CN106682621B - SAR image recognition methods based on Bayesian network - Google Patents
SAR image recognition methods based on Bayesian network Download PDFInfo
- Publication number
- CN106682621B CN106682621B CN201611237231.XA CN201611237231A CN106682621B CN 106682621 B CN106682621 B CN 106682621B CN 201611237231 A CN201611237231 A CN 201611237231A CN 106682621 B CN106682621 B CN 106682621B
- Authority
- CN
- China
- Prior art keywords
- work
- cut zone
- sar image
- probability
- pixels
- 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
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/29—Graphical models, e.g. Bayesian networks
Abstract
The SAR image recognition methods based on Bayesian network that the invention discloses a kind of.Mainly solve the problems, such as that target identification type is less in the prior art and needs training sample.Implementation step are as follows: 1., according to the method based on level vision semanteme and the hidden model of adaptive neighborhood multinomial, obtain SAR image segmentation result；2. constructing the corresponding relationship between image grayscale and targeted species according to the segmentation result of SAR image；3. constructing Bayesian network according to the corresponding relationship between image grayscale and targeted species；4. determining the targeted species of cut zone according to Bayesian network.The present invention realizes the good recognition effect of SAR image, can be used for target following.
Description
Technical field
The invention belongs to technical field of image processing, in particular to SAR image recognition methods can be used for image co-registration, match
Quasi- and target following.
Background technique
SAR is a kind of important earth observation means, and carrying out target identification using SAR image has very important meaning
Justice.Typical SAR image recognition methods is the method based on template matching.Mainly first in SAR image in such method
Objective extraction feature, such as shape feature, boundary characteristic, gray feature, resettle the template of target.According to the template, design
The type of similarity measure target for identification.Such method is stronger to the dependence of target template, and flexibility is poor.Work as target
When larger deformation occurs, target cannot be identified well, and the targeted species of such method identification are relatively simple.
In order to improve the effect of SAR image target identification, the SAR image recognition methods based on statistical model is applied and is given birth to.
Such method is not only utilized the feature of SAR image but also considers the spatial context relationship of target.The feature master of SAR image
It is indicated with likelihood probability, is typically expressed as Gaussian Profile, gamma distribution etc..The spatial context relationship of target mainly uses elder generation
Probability is tested to indicate, is typically expressed as Gibbs distribution and multinomial logistic regression function.By by likelihood probability and priori
Probability multiplication calculates the type of target in SAR image.It is extensive since such method has the advantage than other recognition methods
It is identified applied to SAR image.
But the SAR image recognition methods based on statistical model needs a large amount of training data, by largely training number
According to clarification of objective information and contextual information is obtained, target identification is realized.However, being obtained a large amount of for SAR image
Training sample be it is extremely difficult even not possible with, therefore limit such method SAR image identification in application.Separately
Outside, such method is not due to accounting for the causality between the feature of SAR image and targeted species, the target species of identification
Class is also relatively simple.
Summary of the invention
It is an object of the invention to propose a kind of SAR figure based on Bayesian network for above-mentioned existing methods deficiency
As recognition methods, to promote the effect of SAR image identification.
Technical thought of the invention is: by improving to the method based on statistical model, improving SAR image identification
Effect, the i.e. causality of the feature of consideration SAR image and targeted species construct shellfish in the case where not needing training sample
This network of leaf, implementation step are as follows:
(1) method based on level vision semanteme and the hidden model of adaptive neighborhood multinomial is used, SAR image is divided
It cuts, obtains cut zone R={ Ri, i ∈ { 1,2 ..., m }, m represents the total number of cut zone, RiRepresent i-th of cut section
Domain；
(2) according to the segmentation result of SAR image, the corresponding relationship between image grayscale and targeted species is constructed:
(2a) calculates cut zone RiInterior average gray Ai:
Wherein, NiIt is cut zone RiInterior total number of pixels, ysIt is cut zone RiInterior s-th of pixel, s ∈ 1,
2,...,Ni}；
(2b) is according to the average gray A of calculatingi, by cut zone Ri3 gray scale intervals are divided into,
If average gray Ai∈ [0,50), then cut zone RiIt is low gray scale interval；
If average gray Ai∈ [50,175), then cut zone RiIt is middle gray scale interval；
If average gray Ai∈ [175,255], then cut zone RiIt is high gray scale interval；
It is 4 kinds of road, waters, wetland and shade targets that (2c) it is corresponding, which to set low gray scale interval, and middle gray scale interval is corresponding
It is land and 2 kinds of farmland target, corresponding high gray scale interval is forest and regular 2 kinds of targets of made Target group；
(3) according to the corresponding relationship between image grayscale and targeted species, Bayesian network is constructed；
(3a) obtains SAR image sketch map according to SAR image sketch model；
(3b) calculates cut zone R according to the sketch map of SAR imageiFeature di:
Wherein, bjIt is the number of pixels that direction is j, j ∈ K, K={ 0,1 ..., 179 } represent 0 degree to 179 degree totally 180
Direction；
(3c) according to step (2c) and (3b), the first layer for constructing Bayesian network is SAR image, and the second layer is SAR figure
The targeted species of picture, third layer are the feature d of SAR image cut zoneiWith total number of pixels Ni；
(4) according to Bayesian network, the formula of the targeted species of cut zone is constructed:
(4a) defines cut zone RiTargeted species zcAre as follows:
Wherein, z={ zc, c ∈ { 1,2 ..., 8 }, z1,z2,z3,z4,z5,z6,z7And z8Respectively represent road, waters, wet
Ground, shade, land, farmland, forest and rule 8 kinds of target types of made Target group；
(4b) obtains identification cut zone R according to (4a)iThe formula of targeted species:
Wherein, α is normalized parameter, p (di|zc) it is feature diConditional probability, p (Ni|zc) it is number of pixels Ni's
Conditional probability, p (zc) it is prior probability；
(5) according to cut zone R in (4b)iThe formula of targeted species calculates separately cut zone RiBelong to 8 kinds of different mesh
The probability for marking type, takes the maximum value of this 8 probability, and the corresponding targeted species of maximum value are cut zone RiTarget type.
The invention has the following advantages over the prior art:
The first, the information that the present invention has excavated image itself carries out target identification, does not need the training samples information of target,
Reduce the difficulty of SAR image identification.
The second, the present invention constructs Bayesian network according to the corresponding relationship between image grayscale and targeted species, uses shellfish
Low gray scale interval in this network of leaf corresponds to 4 kinds of road, waters, shade and wetland targets, and middle gray scale interval corresponds to land and agriculture
2 kinds of field target, high gray scale interval correspond to 2 kinds of targets of made Target group of forest and rule, which enriches target
Information improves the targeted species of SAR image identification.
Detailed description of the invention
Fig. 1 is implementation flow chart of the invention；
Fig. 2 is the corresponding relationship of image grayscale and targeted species in the present invention；
Fig. 3 is the bayesian network structure figure in the present invention；
Fig. 4 is the SAR image recognition result figure for being 1 meter to Ku wave band resolution ratio with the present invention；
Fig. 5 is the SAR image recognition result figure for being 3 meters to C-band resolution ratio with the present invention.
Specific embodiment
Referring to Fig.1, implementation steps of the invention are as follows:
Step 1, using the method based on level vision semanteme and the hidden model of adaptive neighborhood multinomial, to SAR image into
Row segmentation.
In the prior art, it may be implemented the segmentation of SAR image there are many method, such as markov random file MRF method,
Condition random field CRF method, the hidden model method of multinomial, based on level vision is semantic and the hidden model of adaptive neighborhood multinomial
Method.The present invention was published in IEEE Transactions in 2016 according to Fang-Liu and Yiping Duan et al.
Article " SAR image segmentation based on Geoscience and Remote Sensing magazine
hierarchical visual semantic and adaptive neighborhood multinomial latent
Model, the SAR image segmentation method based on level vision semanteme and the hidden model of adaptive neighborhood multinomial " proposed in mould
Type is split SAR image.
Region in segmentation result with same tag is cut zone R={ Ri, i ∈ { 1,2 ..., m }, m, which are represented, to be divided
Cut the total number in region, RiRepresent i-th of cut zone.
Step 2, according to the segmentation result of SAR image, the corresponding relationship between image grayscale and targeted species is constructed.
(2a) calculates cut zone RiInterior average gray Ai:
Wherein, NiIt is cut zone RiInterior total number of pixels, ysIt is cut zone RiInterior s-th of pixel, s ∈ 1,
2,...,Ni}；
(2b) is according to the average gray A of calculatingi, by cut zone Ri3 gray scale intervals are divided into,
If average gray Ai∈ [0,50), then cut zone RiIt is low gray scale interval；
If average gray Ai∈ [50,175), then cut zone RiIt is middle gray scale interval；
If average gray Ai∈ [175,255], then cut zone RiIt is high gray scale interval；
It is 4 kinds of road, waters, wetland and shade targets that (2c) it is corresponding, which to set low gray scale interval, and middle gray scale interval is corresponding
It is land and 2 kinds of farmland target, corresponding high gray scale interval is forest and regular 2 kinds of targets of made Target group, such as Fig. 2 institute
Show.
Step 3, according to the corresponding relationship between image grayscale and targeted species, Bayesian network is constructed.
(3a) was published in IEEE Transactions in 2014 according to Jie-Wu and Fang-Liu et al.
Article " Local maximal homogenous region on Geoscience and Remote Sensing magazine
search for SAR speckle reduction with sketch-based geometrical kernel
Function " proposed in model obtain SAR image sketch map；
(3b) calculates cut zone R according to the sketch map of SAR imageiFeature di:
Wherein, bjIt is the number of pixels that direction is j, j ∈ K, K={ 0,1 ..., 179 } represent 0 degree to 179 degree totally 180
Direction；
(3c) according to step (2c) and (3b), building first layer is SAR image, and the second layer is the targeted species of SAR image,
Third layer is the feature d of SAR image cut zoneiWith total number of pixels NiBayesian network, as shown in Figure 3.
Step 4, according to Bayesian network, the formula of the targeted species of cut zone is constructed.
(4a) defines cut zone RiTargeted species zcAre as follows:
Wherein, z={ zc, c ∈ { 1,2 ..., 8 }, z1,z2,z3,z4,z5,z6,z7And z8Respectively represent road, waters, wet
Ground, shade, land, farmland, forest and rule 8 kinds of target types of made Target group；
(4b) obtains identification cut zone R according to (4a)iThe formula of targeted species:
Wherein, α is normalized parameter, p (di|zc) it is feature diConditional probability, p (Ni|zc) it is number of pixels Ni's
Conditional probability, p (zc) it is prior probability；
(4c) determines feature diConditional probability p (di|zc):
(4c1) is according to the feature d for counting road in 100 width SAR imagesi, wherein 90 width have feature di, 10 width do not have
Feature di, therefore, work as zcWhen being road, p (d is takeni|zc)=0.9；
(4c2) is according to the feature d for counting waters in 100 width SAR imagesi, wherein 40 width have feature di, 60 width do not have
Feature di, therefore, work as zcWhen being waters, p (d is takeni|zc)=0.4；
(4c3) is according to the feature d for counting wetland in 100 width SAR imagesi, wherein 20 width have feature di, 80 width do not have
Feature di, therefore, work as zcWhen being wetland, p (d is takeni|zc)=0.2；
(4c4) is according to the feature d for counting shade in 100 width SAR imagesi, wherein 20 width have feature di, 80 width do not have
Feature di, therefore, work as zcWhen being shade, p (d is takeni|zc)=0.2；
(4c5) is according to the feature d for counting land in 100 width SAR imagesi, wherein 40 width have feature di, 60 width do not have
Feature di, therefore, work as zcWhen being land, p (d is takeni|zc)=0.4；
(4c6) is according to the feature d for counting farmland in 100 width SAR imagesi, wherein 80 width have feature di, 20 width do not have
Feature di, therefore, work as zcWhen being farmland, p (d is takeni|zc)=0.8；
(4c7) is according to the feature d for counting forest in 100 width SAR imagesi, wherein 10 width have feature di, 90 width do not have
Feature di, therefore, work as zcWhen being forest, p (d is takeni|zc)=0.1；
(4c8) is according to the feature d for counting made Target group regular in 100 width SAR imagesi, wherein 90 width have feature
di, 10 width are without feature di, therefore, work as zcWhen being the made Target group of rule, p (d is takeni|zc)=0.9；
(4d) determines number of pixels NiConditional probability p (Ni|zc):
(4d1) is according to the number of pixels N for counting road in 100 width SAR imagesi, wherein 90 width have number of pixels Ni, 10
Width does not have number of pixels Ni, therefore, work as zcWhen being road, p (N is takeni|zc)=0.9；
(4d2) is according to the number of pixels N for counting waters in 100 width SAR imagesi, wherein 90 width have number of pixels Ni, 10
Width does not have number of pixels Ni, therefore, work as zcWhen being waters, p (N is takeni|zc)=0.9；
(4d3) is according to the number of pixels N for counting wetland in 100 width SAR imagesi, wherein 60 width have number of pixels Ni, 40
Width does not have number of pixels Ni, therefore, work as zcWhen being wetland, p (N is takeni|zc)=0.6；
(4d4) is according to the number of pixels N for counting shade in 100 width SAR imagesi, wherein 10 width have number of pixels Ni, 90
Width does not have number of pixels Ni, therefore, work as zcWhen being shade, p (N is takeni|zc)=0.1；
(4d5) is according to the number of pixels N for counting land in 100 width SAR imagesi, wherein 80 width have number of pixels Ni, 20
Width does not have number of pixels Ni, therefore, work as zcWhen being land, p (N is takeni|zc)=0.8；
(4d6) is according to the number of pixels N for counting farmland in 100 width SAR imagesi, wherein 80 width have number of pixels Ni, 20
Width does not have number of pixels Ni, therefore, work as zcWhen being farmland, p (N is takeni|zc)=0.8；
(4d7) is according to the number of pixels N for counting forest in 100 width SAR imagesi, wherein 90 width have number of pixels Ni, 10
Width does not have number of pixels Ni, therefore, work as zcWhen being forest, p (N is takeni|zc)=0.9；
(4d8) is according to the number of pixels N for counting made Target group regular in 100 width SAR imagesi, wherein 90 width have
Number of pixels Ni, 10 width are without number of pixels Ni, therefore, work as zcWhen being the made Target group of rule, p (N is takeni|zc)=0.9；
(4e) determines prior probability p (zc):
(4e1) is according to 100 width SAR images of statistics, wherein 25 width include therefore road works as zcWhen being road, p (z is takenc)=
0.25；
(4e2) is according to 100 width SAR images of statistics, wherein 25 width include therefore z is worked as in waterscWhen being waters, p (z is takenc)=
0.25；
(4e3) is according to 100 width SAR images of statistics, wherein 25 width include therefore wetland works as zcWhen being wetland, p (z is takenc)=
0.25；
(4e4) is according to 100 width SAR images of statistics, wherein 25 width include therefore shade works as zcWhen being shade, p (z is takenc)=
0.25；
(4e5) is according to 100 width SAR images of statistics, wherein 50 width include therefore z is worked as in farmlandcWhen being farmland, p (z is takenc)=
0.5；
(4e6) is according to 100 width SAR images of statistics, wherein 50 width include therefore z is worked as in landcWhen being land, p (z is takenc)=
0.5；
(4e7) is according to 100 width SAR images of statistics, wherein 50 width include therefore forest works as zcWhen being forest, p (z is takenc)=
0.5；
(4e7) is according to 100 width SAR images of statistics, wherein therefore the made Target group that 50 width include rule works as zcIt is rule
When made Target group then, p (z is takenc)=0.5；
(4f) is according to feature d obtained aboveiConditional probability p (di|zc), number of pixels NiConditional probability p (Ni|
zc), prior probability p (zc), calculate normalized parameter α:
Wherein, p (~zc) be and p (zc) complementary probability, p (~zc)=1-p (zc)；p(di) it is feature diWhat is occurred is general
Rate, according to a width SAR image its with feature diWith do not have feature diThe equal characteristic of probability, take p (di)=0.5；p
(Ni) it is number of pixels NiThe probability of appearance, according to a width SAR image its with number of pixels NiWith do not have number of pixels NiGenerally
The equal characteristic of rate, takes p (Ni)=0.5.
Step 5, according to step 4, the final recognition result of SAR image is obtained.
To each cut zone Ri, calculate separately the probability that the cut zone belongs to 8 kinds of targeted species:
Cut zone RiBelong to road z1Probability p (z1|diNi)；
Cut zone RiBelong to waters z2Probability p (z2|diNi)；
Cut zone RiBelong to wetland z3Probability p (z3|diNi)；
Cut zone RiBelong to shade z4Probability p (z4|diNi)；
Cut zone RiBelong to land z5Probability p (z5|diNi)；
Cut zone RiBelong to farmland z6Probability p (z6|diNi)；
Cut zone RiBelong to forest z7Probability p (z7|diNi)；
Cut zone RiBelong to the made Target group z of rule8Probability p (z8|diNi)；
The maximum value of this 8 probability is taken, the corresponding targeted species of maximum value are cut zone RiTargeted species.For example,
Maximum value is cut zone RiBelong to road z1Probability p (z1|diNi), then cut zone RiTargeted species be road.
Effect of the invention is further illustrated by the data and image of following emulation.
1. simulated conditions
The hardware condition that the present invention emulates are as follows: Intellisense and image understanding laboratory graphics workstation；
The present invention emulate used in SAR image are as follows: the SAR image and C-band resolution ratio that Ku wave band resolution ratio is 1 meter be
3 meters of SAR image.
2. emulation content and result
Emulation 1: being identified with the SAR image that the present invention is 1 meter to Ku wave band resolution ratio, as a result such as Fig. 4, wherein Fig. 4
It (a) is original SAR image that Ku wave band resolution ratio is 1 meter, Fig. 4 (b) is SAR image recognition result figure.
Emulation 2: being identified with the SAR image that the present invention is 3 meters to C-band resolution ratio, as a result such as Fig. 5, wherein Fig. 5
It (a) is original SAR image that C-band resolution ratio is 3 meters, Fig. 5 (b) is SAR image recognition result figure.
From Fig. 4 (b) as can be seen that the present invention has identified the made Target group of wetland, land, forest and rule.
From Fig. 5 (b) as can be seen that the present invention has identified the made Target group of land, road and rule.
To sum up, the present invention does not need training sample and the targeted species of identification are more, improves the effect of image recognition.
Claims (5)
1. the SAR image recognition methods based on Bayesian network, comprising:
(1) method based on level vision semanteme and the hidden model of adaptive neighborhood multinomial is used, SAR image is split,
Obtain cut zone R={ Ri, i ∈ { 1,2 ..., m }, m represents the total number of cut zone, RiRepresent i-th of cut zone；
(2) according to the segmentation result of SAR image, the corresponding relationship between image grayscale and targeted species is constructed:
(2a) calculates cut zone RiInterior average gray Ai:
Wherein, NiIt is cut zone RiInterior total number of pixels, ysIt is cut zone RiInterior s-th of pixel, s ∈ 1,2 ...,
Ni}；
(2b) is according to the average gray A of calculatingi, by cut zone Ri3 gray scale intervals are divided into,
If average gray Ai∈ [0,50), then cut zone RiIt is low gray scale interval；
If average gray Ai∈ [50,175), then cut zone RiIt is middle gray scale interval；
If average gray Ai∈ [175,255], then cut zone RiIt is high gray scale interval；
It is 4 kinds of road, waters, wetland and shade targets that (2c) it is corresponding, which to set low gray scale interval, and corresponding middle gray scale interval is land
Ground and 2 kinds of farmland target, corresponding high gray scale interval is forest and regular 2 kinds of targets of made Target group；
(3) according to the corresponding relationship between image grayscale and targeted species, Bayesian network is constructed；
(3a) obtains SAR image sketch map according to SAR image sketch model；
(3b) calculates cut zone R according to the sketch map of SAR imageiFeature di:
Wherein, bjIt is the number of pixels that direction is j, j ∈ K, K={ 0,1 ..., 179 } represent 0 degree to 179 degree totally 180 direction；
(3c) according to step (2c) and (3b), the first layer for constructing Bayesian network is SAR image, and the second layer is SAR image
Targeted species, third layer are the feature d of SAR image cut zoneiWith total number of pixels Ni；
(4) according to Bayesian network, the formula of the targeted species of cut zone is constructed:
(4a) defines cut zone RiTargeted species zcAre as follows:
Wherein, z={ zc, c ∈ { 1,2 ..., 8 }, z1,z2,z3,z4,z5,z6,z7And z8Respectively represent road, waters, wetland,
Shade, land, farmland, forest and rule 8 kinds of target types of made Target group；
(4b) obtains identification cut zone R according to (4a)iThe formula of targeted species:
Wherein, α is normalized parameter, p (di|zc) it is feature diConditional probability, p (Ni|zc) it is number of pixels NiCondition
Probability, p (zc) it is prior probability；
(5) according to cut zone R in (4b)iThe formula of targeted species calculates separately cut zone RiBelong to 8 kinds of different target kinds
The probability of class, takes the maximum value of this 8 probability, and the corresponding targeted species of maximum value are cut zone RiTarget type.
2. method according to claim 1, the wherein feature d in step (4b)iConditional probability p (di|zc), it is according to system
Count the feature d of 8 kinds of different target types in 100 width SAR imagesi, it obtains such as lower probability:
Work as zcWhen being road, p (di|zc)=0.9；
Work as zcWhen being waters, p (di|zc)=0.4；
Work as zcWhen being wetland, p (di|zc)=0.2；
Work as zcWhen being shade, p (di|zc)=0.2；
Work as zcWhen being land, p (di|zc)=0.4；
Work as zcWhen being farmland, p (di|zc)=0.8；
Work as zcWhen being forest, p (di|zc)=0.1；
Work as zcWhen being the made Target group of rule, p (di|zc)=0.9.
3. method according to claim 1, the wherein number of pixels N in step (4b)iConditional probability p (Ni|zc), it is root
According to statistics in 100 width SAR images 8 kinds of different target types number of pixels Ni, it obtains such as lower probability:
Work as zcWhen being road, p (Ni|zc)=0.9；
Work as zcWhen being waters, p (Ni|zc)=0.9；
Work as zcWhen being wetland, p (Ni|zc)=0.6；
Work as zcWhen being shade, p (Ni|zc)=0.1；
Work as zcWhen being land, p (Ni|zc)=0.8；
Work as zcWhen being farmland, p (Ni|zc)=0.8；
Work as zcWhen being forest, p (Ni|zc)=0.9；
Work as zcWhen being the made Target group of rule, p (Ni|zc)=0.9.
4. method according to claim 1, the wherein prior probability p (z in step (4b)c), it is according to statistics 100 width SAR figure
As the number that 8 kinds of different target types occur, obtain such as lower probability:
Work as zcWhen being road, p (zc)=0.25；
Work as zcWhen being waters, p (zc)=0.25；
Work as zcWhen being wetland, p (zc)=0.25；
Work as zcWhen being shade, p (zc)=0.25；
Work as zcWhen being farmland, p (zc)=0.5；
Work as zcWhen being land, p (zc)=0.5；
Work as zcWhen being forest, p (zc)=0.5；
Work as zcWhen being the made Target group of rule, p (zc)=0.5.
5. method according to claim 1, the wherein normalized parameter alpha in step (4b), is calculated by following formula:
Wherein, p (~zc) be and p (zc) complementary probability, p (~zc)=1-p (zc)；p(di) it is feature diThe probability of appearance, p
(di)=0.5；p(Ni) it is number of pixels NiThe probability of appearance, p (Ni)=0.5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611237231.XA CN106682621B (en) | 2016-12-28 | 2016-12-28 | SAR image recognition methods based on Bayesian network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611237231.XA CN106682621B (en) | 2016-12-28 | 2016-12-28 | SAR image recognition methods based on Bayesian network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106682621A CN106682621A (en) | 2017-05-17 |
| CN106682621B true CN106682621B (en) | 2019-01-29 |
Family
ID=58871974
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611237231.XA Active CN106682621B (en) | 2016-12-28 | 2016-12-28 | SAR image recognition methods based on Bayesian network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106682621B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110222700A (en) * | 2019-05-30 | 2019-09-10 | 五邑大学 | SAR image recognition methods and device based on Analysis On Multi-scale Features and width study |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8977062B2 (en) * | 2013-02-25 | 2015-03-10 | Raytheon Company | Reduction of CFAR false alarms via classification and segmentation of SAR image clutter |
| CN103713288B (en) * | 2013-12-31 | 2015-10-28 | 电子科技大学 | Sparse Bayesian reconstruct linear array SAR formation method is minimized based on iteration |
| CN104537384B (en) * | 2015-01-21 | 2017-09-01 | 西安电子科技大学 | A kind of SAR target discrimination methods of combination likelihood ratio judgement |
-
2016
- 2016-12-28 CN CN201611237231.XA patent/CN106682621B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN106682621A (en) | 2017-05-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110097101B (en) | Remote sensing image fusion and coastal zone classification method based on improved reliability factor | |
| CN109871902B (en) | SAR small sample identification method based on super-resolution countermeasure generation cascade network | |
| EP4158540A1 (en) | Method for property feature segmentation | |
| CN102073879B (en) | Method for identifying characteristic land categories of ocean remote sensing images of coast on basis of semi-supervised learning | |
| CN101799921B (en) | Cloud detection method of optic remote sensing image | |
| CN110765912B (en) | SAR image ship target detection method based on statistical constraint and Mask R-CNN | |
| CN103971115A (en) | Automatic extraction method for newly-increased construction land image spots in high-resolution remote sensing images based on NDVI and PanTex index | |
| CN104794730B (en) | SAR image segmentation method based on super-pixel | |
| CN102314610B (en) | Object-oriented image clustering method based on probabilistic latent semantic analysis (PLSA) model | |
| CN104346814B (en) | Based on the SAR image segmentation method that level vision is semantic | |
| CN112766089B (en) | Cross-domain road extraction method based on global-local confrontation learning framework | |
| CN102542293A (en) | Class-I extraction and classification method aiming at high-resolution SAR (Synthetic Aperture Radar) image scene interpretation | |
| CN103871039A (en) | Generation method for difference chart in SAR (Synthetic Aperture Radar) image change detection | |
| Veeravasarapu et al. | Adversarially tuned scene generation | |
| Tang et al. | A multiple-point spatially weighted k-NN method for object-based classification | |
| CN109712149A (en) | A kind of image partition method based on wavelet energy and fuzzy C-mean algorithm | |
| CN104408731B (en) | Region graph and statistic similarity coding-based SAR (synthetic aperture radar) image segmentation method | |
| CN107992856A (en) | High score remote sensing building effects detection method under City scenarios | |
| CN104102928A (en) | Remote sensing image classification method based on texton | |
| Li et al. | Unsupervised road extraction via a Gaussian mixture model with object-based features | |
| Ko et al. | Effective training strategies for deep-learning-based precipitation nowcasting and estimation | |
| Pan et al. | An adaptive multifeature method for semiautomatic road extraction from high-resolution stereo mapping satellite images | |
| Wang et al. | Land contained sea area ship detection using spaceborne image | |
| Song et al. | Small UAV-based multi-temporal change detection for monitoring cultivated land cover changes in mountainous terrain | |
| Zhang et al. | Analyzing land use and land cover change patterns and population dynamics of fast-growing US cities: Evidence from Collin County, Texas |
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 |