CN105989595B - 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
- CN105989595B CN105989595B CN201510075774.5A CN201510075774A CN105989595B CN 105989595 B CN105989595 B CN 105989595B CN 201510075774 A CN201510075774 A CN 201510075774A CN 105989595 B CN105989595 B CN 105989595B
- Authority
- CN
- China
- Prior art keywords
- samples
- change
- sample
- unchanged
- remote sensing
- 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
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 101
- 238000012360 testing method Methods 0.000 claims abstract description 70
- 238000000034 method Methods 0.000 claims abstract description 30
- 238000011176 pooling Methods 0.000 claims abstract description 7
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 18
- 238000004422 calculation algorithm Methods 0.000 description 15
- 230000003044 adaptive effect Effects 0.000 description 10
- 238000004458 analytical method Methods 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 4
- 238000012847 principal component analysis method Methods 0.000 description 4
- 230000007547 defect Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 238000002372 labelling Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 238000013526 transfer learning Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000002844 melting Methods 0.000 description 1
- 230000008018 melting Effects 0.000 description 1
- 238000000513 principal component analysis Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
A multi-temporal remote sensing image change detection method based on joint dictionary learning comprises the following steps: 1) extracting a large number of unchanged samples from the multi-temporal remote sensing image, and performing joint dictionary learning on the samples to obtain the basis of the unchanged samples; 2) taking the rest multi-temporal samples which are not selected in the step 1) as test set samples; carrying out sparse reconstruction on the samples in the test set by using the bases of the unchanged samples; obtaining a difference image by subtracting the test set sample and the reconstructed test set sample; 3) selecting a small amount of change samples from the multi-temporal remote sensing image; sparse reconstructing the changed samples of different time phases by using the basis of the unchanged samples; obtaining a change threshold value of the change sample through pooling operation by using a difference value between reconstructed images of different phase change samples; 4) and (4) judging a change area of the multi-temporal remote sensing image by combining the difference image and a change threshold of the change sample, and counting the detection rate. The method can greatly reduce the use of the marked sample, does not need to manually select the change threshold value, and can improve the detection rate of the change of the remote sensing image.
Description
Technical Field
The invention belongs to the technical field of information processing, relates to a multi-temporal multispectral image change detection method, and particularly relates to a multi-temporal remote sensing image change detection method based on joint dictionary learning.
Background
Since the 20 th century, the development of information technology and space technology has profoundly changed the way in which humans observed the earth. "want to be in the order of thousands of miles, one storey above", from the first satellite to the sky, humans begin to overlook the atlantoan at an unprecedented height. With the advent of remote sensing technology, people can more intuitively understand the change of the earth every day. The rapid development of the earth observation technology enables the remote sensing images of different time phases in the same region to be acquired. The remote sensing images in multiple time phases can provide important technical support for detection of geographical national conditions and real-time detection, and detection of changes of the remote sensing images can play an important role in aspects of environment monitoring, glacier melting, disaster assessment, city expansion, military target detection, land utilization and the like.
At present, detection methods for multi-temporal remote sensing image changes are mainly classified into two types:
one type is an unsupervised remote sensing image change detection method, which is most common and popular in the change detection of multi-temporal high-resolution remote sensing images and hyperspectral remote sensing images, because the method not only has lower calculation complexity, but also can obtain better effect. For example, Bovolo et al, in the reference "A the organic frame for the underlying change detection based on change vector analysis in the polar domain. IEEE Transactions on Geoscience and Remote Sensing,45(1):218, 236, 2007", propose a change detection method based on a change vector method, which not only demonstrates the feasibility of the method from mathematical formulas, but also gives a detailed description of the specific principles. However, the method has the following disadvantages: the vector difference value is directly calculated and solved on the multi-temporal remote sensing image, and the interference of different noises of the image and different shooting angles of remote sensing platforms is easily caused, so that the detection rate is low.
