CN107491734B - Semi-supervised polarimetric SAR image classification method based on multi-core fusion and space Wishart LapSVM - Google Patents

Semi-supervised polarimetric SAR image classification method based on multi-core fusion and space Wishart LapSVM Download PDF

Info

Publication number
CN107491734B
CN107491734B CN201710589220.6A CN201710589220A CN107491734B CN 107491734 B CN107491734 B CN 107491734B CN 201710589220 A CN201710589220 A CN 201710589220A CN 107491734 B CN107491734 B CN 107491734B
Authority
CN
China
Prior art keywords
wishart
spatial
polarization
sar image
lapsvm
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
Application number
CN201710589220.6A
Other languages
Chinese (zh)
Other versions
CN107491734A (en
Inventor
王敏
王勇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Suzhou Wenjie Sensing Technology Co ltd
Original Assignee
Suzhou Wenjie Sensing Technology Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Suzhou Wenjie Sensing Technology Co ltd filed Critical Suzhou Wenjie Sensing Technology Co ltd
Priority to CN201710589220.6A priority Critical patent/CN107491734B/en
Publication of CN107491734A publication Critical patent/CN107491734A/en
Application granted granted Critical
Publication of CN107491734B publication Critical patent/CN107491734B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Astronomy & Astrophysics (AREA)
  • Remote Sensing (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a semi-supervised polarimetric SAR image classification method based on multi-core fusion and space Wishart LapSVM, which mainly solves the problem of low classification precision caused by fewer labeled samples of polarimetric synthetic aperture radar fully polarimetric SAR images in the existing classification method. The method comprises the following implementation steps: obtaining a polarization correlation matrix T, extracting a polarization characteristic vector of the polarization correlation matrix T, carrying out normalization processing, establishing a training sample set, constructing a Spatial-Wishart manifold regular term and the like, calculating the classification accuracy and outputting a polarization SAR image classification result. The method solves the problem of low classification accuracy of the traditional unsupervised polarized SAR image, also avoids the defects of difficult manual marking and high cost caused by the fact that a large amount of label data is needed in a supervised classification method, obtains a better classification effect by jointly utilizing a small amount of labeled data and a large amount of unlabeled cheap data, and can be used for target classification, detection and identification of the polarized SAR image.

