CN116797845A - Unsupervised reduced polarization classification method based on scattering mechanism - Google Patents
Unsupervised reduced polarization classification method based on scattering mechanism Download PDFInfo
- Publication number
- CN116797845A CN116797845A CN202310818264.7A CN202310818264A CN116797845A CN 116797845 A CN116797845 A CN 116797845A CN 202310818264 A CN202310818264 A CN 202310818264A CN 116797845 A CN116797845 A CN 116797845A
- Authority
- CN
- China
- Prior art keywords
- scattering mechanism
- representing
- class
- scattering
- polarization
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 230000007246 mechanism Effects 0.000 title claims abstract description 91
- 230000010287 polarization Effects 0.000 title claims abstract description 75
- 238000000034 method Methods 0.000 title claims abstract description 28
- 238000012549 training Methods 0.000 claims abstract description 8
- 238000007476 Maximum Likelihood Methods 0.000 claims abstract description 7
- 238000004458 analytical method Methods 0.000 claims abstract description 7
- 230000008859 change Effects 0.000 claims abstract description 7
- 230000008569 process Effects 0.000 claims abstract description 5
- 238000012804 iterative process Methods 0.000 claims abstract description 4
- 239000013598 vector Substances 0.000 claims description 32
- 239000011159 matrix material Substances 0.000 claims description 31
- 238000005259 measurement Methods 0.000 claims description 10
- 238000013507 mapping Methods 0.000 claims description 6
- 238000000342 Monte Carlo simulation Methods 0.000 claims description 3
- 230000001427 coherent effect Effects 0.000 claims description 3
- 238000012512 characterization method Methods 0.000 claims description 2
- 238000000354 decomposition reaction Methods 0.000 abstract description 4
- 238000007635 classification algorithm Methods 0.000 description 6
- 230000008901 benefit Effects 0.000 description 3
- 238000003384 imaging method Methods 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000010899 nucleation Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/761—Proximity, similarity or dissimilarity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
Abstract
The invention provides an unsupervised reduced polarization classification method based on a scattering mechanism, which comprises the following steps: the ground objects are roughly classified into three types according to modeling analysis and judgment of a surface scattering mechanism, a dihedral angle scattering mechanism and a bulk scattering mechanism. And dividing subclasses in each class according to the scattered power intensity, taking elements with median power in the subclasses as initial class centers of the Weisauter classifier, and obtaining initial classification results. And (3) taking the data of each class as a training sample to estimate the class center, and reclassifying according to the Wishade maximum likelihood distance metric. And repeating the process, and ending the iterative process when the change of the class center is negligible and the quantity meets the preset value, so as to obtain a final classification result. The model-based polarization decomposition has definite physical significance and can provide basis for training of classified samples.
Description
Technical Field
The invention belongs to the SAR field, and particularly relates to an unsupervised reduced polarization classification method based on a scattering mechanism.
Background
The reduced polarization is a novel system for balancing system resources and observation dimensions of ground feature information. The circular polarization transmitting linear polarization receiving (CTLR) mode is used as a typical simplified polarization mode and can achieve performance similar to full polarization in the fields of land mapping, agricultural monitoring, marine application and the like, so that the circular polarization transmitting linear polarization receiving (CTLR) mode is widely applied. Compared with the traditional full polarization mode, the simplified polarization system is relatively simple in complexity, low in power consumption and capable of achieving flatter and more balanced distance ambiguity performance. In addition, the radar imaging regime without alternate emission allows for a reduced polarization mode to select smaller pulse repetition frequencies and obtain a larger swath. Therefore, research on reduced polarization has received extensive attention from students in the field of remote sensing.
The simplified polarization system has the advantages of low power consumption, lower data downlink speed under the same mapping breadth and the same mode of transmitting circularly polarized waves as the ground-based month observation SAR radar, so that the simplified polarization system is used for month observation at the earliest. In 2008, the united states and india respectively transmitted LRO and Chandrayaan-1, two satellites carrying the compressed polarized Synthetic Aperture Radar (SAR) load, and realized the first in-orbit test of the compressed polarized mode. The RISAT-1 satellite transmitted in 2012 in india enabled for the first time earth observation in the reduced polarization mode. To date, the countries such as the United states, india, japan, canada, argentina and the like are sequentially launched to carry satellites capable of realizing the load of the reduced polarization mode, so that reliable support is provided for the research of reduced polarization. China first can realize that satellite terrestrial survey No. one (LT-1) of the reduced polarization mode earth observation is successfully operated in orbit.
The method is characterized in that the preliminary classification of the ground object type is realized based on the polarization characteristic parameters extracted from the observation vector of the ground object, and the unsupervised classification is realized by combining various classifiers. The traditional unsupervised classification algorithm based on abbreviated polarization H/alpha classification and m- χ decomposition reduces the accuracy of the results of the preliminary classification due to the reduced data dimensions.
