CN108564585A - A kind of image change detection method based on Self-organizing Maps and deep neural network - Google Patents
A kind of image change detection method based on Self-organizing Maps and deep neural network Download PDFInfo
- Publication number
- CN108564585A CN108564585A CN201810397032.8A CN201810397032A CN108564585A CN 108564585 A CN108564585 A CN 108564585A CN 201810397032 A CN201810397032 A CN 201810397032A CN 108564585 A CN108564585 A CN 108564585A
- Authority
- CN
- China
- Prior art keywords
- pixel
- disparity map
- class
- variation
- mean
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 20
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 15
- 238000000034 method Methods 0.000 claims abstract description 22
- 238000010586 diagram Methods 0.000 claims abstract description 21
- 238000013507 mapping Methods 0.000 claims abstract description 6
- 238000012549 training Methods 0.000 claims abstract description 6
- 238000012545 processing Methods 0.000 claims description 8
- 238000001914 filtration Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 description 7
- 238000012544 monitoring process Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000001427 coherent effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000010287 polarization Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/10—Image enhancement or restoration using non-spatial domain filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
- G06T2207/10044—Radar image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
- G06T2207/20032—Median filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of image change detection method based on Self-organizing Maps and deep neural network, includes the following steps：1, it is handled using phasor when the MEAN couples of two width SAR of average ratio operator M based on median filter process, generates disparity map；2, Unsupervised clustering is carried out to disparity map using self-organized mapping network SOM, the pixel in disparity map is divided into unchanged class, noise class and variation class, obtains preliminary variation diagram；3, deep neural network DNN is built using automatic noise reduction codes device, DNN networks is trained using noise class pixel as training set, allow it to the feature of identification noise region；4, variation class pixel is inputted into trained DNN networks, remaining noise class pixel in variation class pixel is judged and remove, to obtain final variation diagram based on preliminary variation diagram.This method advantageously reduces the omission factor and false alarm rate of diameter radar image variation detection.
Description
Technical field
The present invention relates to Image Change Detection technical fields, especially a kind of to be based on Self-organizing Maps and deep neural network
Image change detection method.
Background technology
Synthetic aperture radar (SAR) monitoring image is as a kind of radar imagery, can be in pole compared with optical monitoring image
High-resolution is obtained under the meteorological condition of end.The application of SAR image is very extensive, as environmental protection, agricultural investigation, urban study,
Forest resource monitoring etc..Change detection techniques based on SAR image can to same geographic range, the SAR image of different time,
The situation of change of physical object between detection observed image.From the point of view of previous document, SAR image change detection techniques have been in
Reveal huge application potential.But SAR image background information is complicated, often texture is not notable, is covered by mottled effect, image
It is second-rate.How mottled effect is effectively handled, and acquisition is more clear available modified-image, has very important meaning
Justice.
It is previous studies have shown that the unsupervised variation detection of single polarization SAR image is made of three steps：1) to same
The SAR image of point different time is pre-processed；2) using treated, image generates disparity map；3) analysis disparity map, which generates, becomes
Change figure.It generates disparity map and analysis disparity map is very important two steps and is carried the purpose is to reduce the influence of coherent noise
The ability of High variation detection.Existing method has been achieved for certain progress, but still remain it is following two aspect the problem of.
1, it in terms of disparity map generation, because of influence of the Multiplicative random noise for SAR image, is directly generated using difference operator
SAR change-detection images has high unstability and extremely low robustness.In this respect, because ratio operator has preferably
Processing noise immune, researcher commonly pay attention to ratio operator use to improve detection performance.But traditional ratio operator without
The method rationally balanced reservation for inhibiting noise and fringe region.2, in terms of difference map analysis, traditional analysis method is often limited
In the generating process of disparity map.It is to need how according to the suitable disparity map analysis model of the different generating modes of disparity map selection
The critical issue to be solved.In fact, difference map analysis is a two-value classification problem.Since the pixel in difference diagram is labeled as
Change class label and constant class label.The pixel changed is marked as 1, and unchanged pixel is labeled as 0, ultimately produces one
The variation diagram of binary number matrix composition.But usually there is false alarm point in variation diagram (has not changing for variation category feature
Pixel is often referred to noise), research both domestic and external, which still lacks, at present is further processed false alarm point in variation diagram.
