CN104504393B - Polarimetric SAR Image semisupervised classification method based on integrated study - Google Patents
Polarimetric SAR Image semisupervised classification method based on integrated study Download PDFInfo
- Publication number
- CN104504393B CN104504393B CN201410748929.2A CN201410748929A CN104504393B CN 104504393 B CN104504393 B CN 104504393B CN 201410748929 A CN201410748929 A CN 201410748929A CN 104504393 B CN104504393 B CN 104504393B
- Authority
- CN
- China
- Prior art keywords
- sar image
- sample
- scattering
- polarized sar
- polarized
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
-
- 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
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
Abstract
The invention discloses a kind of Polarimetric SAR Image semisupervised classification method based on integrated study, implementation step is：(1) a width Polarimetric SAR Image is inputted；(2) scattering signatures of Polarimetric SAR Image, polarization parameter and textural characteristics are extracted；(3) the anti-noise factor of sample is calculated using Knn algorithms；(4) according to the weight of sample, training sample is extracted；(5) training obtains grader, and calculates error rate；(6) weight of sample is adjusted according to error rate；(7) according to Wishart distances, the high unmarked sample of confidence level is selected, is added to training set, and assign weight；(8) judge whether to meet stop condition, satisfaction then terminates and output category result, otherwise goes to (4)；Present invention combination polarization parameter, scattering signatures and textural characteristics, can be described more fully with atural object truth, and utilize the unmarked sample in part, enrich original training sample, obtain preferable nicety of grading.
Description
Technical Field
The invention belongs to the technical field of image processing, and particularly relates to a polarimetric SAR image semi-supervised classification method based on ensemble learning.
Background
Due to its great advantages, polarized SAR (synthetic aperture radar) is widely used in both civilian and military industries, such as target detection, ground object classification, parametric inversion, and so on. Among them, the application in the aspect of ground object classification is a very important one of many applications of polarized SAR.
The polarized SAR image classification method roughly comprises two steps, namely feature selection and classifier determination. How to obtain valuable features and a good classification strategy is the key point of the polarized SAR image classification problem. The characteristic aspects are polarization parameters, scattering matrixes and texture characteristics which are commonly used. Cloude in 1997 proposes a well-known H/alpha classification method, i.e. the scattering entropy H and the scattering angle alpha are obtained by scattering matrix decomposition, and one polarized SAR image can be classified into 8 types according to the two values. And then, various polarimetric SAR image classification methods based on different decompositions appear. Since the real feature has a complex class structure, it is not comprehensive to characterize the feature by a certain decomposition.
Generally, the labels of the polarized SAR images are obtained by manual labeling, and thus, errors are easily introduced by human factors, and the influence of the error labels on the classifier is fatal. In the general polarization SAR classification method, the problem is rarely considered, and actually, the influence on the classifier due to error labels introduced by human factors is not negligible.
The polarized SAR obtains rich and comprehensive target information by carrying out full-polarization measurement on the target, so that the polarized SAR has very obvious advantages in the classification problem. At present, no matter the airborne polarized SAR data or the satellite-borne polarized SAR data can be very conveniently acquired on the network, and the polarized SAR data is also very abundant. But the development of automatic interpretation methods of polarimetric SAR images is not as rapid as imaginable. How to analyze the polarized SAR data of mass level becomes an important research point for remote sensing image processing in the future.
The classification of polarized SAR images is roughly divided into two major categories, unsupervised and supervised. For supervised classification of polarized SAR images, it is most often done based on statistical distribution of the polarization data. Such as a complex Wishart distribution based on covariance matrices, a multivariate complex gaussian distribution based on polarization measurement vectors, etc. While supervised classification requires knowing a certain number of labeled samples, for polarimetric SAR data, the process of manually labeling training samples usually takes a lot of labor and time, and the correctness of the training samples cannot be guaranteed by manual labeling, and the distribution of the entire sample cannot be estimated according to the limited training samples.
Disclosure of Invention
The invention aims to provide a polarimetric SAR image semi-supervised classification method based on ensemble learning aiming at the defects of the prior art, on one hand, the influence of label noise is inhibited through a K nearest neighbor (Knn) algorithm, on the other hand, a part of unmarked samples are added through a Wishart distance to assist classification, and the classification precision is improved.