Disclosure of Invention
In order to solve the technical problems, the invention provides an unsupervised reduced polarization classification method based on a scattering mechanism based on real reduced polarization data provided by LT-1, which is an unsupervised classification algorithm of ground features and compares the real ground information to verify the effectiveness of the process.
In order to achieve the above purpose, the invention adopts the following technical scheme:
an unsupervised reduced polarization classification method based on a scattering mechanism comprises the following steps:
step 1, judging a main scattering mechanism according to modeling analysis of a surface scattering mechanism, a dihedral angle scattering mechanism and a bulk scattering mechanism, roughly dividing ground objects into three categories, dividing subclasses in each category according to scattering power intensity, taking an element with the scattering power intensity being a median in the subclasses as an initial class center of a Weisauter classifier, and obtaining an initial classification result;
step 2, taking the data of each category as a training sample to update the class center in an iterative manner, and reclassifying according to the maximum likelihood distance metric of the Wishade; when the number of the categories is larger than the required clustering number, merging the most similar categories in the center of the categories; if a plurality of classes with close similarity exist, preferentially merging the classes with few elements; and repeating the process, and ending the iterative process when the change of the class center is ignored and the number meets the preset value, so as to obtain a final classification result.
Further, in the step 1, the method for roughly classifying the ground objects into three categories includes:
the relationship between Stokes vector and full polarization coherence matrix is:
wherein S represents a ground object scattering matrix, S xy Representing the complex scattering coefficient of the reflected y polarized wave of the received x polarized wave, x representing H or V, y representing H or V, the superscript T representing the transpose, the superscript x representing the conjugate, k representing the three-dimensional target vector, T representing the polarization coherence matrix,jones vector, E representing a polarized wave of L or R in CTLR mode H Representing the H reception channel sampled signal, E V A signal representing V receiving channel samples, H representing vertical, V representing horizontal, j representing imaginary units,/>Then the corresponding Stokes vector is represented, re (·) represents the real part of the complex number, im (·) represents the imaginary part of the complex number;
from formula (1) -formula (3):
wherein ,Tij Representing the ith row and the jth column elements of the polarization coherent matrix; q 0 ,q 1 ,q 2 ,q 3 Representing elements in a Stokes vector; deriving Stokes vectors under corresponding CTLR modes according to a model of polarization coherence matrix characterization established by the full polarization on a surface scattering mechanism, a dihedral angle scattering mechanism and a bulk scattering mechanism;
stokes vector Q of surface scattering model s The form of (2) is:
wherein β represents the ratio of the reflection coefficients of the horizontal and vertical polarizations of the reflecting surface; for an ideal surface scattering mechanism, q 3 ＝1；
Stokes vector Q of dihedral angle scattering model d The form of (2) is as follows:
wherein , x polarization reflection coefficient representing nth scattering, n=1, 2, X represents H or V, γ H and γV Representing phase changes during electromagnetic wave propagation; for an ideal dihedral scattering mechanism, q 3 ＝-1；
Stokes vector Q of volume scattering model v The form of (c) is derived from the Arii volume scattering model as follows:
wherein ,mv Representing the degree of polarization of the model, phi representing the average orientation angle; the power of the bulk scattering mechanism is obtained assuming that the surface scattering mechanism and the dihedral angle scattering mechanism are ideal conditions and maximizing the power occupied by the bulk scattering mechanismAnd satisfies the following:
the polarization degree m of the model obtained by analysis v As an index for judging whether the bulk scattering mechanism is dominant, a threshold for distinguishing whether the bulk scattering mechanism is dominant is setValue m th The method comprises the steps of carrying out a first treatment on the surface of the In the region where the bulk scattering mechanism is not dominant, q is obtained from the formula (5) -formula (6) 3 Is used to distinguish the dominance of the surface scattering mechanism from the dihedral scattering mechanism; q when transmitting right-hand R polarized wave 3 0 indicates that the surface scattering mechanism is dominant, q 3 < 0 means that the dihedral scattering mechanism dominates; q when L polarized wave is transmitted 3 < 0 means that the surface scattering mechanism is dominant, q 3 0 indicates that the dihedral scattering mechanism dominates.
Further, in the step 2, fine classification is achieved by means of a generalized weisat classifier, including:
the covariance matrix C of the reduced polarization is:
it obeys the complex weisalde distribution, the probability density function P of which is as follows:
wherein M represents a class center, L is a multiview number, Γ 2 (L) is a constant, exp (·) represents an exponent based on e, det (·) represents a determinant of the matrix, tr (·) represents a trace of the matrix;
based on the statistical characteristics of the covariance matrix C of the reduced polarization, a distance measure d which describes the difference of the two covariance matrices is obtained, wherein the distance measure d is shown in the following formula:
wherein the subscripts m and n are used to distinguish different covariance matrices, when C m And C n At the same time, the distance metric d=0;
the measure of similarity is described by mapping the distance measure d to an interval of 0 to 1, similarity parameters obtained by laplace kernel mappingλ γ The following formula is shown:
λ γ (C m ,C n )＝e -γd (12)
wherein, gamma is a super parameter, and is determined by Monte Carlo experiments;
counting the distance measurement of 1000 pairs of adjacent pixels randomly selected from the simplified polarization data, wherein the median of the counted distance measurement is d mid Taking out
Each pixel is classified into a class with the smallest distance measure from the class center, and unsupervised classification is realized.