Description

Semi-supervised polarimetric SAR image classification method based on multi-core fusion and space Wishart LapSVM
Technical Field
The invention belongs to the technical field of image processing, and relates to a polarized SAR image classification method which can be used in the technical fields of ground feature classification, target identification and the like of polarized SAR images.
Background
The polarimetric sar (polarimetric sar) is a synthetic aperture radar capable of performing full-polarization measurement on a target, and performs full-polarization measurement imaging on the target by measuring and recording phase difference information of combined echoes of different polarization states. The polarized SAR data contain richer target scattering information, can more comprehensively express and describe a target, improves the identification capability of ground objects, has the advantages of all weather, all-day time, high resolution and the like, and has very outstanding advantages in the aspects of target detection and identification, classification, parameter inversion and the like, so that the polarized SAR data are widely applied to various fields of military affairs, agriculture, navigation and the like. At present, the polarized SAR imaging technology has been developed rapidly, but the corresponding polarized SAR image processing technology can not meet the existing requirements. Therefore, there is an urgent need to develop an image processing technique capable of omnidirectionally interpreting a polarized SAR image.
The existing polarized SAR image classification method can be classified into supervised classification and unsupervised classification according to whether labeled data is needed in the learning process. Supervised learning is to obtain an optimal model through training of a large number of labeled samples, and then realize prediction of unlabeled data by using the optimal model, such as a polarization covariance matrix supervised classification method based on complex Wishart distribution proposed by Lee et al, a classification method based on a back propagation neural network proposed by Heermann et al, and the like. The unsupervised learning finishes classification or clustering by mining the internal structure and inherent attributes of data, and the learning process does not need label data, such as an H/alpha unsupervised classification method proposed by Cloude et al, a polarized SAR image unsupervised classification algorithm based on Freeman decomposition proposed by Lee et al, and the like.
The semi-supervised learning is a learning method combining supervised learning and unsupervised learning, a small number of marked samples are utilized, a large number of cheap unmarked samples are combined, inherent structures and information contained in the unmarked samples are fully utilized to improve the classification effect, the problems of difficult manual marking, high cost and the like caused by using a large amount of marked data in the supervised learning are avoided, and the defect of low classification precision of the unsupervised learning is effectively overcome.
Because the existing supervised and unsupervised polarimetric SAR image classification methods have certain limitations, it is urgent in the technical field to research an effective semi-supervised polarimetric SAR image classification method.
Disclosure of Invention
The invention aims to overcome the inherent defects and shortcomings of a supervised and unsupervised polarimetric SAR image classification method in the current field, provides a semi-supervised polarimetric SAR image classification method based on multi-core fusion and space Wishart-LapSVM, and improves classification accuracy by using a large amount of cheap unmarked samples on the basis of a small amount of marked samples.
The technical scheme of the invention is based on clustering hypothesis and space consistency hypothesis of polarized SAR data, a Spatial-Wishart manifold regular term is constructed, and high-dimensional mapping is realized on polarized feature vectors by using a multi-core weighting fusion mode, so that semi-supervised polarized SAR terrain classification based on LapSVM is realized, and a large amount of cheap unmarked samples are fully utilized to improve the classification effect. The specific implementation scheme is as follows:
(1) inputting a polarized SAR image to be classified to obtain a polarized coherent matrix T of the image;
(2) constructing a polarization characteristic vector based on a polarization coherent matrix T in the polarization SAR image and combining spatial information, and performing characteristic normalization processing;
(3) randomly selecting 1% of data from each type of polarized SAR images to be classified for marking, and combining 30% of non-marked data to jointly form a training sample set;
(4) designing a similarity measurement criterion between polarized SAR image pixel points based on a space consistency assumption and a polarized coherent matrix of polarized SAR data obeying complex Wishart distribution, and constructing a Spatial-Wishart manifold regular term according to a clustering assumption;