The other kind is a supervision method based on post-classification change detection, the method is to supervise and learn remote sensing images of multiple time phases and improve the classification precision of the ground objects on a single image, thereby improving the final detection precision, and the method is called as a 'From-to' model by the professor of the remote sensing experts of the university of Wuhan. For example, B.Demir et al, in the reference "update land-Cover Maps by Classification of Image Time Series: A Novel Change-Detection-drive Transfer Learning approach, IEEE Transactions on geoscience and remove Sensing,51(1):300-312, 2013", propose a method based on Transfer Learning, which uses Transfer Learning to classify the ground features of multi-temporal Remote Sensing images, and finally realize high-precision Change Detection. However, the method has the following disadvantages: the method based on the traditional supervision and classification needs a large amount of experts to label the ground features in a time-consuming and labor-consuming manner, so that the method has certain limitation in popularization; meanwhile, because the selection of the change threshold depends on manual selection or other clustering methods, the development of supervised change detection methods is limited.
Disclosure of Invention
In order to solve the technical problems in the background art, the invention provides the joint dictionary learning-based multi-temporal remote sensing image change detection method which greatly reduces the use of the marked samples, does not need to manually select the change threshold value and can improve the remote sensing image change detection rate.
The technical solution of the invention is as follows:
the invention provides a multi-temporal remote sensing image change detection method based on joint dictionary learning, which is characterized by comprising the following steps: the multi-temporal remote sensing image change detection method based on the joint dictionary learning comprises the following steps:
1) extracting a large number of unchanged samples from the multi-temporal remote sensing image, and performing joint dictionary learning on the extracted large number of unchanged samples to obtain the basis of the unchanged samples;
2) marking all other multi-temporal samples which are not selected in the step 1) as test set samples, wherein the test set samples comprise changed samples and unchanged samples; carrying out sparse reconstruction on the test set sample by using the basis of the unchanged sample obtained in the step 1) to obtain a reconstructed test set sample; the difference between the test set sample and the reconstructed test set sample is obtained to obtain a difference image;
3) selecting a small amount of change samples from the multi-temporal remote sensing image; performing sparse reconstruction on the changed samples in different time phases according to the basis of the unchanged samples to obtain reconstructed images of the changed samples in different time phases; obtaining a change threshold value of the change sample by using a difference value between reconstructed images of different phase change samples through pooling operation;
4) and (3) judging a change area of the multi-temporal remote sensing image by combining the difference image obtained in the step 2) and the change threshold of the change sample obtained in the step 3), and counting the detection rate.
The specific implementation manner of the step 1) is as follows:
1.1) preprocessing the multi-temporal remote sensing image, and selecting a large number of unchanged samples in different phases at the same place;
1.2) splicing a large number of unchanged samples selected from different phases together, and obtaining the basis of the unchanged samples by using a sparse expression method, namely:
X1＝D1S1
wherein, X1An unchanged sample; d1A dictionary of unchanged samples, i.e. the basis of unchanged samples; s1Is notAnd changing sparse expression coefficients of the sample.
The specific implementation manner of the step 2) is as follows:
2.1) after the test set sample is established, decomposing the test set sample into a new dictionary and a new sparse expression coefficient by using a sparse expression mechanism, namely:
X2＝D2S2
wherein, X2Is a test set sample; d2A dictionary of test set samples, i.e., the bases of the test set samples; s2Sparse expression coefficients of the test set samples are obtained;
2.2) Using the base D of the unchanged sample obtained in step 1.2)1Base D of the test set sample obtained in step 2.1) is replaced2Reconstructing a test set sample of the multi-temporal remote sensing image under a sparse expression frame to obtain a reconstructed test set sample X2′；
2.3) for test set sample X2And reconstructing the test set sample X2' carry out difference operation to obtain difference image.
The specific implementation manner of the step 3) is as follows:
3.1) for each single-time phase change sample set in the selected small number of change samples, respectively utilizing the sparse expression coefficient and the base D of the unchanged sample obtained in the step 1.2)1Reconstructing to obtain a reconstructed image of the sample set with different phase changes;
3.2) performing difference operation on the reconstructed images of the different phase change sample sets obtained in the step 3.1) to obtain a reconstructed error image of the changed samples on the basis of the unchanged samples;
3.3) performing pooling operation on the reconstruction error image obtained in the step 3.2) to obtain a reconstruction error, namely a change threshold of the change sample.
The specific implementation manner of the step 4) is as follows:
4.1) comparing the pixel value of the difference image obtained in the step 2.3) with the change threshold value of the change sample obtained in the step 3.3); if the pixel value of the difference image is larger than or equal to the change threshold of the change sample, marking the area corresponding to the difference image as a change area; if the pixel value of the difference image is smaller than the change threshold of the change sample, marking the area corresponding to the difference image as an unchanged area;
4.2) judging the change area of the multi-temporal remote sensing image according to the step 4.1), wherein the average judging accuracy is the detection rate, and the calculation mode of the detection rate is as follows: and judging the percentage of the number of the correct pixels to the total number of the pixels.
The invention has the advantages that:
the invention provides a multi-temporal remote sensing image change detection method based on joint dictionary learning, which introduces a large number of unchanged samples into a sparse expression mechanism and only utilizes a small number of changed samples to learn a change threshold, thereby overcoming the difficulty that a traditional supervision method needs a large number of manual labels and improving the change detection precision of multi-temporal remote sensing images; meanwhile, the change threshold of the change sample is acquired from the experimental image in a self-adaptive manner, so that the defect that the change threshold needs to be manually selected or other algorithms are used for threshold learning in the traditional method is overcome; on the test result, the adaptive threshold selection strategy adopted by the invention has better capability, can avoid the influence of different noises and shooting angles on the detection result, and obtains a better identification result, thereby providing better technical support for the aspects of geographical national condition detection, military reconnaissance, environmental monitoring and the like.
Drawings
FIG. 1 is a flow chart of a multi-temporal remote sensing image change detection method based on joint dictionary learning according to the present invention;
FIG. 2a is a 2000 year Queenshan region multispectral data map taken with a resource number three satellite;
FIG. 2b is a diagram of multispectral data taken from a satellite number three resource in the Kunshan region of 2003;
FIG. 3a is a graph of multispectral data from a 2000 year Taizhou region taken from resource number three satellite;
FIG. 3b is a graph of multispectral data from the Taizhou region 2003 taken from satellite number three of the asset;
FIG. 4a is a diagram illustrating a detection result of multi-temporal remote sensing image change in a Kun-shan region by using a change vector analysis method;
FIG. 4b is a diagram illustrating the detection result of the multi-temporal remote sensing image change in the Kun-shan region by using a principal component analysis method;
FIG. 4c is a diagram of a detection result of a multi-temporal remote sensing image change in the Kun-shan region by using a multivariate detection algorithm of iterative weights;
FIG. 4d is a diagram of a detection result of multi-temporal remote sensing image changes in the Kun-shan region by using a semi-supervised saliency detection algorithm;
FIG. 4e is a diagram of a detection result of multi-temporal remote sensing image changes in the Kun-shan region by using a supervised slow feature algorithm;
FIG. 4f is a diagram showing the detection result of the multi-temporal remote sensing image change in the Kunzan region by using the detection method (without the adaptive threshold selection strategy) of the present invention;
FIG. 4g is a diagram showing the detection result of the multi-temporal remote sensing image change in the Kun-shan region by using the detection method (including the adaptive threshold selection strategy) of the present invention;
FIG. 4h is a real change region labeling diagram of a multi-temporal remote sensing image in the Kun mountain area;
FIG. 5a is a diagram showing the detection result of the change of the multi-temporal remote sensing image in the Taizhou region by using the change vector analysis method;
FIG. 5b is a diagram showing the detection result of the change of the multi-temporal remote sensing image in the Taizhou region by using the principal component analysis method;
FIG. 5c is a diagram showing the results of the multi-temporal remote sensing image change in the Taizhou region using the iterative weighted multi-element detection algorithm;
FIG. 5d is a diagram of the results of a multi-temporal remote sensing image change in the Taizhou region using a semi-supervised saliency detection algorithm;
FIG. 5e is a diagram of the detection result of the change of the multi-temporal remote sensing image in the Taizhou region by using the supervised slow feature algorithm;
FIG. 5f is a diagram showing the results of the detection of the multi-temporal remote sensing image changes in the Taizhou region using the detection method of the present invention (without the adaptive threshold selection strategy);
FIG. 5g is a diagram showing the detection result of the multi-temporal remote sensing image change in the Taizhou region by using the detection method (including the adaptive threshold selection strategy) of the present invention;
FIG. 5h is a true change region labeling diagram of the multi-temporal remote sensing image in the Taizhou region.
Detailed Description
Referring to fig. 1, the invention provides a multi-temporal remote sensing image change detection method based on joint dictionary learning, which comprises the following steps:
1) extracting a large number of unchanged samples from the multi-temporal remote sensing image (generally, 20% -50% of unchanged samples can be extracted from the multi-temporal remote sensing image), and performing joint dictionary learning on the extracted large number of unchanged samples to obtain a base of the unchanged samples;
1.1) preprocessing the multi-temporal remote sensing image, and selecting a large number of unchanged samples in different phases at the same place;
1.2) splicing a large number of unchanged samples selected from different phases together, and obtaining the basis of the unchanged samples by using a sparse expression method, namely:
X1＝D1S1
wherein, X1An unchanged sample; d1A dictionary of unchanged samples, i.e. the basis of unchanged samples; s1Sparse representation coefficients for unchanged samples.
In the invention, the joint dictionary learning is to splice samples of the same category in remote sensing images of different time phases together to be used as a new joint sample, carry out sparse learning on the joint sample to obtain a dictionary of the joint sample, and the dictionary obtained by the joint sample is called as a joint dictionary. In the step, unchanged samples of the remote sensing images of different time phases are selected, so that a combined dictionary of the unchanged samples, namely a base of the unchanged samples, is obtained.
2) Marking all other multi-temporal samples which are not selected in the step 1) as test set samples, wherein the test set samples comprise changed samples and unchanged samples; carrying out sparse reconstruction on the test set sample by using the basis of the unchanged sample obtained in the step 1) to obtain a reconstructed test set sample; the difference between the test set sample and the reconstructed test set sample is obtained to obtain a difference image;
2.1) after the test set sample is established, decomposing the test set sample into a new dictionary and a new sparse expression coefficient by using a sparse expression mechanism, namely:
X2＝D2S2
wherein, X2Is a test set sample; d2A dictionary of test set samples, i.e., the bases of the test set samples; s2Sparse expression coefficients of the test set samples are obtained;
2.2) Using the base D of the unchanged sample obtained in step 1.2)1Base D of the test set sample obtained in step 2.1) is replaced2Reconstructing a test set sample of the multi-temporal remote sensing image under a sparse expression frame to obtain a reconstructed test set sample X2′；
2.3) for test set sample X2And reconstructing the test set sample X2' carry out difference operation to obtain difference image.
3) Selecting a small amount of change samples from the multi-temporal remote sensing images (generally, 2% -10% of change samples can be selected from the multi-temporal remote sensing images); performing sparse reconstruction on the changed samples in different time phases according to the basis of the unchanged samples to obtain reconstructed images of the changed samples in different time phases; obtaining a change threshold value of the change sample by using a difference value between reconstructed images of different phase change samples through pooling operation;
3.1) for each single-time phase change sample set in the selected small number of change samples, respectively utilizing the sparse expression coefficient and the base D of the unchanged sample obtained in the step 1.2)1Reconstructing to obtain a reconstructed image of the sample set with different phase changes;
3.2) performing difference operation on the reconstructed images of the different phase change sample sets obtained in the step 3.1) to obtain a reconstructed error image of the changed samples on the basis of the unchanged samples;
3.3) performing pooling operation on the reconstruction error image obtained in the step 3.2) to obtain a reconstruction error, namely a change threshold of the change sample. Because the change threshold is acquired from the experimental image in a self-adaptive manner, the defect that the change threshold is selected manually or depends on other mature algorithms can be overcome better, and the self-adaptive threshold selection strategy of the invention has better capability on the test result.
4) Judging a change area of the multi-temporal remote sensing image by combining the difference image obtained in the step 2) and the change threshold of the change sample obtained in the step 3), and counting the detection rate;
4.1) comparing the pixel value of the difference image obtained in the step 2.3) with the change threshold value of the change sample obtained in the step 3.3); if the pixel value of the difference image is larger than or equal to the change threshold of the change sample, marking the area corresponding to the difference image as a change area; if the pixel value of the difference image is smaller than the change threshold of the change sample, marking the area corresponding to the difference image as an unchanged area;
4.2) judging the change area of the multi-temporal remote sensing image according to the step 4.1), wherein the average judging accuracy is the detection rate, and the calculation mode of the detection rate is as follows: and judging the percentage of the number of the correct pixels to the total number of the pixels.