Further, in the step 2, the step of iteratively updating the class center by using the data of each class as the training sample includes the following steps:
step (1): super parameter m for selecting unsupervised classification algorithm th And a filter, and inputting parameters required by the filter; setting the output class number and the convergence threshold delta of the decision class center d ；
Step (2): the ground features are divided into three main classes according to the dominance of a surface scattering mechanism, a dihedral angle scattering mechanism and a volume scattering mechanism, the ground features are divided into K classes according to the scattering power intensity of the ground features, and a covariance matrix of a sample with the scattering power being the median in each class is taken as an initial class centerThe superscript (0) represents the initial value of the iteration, and the subscript k represents the kth class; further suppressing the coherence spots on the observed data by using a filter;
step (3): calculating element C in each sample k,n The maximum likelihood distance measurement with each class center marks the element as the class to which the class center most similar to the element belongs, the subscript k, n represents the nth element in the kth class, and the class center is updated according to the following formula:
wherein ,Nk Representing the total number of elements contained in the kth class; when the class center is larger than the output class number, the class with the most similar and least element number is preferentially combined by adopting the following formula:
wherein ,k1 ,k 2 To distinguish two different classes, k 1 ∪k 2 Union of two types of mergers, normalized weightsObtained by the following formula:
step (3) is circularly executed until the change of the class center is negligible and the number is the required classification number.
The beneficial effects are that:
the method has definite physical significance for polarization decomposition based on the model, and can provide basis for training of classified samples. In addition, compared with the preprocessing steps of the rest simplified polarization unsupervised classification algorithm, the method has the advantage of less calculation amount.
Drawings
FIG. 1 is an optical image of an observation area (from Google Earth);
FIG. 2 is an exploded view of m- χ polarization;
FIG. 3 is a graph of the result of a coarse classification based on a dominant scattering mechanism, where graph (a) is where dihedral angle scattering mechanisms are dominant, graph (b) is where bulk scattering mechanisms are dominant, and graph (c) is where surface scattering mechanisms are dominant;
fig. 4 is a diagram of a result of iterative clustering of weishat (Wishart).
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention. In addition, the technical features of the embodiments of the present invention described below may be combined with each other as long as they do not collide with each other.
The invention discloses an unsupervised reduced polarization classification method based on a scattering mechanism, which comprises the following steps:
and 1, judging a main scattering mechanism according to modeling analysis of a surface scattering mechanism, a dihedral angle scattering mechanism and a bulk scattering mechanism, and roughly classifying ground objects into three categories. The subclasses are divided according to the scattering power intensity in each class, and the element with the scattering power intensity being the median in the subclasses is used as the initial class center of the Wishare classifier.
And 2, estimating class centers by taking the data of each class as training samples, and reclassifying according to the Wishare maximum likelihood distance measurement. When the number of categories is greater than the number of clusters required, the most similar categories in the center of the categories are merged. If there are multiple classes with close similarity, the classes with fewer elements are preferentially merged. And repeating the process, and ending the iterative process when the change of the class center is negligible and the quantity meets the preset value, so as to obtain a final classification result.
Specifically, in the step 1, the ground objects are roughly classified into three categories by an unsupervised rough classification method.
The reduced polarization mode adopted by LT-1 is a circularly polarized transmission line polarized reception (CTLR) mode. For the reduced polarization mode, power values are often used to define the electromagnetic wave polarization state, i.e., stokes (Stokes) vector feature definition. Since the study of the different scattering mechanisms by the fully polarized mode is very mature, the relationship between Stokes vectors and the fully polarized coherence matrix is given below.
Wherein S represents a ground object scattering matrix, S xy Representing the complex scattering coefficient of the reflected y polarized wave of the received x polarized wave, x representing H or V, y representing H or V, the superscript T representing the transpose, the superscript x representing the conjugate, k representing the three-dimensional target vector, T representing the polarization coherence matrix,representing Jones vector of a left-hand (L) or right-hand (R) polarized wave emitted in CTLR mode, E H Representing the H reception channel sampled signal, E V A signal representing V receiving channel samples, H representing vertical, V representing horizontal, j representing imaginary units, and +.>Then the corresponding Stokes vector is represented, re (-) represents the real part of the complex number and Im (-) represents the imaginary part of the complex number. From the formula (1) -formula (3):
wherein ,q0 ,q 1 ,q 2 ,q 3 Representing elements in Stokes vectors, T ij Representing the ith row and jth column elements of the polarization coherence matrix.