Invention content
The purpose of the present invention is to provide a kind of Image Change Detection side based on Self-organizing Maps and deep neural network
Method, this method advantageously reduce the omission factor and false alarm rate of diameter radar image variation detection.
To achieve the above object, the technical scheme is that：It is a kind of based on Self-organizing Maps and deep neural network
Image change detection method includes the following steps：
Step S1：It is handled using phasor when MEAN couples of two width SAR of average ratio operator M based on median filter process,
Generate disparity map；
Step S2：Unsupervised clustering is carried out to the disparity map of generation using self-organized mapping network SOM, it will be in disparity map
Pixel is divided into unchanged class, noise class and variation class, obtains preliminary variation diagram；
Step S3：Deep neural network DNN is built using automatic noise reduction codes device, using noise class pixel as training
Collection input DNN networks, are trained DNN networks, allow it to the feature of identification noise region；
Step S4：Variation class pixel is inputted into trained DNN networks, judges and remove to remain in variation class pixel
Noise class pixel, to based on preliminary variation diagram obtain final variation diagram.
Further, in step S1, using MEAN pairs of two width SAR phases of average ratio operator M based on median filter process
Figure is handled, and is included the following steps：
Step S11:Using average ratio operator respectively to two width SAR when phasor in each neighborhood of a point window carry out it is equal
Value filtering processing, and then average ratio disparity map is obtained, calculation formula is as follows：
Wherein, u1(i,j)、u2(i, j) indicates phasor I when two width from same place different time respectively1And I2Point
The local mean value of (i, j), N (i, j) indicate the neighborhood window of point (i, j),The first width figure and the second width figure are indicated respectively
The value of pixel s in middle neighborhood N (i, j), n indicate the points in neighborhood N (i, j), XmeanIndicate that phasor is in picture when to two width
The average ratio ratio that vegetarian refreshments (i, j) and its neighborhood acquire；All the points in clock synchronization phasor carry out above-mentioned processing, that is, obtain average ratio
Disparity map；
Step S12：Medium filtering calculating is carried out to the average ratio disparity map of generation：
Xmedian-mean=median ({ xs|s∈Nmean(i,j)})
Wherein, Xmedian-meanExpression uses the value obtained after median filter process, median ({ x to average ratio ratios|s
∈Nmean(i, j) }) indicate average ratio ratio to the point of each pixel (i, j) and its neighborhood, find their intermediate value
Instead of the value of point (i, j) script, Nmean(i, j) indicates point (i, j) in XmeanNeighborhood window.
Further, in step S2, Unsupervised clustering is carried out to the disparity map of generation using self-organized mapping network SOM,
Pixel in disparity map is divided into unchanged class, noise class and variation class, preliminary variation diagram is obtained, includes the following steps：
Step S21：The input layer of SOM networks and the number of nodes of competition layer are set, and each numerical value in disparity map is made
SOM networks are inputted for input data, each value in disparity map indicates the gray-value variation of phasor corresponding position when two width SAR
Amplitude；
Step S22：Interior star weight vector corresponding to each node i of competition layer is normalized
Step S23：By input layer vector XmIt is compared with the interior star weight vector of each node of competition layer, wherein m is
Input layer length, the i.e. number of nodes of input layer,
Step S24：With input layer vector XmThe shortest competition layer node i * of distance d win, update its interior star weight vector,
The update method of interior star weight vector is：
Wherein,It indicates to calculate XmWithEuclidean distance；
Step S25：Renewal learning rate a, and judge whether a reaches threshold value is then output as a result, otherwise return to step S22；
To which the pixel in disparity map is divided into three classes：1, numerical value is 0 in disparity map, indicates two width SAR phases
The gray value for scheming corresponding pixel is non-changing, is denoted as S_unchange；2, numerical value is more than the big of setting in disparity map
Threshold epsilon1, it indicates that the gray value of phasor corresponding pixel when two width SAR is variation, is denoted as S_change；3, in difference
Numerical value is less than the small threshold epsilon of setting in figure2, wherein ε2<ε1, indicate that the corresponding pixel of phasor is noise when two width SAR, note
For S_noise.