In order to achieve the above object, the present invention comprises the steps of:
(1) Extracting the characteristics of a polarized SAR image, wherein the characteristics comprise a polarization parameter F1, a scattering characteristic F2 and a texture characteristic F3;
(1a) Extracting a polarization parameter F1 of the polarized SAR image based on the S matrix;
(1b) Extracting a scattering characteristic F2 of the polarized SAR image based on polarized scattering decomposition;
(1c) Extracting texture features F3 of the polarized SAR image based on the gray level co-occurrence matrix;
(2) Selecting 1% of samples from marked samples of the polarized SAR image as training samples, and initializing the weight { W of the training samples n }；
(3) Calculating the anti-noise factor C of each training sample by using a K nearest neighbors (K nearest neighbors) algorithm n ,n＝1,...,N；
(3a) Calculating Euclidean distances between each marked sample point in the polarized SAR image and eight adjacent pixel points of the sample point;
(3b) Selecting k nearest neighbors of the sample point, calculating the ratio of the k nearest neighbors to the sample point belonging to the same class, namely an anti-noise factor, wherein the k value is 3;
(4) Entering an iteration process, recording the current iteration times t, wherein the initial t =1, and usingTraining the training data of the weak classifier B to obtain a base classifier h t ；
(5) Calculating h t Error rate of
Wherein, W i t Is the weight of the ith sample in the t iteration, h t (x i ) Is the class in which the ith sample is predicted, y i Is the true class of the ith sample, and N is the number of samples.
(6) Compute basis classifier h t Weight of (2)
Where ln is a logarithmic function, ε t Is the error rate.
(7) Updating sample weights
Wherein Z is t Is a normalization factor.
(8) By means of a classifier h t For the prediction of the unlabeled sample set U, respectively calculating the distance from the sample predicted as the m-th class to the center of the class m, taking two samples with the shortest distance, adding the two samples into the labeled sample, and weighting:
wherein, the first and the second end of the pipe are connected with each other,is the ith sample in the unlabeled sample set,is the jth sample among the marked samples.
(9) Repeating the steps (4) - (8), if the set maximum iteration times is reached or the error rate is greater than 0.5, ending, otherwise, returning to the cycle;
(10) A combination base classifier:
wherein alpha is t Is the weight of the classifier for the t-th iteration.
(11) Outputting a classification result of the polarized SAR image;
compared with the prior art, the invention has the following advantages:
1. the method combines the polarization parameters, scattering characteristics and texture characteristics of the polarized SAR, and well represents the surface feature characteristics, so that the classification precision of the polarized SAR image is improved;
2. according to the method, anti-noise factors are introduced by utilizing Knn, so that the tag noise in the polarized SAR image classification is suppressed;
3. according to the method, wishart distance measurement is utilized, a part of unlabeled samples are selected and added into a training set, so that a better classifier can be obtained by a new training set;
drawings
FIG. 1 is a general flow chart of the present invention;
FIG. 2 is a Pauli decomposition synthetic graph of a polarized SAR image inputted by the present invention;
FIG. 3 is a true clutter map of a polarized SAR image used with the present invention;
fig. 4 is a diagram of the final classification result of the polarized SAR image used in the present invention.
Detailed Description
The image processing method of the present invention will be described below with reference to the drawings showing examples of implementation.
Example 1 is described with reference to fig. 1.
The implementation method comprises the following steps: the polarized SAR image semi-supervised classification method based on ensemble learning comprises the following steps:
step 1, inputting a polarized SAR image;
step 2, extracting the characteristics of the polarized SAR image;
the covariance matrix C of any one pixel in the polarized SAR image is given by:
wherein H denotes horizontal polarization, V denotes vertical polarization, S ij Data representing i-direction transmission of the polarized SAR image and j-direction reception of the polarized wave represent conjugation.
The covariance matrix C is converted to a coherence matrix T according to:
wherein A = S HH +S HH ,B＝S HH -S HH ,C＝2S HV ，
S HH Indicating horizontally transmitted and horizontally received polarized wave echo data, S HV Indicating horizontally transmitted and vertically received echo data of polarized waves, S VH Representing the transmission of horizontally received echo data of polarized waves in the vertical direction, S VV Showing the vertical direction transmitting the polarization wave echo data received in the vertical direction.
