CN105989595A - Multi-temporal remote sensing image change detection method based on joint dictionary learning - Google Patents
Multi-temporal remote sensing image change detection method based on joint dictionary learning Download PDFInfo
- Publication number
- CN105989595A CN105989595A CN201510075774.5A CN201510075774A CN105989595A CN 105989595 A CN105989595 A CN 105989595A CN 201510075774 A CN201510075774 A CN 201510075774A CN 105989595 A CN105989595 A CN 105989595A
- Authority
- CN
- China
- Prior art keywords
- sample
- change
- remote sensing
- test set
- difference
- 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.)
- Granted
Links
Abstract
The invention discloses a multi-temporal remote sensing image change detection method based on joint dictionary learning. The method comprises steps: 1) a large number of unchanged samples are extracted from the multi-temporal remote sensing image, joint dictionary learning is carried out on the samples, and the base for the unchanged samples is obtained; 2) the remaining multi-temporal samples not extracted in the first step serve as testing samples, the base for the unchanged samples is used for carrying out sparse reconstruction on the testing samples, and difference between the testing samples and reconstructed testing samples is carried out to obtain a difference image; 3) a small number of changed samples are selected from the multi-temporal remote sensing image, the base for the unchanged samples is used for carrying out sparse reconstruction on different-temporal changed samples, and by using the difference between reconstructed images for different-temporal changed samples, a change threshold for the changed samples is obtained through pooling operation; and 4) the difference image and the change threshold for the changed samples are combined to judge the change area for the multi-temporal remote sensing image, and the detection rate is counted. Use of labeled samples can be greatly reduced, the change threshold does not need to be chosen manually, and the remote sensing image change detection rate can be improved.
Description
Technical field
The invention belongs to technical field of information processing, relate to a kind of multidate multispectral image change detecting method,
Particularly relate to a kind of Multitemporal Remote Sensing Images change detecting method based on associating dictionary learning.
Background technology
Since 20th century, the development of information technology and space technology profoundly changes the mankind and observes the side of the earth
Formula." ascending another storey to see a thousand miles further ", from the beginning of first artificial satellite lift-off, the mankind just start
The whole world is got a bird's eye view with a kind of unprecedented height.Along with the appearance of remote sensing technology, people can be more intuitively
Understand the change of earth every day.Wherein, make to obtain same due to the fast development of earth observation technology
The remote sensing image of district's difference phase is possibly realized.The remote sensing image utilizing multidate can be geographical national conditions detection
Thering is provided important technical guarantee and provide the real-time of detection, the change of remote sensing images carries out detection can be
The aspects such as environmental monitoring, Melting Glaciers, Disaster Assessment, urban sprawl, military target detection, Land_use change
Play an important role.
Detection method currently for Multitemporal Remote Sensing Images change is broadly divided into two classes:
One class is that the method is in the high score remote sensing of multidate based on unsupervised remote sensing image variation detection method
Image, target in hyperspectral remotely sensed image change-detection in be the most general and popular because the method not only calculates
Complexity ratio is relatively low, also can obtain preferable effect simultaneously.Such as, F.Bovolo et al. is at list of references " A
theoretical framework for unsupervised change detection based on change vector
analysis in the polar domain.IEEE Transactions on Geoscience and Remote Sensing,
45 (1): 218-236,2007. " a kind of change detecting method based on diverse vector method is proposed in, not only
From mathematical formulae, prove the feasibility of the method, give also detailed description simultaneously for concrete principle.
But the deficiency that the method exists is: directly ask in Multitemporal Remote Sensing Images enterprising row vector difference size computing
Solve, be easily subject to image difference noise and the interference of different remote sensing platform shooting angle, cause verification and measurement ratio low.