The beneficial effects of the multi-temporal remote sensing image change detection method based on the joint dictionary learning provided by the invention are explained by using a simulation experiment as follows:
1) simulation conditions
Performing simulation by using MATLAB software on an operating system with a central processing unit of Intel (R) Core i 3-5302.93 GHZ and a memory of 4G, WINDOWS 7; the test images used in the experiment were multispectral data of the Kunlshan region and the Taizhou region taken by the resource number three satellite (see FIGS. 2a, 2b, 3a and 3 b).
2) Emulated content
The detection method provided by the invention is adopted for carrying out the experiment:
firstly, selecting a large number of unchanged area sample pairs and a small number of changed area sample pairs from two databases of the Kunma area and the Taizhou area respectively, and taking all the remaining sample pairs as test set samples;
secondly, learning the basis of the unchanged samples on a large number of unchanged area sample pairs by a sparse expression method, constructing a test set sample by using the basis of the unchanged samples, and obtaining a difference image between the test set sample and the reconstructed test set sample;
then, an adaptive change threshold is learned for a small number of change samples, and discrimination is performed on the difference image using the change threshold.
The experimental results obtained with the detection method of the invention are compared with those obtained with the conventional detection method, wherein:
FIGS. 4a and 5a show the results of tests performed in the Kunming region and the Taizhou region, respectively, using the variation vector analysis method (see the references: A the experimental frame for the unsupervised change based on change vector analysis in the polar domain. IEEE transactions. on Geoscience and Remote Sensing,45(1),218- & 236, 2007). It can be found that: on a Kunshan database and a Taizhou database, the detection effect of the change vector analysis method is not ideal, the error distribution area is more, and the noise points are more. This is because the change vector analysis method cannot extract difference information well for a single-band remote sensing image.
FIGS. 4b and 5b show the results of tests on Kunmansi and Taizhou regions, respectively, using principal component analysis (see references: Unstand modified detection with kernel. IEEE geosci. remote Sens. Lett,9(6): 1026-. It can be found that: on the Kun mountain database and the Taizhou database, good test results cannot be obtained by adopting a principal component analysis method, the number of noise points is large, and a detected image is not smooth. The reason is that the change area and the unchanged area cannot be well distinguished in the principal component extraction process, which indicates that the principal component analysis method is not well applicable to the detection research of the change of the remote sensing image.
FIGS. 4c and 5c are The results of experiments for The Kunming region and The Taizhou region, respectively, using an iterative weighted multivariate detection algorithm (see references: The regularized detection method for change detection in multi-and hyperspectral data. IEEE trans. on Image Process,16(2):463 and 478, 2007). It can be found that: on a Kun mountain database and a Taizhou database, the detection results of the method are not good as those of the detection method provided by the invention, the detection accuracy is not high, and particularly, the error division area is more. This is because the iterative weighted multivariate detection algorithm has a high probability of erroneous judgment when distinguishing similar features.
FIGS. 4d and 5d show results of tests using Semi-supervised significance detection algorithms (see references: Semi-supervised significance detection using sm entry solution path. IEEE Trans. on Geoscience and remove Sensing,51(4-1):1939-1950,2013.) for the Kunzan area and the Taizhou area, respectively. It can be found that: on a Kunshan database and a Taizhou database, the method has poor performance, low detection rate and high false detection probability, and simultaneously, the detected image is not smooth. This is because the method is heavily dependent on the desired features, and it is time-consuming and laborious to select different features for different images, and thus the method is not universal.
FIGS. 4e and 5e are the results of experiments in the Kunming region and Taizhou region, respectively, using a supervised Slow feature algorithm (see references: Slow feature analysis for change detection in multispectral image, IEEE transactions. on Geoscience and RemoteSensing,52(5):2858 and 2874, 2014.). It can be found that: on the Kunman database and the Taizhou database, the method has good performance, high detection rate and less area errors of the images; although the noise points are few, the method cannot well express the change information for the area with insignificant change, so the method is to be improved in detection performance.
Fig. 4f and fig. 5f are test results of the kunzan area and the taizhou area respectively by using the detection method provided by the present invention (instead of using the adaptive threshold selection strategy, the conventional clustering algorithm is used). It can be found that: on a Kunshan database and a Taizhou database, the detection result of the detection method (a traditional clustering algorithm is selected instead of an adaptive threshold selection strategy) is similar to the detection result of a supervised slow feature algorithm. Therefore, without step 3) (i.e., without the adaptive threshold selection strategy), the joint dictionary learning method of the present invention performs well in extracting variation information.
Fig. 4g and 5g are the test results for the kunzan region and the taizhou region, respectively, using the detection method provided by the present invention (including the steps of the adaptive threshold selection strategy). It can be found that: on a Kunman database and a Taizhou database, the test result of the detection method is the best, the classification accuracy is high, and the noise points are few; meanwhile, the self-adaptive change threshold learned in the step 3) can be used for better distinguishing the changed area from the unchanged area.
Fig. 4h and 5h are real change region labeling diagrams of multi-temporal remote sensing images of the Kunmshan region and the Taizhou region respectively.
Finally, the test results of different detection methods are compared with the real standard (i.e., fig. 4h and fig. 5h), and the comparison result is counted as the detection accuracy for the multi-temporal remote sensing image change detection, with the result shown in table 1.
As can be seen from the table 1, the multi-temporal remote sensing image change detection method based on the joint dictionary learning has better performance, and the detection rate of the method is higher than that of the existing direct detection method. The invention takes the information of a large number of unchanged samples into full consideration, thereby overcoming the defect of insufficient utilization of the large number of unchanged samples in the traditional method; and the invention can obtain the change threshold value suitable for the image from the information of the image through the selection strategy of the self-adaptive threshold value, and can avoid the influence of different noises and shooting angles on the detection result, thereby obtaining a better identification result, further verifying the advancement of the multi-temporal remote sensing image change detection method based on the joint dictionary learning, and further providing better technical support for the aspects of geographical and national condition detection, military reconnaissance, environmental monitoring and the like.
TABLE 1 detection rates of different detection methods
Claims (3)
1. A multi-temporal remote sensing image change detection method based on joint dictionary learning is characterized by comprising the following steps: the multi-temporal remote sensing image change detection method based on the joint dictionary learning comprises the following steps:
1) extracting a large number of unchanged samples from the multi-temporal remote sensing image, and performing joint dictionary learning on the extracted large number of unchanged samples to obtain the basis of the unchanged samples;
the specific implementation manner of the step 1) is as follows:
1.1) preprocessing the multi-temporal remote sensing image, and selecting a large number of unchanged samples in different phases at the same place;
1.2) splicing a large number of unchanged samples selected from different phases together, and obtaining the basis of the unchanged samples by using a sparse expression method, namely:
X1＝D1S1
wherein, X1An unchanged sample; d1A dictionary of unchanged samples, i.e. the basis of unchanged samples; s1Sparse representation coefficients for unchanged samples;
2) marking all other multi-temporal samples which are not selected in the step 1) as test set samples, wherein the test set samples comprise changed samples and unchanged samples; carrying out sparse reconstruction on the test set sample by using the basis of the unchanged sample obtained in the step 1) to obtain a reconstructed test set sample; the difference between the test set sample and the reconstructed test set sample is obtained to obtain a difference image;
3) selecting a small amount of change samples from the multi-temporal remote sensing image; performing sparse reconstruction on the changed samples in different time phases according to the basis of the unchanged samples to obtain reconstructed images of the changed samples in different time phases; obtaining a change threshold value of the change sample by using a difference value between reconstructed images of different phase change samples through pooling operation;
the specific implementation manner of the step 3) is as follows:
3.1) for each single-time phase change sample set in the selected small number of change samples, respectively utilizing the sparse expression coefficient and the base D of the unchanged sample obtained in the step 1.2)1Reconstructing to obtain a reconstructed image of the sample set with different phase changes;
3.2) performing difference operation on the reconstructed images of the different phase change sample sets obtained in the step 3.1) to obtain a reconstructed error image of the changed samples on the basis of the unchanged samples;
3.3) performing pooling operation on the reconstruction error image obtained in the step 3.2) to obtain a reconstruction error, namely a change threshold of the change sample;
4) and (3) judging a change area of the multi-temporal remote sensing image by combining the difference image obtained in the step 2) and the change threshold of the change sample obtained in the step 3), and counting the detection rate.
2. The joint dictionary learning-based multi-temporal remote sensing image change detection method according to claim 1, characterized in that: the specific implementation manner of the step 2) is as follows:
2.1) after the test set sample is established, decomposing the test set sample into a new dictionary and a new sparse expression coefficient by using a sparse expression mechanism, namely:
X2＝D2S2
wherein, X2Is a test set sample; d2A dictionary of test set samples, i.e., the bases of the test set samples; s2Sparse expression coefficients of the test set samples are obtained;
2.2) Using the base D of the unchanged sample obtained in step 1.2)1Base D of the test set sample obtained in step 2.1) is replaced2Reconstructing a test set sample of the multi-temporal remote sensing image under a sparse expression frame to obtain a reconstructed test set sample X2′；
2.3) for test set sample X2And reconstructing the test set sample X2' carry out difference operation to obtain difference image.
3. The joint dictionary learning-based multi-temporal remote sensing image change detection method according to claim 2, characterized in that: the specific implementation manner of the step 4) is as follows:
4.1) comparing the pixel value of the difference image obtained in the step 2.3) with the change threshold value of the change sample obtained in the step 3.3); if the pixel value of the difference image is larger than or equal to the change threshold of the change sample, marking the area corresponding to the difference image as a change area; if the pixel value of the difference image is smaller than the change threshold of the change sample, marking the area corresponding to the difference image as an unchanged area;
4.2) judging the change area of the multi-temporal remote sensing image according to the step 4.1), wherein the average judging accuracy is the detection rate, and the calculation mode of the detection rate is as follows: and judging the percentage of the number of the correct pixels to the total number of the pixels.
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 CN105989595A (en) | 2016-10-05 |
| CN105989595B true 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) |
Families Citing this family (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 |
| 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 |
-
2015
- 2015-02-12 CN CN201510075774.5A patent/CN105989595B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN105989595A (en) | 2016-10-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110443143B (en) | Multi-branch convolutional neural network fused remote sensing image scene classification method | |
| Kang et al. | Deep unsupervised embedding for remotely sensed images based on spatially augmented momentum contrast | |
| CN113378632B (en) | Pseudo-label optimization-based unsupervised domain adaptive pedestrian re-identification method | |
| CN113449594B (en) | Multilayer network combined remote sensing image ground semantic segmentation and area calculation method | |
| CN107424159B (en) | Image semantic segmentation method based on super-pixel edge and full convolution network | |
| CN107133569B (en) | Monitoring video multi-granularity labeling method based on generalized multi-label learning | |
| CN106408030B (en) | SAR image classification method based on middle layer semantic attribute and convolutional neural networks | |
| CN105528595A (en) | Method for identifying and positioning power transmission line insulators in unmanned aerial vehicle aerial images | |
| CN105989336B (en) | Scene recognition method based on deconvolution deep network learning with weight | |
| CN102385592B (en) | Image concept detection method and device | |
| CN103440508B (en) | The Remote Sensing Target recognition methods of view-based access control model word bag model | |
| Zhao et al. | Sparsity-guided saliency detection for remote sensing images | |
| CN109635726B (en) | Landslide identification method based on combination of symmetric deep network and multi-scale pooling | |
| CN113808166B (en) | Single-target tracking method based on clustering difference and depth twin convolutional neural network | |
| CN111931703B (en) | Object detection method based on human-object interaction weak supervision label | |
| CN113723492A (en) | Hyperspectral image semi-supervised classification method and device for improving active deep learning | |
| CN106557740A (en) | The recognition methods of oil depot target in a kind of remote sensing images | |
| Hu et al. | A comparative study of sampling analysis in scene classification of high-resolution remote sensing imagery | |
| CN106096622B (en) | Semi-supervised Classification of hyperspectral remote sensing image mask method | |
| CN105868711B (en) | Sparse low-rank-based human behavior identification method | |
| CN114119621A (en) | SAR remote sensing image water area segmentation method based on depth coding and decoding fusion network | |
| CN107203779A (en) | Hyperspectral dimensionality reduction method based on spatial-spectral information maintenance | |
| CN114565824B (en) | Single-stage rotating ship detection method based on full convolution network | |
| CN112270285A (en) | SAR image change detection method based on sparse representation and capsule network | |
| Duan et al. | Multi-scale convolutional neural network for SAR image semantic segmentation |
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 |