CN106446935A - Kernel sparse representation and spatial constraint-based polarimetric SAR image classification method - Google Patents
Kernel sparse representation and spatial constraint-based polarimetric SAR image classification method Download PDFInfo
- Publication number
- CN106446935A CN106446935A CN201610802709.2A CN201610802709A CN106446935A CN 106446935 A CN106446935 A CN 106446935A CN 201610802709 A CN201610802709 A CN 201610802709A CN 106446935 A CN106446935 A CN 106446935A
- Authority
- CN
- China
- Prior art keywords
- scattering
- polarimetric sar
- sar image
- sorted
- classification
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Radar Systems Or Details Thereof (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kernel sparse representation and spatial constraint-based polarimetric SAR image classification method and belongs to the technical field of polarimetric SAR image classification. The method is used for solving the problem of low polarimetric SAR image classification efficiency caused by the fact that a pixel-based polarimetric SAR remote sensing image classification method has a high dimension characteristic and an excessively large calculation amount. The method comprises the classification steps of preprocessing an original polarimetric SAR image data to obtain to-be-classified polarimetric SAR image data; determining image types and training samples; performing polarimetric SAR image feature extraction based on a multicomponent scattering model, and eigenvalue decomposition and gray-level co-concurrence matrixes; constructing an over-complete dictionary; obtaining optimal sparse coefficients of the training samples of the image types; and performing signal reconstruction by utilizing the optimal sparse coefficients, and determining the image types according to reconstructed signals of the training samples of the image types and errors of to-be-classified polarimetric SAR images, thereby realizing the classification of the to-be-classified polarimetric SAR images. The method is used for the polarimetric SAR image classification.
Description
Technical field
The present invention relates to the Classification of Polarimetric SAR Image method based on nuclear sparse expression and space constraint, belong to polarization SAR figure
As sorting technique field.
Background technology
Polarimetric synthetic aperture radar (SAR) as a kind of advanced microwave remote sensing means, have round-the-clock, round-the-clock, point
The advantage that resolution is high, large area covers, it passes through the echo information under observation different transmitting-receiving polarization combination, distinguishes the careful of object
The parameters such as structure, target sensing and material composition carry out the backscattering characteristic that more adding system comprehensively reflects target, thus obtaining
Obtain the terrestrial object information more fully enriching, have broad application prospects in remote sensing fields.
Come into operation with increasing polarization SAR system, the polarimetric SAR image data of acquisition is more and more abundanter, such as
What is made to these view data and quickly and accurately interpreting, a difficult problem always in the urgent need to address.Polarimetric SAR Image
Classification refers to the polarization SAR Interpretation Technology classified of atural object occurring in scene, merged image procossing, pattern recognition,
The research of the multiple fields such as artificial intelligence.For the classification of atural object, mainly include supervised classification and two kinds of unsupervised classification.Supervision
Classification is classification to be previously set according to the comparison diagram of scattered information in existing atural object classification figure or Polarimetric SAR Image, and
The training sample choosing each class is trained.Unsupervised classification does not have human intervention for atural object classification, and program is according to certain
Criterion carries out cluster and obtains classification results.
Sparse representation theory is the important tool that data efficient represents, is the once breakthrough Theory Advance that signal represents,
The rarefaction representation of signal can be realized by excessively complete dictionary, be significant in the epoch of information-intensive society data explosion.
At present, in recognition of face, the aspect such as classification hyperspectral imagery and SAR automatic target detection achieves significantly rarefaction representation
Achievement.
With the raising of SAR system data retrieval capabilities and image resolution ratio, in image, target type is more finely various,
Point target in low resolution SAR image becomes target area in high-definition picture, is provided with region and Texture eigenvalue.?
In image classification and interpretation, except the scattering propertiess of target itself are in addition it is also necessary to consider its spatial relationship.Research shows, in remote sensing
In image, neighbor tends to belong to same category, and its Electromagnetic Scattering Characteristics and other features often have larger similarity.
Therefore, during rarefaction representation in addition to limiting sparse condition and reconstruction accuracy, it should also be taken into account that the space of neighbor
Constraint.Meet hypothesis below:For the adjacent pixel in space, approximately linear can be carried out with identical training sample and represent,
But the weight coefficient of corresponding training sample is different.The i.e. support of the atom in excessively complete dictionary in the rarefaction representation of neighbor
Collection is identical, and different is only weight coefficient.
Higher-dimension that existing sorting technique based on pixel exists and the excessive defect of amount of calculation, and its specification area one
Cause property is poor, has had a strong impact on the classification effectiveness of Polarimetric SAR Image.
Content of the invention
The invention aims to solve existing based on the area based on polarimetric SAR image classification method of pixel exist higher-dimension and
The excessive defect of amount of calculation, causes the low problem of Classification of Polarimetric SAR Image efficiency, there is provided one kind is based on nuclear sparse expression and sky
Between constraint Classification of Polarimetric SAR Image method.
Classification of Polarimetric SAR Image method based on nuclear sparse expression and space constraint of the present invention, it includes following step
Suddenly:
Step one:Collection original polarization SAR image data, carries out pretreatment, obtains polarimetric SAR image data to be sorted;
Step 2:Determine the image category of polarimetric SAR image data to be sorted and the training sample of each image category;
Step 3:Multicomponent scattering model, feature are based on successively to Polarimetric SAR Image to be sorted according to described image classification
Value is decomposed and gray level co-occurrence matrixes carry out Polarimetric SAR Image feature extraction:
A:Polarimetric SAR Image scattering signatures based on multicomponent scattering model extract:Polarimetric SAR Image to be sorted is carried out
Polarization target decomposition, obtains the scattering that odd scattering, even scattering, volume scattering, line scattering and spiral scatter five kinds of scattering compositions
Power;
B:The Polarimetric SAR Image polarization characteristic that feature based value is decomposed extracts:To Polarimetric SAR Image feature based to be sorted
Value decomposes three eigenvalues of acquisition and Polarization scattering entropy, average scattering angle and anisotropy amount;
C:Polarimetric SAR Image texture feature extraction based on gray level co-occurrence matrixes:Energy is extracted to Polarimetric SAR Image to be sorted
Amount and contrast textural characteristics;
Step 4:All features that in step 3, Polarimetric SAR Image to be sorted is carried out with feature extraction acquisition are carried out whole
Close and kernel function mapping, by the use of each described image classification training sample characteristic vector as the atom of dictionary, built complete
Standby dictionary;
Step 5:According to the excessively complete dictionary obtaining and space constraint relation, using synchronous orthogonal matching pursuit algorithm pair
Polarimetric SAR Image to be sorted carries out rarefaction representation, obtains the optimum sparse coefficient of each image category training sample;
Step 6:Optimum sparse coefficient using each the image category training sample obtaining in step 5 carries out signal weight
Build, the image class belonging to error decision of the reconstruction signal according to each image category training sample and Polarimetric SAR Image to be sorted
Not, realize the classification to Polarimetric SAR Image to be sorted.
Advantages of the present invention:The inventive method is applied to field of remote sensing image processing, realizes the classification of Polarimetric SAR Image.
The polarization characteristic of Polarimetric SAR Image and textural characteristics are combined by it, preferably describe characters of ground object, and are realized by nuclear mapping
Non-linear transfer, based on sparse representation theory and space constraint, makes full use of the similarity of adjacent space, decreases the dimension of classification
Degree and complexity, solve the problems, such as that the specification area concordance based on pixel is poor, improve the classification effect of Polarimetric SAR Image
Rate.
Brief description
Fig. 1 is the flow chart of the Classification of Polarimetric SAR Image method based on nuclear sparse expression and space constraint of the present invention;
Fig. 2 is the magnitude image of Polarimetric SAR Image hh passage to be sorted;
Fig. 3 is multicomponent goal decomposition result schematic diagram;
Fig. 4 is Eigenvalues Decomposition result schematic diagram;
Fig. 5 is the energy diagram of gray level co-occurrence matrixes;
Fig. 6 is gray level co-occurrence matrixes contrast schematic diagram;
Fig. 5 and Fig. 6 is as the textural characteristics schematic diagram of gray level co-occurrence matrixes;
Fig. 7 is the sparse representation model of space constraint;
Fig. 8 is the classification results schematic diagram of present invention Polarimetric SAR Image to be sorted.
Specific embodiment
Specific embodiment one:With reference to Fig. 1 to Fig. 8, present embodiment is described, is based on core described in present embodiment dilute
Relieving the exterior syndrome shows the Classification of Polarimetric SAR Image method with space constraint, and it comprises the following steps:
Step one:Collection original polarization SAR image data, carries out pretreatment, obtains polarimetric SAR image data to be sorted;
Step 2:Determine the image category of polarimetric SAR image data to be sorted and the training sample of each image category;
Step 3:Multicomponent scattering model, feature are based on successively to Polarimetric SAR Image to be sorted according to described image classification
Value is decomposed and gray level co-occurrence matrixes carry out Polarimetric SAR Image feature extraction:
A:Polarimetric SAR Image scattering signatures based on multicomponent scattering model extract:Polarimetric SAR Image to be sorted is carried out
Polarization target decomposition, obtains the scattering that odd scattering, even scattering, volume scattering, line scattering and spiral scatter five kinds of scattering compositions
Power;
B:The Polarimetric SAR Image polarization characteristic that feature based value is decomposed extracts:To Polarimetric SAR Image feature based to be sorted
Value decomposes three eigenvalues of acquisition and Polarization scattering entropy, average scattering angle and anisotropy amount;
C:Polarimetric SAR Image texture feature extraction based on gray level co-occurrence matrixes:Energy is extracted to Polarimetric SAR Image to be sorted
Amount and contrast textural characteristics;
Step 4:All features that in step 3, Polarimetric SAR Image to be sorted is carried out with feature extraction acquisition are carried out whole
Close and kernel function mapping, by the use of each described image classification training sample characteristic vector as the atom of dictionary, built complete
Standby dictionary;
Step 5:According to the excessively complete dictionary obtaining and space constraint relation, using synchronous orthogonal matching pursuit algorithm pair
Polarimetric SAR Image to be sorted carries out rarefaction representation, obtains the optimum sparse coefficient of each image category training sample;
Step 6:Optimum sparse coefficient using each the image category training sample obtaining in step 5 carries out signal weight
Build, the image class belonging to error decision of the reconstruction signal according to each image category training sample and Polarimetric SAR Image to be sorted
Not, realize the classification to Polarimetric SAR Image to be sorted.
In the A of step 3, the Polarimetric SAR Image scattering signatures extraction detailed process based on multicomponent scattering model is:
The covariance matrix of Polarimetric SAR Image to be sorted is decomposed into odd scattering, even scattering, volume scattering, spiral dissipate
Penetrate the weighted sum scattering five kinds of basic scattering mechanisms with line:
C=foddCodd+fdoubleCdouble+fvolumeCvolume+fhelixChelix+fwireCwire, (1)
In formula, C is the covariance matrix of Polarimetric SAR Image to be sorted, CoddScatter covariance matrix, C for odddoubleFor
Even scatters covariance matrix, CvolumeFor volume scattering covariance matrix, ChelixScatter covariance matrix, C for spiralwireFor line
Scattering covariance matrix, foddFor the weight coefficient of odd scattering, fdoubleFor the weight coefficient of even scattering, fvolumeDissipate for body
The weight coefficient penetrated, fhelixFor the weight coefficient of spiral scattering, fwireWeight coefficient for line scattering;
The covariance matrix of Polarimetric SAR Image to be sorted<C>For:
Each element S in formulapq(p, q=h v) represent that, with the transmitting of q polarization mode, target when p-polarization mode receives is backward
Complex scattering coefficients,<>Represent spacial average;
Make formula (1) equal sign left and right sides items correspondent equal, that is,:
In formula, β is the ratio of hh backscattering coefficient in odd scattering and vv backscattering coefficient, and α is hh in even scattering
Backscattering coefficient and the ratio of vv backscattering coefficient, γ is the ratio of hh polarization and vv polarization backscattering coefficient in line scattering
Value, ρ is the ratio of hv polarization in line scattering and vv polarization backscattering coefficient;
Solve the weight coefficient f that above formula obtains five kinds of scattering mechanismsodd、fdouble、fvolume、fhelixAnd fwire, and then solve
Obtain scattered power P of odd scatteringodd, even scattering scattered power Pdouble, scattered power P of volume scatteringvolume, spiral
Scattered power P of scatteringhelix, line scattering scattered power PwireAnd total scattering power P:
Thus Polarimetric SAR Image to be sorted is decomposed into the power of five kinds of Polarization scattering mechanism as Polarization scattering feature.
The Polarimetric SAR Image polarization characteristic that in the B of step 3, feature based value is decomposed extracts detailed process:
The coherence matrix of Polarimetric SAR Image to be sorted<T>Eigenvalues Decomposition be expressed as follows:
In formula, H is Matrix Conjugate transposition, and T is matrix transpose, Λ=diag { λ1,λ2,λ3, it is characterized value diagonal matrix, should
The diagonal of eigenvalue diagonal matrix is three eigenvalues of coherence matrix, λ1≥λ2≥λ3>=0,It is characterized arrow
Moment matrix, its column vector corresponds to<T>Orthogonal characteristic vectorWithCharacteristic vector is expressed as:
α in formulaiRepresent the internal degree of freedom of scattering object, span is 0 °≤αi≤90°;βiFor scattering object with respect to thunder
Reach the deflection of sight line, span is -180 °≤βi≤180°;δiFor the first scattering phase angle of scattering object, γiFor scattering
Second scattering phase angle of body;
Define Polarization scattering entropy H, average scattering angle according to three eigenvaluesWith anisotropy amount A, respectively:
P in formulaiFor:pi=λi/ ∑ λ, wherein λ=λ1+λ2+λ3;
For Polarization scattering entropy H, 0≤H≤1, reflection Polarimetric SAR Image dissipates from isotropic scatterning H=0 to completely random
Penetrate the unordered degree of statistics of H=1;Average scattering angle Represent the scattering mechanism of Polarimetric SAR Image;Each to different
Property amount A reflect the size of second, third eigenvalue, be the supplementary parameter of scattering entropy.
In the C of step 3, the Polarimetric SAR Image texture feature extraction detailed process based on gray level co-occurrence matrixes is:
Based on Probability p (a, the b) computing formula of gray level co-occurrence matrixes it is:
Above formula represents that on a position in Polarimetric SAR Image to be sorted, gray level is the pixel of a to give apart from d and side
To θ on another location gray level be b pixel occur Probability p (a, b | θ, d);
The matrix form of gray level co-occurrence matrixes GLCM is:
In formula, L is gradation of image series;
ENERGY E NY based on gray level co-occurrence matrixes and contrast C ON are used for describing the texture spy of Polarimetric SAR Image to be sorted
Levy, be defined as:
ENERGY E NY represents the uniformity of intensity profile in Polarimetric SAR Image to be sorted;
Contrast C ON represents the definition of Polarimetric SAR Image to be sorted.
The detailed process building complete dictionary in step 4 is:
All features that Polarimetric SAR Image to be sorted in step 3 is carried out feature extraction acquisition are integrated, and acquisition is retouched
State the characteristic vector of training sample, recycle l2Norm method is normalized, and the characteristic vector after normalization is carried out kernel function
Space reflection, the atom as dictionary built complete dictionary;
Set image category number as Q, for classification i, i=1,2 ..., Q, the N of selectioniThe M Wei Te of individual training sample
Levy vector to be normalized, obtain NiIndividual characteristic vectorExcessively complete dictionary as atomic building classification i
If being mapped as φ from data space to higher dimensional space, carried out in characteristic space vector using gaussian radial basis function
The long-pending kernel function calculatingFor:
In formula, σ is constant;For space pixel p in Polarimetric SAR Image to be sortedmCharacteristic vector,For polarization to be sorted
Space pixel p in SAR imagenCharacteristic vector;
Mapped by kernel function, atomIt is each mapped toCross complete dictionary DiIt is mapped as
Dφi, the excessively complete dictionary of all categoriesAfter being mapped by kernel function, composition mapped complete dictionary
Dφ:Dφ=[Dφ1,…,DφQ]∈RM*N(N=N1+…+NQ).
The detailed process obtaining the optimum sparse coefficient of each image category training sample in step 5 is:
If Polarimetric SAR Image spatial neighborhood pixels p to be sortedmAnd pn, the characteristic vector of extraction is respectivelyWithKernel function is mapped asWithFor classification i,By excessively complete dictionary DφiRarefaction representation is:
In formula,It is characterized vectorKernel function mapping vector,For rarefaction representation
Coefficient vector, xim1,xim2,L,For dictionary D excessively complete during rarefaction representationφiIn each atom weight coefficient, wherein non-zero
The subscript collection of coefficient is combined into
According to space constraint relation, neighbor pnThe characteristic vector extractingNuclear mapping vectorFor classification i, use
Pixel pmIdentical atomBut different coefficients carry out linear expression:
It is set with u spatial neighborhood pixels, the eigenmatrix that the feature of extraction builds after kernel function mapping isFor classification i, by excessively complete dictionary DφiRarefaction representation is:
In formula, coefficient matrix XiMiddle nonzero term has identical bottom to mark
Therefore, the sparse representation model of space constraintFor:
In formula, | | X | |OK, 0Represent coefficient matrix XiIn non-zero row number, referred to as degree of rarefication K;
The sparse representation model of space constraint is converted into using degree of rarefication K as relaxation condition:
In formula | | | |FFor Frobenius norm,Optimum sparse coefficient training sample tried to achieve for classification i.
Realize in step 6 to the detailed process of the classification of all Polarimetric SAR Images to be sorted be:
Q excessively complete dictionary D of Q class image category composition1,D2,...DQ, after kernel function mapping, obtain each class respectively
The optimum sparse coefficient of other training sampleAgain solve acquisition Polarimetric SAR Image to be sorted with different classes of
Error r of excellent sparse coefficient reconstruction signali(Φφ):
In formula | | | |2For norm;
Polarimetric SAR Image generic to be sorted is judged by minimum residual method, chooses residual with Polarimetric SAR Image to be sorted
The minimum classification representated by reconstruction signal of difference is as affiliated image category Class (Φ) of this Polarimetric SAR Image to be sorted:
In present embodiment, orthogonal characteristic vectorWithCan be understood as certain scattering mechanism.αiCorresponding certain
Scattering mechanism type, unrelated with the direction of scattering object;βiRelevant with the rotation of sensor coordinate system, if adopting Pauli base
Vector quantization is carried out to collision matrix, then βiFor scattering object with respect to radar line of sight deflection.In described polarization SAR to be sorted
In image, the rill of texture is deeper, then its contrast C ON is bigger, and the visual effect of image is clear.Constant σ practical application
When take appropriate value as needed.
According to space constraint relation, the adjacent pixel in space can carry out approximately linear with identical training sample and represent,
But corresponding weight coefficient is different, and that is, in the rarefaction representation of neighbor, the atom in excessively complete dictionary is identical, different
It is only weight coefficient.
Embodiment 1:The effect of the present invention can be further illustrated by following experiment.
1st, experimental data
The present invention experiment used by data be EMISAR Foulum area Polarimetric SAR Image, image size be 886 ×
1100 pixels.This area has the multiformity of atural object, and its hh channel amplitude image is as shown in Fig. 2 cut according on Google Maps
The optical picture taking understand Foulum area atural object can be divided into forest, city, bare area, thin through crop, broad leaf crop five class.Described
The optical picture the intercepting same time non-with polarization SAR data obtains.
2nd, experiment content and analysis
To Polarimetric SAR Image, classified based on the Classification of Polarimetric SAR Image method of nuclear sparse expression and space constraint,
Result is as shown in Figure 8.
For test sample, the nicety of grading of the Classification of Polarimetric SAR Image method based on nuclear sparse expression and space constraint
Confusion matrix is as shown in table 1.Classification of Polarimetric SAR Image method based on nuclear sparse expression and space constraint and dividing based on SVM
Class classification Comparative result is as shown in table 2.
Table 1
Table 2
By the classification results of Fig. 8 it can be seen that Classification of Polarimetric SAR Image side based on nuclear sparse expression and space constraint
Method can obtain classification results well, especially homogeneous area, due to space constraint, atural object can be had and more comprehensively retouch
State, it is possible to reduce the noise of classification results, thus further increase classifying quality.From table 2 it can be seen that except improving classification
Precision, in terms of classification effectiveness, the inventive method takes as 241s, and is taken based on the rarefaction representation sorting technique of pixel and be
1154s.This is because combining the rarefaction representation of space constraint in categorizing process, the rarefaction representation process of neighbor is simultaneously
Carry out, and in the categorizing process of rarefaction representation, rarefaction representation is carried out based on single pixel.Therefore, based on sparse table
Show that the Classification of Polarimetric SAR Image method with space constraint not only make use of spatial information, improves nicety of grading consistent with region
Property, and classification effectiveness is greatly improved.
Claims (7)
1. a kind of Classification of Polarimetric SAR Image method based on nuclear sparse expression and space constraint it is characterised in that it include following
Step:
Step one:Collection original polarization SAR image data, carries out pretreatment, obtains polarimetric SAR image data to be sorted;
Step 2:Determine the image category of polarimetric SAR image data to be sorted and the training sample of each image category;
Step 3:Multicomponent scattering model is based on successively to Polarimetric SAR Image to be sorted according to described image classification, eigenvalue divides
Solution and gray level co-occurrence matrixes carry out Polarimetric SAR Image feature extraction:
A:Polarimetric SAR Image scattering signatures based on multicomponent scattering model extract:Polarimetric SAR Image to be sorted is polarized
Goal decomposition, obtains the scattered power that odd scattering, even scattering, volume scattering, line scattering and spiral scatter five kinds of scattering compositions;
B:The Polarimetric SAR Image polarization characteristic that feature based value is decomposed extracts:Polarimetric SAR Image feature based value to be sorted is divided
Solution obtains three eigenvalues and Polarization scattering entropy, average scattering angle and anisotropy amount;
C:Polarimetric SAR Image texture feature extraction based on gray level co-occurrence matrixes:To Polarimetric SAR Image to be sorted extract energy and
Contrast textural characteristics;
Step 4:All features that in step 3, Polarimetric SAR Image to be sorted is carried out with feature extraction acquisition are integrated and
Kernel function map, by the use of each described image classification training sample characteristic vector as the atom of dictionary, built complete word
Allusion quotation;
Step 5:According to the excessively complete dictionary obtaining and space constraint relation, treated point using synchronous orthogonal matching pursuit algorithm
Class Polarimetric SAR Image carries out rarefaction representation, obtains the optimum sparse coefficient of each image category training sample;
Step 6:Optimum sparse coefficient using each the image category training sample obtaining in step 5 carries out signal reconstruction,
The image category belonging to error decision of the reconstruction signal according to each image category training sample and Polarimetric SAR Image to be sorted,
Realize the classification to Polarimetric SAR Image to be sorted.
2. the Classification of Polarimetric SAR Image method based on nuclear sparse expression and space constraint according to claim 1, its feature
It is, in the A of step 3, the Polarimetric SAR Image scattering signatures extraction detailed process based on multicomponent scattering model is:
By the covariance matrix of Polarimetric SAR Image to be sorted be decomposed into odd scattering, even scattering, volume scattering, spiral scattering and
Line scatters the weighted sum of five kinds of basic scattering mechanisms:
C=foddCodd+fdoubleCdouble+fvolumeCvolume+fhelixChelix+fwireCwire, (1)
In formula, C is the covariance matrix of Polarimetric SAR Image to be sorted, CoddScatter covariance matrix, C for odddoubleFor even
Scattering covariance matrix, CvolumeFor volume scattering covariance matrix, ChelixScatter covariance matrix, C for spiralwireFor line scattering
Covariance matrix, foddFor the weight coefficient of odd scattering, fdoubleFor the weight coefficient of even scattering, fvolumeFor volume scattering
Weight coefficient, fhelixFor the weight coefficient of spiral scattering, fwireWeight coefficient for line scattering;
The covariance matrix of Polarimetric SAR Image to be sorted<C>For:
Each element S in formulapq(p, q=h v) represent that, with the transmitting of q polarization mode, target when p-polarization mode receives dissipates backward again
Penetrate coefficient,<>Represent spacial average;
Make formula (1) equal sign left and right sides items correspondent equal, that is,:
In formula, β is the ratio of hh backscattering coefficient in odd scattering and vv backscattering coefficient, and α is backward for hh in even scattering
Scattering coefficient and the ratio of vv backscattering coefficient, γ is the ratio of hh polarization and vv polarization backscattering coefficient in line scattering, ρ
Ratio for hv polarization in line scattering and vv polarization backscattering coefficient;
Solve the weight coefficient f that above formula obtains five kinds of scattering mechanismsodd、fdouble、fvolume、fhelixAnd fwire, and then solve acquisition
Scattered power P of odd scatteringodd, even scattering scattered power Pdouble, scattered power P of volume scatteringvolume, spiral scattering
Scattered power Phelix, line scattering scattered power PwireAnd total scattering power P:
Podd=fodd(1+|β|2)
Pdouble=fdouble(1+|α|2)
Pvolume=8fvolume/ 3, (4) Phelix=fhelix
Pwire=fwire(1+|γ|2+2|ρ|2)
P=Podd+Pdouble+Pvolume+Phelix+Pwire
Thus Polarimetric SAR Image to be sorted is decomposed into the power of five kinds of Polarization scattering mechanism as Polarization scattering feature.
3. the Classification of Polarimetric SAR Image method based on nuclear sparse expression and space constraint according to claim 2, its feature
It is, the Polarimetric SAR Image polarization characteristic that in the B of step 3, feature based value is decomposed extracts detailed process and is:
The coherence matrix of Polarimetric SAR Image to be sorted<T>Eigenvalues Decomposition be expressed as follows:
In formula, H is Matrix Conjugate transposition, and T is matrix transpose, Λ=diag { λ1,λ2,λ3, it is characterized value diagonal matrix, this feature
The diagonal of value diagonal matrix is three eigenvalues of coherence matrix, λ1≥λ2≥λ3>=0,It is characterized vector moment
Battle array, its column vector corresponds to<T>Orthogonal characteristic vectorWithCharacteristic vector is expressed as:
α in formulaiRepresent the internal degree of freedom of scattering object, span is 0 °≤αi≤90°;βiRegard with respect to radar for scattering object
The deflection of line, span is -180 °≤βi≤180°;δiFor the first scattering phase angle of scattering object, γiFor scattering object
Second scattering phase angle;
Define Polarization scattering entropy H, average scattering angle according to three eigenvaluesWith anisotropy amount A, respectively:
P in formulaiFor:pi=λi/ ∑ λ, wherein λ=λ1+λ2+λ3;
For Polarization scattering entropy H, 0≤H≤1, reflection Polarimetric SAR Image scatters H from isotropic scatterning H=0 to completely random
=1 unordered degree of statistics;Average scattering angle Represent the scattering mechanism of Polarimetric SAR Image;Anisotropy amount
A reflects the size of second, third eigenvalue, is the supplementary parameter of scattering entropy.
4. the Classification of Polarimetric SAR Image method based on nuclear sparse expression and space constraint according to claim 3, its feature
It is, in the C of step 3, the Polarimetric SAR Image texture feature extraction detailed process based on gray level co-occurrence matrixes is:
Based on Probability p (a, the b) computing formula of gray level co-occurrence matrixes it is:
Above formula represents that on a position in Polarimetric SAR Image to be sorted, gray level is the pixel of a to give apart from d and direction θ
On another location gray level be b pixel occur Probability p (a, b | θ, d);
The matrix form of gray level co-occurrence matrixes GLCM is:
In formula, L is gradation of image series;
ENERGY E NY based on gray level co-occurrence matrixes and contrast C ON are used for describing the textural characteristics of Polarimetric SAR Image to be sorted,
It is defined as:
ENERGY E NY represents the uniformity of intensity profile in Polarimetric SAR Image to be sorted;
Contrast C ON represents the definition of Polarimetric SAR Image to be sorted.
5. the Classification of Polarimetric SAR Image method based on nuclear sparse expression and space constraint according to claim 4, its feature
It is, the detailed process building complete dictionary in step 4 is:
All features that Polarimetric SAR Image to be sorted in step 3 is carried out feature extraction acquisition are integrated, and obtain description instruction
Practice the characteristic vector of sample, recycle l2Norm method is normalized, and the characteristic vector after normalization is carried out kernel function space
Mapping, the atom as dictionary built complete dictionary;
Set image category number as Q, for classification i, i=1,2 ..., Q, the N of selectioniThe M D feature vectors of individual training sample
It is normalized, obtain NiIndividual characteristic vectorExcessively complete dictionary as atomic building classification i
If being mapped as φ from data space to higher dimensional space, characteristic space vector inner product meter is carried out using gaussian radial basis function
The kernel function calculatedFor:
In formula, σ is constant;For space pixel p in Polarimetric SAR Image to be sortedmCharacteristic vector,For polarization SAR to be sorted
Space pixel p in imagenCharacteristic vector;
Mapped by kernel function, atomIt is each mapped toCross complete dictionary DiIt is mapped as Dφi, institute
There is the excessively complete dictionary of classificationAfter being mapped by kernel function, composition mapped complete dictionary Dφ:Dφ=
[Dφ1..., DφQ]∈RM*N(N=N1++NQ).
6. the Classification of Polarimetric SAR Image method based on nuclear sparse expression and space constraint according to claim 5, its feature
It is, the detailed process obtaining the optimum sparse coefficient of each image category training sample in step 5 is:
If Polarimetric SAR Image spatial neighborhood pixels p to be sortedmAnd pn, the characteristic vector of extraction is respectivelyWith
Kernel function is mapped asWithFor classification i,By excessively complete dictionary DφiRarefaction representation is:
In formula,It is characterized vectorKernel function mapping vector,For rarefaction representation it is
Number vector, xim1,xim2,L,For dictionary D excessively complete during rarefaction representationφiIn each atom weight coefficient, wherein nonzero coefficient
Subscript collection be combined into
According to space constraint relation, neighbor pnThe characteristic vector extractingNuclear mapping vectorFor classification i, use pixel pm
Identical atomBut different coefficients carry out linear expression:
It is set with u spatial neighborhood pixels, the eigenmatrix that the feature of extraction builds after kernel function mapping isFor classification i, by excessively complete dictionary DφiRarefaction representation is:
In formula, coefficient matrix XiMiddle nonzero term has identical bottom to mark
Therefore, the sparse representation model of space constraintFor:
In formula, | | X | |OK, 0Represent coefficient matrix XiIn non-zero row number, referred to as degree of rarefication K;
The sparse representation model of space constraint is converted into using degree of rarefication K as relaxation condition:
In formula | | | |FFor Frobenius norm,Optimum sparse coefficient training sample tried to achieve for classification i.
7. the Classification of Polarimetric SAR Image method based on nuclear sparse expression and space constraint according to claim 6, its feature
Be, realize in step 6 to the detailed process of the classification of all Polarimetric SAR Images to be sorted be:
Q excessively complete dictionary D of Q class image category composition1,D2,...DQ, obtain each classification instruction after kernel function mapping respectively
Practice the optimum sparse coefficient of sampleSolve acquisition Polarimetric SAR Image to be sorted diluter with different classes of optimum
Error r of sparse coefficient reconstruction signali(Φφ):
In formula | | | |2For norm;
Polarimetric SAR Image generic to be sorted is judged by minimum residual method, chooses with Polarimetric SAR Image residual error to be sorted
Classification representated by little reconstruction signal is as affiliated image category Class (Φ) of this Polarimetric SAR Image to be sorted:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610802709.2A CN106446935A (en) | 2016-09-05 | 2016-09-05 | Kernel sparse representation and spatial constraint-based polarimetric SAR image classification method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610802709.2A CN106446935A (en) | 2016-09-05 | 2016-09-05 | Kernel sparse representation and spatial constraint-based polarimetric SAR image classification method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106446935A true CN106446935A (en) | 2017-02-22 |
Family
ID=58163971
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610802709.2A Pending CN106446935A (en) | 2016-09-05 | 2016-09-05 | Kernel sparse representation and spatial constraint-based polarimetric SAR image classification method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106446935A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107239799A (en) * | 2017-05-26 | 2017-10-10 | 西安电子科技大学 | Polarization SAR image classification method with depth residual error net is decomposed based on Pauli |
CN109272017A (en) * | 2018-08-08 | 2019-01-25 | 太原理工大学 | The vibration signal mode identification method and system of distributed fiberoptic sensor |
CN109460751A (en) * | 2018-12-28 | 2019-03-12 | 内蒙古工业大学 | A method of the terrain classification based on scattering similitude |
CN109946698A (en) * | 2019-04-15 | 2019-06-28 | 北京市遥感信息研究所 | A kind of polarimetric synthetic aperture radar typical target feature base construction method and device |
CN111160296A (en) * | 2019-12-31 | 2020-05-15 | 中国科学院电子学研究所 | Landslide disaster detection method |
CN112633075A (en) * | 2020-11-30 | 2021-04-09 | 东南大学 | SAR image classification method based on depth-improved residual error network |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102122355A (en) * | 2011-03-15 | 2011-07-13 | 西安电子科技大学 | SAR (synthetic aperture radar) target identification method based on nuclear sparse representation |
CN104123555A (en) * | 2014-02-24 | 2014-10-29 | 西安电子科技大学 | Super-pixel polarimetric SAR land feature classification method based on sparse representation |
CN104182978A (en) * | 2014-08-22 | 2014-12-03 | 哈尔滨工程大学 | Hyper-spectral image target detection method based on spatially spectral kernel sparse representation |
CN104361346A (en) * | 2014-10-21 | 2015-02-18 | 西安电子科技大学 | K-SVD and sparse representation based polarization SAR (synthetic aperture radar) image classification method |
-
2016
- 2016-09-05 CN CN201610802709.2A patent/CN106446935A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102122355A (en) * | 2011-03-15 | 2011-07-13 | 西安电子科技大学 | SAR (synthetic aperture radar) target identification method based on nuclear sparse representation |
CN104123555A (en) * | 2014-02-24 | 2014-10-29 | 西安电子科技大学 | Super-pixel polarimetric SAR land feature classification method based on sparse representation |
CN104182978A (en) * | 2014-08-22 | 2014-12-03 | 哈尔滨工程大学 | Hyper-spectral image target detection method based on spatially spectral kernel sparse representation |
CN104361346A (en) * | 2014-10-21 | 2015-02-18 | 西安电子科技大学 | K-SVD and sparse representation based polarization SAR (synthetic aperture radar) image classification method |
Non-Patent Citations (1)
Title |
---|
孙良洁: "基于稀疏表示及上下文信息的PolSAR图像分类", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107239799A (en) * | 2017-05-26 | 2017-10-10 | 西安电子科技大学 | Polarization SAR image classification method with depth residual error net is decomposed based on Pauli |
CN109272017A (en) * | 2018-08-08 | 2019-01-25 | 太原理工大学 | The vibration signal mode identification method and system of distributed fiberoptic sensor |
CN109272017B (en) * | 2018-08-08 | 2022-07-12 | 太原理工大学 | Vibration signal mode identification method and system of distributed optical fiber sensor |
CN109460751A (en) * | 2018-12-28 | 2019-03-12 | 内蒙古工业大学 | A method of the terrain classification based on scattering similitude |
CN109946698A (en) * | 2019-04-15 | 2019-06-28 | 北京市遥感信息研究所 | A kind of polarimetric synthetic aperture radar typical target feature base construction method and device |
CN111160296A (en) * | 2019-12-31 | 2020-05-15 | 中国科学院电子学研究所 | Landslide disaster detection method |
CN111160296B (en) * | 2019-12-31 | 2024-01-26 | 中国科学院电子学研究所 | Landslide hazard detection method |
CN112633075A (en) * | 2020-11-30 | 2021-04-09 | 东南大学 | SAR image classification method based on depth-improved residual error network |
CN112633075B (en) * | 2020-11-30 | 2022-03-29 | 东南大学 | SAR image classification method based on depth-improved residual error network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Wu et al. | ORSIm detector: A novel object detection framework in optical remote sensing imagery using spatial-frequency channel features | |
CN107316013B (en) | Hyperspectral image classification method based on NSCT (non-subsampled Contourlet transform) and DCNN (data-to-neural network) | |
CN106446935A (en) | Kernel sparse representation and spatial constraint-based polarimetric SAR image classification method | |
Jia et al. | Gabor feature-based collaborative representation for hyperspectral imagery classification | |
CN102096825B (en) | Graph-based semi-supervised high-spectral remote sensing image classification method | |
Mishra et al. | Land cover classification of PALSAR images by knowledge based decision tree classifier and supervised classifiers based on SAR observables | |
CN107527023B (en) | Polarized SAR image classification method based on superpixels and topic models | |
Zhao et al. | Contrastive-regulated CNN in the complex domain: A method to learn physical scattering signatures from flexible PolSAR images | |
CN103886336B (en) | Polarized SAR image classifying method based on sparse automatic encoder | |
CN107992891B (en) | Multispectral remote sensing image change detection method based on spectral vector analysis | |
CN110458192B (en) | Hyperspectral remote sensing image classification method and system based on visual saliency | |
Cui et al. | Ratio-detector-based feature extraction for very high resolution SAR image patch indexing | |
Jia et al. | A two-stage feature selection framework for hyperspectral image classification using few labeled samples | |
CN109389080A (en) | Hyperspectral image classification method based on semi-supervised WGAN-GP | |
CN111881933A (en) | Hyperspectral image classification method and system | |
CN107330457B (en) | A kind of Classification of Polarimetric SAR Image method based on multi-feature fusion | |
CN104732244A (en) | Wavelet transform, multi-strategy PSO (particle swarm optimization) and SVM (support vector machine) integrated based remote sensing image classification method | |
CN109840542B (en) | Self-adaptive dimension decision tree classification method based on polarization characteristics | |
CN104008394B (en) | Semi-supervision hyperspectral data dimension descending method based on largest neighbor boundary principle | |
CN101853509A (en) | SAR (Synthetic Aperture Radar) image segmentation method based on Treelets and fuzzy C-means clustering | |
Gou et al. | Coastal zone classification with fully polarimetric SAR imagery | |
CN105913092A (en) | Figure regular hyperspectral image band selection method based on subspace learning | |
CN105894013A (en) | Method for classifying polarized SAR image based on CNN and SMM | |
CN111680579B (en) | Remote sensing image classification method for self-adaptive weight multi-view measurement learning | |
CN104680182B (en) | Polarization SAR sorting technique based on NSCT and differentiation dictionary learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170222 |
|
RJ01 | Rejection of invention patent application after publication |