From the polarization coherence matrix representation model established by the full polarization versus surface scattering, dihedral angle scattering and bulk scattering mechanisms, the Stokes vector in the corresponding CTLR mode can be derived.
Stokes vector Q of surface scattering model s The form of (2) is:
where β represents the ratio of the reflection coefficients of the horizontal (H) and vertical (V) polarizations of the reflecting surface. For an ideal surface scattering mechanism, q 3 ＝1。
Stokes vector Q of dihedral angle scattering model d The form of (2) is as follows:
wherein , x-polarization reflection coefficient, γ, representing nth (n=1, 2) scattering H and γV Representing the phase change during the propagation of an electromagnetic wave. For an ideal dihedral scattering mechanism, q 3 ＝-1。
Similarly, stokes vector Q of volume scattering model v The form of (c) is derived from the Arii volume scattering model as follows:
wherein ,mv Representing the degree of polarization of the model, phi representing the average orientation angle. Assuming that the surface scattering mechanism and the dihedral scattering mechanism are ideal conditions and maximizing the power occupied by the bulk scattering mechanism, the bulk scattering mechanism power can be obtained asAnd the requirements are as follows:
by analysis to obtain m v Can be used for judging whether the bulk scattering mechanism is dominantIndex, setting threshold value m for distinguishing whether volume scattering mechanism is dominant or not th . In the region where the bulk scattering mechanism is not dominant, q is known from formulae (5) - (6) 3 The sign of (c) can distinguish between the dominance of the surface scattering mechanism and the dihedral scattering mechanism. Q when transmitting R polarized wave 3 0 indicates that the surface scattering mechanism is dominant, q 3 A < 0 indicates that the dihedral scattering mechanism is dominant. When an L polarized wave is transmitted, the discrimination conditions are reversed.
In the step 2, the implementation of fine classification by means of a generalized Wishart classifier includes:
the fully polarized target vector has been shown to satisfy the multivariate complex gaussian distribution, so the fully polarized coherence matrix satisfies the complex Wishart distribution, and equation (2) has demonstrated that the receiving vector of the reduced polarization can be linearly represented by the target vector, so the covariance matrix C of the reduced polarization is:
it also obeys a complex Wishart distribution, the probability density function P of which is shown below:
wherein M represents a class center, L is a multiview number, Γ 2 (L) is a constant, exp (·) represents an exponent based on e, det (·) represents a determinant of the matrix, and tr (·) represents a trace of the matrix. Based on the statistical characteristics of the simplified polarization covariance matrix, a distance measurement index d which describes the difference between two covariance matrices and is shown in the following formula can be obtained:
wherein the subscripts m and n are used to distinguish different covariance matrices, when C m And C n At the same time, the distance metric d=0. The measure of similarity is typically measured by the distanceThe separation amount is mapped to an interval of 0 to 1 to describe the similarity parameter lambda obtained by Laplace kernel function mapping γ The following formula is shown:
λ γ (C m ,C n )＝e -γd (12)
wherein, gamma is a super parameter, which can be determined by Monte Carlo experiments. Counting the distance measurement of 1000 pairs of adjacent pixels randomly selected from the simplified polarization data, wherein the median of the counted distance measurement is d mid Taking out
Each pixel is classified as a class with the smallest distance measure from the class center, and when the class center can be determined unsupervised, unsupervised classification can be achieved. The method of determining the class center is given below.
Step 1: selecting a hyper-parameter m of a suitable unsupervised classification algorithm th . The experiment selects a sliding average Filter (Box Car Filter), the window size of the Filter needs to be determined, and the Filter can also select a fine Lee Filter (Lee Refined Filter) and a Non-Local Filter (Non-Local Filter) and input the parameters needed by the Filter. Setting the output class number and the convergence threshold delta of the decision class center d 。
Step 2: dividing the ground object into three main classes according to the dominance of three scattering mechanisms, dividing the ground object into K classes according to the scattering power intensity of the ground object, and taking a covariance matrix of a sample with the scattering power being the median value in each class as an initial class centerThe superscript (0) indicates the initial value of the iteration and the subscript k indicates the kth class. And the coherent spots are further suppressed on the observed data by using the selected filter, so that the robustness of the classification algorithm is improved.
Step 3: calculating element C in each sample k,n The maximum likelihood distance metric from each class center marks the element as the class to which its most similar class center belongs, and the subscript k, n represents the nth element in the kth class. According toThe following update class centers:
wherein ,Nk Indicating the total number of elements contained in the k-th class. When the class center is larger than the output class number, the class with the most similar and least element number is preferentially combined by adopting the following formula:
wherein ,k1 ,k 2 To distinguish two different classes, k 1 ∪k 2 Union of two types of mergers, normalized weightsObtained by the following formula:
step 3 is performed in a loop until the class center varies little and the number is the required number of classifications.