Obtaining a gray level co-occurrence matrix of the polarization SAR data according to the following formula:
G(i,j|θ)＝[(x,y),(x+dx,y+dy)]
wherein i and j represent two pixels in a polarized SAR image space respectively, x represents the abscissa of the pixel, y represents the ordinate of the pixel, dx and dy represent the offset in respective directions respectively, theta represents the direction of the offset, and the total of four values are 0, 45, 90 and 135 degrees.
Solving F1 according to the following formula:
HH channel backscattering coefficient:
HV channel backscattering coefficient:
VV channel backscattering coefficient:
co-polarization ratio: r is a radical of hydrogen vvhh ＝10×log(|S vv | 2 /|S hh | 2 )
Cross polarization ratio: r is hvhh ＝10×log(|S hv | 2 /|S hh | 2 )
HV/VV passage ratio: r is hvvv ＝10×log(|S hv | 2 /|S vv | 2 )
VV/HH backscattering coefficient ratio:
HV/HH backscattering coefficient ratio:
HV/VV backscattering coefficient ratio:
HH-VV phase:
depolarizationThe ratio is as follows:
degree of polarization:
wherein R is 1 Representing the frequency-amplitude difference, R, of the horizontal and vertical components in a polarimetric SAR image 2 Representing the power difference between the 45 and 135 degree components of the two directions, R 3 Representing the power difference, R, between the components of the electromagnetic wave of the polarized SAR image on the left and right circular polarization basis 0 Representing the total amplitude value.
F2 is calculated according to the following formula:
pauli decomposition:
calculating the parameter | a 2 ,|b| 2 ,|c| 2
Where T is a coherence matrix, T (1, 1) T (2, 2) T (3, 3) are respectively three elements on a diagonal of the coherence matrix, | a 2 Representing the energy corresponding to odd scattering in the polarized SAR scattering matrix, | b | n 2 Representing the energy corresponding to even scattering in the polarized SAR scattering matrix, | c | n | 2 Representing the energy corresponding to even scattering at an angle of 45 degrees in a polarized SAR scattering matrix;
cloude decomposition:
the parameters H, A, alpha are calculated by the following formula
Wherein λ 1 ，λ 2 ，λ 3 Is three eigenvalues of a coherent matrix, H represents the scattering entropy of the scattering matrix of the polarized SAR image, P i Is the ith eigenvalue of the coherence matrix divided by the sum of all eigenvaluesThe ratio, a, represents the inverse entropy of the scattering matrix of the polarized SAR image, and α represents the scattering type.
Freeman-Druden decomposition:
the parameter P is calculated by the following formula s ,P v ,P d
＜|S HH | 2 ＞＝f s |β| 2 +f d |α| 2 +f d
＜|S HH | 2 ＞＝f s +f d +f v
＜|S HV | 2 ＞＝f v /3
Wherein f is s Representing the coefficient of the planar scattering component, f d Representing the dihedral scattering component coefficient, f v And (3) expressing the volume scattering component coefficient, and calculating three scattering power components according to the following formula on the basis of the volume scattering component coefficient:
P s ＝f s (1+|β| 2 )
P d ＝f d (1+|α| 2 )
wherein P is s Denotes the surface scattered power, P d Denotes the surface scattered power, P v Represents the volume scattering power, beta represents the ratio of the backscattering coefficient received horizontally by horizontal emission to the backscattering coefficient received vertically by vertical emission, when S hh Andwhen the real part obtained by inner product is smaller than zero, beta is 1, and when S is smaller hh Andwhen the real part obtained by the inner product is not smaller than zero, taking alpha as-1; the R value represents the homopolarization ratio of the polarized SAR image.
Krogager decomposition:
the S matrix is decomposed according to the following equation:
wherein S represents a scattering matrix of the polarized SAR image,denotes the absolute phase value of the scattering target, j denotes the imaginary sign of the complex number,representing the offset, k, relative to the dihedral and helical scatter components s Sphere component, k, of a scattering matrix representing a polarized SAR image d Dihedral component, k, of a scattering matrix representing a polarized SAR image h Representing the volume scattering component in the scattering matrix of the polarized SAR image; solving by undetermined coefficient method to obtain k s ,k d ,k h Values of three parameters.