(5) selecting a group of kernel functions, calculating a fusion kernel matrix based on a multi-kernel weighted fusion strategy, and performing high-dimensional mapping on the polarization feature vectors;
(6) training Spatial-Wishart LapSVM by using a training sample set, and performing rapid optimization solution based on a PCG algorithm;
(7) label prediction is carried out on the unmarked training sample and the test sample by utilizing the trained Spatial-Wishart LapSVM model and based on a one-vs-one multi-classification strategy;
(8) calculating classification accuracy and outputting a polarized SAR image classification result;
compared with the prior art, the invention has the following advantages:
1. the invention belongs to a semi-supervised polarimetric SAR image classification algorithm, which can obtain the internal inherent structure and attribute of a data set by only using a small amount of expensive marked data and simultaneously fully using a large amount of cheap unmarked samples, thereby improving the classification precision and obtaining more advantageous classification effect with less marking cost;
2. based on the clustering hypothesis and the Spatial consistency hypothesis of the polarized SAR image data, the method utilizes the characteristic that a polarized coherent matrix obeys complex Wishart distribution, combines Spatial neighborhood information, and constructs a similarity measurement criterion among polarized SAR pixel points, thereby constructing a Spatial-Wishart manifold regular term, improving the classification precision by utilizing a large number of unmarked samples, and improving the classification effect;
3. the invention carries out high-dimensional feature mapping of the polarization feature vector in a multi-core fusion mode, thereby effectively solving the heterogeneous characteristic of the polarization feature and avoiding the defect of a single kernel function.
4. According to the invention, the rapid optimization solution of the LapSVM model is carried out by utilizing the PCG algorithm in an original form, so that the solution speed is greatly improved.
Drawings
FIG. 1 is a block flow diagram of the steps performed by the present invention;
FIG. 2 is a PauliRGB composite plot of polarized SAR data used in the simulation of the present invention;
FIG. 3 is a true terrain labeling map of a polarized SAR image used by the present invention;
FIG. 4 is a graph of the results of a polarized SAR image classification experiment according to the present invention (Spatial-Wishart LapSVM);
fig. 5 is a diagram of a result of a classification experiment of a polarized SAR image of a supervised Wishart;
FIG. 6 is a graph of the classification experimental results of the polarization SAR image of KNN;
fig. 7 is a schematic diagram of distances between spatial neighborhood points.
Detailed Description
The invention relates to a semi-supervised polarimetric SAR image classification method based on multi-core fusion and space Wishart LapSVM, which comprises the following specific implementation steps of referring to FIG. 1:
step 1, inputting a polarized SAR image to be classified to obtain a polarized coherent matrix T of the image.
Referring to fig. 2, the polarized SAR image is a netherlands farmland map, the land feature classes to be classified comprise bare land, potato, beet, barley, pea and wheat, different colors in the map represent different land feature classes, and the true land feature class label map is shown in fig. 3.
The invention realizes the semi-supervised ground object classification of the polarized SAR image, and verifies the actual classification effect of the invention by using the classification experiment of 6 types of ground objects in the image.
And 2, obtaining a polarization characteristic vector based on the polarization coherent matrix T in the polarization SAR image, and performing characteristic normalization processing.
(2a) The polarization coherent matrix of each pixel point of the polarization SAR image is expressed by a matrix with the dimension of 3 multiplied by 3:
Figure GDA0002913594590000031
wherein, Tij＝Tji *,i≠j。
(2b) The polarization coherent matrix contains all polarization information of the polarization SAR data and has the capability of expressing the characteristics of the polarization SAR data. Accordingly, we express the polarization vector form of a single pixel point as a form with dimensions of 9 × 1 as follows:
I＝(|T11|2,|T22|2,|T33|2,|Re[T12]|2,|Re[T13]|2,|Re[T23]|2,|Im[T21]|2,|Im[T23]|2,|Im[T31]|2)
(2c) by using spatial information, the feature vector of each pixel point is expressed as a combination of the features of a plurality of pixel points in the neighborhood around the pixel point, and can be expressed as:
xi＝[......,Ii-1,Ii,Ii+1,......]；
(2c) carrying out normalization processing on the sample characteristic vector of the whole to-be-classified polarized SAR image;
and 3, randomly selecting a training sample set.
(3a) Randomly selecting 1% of sample points from each category of the polarized SAR images to be classified for marking, and using the sample points as a marked sample set;
(3b) uniformly selecting 30% of unmarked sample points of the polarized SAR image to be classified as an unmarked sample set;
(3c) combining the marked sample set and the unmarked sample set to jointly form a training sample set;
and 4, constructing a Spatial-Wishart manifold regular term based on the clustering assumption and the Spatial consistency assumption of the polarized SAR data.
(4a) The polarization coherent matrix based on the polarization SAR obeys complex Wishart distribution, and the complex Wishart distance between any two pixel points is designed as follows:
the polarization coherence matrix T of the samples with view n obeys a complex Wishart distribution, whose probability density function is:
Figure GDA0002913594590000041
wherein k (n, q) ═ piq(q-1)/2Γ(n)···Γ(n-q+1)
In the formula, sigma represents the mathematical expectation of T, n is a visual number, k is a normalization coefficient, Γ is a Gamma function, and Tr is the trace of the matrix. Giving a polarization coherent matrix T of a pixel point j and a pixel point iiThe probability of occurrence is
Figure GDA0002913594590000042
The above formula takes log likelihood, and the similarity between the pixel point i and the pixel point j can be obtained:
Lij(Ti|Tj)＝qnln(n)+(n-q)ln|Ti|-nTr(Tj -1Ti)-nln(Tj)-ln(k(n,q))
similarly, the similarity between the pixel point j and the pixel point i can be obtained:
Lji(Tj|Ti)＝qnln(n)+(n-q)ln|Tj|-nTr(Ti -1Tj)-nln(Ti)-ln(k(n,q))
therefore, the similarity between any two pixel points in the polarized SAR image is defined as:
Figure GDA0002913594590000043
where C is a constant.
Suppose the prior probability p (T) of all pixel pointsi) Is equal to, if
Figure GDA0002913594590000044
The similarity between pixel points i and j is greater than the similarity between pixel points k and j.
Further simplifying the logarithm likelihood function, removing a constant term and an irrelevant term in the formula, and taking an inverse number to obtain the complex Wishart distance between the pixel points:
Figure GDA0002913594590000051
(4b) and (4) constructing Spatial-Wishart similarity between the polarized SAR pixel points based on the complex Wishart distance measurement criterion and the Spatial consistency assumption between the polarized SAR pixel points defined in the step (4a) and the Spatial neighborhood information.
The actual ground object distance from the central pixel point in the simulated spatial neighbor window, and the distance r between the spatial neighbor points is defined as shown in fig. 7:
wherein width is the width of the adjacent window, r1＝1，
Figure GDA0002913594590000052
r3＝2，
Figure GDA0002913594590000053
Namely:
1-near neighbor point (such as point a) which is 1 unit away from the central pixel point in space;
spatial distance from center pixel
Figure GDA0002913594590000054
Of a unit
Figure GDA0002913594590000055
-a neighborhood (e.g. point b);
2-near neighbor points (such as c points) which are 2 units away from the central pixel point in space;
spatial distance from center pixel
Figure GDA0002913594590000056
Of a unit
Figure GDA0002913594590000057
-neighbors (e.g. d);
spatial distance from center pixel
Figure GDA0002913594590000058
Of a unit
Figure GDA0002913594590000059
-a neighbor point (e.g. point e);
setting k neighbor number and width of a Spatial neighborhood window, wherein the final Spatial-Wishart similarity is defined as follows:
Figure GDA00029135945900000510
wherein r is the space distance between pixel points in the space neighborhood window, sigma is the parameter of the Gaussian function, and Nk(Ti) Is TiK-nearest neighbors of, Ns(Ti) Is TiThe spatial neighbors.
The Spatial-Wishat similarity based on the Spatial neighbor constraint can fully utilize Spatial neighborhood information, eliminate noise interference and improve classification accuracy.
(4c) The inherent manifold structure of the data is represented by constructing a neighbor graph in which nodes represent labeled and unlabeled samples, which are represented by edge weights WijAnd (4) connecting. By using a combination of labelled and unlabelled samplesBased on the clustering assumption and the Spatial consistency assumption, the structured Spatial-Wishart manifold regular term of the LapSVM is defined as follows:
Figure GDA0002913594590000061
wherein L is the graph laplace matrix L ═ D-W; d is a diagonal matrix whose diagonal elements are degrees of each vertex, i.e.
Figure GDA0002913594590000062
W is an edge weight matrix whose element calculation is as shown in (4 b).
And 5, calculating a fusion kernel matrix based on a multi-kernel weighted fusion strategy, and performing high-dimensional mapping on the polarized feature vectors.
(5a) Selecting a group of kernel functions, wherein a Gaussian kernel, a linear kernel and a polynomial kernel are selected and respectively marked as kr,kl,kp；
(5b) Based on the set kernel function group, performing multi-kernel weighted fusion high-dimensional mapping on the polarized SAR feature vector, namely calculating a Gram matrix as follows:
K＝(kij)＝((1-μ1-μ2)·kr(xi,xj)+μ1·kl(xi,xj)+μ2·kp(xi,xj))
＝(1-μ1-μ2)Kr+μ1Kl+μ2Kp i,j＝1,···l+u
wherein K is a Gram matrix, mu1,μ2∈[0，1]An adjustment factor for the proportion of each kernel function;
step 6, training Spatial-Wishart LapSVM by using a training sample set, and carrying out rapid optimization solution based on a PCG algorithm;
(6a) as described in step (4), constructing a Spatial-Wishart manifold regularization term by combining the clustering assumption and the Spatial consistency assumption on the polarized SAR data:
Figure GDA0002913594590000063
(6b) constructing a Spatial-Wishart LapSVM model by including the Spatial-Wishart manifold regularization term:
Figure GDA0002913594590000064
(6c) the representation theorem indicates that the LapSVM semi-supervised framework is in HKThe solution in space can be expressed as:
Figure GDA0002913594590000065
thus, the problem solved optimally in raw form can be expressed as:
Figure GDA0002913594590000071
see s.melacci, m.belkin.display Support Machines Trained in the primary, journal of Machine Learning Research,2011,12(3): 1149-.
And 7, performing label prediction on the unmarked training sample and the test sample by using the trained LapSVM model and based on a one-vs-one-many classification strategy.
(7a) According to the class number m of the samples to be classified and the one-vs-one multi-classification strategy, training is needed
Figure GDA0002913594590000072
A Spatial-Wishart LapSVM classification model;
(7b) for unlabelled sample sets and test sample sets, using training
Figure GDA0002913594590000073
A SpatiaAnd respectively predicting and voting by the l-Wishart LapSVM binary classification model, and taking the category with the maximum votes as the category of the pixel point.
And 8, calculating the classification accuracy and outputting the classification result of the polarized SAR image.
The effects of the present invention can be further illustrated by the following simulations:
1. simulation conditions and simulation contents:
the example completes the classification simulation experiment of the KNN and supervision Wishart polarization SAR image and the invention on a Matlab R2013a running platform under an Intel (R) core (TM) i3 CPU 2.53GHz Windows 7 system.
2. Content of simulation experiment
A. Simulation of polarized SAR image classification algorithm
The present invention was applied to a polarized SAR image of a 300 × 270 dutch farmland, as shown in fig. 2, of which the ground object categories to be classified include six regions of bare land, potato, beet, barley, pea and wheat. FIG. 4 is a diagram showing the results of a simulation experiment in which the method of the present invention is used to classify the image of FIG. 2, and the classification results are labeled as shown in the figure.
B. Simulation for supervising Wishart and KNN polarization SAR image classification algorithm
The existing supervised Wishart polarimetric SAR image classification algorithm is applied to the polarimetric SAR image of 300 x 270 Dutch farmland as shown in figure 2, the simulation experiment result is shown in figure 5, and various classification result labels are shown in the figure.
The KNN polarized SAR classification algorithm is applied to the polarized SAR image of 300 x 270 Dutch farmland as shown in figure 2, the simulation experiment result is shown in figure 6, and the classification result labels of various types are shown in the figure.
3. Simulation experiment results
As can be seen from fig. 4, the method has a good subjective visual effect on the simulation experiment result obtained by classifying the polarized SAR image, a high classification accuracy, high regional consistency, and good differentiation on the classification result of the 6 types of ground objects to be classified in fig. 2.
As can be seen from fig. 5 and 6, the simulation experiment result obtained by the existing supervised Wishart has a general subjective visual effect, serious misclassification, blurred edges, and low region consistency, and has poor differentiation on the classification result of the 6 types of ground objects to be classified in fig. 2.
The simulation experiments can show that the method has certain advantages aiming at the classification of the polarized SAR images, overcomes the defects of the prior art applied to the polarized SAR images, can obtain the effect of higher classification accuracy by only using a small amount of labeled samples, and greatly reduces the labeling cost of the samples while obtaining accurate classification.
In conclusion, the classification effect of the method for the polarized SAR image is obviously better than that of the existing KNN and supervision Wishart classification technology for the polarized SAR image. Therefore, the invention effectively overcomes various defects in the prior art and has high industrial utilization value.
The foregoing embodiments are merely illustrative of the principles and utilities of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or change the above-mentioned embodiments without departing from the spirit and scope of the present invention. Accordingly, it is intended that all equivalent modifications or changes which can be made by those skilled in the art without departing from the spirit and technical spirit of the present invention be covered by the claims of the present invention.

Claims (6)

1. A semi-supervised polarimetric SAR image classification method based on multi-core fusion and space Wishart LapSVM comprises the following steps:
(1) inputting a polarized SAR image to be classified to obtain a polarized coherent matrix T of the image;
(2) obtaining a polarization characteristic vector based on a polarization coherent matrix T in the polarization SAR image, and performing characteristic normalization processing;
(3) randomly selecting 1% of data from each type of polarized SAR images to be classified for marking, and combining 30% of non-marked data to jointly form a training sample set;
(4) designing a similarity measurement criterion between polarized SAR image pixel points based on a space consistency assumption and a polarized coherent matrix of polarized SAR data obeying complex Wishart distribution, and constructing a Spatial-Wishart manifold regular term according to a clustering assumption;
(5) selecting a group of kernel functions, calculating a fusion kernel matrix based on a multi-kernel weighted fusion strategy, and performing high-dimensional mapping on the polarization feature vectors;
(6) training Spatial-Wishart LapSVM by using a training sample set, and performing rapid optimization solution based on a PCG algorithm;
(7) label prediction is carried out on the unmarked training sample and the test sample by utilizing the trained Spatial-Wishart LapSVM model and based on a one-vs-one multi-classification strategy;
(8) and calculating the classification accuracy and outputting a polarized SAR image classification result.
2. The semi-supervised polarimetric SAR image classification method based on multi-core fusion and space Wishart LapSVM as claimed in claim 1, wherein the step (2) of obtaining the eigenvector of each pixel point based on the polarimetric SAR image polarization correlation matrix is performed according to the following steps:
(2a) the polarization coherent matrix of each pixel point of the polarization SAR image is expressed by a matrix with the dimension of 3 multiplied by 3:
Figure FDA0002901592280000011
wherein, Tij＝Tji *,i≠j，
(2b) The polarization coherent matrix contains all polarization information of the polarization SAR data and has the capability of expressing the characteristics of the polarization SAR data, and accordingly, the polarization vector form of a single pixel point is expressed as a form with the following dimensions of 1 multiplied by 9:
I＝(|T11|2,|T22|2,|T33|2,|Re[T12]|2,|Re[T13]|2,|Re[T23]|2,|Im[T21]|2,|Im[T23]|2,|Im[T31]|2)
(2c) by using spatial information, the feature vector of each pixel point is expressed as a combination of the features of a plurality of pixel points in the neighborhood around the pixel point, and can be expressed as:
xi＝[......,Ii-1,Ii,Ii+1,......]；
(2d) and carrying out normalization processing on the sample characteristic vector of the whole to-be-classified polarized SAR image.
3. The semi-supervised polarimetric SAR image classification method based on multi-kernel fusion and Spatial Wishart LapSVM as claimed in claim 1, wherein the constructing Spatial-Wishart manifold regularization term in step (4) is performed by the following steps:
(4a) a polarization coherent matrix based on polarization SAR pixel points obeys complex Wishart distribution, and the Spatial-Wishart similarity between any two pixel points is designed:
the polarization coherence matrix T of the samples with view n obeys q-dimensional complex Wishart distribution, and its probability density function is:
Figure FDA0002901592280000021
wherein k (n, q) ═ piq(q-1)/2Γ(n)···Γ(n-q+1)
In the formula, sigma represents the mathematical expectation of T, n is a visual number, k is a normalization coefficient, Gamma is a Gamma function, and Tr is the trace of the matrix; giving a polarization coherent matrix T of a pixel point j and a pixel point iiThe probability of occurrence is
Figure FDA0002901592280000022
Taking the log likelihood of the above equation, we get:
L(Ti|Tj)＝qnln(n)+(n-q)ln|Ti|-nTr(Tj -1Ti)-nln(Tj)-ln(k(n,q))
according to the maximum likelihood criterion, the prior probability p (T) of all pixel points is assumedi) Is equal toIf, if
Figure FDA0002901592280000023
The similarity between the pixel points i and j is greater than the similarity between the pixel points k and j;
further simplifying the logarithm likelihood function, removing a constant term and an irrelevant term in the formula, and taking an inverse number to obtain the complex Wishart distance between the pixel points:
Figure FDA0002901592280000031
(4b) based on the similarity measurement criterion between the polarized SAR pixel points defined in the step (4a), simultaneously, the Spatial neighborhood information is jointly utilized to effectively overcome noise interference existing in the polarized SAR image, and Spatial-Wishart similarity is constructed:
setting k neighbor number and width of a Spatial neighborhood window, wherein the final Spatial-Wishart similarity is defined as follows:
Figure FDA0002901592280000032
wherein r is the space distance between pixel points in the space neighborhood window, sigma is the parameter of the Gaussian function, and Nk(Ti) Is TiK-nearest neighbors of, Ns(Ti) Is TiSpatial neighbor;
the Spatial-Wishat similarity based on the Spatial neighbor constraint can fully utilize Spatial neighborhood information, eliminate noise interference and improve classification accuracy;
(4c) the inherent manifold structure of the data is represented by constructing a neighbor graph in which nodes represent labeled and unlabeled samples, which are represented by edge weights WijConnection of；Spatial-Wisha of constructed LapSVM based on clustering hypothesis and Spatial consistency hypothesis by jointly utilizing labeled and unlabeled samplesThe rt manifold regularization term is defined as follows
Figure FDA0002901592280000033
Wherein L is the graph laplace matrix L ═ D-W; d is a diagonal matrix whose diagonal elements are degrees of each vertex, i.e.
Figure FDA0002901592280000034
W is an edge weight matrix whose element calculation is as shown in (4 b).
4. The semi-supervised polarimetric SAR image classification method based on multi-core fusion and space Wishart LapSVM as claimed in claim 1, wherein the multi-core fusion weighting-based strategy in step (5) realizes high-dimensional mapping of feature vectors, and the method comprises the following steps:
(5a) selecting a group of kernel functions, selecting a Gaussian kernel, a linear kernel and a polynomial kernel which are respectively marked as Kr、Kl、Kp；
(5b) Based on the set kernel function group, performing multi-kernel weighted fusion high-dimensional mapping on the polarized SAR feature vector, namely:
K＝(kij)＝((1-μ1-μ2)·kr(xi，xj)+μ1·kl(xi，xj)+μ2·kp(xi，xj))
＝(1-μ1-μ2)Kr+μ1Kl+μ2Kp i，j＝1，…l+u
k is Gram matrix,. mu.1,μ2∈[0,1]And u represents the number of unmarked samples as a regulating factor of the proportion occupied by each kernel function.
5. The semi-supervised polarimetric SAR image classification method based on multi-core fusion and Spatial Wishart LapSVM as claimed in claim 4, wherein the training and solving based on Spatial-Wishart LapSVM in step (6) are performed according to the following steps:
(6a) as described in step (4), constructing a Spatial-Wishart manifold regularization term by combining the clustering assumption and the Spatial consistency assumption on the polarized SAR data:
Figure FDA0002901592280000041
(6b) constructing a Spatial-Wishart LapSVM model by including the Spatial-Wishart manifold regularization term:
Figure FDA0002901592280000042
(6c) the representation theorem indicates that the LapSVM semi-supervised framework is in HKThe solution in space can be expressed as:
Figure FDA0002901592280000043
thus, the problem solved optimally in raw form can be expressed as:
Figure FDA0002901592280000044
6. the semi-supervised polarimetric SAR image classification method based on multi-core fusion and Spatial Wishart LapSVM as claimed in claim 1, wherein the label prediction is performed on the unlabeled sample set and the test sample set based on the trained Spatial-Wishart LapSVM model in step (7) according to the following steps:
(7a) according to the class number m of the samples to be classified and the one-vs-one multi-classification strategy, training is needed
Figure FDA0002901592280000045
A Spatial-Wishart LapSVM classification model;
(7b) for unlabelled sample sets and test sample sets, using training
Figure FDA0002901592280000046
And (4) respectively predicting by using the Spatial-Wishart LapSVM binary classification model, voting, and taking the category with the maximum vote number as the category of the pixel point.
CN201710589220.6A 2017-07-19 2017-07-19 Semi-supervised polarimetric SAR image classification method based on multi-core fusion and space Wishart LapSVM Active CN107491734B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710589220.6A CN107491734B (en) 2017-07-19 2017-07-19 Semi-supervised polarimetric SAR image classification method based on multi-core fusion and space Wishart LapSVM

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710589220.6A CN107491734B (en) 2017-07-19 2017-07-19 Semi-supervised polarimetric SAR image classification method based on multi-core fusion and space Wishart LapSVM

Publications (2)

Publication Number Publication Date
CN107491734A CN107491734A (en) 2017-12-19
CN107491734B true CN107491734B (en) 2021-05-07

Family

ID=60643863

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710589220.6A Active CN107491734B (en) 2017-07-19 2017-07-19 Semi-supervised polarimetric SAR image classification method based on multi-core fusion and space Wishart LapSVM

Country Status (1)

Country Link
CN (1) CN107491734B (en)

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108960276B (en) * 2018-05-08 2022-02-15 南京理工大学 Sample expansion and consistency discrimination method for improving spectral image supervision classification performance
CN109670537A (en) * 2018-12-03 2019-04-23 济南大学 The full attribute weight fuzzy clustering method of multicore based on quasi- Monte Carlo feature
CN110008924A (en) * 2019-04-15 2019-07-12 中国石油大学(华东) A kind of semi-supervised automark method and device towards atural object in Hyperspectral imaging
CN110717513B (en) * 2019-09-05 2023-04-07 华南理工大学 Zero-sample deep-sea biological picture classification method based on multiple classifiers
CN111695466B (en) * 2020-06-01 2023-03-24 西安电子科技大学 Semi-supervised polarization SAR terrain classification method based on feature mixup
CN112580687A (en) * 2020-11-23 2021-03-30 北京化工大学 Unsupervised classification method based on millimeter wave full-polarization SAR image
CN112507621B (en) * 2020-12-16 2023-04-07 电子科技大学 Radar interference semi-supervised identification method based on label propagation
CN113409891B (en) * 2021-05-25 2023-02-03 电子科技大学长三角研究院(衢州) Method, device, equipment and storage medium for predicting DNA6mA modification class
CN115223050B (en) * 2022-04-28 2023-08-18 湖北工程学院 Polarized satellite image analysis method

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104408466A (en) * 2014-11-17 2015-03-11 中国地质大学（武汉） Semi-supervision and classification method for hyper-spectral remote sensing images based on local stream type learning composition
CN106067042A (en) * 2016-06-13 2016-11-02 西安电子科技大学 Polarization SAR sorting technique based on semi-supervised degree of depth sparseness filtering network
CN106096627A (en) * 2016-05-31 2016-11-09 河海大学 The Polarimetric SAR Image semisupervised classification method that considering feature optimizes
CN106127224A (en) * 2016-06-13 2016-11-16 西安电子科技大学 Based on the semi-supervised polarization SAR sorting technique quickly auctioning figure
CN106203487A (en) * 2016-06-30 2016-12-07 北京航空航天大学 A kind of image classification method based on Multiple Kernel Learning Multiple Classifier Fusion and device

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104408466A (en) * 2014-11-17 2015-03-11 中国地质大学（武汉） Semi-supervision and classification method for hyper-spectral remote sensing images based on local stream type learning composition
CN106096627A (en) * 2016-05-31 2016-11-09 河海大学 The Polarimetric SAR Image semisupervised classification method that considering feature optimizes
CN106067042A (en) * 2016-06-13 2016-11-02 西安电子科技大学 Polarization SAR sorting technique based on semi-supervised degree of depth sparseness filtering network
CN106127224A (en) * 2016-06-13 2016-11-16 西安电子科技大学 Based on the semi-supervised polarization SAR sorting technique quickly auctioning figure
CN106203487A (en) * 2016-06-30 2016-12-07 北京航空航天大学 A kind of image classification method based on Multiple Kernel Learning Multiple Classifier Fusion and device

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
Laplacian Support Vector Machines Trained in the Primal;Stefano Melacci等;《Journal of Machine Learning Research》;20110311;第12卷(第3期);1149-1184 *
一种改进的Laplacian_SVM的SAR图像分割算法;刘若辰等;《红外与毫米波学报》;20110615;第30卷(第3期);250-254、259 *
半监督支持向量机分类方法研究;陈永健;《中国优秀硕士学位论文全文数据库_信息科技辑》;20150215;1-65 *

Also Published As

Publication number Publication date
CN107491734A (en) 2017-12-19

Similar Documents

Publication Publication Date Title
CN107491734B (en) Semi-supervised polarimetric SAR image classification method based on multi-core fusion and space Wishart LapSVM
Li et al. Learning deep semantic segmentation network under multiple weakly-supervised constraints for cross-domain remote sensing image semantic segmentation
Li et al. A deep learning method for change detection in synthetic aperture radar images
CN108388927B (en) Small sample polarization SAR terrain classification method based on deep convolution twin network
CN102096825B (en) Graph-based semi-supervised high-spectral remote sensing image classification method
Liu et al. Enhancing spectral unmixing by local neighborhood weights
Ayerdi et al. Spatially regularized semisupervised ensembles of extreme learning machines for hyperspectral image segmentation
Wu et al. Multiscale CNN with autoencoder regularization joint contextual attention network for SAR image classification
Chen et al. Agricultural remote sensing image cultivated land extraction technology based on deep learning
Liu et al. Deep fuzzy graph convolutional networks for PolSAR imagery pixelwise classification
Zhang et al. Polygon structure-guided hyperspectral image classification with single sample for strong geometric characteristics scenes
CN109034213B (en) Hyperspectral image classification method and system based on correlation entropy principle
CN108985161B (en) Low-rank sparse representation image feature learning method based on Laplace regularization
He et al. Multi-spectral remote sensing land-cover classification based on deep learning methods
CN109712149A (en) A kind of image partition method based on wavelet energy and fuzzy C-mean algorithm
Wu et al. Cost-sensitive latent space learning for imbalanced PolSAR image classification
CN101877134B (en) Robust tracking method of target in airport monitoring video
Wang et al. Spatial-spectral clustering with anchor graph for hyperspectral image
CN107423771B (en) Two-time-phase remote sensing image change detection method
Bagwari et al. A comprehensive review on segmentation techniques for satellite images
Yao et al. Cloud Detection in Optical Remote Sensing Images with Deep Semi-supervised and Active Learning
Huang et al. Aerial image classification by learning quality-aware spatial pyramid model
Jiang et al. A robust end-to-end deep learning framework for detecting Martian landforms with arbitrary orientations
Jia et al. Structure-Adaptive Convolutional Neural Network for Hyperspectral Image Classification
Ma et al. MBSI-Net: Multimodal Balanced Self-Learning Interaction Network for Image Classification

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