Examples:
based on the abbreviated polarization data acquired by the LT-1 system, the emission polarization type is L polarization, the imaging time is 2023, 4 months and 9 days, and the imaging area is Hami city of Xinjiang, china. The data used are multi-view data of azimuth 8-view and distance 5-view after the average processing of the original single-view complex image (SLC) data, and the image size is 2048 multiplied by 2048. Through experiments, the algorithm provided for m th The external super parameters are not sensitive, and part of parameters of the experimental display result are shown as I.
TABLE I
The optical image of the observation area, the decomposition result, and the classification result are shown in fig. 1 and 2, respectively. FIG. 3 illustrates a coarse classification result based on a dominant scattering mechanism, wherein FIG. 3, panel (a) shows that a dihedral scattering mechanism is dominant, FIG. 3, panel (b) shows that a bulk scattering mechanism is dominant, and FIG. 3, panel (c) shows that a surface scattering mechanism is dominant; fig. 4 shows the result of Wishart iterative clustering. After the iteration is finished, the ground objects are divided into 12 types (the samples of each dominant scattering mechanism are divided into 4 types), the classification method can effectively distinguish sparse vegetation areas, gobi areas, fallow or seeding areas, nursery and urban areas, and the land coverage types which are not visually distinguished in the optical images and the decomposed images are also distinguished from the classification results, so that the effectiveness of the method is proved.
It will be readily appreciated by those skilled in the art that the foregoing description is merely a preferred embodiment of the invention and is not intended to limit the invention, but any modifications, equivalents, improvements or alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (4)
1. An unsupervised reduced polarization classification method based on a scattering mechanism is characterized by comprising the following steps:
step 1, judging a main scattering mechanism according to modeling analysis of a surface scattering mechanism, a dihedral angle scattering mechanism and a bulk scattering mechanism, roughly dividing ground objects into three categories, dividing subclasses in each category according to scattering power intensity, taking an element with the scattering power intensity being a median in the subclasses as an initial class center of a Weisauter classifier, and obtaining an initial classification result;
step 2, taking the data of each category as a training sample to update the class center in an iterative manner, and reclassifying according to the maximum likelihood distance metric of the Wishade; when the number of the categories is larger than the required clustering number, merging the most similar categories in the center of the categories; if a plurality of classes with close similarity exist, preferentially merging the classes with few elements; and repeating the process, and ending the iterative process when the change of the class center is ignored and the number meets the preset value, so as to obtain a final classification result.
2. The method for classifying the unsupervised simplified polarization based on the scattering mechanism according to claim 1, wherein in the step 1, the method for classifying the ground features into three categories by using an unsupervised coarse classification method comprises the following steps:
the relationship between Stokes vector and full polarization coherence matrix is:
wherein S represents a ground object scattering matrix, S xy Representing the complex scattering coefficient of the reflected y polarized wave of the received x polarized wave, x representing H or V, y representing H or V, the superscript T representing the transpose, the superscript x representing the conjugate, k representing the three-dimensional target vector, T representing the polarization coherence matrix,jones vector, E representing a polarized wave of L or R in CTLR mode H Representing the H reception channel sampled signal, E V A signal representing V receiving channel samples, H representing vertical, V representing horizontal, j representing imaginary units,/>Then the corresponding Stokes vector is represented, re (·) represents the real part of the complex number, im (·) represents the imaginary part of the complex number;
from formula (1) -formula (3):
wherein ,Tij Representing the ith row and the jth column elements of the polarization coherent matrix; q 0 ,q 1 ,q 2 ,q 3 Representing elements in a Stokes vector; deriving Stokes vectors under corresponding CTLR modes according to a model of polarization coherence matrix characterization established by the full polarization on a surface scattering mechanism, a dihedral angle scattering mechanism and a bulk scattering mechanism;
stokes vector Q of surface scattering model s The form of (2) is:
wherein β represents the ratio of the reflection coefficients of the horizontal and vertical polarizations of the reflecting surface; for an ideal surface scattering mechanism, q 3 ＝1；
Stokes vector Q of dihedral angle scattering model d The form of (2) is as follows:
wherein , x polarization reflection coefficient representing nth scattering, n=1, 2, X represents H or V, γ H and γV Representing phase changes during electromagnetic wave propagation; for an ideal dihedral scattering mechanism, q 3 ＝-1；
Stokes vector Q of volume scattering model v The form of (c) is derived from the Arii volume scattering model as follows:
wherein ,mv Representing the degree of polarization of the model, phi representing the average orientation angle; the power of the bulk scattering mechanism is obtained assuming that the surface scattering mechanism and the dihedral angle scattering mechanism are ideal conditions and maximizing the power occupied by the bulk scattering mechanismAnd satisfies the following:
the polarization degree m of the model obtained by analysis v As an index for judging whether the bulk scattering mechanism is dominant, a threshold value m for distinguishing whether the bulk scattering mechanism is dominant is set th The method comprises the steps of carrying out a first treatment on the surface of the In the region where the bulk scattering mechanism is not dominant, q is obtained from the formula (5) -formula (6) 3 Is used to distinguish the dominance of the surface scattering mechanism from the dihedral scattering mechanism; q when transmitting right-hand R polarized wave 3 0 indicates that the surface scattering mechanism is dominant, q 3 < 0 means that the dihedral scattering mechanism dominates; q when L polarized wave is transmitted 3 < 0 means that the surface scattering mechanism is dominant, q 3 0 indicates that the dihedral scattering mechanism dominates.