Huynen decomposition:
the coherence matrix T is decomposed according to the following equation:
wherein, a represents the symmetry factor of the polarized SAR image, c represents the configuration factor of the polarized SAR image, i represents the complex imaginary sign, d represents the local curvature difference of the polarized SAR image, h represents the directivity of the polarized SAR image, g represents the coupling degree of the symmetric part and the asymmetric part of the polarized SAR image, and b is tableShowing the non-regularity factor of the polarized SAR image, e showing the surface torsion of the polarized SAR image, f showing the helicity of the polarized SAR image, and l showing the asymmetry factor of the polarized SAR image, and solving by using an undetermined coefficient method to obtain A 0 ,B 0 +B,B 0 Several parameter values of-B, C, D, E, F, G, H.
Texture feature F3 is solved according to the following equation:
entropy:
angular second moment:
moment of dissimilarity:
contrast ratio:
correlation degree:
mean value:
variance:
wherein i and j respectively represent power values of two positions corresponding to the polarized SAR image, x represents an abscissa of a position corresponding to a pixel of the polarized SAR image, y represents an ordinate of a position corresponding to a pixel of the polarized SAR image, dx represents a transverse offset of the pixel of the polarized SAR image, and dy represents a longitudinal offset of the pixel of the polarized SAR image.
Step 3, calculating the anti-noise factor C of each training sample by using a KNN algorithm n ,n＝1,...,N
Calculating Euclidean distances between each marked sample point in the polarized SAR image and corresponding eight neighbors, then selecting k nearest neighbors, calculating a ratio r of the k neighbors to the k neighbors belonging to the same class, wherein if the ratio r is higher, the central sample point is high in similarity with the k neighbors, and otherwise, if the ratio r is lower, the central sample point is not similar to the k neighbors, and the label is likely to be manually pasted. Therefore, the magnitude of the likelihood that the center sample point is the tag noise can be estimated by calculating the magnitude of the value of r, which is named the anti-noise factor.
Step 4, entering an iteration process, recording the current iteration times t, initially t =1, training the training data of the round by using a weak classifier B, and obtaining a classifier h t The weak classifier selects the classical CART.
Step 5, calculating h t Error rate of
The weight is increased for the sample with wrong classification, and the weight is reduced for the sample with correct classification;
step 6, calculating a base classifier h t Weight of (2)
Because the confidence of each classifier is different, a weight is given to the classification itself according to the classification result of each classifier.
Step 7, updating the sample weight
The weights are updated in order to give more attention to the samples that were not correctly classified in the previous iteration in the next iteration, and it is desirable that such samples get more weight, giving them a greater chance to be selected as training samples, and conversely, the samples that were correctly classified in the previous iteration are weighted down, because they are better distinguished, and the classifier should give more attention to those samples that are difficult to distinguish.
Step 8 with classifier h t For the prediction of the unlabeled sample set U, the distances from the sample predicted as the m-th class to the center of the class m are respectively calculated, two samples with the shortest distance are taken and added into the labeled sample, and the weight is given:
here, the samples with high confidence coefficient are selected by using the wishart distance and added into the training set to enrich the original training set.
The Wishart distance formula is as follows:
d(＜T＞,V m )＝n[ln|V m |+Tr(V m -1 ＜T＞)]
step 9, repeating the steps (4) to (8), ending if the stop condition is met, and returning to the circulation if the stop condition is not met;
step 10, combining the base classifiers:
the plurality of classifiers obtained in the above manner are combined according to their respective weights, and a result is given by voting.
And 11, outputting a classification result of the polarized SAR image.
The effect of the present invention can be further illustrated by the following experimental simulation:
1. simulation conditions
The polarized SAR image used in the simulation experiment of the invention is an AIRSAR image with the size of 300 multiplied by 270, and describes the ground feature condition of the Fleviland area in the Netherlands.
The hardware platform is as follows: intel Core2Duo CPU E6550@2.33GHZ, 2GB RAM;
the software platform is as follows: MATLAB R2009a;
2. simulation content and results
Compared with the Knn. AdaBoost classification method, the invention is used for carrying out experiment comparison, the training sample ratio is from 0.001 to 0.01, the step length is increased according to 0.001, and the experiment result is repeated for 10 times to obtain the average value.
Simulation 1: the polar SAR images Flevoland are classified by a comparative experiment algorithm knn.
Simulation 2: the method classifies the polarized SAR image Flevoland, and the classification result is shown in figure 4.
3. Analysis of Experimental results
Wherein semi.knn.adaboost is the method proposed by the present invention.