Compared to the prior art, the beneficial effects of the invention are as follows：First, this method improves mean value using median filter
Than operator generate differential image, reduce image open country spot noise for generate disparity map influence, and with logarithm ratio operator
The disparity map of formation is compared, and the marginal information of image is greatly remained, and reduces omission factor；Secondly, Self-organizing Maps net is utilized
Network pre-processes disparity map, obtains preliminary variation diagram, and the pixel of disparity map is made to be divided into unchanged class, noise class and variation class three
Type；Finally, network training is carried out to noise class training set using deep neural network, is made an uproar with reducing the residue changed in class
Sound carries out analysis to preliminary variation diagram and obtains final variation diagram.Compared with other current main stream approach, this method is directly against making an uproar
Sound is handled, and lower omission factor is shown in diameter radar image data set with main more preferably false alarm
Rate, and there is universality to various data sets, it is highly practical.
Description of the drawings
Fig. 1 is the implementation flow chart of the method for the present invention.
Specific implementation mode
Below in conjunction with the accompanying drawings and specific embodiment the invention will be further described.
The present invention provides a kind of image change detection method based on Self-organizing Maps and deep neural network, such as Fig. 1 institutes
Show, includes the following steps：
Step S1：It is handled using phasor when MEAN couples of two width SAR of average ratio operator M based on median filter process,
Generate disparity map.Specifically include following steps：
Step S11:Using average ratio operator respectively to two width SAR when phasor in each neighborhood of a point window carry out it is equal
Value filtering processing, and then average ratio disparity map is obtained, calculation formula is as follows：
Wherein, u1(i,j)、u2(i, j) indicates phasor I when two width from same place different time respectively1And I2Point
The local mean value of (i, j), N (i, j) indicate the neighborhood window of point (i, j),The first width figure and the second width figure are indicated respectively
The value of pixel s in middle neighborhood N (i, j), thenIt is indicated respectively to the first width figure and the second width
The value of the interior all pixels points of neighborhood N (i, j) sums up in figure, and n indicates the points in neighborhood N (i, j), XmeanIt indicates to two width
The average ratio ratio that Shi Xiangtu is acquired in pixel (i, j) and its neighborhood；All the points in clock synchronization phasor carry out above-mentioned processing, i.e.,
Obtain average ratio disparity map；
Step S12：Medium filtering calculating is carried out to the average ratio disparity map of generation：
Xmedian-mean=median ({ xs|s∈Nmean(i,j)})
Wherein, Xmedian-meanExpression uses the value obtained after median filter process, median ({ x to average ratio ratios|s
∈Nmean(i, j) }) it indicates, to the average ratio ratio of each pixel (i, j) and surrounding point, to find their intermediate value
Instead of the value of point (i, j) script, xsForOrNmean(i, j) indicates point (i, j) in XmeanNeighborhood window.
Step S2：Unsupervised clustering is carried out to the disparity map of generation using self-organized mapping network SOM, it will be in disparity map
Pixel is divided into unchanged class, noise class and variation class, obtains preliminary variation diagram.Specifically include following steps：
Step S21：The input layer of SOM networks and the number of nodes of competition layer are set, in the present embodiment, by input layer
Number of nodes is set as 1, and the number of nodes of competition layer is set as 3, is then inputted each numerical value in disparity map as input data
SOM networks, each value in disparity map indicate the gray-value variation amplitude of phasor corresponding position when two width SAR；
Step S22：Interior star weight vector corresponding to each node i of competition layer is normalized
Step S23：By input layer vector XmIt is compared with the interior star weight vector of each node of competition layer, wherein m is
Input layer length, the i.e. number of nodes of input layer,
Step S24：With input layer vector XmThe shortest competition layer node i * of distance d win, update its interior star weight vector,
The update method of interior star weight vector is：
Wherein,It indicates to calculate XmWithEuclidean distance；
Step S25：Renewal learning rate a, and judge whether a reaches threshold value is then output as a result, otherwise return to step S22；
To which the pixel in disparity map is divided into three classes：1, numerical value is 0 in disparity map, indicates two width SAR phases
The gray value for scheming corresponding pixel is non-changing, is denoted as S_unchange；2, numerical value is more than the big of setting in disparity map
Threshold epsilon1, it indicates that the gray value of phasor corresponding pixel when two width SAR is variation, is denoted as S_change；3, in difference
Numerical value is less than the small threshold epsilon of setting in figure2, wherein ε2<ε1, indicate that the corresponding pixel of phasor is noise when two width SAR, note
For S_noise.
Step S3：Deep neural network DNN is built using automatic noise reduction codes device, using noise class pixel as training
Collection input DNN networks, are trained DNN networks, allow it to the feature of identification noise region.
Step S4：Variation class pixel is inputted into trained DNN networks, judges and remove to remain in variation class pixel
Noise class pixel, to based on preliminary variation diagram obtain final variation diagram.
The above are preferred embodiments of the present invention, all any changes made according to the technical solution of the present invention, and generated function is made
When with range without departing from technical solution of the present invention, all belong to the scope of protection of the present invention.
Claims (3)
1. a kind of image change detection method based on Self-organizing Maps and deep neural network, which is characterized in that including following
Step：
Step S1：It is handled, is generated using phasor when MEAN couples of two width SAR of average ratio operator M based on median filter process
Disparity map；
Step S2：Unsupervised clustering is carried out to the disparity map of generation using self-organized mapping network SOM, by the pixel in disparity map
Point is divided into unchanged class, noise class and variation class, obtains preliminary variation diagram；
Step S3：Deep neural network DNN is built using automatic noise reduction codes device, it is defeated as training set using noise class pixel
Enter DNN networks, DNN networks are trained, allows it to the feature of identification noise region；
Step S4：Variation class pixel is inputted into trained DNN networks, judge and removes remaining in variation class pixel make an uproar
Sound class pixel, to obtain final variation diagram based on preliminary variation diagram.
2. a kind of image change detection method based on Self-organizing Maps and deep neural network according to claim 1,
It is characterized in that, in step S1, carried out using phasor when MEAN couples of two width SAR of average ratio operator M based on median filter process
Processing, includes the following steps：
Step S11:Using average ratio operator respectively to two width SAR when phasor in each neighborhood of a point window carry out mean value filter
Wave processing, and then average ratio disparity map is obtained, calculation formula is as follows：
Wherein, u1(i,j)、u2(i, j) indicates phasor I when two width from same place different time respectively1And I2In point (i, j)
Local mean value, N (i, j) indicate point (i, j) neighborhood window,Neighborhood in the first width figure and the second width figure is indicated respectively
The value of pixel s in N (i, j), n indicate the points in neighborhood N (i, j), XmeanIndicate to two width when phasor pixel (i,
And its average ratio ratio that acquires of neighborhood j)；All the points in clock synchronization phasor carry out above-mentioned processing, that is, obtain average ratio disparity map；
Step S12：Medium filtering calculating is carried out to the average ratio disparity map of generation：
Xmedian-mean=median ({ xs|s∈Nmean(i,j)})
Wherein, Xmedian-meanExpression uses the value obtained after median filter process, median ({ x to average ratio ratios|s∈Nmean
(i, j) }) it indicates, to the average ratio ratio of the point of each pixel (i, j) and its neighborhood, to find their intermediate value to replace a little
The value of (i, j) script, Nmean(i, j) indicates point (i, j) in XmeanNeighborhood window.
3. a kind of image change detection method based on Self-organizing Maps and deep neural network according to claim 2,
It is characterized in that, in step S2, Unsupervised clustering is carried out to the disparity map of generation using self-organized mapping network SOM, by difference
Pixel in figure is divided into unchanged class, noise class and variation class, obtains preliminary variation diagram, includes the following steps：
Step S21：The input layer of SOM networks and the number of nodes of competition layer are set, and using each numerical value in disparity map as defeated
Enter data input SOM networks, each value in disparity map indicates the gray-value variation width of phasor corresponding position when two width SAR
Degree；
Step S22：Interior star weight vector corresponding to each node i of competition layer is normalized
Step S23：By input layer vector XmIt is compared with the interior star weight vector of each node of competition layer, wherein m is input layer
Length, the i.e. number of nodes of input layer,
Step S24：With input layer vector XmThe shortest competition layer node i * of distance d win, update its interior star weight vector, interior star
The update method of weight vector is：
Wherein,It indicates to calculate XmWithEuclidean distance；
Step S25：Renewal learning rate a, and judge whether a reaches threshold value is then output as a result, otherwise return to step S22；
To which the pixel in disparity map is divided into three classes：1, numerical value is 0 in disparity map, indicates phasor pair when two width SAR
The gray value for the pixel answered is non-changing, is denoted as S_unchange；2, numerical value is more than the big threshold value set in disparity map
ε1, it indicates that the gray value of phasor corresponding pixel when two width SAR is variation, is denoted as S_change；3, in disparity map
Numerical value is less than the small threshold epsilon of setting2, wherein ε2<ε1, indicate that the corresponding pixel of phasor is noise when two width SAR, is denoted as S_
noise。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810397032.8A CN108564585B (en) | 2018-04-27 | 2018-04-27 | Image change detection method based on self-organizing mapping and deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810397032.8A CN108564585B (en) | 2018-04-27 | 2018-04-27 | Image change detection method based on self-organizing mapping and deep neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108564585A true CN108564585A (en) | 2018-09-21 |
| CN108564585B CN108564585B (en) | 2022-01-18 |
Family
ID=63537397
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810397032.8A Expired - Fee Related CN108564585B (en) | 2018-04-27 | 2018-04-27 | Image change detection method based on self-organizing mapping and deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108564585B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113269235A (en) * | 2021-05-10 | 2021-08-17 | 青岛理工大学 | Assembly body change detection method and device based on unsupervised learning |
| CN113489514A (en) * | 2021-07-05 | 2021-10-08 | 国网湖南省电力有限公司 | Power line communication noise identification method and device based on self-organizing mapping neural network |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101383023A (en) * | 2008-10-22 | 2009-03-11 | 西安交通大学 | Neural network short-term electric load prediction based on sample dynamic organization and temperature compensation |
| CN102867187A (en) * | 2012-07-04 | 2013-01-09 | 西安电子科技大学 | NSST (NonsubsampledShearlet Transform) domain MRF (Markov Random Field) and adaptive threshold fused remote sensing image change detection method |
| CN103544546A (en) * | 2013-10-31 | 2014-01-29 | 浙江大学 | Method for online prediction of quiescent voltage stability limit of electric system |
-
2018
- 2018-04-27 CN CN201810397032.8A patent/CN108564585B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101383023A (en) * | 2008-10-22 | 2009-03-11 | 西安交通大学 | Neural network short-term electric load prediction based on sample dynamic organization and temperature compensation |
| CN102867187A (en) * | 2012-07-04 | 2013-01-09 | 西安电子科技大学 | NSST (NonsubsampledShearlet Transform) domain MRF (Markov Random Field) and adaptive threshold fused remote sensing image change detection method |
| CN103544546A (en) * | 2013-10-31 | 2014-01-29 | 浙江大学 | Method for online prediction of quiescent voltage stability limit of electric system |
Non-Patent Citations (4)
| Title |
|---|
| MAOGUO GONG 等: "Feature learning and change feature classification based on deep learning for ternary change detection in SAR images", 《ELSEVIER》 * |
| SHOUHEI OHNO 等: "Efficient automatic target recognition method for aircraft SAR image using supervised SOM clustering", 《IEEE》 * |
| 刘逸: "粒子群优化算法的改进及应用研究", 《中国博士学位论文全文数据库信息科技辑》 * |
| 杨国栋: "基于分布式并行聚类的SAR图像变化检测算法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113269235A (en) * | 2021-05-10 | 2021-08-17 | 青岛理工大学 | Assembly body change detection method and device based on unsupervised learning |
| CN113269235B (en) * | 2021-05-10 | 2022-12-27 | 青岛理工大学 | Assembly body change detection method and device based on unsupervised learning |
| CN113489514A (en) * | 2021-07-05 | 2021-10-08 | 国网湖南省电力有限公司 | Power line communication noise identification method and device based on self-organizing mapping neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108564585B (en) | 2022-01-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108537192B (en) | Remote sensing image earth surface coverage classification method based on full convolution network | |
| Yuan et al. | Deep-learning-based multispectral satellite image segmentation for water body detection | |
| CN111259905B (en) | Feature fusion remote sensing image semantic segmentation method based on downsampling | |
| Li et al. | Survey of single image super‐resolution reconstruction | |
| CN111709902A (en) | Infrared and visible light image fusion method based on self-attention mechanism | |
| CN106295637B (en) | A kind of vehicle identification method based on deep learning and intensified learning | |
| CN107368845A (en) | A kind of Faster R CNN object detection methods based on optimization candidate region | |
| CN111832504A (en) | Space information intelligent integrated generation method for satellite in-orbit application | |
| Thoonen et al. | Classification of heathland vegetation in a hierarchical contextual framework | |
| CN108564585A (en) | A kind of image change detection method based on Self-organizing Maps and deep neural network | |
| CN104463822B (en) | Multi-focus image fusing method and its device based on multiple dimensioned global filtering | |
| Xiao et al. | Physics-based GAN with iterative refinement unit for hyperspectral and multispectral image fusion | |
| CN115810149A (en) | High-resolution remote sensing image building extraction method based on superpixel and image convolution | |
| Duan et al. | Multi-scale convolutional neural network for SAR image semantic segmentation | |
| Zhou et al. | MSAR‐DefogNet: Lightweight cloud removal network for high resolution remote sensing images based on multi scale convolution | |
| CN112330562B (en) | Heterogeneous remote sensing image transformation method and system | |
| CN114708494A (en) | Rural homestead building identification method and system | |
| CN112767277B (en) | Depth feature sequencing deblurring method based on reference image | |
| CN113901931A (en) | Knowledge distillation model-based behavior recognition method for infrared and visible light videos | |
| Chen et al. | Attentive generative adversarial network for removing thin cloud from a single remote sensing image | |
| Alhichri | RS-DeepSuperLearner: fusion of CNN ensemble for remote sensing scene classification | |
| Rangzan et al. | Improved water quality mapping based on cross‐fusion of Sentinel‐2 and Landsat‐8 imageries | |
| Liu et al. | SI-SA GAN: A generative adversarial network combined with spatial information and self-attention for removing thin cloud in optical remote sensing images | |
| CN112729562B (en) | Sea ice distribution detection method based on improved U-shaped convolutional neural network | |
| CN115953312A (en) | Joint defogging detection method and device based on single image and storage medium |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20220118 |