Another kind of is measure of supervision based on rear Classification Change Detection, and this method is by the remote sensing figure of multidate
As exercising supervision study, single image improves self terrain classification precision, thus improves final detection
Precision, this method of remote sensing expert professor Zhang Liangpei of Wuhan University is called " From-to " model.Such as,
B.Demir et al. is at list of references " Updating Land-Cover Maps by Classification of Image
Time Series:A Novel Change-Detection-Driven Transfer Learning Approach.IEEE
Transactions on Geoscience and Remote Sensing, 51 (1): 300-312,2013. " propose in
A kind of method based on transfer learning, utilizes transfer learning that multi-temporal remote sensing image is carried out terrain classification,
After realize high-precision change-detection.But the deficiency that the method exists is: side based on tradition supervised classification
What the method a large amount of experts of needs were wasted time and energy carries out atural object mark so that the method has certain limitation on promoting
Property；Simultaneously as the selection of change threshold depends on artificial selection or other clustering methods so that supervision
The development of change detecting method be restricted.
Summary of the invention
In order to solve above-mentioned technical problem present in background technology, the invention provides one and significantly reduce mark
The use of note sample, without artificial selection's change threshold and base that Remote Sensing Imagery Change Detection rate can be improved
Multitemporal Remote Sensing Images change detecting method in associating dictionary learning.
The technical solution of the present invention is:
The invention provides a kind of Multitemporal Remote Sensing Images change detecting method based on associating dictionary learning, its
Be characterized in that described Multitemporal Remote Sensing Images change detecting method based on associating dictionary learning include with
Lower step:
1) extract from Multitemporal Remote Sensing Images and do not change sample in a large number, in a large number do not change sample enter extract
Row associating dictionary learning, is not changed the base of sample；
2) by step 1) in remaining all multidate sample labeling of not choosing be test set sample, test set
Sample includes changing sample and not changing sample；Utilizing step 1) gained do not changes the base of sample to test set sample
Originally carry out sparse reconstruct, obtain reconstructing test set sample；Test set sample and reconstruct test set sample are asked poor,
Obtain Difference image；
3) choose from Multitemporal Remote Sensing Images and change sample on a small quantity；According to not changing the base of sample, to difference
The change sample of phase carries out sparse reconstruct, obtains the reconstruct image of different Temporal variation sample；Utilize difference
Difference between the reconstruct image of Temporal variation sample, operates through pondization, obtains changing the change threshold of sample
Value；
4) integrating step 2) gained Difference image and step 3) change threshold of gained change sample, it determines
Go out the region of variation of Multitemporal Remote Sensing Images, and statistic mixed-state rate.
Above-mentioned steps 1) specific implementation be:
1.1) by after Multitemporal Remote Sensing Images pretreatment, not changing in a large number in the difference phase of same place is chosen
Sample；
1.2) a large amount of do not change sample by what different phases were chosen and be stitched together, utilize the side of sparse expression
Method is not changed the base of sample, it may be assumed that
X1=D1S1
Wherein, X1For not changing sample；D1For not changing the dictionary of sample, i.e. do not change the base of sample；S1
For not changing the sparse expression coefficient of sample.
Above-mentioned steps 2) specific implementation be:
2.1), after setting up test set sample, utilize sparse expression mechanism that test set sample is decomposed into new dictionary
With new sparse expression coefficient, it may be assumed that
X2=D2S2
Wherein, X2For test set sample；D2For the dictionary of test set sample, the i.e. base of test set sample；S2
Sparse expression coefficient for test set sample；
2.2) utilize step 1.2) in gained do not change the base D of sample1Replacement step 2.1) in gained test set sample
This base D2, under the framework of sparse expression, reconstruct the test set sample of Multitemporal Remote Sensing Images, reconstructed
Test set sample X2′；
2.3) to test set sample X2With reconstruct test set sample X2' carry out seeking difference operation, obtain Difference image.
Above-mentioned steps 3) specific implementation be:
3.1) change the change sample set of each Mono temporal in sample on a small quantity for choose, be utilized respectively it
Sparse expression coefficient and step 1.2) in gained do not change the base D of sample1It is reconstructed, obtains phase transformation during difference
Change the reconstruct image of sample set；
3.2) for step 3.1) the reconstruct image of gained difference Temporal variation sample set carries out difference operation,
To change sample reconstructed error image on the base not changing sample；
3.3) by step 3.2) in gained reconstructed error image carry out pondization operation, obtain reconstructed error, i.e. become
Change the change threshold of sample.
Above-mentioned steps 4) specific implementation be:
4.1) by step 2.3) pixel value of gained Difference image and step 3.3) the change threshold of gained change sample
Value contrasts；If the pixel value of Difference image is more than or equal to the change threshold of change sample, then by this difference
The area marking that image is corresponding is region of variation；If the pixel value of Difference image is less than the change threshold of change sample
Value, then be non-region of variation by area marking corresponding for this Difference image；
4.2) according to step 4.1) determine the region of variation of Multitemporal Remote Sensing Images after, it determines average accurately
Rate is verification and measurement ratio, and the calculation of verification and measurement ratio is: differentiate that correct number of pixels accounts for the hundred of total number of pixels
Proportion by subtraction.
The invention have the advantage that
The invention provides a kind of Multitemporal Remote Sensing Images change detecting method based on associating dictionary learning, should
The substantial amounts of sample that do not changes is incorporated in sparse expression mechanism by detection method, and only make use of and change sample on a small quantity
This study change threshold, overcoming tradition measure of supervision needs the difficulty of a large amount of artificial mark, improves many
The change-detection precision of phase remote sensing images；Meanwhile, the change threshold changing sample in the present invention is self adaptation
Taking from experimental image middle school, overcoming traditional method needs artificial selection's change threshold or relies on other to calculate
Method carries out the deficiency of threshold learning；In result of the test, the adaptive threshold selection Policy Table that the present invention takes
Reveal preferable ability, it is possible to avoid different noise and the shooting angle impact on testing result, it is thus achieved that more
Good recognition result, thus provide preferably for aspects such as geographical national conditions detection, military surveillance and environmental monitorings
Technical support.
Accompanying drawing explanation
Fig. 1 is the Multitemporal Remote Sensing Images change detecting method based on associating dictionary learning that the present invention provides
Flow chart；
Fig. 2 a is the multispectral data figure of Kunshan Region, souths in 2000 of No. three satellite shootings of resource；
Fig. 2 b is the multispectral data figure of Kunshan Region, souths in 2003 of No. three satellite shootings of resource；
Fig. 3 a is the multispectral data figure of Taizhou Prefectures in 2000 of No. three satellite shootings of resource；
Fig. 3 b is the multispectral data figure of Taizhou Prefectures in 2003 of No. three satellite shootings of resource；
Fig. 4 a is the testing result utilizing Change vector Analysis method to change Kunshan Region, south Multitemporal Remote Sensing Images
Figure；
Fig. 4 b is the testing result utilizing principal component analytical method to change Kunshan Region, south Multitemporal Remote Sensing Images
Figure；
Fig. 4 c is the inspection that Kunshan Region, south Multitemporal Remote Sensing Images is changed by the multivariate detection algorithm utilizing iteration weight
Survey result figure；
Fig. 4 d is the detection utilizing semi-supervised notable detection algorithm to change Kunshan Region, south Multitemporal Remote Sensing Images
Result figure；
Fig. 4 e is the testing result utilizing the slow characteristics algorithm of supervision to change Kunshan Region, south Multitemporal Remote Sensing Images
Figure；
Fig. 4 f be utilize detection method (not comprising adaptive threshold selection strategy) to Kunshan Region, south many time
The testing result figure of phase Remote Sensing Imagery Change；
Fig. 4 g be utilize detection method (comprising adaptive threshold selection strategy) to Kunshan Region, south many time
The testing result figure of phase Remote Sensing Imagery Change；
Fig. 4 h is the real change area marking figure of Kunshan Region, south Multitemporal Remote Sensing Images；
Fig. 5 a is the testing result utilizing Change vector Analysis method to change Taizhou Prefecture's Multitemporal Remote Sensing Images
Figure；
Fig. 5 b is the testing result utilizing principal component analytical method to change Taizhou Prefecture's Multitemporal Remote Sensing Images
Figure；
Fig. 5 c is the inspection that Taizhou Prefecture's Multitemporal Remote Sensing Images is changed by the multivariate detection algorithm utilizing iteration weight
Survey result figure；
Fig. 5 d is the detection utilizing semi-supervised notable detection algorithm to change Taizhou Prefecture's Multitemporal Remote Sensing Images
Result figure；
Fig. 5 e is the testing result utilizing the slow characteristics algorithm of supervision to change Taizhou Prefecture's Multitemporal Remote Sensing Images
Figure；
Fig. 5 f be utilize detection method (not comprising adaptive threshold selection strategy) to Taizhou Prefecture many time
The testing result figure of phase Remote Sensing Imagery Change；
Fig. 5 g be utilize detection method (comprising adaptive threshold selection strategy) to Taizhou Prefecture many time
The testing result figure of phase Remote Sensing Imagery Change；
Fig. 5 h is the real change area marking figure of Taizhou Prefecture's Multitemporal Remote Sensing Images.