The experimental results show that the method combines the polarization parameters, the scattering characteristics and the texture characteristics to represent the ground feature characteristics, and can more comprehensively describe the real situation of the ground feature compared with the existing method which only uses the polarization information, thereby obtaining better classification precision. In the method, wishart distance measurement is utilized, a part of unlabeled samples are selected and added into a training set, and a better classifier is obtained through training.
Claims (4)
1. A polarized SAR image semi-supervised classification method based on ensemble learning comprises the following steps:
(1) Extracting the characteristics of a polarized SAR image, wherein the characteristics comprise a polarization parameter F1, a scattering characteristic F2 and a texture characteristic F3;
(1a) Extracting a polarization parameter F1 of the polarized SAR image based on the S matrix;
(1b) Extracting a scattering characteristic F2 of the polarized SAR image based on polarized scattering decomposition;
(1c) Extracting texture features F3 of the polarized SAR image based on the gray level co-occurrence matrix;
(2) Selecting from labeled samples of polarized SAR imagesSelecting a sample with a ratio of 1% as a training sample, wherein the number of the training samples is N, and initializing the sample weight { W) of the training sample n And all initial values are set to be 1/N;
(3) Calculating the anti-noise factor C of each training sample by using a K nearest neighbors (K nearest neighbors) algorithm n ,n＝1,...,N；
(3a) Calculating Euclidean distances between each training sample in the step (2) and eight adjacent pixel points of the sample;
(3b) Selecting eight adjacent pixel points of the sample and k pixel points with the minimum Euclidean distance of the sample, and calculating the ratio of the k adjacent pixel points to the sample belonging to the same class, namely an anti-noise factor, wherein the k value is 3;
(4) Training the training sample by using a weak classifier B according to the characteristics of the polarized SAR image of the training sample extracted in the step (1) to obtain a base classifier h t And recording the current time t, wherein the initial t =1;
(5) Calculating a base classifier h obtained by training in the step (4) t Error rate of
Wherein, W i t Is the weight of the ith sample in the t iteration, h t (x i ) Is the class in which the ith sample is predicted, y i Is the true category of the ith sample, and N is the number of samples;
(6) Calculating a base classifier h obtained by training in the step (4) t Weight of (2)
Where ln is a logarithmic function, ε t Is the error rate, C t Is an anti-noise factor;
(7) Updating the sample weight of the training sample in the step (2) according to the anti-noise factor of the training sample calculated in the step (3)
Wherein, W i t+1 Is the weight of the ith sample in the t +1 th iteration, α t Is base classifier h t Weight of (C) t Is an antinoise factor, Z t Is a normalization factor, W n Then the sample weight of the first nth sampleThe abbreviation of (c);
(8) Using the base classifier h obtained by training in (4) t Respectively calculating the distance from the sample predicted as the m-th class to the center of the m-th class for the unlabeled sample set U prediction of the polarized SAR image, adding the labeled sample into the two samples closest to each other, and weighting the two samples:
wherein, the first and the second end of the pipe are connected with each other,is the ith sample in the unlabeled sample set u,is the jth sample, N, in the marked sample set l l Is the total number of samples in the marked sample set l, and m is any category mark;
(9) Repeating the steps (4) - (8), if the set maximum iteration times is reached or the error rate is greater than 0.5, ending and skipping to the step (10), otherwise, returning to the step (4) for continuation;
(10) Combining the base classifiers generated in the steps (4) to (8) to obtain a combined classifier:
wherein alpha is t Weights of classifiers that are the t-th iteration，h t (x) Is a base classifier h t Sign is a sign function for the prediction of sample x;
(11) And (5) predicting and classifying the polarized SAR image by using the combined classifier in the step (10), and outputting a classification result of the polarized SAR image.