3. An unsupervised reduced polarization classification method based on scattering mechanism according to claim 2, wherein in step 2, fine classification is achieved by means of a generalized weisat classifier, comprising:
the covariance matrix C of the reduced polarization is:
it obeys the complex weisalde distribution, the probability density function P of which is as follows:
wherein M represents a class center, L is a multiview number, Γ 2 (L) is a constant, exp (·) represents an exponent based on e, det (·) represents a determinant of the matrix, tr (·) represents a trace of the matrix;
based on the statistical characteristics of the covariance matrix C of the reduced polarization, a distance measure d which describes the difference of the two covariance matrices is obtained, wherein the distance measure d is shown in the following formula:
wherein the subscripts m and n are used to distinguish different covariance matrices, when C m And C n At the same time, the distance metric d=0;
the measure of similarity is described by mapping the distance measure d to an interval of 0 to 1, the similarity parameter lambda being obtained by the laplace kernel mapping γ The following formula is shown:
λ γ (C m ,C n )＝e -γd (12)
wherein, gamma is a super parameter, and is determined by Monte Carlo experiments;
counting the distance measurement of 1000 pairs of adjacent pixels randomly selected from the simplified polarization data, wherein the median of the counted distance measurement is d mid Taking out
Each pixel is classified into a class with the smallest distance measure from the class center, and unsupervised classification is realized.
4. A method for classifying an unsupervised reduced polarization based on a scattering mechanism according to claim 3, wherein in the step 2, the iterative updating of the class center using the data of each class as the training sample comprises the steps of:
step (1): selecting an unsupervised classification algorithmSuper parameter m of (2) th And a filter, and inputting parameters required by the filter; setting the output class number and the convergence threshold delta of the decision class center d ；
Step (2): the ground features are divided into three main classes according to the dominance of a surface scattering mechanism, a dihedral angle scattering mechanism and a volume scattering mechanism, the ground features are divided into K classes according to the scattering power intensity of the ground features, and a covariance matrix of a sample with the scattering power being the median in each class is taken as an initial class centerThe superscript (0) represents the initial value of the iteration, and the subscript k represents the kth class; further suppressing the coherence spots on the observed data by using a filter;
step (3): calculating element C in each sample k,n The maximum likelihood distance measurement with each class center marks the element as the class to which the class center most similar to the element belongs, the subscript k, n represents the nth element in the kth class, and the class center is updated according to the following formula:
wherein ,Nk Representing the total number of elements contained in the kth class; when the class center is larger than the output class number, the class with the most similar and least element number is preferentially combined by adopting the following formula:
wherein ,k1 ,k 2 To distinguish two different classes, k 1 ∪k 2 Union of two types of mergers, normalized weightsObtained by the following formula:
step (3) is circularly executed until the change of the class center is negligible and the number is the required classification number.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310818264.7A CN116797845B (en) | 2023-07-05 | 2023-07-05 | Unsupervised reduced polarization classification method based on scattering mechanism |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310818264.7A CN116797845B (en) | 2023-07-05 | 2023-07-05 | Unsupervised reduced polarization classification method based on scattering mechanism |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116797845A true CN116797845A (en) | 2023-09-22 |
| CN116797845B CN116797845B (en) | 2024-01-26 |
Family
ID=88039970
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310818264.7A Active CN116797845B (en) | 2023-07-05 | 2023-07-05 | Unsupervised reduced polarization classification method based on scattering mechanism |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116797845B (en) |
Citations (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101498789A (en) * | 2009-02-25 | 2009-08-05 | 中国测绘科学研究院 | Ground object target classification method and apparatus based on polarimetric synthetic aperture radar |