Detailed description of the invention
See Fig. 1, the invention provides a kind of Multitemporal Remote Sensing Images change-detection based on associating dictionary learning
Method, it comprises the following steps:
1) extract from Multitemporal Remote Sensing Images in a large number do not change sample (typically can be from Multitemporal Remote Sensing Images
Middle extraction 20%-50% does not changes sample), a large amount of do not change sample to extract and carry out associating dictionary learning,
Do not changed the base of sample；
1.1) by after Multitemporal Remote Sensing Images pretreatment, not changing in a large number in the difference phase of same place is chosen
Sample；
1.2) a large amount of do not change sample by what different phases were chosen and be stitched together, utilize the side of sparse expression
Method is not changed the base of sample, it may be assumed that
X1=D1S1
Wherein, X1For not changing sample；D1For not changing the dictionary of sample, i.e. do not change the base of sample；S1
For not changing the sparse expression coefficient of sample.
In the present invention, associating dictionary learning is to splice the generic sample in different phase remote sensing images one
Rise, as a new associating sample, this associating sample is carried out sparse study and asks for the word of this associating sample
Allusion quotation, the dictionary asking for this associating sample is referred to as associating dictionary.Choose in this step is different phase remote sensing
Image do not change sample, thus try to achieve the associating dictionary not changing sample, i.e. do not change the base of sample.
2) by step 1) in remaining all multidate sample labeling of not choosing be test set sample, test set
Sample includes changing sample and not changing sample；Utilizing step 1) gained do not changes the base of sample to test set sample
Originally carry out sparse reconstruct, obtain reconstructing test set sample；Test set sample and reconstruct test set sample are asked poor,
Obtain Difference image；
2.1), after setting up test set sample, utilize sparse expression mechanism that test set sample is decomposed into new dictionary
With new sparse expression coefficient, it may be assumed that
X2=D2S2
Wherein, X2For test set sample；D2For the dictionary of test set sample, the i.e. base of test set sample；S2
Sparse expression coefficient for test set sample；
2.2) utilize step 1.2) in gained do not change the base D of sample1Replacement step 2.1) in gained test set sample
This base D2, under the framework of sparse expression, reconstruct the test set sample of Multitemporal Remote Sensing Images, reconstructed
Test set sample X2′；
2.3) to test set sample X2With reconstruct test set sample X2' carry out seeking difference operation, obtain Difference image.
3) choose from Multitemporal Remote Sensing Images change sample on a small quantity (typically can be from Multitemporal Remote Sensing Images
Choose the change sample of 2%-10%)；According to not changing the base of sample, the change sample of different phases is carried out
Sparse reconstruct, obtains the reconstruct image of different Temporal variation sample；Utilize the reconstruct of different Temporal variation sample
Difference between image, operates through pondization, obtains changing the change threshold of sample；
3.1) change the change sample set of each Mono temporal in sample on a small quantity for choose, be utilized respectively it
Sparse expression coefficient and step 1.2) in gained do not change the base D of sample1It is reconstructed, obtains phase transformation during difference
Change the reconstruct image of sample set；
3.2) for step 3.1) the reconstruct image of gained difference Temporal variation sample set carries out difference operation,
To change sample reconstructed error image on the base not changing sample；
3.3) by step 3.2) in gained reconstructed error image carry out pondization operation, obtain reconstructed error, i.e. become
Change the change threshold of sample.Owing to this change threshold is that self adaptation takes from experimental image middle school, can be preferable
The Traditional Man that overcomes choose change threshold or depend on the deficiency of other ripe algorithms, and in result of the test
On, the adaptive threshold selection Policy Table of the present invention reveals preferable ability.
4) integrating step 2) gained Difference image and step 3) change threshold of gained change sample, it determines
Go out the region of variation of Multitemporal Remote Sensing Images, and statistic mixed-state rate；
4.1) by step 2.3) pixel value of gained Difference image and step 3.3) the change threshold of gained change sample
Value contrasts；If the pixel value of Difference image is more than or equal to the change threshold of change sample, then by this difference
The area marking that image is corresponding is region of variation；If the pixel value of Difference image is less than the change threshold of change sample
Value, then be non-region of variation by area marking corresponding for this Difference image；
4.2) according to step 4.1) determine the region of variation of Multitemporal Remote Sensing Images after, it determines average accurately
Rate is verification and measurement ratio, and the calculation of verification and measurement ratio is: differentiate that correct number of pixels accounts for the hundred of total number of pixels
Proportion by subtraction.
Below with emulation experiment, the multi-temporal remote sensing figure based on associating dictionary learning that the present invention provides is described
Beneficial effect as change detecting method:
1) simulated conditions
Central processing unit be Intel (R) Core i3-530 2.93GHZ, internal memory 4G, WINDOWS7 operation
MATLAB software is used to emulate in system；The test image used in experiment is that No. three satellites of resource are clapped
The Kunshan Region, south taken the photograph and the multispectral data (seeing Fig. 2 a, Fig. 2 b, Fig. 3 a and Fig. 3 b) of Taizhou Prefecture.
2) emulation content
The detection method using the present invention to provide is tested:
First, Kunshan Region, south and two, Taizhou Prefecture data base are selected respectively and do not change zone sample in a large number
To with a small amount of region of variation sample pair, and using remaining all samples to as test set sample；
Secondly, by the method for sparse expression respectively do not change in a large number zone sample on learn unchangedization sample
This base, utilizes the basic weight structure test set sample not changing sample, and obtains test set sample and reconstruct survey
Differential image between examination collection sample；
Then, a small amount of change sample learns adaptive change threshold value, utilizes change threshold at differential image
On differentiate.
Below, detection method experimental results will be used and use the experiment of traditional detection method gained
Result compares, wherein:
Fig. 4 a and Fig. 5 a is that (see reference employing Change vector Analysis method document: A theoretical framework
for unsupervised change detection based on change vector analysis in the polar
Domain.IEEE Trans.On Geoscience and Remote Sensing, 45 (1), 218-236,2007.) point
Other to Kunshan Region, south with the result of the test of Taizhou Prefecture.It appeared that: Kunshan data base and Taizhou data base
On, the Detection results using Change vector Analysis method is undesirable, and misclassification region is more, and noise spot is the most very
Many.This is because for single-range remote sensing images, Change vector Analysis method can not well extract difference
Different information.
Fig. 4 b and Fig. 5 b is that (see reference employing principal component analytical method document: Unsupervised change
Detection with kernels.IEEE Geosci.Remote Sens.Lett, 9 (6): 1026-1030,2012.) point
Other to Kunshan Region, south with the result of the test of Taizhou Prefecture.It appeared that: Kunshan data base and Taizhou data base
On, use principal component analytical method all to fail to obtain good result of the test, its noise spot is more, detects image
Unsmooth.This is to distinguish region of variation and non-region of variation well owing to failing during Principle component extraction,
Illustrate that principal component analytical method is not ideally suited for the detection research of Remote Sensing Imagery Change.
Fig. 4 c and Fig. 5 c be use iteration weight multivariate detection algorithm (see reference document: The regularized
iteratively reweighted mad method for change detection in multi-and hyper spectral
Data.IEEE Trans.on Image Process, 16 (2): 463-478,2007.) respectively to Kunshan Region, south and Taizhou
The result of the test in area.It appeared that: on Kunshan data base and Taizhou data base, the detection knot of the method
The detection method that fruit does not all have the present invention to provide is good, and its Detection accuracy is the highest, and especially misclassification region is more.