2. The semi-supervised classification method for polarimetric SAR images based on ensemble learning of claim 1, wherein the extracting polarization parameters F1 in step (1 a) comprises the following steps:
firstly, respectively calculating 12 groups of characteristic parameters according to the following formula:
HH channel backscattering coefficient:
HV channel backscattering coefficient:
VV channel backscattering coefficient:
co-polarization ratio: r is vvhh ＝10×log(|S vv | 2 /|S hh | 2 )
Cross polarization ratio: r is hvhh ＝10×log(|S hv | 2 /|S hh | 2 )
HV/VV passage ratio: r is hvvv ＝10×log(|S hv | 2 /|S vv | 2 )
Ratio of VV/HH backscattering coefficient:
HV/HH backscattering coefficient ratio:
HV/VV backscattering coefficient ratio:
HH-VV phase:
depolarization ratio:
degree of polarization:
wherein S hh Indicating horizontally transmitted and horizontally received echo data of polarized waves, S hv Indicating horizontally transmitted and vertically received echo data of polarized waves, S vh Indicating transmitted and received horizontally polarized wave echo data in vertical direction, S vv Representing vertically transmitted and vertically received echo data of polarized waves, R 1 Representing the frequency-amplitude difference, R, of the horizontal and vertical components in a polarimetric SAR image 2 Representing the power difference between the 45 and 135 degree components of the two directions, R 3 Representing the power difference, R, between the components of the electromagnetic wave of the polarized SAR image on the left and right circularly polarized basis 0 Representing the total amplitude value, representing the conjugate;
and secondly, combining the 12 sets of characteristic parameters to obtain a polarization parameter characteristic F1.
3. The integrated learning-based polarimetric SAR image semi-supervised classification method according to claim 2, wherein the step (1 b) of extracting the scattering features F2 comprises the following steps:
first, pauli decomposition:
calculating the parameter | a 2 ,|b| 2 ,|c| 2
Where T is a coherence matrix, T (1, 1) T (2, 2) T (3, 3) are respectively three elements on a diagonal of the coherence matrix, | a 2 Representing the energy corresponding to odd scattering in the polarized SAR scattering matrix, | b $ Y 2 Representing the energy corresponding to even scattering in the polarized SAR scattering matrix, | c $ 2 Representing the energy corresponding to the even scattering at an angle of 45 degrees in the polarization SAR scattering matrix;
second, the cloud decomposition:
the parameters H, A, alpha are calculated by the following formula
Wherein λ 1 ，λ 2 ，λ 3 Is three eigenvalues of a coherent matrix, H represents the scattering entropy of the scattering matrix of the polarized SAR image, P i The ratio is obtained by dividing the ith eigenvalue of the coherent matrix by the sum of all eigenvalues, A represents the inverse entropy of the scattering matrix of the polarized SAR image, and alpha represents the scattering type;
thirdly, freeman-Druden decomposition:
the parameter P is calculated by the following formula s ,P v ,P d
<|S HH | 2 >＝f s |β| 2 +f d |α| 2 +f d
<|S HH | 2 >＝f s +f d +f v
<|S HV | 2 >＝f v /3
Wherein f is s Coefficient of plane scattering component, f d Representing dihedral angleCoefficient of scattered component, f v And (3) expressing the volume scattering component coefficient, and calculating three scattering power components according to the following formula on the basis of the volume scattering component coefficient:
P s ＝f s (1+|β| 2 )
P d ＝f d (1+|α| 2 )
wherein P is s Denotes the surface scattered power, P d Denotes the surface scattered power, P v Represents the volume scattering power, beta represents the ratio of the backscattering coefficient received horizontally by horizontal emission to the backscattering coefficient received vertically by vertical emission, when S hh And withWhen the real part obtained by inner product is smaller than zero, beta is 1, and S is hh And withWhen the real part obtained by the inner product is not smaller than zero, taking alpha as-1; the R value represents the homopolarization ratio of the polarized SAR image;
fourthly, decomposing by Krogager:
the S matrix is decomposed according to the following equation:
wherein S represents a scattering matrix of the polarized SAR image,representing absolute phase values of scattering targets, j representing complex numbersThe sign of the imaginary part is such that,representing the offset, k, relative to the dihedral and helical scatter components s Sphere component, k, of a scattering matrix representing a polarized SAR image d Dihedral component, k, of a scattering matrix representing a polarized SAR image h Representing the volume scattering component in the scattering matrix of the polarized SAR image; solved by undetermined coefficient method to obtain
k s ,k d ,k h The values of the three parameters;
fifthly, huynen decomposition:
the coherence matrix T is decomposed according to the following equation:
the method comprises the steps that a represents a symmetry factor of a polarized SAR image, c represents a configuration factor of the polarized SAR image, i represents a complex imaginary sign, d represents a local curvature difference of the polarized SAR image, h represents the directivity of the polarized SAR image, g represents the coupling degree of the symmetrical part and the asymmetrical part of the polarized SAR image, b represents an irregular factor of the polarized SAR image, e represents the surface torsion of the polarized SAR image, f represents the helicity of the polarized SAR image, l represents an asymmetrical factor of the polarized SAR image, and A is obtained by solving with a undetermined coefficient method 0 ,B 0 +B,B 0 -B, C, D, E, F, G, H;
and sixthly, combining the parameters to obtain the scattering characteristic F2.
