CN105844279B - SAR image change detection based on deep learning and SIFT feature - Google Patents
SAR image change detection based on deep learning and SIFT feature Download PDFInfo
- Publication number
- CN105844279B CN105844279B CN201610163983.XA CN201610163983A CN105844279B CN 105844279 B CN105844279 B CN 105844279B CN 201610163983 A CN201610163983 A CN 201610163983A CN 105844279 B CN105844279 B CN 105844279B
- Authority
- CN
- China
- Prior art keywords
- sar image
- image
- feature
- neural network
- sar
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2137—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on criteria of topology preservation, e.g. multidimensional scaling or self-organising maps
- G06F18/21375—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on criteria of topology preservation, e.g. multidimensional scaling or self-organising maps involving differential geometry, e.g. embedding of pattern manifold
-
- 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
- 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
- G06T2207/20081—Training; Learning
Abstract
The SAR image change detection based on deep learning and SIFT feature that the invention discloses a kind of mainly solves the problem that the prior art is sensitive to the speckle noise of SAR image, causes final variation testing result precision not high.The specific steps of the present invention are as follows: (1) reading in SAR image；(2) it normalizes；(3) training characteristics are constructed；(4) training deep neural network；(5) SAR image read in two width makees log ratio operation, obtains log ratio differential image；(6) the neighborhood characteristics sample matrix of log ratio differential image is constructed；(7) log ratio differential image is detected；(8) output variation testing result figure.The present invention takes full advantage of SIFT feature to the stability characteristic (quality) of SAR image speckle noise, overcomes the influence of SAR image speckle noise, improves the accuracy rate of SAR image variation detection.
Description
Technical field
The invention belongs to technical field of image processing, further relate to one in Remote Sensing Imagery Change Detection technical field
Kind is based on deep learning and shift-invariant operator transformation SIFT (Scale-invariant Feature Transform, SIFT)
Synthetic aperture radar SAR (Synthetic Aperture Radar, SAR) image change detection method of feature.The present invention can
Region of variation for the SAR image to areal different periods detects.
Background technique
Radar imaging technology is to grow up the fifties in last century, which has been achieved for the hair of great-leap-forward so far
Exhibition.At present the technology military affairs, agricultural, ocean, geology, in terms of be widely used.
Synthetic aperture radar as a kind of active microwave sensor, have high resolution, round-the-clock, round-the-clock work and
The strong feature of penetration power, therefore SAR is not influenced by correlated conditions such as atmospheric conditions and cloud covers.SAR emergency event, from
The detection of right disaster and assessment etc. the advantage incomparable with remote sensing image.The technology has been widely used in
The fields such as military, agricultural and scientific research.SAR image variation detection be by the two width SAR images to areal different times into
Row analysis, to obtain the change information of atural object or target.SAR image change detection techniques play in many applications at present
Increasingly important role, such as the detection and assessment of natural calamity, the monitoring of environment, city management planning and military investigation
Deng.
At present about SAR image variation detection two class of method rough classification: (1) classification and predicting method, also referred to as after classify
Comparison method.This method carries out independent sorting to two width original images first respectively, then carries out pixel-by-pixel to the image that two width have been classified
Ground compares, and then obtains final variation testing result；(2) classification, also referred to as disparity map classification method after comparing.The party is first
The disparity map of two images is obtained by differential technique, ratio method or log ratio method etc., then the disparity map is analyzed.Its
In the second class method be popular method at present, therefore the construction of multidate SAR image disparity map is very crucial.It uses at present
Obtaining more method includes image differential technique, image ratio method and log ratio method etc., then to the difference obtained more afterwards
Figure is further analyzed, including transformation, probability distribution etc., obtains final variation testing result.Second class method is simply straight
It sees, obtained variation details is more significant.
" SAR image change detection based on gauss hybrid models " that Nanjing electronic technology research institute delivers at it
A kind of SAR image variation detection based on gauss hybrid models is proposed in (modern radar, 2014,36 (9): 34-36) paper
Method.This method passes through the disparity map of log ratio method construction SAR image first, then analyzes the probability statistics of disparity map, then
Probability distribution is fitted using the mixed model of four Gaussian functions, obtains final variation testing result.This method is deposited
Shortcoming be firstly, this method is sensitive to the speckle noise of SAR image, to cause the precision for finally changing detection not high.
