimation is not high.

For the problems referred to above, it is necessary to study new High Resolution Visible Light Remote Sensing Imagery Change Detection technology and come effective gram Take above-mentioned difficult point.

Content of the invention

In order to solve the technical problem of above-mentioned background technology proposition, the present invention is intended to provide card side is converted with reference to MRF model Multi-temporal remote sensing image change detecting method, overcome prior art and be difficult to solve high spatial resolution remote sense image background letter The problem that breath is complicated, noise jamming is serious.

In order to realize above-mentioned technical purpose, the technical scheme is that：

Card side converts the multi-temporal remote sensing image change detecting method with reference to MRF model, comprises the following steps：

(1) the high-resolution optical remote sensing image of two phases is input into, is designated as X respectively₁And X₂；

(2) to X₁And X₂Carry out Image registration；

(3) using Multivariate alteration detection method to X₁And X₂Carry out radiation normalization correction；

(4) multidate difference image D is calculated_X=X₁-X₂；

(5) multidate difference image D is initialized_XIn non-changing region, calculate the mean value vector in non-changing region and side Difference matrix, and calculate the chi-square value of each point in multidate difference image；

(6) on the basis of given confidence level, detection threshold value is calculated, and is detected according to detection threshold value, determine Region of variation in multidate difference image and non-changing region；

(7) the non-changing region that the non-changing region for determining step (6) is determined with step (5) is compared, if two Person is consistent, then the testing result for obtaining step (6), as change testing result step (6) is determined if both are different Non-changing region as new non-changing region, return to step (5), loop iteration；

(8) the change testing result for obtaining step (7) is used as the input of MRF model, and is based on multidate difference image D_XModulus value obtain final change testing result.

Further, the detailed process of step (8) is as follows：

A () calculates multidate difference image D_XModulus value：

In above formula, X_1bAnd X_2bRepresent X respectively₁And X₂The image of b wave band, B represent each phase remote sensing image Wave band number, (i, j) are the coordinates of image；

B () builds the energy function of MRF model：

In above formula, U_dataRepresent data item, U_contextRepresentation space local energy item, M₁And M₂Represent the height of image respectively And width, the value of Y (i, j) expression change testing result coordinate (i, j), Y^S(i, j) is the neighborhood system of coordinate (i, j)；

C () solves U using ICM optimized algorithm and minimizes, obtain final change testing result.

Further, U_dataIt is further represented as,

In above formula,μ_Y(i,j)∈{μ_n,μ_c,And μ_nRepresent the variance and all in non-changing region respectively Value,And μ_cRepresent variance and the average of region of variation respectively.

Further, it is characterised in that U_contextIt is further represented as,

In above formula, (p, q) is the neighborhood coordinate of (i, j), and β is the parameter for controlling space local energy item, and I is to indicate letter Number, is defined as follows：

Further, in step (2), to X₁And X₂Carrying out Image registration includes geometric approximate correction and geometric accurate correction, institute State the process of geometric approximate correction：

(A) X is selected₁And X₂Respectively as reference images and image to be corrected；

(B) gather ground control point in reference images and image to be corrected respectively, the quantity of ground control point more than etc. In 9, and ground control point is evenly distributed on image；

(C) mean square error at calculating benchmark image and each ground control point of image to be corrected；

(D) treat correcting image using polynomial method to be corrected；

(E) treating correcting image using bilinear interpolation carries out resampling；

The geometric accurate correction is by the remote sensing image through geometric approximate correction, is entered with Triangulation Method using Auto-matching Row correction.

Further, in step (5), detection threshold value is calculated using following formula：

In above formula, 1- α is confidence level, C_ijRepresent D_XIn the chi-square value at coordinate (i, j) place,For detection threshold value.

The beneficial effect that is brought using technique scheme：

After MRF model to be incorporated into the present invention CST testing result so that the change converted based on CST is detected with sky Between restriction ability.In change detection, using the method for iteration, estimate average and the standard deviation in non-changing region, only overcome The deficiency of average and variance matrix is calculated merely with whole multiband difference image so that the result of change detection is relatively reliable, Also more there is robustness.

Description of the drawings

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

Fig. 2 (a), 2 (b) are regional the 3rd ripple of high-resolution IKONOS image of Saudi Arabia Mina in January, 2007 respectively Regional the 3rd wave band schematic diagram of high-resolution IKONOS image of section schematic diagram, the Saudi Mina in December, 2007；

Fig. 3 is the reference picture of change detection；

Fig. 4 (a), 4 (b), 4 (c), 4 (d) are EM-CVA algorithm, ICST algorithm, RCST algorithm, inventive algorithm respectively Testing result schematic diagram.

Specific embodiment

Below with reference to accompanying drawing, technical scheme is described in detail.

As shown in figure 1, card side converts the multi-temporal remote sensing image change detecting method with reference to MRF model, step is as follows：

Step 1：Input the same area, two panel height resolution Optical remote sensing images of different phases, are designated as respectively：X₁And X₂.

Step 2：To X₁And X₂Image registration is carried out, is divided into thick correction and two steps of fine correction：

For geometric approximate correction, realized using the correlation function in ENVI4.8 software, concrete operation step is：

(1) reference images and image to be corrected are shown；

(2) gather ground control point GCPs, GCPs should be evenly distributed in entire image, the number of GCPs at least above etc. In 9；

(3) mean square error is calculated；

(4) treat correcting image using polynomial method to be corrected；

(5) resampling output is carried out using bilinear interpolation, if unknown function f is sought in the value of point P=(x, y), it is assumed that Know function f in Q₁₁=(x₁,y₁),Q₁₂=(x₁,y₂),Q₂₁=(x₂,y₁), and Q₂₂=(x₂,y₂) four points value, if selected One coordinate system causes the coordinate of this four points to be respectively (0,0), (0,1), (1,0) and (1,1), then bilinear interpolation is public Formula can just be expressed as：

F (x, y) ≈ f (0,0) (1-x) (1-y)+f (1,0) x (1-y)+f (0,1) (1-x) y+f (1,1) xy is for geometry essence Correction, the multi-spectrum remote sensing image data through geometric approximate correction carry out geometry essence using Auto-matching and Triangulation Method Correction.Delaunay triangulation network is built using incremental algorithm, to each triangle, using its three summits ranks number with The geographical coordinate of its corresponding reference images same place determining the affine Transform Model parameter of the triangle interior, to be corrected Image is corrected, the remote sensing shadow after being corrected.

Step 3：Using Multivariate alteration detection (Multivariate Alteration Detection, MAD) method to X₁ And X₂Carry out radiation normalization correction.X is found first₁And X₂One linear combination of each wave band brightness value, obtains change information increasing Strong difference image, determines change and non-region of variation by threshold value, then by the corresponding two phases pixel of non-region of variation To mapping equation, complete relative detector calibration.

Step 4：To the multidate high resolution image X being input into₁And X₂, calculate multidate difference image D_X：

D_X=X₁-X₂

Step 5：By whole D_XIt is considered as non-changing region, and its mean value vector m and variance matrix Σ is calculated, and calculates difference The chi-square value of each point of image：

C_ij=(x_ij-m)^TΣ^-1(x_ij- m)～χ²(b)

In above formula, C_ijRepresent the chi-square value of difference image (i, j) coordinate points, which obeys chi square distribution of the free degree for b； x_ijRepresent vector value of the difference image in (i, j) coordinate points；Σ^-1Represent the inverse matrix of variance matrix；B represents difference image Wave band number.

Step 6：Given confidence level 1- α, calculates detection threshold value using following formula：

When confidence level is for 1- α, C_ijValue be more thanProbability be α.If α value is less, it is more thanC_ijOut-of-bounds point (outlier) or change point can be considered as, thereby determine that threshold value isAnd according to the threshold Value obtains preliminary testing result.

Step 7：The non-changing region that the non-changing region determined in step 6 testing result and step 5 are determined is (whole D_X) be compared, if both are consistent, the testing result that step 6 is obtained as change testing result, if both are not With the non-changing region for then determining step 6 is used as new non-changing region, return to step 5, loop iteration.

Step 8：The change testing result that step 7 is obtained is used as the input of MRF model, and the modulus value based on difference image Obtain final change testing result.Implementing step is：

(1) D is calculated first_XModulus value X_M：

In above formula, X_1bAnd X_2bRepresent first phase and the second phase multispectral image X respectively₁And X₂B wave band image, B Represent the wave band number of each phase remote sensing image, (i, j) is the coordinate of image；

(2) energy function for building MRF model is as follows：

In above formula, U_dataRepresent data item, U_contextRepresentation space local energy item, M₁And M₂Represent the height of image respectively And width, the value of Y (i, j) expression change testing result coordinate (i, j), Y^S(i, j) is the neighborhood system of coordinate (i, j).

Wherein, data item U_dataCan be further represented as：

In above formula,μ_Y(i,j)∈{μ_n,μ_c,And μ_nRespectively represent non-changing region variance and Average,And μ_cRepresent variance and the average of region of variation respectively.

Wherein, local local spatial energy item U_contextCan be further represented as：

In above formula, (p, q) is neighborhood coordinate.8 neighborhood systems are adopted in the present invention, and β is control space local energy item Parameter, I are indicator function, are defined as follows：

(3) U is solved using ICM optimized algorithm to minimize, obtain final change testing result.

The effect of the present invention can be further illustrated by following experimental result and analysis：

The experimental data of the present invention is the multidate IKNOS high-resolution image data in Saudi Mina area, figure As size is 700 × 950, using tri- wave bands of B1, B2 and B3, as the remote sensing shadow of B3 wave band when Fig. 2 (a) and Fig. 2 (b) are two Picture.

In order to verify effectiveness of the invention, change detecting method of the present invention is compared with following change detecting methods Right：

(1) [Bruzzone L. of Italy etc. is in article " Automatic for EM method (CVA-EM) based on CVA analysis of difference image for unsupervised change detection”(IEEE Transactions on Geoscience and Remote Sensing,2000,38(3):1171-1182.) in carried Detection method].

(2) [B.Descl é e, P.Bogaert, and P.Defourny is in text for CST detection (ICST) method based on iteration Chapter " Forest change detection by statistical object-based method " (Remote Sensing of Environment,2006,102(1-2):The method carried in 1-12.)].

(3) [Aiye Shi etc. is in article " Unsupervised for CST detection (RCST) method based on Robust Estimation change detection based on robust chi-squared transform for bitemporal remotely sensed images”(International of Remote Sensing,2006,102(1-2):1-12.) Middle carried method].

(4) the inventive method.

Fig. 3 is the reference picture of change detection.Under conditions of confidence level 0.99, the testing result of above-mentioned four kinds of methods As shown in Fig. 4 (a), 4 (b), 4 (c), 4 (d), detection performance false retrieval number FP, missing inspection number FN, total error number OE and Kappa coefficient Four indexs are weighing.FP, FN and OE are closer to 0, Kappa coefficient and are closer to 1, show that the performance of change detecting method is got over Good.Testing result is as shown in table 1.From table 1, the detection method performance carried by the present invention is better than other three kinds of detection methods, This shows that the change detecting method carried by the present invention is effective.

Table 1

Above example technological thought only to illustrate the invention, it is impossible to which protection scope of the present invention is limited with this, every According to technological thought proposed by the present invention, any change that is done on the basis of technical scheme, the scope of the present invention is each fallen within Within.

Claims (6)

1. the conversion of card side is with reference to the multi-temporal remote sensing image change detecting method of MRF model, it is characterised in that including following step Suddenly：

(1) the high-resolution optical remote sensing image of two phases is input into, is designated as X respectively₁And X₂；

(2) to X₁And X₂Carry out Image registration；

(3) using Multivariate alteration detection method respectively to X₁And X₂Carry out radiation normalization correction；

(4) multidate difference image D is calculated_X=X₁-X₂；

(5) multidate difference image D is initialized_XIn non-changing region, calculate the mean value vector in non-changing region and variance square Battle array, and calculate the chi-square value of each point in multidate difference image；

(6) on the basis of given confidence level, detection threshold value is calculated, and is detected according to detection threshold value, when determining many Region of variation in facial difference image and non-changing region；

(7) the non-changing region that the non-changing region for determining step (6) is determined with step (5) is compared, if both one Cause, then, used as change testing result, if both are different, what step (6) was determined is non-for the testing result for obtaining step (6) Region of variation is used as new non-changing region, return to step (5), loop iteration；

(8) the change testing result for obtaining step (7) is used as the input of MRF model, and is based on multidate difference image D_XMould It is worth to final change testing result.

2. card side converts the multi-temporal remote sensing image change detecting method with reference to MRF model, step according to claim 1 (8) detailed process is as follows：

A () calculates multidate difference image D_XModulus value：

$X_M(i,j)=\sqrt{\underset{b=1}{\overset{B}{\mathrm{\Σ}}}{\left(X_{1b}\right(i,j)-X_{2b}(i,j\left)\right)}^2}$

In above formula, X_1bAnd X_2bRepresent X respectively₁And X₂The image of b wave band, B represent the wave band of each phase remote sensing image Number, (i, j) are the coordinates of image；

B () builds the energy function of MRF model：

$U=\underset{i=1}{\overset{M_1}{\mathrm{\Σ}}}\underset{j}{\overset{M_2}{\mathrm{\Σ}}}{U}_{data}\left(X_M\right(i,j),Y(i,j\left)\right)+U_{context}\left(Y\right(i,j),Y^S(i,j\left)\right)$

In above formula, U_dataRepresent data item, U_contextRepresentation space local energy item, M₁And M₂Represent the height and width of image, Y respectively (i, j) represents the value of change testing result coordinate (i, j), Y^S(i, j) is the neighborhood system of coordinate (i, j)；

C () solves U using ICM optimized algorithm and minimizes, obtain final change testing result.

3. card side converts the multi-temporal remote sensing image change detecting method with reference to MRF model, its feature according to claim 2 It is：U_dataIt is further represented as,

$U_{data}\left(X_M\right(i,j),Y(i,j\left)\right)=\frac{1}{2}ln\left|2{\mathrm{\π\σ}}_{Y(i,j)}^2\right|+\frac{1}{2}{\left(Y\right(i,j)-{\mathrm{\μ}}_{Y(i,j)})}^2{\[{\mathrm{\σ}}_{Y(i,j)}^2\]}^{-1}$

In above formula,μ_Y(i,j)∈{μ_n,μ_c,And μ_nRepresent variance and the average in non-changing region respectively,And μ_cRepresent variance and the average of region of variation respectively.

4. card side converts the multi-temporal remote sensing image change detecting method with reference to MRF model, its feature according to claim 2 It is：U_contextIt is further represented as,

$U_{context}\left(Y\right(i,j),Y^S(i,j\left)\right)=\underset{(p,q)\∈Y^S(i,j)}{\Σ}\mathrm{\β}I\left(Y\right(i,j),Y(p,q\left)\right)$

In above formula, (p, q) is the neighborhood coordinate of (i, j), and β is the parameter for controlling space local energy item, and I is indicator function, fixed Justice is as follows：

$I\left(Y\right(i,j),Y(p,q\left)\right)=\left\{\begin{array}{cc}-1& ifY(i,j)=Y(p,q)\\ 0& Otherwise\end{array}\right..$

5. card side converts the multi-temporal remote sensing image change detecting method with reference to MRF model, its feature according to claim 1 It is：In step (2), to X₁And X₂Carrying out Image registration includes geometric approximate correction and geometric accurate correction, the geometric approximate correction Process：

(A) X is selected₁And X₂Respectively as reference images and image to be corrected；

(B) ground control point is gathered in reference images and image to be corrected respectively, and the quantity of ground control point is more than or equal to 9, And ground control point is evenly distributed on image；

(C) mean square error at calculating benchmark image and each ground control point of image to be corrected；

(D) treat correcting image using polynomial method to be corrected；

(E) treating correcting image using bilinear interpolation carries out resampling；

The geometric accurate correction is by the remote sensing image through geometric approximate correction, carries out school using Auto-matching and Triangulation Method Just.

6. card side converts the multi-temporal remote sensing image change detecting method with reference to MRF model, its feature according to claim 1 It is, in step (5), detection threshold value is calculated using following formula：

$P(C_{ij}<{\mathrm{\&chi;}}_{1-\mathrm{\&alpha;}}^2(b\left)\right)=1-\mathrm{\&alpha;}$

In above formula, 1- α is confidence level, C_ijRepresent D_XIn the chi-square value at coordinate (i, j) place,For detection threshold value.