This is because the multivariate detection algorithm of iteration weight is when distinguishing similar atural object, probability of miscarriage of justice is bigger.
Fig. 4 d and Fig. 5 d is that (see reference the semi-supervised notable detection algorithm of employing document: Semi-supervised
novelty detection using svm entire solution path.IEEE Trans.On Geoscience and
Remote Sensing, 51 (4-1): 1939-1950,2013.) respectively the test of Kunshan Region, south and Taizhou Prefecture is tied
Really.It appeared that: on Kunshan data base and Taizhou data base, the performance of the method is not good enough, and verification and measurement ratio is relatively
Low, probability of false detection is relatively big, and detection image is unsmooth simultaneously.This is owing to the method depends critically upon required
Feature, and it is more laborious that different images is chosen different characteristic, thus the method does not have pervasive
Property.
Fig. 4 e and Fig. 5 e be use supervision slow characteristics algorithm (see reference document: Slow feature analysis for
change detection in multispectral imagery,IEEE Trans.On Geoscience and Remote
Sensing, 52 (5): 2858-2874,2014.) respectively to Kunshan Region, south and the result of the test of Taizhou Prefecture.Permissible
Find: on Kunshan data base and Taizhou data base, the method all shows well, and verification and measurement ratio is higher, image
There is less region misclassification；Although its result of the test noise spot is few, but for changing unconspicuous region, should
Method can not well express change information, and therefore the method haves much room for improvement at detection aspect of performance.
Fig. 4 f and Fig. 5 f be use the present invention provide detection method (do not use adaptive threshold selection strategy, and
Select traditional clustering algorithm) respectively to Kunshan Region, south and the result of the test of Taizhou Prefecture.It appeared that: elder brother
On mountain data base and Taizhou data base, detection method (do not use adaptive threshold selection strategy, and
Select traditional clustering algorithm) testing result similar with the testing result of the slow characteristics algorithm of supervision.Therefore,
There is no step 3) in the case of (that is, not using adaptive threshold selection strategy), the associating word of the present invention
Allusion quotation learning method has preferably performance on extraction change information.
Fig. 4 g and Fig. 5 g is that the detection method using the present invention to provide (comprises the step of adaptive threshold selection strategy
Suddenly) respectively to Kunshan Region, south and the result of the test of Taizhou Prefecture.It appeared that: in Kunshan data base and Taizhou
On data base, the result of the test of this detection method is best, and its classification accuracy is higher, and noise spot is less；
Also illustrate simultaneously, utilize step 3) study to adaptive change threshold value can preferably distinguish region of variation and
Non-region of variation.
Fig. 4 h and Fig. 5 h is the real change region of the Multitemporal Remote Sensing Images of Kunshan Region, south and Taizhou Prefecture respectively
Mark figure.
Finally, result of the test and the true standard (that is, Fig. 4 h and Fig. 5 h) of different detection methods are compared
Right, add up comparison result, as the accuracy in detection to Multitemporal Remote Sensing Images change-detection, result such as table 1
Shown in.
As it can be seen from table 1 the Multitemporal Remote Sensing Images change inspection based on associating dictionary learning that the present invention provides
Survey method has preferably performance, and the verification and measurement ratio of the present invention is higher than the verification and measurement ratio of existing direct detecting method.This
It is owing to the present invention has taken into full account a large amount of information not changing sample, thus overcomes in traditional method in a large number
Do not change the drawback that sample is under-utilized；And the selection strategy that the present invention is by adaptive threshold, can be fine
From image self-information, obtain being suitable for the change threshold of image self, it is possible to avoid different noise and shooting
The angle impact on testing result, thus obtain more preferable recognition result, demonstrate the present invention further and carry
The advance of the Multitemporal Remote Sensing Images change detecting method based on associating dictionary learning of confession such that it is able to for
The aspects such as geographical national conditions detection, military surveillance and environmental monitoring provide superior technique support.
The verification and measurement ratio of the different detection method of table 1
Claims (5)
1. a Multitemporal Remote Sensing Images change detecting method based on associating dictionary learning, it is characterised in that:
Described Multitemporal Remote Sensing Images change detecting method based on associating dictionary learning comprises the following steps:
1) extract from Multitemporal Remote Sensing Images and do not change sample in a large number, in a large number do not change sample enter extract
Row associating dictionary learning, is not changed the base of sample；
2) by step 1) in remaining all multidate sample labeling of not choosing be test set sample, test set
Sample includes changing sample and not changing sample；Utilizing step 1) gained do not changes the base of sample to test set sample
Originally carry out sparse reconstruct, obtain reconstructing test set sample；Test set sample and reconstruct test set sample are asked poor,
Obtain Difference image；
3) choose from Multitemporal Remote Sensing Images and change sample on a small quantity；According to not changing the base of sample, to difference
The change sample of phase carries out sparse reconstruct, obtains the reconstruct image of different Temporal variation sample；Utilize difference
Difference between the reconstruct image of Temporal variation sample, operates through pondization, obtains changing the change threshold of sample
Value；
4) integrating step 2) gained Difference image and step 3) change threshold of gained change sample, it determines
Go out the region of variation of Multitemporal Remote Sensing Images, and statistic mixed-state rate.
Multitemporal Remote Sensing Images change-detection side based on associating dictionary learning the most according to claim 1
Method, it is characterised in that: described step 1) specific implementation be:
1.1) by after Multitemporal Remote Sensing Images pretreatment, not changing in a large number in the difference phase of same place is chosen
Sample；
1.2) a large amount of do not change sample by what different phases were chosen and be stitched together, utilize the side of sparse expression
Method is not changed the base of sample, it may be assumed that
X1=D1S1
Wherein, X1For not changing sample；D1For not changing the dictionary of sample, i.e. do not change the base of sample；S1
For not changing the sparse expression coefficient of sample.
Multitemporal Remote Sensing Images change-detection side based on associating dictionary learning the most according to claim 2
Method, it is characterised in that: described step 2) specific implementation be:
2.1), after setting up test set sample, utilize sparse expression mechanism that test set sample is decomposed into new dictionary
With new sparse expression coefficient, it may be assumed that
X2=D2S2
Wherein, X2For test set sample；D2For the dictionary of test set sample, the i.e. base of test set sample；S2
Sparse expression coefficient for test set sample；
2.2) utilize step 1.2) in gained do not change the base D of sample1Replacement step 2.1) in gained test set sample
This base D2, under the framework of sparse expression, reconstruct the test set sample of Multitemporal Remote Sensing Images, reconstructed
Test set sample X2′；
2.3) to test set sample X2With reconstruct test set sample X2' carry out seeking difference operation, obtain Difference image.
Multitemporal Remote Sensing Images change-detection side based on associating dictionary learning the most according to claim 3
Method, it is characterised in that: described step 3) specific implementation be:
3.1) change the change sample set of each Mono temporal in sample on a small quantity for choose, be utilized respectively it
Sparse expression coefficient and step 1.2) in gained do not change the base D of sample1It is reconstructed, obtains phase transformation during difference
Change the reconstruct image of sample set；
3.2) for step 3.1) the reconstruct image of gained difference Temporal variation sample set carries out difference operation,
To change sample reconstructed error image on the base not changing sample；
3.3) by step 3.2) in gained reconstructed error image carry out pondization operation, obtain reconstructed error, i.e. become
Change the change threshold of sample.