| CN102540157A (en) * | 2010-12-08 | 2012-07-04 | 中国科学院电子学研究所 | Ground feature classifying method based on simplified polarization synthetic aperture radar data |
| CN102799896A (en) * | 2012-06-29 | 2012-11-28 | 中山大学 | POLSAR image unsupervised classification method based on target scattering identification |
| CN102968640A (en) * | 2012-10-25 | 2013-03-13 | 西安电子科技大学 | Polarized synthetic aperture radar (SAR) image classification method based on Freeman decomposition and data distribution characteristics |
| CN102999762A (en) * | 2012-10-25 | 2013-03-27 | 西安电子科技大学 | Method for classifying polarimetric SAR (synthetic aperture radar) images on the basis of Freeman decomposition and spectral clustering |
| CN103593669A (en) * | 2013-11-22 | 2014-02-19 | 中国电子科技集团公司第五十四研究所 | Method for decomposing image four components of polarization synthetic aperture radar |
| CN103870841A (en) * | 2014-03-11 | 2014-06-18 | 西安电子科技大学 | Polarization SAR (synthetic aperture radar) image classification method based on Freeman decomposition and PSO (particle swarm optimization) |
| CN103941245A (en) * | 2014-05-09 | 2014-07-23 | 中国科学院电子学研究所 | Method for Freeman-Durden target decomposition of condensed polarization interference data |
| CN107203791A (en) * | 2017-06-30 | 2017-09-26 | 西安电子科技大学 | Based on the heteropolar Classification of Polarimetric SAR Image method than with scattered power entropy |
| CN107392140A (en) * | 2017-07-19 | 2017-11-24 | 苏州闻捷传感技术有限公司 | Polarization SAR terrain classification method with incremental learning is decomposed based on low-rank sparse |
| CN109615025A (en) * | 2018-12-28 | 2019-04-12 | 内蒙古工业大学 | Quick atural object classification method |
| CN110516698A (en) * | 2019-07-15 | 2019-11-29 | 中国科学院电子学研究所 | Complete polarization image polarization decomposing method, apparatus, electronic equipment and storage medium |
| CN114202674A (en) * | 2021-12-15 | 2022-03-18 | 北京化工大学 | Interactive CNN classification method for simplified polarization SAR based on scattering mechanism |
| CN114509737A (en) * | 2022-01-25 | 2022-05-17 | 国家卫星海洋应用中心 | CTLR compact polarization SAR data scattering component decomposition method, device, equipment and medium |
| CN114565694A (en) * | 2022-03-03 | 2022-05-31 | 西北工业大学 | Satellite-borne CTLR mode compact polarization SAR reconstruction method |
| CN115542284A (en) * | 2022-12-05 | 2022-12-30 | 中国科学院空天信息创新研究院 | Lunar surface dielectric constant inversion method based on simple polarimetric SAR |
| WO2023104167A1 (en) * | 2021-12-10 | 2023-06-15 | 深圳先进技术研究院 | Fully polarimetric synthetic aperture radar image classification method and system, and terminal |
-
2023
- 2023-07-05 CN CN202310818264.7A patent/CN116797845B/en active Active
Patent Citations (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101498789A (en) * | 2009-02-25 | 2009-08-05 | 中国测绘科学研究院 | Ground object target classification method and apparatus based on polarimetric synthetic aperture radar |
| CN102540157A (en) * | 2010-12-08 | 2012-07-04 | 中国科学院电子学研究所 | Ground feature classifying method based on simplified polarization synthetic aperture radar data |
| CN102799896A (en) * | 2012-06-29 | 2012-11-28 | 中山大学 | POLSAR image unsupervised classification method based on target scattering identification |
| CN102968640A (en) * | 2012-10-25 | 2013-03-13 | 西安电子科技大学 | Polarized synthetic aperture radar (SAR) image classification method based on Freeman decomposition and data distribution characteristics |
| CN102999762A (en) * | 2012-10-25 | 2013-03-27 | 西安电子科技大学 | Method for classifying polarimetric SAR (synthetic aperture radar) images on the basis of Freeman decomposition and spectral clustering |
| CN103593669A (en) * | 2013-11-22 | 2014-02-19 | 中国电子科技集团公司第五十四研究所 | Method for decomposing image four components of polarization synthetic aperture radar |
| CN103870841A (en) * | 2014-03-11 | 2014-06-18 | 西安电子科技大学 | Polarization SAR (synthetic aperture radar) image classification method based on Freeman decomposition and PSO (particle swarm optimization) |
| CN103941245A (en) * | 2014-05-09 | 2014-07-23 | 中国科学院电子学研究所 | Method for Freeman-Durden target decomposition of condensed polarization interference data |
| CN107203791A (en) * | 2017-06-30 | 2017-09-26 | 西安电子科技大学 | Based on the heteropolar Classification of Polarimetric SAR Image method than with scattered power entropy |
| CN107392140A (en) * | 2017-07-19 | 2017-11-24 | 苏州闻捷传感技术有限公司 | Polarization SAR terrain classification method with incremental learning is decomposed based on low-rank sparse |
| CN109615025A (en) * | 2018-12-28 | 2019-04-12 | 内蒙古工业大学 | Quick atural object classification method |
| CN110516698A (en) * | 2019-07-15 | 2019-11-29 | 中国科学院电子学研究所 | Complete polarization image polarization decomposing method, apparatus, electronic equipment and storage medium |
| WO2023104167A1 (en) * | 2021-12-10 | 2023-06-15 | 深圳先进技术研究院 | Fully polarimetric synthetic aperture radar image classification method and system, and terminal |
| CN114202674A (en) * | 2021-12-15 | 2022-03-18 | 北京化工大学 | Interactive CNN classification method for simplified polarization SAR based on scattering mechanism |
| CN114509737A (en) * | 2022-01-25 | 2022-05-17 | 国家卫星海洋应用中心 | CTLR compact polarization SAR data scattering component decomposition method, device, equipment and medium |
| CN114565694A (en) * | 2022-03-03 | 2022-05-31 | 西北工业大学 | Satellite-borne CTLR mode compact polarization SAR reconstruction method |
| CN115542284A (en) * | 2022-12-05 | 2022-12-30 | 中国科学院空天信息创新研究院 | Lunar surface dielectric constant inversion method based on simple polarimetric SAR |
Non-Patent Citations (6)
| Title |
|---|
| V. KUMAR等: "Classification Assessment of Real Versus Simulated Compact and Quad-Pol Modes of ALOS-2", 《IEEE GEOSCIENCE AND REMOTE SENSING LETTERS》, vol. 16, no. 9, pages 1497 - 1501, XP011743014, DOI: 10.1109/LGRS.2019.2899268 * |
| Y. WANG等: "A Modified Four-Component Decomposition Method With Refined Volume Scattering Models", 《IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING》, vol. 13, pages 1946 - 1958, XP011789814, DOI: 10.1109/JSTARS.2020.2990691 * |
| 刘高峰: "极化SAR图像特征提取与分类方法研究", 《中国博士学位论文全文数据库信息科技辑》, no. 10, pages 136 - 141 * |
| 廖露等: "一种基于改进极化散射矩阵和H/α-Wishart分类器的极化SAR影像分类方法", 《电脑知识与技术》, vol. 15, no. 30, pages 215 - 216 * |
| 王宇等: "基于极化特征参数和极化干涉最优参数的改进四元素分解方法", 《电子与信息学报》, vol. 41, no. 12, pages 2881 - 2888 * |
| 项德良: "SAR/PolSAR图像建筑物信息提取技术研究", 《中国博士学位论文全文数据库信息科技辑》, no. 1, pages 136 - 315 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116797845B (en) | 2024-01-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Rignot et al. | Segmentation of polarimetric synthetic aperture radar data | |
| Beaulieu et al. | Segmentation of textured polarimetric SAR scenes by likelihood approximation | |
| CN103824084B (en) | Classification of Polarimetric SAR Image method based on SDIT and SVM | |
| CN101498789B (en) | Ground object target classification method and apparatus based on polarimetric synthetic aperture radar | |
| CN101908138B (en) | Identification method of image target of synthetic aperture radar based on noise independent component analysis | |
| CN102208031B (en) | Freeman decomposition and homo-polarization rate-based polarized synthetic aperture radar (SAR) image classification method | |
| CN102982338B (en) | Classification of Polarimetric SAR Image method based on spectral clustering | |
| CN102968640B (en) | Decompose and the Classification of Polarimetric SAR Image method of data distribution characteristics based on Freeman | |
| CN103839073B (en) | Polarization SAR image classification method based on polarization features and affinity propagation clustering | |
| CN104239901B (en) | Classification of Polarimetric SAR Image method based on Fuzzy particle swarm artificial and goal decomposition | |
| Park et al. | Unsupervised classification of scattering mechanisms in polarimetric SAR data using fuzzy logic in entropy and alpha plane | |
| CN110349160A (en) | One kind is based on super-pixel and fuzzy C-means clustering SAR image segmentation method | |
| CN109359525B (en) | Polarized SAR image classification method based on sparse low-rank discrimination spectral clustering | |
| Schou et al. | Restoration of polarimetric SAR images using simulated annealing | |
| Bagan et al. | Combination of AVNIR-2, PALSAR, and polarimetric parameters for land cover classification | |
| CN107203791B (en) | Polarimetric SAR image classification method based on heteropolarity ratio and scattering power entropy | |
| Kaplan | Analysis of multiplicative speckle models for template-based SAR ATR | |
| CN102999762A (en) | Method for classifying polarimetric SAR (synthetic aperture radar) images on the basis of Freeman decomposition and spectral clustering | |
| CN107742133A (en) | A kind of sorting technique for Polarimetric SAR Image | |
| CN103365985B (en) | The adaptive polarization SAR sorting technique of one kind | |
| CN105138966A (en) | Quick density peak value clustering based polarimetric SAR image classification method | |
| CN116797845B (en) | Unsupervised reduced polarization classification method based on scattering mechanism | |
| CN112989940A (en) | Raft culture area extraction method based on high-resolution three-satellite SAR image | |
| Hu et al. | Model-based Polarimetric Target Decomposition with Power Redistribution for Urban Areas | |
| Kim et al. | Detection and analysis of anisotropic scattering in SAR data |
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 |