4. The integrated learning-based semi-supervised classification method for polarized SAR images as claimed in claim 1, wherein the classification result of the polarized SAR images in step (11) comprises the following steps:
r, G and B are three primary colors and are mixed according to a certain ratio to obtain green, grass green, blue, red, purple and orange, and different colors are respectively distinguished in different categories.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410748929.2A CN104504393B (en) | 2014-12-04 | 2014-12-04 | Polarimetric SAR Image semisupervised classification method based on integrated study |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410748929.2A CN104504393B (en) | 2014-12-04 | 2014-12-04 | Polarimetric SAR Image semisupervised classification method based on integrated study |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104504393A CN104504393A (en) | 2015-04-08 |
| CN104504393B true CN104504393B (en) | 2018-01-05 |
Family
ID=52945789
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410748929.2A Active CN104504393B (en) | 2014-12-04 | 2014-12-04 | Polarimetric SAR Image semisupervised classification method based on integrated study |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104504393B (en) |
Families Citing this family (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104820687A (en) * | 2015-04-22 | 2015-08-05 | 中国科学院信息工程研究所 | Construction method of directed link type classifier and classification method |
| CN104850832B (en) * | 2015-05-06 | 2018-10-30 | 中国科学院信息工程研究所 | A kind of large-scale image sample mask method and system based on classification iteration |
| CN105138970B (en) * | 2015-08-03 | 2018-11-16 | 西安电子科技大学 | Classification of Polarimetric SAR Image method based on spatial information |
| CN105160353B (en) * | 2015-08-18 | 2018-07-17 | 西安电子科技大学 | Polarization SAR data terrain classification method based on multiple features collection |
| CN105574547B (en) * | 2015-12-22 | 2019-03-05 | 北京奇虎科技有限公司 | Adapt to integrated learning approach and device that dynamic adjusts base classifier weight |
| CN106067042B (en) * | 2016-06-13 | 2019-02-15 | 西安电子科技大学 | Polarization SAR classification method based on semi-supervised depth sparseness filtering network |
| CN106529385A (en) * | 2016-08-12 | 2017-03-22 | 上海卫星工程研究所 | Unsupervised classification for polarimetric interferometric SAR image based on coherence spectrum parameters |
| CN107133653B (en) * | 2017-05-23 | 2020-11-03 | 西安电子科技大学 | High-resolution SAR image classification method based on deep ladder network |
| CN107203791B (en) * | 2017-06-30 | 2019-12-31 | 西安电子科技大学 | Polarimetric SAR image classification method based on heteropolarity ratio and scattering power entropy |
| CN107392140B (en) * | 2017-07-19 | 2021-01-08 | 苏州闻捷传感技术有限公司 | Polarized SAR terrain classification method based on low-rank sparse decomposition and incremental learning |
| CN107451617B (en) * | 2017-08-08 | 2020-08-04 | 西北大学 | Graph transduction semi-supervised classification method |
| CN108038448A (en) * | 2017-12-13 | 2018-05-15 | 河南理工大学 | Semi-supervised random forest Hyperspectral Remote Sensing Imagery Classification method based on weighted entropy |
| CN113505840A (en) * | 2021-07-19 | 2021-10-15 | 西北农林科技大学 | Polarization SAR image combination classification method based on confusion matrix |
| CN115565610A (en) * | 2022-09-29 | 2023-01-03 | 四川大学 | Method and system for establishing recurrence transfer analysis model based on multiple sets of mathematical data |
| CN116807479B (en) * | 2023-08-28 | 2023-11-10 | 成都信息工程大学 | Driving attention detection method based on multi-mode deep neural network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8125370B1 (en) * | 2007-04-16 | 2012-02-28 | The United States Of America As Represented By The Secretary Of The Navy | Polarimetric synthetic aperture radar signature detector |
| CN102651073A (en) * | 2012-04-07 | 2012-08-29 | 西安电子科技大学 | Sparse dynamic ensemble selection-based SAR (synthetic aperture radar) image terrain classification method |
| CN102737381A (en) * | 2012-06-13 | 2012-10-17 | 西安电子科技大学 | Image partitioning method based on mixed bipartite graph clustering integration |
| CN103955702A (en) * | 2014-04-18 | 2014-07-30 | 西安电子科技大学 | SAR image terrain classification method based on depth RBF network |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8193967B2 (en) * | 2008-12-10 | 2012-06-05 | The United States Of America As Represented By The Secretary Of The Army | Method and system for forming very low noise imagery using pixel classification |