Secondly as the probability distribution of different SAR images is different, thus this method detects robustness for the variation of different SAR images
It is not high.
In the patent of its application, " SAR image based on unsupervised deep neural network changes inspection for Xian Electronics Science and Technology University
Survey " propose in (number of patent application: 201410818305.3, publication number: CN104517124A) it is a kind of based on unsupervised depth
The SAR image of neural network changes detection.This method carries out joint FCM points to the SAR image of areal difference phase first
Class obtains coarse variation testing result；Then the non-noise point for selecting possibility big according to initial variation testing result as
The training sample of deep neural network；These samples are input in deep neural network again and are trained；Finally two width are waited for
The image of detection, which inputs, obtains final variation testing result in trained deep neural network.Deficiency existing for this method
Place is that this method does not consider the influence of SAR image speckle noise when joint classification, leads to selected training sample point
Reliability is inadequate.
Summary of the invention
It is an object of the invention to overcome above-mentioned prior art, propose a kind of special based on deep learning and SIFT
The SAR image change detection of sign, to realize the accurate detection to SAR image change region.This method combines depth
Two methods of habit and SIFT feature, directly by SIFT feature training deep neural network, since SIFT feature can reflect figure
The local feature of picture all has invariance to image rotation, scaling and brightness change, and to affine transformation and noise
Also a degree of stability is kept, thus can be used as the reliable training sample of deep neural network.This method thinking is simple
Clear, the feature by efficiently using original image improves the precision of variation detection.
The present invention realizes that the thinking of above-mentioned purpose is: extracting original image with Scale invariant features transform method first
SIFT feature, using this as training sample, one deep neural network of training.Log ratio method is recycled to obtain original image
Disparity map, extract the domain features of each pixel of the disparity map, in this, as test data, be input to trained depth
It is tested in neural network, exports final variation testing result.
The specific steps that the present invention realizes include the following:
(1) SAR image is read in:
Read in SAR image I and J that two width of areal difference phase have been registrated and have corrected；
(2) it normalizes:
According to the following formula, SAR image I and J are normalized, the SAR image after being normalized:
Wherein, I' indicates the SAR image after SAR image I normalization, and min expression is minimized operation, and max expression takes most
Big Value Operations, J' indicate the SAR image after SAR image J normalization；
(3) training characteristics are constructed:
(3a) converts SIFT method using shift-invariant operator, extracts SAR image I' and J' after two width normalize respectively
Shift-invariant operator converts SIFT feature S1And S2；
(3b) converts SIFT feature S to two groups of shift-invariant operators1And S2Carry out cascade operation, the feature after being cascaded
S；
(3c) carries out dimensionality reduction to the feature S after cascade, using principal component analysis PCA algorithm, the feature S' after obtaining dimensionality reduction；
(4) the feature S' after dimensionality reduction is input in deep neural network, training deep neural network；
(5) according to the following formula, the log ratio differential image of two width SAR images of reading is calculated:
Wherein, D indicates the log ratio differential image for the two width SAR images read in, and log expression is derived from right log operations,
| | indicate the operation that takes absolute value, I and J respectively indicate the SAR image of reading；
(6) the neighborhood characteristics sample matrix of log ratio differential image D is constructed:
(6a) uses neighborhood characteristics extracting method, extracts from the pixel matrix that log ratio differential image D is constituted every
The neighborhood characteristics vector of a pixel；
Log ratio differential image D all pixels neighborhood of a point feature vector is formed M × N-dimensional neighborhood spy by (6b)
Levy sample matrix, wherein M indicates that the sum of all pixels point in log ratio differential image D, N indicate log ratio disparity map
As the dimension of the neighborhood characteristics vector of pixel each in D；
(7) log ratio differential image is detected:
The neighborhood characteristics sample matrix of log ratio differential image D is input in trained deep neural network, is examined
Log ratio differential image D is surveyed, obtaining each pixel detection in log ratio differential image D is to change the inspection of class or non-changing class
Survey classification；
(8) output detection classification.
Compared with the prior art, the present invention has the following advantages:
First, due to being extracted the SIFT for reading in SAR image present invention employs Scale invariant features transform SIFT algorithm
Feature, and deep neural network is trained using this feature, the selection for overcoming training sample in existing method is unreliable
The problem of, so that the present invention improves the precision of SAR image variation detection.
Second, since the present invention is extracted the SIFT feature of reading SAR image, the part that this feature can reflect image is special
Sign, and a degree of stability is also kept to affine transformation and noise, overcome affected by noise in existing method cause
The problem of cannot effectively detecting region of variation, so that the present invention improves the precision of SAR image variation detection.
Third, since the present invention is extracted the SIFT feature of reading SAR image, this feature is to image rotation, scaling
And brightness change all has invariance, thus there is stability to a certain extent to the feature extraction of different images, gram
Problem not high for different SAR image variation detection robustness in existing method is taken, so that the present invention is for different
SAR image information has stronger adaptability.
Detailed description of the invention
Fig. 1 is flow chart of the invention；
Fig. 2 is in emulation experiment of the present invention to the variation testing result figure of the area Bern SAR image；
Fig. 3 is in emulation experiment of the present invention to the variation testing result figure of the area Ottawa SAR image；
Fig. 4 is in emulation experiment of the present invention to the variation testing result figure of the Yellow River estuary area SAR image.
Specific embodiment
The invention will be further described below in conjunction with the accompanying drawings.
It is referring to Fig.1, of the invention that the specific implementation steps are as follows:
Step 1, SAR image is read in.
Read in SAR image I and J that two width of areal difference phase have been registrated and have corrected.
Step 2, it normalizes.
According to the following formula, SAR image I and J are normalized, the SAR image after being normalized:
Wherein, I' indicates the SAR image after SAR image I normalization, and min expression is minimized operation, and max expression takes most
Big Value Operations, J' indicate the SAR image after SAR image J normalization.
Step 3, training characteristics are constructed.
SIFT method is converted using shift-invariant operator, extracts the translation of SAR image I' and J' after the normalization of two width respectively
Invariant features convert SIFT feature S1And S2。
SIFT feature S is converted to two groups of shift-invariant operators1And S2Carry out cascade operation, the feature S after being cascaded.
To the feature S after cascade, dimensionality reduction is carried out using principal component analysis PCA algorithm, the feature S' after obtaining dimensionality reduction.
Step 4, the feature S' after dimensionality reduction is input in deep neural network, training deep neural network.
Training deep neural network specific steps are as follows:
The first step initializes the parameter of limited Boltzmann machine (RBM)；
Second step, feature S ' to be trained are trained using limited Boltzmann machine (RBM), obtain weight and biasing,
The network number of plies is set as 3 hidden layers, and each node layer number is respectively 250,150,100, and it is every to be limited Boltzmann machine (RBM)
One layer of 50 generation of training；
Third step is finely adjusted RBM training network using the BP neural network based on minimum cross entropy, and training algebra is
50 generations；
4th step obtains trained deep neural network.
Step 5, according to the following formula, the log ratio differential image of two width SAR images of reading is calculated:
Wherein, D indicates the log ratio differential image for the two width SAR images read in, and log expression is derived from right log operations,
| | indicate the operation that takes absolute value, I and J respectively indicate the SAR image of reading.
Step 6, the neighborhood characteristics sample matrix of log ratio differential image D is constructed.
Using neighborhood characteristics extracting method, each picture is extracted from the pixel matrix that log ratio differential image D is constituted
The neighborhood characteristics vector of vegetarian refreshments.
Neighborhood characteristics extracting method specific steps are as follows:
The first step, choosing a size on log ratio differential image D is n × n-pixel sliding window, by selected window
The value of mouth all pixels point pulls into the feature vector of a 1 × N-dimensional, wherein n is the size of sliding window, N=n × n；
Second step from left to right, successively sliding window from top to bottom obtains all pixels on log ratio differential image D
Neighborhood of a point feature vector.
Log ratio differential image D all pixels neighborhood of a point feature vector is formed into M × N-dimensional neighborhood characteristics sample
This matrix, wherein M indicates that the sum of all pixels point in log ratio differential image D, N indicate in log ratio differential image D
The dimension of the neighborhood characteristics vector of each pixel.
Step 7, log ratio differential image D is detected.
The neighborhood characteristics sample matrix of log ratio differential image D is input in trained deep neural network, is examined
Log ratio differential image D is surveyed, obtaining each pixel detection in log ratio differential image D is to change the inspection of class or non-changing class
Survey classification.
Step 8, output detection classification.
Effect of the invention is described further below with reference to emulation experiment.
1, simulated conditions:
Emulation experiment of the invention is Intel Pentium (R) Dual-Core CPU, the memory in dominant frequency 2.30GHz
It is carried out under the hardware environment of 5GB and the software environment of MATLAB R2015a.
Simulation parameter used in emulation experiment of the present invention is as follows:
Missing inspection number: the number of pixels in the region that changes in statistical experiment result figure, with the picture with reference to region of variation in figure
Plain number compares, and with reference to changing in figure but being detected as unchanged number of pixels in experimental result picture, referred to as leaks
Examine number FN.
Erroneous detection number: the number of pixels in the region that do not change in statistical experiment result figure, and with reference to region of variation non-in figure
Number of pixels compare, with reference to do not change in figure but be detected as in experimental result picture variation number of pixels, claim
For erroneous detection number FP.
Total error number/the total pixel number of accuracy PCC:PCC=1-.
Measure testing result figure and the Kappa coefficient with reference to figure consistency:
Wherein, accuracy PCC indicates actual concordance rate, the concordance rate of PRE representation theory.
2, emulation content and interpretation of result:
Emulation experiment of the invention has used three groups of true SAR image data and corresponding variation detection with reference to figure, imitates
The experimental image data used in true experiment are as follows:
The first group of true SAR image data and corresponding variation that emulation experiment of the present invention uses are detected with reference to figure
The SAR image in the area Bern, as shown in Fig. 2, it is the area in April, 1999 Bern that image size, which is 301 × 301, Fig. 2 (a),
SAR image, Fig. 2 (b) are the SAR images in the area in May, 1999 Bern, and Fig. 2 (c) is the corresponding variation detection reference in the area Bern
Figure.
The second group of true SAR image data and corresponding variation that emulation experiment of the present invention uses are detected with reference to figure
The SAR image in the area Ottawa, as shown in figure 3, it is the area in May, 1997 Ottawa that image size, which is 290 × 350, Fig. 3 (a),
SAR image, Fig. 3 (b) is the SAR image in the area the Ottawa of in August, 1997, and Fig. 3 (c) is the corresponding variation inspection in the area Ottawa
It surveys with reference to figure.
It is yellow that the true SAR image data of the third group that emulation experiment of the present invention uses and corresponding variation, which are detected with reference to figure,
The SAR image in river estuary area, as shown in figure 4, image size be 306 × 291, Fig. 4 (a) be in June, 2008 the Yellow River enter sea
The SAR image in mouthful area, Fig. 4 (b) is the SAR image in June, 2009 the Yellow River estuary area, and Fig. 4 (c) is the Yellow River estuary
Area changes detection with reference to figure accordingly.
Emulation experiment of the invention is using broad sense KI threshold value GKI method, fuzzy local message C mean cluster FLICM method
With using the method for the present invention, detection is changed to the area Bern, the area Ottawa and the Yellow River estuary area SAR image respectively
Testing result compare.
Emulation experiment one: using broad sense KI threshold value GKI method, fuzzy local message C mean cluster FLICM method and use
The variation testing result that the method for the present invention obtains as shown in Fig. 2 (d) to 2 (f), is shown in the specific comparative analysis of testing result respectively
Table 1.From the visual effect of Fig. 2 can be seen that using testing result figure of the invention with it is closest with reference to figure.It can be with by table 1
Find out, the pixel number of false retrieval of the present invention is 401 and 497 fewer than GKI and FLICM respectively, and total erroneous pixel number also divides
183 and 278 not fewer than the two, Kappa coefficient also distinguishes high by 2.69% and 5.36% than the two.
1 area Bern of table changes testing result
| Method | Missing inspection pixel number | False retrieval pixel number | Total erroneous pixel number | It detects accuracy (%) | Kappa coefficient (%) |
| GKI | 56 | 513 | 569 | 99.37 | 79.13 |
| FLICM | 55 | 609 | 664 | 99.27 | 76.46 |
| The present invention | 274 | 112 | 386 | 99.57 | 81.82 |
Emulation experiment two: using broad sense KI threshold value GKI method, fuzzy local message C mean cluster FLICM method and use
The variation testing result that the method for the present invention obtains as shown in Fig. 3 (d) to 3 (f), is shown in the specific comparative analysis of testing result respectively
Table 2.From the visual effect of Fig. 3 can be seen that using testing result figure of the invention with it is closest with reference to figure.It can be with by table 2
Find out, the pixel number ratio GKI of false retrieval of the present invention lacked 1189, the pixel number of missing inspection fewer than GKI and FLICM respectively 1814
It is a and 1493, and total erroneous pixel number is 3003 and 415 fewer than the two respectively, Kappa coefficient is also distinguished than the two
High 11.00% and 3.98%.
2 area Ottawa of table changes testing result
| Method | Missing inspection pixel number | False retrieval pixel number | Total erroneous pixel number | It detects accuracy (%) | Kappa coefficient (%) |
| GKI | 2962 | 2391 | 5353 | 94.73 | 80.29 |
| FLICM | 2641 | 124 | 2765 | 97.28 | 87.31 |
| The present invention | 1148 | 1202 | 2350 | 97.68 | 91.29 |
Emulation experiment three: using broad sense KI threshold value GKI method, fuzzy local message C mean cluster FLICM method and use
The variation testing result that the method for the present invention obtains as shown in Fig. 4 (d) to 4 (f), is shown in the specific comparative analysis of testing result respectively
Table 3.From the visual effect of Fig. 4 can be seen that using testing result figure of the invention with it is closest with reference to figure.It can be with by table 3
Find out, the pixel number of false retrieval of the present invention is 2392 and 30 fewer than GKI and FLICM respectively, and the pixel number of missing inspection compares GKI respectively
Lack 2037 and 88 with FLICM, and total erroneous pixel number is 4429 and 118 fewer than the two respectively, Kappa system
Number also distinguishes high by 44.27% and 0.79% than the two.
3 the Yellow River estuary area of table changes testing result
| Method | Missing inspection pixel number | False retrieval pixel number | Total erroneous pixel number | It detects accuracy (%) | Kappa coefficient (%) |
| GKI | 2988 | 2836 | 5824 | 93.46 | 41.00 |
| FLICM | 1039 | 474 | 1513 | 98.30 | 84.48 |
| The present invention | 951 | 444 | 1395 | 98.43 | 85.27 |
Claims (3)
1. a kind of SAR image change detection based on deep learning and SIFT feature, includes the following steps:
(1) SAR image is read in:
Read in SAR image I and J that two width of areal difference phase have been registrated and have corrected；
(2) it normalizes:
According to the following formula, SAR image I and J are normalized, the SAR image after being normalized:
Wherein, I' indicates the SAR image after SAR image I normalization, and min expression is minimized operation, and max expression is maximized
Operation, J' indicate the SAR image after SAR image J normalization；
(3) training characteristics are constructed:
(3a) converts SIFT method using shift-invariant operator, extracts the translation of SAR image I' and J' after the normalization of two width respectively
Invariant features convert SIFT feature S1And S2；
(3b) converts SIFT feature S to two groups of shift-invariant operators1And S2Carry out cascade operation, the feature S after being cascaded；
(3c) carries out dimensionality reduction to the feature S after cascade, using principal component analysis PCA algorithm, the feature S' after obtaining dimensionality reduction；
(4) the feature S' after dimensionality reduction is input in deep neural network, training deep neural network；
(5) according to the following formula, the log ratio differential image of two width SAR images of reading is calculated:
Wherein, D indicates the log ratio differential image for the two width SAR images read in, and log expression is derived from right log operations, | |
Expression takes absolute value operation, and I and J respectively indicate the SAR image of reading；
(6) the neighborhood characteristics sample matrix of log ratio differential image D is constructed:
(6a) uses neighborhood characteristics extracting method, extracts each picture from the pixel matrix that log ratio differential image D is constituted
The neighborhood characteristics vector of vegetarian refreshments；
Log ratio differential image D all pixels neighborhood of a point feature vector is formed M × N-dimensional neighborhood characteristics sample by (6b)
This matrix, wherein M indicates that the sum of all pixels point in log ratio differential image D, N indicate in log ratio differential image D
The dimension of the neighborhood characteristics vector of each pixel；
(7) the variation testing result of log ratio differential image D is obtained:
The neighborhood characteristics sample matrix of log ratio differential image D is input in trained deep neural network, is obtained pair
Each pixel detection is to change the detection classification of class or non-changing class in number ratio difference image D；
(8) output detection classification.
2. the SAR image change detection according to claim 1 based on deep learning and SIFT feature, feature exist
In: specific step is as follows for training deep neural network described in step (4):
The first step initializes deep neural network, and the hidden layer number of plies of deep neural network is 3, deep neural network hidden layer
Interstitial content be respectively 250,150,100, each hidden layer of deep neural network is a limited Boltzmann machine RBM；
Second step is trained the feature S ' after dimensionality reduction using deep neural network, obtains each hidden layer and is limited Bohr hereby
The weight and biasing of graceful machine RBM；
Third step, to limited Boltzmann machine RBM training 50 times of each hidden layer；
4th step is finely adjusted deep neural network using the back-propagation algorithm based on minimum cross entropy, is trained
Deep neural network.
3. the SAR image change detection according to claim 1 based on deep learning and SIFT feature, feature exist
In: specific step is as follows for neighborhood characteristics extracting method described in step (6a):
The first step, choosing a size on disparity map D is n × n-pixel sliding window, by selected window all pixels point
Value pulls into the feature vector of a 1 × N-dimensional, wherein n is the size of sliding window, N=n × n；
Second step, from left to right, successively sliding window from top to bottom, obtain the domain features of each pixel of differential image D to
Amount.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610163983.XA CN105844279B (en) | 2016-03-22 | 2016-03-22 | SAR image change detection based on deep learning and SIFT feature |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610163983.XA CN105844279B (en) | 2016-03-22 | 2016-03-22 | SAR image change detection based on deep learning and SIFT feature |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105844279A CN105844279A (en) | 2016-08-10 |
| CN105844279B true CN105844279B (en) | 2019-04-23 |
Family
ID=56587777
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610163983.XA Active CN105844279B (en) | 2016-03-22 | 2016-03-22 | SAR image change detection based on deep learning and SIFT feature |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105844279B (en) |
Families Citing this family (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10896367B2 (en) * | 2017-03-07 | 2021-01-19 | Google Llc | Depth concatenation using a matrix computation unit |
| CN107230201B (en) * | 2017-05-15 | 2020-11-03 | 西安电子科技大学 | Sample self-calibration ELM-based on-orbit SAR (synthetic aperture radar) image change detection method |
| CN107154054B (en) * | 2017-05-22 | 2020-12-08 | 西安电子科技大学 | Polarization SAR image change detection method based on depth confidence network |
| CN107274392B (en) * | 2017-06-09 | 2019-10-08 | 西安电子科技大学 | Multiple-objection optimization SAR image change detection based on deep belief network |
| CN107358162B (en) * | 2017-06-12 | 2020-04-07 | 西安电子科技大学 | Polarization SAR image change detection method based on depth stack network |
| CN107392940B (en) * | 2017-06-12 | 2020-10-30 | 西安电子科技大学 | SAR image change detection method based on stack semi-supervised self-adaptive denoising self-encoder |
| CN107358261B (en) * | 2017-07-13 | 2020-05-01 | 西安电子科技大学 | High-resolution SAR image change detection method based on curvelet SAE |
| CN107644413B (en) * | 2017-08-25 | 2019-11-01 | 西安电子科技大学 | Based on neighborhood ratio and the SAR image change region detection method learnt from step |
| CN108171119B (en) * | 2017-12-08 | 2020-01-07 | 西安电子科技大学 | SAR image change detection method based on residual error network |
| CN108447057B (en) * | 2018-04-02 | 2021-11-30 | 西安电子科技大学 | SAR image change detection method based on significance and depth convolution network |
| CN108846381B (en) * | 2018-07-06 | 2022-03-18 | 西安电子科技大学 | SAR image change detection method based on maximum edge metric learning |
| CN109343060B (en) * | 2018-12-07 | 2021-01-29 | 电子科技大学 | ISAR imaging method and system based on deep learning time-frequency analysis |
| CN109977968B (en) * | 2019-03-25 | 2021-03-12 | 合肥工业大学 | SAR change detection method based on deep learning classification comparison |
| CN110728665B (en) * | 2019-09-30 | 2023-04-18 | 西安电子科技大学 | SAR image change detection method based on parallel probabilistic neural network |
| CN110765291A (en) * | 2019-10-28 | 2020-02-07 | 广东三维家信息科技有限公司 | Retrieval method and device and electronic equipment |
| CN112464803A (en) * | 2020-11-26 | 2021-03-09 | 泰康保险集团股份有限公司 | Image comparison method and device |
| CN112990046B (en) * | 2021-03-25 | 2023-08-04 | 北京百度网讯科技有限公司 | Differential information acquisition method, related device and computer program product |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102096921A (en) * | 2011-01-10 | 2011-06-15 | 西安电子科技大学 | SAR (Synthetic Aperture Radar) image change detection method based on neighborhood logarithm specific value and anisotropic diffusion |
-
2016
- 2016-03-22 CN CN201610163983.XA patent/CN105844279B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102096921A (en) * | 2011-01-10 | 2011-06-15 | 西安电子科技大学 | SAR (Synthetic Aperture Radar) image change detection method based on neighborhood logarithm specific value and anisotropic diffusion |
Non-Patent Citations (2)
| Title |
|---|
| Difference representation learning using stacked restricted Boltzmann machines for change detection in SAR images;Jia Liu等;《IEEE》;20140921;第20卷(第12期);全文 |
| 基于小波域Fisher分类器的SAR图像变化检测;辛芳芳等;《基于小波域Fisher分类器的SAR图像变化检测》;20110430;第30卷(第2期);全文 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105844279A (en) | 2016-08-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105844279B (en) | SAR image change detection based on deep learning and SIFT feature | |
| Zhan et al. | Log-based transformation feature learning for change detection in heterogeneous images | |
| Gong et al. | Feature learning and change feature classification based on deep learning for ternary change detection in SAR images | |
| Nath et al. | A survey of image classification methods and techniques | |
| CN108319964B (en) | Fire image recognition method based on mixed features and manifold learning | |
| Zhang et al. | Sparse transfer manifold embedding for hyperspectral target detection | |
| Benedek et al. | Change detection in optical aerial images by a multilayer conditional mixed Markov model | |
| Khazai et al. | An approach for subpixel anomaly detection in hyperspectral images | |
| Niemeyer et al. | Change detection using object features | |
| Shi et al. | Discriminative feature learning with distance constrained stacked sparse autoencoder for hyperspectral target detection | |
| CN107516082B (en) | Based on the SAR image change region detection method from step study | |
| Jia et al. | SAR image change detection based on correlation kernel and multistage extreme learning machine | |
| CN109191418A (en) | A kind of method for detecting change of remote sensing image based on contraction self-encoding encoder feature learning | |
| Xu et al. | Generalized morphological component analysis for hyperspectral unmixing | |
| Devi et al. | Change detection techniques–A survey | |
| Kaur et al. | A methodology for the performance analysis of cluster based image segmentation | |
| Huang et al. | Superpixel-based change detection in high resolution sar images using region covariance features | |
| CN107644413B (en) | Based on neighborhood ratio and the SAR image change region detection method learnt from step | |
| CN112784777B (en) | Unsupervised hyperspectral image change detection method based on countermeasure learning | |
| Majidi et al. | A saliency-guided neighbourhood ratio model for automatic change detection of SAR images | |
| Shen et al. | Cropland extraction from very high spatial resolution satellite imagery by object-based classification using improved mean shift and one-class support vector machines | |
| CN111460943A (en) | Remote sensing image ground object classification method and system | |
| Zhang et al. | Region-of-interest extraction based on spectrum saliency analysis and coherence-enhancing diffusion model in remote sensing images | |
| Hong et al. | Random forest fusion classification of remote sensing PolSAR and optical image based on LASSO and IM factor | |
| Akhlaq et al. | Comparative analysis of object-based and pixel-based classification of high-resolution remote sensing images for mapping coral reef geomorphic zones |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |