CN109740475A - A kind of remote sensing images ground scene classification method - Google Patents
A kind of remote sensing images ground scene classification method Download PDFInfo
- Publication number
- CN109740475A CN109740475A CN201811594292.0A CN201811594292A CN109740475A CN 109740475 A CN109740475 A CN 109740475A CN 201811594292 A CN201811594292 A CN 201811594292A CN 109740475 A CN109740475 A CN 109740475A
- Authority
- CN
- China
- Prior art keywords
- polarization
- scattering
- degree
- follows
- membership
- 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
Landscapes
- Radar Systems Or Details Thereof (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
A kind of remote sensing images ground scene classification method, comprising the following steps: be directed to polarimetric SAR image, construct its polarization covariance matrix and polarization coherence matrix；Polarization coherence matrix is carried outIt decomposes, obtains Polarization scattering entropy H, degree of anisotropy A and average scattering angleCalculate the span image of polarimetric SAR image, and designed image adaptive neighborhood on this basis；Based on image adaptive neighborhood, self-adaptive averaging factor is designed, and constructs MRF priori energy function；By the spatial correlation information based on MRF in conjunction with polarization information, is classified using fuzzy C-mean algorithm and construct objective function, iteratively solved optimal degree of membership and class categories center, polarimetric SAR image is made to be referred to the classification with maximum membership degree pixel-by-pixel.The present invention is able to suppress the noise phenomenon in classification results, and can preferably avoid smoothing effect, retaining space detailed information.
Description
Technical field
The invention belongs to remote sensing image interpretation fields, and in particular to a kind of remote sensing images ground scene classification method is based on
The spatial correlation characteristic of pixel in neighborhood establishes the prior probability of classification using Markov model in specific neighborhood, and
Combining space information carries out remote sensing image classification on this basis, to provide classification for analysis of image content and scene classification
Attribute information.
Background technique
Classification of Polarimetric SAR Image is research contents mostly important in remote sensing image interpretation.In polarization SAR application, very
It needs to divide an image into different regions in more situations.For example, need to distinguish different agrotypes in crops assessment,
In geographic mapping, urban planning application, need to distinguish different types of atural object.The result of classification can be also used for target detection
With the fields such as identification and variation detection, can also be provided in real time accurately directly as final output for user and policymaker
Classification and information.The characteristics of in view of SAR image coherent speckle noise, analysis and processing side based on region/object
Method has become the hot spot of area based on polarimetric SAR image interpretation, and it is then subsequent for obtaining accurately and reliably Target scalar classification results
The key of processing and application.Classification of Polarimetric SAR Image method based on statistical model is the main contents of current research, such side
Method consider polarization SAR data statistics randomness, constructed using the statistical property of coherent speckle noise specific distance function or
Similarity measurement.In middle low resolution, polarization SAR data, which always assume that, meets Gaussian statistics, this statistical
Cloth can preferably describe the back scattering of distributed object and homogeneous area.At this stage, SAR and polarization SAR system can be real
Existing high-resolution imaging, resolution ratio have reached meter level even sub-meter grade.However the raising of spatial resolution and the appearance of texture
So that the statistical property of polarization SAR data is changed, shows apparent non-Gaussian feature.In addition, in the nonuniformitys area such as city
There is the coherent scattering mixed and incoherent scattering in domain, the backscattering characteristic in the region is caused to be very difficult to model.This is to polarization
The atural object Accurate classification of SAR image brings new challenge, and traditional method based on Gaussian statistics cannot be satisfied with
Result.Therefore, carrying out exact classification to high-resolution polarimetric SAR data is worth further further investigation.
Conventional Classification of Polarimetric SAR Image method generally only uses the statistical property and/or scattering properties of single pixel, and
Ignore the spatial information (si) of image.Spatial information (si) refers to that image pixel forms letter provided by certain natural mode in spatial domain
Breath mainly includes the information such as spatial coherence, texture, edge, region.Since target and the architectural characteristic of atural object are often embodied in
In spatial distribution in figure, therefore spatial information (si) can provide auxiliary information for Classification of Polarimetric SAR Image and segmentation, effectively make up list
The deficiency of a pixel information contained.For example, the spatial autocorrelation information in neighborhood can effectively inhibit the image of coherent speckle noise,
Remove " spiced salt " noise effect of final classification result.Texture information, marginal information can be combined with Polarization scattering information, be used for
Improve the precision of SAR remote sensing image classification.The spatial information (si) in polarimetric SAR image is sufficiently excavated, and it is special with polarization statistics
Property and scattering properties effectively combined, for promoted image interpretation especially terrain classification effect be of great significance.Together
When, the research of this respect still needs to be broken through there are many difficult point.Markov random field (Markov Random Field, MRF) is theoretical
It is the basic skills for introducing spatial coherence.It is relied on based on Markov it is assumed that MRF considers surrounding pixel to the shadow of current pixel
It rings, the prior probability of pixel class is constructed by bayesian theory.Wherein, iterated conditional modes (Iterated
Conditional Modes, ICM) it is a kind of most common MRF method, this method seeks locally optimal solution instead of global optimum
Solution can simplify calculating, and have preferable robustness.In SAR and Classification of Polarimetric SAR Image, obtained using the method for MRF
More smooth result.However, traditional MRF analysis frequently resulted in smooth phenomenon using fixed neighbour structure, it is some
Spatial detail feature thickens.In order to solve this problem, some adaptive M RF (Adaptive MRF, AMRF) methods are adopted
With adaptive neighborhood and parameter, it can preferably retain the minutia of image, and condition random field (Conditional
Random Field, CRF) be MRF a kind of variant, SAR and polarimetric SAR image processing in achieve certain application.
Summary of the invention
It is an object of the invention to be directed to above-mentioned the problems of the prior art, a kind of remote sensing images ground scene classification is provided
Method is analyzed and is studied to polarimetric SAR image Polarization target decomposition emphatically, establishes class in specific neighborhood by MRF
Other prior probability, influence caused by overcoming classification boundaries uncertain are effective to promote terrain classification effect.
To achieve the goals above, the technical solution adopted by the present invention the following steps are included:
For polarimetric SAR image, its polarization covariance matrix and polarization coherence matrix are constructed；
Polarization coherence matrix is carried outIt decomposes, obtains Polarization scattering entropy H, degree of anisotropy A and average scattering angle
Calculate the span image of polarimetric SAR image, and designed image adaptive neighborhood on this basis；
Based on image adaptive neighborhood, self-adaptive averaging factor is designed, and constructs MRF priori energy function；
By the spatial correlation information based on MRF in conjunction with polarization information, is classified using fuzzy C-mean algorithm and constructs objective function,
Optimal degree of membership and class categories center are iteratively solved, polarimetric SAR image is made to be referred to the class with maximum membership degree pixel-by-pixel
Not.
Constructing its polarization covariance matrix and polarization coherence matrix, detailed process is as follows:
Polarization radar collision matrix is expressed asWherein, H and V respectively indicates horizontal polarization and vertical
Polarization, SPQ(P, Q=H, V) is
Number, then has S in the back scattering for meeting reciprocal theoremHV=SVH；The expression formula of Pauli Scattering of Vector is as follows:
In formula, subscriptTRepresenting matrix transposition；
The expression formula of polarization coherence matrix is as follows:
In formula, subscriptThe conjugate transposition of representing matrix, subscript*Indicate complex conjugate；
The expression formula of polarization covariance matrix is as follows:
Wherein,
Polarization coherence matrix is carried outDetailed process is as follows for decomposition:
By carrying out characteristic value or eigendecomposition to coherence matrix T, a simple statistical model is constructed, T is expanded
Exhibition is the sum of the response of three pinpoint targets, each target corresponds to a scattering mechanism, by unit character vector uiIt determines,
And corresponding eigenvalue λiRepresent specific gravity of the scattering mechanism in entire scattering process；The expression formula of decomposable process are as follows:
Polarization scattering entropy H characterization decomposes unordered degree of the obtained different scattering types in statistical significance, and expression formula is such as
Under:
Wherein
Wherein, PiThe pseudo- probability that type occurs is scattered for i-th kind, then the value range for the entropy H that polarizes is 0≤H≤1；
Polarisation Anisotropy degree A is defined to show λ2And λ3Between relationship, expression formula are as follows:
In formula, the value range of A is 0≤A≤1；
Average scattering angleIs defined as:
Work as αiAt=0 °, target scattering corresponds to surface scattering, withIt is gradually increased, scattering mechanism becomes Prague table
Area scattering；
Work as αi=45 ° then represent dipole scattering, and then scattering type becomes the rescattering of two dielectric surfaces,
Work as αiBecome the dihedral angle scattering of metal surface under=90 ° of extreme case.
The back scattering general power span calculating formula of area based on polarimetric SAR image are as follows:
Span=| Shh|2+2|Shv|2+|Svv|2=Tr (C)=Tr (T) chooses the neighbour in span image with minimum variance
Domain carries out Markov model analysis, and the smoothing effect of MRF is limited in selected neighborhood.
Based on selected neighborhood ηsAnd adaptive smoothing factor ψsrIt constructs MRF priori energy function U (), expression formula
It is as follows:
Wherein, | ηs| and bsThe respectively size of neighborhood space and homogeneous measurement；
Smoothing factor ψsrCalculation formula it is as follows:
Wherein, LcIt is the current class of pixel s,Indicate that the pixel belongs to classification LcFuzzy degree of membership.
According to being defined onTwo dimension fuzzy ownership function in plane obtains the value of initially fuzzy degree of membershipIt indicates
Pixel s belongs to the scale factor of classification k, and meetsInitial cluster centre is calculated on this basisThen it is iterated and optimizes using value of the Fuzzy C-Means Cluster Algorithm to fuzzy degree of membership.
The step of iteration, is as follows:
Firstly, calculating coherence matrix < T of pixel ss> and classification k cluster centreBetween Wishart distance:
Based on obtained Wishart distance, fuzzy degree of membership is updated:
Wherein, mcIt is a constant, ρ () is Huber function:
Pass through the fuzzy degree of membership in conjunction with spatial correlation information, generating enhancing:
Wherein,The prior probability extracted is analyzed by local AMRF；
Finally, the fuzzy degree of membership based on enhancing, is updated cluster centre:
Compared with prior art, the present invention have it is following the utility model has the advantages that by by the spatial correlation information based on MRF with
Polarization information combines, and is classified using fuzzy C-mean algorithm and constructs objective function, iteratively solves optimal degree of membership and class categories center,
Polarimetric SAR image is finally set to be referred to the classification with maximum membership degree pixel-by-pixel.The fuzzy C-mean algorithm classification method combines certainly
Adapt to neighborhood MRF and SAR image polarization information, the spatial correlation information that local auto-adaptive neighborhood MRF is analyzed with
Fuzzy degree of membership based on polarization information combines, and inhibits the noise phenomenon in classification results.The present invention uses adaptive neighbour
Domain, while self-adaptive averaging factor is proposed, so as to preferably avoid smoothing effect, retaining space detailed information.
Detailed description of the invention
The schematic diagram of Fig. 1 MRF adaptive neighborhood；
The flow chart of Fig. 2 classification method of the present invention；
Fig. 3 distinct methods cover schematic diagram for AIR data classification result and ground truth data:
(a)- Wishart classifier classification results schematic diagram；(b)- Wishart classifier classification results ground
Live data covers schematic diagram；
(c) it obscuresClassifier classification results schematic diagram；(d) it obscuresClassifier classification results ground truth data
Cover schematic diagram；
(e) Potts-MRF classification results schematic diagram；(f) Potts-MRF classification results ground truth data covers signal
Figure；
(g) adaptive M RF classification results schematic diagram；(h) adaptive M RF classification results ground truth data covers signal
Figure；
Fig. 4 distinct methods are directed to ESAR data classification result figure:
(a) ESAR data Pauli image；(b)- Wishart classifier classification results figure；
(c) it obscuresClassifier classification results figure；(d) adaptive M RF classification results figure.
Specific embodiment
Present invention will be described in further detail below with reference to the accompanying drawings.
Referring to fig. 2, remote sensing images ground scene classification method of the invention the following steps are included:
Step 1: under horizontal polarization and vertical polarization base, collision matrix are as follows:In polarization SAR data
Analysis and treatment process in, polarization scattering matrix is usually subjected to vectorization under orthogonal matrix base.The process table of vectorization
It is shown asWherein subscript T indicates that transposition, Tr () indicate to seek the mark of matrix, and Ψ is
Orthogonal one group 2 × 2 plural basic matrix set under Hermitian inner product.Under Lexicographic matrix base, obtain to
Amount is known as Lexicographic scattering vector, indicates are as follows: k4L=[Shh Shv Svh Svv]T, under Pauli matrix base, obtain
Pauli scattering vector are as follows:In singly station system and completely
In the case where sufficient reciprocal theorem, polarization scattering matrix S is symmetrical, that is, Shv=Svh。
The dimension of scattering vector is reduced to 3 from 4, and corresponding Lexicographic scattering vector and Pauli scattering vector simplify
Are as follows:With
The polarization covariance matrix of target is defined as the second moment of Lexicographic scattering vector, it may be assumed that
Wherein<>indicates multiple look processing or space average, subscript * andRespectively indicate complex conjugate and complex vector
Conjugate transposition.Similarly, polarization coherence matrix is the second moment of Pauli scattering vector, is indicated are as follows:
Covariance matrix C and coherence matrix T is Positive Semidefinite Hermitian Matrix, can pass through the unit tenth of the twelve Earthly Branches between the two
Matrix U is mutually converted, therefore C and T characteristic value having the same.
The transformational relation of the two and the form of unitary matrice U are as follows:
T=UCU-1Or C=U-1TU, wherein
Step 2: Polarization scattering entropy characterization decomposes unordered degree of the obtained different scattering types in statistical significance, meter
Formula are as follows:
Wherein
Wherein, PiThe pseudo- probability that type occurs is scattered for i-th kind, then the value range for the entropy H that polarizes is 0≤H≤1.Polarization
The proportionate relationship of characteristic value can not be fully described in entropy H, and Polarisation Anisotropy degree A can be used to show λ2And λ3Between relationship:
The value range of A is also 0≤A≤1.Average scattering angleIt is defined asDue to average scattering angle
Value it is related to the physical property of scattering process behind,It is the key parameter of main scattering mechanism for identification.
Step 3: the back scattering general power (span) of target is defined as the sum of the intensity of each POLARIZATION CHANNEL scattering coefficient,
Can be calculated by covariance matrix or coherence matrix, be span=| Shh|2+2|Shv|2+|Svv|2=Tr (C)=Tr
(T)。
Traditional neighborhood configuration mainly includes first order neighbors and second order neighborhood, then using traditional horse of fixed neighborhood configuration
Er Kefu model can cause smoothing effect, lead to the loss of finer structures information in figure.
In order to preferably protect these spatial informations, the present invention carries out polarization SAR figure using the MRF based on adaptive neighborhood
As classification.These adaptive neighborhoods are specifically to be made of five kinds of neighborhood alternate items, their shape respectively corresponds different spaces
Situation, including local uniform region and linear edge etc..For having the neighborhood of minimum variance, its homogeneity in span image
Highest, therefore optimum carries out MRF analysis.In this way, the smoothing effect of MRF is limited in selected neighborhood, from
And it can preferably keep the spatial details information such as edge and other finer structures.
Step 4: the present invention proposes an adaptive smoothing factor ψ when building priori energy function U ()sr.It is not
Only consider the classification information in neighborhood, while also considering pixel corresponding to different classes of fuzzy degree of membership.
Based on selected neighborhood ηsAnd smoothing factor ψsr, the calculating formula of energy function U () are as follows:
Wherein, | ηs| and bsThe respectively size of neighborhood space and homogeneous measurement.The smoothing factor ψ proposedsrCalculating formula
Are as follows:
Wherein, LcIt is the current class of pixel s,Indicate that the pixel belongs to classification LcFuzzy degree of membership.
It should be pointed out that when all fuzzy degree of membership are set as identical value, ψsrForm be δ (Ls-Lr), it is equivalent
In classical Potts model.Therefore, Potts model is considered as a special case of proposed model.
Step 5: according toTwo dimension fuzzy ownership function in plane can obtain the value of initially fuzzy degree of membershipIt indicates that pixel s belongs to the scale factor of classification k, and meetsIt can calculate on this basis initially
Cluster centreThen it can be changed to the value of fuzzy degree of membership using Fuzzy C-Means Cluster Algorithm
Generation and optimization.The present invention combines the spatial correlation information based on MRF with polarization information in iterative optimization procedure.
Different the number of iterations is marked using subscript j, the key step of iterative process is as follows:
5a) calculate coherence matrix < T of pixel ss> and classification k cluster centreBetween Wishart distance:
5b) based on obtained Wishart distance, fuzzy degree of membership is updated:
Wherein, mcIt is constant, ρ () is Huber function, and formula isWhen with cluster
When centre distance is closer, ρ () is the form of quadratic power, distance farther out when be then linear forms, to reduce the shadow of exceptional value
It rings.
5c) by conjunction with spatial correlation information, generating the fuzzy degree of membership of enhancing:
WhereinThe prior probability extracted is analyzed by local AMRF.By by space
Information is combined with original fuzzy degree of membership, available relatively reliable fuzzy degree of membership, while more with space background
Unanimously.
5d) the fuzzy degree of membership based on enhancing, is updated cluster centre:
Step 6: after loop termination, so that it may by the final segmentation classification C of pixel ssIt is assigned as with maximum membership degree
Respective classes, i.e.,Polarimetric SAR image is based onThe classification method of decomposition is utilized and is preset
Boundary distinguish different classifications, there are certain limitations on the boundary of this fixation.The cluster centre of certain class target or atural object
It is likely located on boundary or near border, different classifications is divided into so as to cause such pixel.It is asked to solve this
Topic, the present invention combine the spatial correlation information that local AMRF is analyzed with the fuzzy degree of membership based on polarization information, can
To inhibit the noise phenomenon in classification results.In addition, the present invention uses adaptive neighborhood, while proposing adaptive smooth system
Number, so as to preferably avoid smoothing effect, retaining space detailed information.Present invention improves over polarimetric SAR image segmentations
Algorithm, the result obtained in image segmentation is inevitable more reasonable, effective and accurate, meets current accuracy application demand.
Contrived experiment verifies classifying quality of the invention:
The present invention carries out the experiment of image terrain classifications using two groups of polarization SAR data, including AIRSAR L- wave band data and
ESAR L- wave band data.Wherein the spatial resolution of AIRSAR L-band data haplopia complex pattern be 6.6 meters (distances to) and
9.3 meters (orientation), data size is 900 × 1000 mainly comprising ground species such as residential quarter, meadow, barley, forest, potatoes
Not.ESAR L- wave band data is to obtain full-polarization SAR remote sensing by ESAR (Experimental SAR) sensor of German DLR
Image, the data are obtained in German Oberpfaffenhofen test area, and the spatial resolution of haplopia complex pattern is 3 meters, number
According to having a size of 1400 × 1000.Include type of ground objects abundant in figure, mainly includes artificial structure, forest, airfield runway, naked
Ground, meadow, three kinds of different types of crops and other vegetation etc..
Experimentation is as follows:
It is verified respectively using the method for the present invention and three kinds of control methods for two groups of polarimetric SAR images.It is empty combining
In the polarization SAR classification method of domain information, using the MRF analytical framework based on traditional Potts model and using adaptive flat
The MRF analytical framework of sliding coefficient, both methods subsequent referred to as Potts-MRF classification method and adaptive M RF classification method.
In addition, as a comparison, also utilizing- Wishart classifier and fuzzyClassifier carries out classification experiments.According to initialThe quantity of classification is fixed as 8 by the region division criterion of plane, the above method.It should be noted that fixed class
Other quantity will limit the application of proposed classification method to a certain extent, but spatial information (si) can also combine in a similar way
Into other Fuzzy classifications.It can be seen that classification results exist in homogeneous area from Fig. 3 (a) and Fig. 3 (c) largely to make an uproar
Sound, these noises are as caused by the difference of coherent spot and same Terrain Scattering characteristic.Pass through the space for analyzing AMRF
Relevant information is combined with polarization information, can significantly improve classification results.As shown in Fig. 3 (e) and Fig. 3 (g), homogenous region
Classification results are very smooth, and shape details have also obtained preferable holding.In addition, self-adaptive averaging factor of the present invention is according to neighbour
In domain pixel obscure degree of membership value and relative size, automatically regulate influence of the MRF spatial analysis to classification, so as into
One step improves final classification results.Compared with Fig. 3 (e), it can be seen that there are less noises in Fig. 3 (g), and edge is thin
Save also available more preferable reservation, such as the boundary in white ovals.One column of right side of Fig. 3 illustrates ground truth data coverage
Partial result.Seldom mistake point phenomenon is only existed in Fig. 3 (h), it was demonstrated that adaptive M RF classification method of the present invention it is effective
Property.
As seen from Figure 4, for ESAR L-band data, adaptive M RF classification method of the present invention achieves optimal
As a result, the salt-pepper noise of homogenous region has obtained obvious inhibition.In Fig. 4 (b) and Fig. 4 (c), many pictures in white ovals region
Element is accidentally divided.However, the result in the region is obviously improved in Fig. 4 (d), and classification results by using spatial information (si)
It is more smooth.In addition, the classification method proposed has used AMRF and adaptive smoothing factor, therefore some knots in image
Structure detailed information is not influenced by close region and classification are excessively smooth.For example, the white square region in Fig. 4 (d), grass
The marginal information on ground is effectively maintained, and very consistent with the scene in Fig. 4 (a).In contrast, Fig. 4 (b) and Fig. 4 (c)
In the region classification results there are obvious noise, structural information is also not clear enough.The above experimental result and analysis shows, this hair
The adaptive M RF classification method of bright proposition can be obviously improved classifying quality, and keep detailed structure information preferably.
Claims (7)
1. a kind of remote sensing images ground scene classification method, which comprises the following steps:
For polarimetric SAR image, its polarization covariance matrix and polarization coherence matrix are constructed；
Polarization coherence matrix is carried outIt decomposes, obtains Polarization scattering entropy H, degree of anisotropy A and average scattering angle
Calculate the span image of polarimetric SAR image, and designed image adaptive neighborhood on this basis；
Based on image adaptive neighborhood, self-adaptive averaging factor is designed, and constructs MRF priori energy function；
By the spatial correlation information based on MRF in conjunction with polarization information, is classified using fuzzy C-mean algorithm and construct objective function, iteration
Optimal degree of membership and class categories center are solved, polarimetric SAR image is made to be referred to the classification with maximum membership degree pixel-by-pixel.
2. remote sensing images ground scene classification method according to claim 1, it is characterised in that:
Constructing its polarization covariance matrix and polarization coherence matrix, detailed process is as follows:
Polarization radar collision matrix is expressed asWherein, H and V respectively indicates horizontal polarization and vertical pole
Change, SPQ(P, Q=H, V) is
Number, then has S in the back scattering for meeting reciprocal theoremHV=SVH；The expression formula of Pauli Scattering of Vector is as follows:
In formula, subscriptTRepresenting matrix transposition；
The expression formula of polarization coherence matrix is as follows:
In formula, subscriptThe conjugate transposition of representing matrix, subscript*Indicate complex conjugate；
The expression formula of polarization covariance matrix is as follows:
Wherein,
3. remote sensing images ground scene classification method according to claim 2, it is characterised in that:
Polarization coherence matrix is carried outDetailed process is as follows for decomposition:
By carrying out characteristic value or eigendecomposition to coherence matrix T, a simple statistical model is constructed, T is extended to
The sum of the response of three pinpoint targets, each target corresponds to a scattering mechanism, by unit character vector uiIt determines, and
Corresponding eigenvalue λiRepresent specific gravity of the scattering mechanism in entire scattering process；The expression formula of decomposable process are as follows:
Polarization scattering entropy H characterization decomposes unordered degree of the obtained different scattering types in statistical significance, and expression formula is as follows:
Wherein
Wherein, PiThe pseudo- probability that type occurs is scattered for i-th kind, then the value range for the entropy H that polarizes is 0≤H≤1；
Polarisation Anisotropy degree A is defined to show λ2And λ3Between relationship, expression formula are as follows:
In formula, the value range of A is 0≤A≤1；
Average scattering angleIs defined as:
Work as αiAt=0 °, target scattering corresponds to surface scattering, withIt is gradually increased, scattering mechanism becomes bragg surfaces and dissipates
It penetrates；
Work as αi=45 ° then represent dipole scattering, and then scattering type becomes the rescattering of two dielectric surfaces,
Work as αiBecome the dihedral angle scattering of metal surface under=90 ° of extreme case.
4. remote sensing images ground scene classification method according to claim 2, it is characterised in that: area based on polarimetric SAR image
Back scattering general power span calculating formula be span=| Shh|2+2|Shv|2+|Svv|2=Tr (C)=Tr (T) chooses span figure
Neighborhood as in minimum variance carries out Markov model analysis, and the smoothing effect of MRF is limited in selected neighborhood.
5. remote sensing images ground scene classification method according to claim 4, which is characterized in that based on selected neighborhood ηs
And adaptive smoothing factor ψsrIt constructs MRF priori energy function U (), expression formula is as follows:
Wherein, | ηs| and bsThe respectively size of neighborhood space and homogeneous measurement；
Smoothing factor ψsrCalculation formula it is as follows:
Wherein, LcIt is the current class of pixel s,Indicate that the pixel belongs to classification LcFuzzy degree of membership.
6. remote sensing images ground scene classification method according to claim 1, it is characterised in that:
According to being defined onTwo dimension fuzzy ownership function in plane obtains the value of initially fuzzy degree of membershipIndicate pixel s
Belong to the scale factor of classification k, and meetsInitial cluster centre is calculated on this basisThen it is iterated and optimizes using value of the Fuzzy C-Means Cluster Algorithm to fuzzy degree of membership.
7. remote sensing images ground scene classification method according to claim 6, which is characterized in that the step of iteration is as follows:
Firstly, calculating coherence matrix < T of pixel ss> and classification k cluster centreBetween Wishart distance:
Based on obtained Wishart distance, fuzzy degree of membership is updated:
Wherein, mcIt is a constant, ρ () is Huber function:
Pass through the fuzzy degree of membership in conjunction with spatial correlation information, generating enhancing:
Wherein,The prior probability extracted is analyzed by local AMRF；
Finally, the fuzzy degree of membership based on enhancing, is updated cluster centre:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811594292.0A CN109740475B (en) | 2018-12-25 | 2018-12-25 | Ground scene classification method for remote sensing image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811594292.0A CN109740475B (en) | 2018-12-25 | 2018-12-25 | Ground scene classification method for remote sensing image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109740475A true CN109740475A (en) | 2019-05-10 |
| CN109740475B CN109740475B (en) | 2020-04-21 |
Family
ID=66359907
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811594292.0A Active CN109740475B (en) | 2018-12-25 | 2018-12-25 | Ground scene classification method for remote sensing image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109740475B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110619349A (en) * | 2019-08-12 | 2019-12-27 | 深圳市识农智能科技有限公司 | Plant image classification method and device |
| CN112465837A (en) * | 2020-12-09 | 2021-03-09 | 北京航空航天大学 | Image segmentation method for sparse subspace fuzzy clustering by utilizing spatial information constraint |
| CN112558066A (en) * | 2020-10-30 | 2021-03-26 | 西南电子技术研究所（中国电子科技集团公司第十研究所） | Dual-polarization SAR image system |
| CN112950492A (en) * | 2021-01-28 | 2021-06-11 | 中国石油大学(华东) | Full-polarization SAR image denoising method based on self-adaptive anisotropic diffusion |
| CN113587941A (en) * | 2020-05-01 | 2021-11-02 | 华为技术有限公司 | High-precision map generation method, positioning method and device |
| CN114219822A (en) * | 2021-12-15 | 2022-03-22 | 深圳先进技术研究院 | Object-oriented fuzzy clustering image segmentation method, system, terminal and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030169942A1 (en) * | 2001-05-16 | 2003-09-11 | Dorin Comaniciu | Systems and methods for automatic scale selection in real-time imaging |
| CN103294792A (en) * | 2013-05-13 | 2013-09-11 | 西安电子科技大学 | Polarimetric SAR (synthetic aperture radar) terrain classification method based on semantic information and polarimetric decomposition |
| CN104598922A (en) * | 2015-01-07 | 2015-05-06 | 河海大学 | Completely-polarized SAR (synthetic aperture radar) image classification method based on fuzzy c-means |
| CN104700110A (en) * | 2015-04-03 | 2015-06-10 | 电子科技大学 | Plant covering information extracting method based on perfect polarization SAR images |
| CN107145860A (en) * | 2017-05-05 | 2017-09-08 | 西安电子科技大学 | Classification of Polarimetric SAR Image method based on spatial information and deep learning |
-
2018
- 2018-12-25 CN CN201811594292.0A patent/CN109740475B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030169942A1 (en) * | 2001-05-16 | 2003-09-11 | Dorin Comaniciu | Systems and methods for automatic scale selection in real-time imaging |
| CN103294792A (en) * | 2013-05-13 | 2013-09-11 | 西安电子科技大学 | Polarimetric SAR (synthetic aperture radar) terrain classification method based on semantic information and polarimetric decomposition |
| CN104598922A (en) * | 2015-01-07 | 2015-05-06 | 河海大学 | Completely-polarized SAR (synthetic aperture radar) image classification method based on fuzzy c-means |
| CN104700110A (en) * | 2015-04-03 | 2015-06-10 | 电子科技大学 | Plant covering information extracting method based on perfect polarization SAR images |
| CN107145860A (en) * | 2017-05-05 | 2017-09-08 | 西安电子科技大学 | Classification of Polarimetric SAR Image method based on spatial information and deep learning |
Non-Patent Citations (2)
| Title |
|---|
| 安健: ""基于极化合成孔径雷达图像分类算法研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 张涛 等: ""采用自适应邻域马尔可夫场的极化SAR图像分割"", 《测试技术学报》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110619349A (en) * | 2019-08-12 | 2019-12-27 | 深圳市识农智能科技有限公司 | Plant image classification method and device |
| CN113587941A (en) * | 2020-05-01 | 2021-11-02 | 华为技术有限公司 | High-precision map generation method, positioning method and device |
| CN112558066A (en) * | 2020-10-30 | 2021-03-26 | 西南电子技术研究所（中国电子科技集团公司第十研究所） | Dual-polarization SAR image system |
| CN112558066B (en) * | 2020-10-30 | 2023-08-18 | 西南电子技术研究所（中国电子科技集团公司第十研究所） | Dual polarized SAR image system |
| CN112465837A (en) * | 2020-12-09 | 2021-03-09 | 北京航空航天大学 | Image segmentation method for sparse subspace fuzzy clustering by utilizing spatial information constraint |
| CN112465837B (en) * | 2020-12-09 | 2022-08-23 | 北京航空航天大学 | Image segmentation method for sparse subspace fuzzy clustering by utilizing spatial information constraint |
| CN112950492A (en) * | 2021-01-28 | 2021-06-11 | 中国石油大学(华东) | Full-polarization SAR image denoising method based on self-adaptive anisotropic diffusion |
| CN114219822A (en) * | 2021-12-15 | 2022-03-22 | 深圳先进技术研究院 | Object-oriented fuzzy clustering image segmentation method, system, terminal and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109740475B (en) | 2020-04-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109740475A (en) | A kind of remote sensing images ground scene classification method | |
| CN103824084B (en) | Classification of Polarimetric SAR Image method based on SDIT and SVM | |
| Schou et al. | CFAR edge detector for polarimetric SAR images | |
| CN102982338B (en) | Classification of Polarimetric SAR Image method based on spectral clustering | |
| CN105930772A (en) | City impervious surface extraction method based on fusion of SAR image and optical remote sensing image | |
| Parida et al. | Polarimetric decomposition methods for LULC mapping using ALOS L-band PolSAR data in Western parts of Mizoram, Northeast India | |
| Yin et al. | SLIC superpixel segmentation for polarimetric SAR images | |
| CN110223311A (en) | Polarimetric radar edge Detection Method of Remote Sensing Images based on power drive adaptive windows | |
| CN106611422B (en) | Stochastic gradient Bayes's SAR image segmentation method based on sketch structure | |
| Fan et al. | Marine floating raft aquaculture detection of GF-3 PolSAR images based on collective multikernel fuzzy clustering | |
| Anfinsen | Statistical analysis of multilook polarimetric radar images with the Mellin transform | |
| CN102402685A (en) | Method for segmenting three Markov field SAR image based on Gabor characteristic | |
| Lang et al. | Mean-shift-based speckle filtering of polarimetric SAR data | |
| Dabboor et al. | A new likelihood ratio for supervised classification of fully polarimetric SAR data: An application for sea ice type mapping | |
| CN106408581A (en) | Rapid three-dimensional point cloud linear extraction method | |
| CN109712149A (en) | A kind of image partition method based on wavelet energy and fuzzy C-mean algorithm | |
| Liu et al. | Variational textured Dirichlet process mixture model with pairwise constraint for unsupervised classification of polarimetric SAR images | |
| Ferro-Famil et al. | Urban area remote sensing from L-band PolSAR data using time-frequency techniques | |
| Dong et al. | Copula-based joint statistical model for polarimetric features and its application in PolSAR image classification | |
| CN103870842A (en) | Polarized SAR image classification method combining polarization feature and watershed | |
| CN109946670A (en) | A kind of polarization radar information extracting method of optical image driving | |
| Weissgerber et al. | A temporal estimation of entropy and its comparison with spatial estimations on PolSAR images | |
| Khan et al. | Unsupervised classification of polarimetric SAR images by EM algorithm | |
| CN112950492B (en) | Full-polarization SAR image denoising method based on self-adaptive anisotropic diffusion | |
| Xu et al. | Polarimetric semivariogram-based spatial scale selection for PolSAR image segmentation with mean-shift algorithm |
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 | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: A ground scene classification method of remote sensing image Effective date of registration: 20201229 Granted publication date: 20200421 Pledgee: Hangzhou High-tech Financing Guarantee Co.,Ltd. Pledgor: HANGZHOU SHIPING INFORMATION & TECHNOLOGY Co.,Ltd. Registration number: Y2020330001331 |
|
| PC01 | Cancellation of the registration of the contract for pledge of patent right | ||
| PC01 | Cancellation of the registration of the contract for pledge of patent right |
Date of cancellation: 20220128 Granted publication date: 20200421 Pledgee: Hangzhou High-tech Financing Guarantee Co.,Ltd. Pledgor: HANGZHOU SHIPING INFORMATION & TECHNOLOGY Co.,Ltd. Registration number: Y2020330001331 |