-
2014
- 2014-12-04 CN CN201410748929.2A patent/CN104504393B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8125370B1 (en) * | 2007-04-16 | 2012-02-28 | The United States Of America As Represented By The Secretary Of The Navy | Polarimetric synthetic aperture radar signature detector |
| CN102651073A (en) * | 2012-04-07 | 2012-08-29 | 西安电子科技大学 | Sparse dynamic ensemble selection-based SAR (synthetic aperture radar) image terrain classification method |
| CN102737381A (en) * | 2012-06-13 | 2012-10-17 | 西安电子科技大学 | Image partitioning method based on mixed bipartite graph clustering integration |
| CN103955702A (en) * | 2014-04-18 | 2014-07-30 | 西安电子科技大学 | SAR image terrain classification method based on depth RBF network |
Non-Patent Citations (2)
| Title |
|---|
| "一种基于集成学习和特征融合的遥感影像分类方法";刘培 等，;《红外与毫米波学报》;20140630;第33卷(第3期);第311-317页 * |
| "基于半监督集成的遥感图像的分割和分类";王婷;《中国优秀硕士学位论文全文数据库 信息科技辑》;20111215(第S2期);第2-5章 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104504393A (en) | 2015-04-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104504393B (en) | Polarimetric SAR Image semisupervised classification method based on integrated study | |
| CN108388927B (en) | Small sample polarization SAR terrain classification method based on deep convolution twin network | |
| CN103714354B (en) | Hyperspectral image wave band selection method based on quantum-behaved particle swarm optimization algorithm | |
| CN109359525B (en) | Polarized SAR image classification method based on sparse low-rank discrimination spectral clustering | |
| CN111308471B (en) | Rain, snow and hail classification monitoring method based on semi-supervised domain adaptation | |
| CN103413146B (en) | Polarimetric SAR Image sophisticated category method based on Freeman entropys and self study | |
| CN105069796B (en) | SAR image segmentation method based on small echo both scatternets | |
| CN105117736B (en) | Classification of Polarimetric SAR Image method based on sparse depth heap stack network | |
| CN103927551A (en) | Polarimetric SAR semi-supervised classification method based on superpixel correlation matrix | |
| CN108388907B (en) | Real-time updating method of polarized SAR data classifier based on multi-view learning | |
| CN109840542B (en) | Self-adaptive dimension decision tree classification method based on polarization characteristics | |
| CN109932717B (en) | ISAR high-resolution imaging method based on environmental statistics modeling | |
| CN107203791B (en) | Polarimetric SAR image classification method based on heteropolarity ratio and scattering power entropy | |
| Mosavi et al. | Classification of sonar targets using OMKC, genetic algorithms and statistical moments | |
| CN109901130A (en) | A kind of rotor wing unmanned aerial vehicle detection and recognition methods converting and improve 2DPCA based on Radon | |
| CN106446935A (en) | Kernel sparse representation and spatial constraint-based polarimetric SAR image classification method | |
| CN104700116A (en) | Polarized SAR (synthetic aperture radar) image object classifying method based on multi-quantum ridgelet representation | |
| Zhai et al. | A novel sense-through-foliage target recognition system based on sparse representation and improved particle swarm optimization-based support vector machine | |
| CN109063750B (en) | SAR target classification method based on CNN and SVM decision fusion | |
| Han et al. | Unsupervised classification for PolSAR images based on multi-level feature extraction | |
| CN111707999B (en) | Sea surface floating small target detection method based on combination of multiple features and ensemble learning | |
| P. Thampy et al. | A convolution neural network approach to Doppler spectra classification of 205 MHz radar | |
| Chakraborty et al. | Seafloor classification using echo-waveforms: a method employing hybrid neural network architecture | |
| Wan et al. | Squeeze excitation densely connected residual convolutional networks for specific emitter identification based on measured signals | |
| Zhai et al. | A new sense-through-foliage target recognition method based on hybrid particle swarm optimization-based wavelet twin support vector machine |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |