CN109903318B - Registration and change detection integrated processing method based on high-resolution remote sensing image - Google Patents

Registration and change detection integrated processing method based on high-resolution remote sensing image Download PDF

Info

Publication number
CN109903318B
CN109903318B CN201910311014.8A CN201910311014A CN109903318B CN 109903318 B CN109903318 B CN 109903318B CN 201910311014 A CN201910311014 A CN 201910311014A CN 109903318 B CN109903318 B CN 109903318B
Authority
CN
China
Prior art keywords
registration
change
model
establishing
change detection
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201910311014.8A
Other languages
Chinese (zh)
Other versions
CN109903318A (en
Inventor
王慧贤
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Institute of Electronics of CAS
Original Assignee
Institute of Electronics of CAS
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Institute of Electronics of CAS filed Critical Institute of Electronics of CAS
Priority to CN201910311014.8A priority Critical patent/CN109903318B/en
Publication of CN109903318A publication Critical patent/CN109903318A/en
Application granted granted Critical
Publication of CN109903318B publication Critical patent/CN109903318B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Image Processing (AREA)
  • Image Analysis (AREA)

Abstract

A building registration and change detection integrated processing method based on a high-resolution remote sensing image takes registration errors as a spectral change decision factor, and change information is iteratively fed back to a resolving process of a variation registration model in a weight mode. In order to more accurately detect the real change of the specific target of the building, the invention adopts the difference of the multi-scale maximum morphological outline building detection indexes as another decision factor. And finally, establishing a probability model for the two decision factors of the change reflected by the registration error and the difference of the building detection index under a D-S evidence theory framework for fusion processing, thereby obtaining the change detection result of the building. The invention effectively solves the limitation of single processing by utilizing an integrated processing idea, fundamentally solves the problems of influence of registration errors on change detection results and reduction of registration accuracy due to changes, and further improves the quality of registration and change detection.

Description

Registration and change detection integrated processing method based on high-resolution remote sensing image
Technical Field
The invention relates to the field of image processing, in particular to a registration and change detection integrated processing method based on a high-resolution remote sensing image.
Background
With the continuous advance of the urbanization process, how to timely and accurately monitor the change of land utilization/coverage around the city, especially the change of the most active buildings, becomes a problem to be solved urgently at present. The remote sensing change detection technology provides an effective technical approach for solving the problem.
The image registration is used as a key link of change detection, and the quality of the image registration directly influences the precision of the change detection. In the conventional change detection and registration method, registration and change detection are often regarded as two independent processing steps, i.e., registration is performed before change detection, and a flowchart is shown in fig. 1. However, in actual processing, not only the registration error may cause an erroneous change detection result, but also the change information may in turn cause a reduction in the registration accuracy and even a failure in the registration. Although there is a scheme in the prior art that a pixel-based change vector is projected onto a polar coordinate system, and the influence of a registration error is weakened through multi-scale analysis according to the characteristic that the change caused by the registration error is reduced along with the reduction of spatial resolution, the influence of the registration error on change detection is not fundamentally solved.
Disclosure of Invention
In view of the above, the main object of the present invention is to provide an integrated process of registration and change detection based on high resolution remote sensing images, so as to at least partially solve the above mentioned technical problems.
In order to achieve the purpose, the technical scheme of the invention is as follows:
a registration and change detection integrated processing method based on a high-resolution remote sensing image comprises the following steps:
step 1, establishing a registration and change detection integrated processing variational model of the remote sensing image, namely adding feedback of surface feature change into the registration model, fully considering the existence of change information, and simultaneously taking the change obtained by registration error as a spectral change decision factor to carry out subsequent change detection so as to realize the integrated processing of registration and change detection;
step 2, establishing a probability model of a change weight w in the registration and change detection integrated processing variation model;
step 3, carrying out optimization solution on the registration and change detection integrated processing variation model by using a minimum variation energy function;
and 4, fusing the registration error change information obtained in the step 3 and the feature change information of the ground object based on an evidence theory, thereby realizing the integrated processing of registration and change detection of the high-resolution remote sensing image.
The land formations include buildings, vegetation and/or bodies of water.
The step 1 specifically comprises the following substeps:
let the later image be f (x, y, t) and the earlier image be
Figure BDA0002030446140000021
Step 1.1, establishing data items in the integrated processing variation model
Figure BDA0002030446140000022
Vector representationHas the formula of (1)
Figure BDA0002030446140000023
Wherein
Figure BDA0002030446140000024
In the formula: u. of1，u4For scale parameters in the X and Y directions in affine transformations, u2，u3As a rotation parameter, u5，u6As a translation parameter, u7As a contrast parameter, u8As a brightness parameter, fx，fyAnd ftIs the partial derivative of f (x, y, t) to space and time, and w is the variation weight of each point participating in the operation;
step 1.2, establishing a smooth prior term in an integrated processing variation model
Figure BDA0002030446140000025
Figure BDA0002030446140000026
Step 1.3, establishing a linear constraint prior term in the integral processing variational model
Figure BDA0002030446140000027
Figure BDA0002030446140000031
Where Nl is the number of lines, l is the l-th line, Np is the number of points on line l, i is the i-th point on line l,
Figure BDA0002030446140000032
and
Figure BDA0002030446140000033
is the transformation parameter of the i-1, i and i +1 point on the straight line l;
step 1.4, establishing a whole variation model;
Figure BDA0002030446140000034
the probability model in step 2 is as follows:
Figure BDA0002030446140000035
wherein
d(x，y)＝f(x，y，t)-f(x，y，t-1)(20)；
I.e. d (x, y) represents the difference in gray levels between the registered image pairs, and the variables c and σ are constants.
The Euler-Lagrange equation of the variation model in the step 3 is as follows:
Figure BDA0002030446140000036
in the formula:
Figure BDA0002030446140000037
is a Laplace operator, and
Figure BDA0002030446140000038
refers to two adjacent points on the same straight line, the left and the right of which take the point as the center
Figure BDA0002030446140000039
Average value of (d);
regularization is performed on equation (8) during the solution process, adding a regularization term L, which becomes
Figure BDA00020304461400000310
In the formula: l is an 8X 8 diagonal element of lambdaiThe off-diagonal element is a diagonal matrix of 0, λ in this contextiAre all 1.0 × 1010；
The discrete expression of formula (9) is
Figure BDA0002030446140000041
In the formula:
Figure BDA0002030446140000042
is in calculating Laplace operator
Figure BDA0002030446140000043
Is solved by direct iteration, i.e. the mean value of adjacent points is obtained
Figure BDA0002030446140000044
In the formula:
Figure BDA0002030446140000045
and
Figure BDA0002030446140000046
is iterated by the current iter
Figure BDA0002030446140000047
To make an estimation, an initial estimation
Figure BDA0002030446140000048
Estimating in a smaller neighborhood by using formula (12);
Figure BDA0002030446140000049
the step 4 comprises the following substeps:
step 4.1, establishing a probability density function of feature change information of the ground features;
step 4.2, establishing a probability density function of the registration error weight variable, and establishing a model by adopting 1-w, wherein the model is as follows:
Figure BDA00020304461400000410
in the formula: p1＝0.01；x1T is a threshold, obtained using a moment keeping threshold (moment keeping threshold), variable τ is used to control the slope of the distribution, maximum values 1 and x of the 1-w variable are used1Is expressed by the difference of (1-x), τ ═ x1) 6; x is the value of the variable 1-w;
and 4.3, establishing a decision model for fusion processing of the change information and the surface feature based on the evidence theory.
Based on the technical scheme, the invention has the beneficial effects that:
the method effectively solves the limitation of independent processing, fundamentally solves the problems that the registration error influences the change detection result and the registration precision is reduced due to the change, further improves the quality of registration and change detection, and simultaneously improves the precision of image registration and change detection.
Drawings
FIG. 1 is a conventional registration and change detection process;
FIG. 2 is a false detection of registration errors and change detection resulting from conventional processing;
FIG. 3 is a flow chart of the integrated process of the present invention;
FIG. 4 is raw data to be processed;
FIG. 5 is a comparison of registration results;
FIG. 6 is a registration error weight comparison of the integrated processing and the separate processing modes;
fig. 7 is a comparison of the integrated process and the divided process variation detection results.
Detailed Description
In order that the objects, technical solutions and advantages of the present invention will become more apparent, the present invention will be further described in detail with reference to the accompanying drawings in conjunction with the following specific embodiments.
It should be noted that in the drawings or description, the same drawing reference numerals are used for similar or identical parts. Implementations not depicted or described in the drawings are of a form known to those of ordinary skill in the art. Additionally, while exemplifications of parameters including particular values may be provided herein, it is to be understood that the parameters need not be exactly equal to the respective values, but may be approximated to the respective values within acceptable error margins or design constraints.
The building as a special target has a certain height, and is inclined to different directions due to different shooting angles, the top surface of the building has different edges, and the side surfaces of the building have different characteristics, which brings great difficulty to the registration and change detection. For example, in a conventional processing manner, the change information of a building may cause the existence of a registration error, and the registration error may cause the wrong change detection result, as shown in fig. 2, the white frame in the figure is the false detection of the registration error and the change detection caused by the conventional processing defect.
The invention takes a building as an example, and improves the registration and change detection results by utilizing a registration and change detection integrated processing thought based on a variational theory, which specifically utilizes a variational method to carry out image registration, if the influence of change information is added in the model establishment, a better result should be obtained by taking a registration error as the change information, namely an integrated thought. This is because the inventor found that the similarity measure of the early and late images in the registration process provides a kind of consistency information for the registration algorithm, and in turn provides a kind of variation information, i.e. the points with large registration errors indicate a high possibility of variation. Therefore, the registration and the change detection are actually an organic whole which is mutually influenced and mutually embodied, and if the registration and the change detection can be uniformly processed to form an integrated frame, the precision of the image registration and the change detection can be simultaneously improved.
The process of the invention is shown in figure 3, the feedback of change is added into the registration model, and the existence of the change information is fully considered; in turn, since the registration model is based on a gray scale process, the change resulting from the registration error can be used as a spectral change decision factor for subsequent change detection. Because the registration error variation only represents a change, points with large registration errors indicate that the change is highly likely, some changes may be true changes, some changes may be false changes due to occlusion and the like, and other surface feature changes may also be possible, so that other decision factors need to be added to decide the change of the building. The method adopts Multi-scale Maximum morphology reconstruction contour Building detection Index MMMPBI (Multi-scale Maximum morphology files Building detection Index) difference delta MMMPBI to make a decision, and finally adopts an evidence theory to fuse, thereby obtaining the final Building change information.
The invention comprises the following steps:
step 1, establishing a registration and change detection integrated processing variation model;
let the later image be f (x, y, t) and the earlier image be
Figure BDA0002030446140000061
Step 1.1, establishing data items in the integrated processing variation model
Figure BDA0002030446140000062
The vector expression is (1);
Figure BDA0002030446140000063
wherein
Figure BDA0002030446140000064
In the formula: u. of1，u4For scale parameters in the X and Y directions in affine transformations, u2，u3As a rotation parameter, u5，u6As a translation parameter, u7As a contrast parameter, u8Is brightDegree parameter, fx，fyAnd ftThe partial derivative of f (x, y, t) on space and time, w is the weight of the change information of each point participating in the operation, and the larger the change is, the smaller the matching probability of the corresponding point is, and the smaller the weight participating in the operation is; the smaller the variation, the greater the matching probability of the corresponding point, and the greater the weight involved in the calculation.
Step 1.2, establishing a smooth prior term in an integrated processing variation model
Figure BDA0002030446140000065
Figure BDA0002030446140000066
Step 1.3, establishing a linear constraint prior term in the integral processing variational model
Figure BDA0002030446140000071
Figure BDA0002030446140000072
Where Nl is the number of lines, l is the l-th line, Np is the number of points on line l, i is the i-th point on line l,
Figure BDA0002030446140000073
and
Figure BDA0002030446140000074
are the transformation parameters of the i-1, i and i +1 points on the straight line l.
Step 1.4, establishing a whole variation model;
Figure BDA0002030446140000075
step 2, establishing a probability model of the variation weight w in the variation model in the step 1;
the similarity measure adopted in the registration model is based on Mean Square Error (MSE) of gray scale, the determination of the weight value must be closely related to the measure, and meanwhile, the larger the registration Error of the point is, the smaller the weight corresponding to the operation is; the smaller the registration error is, the larger the weight corresponding to the operation is. Based on the above principle, the invention selects the following probability model:
Figure BDA0002030446140000076
wherein
d(x，y)＝f(x，y，t)-f(x，y，t-1) (33)；
I.e. d (x, y) represents the difference in gray levels between the registered image pairs, and the variables c and σ are constants.
Step 3, carrying out optimization solution on the registration and change detection integrated processing variation model in the step 1 by using a minimum variation energy function;
the Euler-Lagrange equation of the variation model in the step 1 is as follows:
Figure BDA0002030446140000077
in the formula:
Figure BDA0002030446140000078
is a Laplace operator, and
Figure BDA00020304461400000711
refers to two adjacent points on the same straight line, the left and the right of which take the point as the center
Figure BDA0002030446140000079
Average value of (a).
In order to prevent the ill-conditioned problem from occurring in the solving process, the formula (8) is subjected to regularization treatment, and a regularization term L is added, which becomes
Figure BDA00020304461400000710
In the formula: l is an 8X 8 diagonal element of lambdaiThe off-diagonal element is a diagonal matrix of 0, λ in this contextiAre all 1.0 × 1010。
The transformation parameters of each point are solved
Figure BDA00020304461400000811
Is a rather huge task, so that the traditional variational solution method is not directly adopted but a discrete form is adopted, and the discrete expression of the formula (9) is
Figure BDA0002030446140000081
In the formula:
Figure BDA0002030446140000082
is in calculating Laplace operator
Figure BDA0002030446140000083
So that the solution can be performed in a direct iterative manner, i.e. in such a way that the mean of adjacent points of (1) is calculated
Figure BDA0002030446140000084
In the formula:
Figure BDA0002030446140000085
and
Figure BDA0002030446140000086
is iterated by the current iter
Figure BDA0002030446140000087
To make an estimation, an initial estimation
Figure BDA0002030446140000088
In a smaller neighborhood using equation (12)And (6) estimating. And when the transformation model is recalculated, updating the weight w so that the point with large error has smaller weight in the next transformation model solution, and the point with small error has larger weight in the next transformation model.
Figure BDA0002030446140000089
Step 4, fusion processing of the change information and the building characteristics is carried out based on an evidence theory;
and based on the change information of the evidence theory and the building characteristic fusion processing, the weight of registration error reaction is used as a spectral change decision factor, and the difference of building detection indexes is used as a decision factor of building change, so that a fusion model of the evidence theory is established.
Step 4.1, establish the probability density function of the delta MMMPBI
Figure BDA00020304461400000810
In the formula: t isin，TdeIs the threshold value of positive and negative difference, and is processed by a Kuper entry threshold value method, and the variable tauin，τdeRespectively used for controlling the slope of positive and negative difference distribution, respectively using 1 and T for effectively controlling variablesinAnd 1 and | TdeThe difference of | represents; x is the value of the variable Δ mmpbi.
Step 4.2, establishing a probability density function of the registration error weight variable, and establishing a model by adopting 1-w, wherein the model is as follows:
Figure BDA0002030446140000091
in the formula: p1＝0.01；x1T is a threshold value obtained by a moment keeping threshold method (moment keeping threshold), the variable τ is used to control the slope of the distribution, and in order to effectively control the variable, the maximum values 1 and x of the 1-w variable are used1Is expressed by the difference of (1-x), τ ═ x1) 6; x is the value of the variable 1-w.
And 4.3, establishing a decision model for fusion processing of the change information and the building characteristics based on the evidence theory.
Table 1 lists decision models for classes in which the meaning of the variables is: building changes (B), other surface changes (S) and no changes (no change, N). And the delta MMMPBI is the difference of the detection index characteristics of the building, and 1-w is the spectral difference expressed by weight variation determined by the registration error. This decision model is used to distinguish between building and other terrain variations. Assume that Δ MMMPBI indicates a building change characteristic and 1-w indicates a characteristic of a general feature spectral change.
TABLE 1 probability Density of each class in Integrated building Change detection
Figure BDA0002030446140000092
Note: k is PΔMMMPBI·(1-P1-w)。
Furthermore, the above definitions of the various elements and methods are not limited to the particular structures, shapes or arrangements of parts mentioned in the examples, which may be easily modified or substituted by one of ordinary skill in the art, for example:
(1) 4.1, other characteristic factors for representing the change of the building can be adopted for the building characteristic difference;
(2) the invention is only described by taking the feature of the building as an example, and for other features such as vegetation, water bodies and the like, the feature change representing the features can be adopted to replace or change the feature change of the building in the step 4.
In order to verify the effectiveness of the invention, WorldView-2 data is selected for experiment, and as shown in FIG. 4, the data are respectively the early stage data shot in 2010 and the late stage data shot in 2012, and the dislocation condition before data registration. As can be seen from fig. 4, the pre-registration misalignment is quite severe, which leads to many spurious changes if the registration is not performed before the change detection process. The data includes the types of ground objects such as grasslands, trees, roads, houses and the like, the house has relatively large change area, and a plurality of newly-built buildings appear. Due to the fact that data resolution is high, the difference of the inclination directions of buildings caused by different shooting angles is more serious, and the difference brings great difficulty to registration and change detection for the buildings in white frames in fig. 4.
In order to verify the advantages of the integrated processing, the integrated processing result is respectively compared with a mode of separate processing of registration and change detection, namely, the registration is firstly carried out and then the change detection is carried out, wherein two registration methods are used for comparison, one is a polynomial registration method based on a correlation coefficient, and the other is a variation registration method driven by vision. The invention evaluates from two aspects of the registration result and the change detection result respectively.
1. Registration result comparison