Multitemporal Remote Sensing Images change-detection side based on associating dictionary learning the most according to claim 4
Method, it is characterised in that: described step 4) specific implementation be:
4.1) by step 2.3) pixel value of gained Difference image and step 3.3) the change threshold of gained change sample
Value contrasts；If the pixel value of Difference image is more than or equal to the change threshold of change sample, then by this difference
The area marking that image is corresponding is region of variation；If the pixel value of Difference image is less than the change threshold of change sample
Value, then be non-region of variation by area marking corresponding for this Difference image；
4.2) according to step 4.1) determine the region of variation of Multitemporal Remote Sensing Images after, it determines average accurately
Rate is verification and measurement ratio, and the calculation of verification and measurement ratio is: differentiate that correct number of pixels accounts for the hundred of total number of pixels
Proportion by subtraction.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510075774.5A CN105989595B (en) | 2015-02-12 | 2015-02-12 | Multi-temporal remote sensing image change detection method based on joint dictionary learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510075774.5A CN105989595B (en) | 2015-02-12 | 2015-02-12 | Multi-temporal remote sensing image change detection method based on joint dictionary learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105989595A true CN105989595A (en) | 2016-10-05 |
| CN105989595B CN105989595B (en) | 2020-10-30 |
Family
ID=57042266
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510075774.5A Active CN105989595B (en) | 2015-02-12 | 2015-02-12 | Multi-temporal remote sensing image change detection method based on joint dictionary learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105989595B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111797737A (en) * | 2020-06-22 | 2020-10-20 | 重庆高新区飞马创新研究院 | Remote sensing target detection method and device |
| CN113011321A (en) * | 2021-03-17 | 2021-06-22 | 中南大学 | Spectral signal denoising method, system, terminal and readable storage medium based on joint dictionary |
-
2015
- 2015-02-12 CN CN201510075774.5A patent/CN105989595B/en active Active
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111797737A (en) * | 2020-06-22 | 2020-10-20 | 重庆高新区飞马创新研究院 | Remote sensing target detection method and device |
| CN113011321A (en) * | 2021-03-17 | 2021-06-22 | 中南大学 | Spectral signal denoising method, system, terminal and readable storage medium based on joint dictionary |
| CN113011321B (en) * | 2021-03-17 | 2022-05-06 | 中南大学 | Spectral signal denoising method, system, terminal and readable storage medium based on joint dictionary |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105989595B (en) | 2020-10-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liu et al. | Scene classification based on multiscale convolutional neural network | |
| CN110490802B (en) | Super-resolution-based satellite image airplane target model identification method | |
| Xia et al. | AID: A benchmark data set for performance evaluation of aerial scene classification | |
| Zhao et al. | Incorporating metric learning and adversarial network for seasonal invariant change detection | |
| Liu et al. | ARC-Net: An efficient network for building extraction from high-resolution aerial images | |
| US9230168B2 (en) | Automatic generation of built-up layers from high resolution satellite image data | |
| Du et al. | A novel binary tree support vector machine for hyperspectral remote sensing image classification | |
| CN114842365B (en) | Unmanned aerial vehicle aerial photography target detection and identification method and system | |
| Yue et al. | Optical remote sensing image understanding with weak supervision: Concepts, methods, and perspectives | |
| Li et al. | A review on deep learning techniques for cloud detection methodologies and challenges | |
| Zhang et al. | Automatically monitoring impervious surfaces using spectral generalization and time series Landsat imagery from 1985 to 2020 in the Yangtze River Delta | |
| Yao et al. | Sensing urban land-use patterns by integrating Google Tensorflow and scene-classification models | |
| CN110110610A (en) | A kind of event detecting method for short-sighted frequency | |
| CN107203779A (en) | The EO-1 hyperion dimension reduction method kept based on empty spectrum information | |
| Codella et al. | Towards large scale land-cover recognition of satellite images | |
| Sathyanarayanan et al. | A multiclass deep learning approach for LULC classification of multispectral satellite images | |
| CN105989595A (en) | Multi-temporal remote sensing image change detection method based on joint dictionary learning | |
| Li et al. | Landslide detection based on shipborne images and deep learning models: a case study in the Three Gorges Reservoir Area in China | |
| Song et al. | Automatic CRP mapping using nonparametric machine learning approaches | |
| Wang et al. | Adaptive scale selection in multiscale segmentation based on the segmented object complexity of GF-2 satellite image | |
| Srivastava et al. | Feature-Based Image Retrieval (FBIR) system for satellite image quality assessment using big data analytical technique | |
| Jiang et al. | IARet: A lightweight multiscale infrared aerocraft recognition algorithm | |
| CN109753896A (en) | A kind of unsupervised heterologous method for detecting change of remote sensing image based on general character autocoder | |
| Lv et al. | Multi-scale attentive region adaptive aggregation learning for remote sensing scene classification | |
| Zhang et al. | Transmission Tower Detection Algorithm Based on Feature-Enhanced Convolutional Network in Remote Sensing Image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | 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 |