Because buildings in the data change more, but grasslands and trees are mostly unchanged, the density is higher, and the check points are not very convenient to select, the registration result is compared, a visual judgment mode is adopted, and the registration result and the reference image are displayed in a superposition mode without quantitative evaluation. Fig. 5 is a comparison of registration results, where fig. 5(a) is a correlation coefficient polynomial registration result, fig. 5(b) is a visual driven variational registration result, and fig. 5(c) is a registration result of the present invention. It can be seen visually that the misalignment after the correlation coefficient polynomial method registration is obviously reduced compared with that before the registration, but a certain registration error still exists, as shown in a white box, the reason for this is mainly that an integral registration model (here, a polynomial is adopted) is adopted, and local deformation cannot be well taken into consideration. And a variation method based on visual driving can take these variations into account, such as white frame positions; however, the lower half of the image, such as a gray frame, has an obvious registration error, even a certain distortion, and the reason for this is that no change information is considered during registration, and the registration model is solved by using the changed pixels and the unchanged pixels with equal authority, thereby causing inaccuracy in model solution. The invention adopts an integrated processing thought, the weights of the solution points participating in the model are fully considered in the solution model, the weight of the point with large error is smaller in the solution, and the weight of the point with small error is larger in the solution, thereby improving the registration accuracy.
2. Comparison of Change detection results
The invention adopts the registration error as a spectral change decision factor to participate in the processing of change detection, and in order to fully verify the influence of the registration errors of different registration modes on the change detection, a relational number polynomial is respectively compared with the registration error weight value graph of the invention. The experimental results are shown in fig. 6, (a) is a correlation coefficient polynomial registration error weight graph, (b) is a registration error weight graph of a vision-driven variation registration method, and (c) is a registration error weight graph adopting the method. The darker the brightness is, the smaller the weight is, that is, the larger the display variation of the registration error is, the conclusion the same as the visual result of the registration error of the upper section can be obtained from fig. 6, for example, a building in a black frame, the registration error weight of the correlation coefficient polynomial method is smaller, and the registration error is larger. The registration error display changes are not necessarily true building change information, such as the area shown by the gray box in fig. 6, which represents the change of vegetation, so that another decision factor is used to determine the change of the building, and the difference based on the multi-scale maximum morphological contour building detection index is used for measurement.
Fig. 7 shows the change detection result, where fig. 7(a) is the reference data of the building change marked by manual interpretation, fig. 7(b) is the detection result of the correlation coefficient polynomial registration error as the spectral change factor, fig. 7(c) is the detection result of the vision-driven variation registration error as the spectral change factor, and fig. 7(d) is the detection result of the registration error of the present invention as the spectral change factor. The detection results showed changes except for a black background. As can be seen from fig. 7(b), the correlation coefficient polynomial based registration method has a large registration error, and if the registration error is large, the roof is misaligned, the luminance area corresponds to the darkness area, and the luminance is greatly different, which results in false detection. In fig. 7(c), the local change of the part of the image is too large, and the difference between the brightness and the contrast considered in the model is not enough to represent the changes, so that the solution of the model is not good, and the registration accuracy is affected, but because the deformation is simulated by adopting a local and global constraint mode, the whole model is not affected, but only the local part is affected, for example, the upper part is not affected, the detection result is better than the whole registration result, the change area in the large-area white frame does not appear, but the house detection result in the lower gray frame is greatly affected, and the false detection appears. It can be seen from fig. 7(d) that the detection result using the integration strategy can sufficiently overcome the above disadvantages, and obtain a satisfactory detection result. The method is mainly characterized in that the influence of the changed pixels is fully considered in the process of establishing the model, the change is used as an iterative feedback factor, so that the weight of the pixels participating in the model calculation is set, the result is optimized, meanwhile, the registration error is used as a change factor to carry out change detection, the problem of false detection caused by the fact that the local deformation cannot be considered in the traditional integral registration is solved, and the defect that a single variational model cannot control the model with large change is overcome.
The above-mentioned embodiments are intended to illustrate the objects, technical solutions and advantages of the present invention in further detail, and it should be understood that the above-mentioned embodiments are only exemplary embodiments of the present invention and are not intended to limit the present invention, and any modifications, equivalents, improvements and the like made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (5)

1. A registration and change detection integrated processing method based on a high-resolution remote sensing image is characterized by comprising the following steps:
step 1, establishing a registration and change detection integrated processing variational model of the remote sensing image, namely adding feedback of surface feature change into the registration model, fully considering the existence of change information, and simultaneously taking the change obtained by registration error as a spectral change decision factor to carry out subsequent change detection so as to realize the integrated processing of registration and change detection;
step 2, establishing a probability model of a change weight w in the registration and change detection integrated processing variation model;
step 3, carrying out optimization solution on the registration and change detection integrated processing variation model by using a minimum variation energy function;
step 4, fusion processing is carried out on the registration error change information obtained in the step 3 and feature change information of the ground object based on an evidence theory, so that the registration and change detection integrated processing of the high-resolution remote sensing image is realized;
the step 1 specifically comprises the following substeps:
let the later image be f (x, y, t) and the earlier image be
Figure FDA0002885221050000011
Step 1.1, establishing data items in the integrated processing variation model
Figure FDA0002885221050000012
The vector expression is (1)
Figure FDA0002885221050000013
Wherein
Figure FDA0002885221050000014
In the formula: u. of1，u4For scale parameters in the X and Y directions in affine transformations, u2，u3As a rotation parameter, u5，u6As a translation parameter, u7As a contrast parameter, u8As a brightness parameter, fx，fyAnd ftIs the partial derivative of f (x, y, t) to space and time, and w is the variation weight of each point participating in the operation;
step 1.2, establishing a smooth prior term in an integrated processing variation model
Figure FDA0002885221050000021
Figure FDA0002885221050000022
Step 1.3, establishing a linear constraint prior term in the integral processing variational model
Figure FDA0002885221050000023
Figure FDA0002885221050000024
Where Nl is the number of lines, l is the l-th line, Np is the number of points on line l, i is the i-th point on line l,
Figure FDA0002885221050000025
and
Figure FDA0002885221050000026
is the transformation parameter of the i-1, i and i +1 point on the straight line l;
step 1.4, establishing a whole variation model;
Figure FDA0002885221050000027
2. the method of claim 1, wherein the terrain comprises buildings, vegetation, and/or bodies of water.
3. The method according to claim 2, wherein the probability model in step 2 is as follows:
Figure FDA0002885221050000028
wherein
d(x，y)＝f(x，y，t)-f(x，y，t-1) (7)；
I.e. d (x, y) represents the difference in gray levels between the registered image pairs, and the variables c and σ are constants.
4. The method according to claim 2, wherein the Euler-Lagrangian equation of the variation model in step 3 is:
Figure FDA0002885221050000029
in the formula:
Figure FDA0002885221050000031
is a Laplace operator, and
Figure FDA0002885221050000032
refers to two adjacent points on the same straight line, the left and the right of which take the point as the center
Figure FDA0002885221050000033
Average value of (d);
regularization is performed on equation (8) during the solution process, adding a regularization term L, which becomes
Figure FDA0002885221050000034
In the formula: l is an 8X 8 diagonal element of lambdaiThe off-diagonal element is a diagonal matrix of 0, λiIs 1.0X 1010；
The discrete expression of formula (9) is
Figure FDA0002885221050000035
In the formula:
Figure FDA0002885221050000036
is in calculating Laplace operator
Figure FDA0002885221050000037
Is solved by direct iteration, i.e. the mean value of adjacent points is obtained
Figure FDA0002885221050000038
In the formula:
Figure FDA0002885221050000039
and
Figure FDA00028852210500000310
is iterated by the current iter
Figure FDA00028852210500000311
To make an estimation, an initial estimation
Figure FDA00028852210500000312
Estimating in a smaller neighborhood by using formula (12);
Figure FDA00028852210500000313
5. the method according to claim 2, characterized in that said step 4 comprises the following sub-steps:
step 4.1, establishing a probability density function of feature change information of the ground features;
step 4.2, establishing a probability density function of the registration error weight variable, and establishing a model by adopting 1-w, wherein the model is as follows:
Figure FDA00028852210500000314
in the formula: p1＝0.01；x1T is a threshold, obtained using a moment keeping threshold (moment keeping threshold), variable τ is used to control the slope of the distribution, maximum values 1 and x of the 1-w variable are used1Is expressed by the difference of (1-x), τ ═ x1) 6; x is the value of the variable 1-w;
and 4.3, establishing a decision model for fusion processing of the change information and the surface feature based on the evidence theory.
CN201910311014.8A 2019-04-17 2019-04-17 Registration and change detection integrated processing method based on high-resolution remote sensing image Active CN109903318B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910311014.8A CN109903318B (en) 2019-04-17 2019-04-17 Registration and change detection integrated processing method based on high-resolution remote sensing image

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910311014.8A CN109903318B (en) 2019-04-17 2019-04-17 Registration and change detection integrated processing method based on high-resolution remote sensing image

Publications (2)

Publication Number Publication Date
CN109903318A CN109903318A (en) 2019-06-18
CN109903318B true CN109903318B (en) 2021-05-11

Family

ID=66954142

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910311014.8A Active CN109903318B (en) 2019-04-17 2019-04-17 Registration and change detection integrated processing method based on high-resolution remote sensing image

Country Status (1)

Country Link
CN (1) CN109903318B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2761589A1 (en) * 2011-09-30 2014-08-06 Fraunhofer Gesellschaft zur Förderung der angewandten Wissenschaft E.V. Joint geometric and photometric multiview image registration
CN107665494A (en) * 2017-10-11 2018-02-06 青岛大学 A kind of adaptive noisy full variation dividing method of SAR image
CN107895377A (en) * 2017-11-15 2018-04-10 国光电器股份有限公司 A kind of foreground target extracting method, device, equipment and storage medium

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN100435167C (en) * 2005-12-23 2008-11-19 中国科学院对地观测与数字地球科学中心 High resolution SAR image registration processing method and system
US8250481B2 (en) * 2008-10-16 2012-08-21 The Curators Of The University Of Missouri Visualizing geographic-area change detected from high-resolution, remotely sensed imagery
CN102496154B (en) * 2011-10-24 2013-10-30 华中科技大学 Multi-temporal remote sensing image change detecting method based on Makov random field
CN103353989B (en) * 2013-06-18 2016-01-20 西安电子科技大学 Based on priori and the SAR image change detection merging gray scale and textural characteristics
CN105551031B (en) * 2015-12-10 2018-11-16 河海大学 Multi-temporal remote sensing image change detecting method based on FCM and evidence theory

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2761589A1 (en) * 2011-09-30 2014-08-06 Fraunhofer Gesellschaft zur Förderung der angewandten Wissenschaft E.V. Joint geometric and photometric multiview image registration
CN107665494A (en) * 2017-10-11 2018-02-06 青岛大学 A kind of adaptive noisy full variation dividing method of SAR image
CN107895377A (en) * 2017-11-15 2018-04-10 国光电器股份有限公司 A kind of foreground target extracting method, device, equipment and storage medium

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
"Improved Variational Image Registration Model and a Fast Algorithm for Its Numerical Approximation";Noppadol Chumchob et al.;《wileyonlinelibrary.com》;20111102;第1966-1995页 *
"一种面向变化检测的建筑物指数";王慧贤 等;《测绘通报》;20181231(第8期);第47-67页 *
"视觉驱动的变分配准方法";王慧贤 等;《测绘学报》;20150831;第44卷(第8期);第893-899页 *

Also Published As

Publication number Publication date
CN109903318A (en) 2019-06-18

Similar Documents

Publication Publication Date Title
CN109615653B (en) Leakage water area detection and identification method based on deep learning and visual field projection model
CN108364272A (en) A kind of high-performance Infrared-Visible fusion detection method
CN106991388B (en) Key point positioning method
CN110766058B (en) Battlefield target detection method based on optimized RPN (resilient packet network)
CN109523583B (en) Infrared and visible light image registration method for power equipment based on feedback mechanism
CN111242026B (en) Remote sensing image target detection method based on spatial hierarchy perception module and metric learning
CN107705290A (en) A kind of AMOLED display screens Mura defects detection method
CN116524062B (en) Diffusion model-based 2D human body posture estimation method
CN107871315B (en) Video image motion detection method and device
CN114578807A (en) Active target detection and obstacle avoidance method for unmanned target vehicle radar vision fusion
CN116229359A (en) Smoke identification method based on improved classical optical flow method model
CN115937518A (en) Pavement disease identification method and system based on multi-source image fusion
TWI383690B (en) Method for image processing
CN111429485A (en) Cross-modal filtering tracking method based on self-adaptive regularization and high-reliability updating
CN110659593A (en) Urban haze visibility detection method based on improved DiracNet
CN112801021B (en) Method and system for detecting lane line based on multi-level semantic information
CN109903318B (en) Registration and change detection integrated processing method based on high-resolution remote sensing image
US20160217556A1 (en) Systems and methods for automatic image enhancement utilizing feedback control
CN105844671B (en) A kind of fast background relief method under the conditions of change illumination
CN110717910B (en) CT image target detection method based on convolutional neural network and CT scanner
CN107657628A (en) A kind of real-time color method for tracking target
CN104008554B (en) Railway line small target detecting method and system based on video
CN115564805A (en) Moving target detection method
CN115830070A (en) Infrared laser fusion positioning method for inspection robot of traction substation
KR20170037804A (en) Robust visual odometry system and method to irregular illumination changes

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant