CN111062445A - Polarimetric SAR image classification method based on collaborative regularization and superpixel - Google Patents
Polarimetric SAR image classification method based on collaborative regularization and superpixel Download PDFInfo
- Publication number
- CN111062445A CN111062445A CN201911333974.0A CN201911333974A CN111062445A CN 111062445 A CN111062445 A CN 111062445A CN 201911333974 A CN201911333974 A CN 201911333974A CN 111062445 A CN111062445 A CN 111062445A
- Authority
- CN
- China
- Prior art keywords
- matrix
- low
- super
- regularization
- superpixel
- 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
- 238000000034 method Methods 0.000 title claims abstract description 107
- 239000011159 matrix material Substances 0.000 claims abstract description 149
- 230000010287 polarization Effects 0.000 claims abstract description 51
- 230000001427 coherent effect Effects 0.000 claims abstract description 32
- 238000012545 processing Methods 0.000 claims abstract description 8
- 238000000354 decomposition reaction Methods 0.000 claims description 29
- 230000001174 ascending effect Effects 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 7
- 238000012163 sequencing technique Methods 0.000 claims description 6
- 238000012935 Averaging Methods 0.000 claims description 5
- 238000010586 diagram Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 230000003044 adaptive effect Effects 0.000 description 4
- 238000010276 construction Methods 0.000 description 4
- 238000004088 simulation Methods 0.000 description 4
- 235000016068 Berberis vulgaris Nutrition 0.000 description 2
- 241000335053 Beta vulgaris Species 0.000 description 2
- 240000002791 Brassica napus Species 0.000 description 2
- 235000004977 Brassica sinapistrum Nutrition 0.000 description 2
- 240000004658 Medicago sativa Species 0.000 description 2
- 244000046052 Phaseolus vulgaris Species 0.000 description 2
- 235000010627 Phaseolus vulgaris Nutrition 0.000 description 2
- 244000061456 Solanum tuberosum Species 0.000 description 2
- 235000002595 Solanum tuberosum Nutrition 0.000 description 2
- 241000209140 Triticum Species 0.000 description 2
- 235000021307 Triticum Nutrition 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 1
- 244000025254 Cannabis sativa Species 0.000 description 1
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 235000010624 Medicago sativa Nutrition 0.000 description 1
- 235000017587 Medicago sativa ssp. sativa Nutrition 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 235000012015 potatoes Nutrition 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000010187 selection method Methods 0.000 description 1
- 239000002689 soil Substances 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Images
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
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Multimedia (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention belongs to the technical field of image processing, and particularly relates to a method, a system and a device for classifying polarized SAR images based on collaborative regularization and superpixels, aiming at solving the problem of low classification precision of the existing polarized SAR image classification method. The method comprises the steps of obtaining a plurality of super pixels and corresponding coherent matrixes thereof through a super pixel generation method based on the obtained polarized SAR image; extracting polarization characteristics of preset dimensions of each super pixel; calculating the Wishart distance and Euclidean distance between the super pixels and other super pixels based on the coherent matrix and polarization characteristics of the super pixels, and constructing a first weight map and a second weight map of the super pixels; obtaining a first low-dimensional feature and a second low-dimensional feature through a dimension reduction model based on collaborative regularization; and obtaining a classification result of the polarized SAR image through a nearest neighbor classifier. According to the method, the classification precision of the polarized SAR image is improved by combining the Wishart distance and the polarization characteristics according to the spatial information of the pixel points.
Description
Technical Field
The invention belongs to the technical field of image processing, and particularly relates to a method, a system and a device for classifying polarized SAR images based on collaborative regularization and superpixels.
Background
The polarized SAR can not only obtain large-area high-resolution images all day long and all weather, but also obtain detailed and rich ground feature information by utilizing various polarization combinations. And thus is widely used in military and civil fields. The ground object classification is an important content for the interpretation of the polarized SAR image.
Lee et al, 1994, proposed a Wishart distance-based Classification method, which was one of the most classical methods (see references: J.S.Lee, M.R.Grunes, R.Kwok, "Classification of multi-lookup polar SAR image based on complex Wishart distribution", int.J.RemoteSensing, vol.15, No.11, 1994.). The Wishart distance used by the method is derived from the distribution of the polarized SAR data through the maximum likelihood, so that the method is suitable for polarized SAR image classification and is widely applied.
In 1999, Lee et al combined H/A/α polarization target decomposition with a Wishart classifier for polarization SAR image classification (see references: Jong-Sen Lee, M.R.Grunes, T.L.Ainsworth, Li-Jen Du, D.L.Schulerand S.R.cloud, "Unverended classified using polar resolution and the complex wiskart classifier," IEEE Transactions on Geoscience and remove Sensing, vol.37, No.5, pp.2249-8, Sept.1999. Anfine et al proposed SAR-based Spectral methods for SAR distance as initial values of the wiskart classifier, proposed as initial values for the wisdom classifier for non-negativity and symmetry distance construction for the wisdom clustering, origin, publication No. 2007, origin, and symmetry, found in the application for construction of the family of weights, see "origin, and symmetry".
In addition, the existing polarized SAR image classification includes two elements of feature extraction and classifier design. Therefore, whether suitable features can be extracted or not greatly influences the classification effect of the polarized SAR image. The polarization features are the features most commonly used for polarization SAR image classification, and can be extracted from polarization SAR data, various polarization target decomposition methods and the like. Tu et al summarize 42-dimensional polarization features and extract low-dimensional features using Laplace feature mapping (LE) for polarized SAR image classification (see references: S.T. Tu, J.Y. Chen, W.Yang, and H.Sun, "Laplacian eigenmaps-base doppler measurement two dimensional reduction for SAR image classification," IEEEtransactions on geometry and motion Sensing, vol.50, No.1, pp.170-179, Jan.2012.). Yang et al summarize the polarization features generated by Decomposition of various polarized objects and propose a CNN-Based feature selection method to select polarization features for polarized SAR Image Classification (see references C. Yang, B. Hou, B. ren, Y.Hu and L.Jiao, "CNN-Based polarized reconstruction feature for polarized SAR Image Classification," in IEEE Transactions on geoscience and Remote Sensing, 881.57, No.11, pp.8796-2, Nov.2019.).
However, the above methods have the following problems: (1) the Wisharp distance and the polarization characteristic play a crucial role in classifying the polarized SAR images, but the Wisharp distance and the polarization characteristic are not considered to be reasonably and fully combined; (2) based on a single pixel point, space information is not considered, the method is sensitive to noise, and the classification effect is not ideal.
Disclosure of Invention
In order to solve the above problems in the prior art, that is, to solve the problem that the existing polarized SAR image classification method does not consider spatial information of a pixel point and does not fully combine the Wishart distance and polarization characteristics, resulting in low classification accuracy, a first aspect of the present invention provides a method for classifying polarized SAR images based on collaborative regularization and superpixel, the method comprising:
step S100, based on the obtained polarized SAR image, obtaining a plurality of super pixels and corresponding coherent matrixes thereof through a super pixel generation method;
step S200, acquiring a covariance matrix of each super pixel; decomposing the coherent matrix and the covariance matrix respectively by presetting a plurality of types of polarized target decomposition methods, and extracting the polarization characteristics of the preset dimensionality of each super pixel by combining the coherent matrix and the covariance matrix;
step S300, calculating Wishart distances and Euclidean distances between each super pixel and other super pixels in a dxd area with each super pixel as a center based on a coherent matrix and polarization characteristics of the super pixel, and sequencing the super pixels in an ascending order; respectively acquiring the weighted values corresponding to the ordered Wishart distance and Euclidean distance by a preset weight assignment method, and constructing a first weight map and a second weight map corresponding to each superpixel;
step S400, constructing a corresponding first matrix and a second matrix for each super pixel based on the first weight map and the second weight map; based on the first matrix and the second matrix, respectively obtaining a first low-dimensional feature and a second low-dimensional feature through a pre-constructed dimension reduction model based on collaborative regularization; the first matrix and the second matrix are Laplacian matrixes of a normalized graph;
step S500, averaging the first low-dimensional features and the second low-dimensional features to obtain average features corresponding to the super pixels, and obtaining a classification result of the polarized SAR image through a nearest neighbor classifier.
In some preferred embodiments, the coherence matrix is an average of the coherence matrix corresponding to each pixel point within each superpixel.
In some preferred embodiments, the predetermined plurality of kinds of polarization target decomposition methods include Pauli decomposition, H/a/α decomposition, Freeman decomposition, Kroggar decomposition, and Huynen decomposition.
In some preferred embodiments, in step S300, "obtaining the weight values corresponding to the sorted Wishart distances and euclidean distances by using a preset weight assignment method respectively" includes:
selecting the first k Wishart distances and Euclidean distances based on the sorted Wishart distances and Euclidean distances, and respectively obtaining corresponding weight values through the following formulas;
wherein, wijThe weight value is sigma max (d) -min (d), and d is a Wishart distance or a Euclidean distance;
and setting the weight values corresponding to the rest Wishart distances and Euclidean distances to be zero.
In some preferred embodiments, in step S400, "construct the corresponding first matrix and second matrix based on the first weight map and the second weight map", the method includes:
step S410, constructing a corresponding first diagonal matrix and a second diagonal matrix based on the first weight map and the second weight map of each superpixel;
step S411, based on the first diagonal matrix and the second diagonal matrix of each super pixel, respectively combining the first weight map and the second weight map to construct a first matrix and a second matrix corresponding to each super pixel.
In some preferred embodiments, the collaborative regularization based dimension reduction model is represented by:
wherein, U(1)Is a first low-dimensional feature, L(1)Is a first matrix, U(2)Is a second low dimensional feature, L(2)For the second matrix, α is a parameter, N is the number of superpixels, M is the dimensionality after dimensionality reduction, tr (-) is the trace of the matrix, T represents the transpose, and R represents a real number.
In some preferred embodiments, in step S400, "obtain the first low-dimensional feature and the second low-dimensional feature through a pre-constructed dimension reduction model based on collaborative regularization", respectively, the method includes:
step S420, respectively sorting the eigenvalues in the first matrix and the second matrix in a descending order, and selecting eigenvectors corresponding to the first M eigenvalues as a first low-dimensional feature and a second low-dimensional feature after sorting;
step A421, based on the first low-dimensional feature, comparing matrix L(2)+αU(1)U(1)TDecomposing the characteristic values to obtain multiple characteristic values, sorting in descending order, and rankingSequentially selecting the eigenvectors corresponding to the first M eigenvalues as updated second low-dimensional characteristics;
step S422, based on the second low-dimensional characteristic, a matrix L is aligned(1)+αU(2)U(2)TDecomposing the eigenvalues to obtain a plurality of eigenvalues, sorting in a descending order, and selecting eigenvectors corresponding to the first M eigenvalues as updated first low-dimensional characteristics after sorting;
step S423, obtaining a current iteration number and a sum of the F-norm of the difference between the first low-dimensional feature and the updated first low-dimensional feature and the F-norm of the difference between the second low-dimensional feature and the updated second low-dimensional feature, outputting the first low-dimensional feature and the second low-dimensional feature if the sum is greater than a preset threshold or the current iteration number is greater than a preset maximum iteration number, otherwise, executing the steps S421 to S422 in a loop.
The invention provides a system for classifying polarimetric SAR images based on collaborative regularization and superpixel, which comprises a superpixel generation module, a polarimetric feature extraction module, a weight graph acquisition module, a feature dimension reduction module and a classification result output module;
the super-pixel generation module is configured to obtain a plurality of super-pixels and corresponding coherent matrixes thereof through a super-pixel generation method based on the obtained polarized SAR image;
the polarization characteristic extraction module is configured to acquire a covariance matrix of each super-pixel; decomposing the coherent matrix and the covariance matrix respectively by presetting a plurality of types of polarized target decomposition methods, and extracting the polarization characteristics of the preset dimensionality of each super pixel by combining the coherent matrix and the covariance matrix;
the weight graph obtaining module is configured to calculate the Wishart distance and the Euclidean distance between each superpixel and other superpixels respectively based on the coherent matrix and the polarization characteristics in a d multiplied by d area taking each superpixel as the center, and perform ascending sequencing; respectively acquiring the weighted values corresponding to the ordered Wishart distance and Euclidean distance by a preset weight assignment method, and constructing a first weight map and a second weight map corresponding to each superpixel;
the characteristic dimension reduction module is configured to construct a first matrix and a second matrix corresponding to each super pixel based on a first weight map and a second weight map of the super pixel; based on the first matrix and the second matrix, respectively obtaining a first low-dimensional feature and a second low-dimensional feature through a pre-constructed dimension reduction model based on collaborative regularization; the first matrix and the second matrix are Laplacian matrixes of a normalized graph;
the output classification result module is configured to average the first low-dimensional features and the second low-dimensional features, obtain average features corresponding to the super pixels, and obtain a classification result of the polarized SAR image through a nearest neighbor classifier.
In a third aspect of the present invention, a storage device is provided, in which a plurality of programs are stored, the programs being loaded and executed by a processor to implement the above-mentioned collaborative regularization and superpixel-based polarization SAR image classification method.
In a fourth aspect of the present invention, a processing apparatus is provided, which includes a processor, a storage device; a processor adapted to execute various programs; a storage device adapted to store a plurality of programs; the program is adapted to be loaded and executed by a processor to implement the above-described co-regularization and superpixel based polarimetric SAR image classification method.
The invention has the beneficial effects that:
according to the method, the classification precision of the polarized SAR image is improved by combining the Wishart distance and the polarization characteristics according to the spatial information of the pixel points. The method performs superpixel segmentation on the polarized SAR image by utilizing the spatial information of the pixel points, and uses the superpixel as a processing unit, thereby greatly reducing the calculation burden and the noise influence.
Meanwhile, the invention obtains a coherent matrix and a covariance matrix of a plurality of super pixels, decomposes the matrix by a plurality of types of polarized object decomposition methods, and fully extracts polarization characteristics. And respectively utilizing the Wishart distance of the coherent matrix and the Euclidean distance of the polarization characteristics, constructing a weight graph by locally searching adjacent samples, and utilizing a collaborative regular mode to extract the characteristics, thereby improving the classification precision.
Drawings
Other features, objects and advantages of the present application will become more apparent upon reading of the following detailed description of non-limiting embodiments thereof, made with reference to the accompanying drawings.
FIG. 1 is a schematic flowchart of a collaborative regularization and super-pixel based polarization SAR image classification method according to an embodiment of the present invention;
FIG. 2 is a block diagram of a collaborative regularization and super-pixel based polarimetric SAR image classification system according to an embodiment of the present invention;
FIG. 3 is a pseudo-color schematic diagram of a polarized SAR image of one embodiment of the present invention;
FIG. 4 is a schematic diagram of a polarized SAR image real terrain marking according to an embodiment of the present invention;
fig. 5 is a schematic diagram of a classification result of a pixel-point-based Wishart classification method according to an embodiment of the present invention;
FIG. 6 is a diagram illustrating classification results of a superpixel-based Wihart classification method according to an embodiment of the present invention;
fig. 7 is a schematic diagram of classification results of a co-regularization and super-pixel based polarization SAR image classification method according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The present application will be described in further detail with reference to the following drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the relevant invention and not restrictive of the invention. It should be noted that, for convenience of description, only the portions related to the related invention are shown in the drawings.
It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict.
The polarimetric SAR image classification method based on collaborative regularization and super-pixel disclosed by the invention comprises the following steps as shown in figure 1:
step S100, based on the obtained polarized SAR image, obtaining a plurality of super pixels and corresponding coherent matrixes thereof through a super pixel generation method;
step S200, acquiring a covariance matrix of each super pixel; decomposing the coherent matrix and the covariance matrix respectively by presetting a plurality of types of polarized target decomposition methods, and extracting the polarization characteristics of the preset dimensionality of each super pixel by combining the coherent matrix and the covariance matrix;
step S300, calculating Wishart distances and Euclidean distances between each super pixel and other super pixels in a dxd area with each super pixel as a center based on a coherent matrix and polarization characteristics of the super pixel, and sequencing the super pixels in an ascending order; respectively acquiring the weighted values corresponding to the ordered Wishart distance and Euclidean distance by a preset weight assignment method, and constructing a first weight map and a second weight map corresponding to each superpixel;
step S400, constructing a corresponding first matrix and a second matrix for each super pixel based on the first weight map and the second weight map; based on the first matrix and the second matrix, respectively obtaining a first low-dimensional feature and a second low-dimensional feature through a pre-constructed dimension reduction model based on collaborative regularization; the first matrix and the second matrix are Laplacian matrixes of a normalized graph;
step S500, averaging the first low-dimensional features and the second low-dimensional features to obtain average features corresponding to the super pixels, and obtaining a classification result of the polarized SAR image through a nearest neighbor classifier.
In order to more clearly describe the method for classifying the polarimetric SAR images based on the co-regularization and the super-pixel, the following describes in detail the steps of an embodiment of the method of the present invention with reference to the accompanying drawings.
And S100, obtaining a plurality of super pixels and corresponding coherence matrixes thereof through a super pixel generation method based on the obtained polarized SAR image.
Synthetic Aperture Radar (SAR) is an all-weather and all-time active microwave remote sensing imaging radar and is widely applied to the fields of geological survey, disaster control, parameter inversion, military and the like. The polarization SAR can record complete polarization scattering information of a ground target, and the accuracy of image interpretation and analysis is greatly improved. While the classification of polarized images is the basis and precondition for object-oriented polarized SAR image interpretation processing.
In this embodiment, a plurality of superpixels are acquired by an adaptive superpixel generation method based on the acquired polarized SAR image. Among them, the adaptive super image generation method can be referred to as: "D.X.Ban, Y.Ban, W.Wang and Y.Su", "Adaptive Superpixel Generation for polar SAR Images With local Adaptive Clustering and SIRV Model", "in IEEE Transactions on Geoscience and Remote Sensing, vol.55, No.6, pp.3115-3131, June 2017. According to the obtained multiple superpixels, calculating the average number of the coherent matrixes containing all the pixel points in each superpixel to be used as the coherent matrix T of each superpixeliAnd i is a subscript.
Step S200, acquiring a covariance matrix of each super pixel; and decomposing the coherent matrix and the covariance matrix respectively by presetting a plurality of types of polarized target decomposition methods, and extracting the polarization characteristics of preset dimensionality of each super pixel by combining the coherent matrix and the covariance matrix.
In this embodiment, a covariance matrix of each super-pixel is obtained, the covariance matrix and the coherence matrix are decomposed by a plurality of types of polarization target decomposition methods, such as Pauli decomposition, H/a/α decomposition, Freeman decomposition, Kroggar decomposition, and Huynen decomposition, and a polarization feature of a preset dimension of each pixel point in each super-pixel is extracted and normalized to 0 to 1 by combining the coherence matrix and the covariance matrix.
Calculating the average of polarization characteristics of all pixel points contained in each super pixel asPolarization characteristic x of super pixeliIf the number of the superpixels is N, the polarization characteristic corresponding to the polarized SAR image is
Step S300, calculating Wishart distances and Euclidean distances between each super pixel and other super pixels in a dxd area with each super pixel as a center based on a coherent matrix and polarization characteristics of the super pixel, and sequencing the super pixels in an ascending order; and respectively acquiring the weight values corresponding to the ordered Wishart distance and Euclidean distance by a preset weight assignment method, and constructing a first weight map and a second weight map corresponding to each superpixel.
In this embodiment, the distances d between the pixels and the Wishart are respectively searched in the d × d area with each super pixel as the centerWEuclidean distance dPIf the nearest k superpixels are adjacent samples of the superpixel i, acquiring corresponding weight values through a formula (1) to construct a sparse weight map W(1)、W(2). Equation (1) is as follows:
wherein, wijAnd d is a Wishart distance or a Euclidean distance.
The specific process of constructing the weight graph is as follows:
within the d × d region centered on each superpixel, the Wishart distance between the superpixel and other superpixels is calculated based on the coherence matrix. In this embodiment, the Wishart distance satisfying the nonnegativity and symmetry is used, and the calculation is shown in formula (2):
wherein j is a subscript and tr (-) is a trace of the matrix.
And (4) performing ascending sorting according to the obtained Wishart distances, selecting the first k Wishart distances, and obtaining the corresponding weight values through a formula (3). Equation (3) is as follows:
wherein,
and the weighted value is corresponding to the Wishart distance.
And setting the weight values corresponding to other Wishart distances as 0, and constructing a first weight map corresponding to each superpixel based on the obtained weight values.
Similarly, in a d × d region centered on each superpixel, the euclidean distances between the superpixel and other superpixels are calculated based on the polarization characteristics of the superpixels, ascending sorting is performed according to the obtained euclidean distances, the first k euclidean distances are selected, and the corresponding weight values are obtained through the formula (4). Equation (4) is as follows:
wherein,
the weight values are corresponding to Euclidean distances.
And setting the weight values corresponding to other Euclidean distances as 0, and constructing a second weight map corresponding to each superpixel based on the obtained weight values.
Step S400, constructing a corresponding first matrix and a second matrix for each super pixel based on the first weight map and the second weight map; based on the first matrix and the second matrix, respectively obtaining a first low-dimensional feature and a second low-dimensional feature through a pre-constructed low-dimensional model based on collaborative regularization; the first matrix and the second matrix are Laplacian matrixes of the normalized graph.
In the present embodiment, for each super pixel, a corresponding first matrix and a corresponding second matrix are constructed based on the first weight map and the second weight map. The specific treatment steps are as follows:
step S410, based on each super pixelThe first weight map and the second weight map, and a corresponding first diagonal matrix D is constructed(1)A second diagonal matrix D(2)。
The construction process is shown in formula (5) (6):
step S411, based on the first diagonal matrix and the second diagonal matrix of each super pixel, respectively combining the first weight map and the second weight map to construct a first matrix L corresponding to each super pixel(1)A second matrix L(2). Wherein L is(1)、L(2)Is W(1)、W(2)Corresponding normalized graph Laplacian matrix.
The construction process is shown in the formula (7) (8):
L(1)＝I-D(1)-1/2W(1)D(1)-1/2(7)
L(2)＝I-D(2)-1/2W(2)D(2)-1/2(8)
wherein I is an identity matrix.
And constructing a dimension reduction model based on collaborative regularization through the first matrix and the second matrix. The constructed dimension reduction model based on the collaborative regularization is shown as a formula (9):
wherein, U(1)Is a first low-dimensional feature, L(1)Is a first matrix, U(2)Is a second low dimensional feature, L(2)For the second matrix, α is a parameter, M is the dimensionality after dimensionality reduction, T represents the transpose, R represents the real number,
is the square of the F-norm. Book of JapaneseThe third term in the formula is a regular term.
For simplicity, the above formula can be expressed as formula (10):
and obtaining a first low-dimensional feature and a second low-dimensional feature through the constructed dimension reduction model based on the collaborative regularization. The specific treatment steps are as follows:
step S420, respectively sorting the eigenvalues in the first matrix and the second matrix in a descending order, and selecting eigenvectors corresponding to the first M eigenvalues as a first low-dimensional characteristic and a second low-dimensional characteristic after sorting;
step A421, based on the first low-dimensional feature, the matrix L(2)+αU(1)U(1)TDecomposing the eigenvalues to obtain a plurality of eigenvalues, sorting in a descending order, and selecting eigenvectors corresponding to the first M eigenvalues as updated second low-dimensional characteristics after sorting; namely fixing the first low-dimensional feature and solving the second low-dimensional feature;
step S422, based on the second low-dimensional feature, the matrix L is processed(1)+αU(2)U(2)TDecomposing the eigenvalues to obtain a plurality of eigenvalues, sorting in a descending order, and selecting eigenvectors corresponding to the first M eigenvalues as updated first low-dimensional characteristics after sorting;
step S423, obtaining the current iteration number and the sum of the F-norm of the difference between the first low-dimensional feature and the updated first low-dimensional feature and the F-norm of the difference between the second low-dimensional feature and the updated second low-dimensional feature, and if the sum is smaller than a preset threshold or the current iteration number is smaller than a preset maximum iteration number, executing the method of steps S421 to S422 in a loop. Wherein, the judgment of the preset threshold is shown as the formula (11):
wherein,
for the first and second low-dimensional features that are not updated,
and epsilon is a preset threshold value for the updated first and second low-dimensional features.
Otherwise, outputting the final first low-dimensional feature and the second low-dimensional feature.
Step S500, averaging the first low-dimensional features and the second low-dimensional features to obtain average features corresponding to the super pixels, and obtaining a classification result of the polarized SAR image through a nearest neighbor classifier.
In this embodiment, the first low-dimensional features and the second low-dimensional features of the respective super pixels are averaged to obtain an average feature as a feature for classification. Wherein, the process of averaging is shown as formula (12):
U＝(U(1)+U(2))/2 (12)
wherein U is the average characteristic.
And obtaining the classification result of the polarized SAR image through a nearest neighbor classifier according to the average characteristic of each super pixel.
In the invention, the method is compared with a Wishart classification method based on pixel points and a Wishart classification method based on superpixels through simulation experiments, so that the effectiveness of the method in classifying the polarized SAR images is proved. And randomly selecting 10% of each class as a training set, and using the rest as a test set to obtain the classification accuracy.
The simulation experiments were performed in a server hardware environment of Intel (R) core (TM) i9-8950HK CPU 2.90GHz 32G RAM and in a software environment of Matlab R2016 a. The experimental data used were the farmland data in the Flevoland (plain) region, with a size of 200 × 320.
Table 1 shows the classification accuracy obtained in the simulation for three methods:
TABLE 1
As can be seen from table 1, the classification accuracy obtained by the method is obviously higher than that of the classic Wishart classification method, and the classification accuracy of the Wishart classification method based on the superpixel is higher than that of the Wishart classification method based on a single pixel point, so that the advantages of the method based on the superpixel are explained, and the method provided by the invention has a good classification effect on the classification of the polarized SAR image. In table 1, english on the left side is the name of crops planted in each field, Potatoes is potato, Grass is grassland, beets is Beet, Lucerne is alfalfa, Wheat is Wheat field, Stem beans is dry bean, Bare soil is Bare land, and Rapeseed is Rapeseed. Total accuracy is the Total accuracy.
In order to better illustrate the classification effect, the present invention is shown by the figure. Fig. 3 is a pseudo-color image of a polarized SAR image for a simulation experiment, fig. 4 is a real ground object labeling image of the polarized SAR image, fig. 5 is a classification result image of a Wishart classification method based on pixel points, fig. 6 is a classification result image of a Wishart classification method based on superpixels, and fig. 7 is a classification result image of the method of the present invention. By comparison, the method of the invention has good classification effect.
A polarized SAR image classification system based on collaborative regularization and superpixel according to a second embodiment of the present invention, as shown in fig. 2, includes: the system comprises a superpixel generation module 100, a polarization feature extraction module 200, a weight map acquisition module 300, a feature dimension reduction module 400 and a classification result output module 500;
the super-pixel generation module 100 is configured to obtain a plurality of super-pixels and corresponding coherence matrices thereof by a super-pixel generation method based on the obtained polarized SAR image;
the polarization feature extraction module 200 is configured to obtain a covariance matrix of each super-pixel; decomposing the coherent matrix and the covariance matrix respectively by presetting a plurality of types of polarized target decomposition methods, and extracting the polarization characteristics of the preset dimensionality of each super pixel by combining the coherent matrix and the covariance matrix;
the weight map obtaining module 300 is configured to calculate the Wishart distance and the euclidean distance between the super-pixel and other super-pixels based on the coherent matrix and the polarization characteristics of the super-pixel in the d × d region with each super-pixel as the center, and perform ascending sorting; respectively acquiring the weighted values corresponding to the ordered Wishart distance and Euclidean distance by a preset weight assignment method, and constructing a first weight map and a second weight map corresponding to each superpixel;
the feature dimension reduction module 400 is configured to construct, for each super pixel, a corresponding first matrix and a corresponding second matrix based on the first weight map and the second weight map; based on the first matrix and the second matrix, respectively obtaining a first low-dimensional feature and a second low-dimensional feature through a pre-constructed dimension reduction model based on collaborative regularization; the first matrix and the second matrix are Laplacian matrixes of a normalized graph;
the output classification result module 500 is configured to average the first low-dimensional features and the second low-dimensional features, obtain average features corresponding to the super pixels, and obtain a classification result of the polarized SAR image through a nearest neighbor classifier.
It can be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working process and related description of the system described above may refer to the corresponding process in the foregoing method embodiment, and details are not described herein again.
It should be noted that, the polarization SAR image classification system based on collaborative regularization and super-pixel provided in the foregoing embodiment is only illustrated by the division of the above functional modules, and in practical applications, the above functions may be allocated to different functional modules according to needs, that is, the modules or steps in the embodiments of the present invention are further decomposed or combined, for example, the modules in the foregoing embodiments may be combined into one module, or may be further split into multiple sub-modules, so as to complete all or part of the above described functions. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the modules or steps, and are not to be construed as unduly limiting the present invention.
A storage device according to a third embodiment of the present invention stores a plurality of programs, and the programs are adapted to be loaded by a processor and to implement the above-mentioned method for classifying a polarimetric SAR image based on co-regularization and super-pixel.
A processing apparatus according to a fourth embodiment of the present invention includes a processor, a storage device; a processor adapted to execute various programs; a storage device adapted to store a plurality of programs; the program is adapted to be loaded and executed by a processor to implement the above-described co-regularization and superpixel based polarimetric SAR image classification method.
It can be clearly understood by those skilled in the art that, for convenience and brevity, the specific working processes and related descriptions of the storage device and the processing device described above may refer to the corresponding processes in the foregoing method examples, and are not described herein again.
Those of skill in the art would appreciate that the various illustrative modules, method steps, and modules described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that programs corresponding to the software modules, method steps may be located in Random Access Memory (RAM), memory, Read Only Memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. To clearly illustrate this interchangeability of electronic hardware and software, various illustrative components and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as electronic hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
The terms "first," "second," and the like are used for distinguishing between similar elements and not necessarily for describing or implying a particular order or sequence.
The terms "comprises," "comprising," or any other similar term are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.
So far, the technical solutions of the present invention have been described in connection with the preferred embodiments shown in the drawings, but it is easily understood by those skilled in the art that the scope of the present invention is obviously not limited to these specific embodiments. Equivalent changes or substitutions of related technical features can be made by those skilled in the art without departing from the principle of the invention, and the technical scheme after the changes or substitutions can fall into the protection scope of the invention.
Claims (10)
1. A polarized SAR image classification method based on collaborative regularization and superpixel is characterized by comprising the following steps:
step S100, based on the obtained polarized SAR image, obtaining a plurality of super pixels and corresponding coherent matrixes thereof through a super pixel generation method;
step S200, acquiring a covariance matrix of each super pixel; decomposing the coherent matrix and the covariance matrix respectively by presetting a plurality of types of polarized target decomposition methods, and extracting the polarization characteristics of the preset dimensionality of each super pixel by combining the coherent matrix and the covariance matrix;
step S300, calculating Wishart distances and Euclidean distances between each super pixel and other super pixels in a dxd area with each super pixel as a center based on a coherent matrix and polarization characteristics of the super pixel, and sequencing the super pixels in an ascending order; respectively acquiring the weighted values corresponding to the ordered Wishart distance and Euclidean distance by a preset weight assignment method, and constructing a first weight map and a second weight map corresponding to each superpixel;
step S400, constructing a corresponding first matrix and a second matrix for each super pixel based on the first weight map and the second weight map; based on the first matrix and the second matrix, respectively obtaining a first low-dimensional feature and a second low-dimensional feature through a pre-constructed dimension reduction model based on collaborative regularization; the first matrix and the second matrix are Laplacian matrixes of a normalized graph;
step S500, averaging the first low-dimensional features and the second low-dimensional features to obtain average features corresponding to the super pixels, and obtaining a classification result of the polarized SAR image through a nearest neighbor classifier.
2. The method for classifying polarimetric SAR images based on co-regularization and superpixels according to claim 1, wherein the coherence matrix is an average value of the coherence matrix corresponding to each pixel point within each superpixel.
3. The collaborative regularization and superpixel based polarimetric SAR image classification method according to claim 1, wherein the preset multiple kinds of polarimetric target decomposition methods include Pauli decomposition, H/a/α decomposition, Freeman decomposition, Kroggar decomposition, Huynen decomposition.
4. The polarimetric SAR image classification method based on the collaborative regularization and the superpixel as claimed in claim 1, wherein in step S300, "the weighted values corresponding to the ordered Wishart distances and euclidean distances are respectively obtained by a preset weight assignment method", and the method thereof is as follows:
selecting the first k Wishart distances and Euclidean distances based on the sorted Wishart distances and Euclidean distances, and respectively obtaining corresponding weight values through the following formulas:
wherein, wijThe weight value is sigma max (d) -min (d), and d is a Wishart distance or a Euclidean distance;
and setting the weight values corresponding to the rest Wishart distances and Euclidean distances to be zero.
5. The method for classifying polarized SAR images based on co-regularization and super-pixel as claimed in claim 1, wherein in step S400, "construct the corresponding first matrix and second matrix based on the first weight map and the second weight map", the method comprises:
step S410, constructing a corresponding first diagonal matrix and a second diagonal matrix based on the first weight map and the second weight map of each superpixel;
step S411, based on the first diagonal matrix and the second diagonal matrix of each super pixel, respectively combining the first weight map and the second weight map to construct a first matrix and a second matrix corresponding to each super pixel.
6. The collaborative regularization and superpixel based polarimetric SAR image classification method according to claim 1, characterized in that said collaborative regularization based dimension reduction model is represented as:
wherein, U(1)Is a first low-dimensional feature, L(1)Is a first matrix, U(2)Is a second low dimensional feature, L(2)For the second matrix, α is a parameter, N is the number of superpixels, M is the dimensionality after dimensionality reduction, tr (-) is the trace of the matrix, T represents the transpose, and R represents a real number.
7. The polarimetric SAR image classification method based on co-regularization and superpixel as claimed in claim 6, wherein in step S400, "obtain the first low-dimensional feature and the second low-dimensional feature respectively through the pre-constructed dimension reduction model based on co-regularization", the method is as follows:
step S420, respectively sorting the eigenvalues in the first matrix and the second matrix in a descending order, and selecting eigenvectors corresponding to the first M eigenvalues as a first reduced-dimension characteristic and a second low-dimension characteristic after sorting;
step A421, based on the first low-dimensional feature, comparing matrix L(2)+αU(1)U(1)TDecomposing the eigenvalues to obtain a plurality of eigenvalues, sorting in a descending order, and selecting eigenvectors corresponding to the first M eigenvalues as updated second low-dimensional characteristics after sorting;
step S422, based on the second low-dimensional characteristic, a matrix L is aligned(1)+αU(2)U(2)TDecomposing the eigenvalues to obtain a plurality of eigenvalues, sorting in a descending order, and selecting eigenvectors corresponding to the first M eigenvalues as updated first low-dimensional characteristics after sorting;
step S423, obtaining a current iteration number and a sum of the F-norm of the difference between the first low-dimensional feature and the updated first low-dimensional feature and the F-norm of the difference between the second low-dimensional feature and the updated second low-dimensional feature, outputting the first low-dimensional feature and the second low-dimensional feature if the sum is greater than a preset threshold or the current iteration number is greater than a preset maximum iteration number, otherwise, executing the steps S421 to S422 in a loop.
8. A polarization SAR image classification system based on collaborative regularization and superpixel is characterized by comprising a superpixel generation module, a polarization feature extraction module, a weight map acquisition module, a feature dimension reduction module and a classification result output module;
the super-pixel generation module is configured to obtain a plurality of super-pixels and corresponding coherent matrixes thereof through a super-pixel generation method based on the obtained polarized SAR image;
the polarization characteristic extraction module is configured to acquire a covariance matrix of each super-pixel; decomposing the coherent matrix and the covariance matrix respectively by presetting a plurality of types of polarized target decomposition methods, and extracting the polarization characteristics of the preset dimensionality of each super pixel by combining the coherent matrix and the covariance matrix;
the weight graph obtaining module is configured to calculate the Wishart distance and the Euclidean distance between each superpixel and other superpixels respectively based on the coherent matrix and the polarization characteristics in a d multiplied by d area taking each superpixel as the center, and perform ascending sequencing; respectively acquiring the weighted values corresponding to the ordered Wishart distance and Euclidean distance by a preset weight assignment method, and constructing a first weight map and a second weight map corresponding to each superpixel;
the characteristic dimension reduction module is configured to construct a first matrix and a second matrix corresponding to each super pixel based on a first weight map and a second weight map of the super pixel; based on the first matrix and the second matrix, respectively obtaining a first low-dimensional feature and a second low-dimensional feature through a pre-constructed dimension reduction model based on collaborative regularization; the first matrix and the second matrix are Laplacian matrixes of a normalized graph;
the output classification result module is configured to average the first low-dimensional features and the second low-dimensional features, obtain average features corresponding to the super pixels, and obtain a classification result of the polarized SAR image through a nearest neighbor classifier.
9. A storage device having stored therein a plurality of programs, wherein said program applications are loaded and executed by a processor to implement the collaborative regularization and superpixel based polarimetric SAR image classification method according to any one of claims 1 to 7.
10. A processing device comprising a processor, a storage device; a processor adapted to execute various programs; a storage device adapted to store a plurality of programs; characterized in that said program is adapted to be loaded and executed by a processor to implement a co-regularization and super-pixel based polarimetric SAR image classification method according to any of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911333974.0A CN111062445B (en) | 2019-12-23 | 2019-12-23 | Polarized SAR image classification method based on collaborative regularization and superpixel |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911333974.0A CN111062445B (en) | 2019-12-23 | 2019-12-23 | Polarized SAR image classification method based on collaborative regularization and superpixel |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111062445A true CN111062445A (en) | 2020-04-24 |
| CN111062445B CN111062445B (en) | 2023-05-19 |
Family
ID=70301487
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911333974.0A Active CN111062445B (en) | 2019-12-23 | 2019-12-23 | Polarized SAR image classification method based on collaborative regularization and superpixel |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111062445B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114004998A (en) * | 2021-11-03 | 2022-02-01 | 中国人民解放军国防科技大学 | Unsupervised polarization SAR image terrain classification method based on multi-view tensor product diffusion |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103294792A (en) * | 2013-05-13 | 2013-09-11 | 西安电子科技大学 | Polarimetric SAR (synthetic aperture radar) terrain classification method based on semantic information and polarimetric decomposition |
| US10042048B1 (en) * | 2014-02-20 | 2018-08-07 | National Technology & Engineering Solutions Of Sandia, Llc | Superpixels for improved structure and terrain classification using multiple synthetic aperture radar image products |
-
2019
- 2019-12-23 CN CN201911333974.0A patent/CN111062445B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103294792A (en) * | 2013-05-13 | 2013-09-11 | 西安电子科技大学 | Polarimetric SAR (synthetic aperture radar) terrain classification method based on semantic information and polarimetric decomposition |
| US10042048B1 (en) * | 2014-02-20 | 2018-08-07 | National Technology & Engineering Solutions Of Sandia, Llc | Superpixels for improved structure and terrain classification using multiple synthetic aperture radar image products |
Non-Patent Citations (2)
| Title |
|---|
| CHEN YANG 等: ""CNN-Based Polarimetric Decomposition Feature Selection for PolSAR Image Classification"", 《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》 * |
| SHANG TAN TU 等: ""Laplacian Eigenmaps-Based Polarimetric Dimensionality Reduction for SAR Image Classification"", 《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114004998A (en) * | 2021-11-03 | 2022-02-01 | 中国人民解放军国防科技大学 | Unsupervised polarization SAR image terrain classification method based on multi-view tensor product diffusion |
| CN114004998B (en) * | 2021-11-03 | 2024-01-19 | 中国人民解放军国防科技大学 | Unsupervised polarization SAR image ground object classification method based on multi-vision tensor product diffusion |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111062445B (en) | 2023-05-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109711446B (en) | Ground feature classification method and device based on multispectral image and SAR image | |
| Erinjery et al. | Mapping and assessment of vegetation types in the tropical rainforests of the Western Ghats using multispectral Sentinel-2 and SAR Sentinel-1 satellite imagery | |
| Teluguntla et al. | A 30-m landsat-derived cropland extent product of Australia and China using random forest machine learning algorithm on Google Earth Engine cloud computing platform | |
| Waske et al. | Classifier ensembles for land cover mapping using multitemporal SAR imagery | |
| Adelabu et al. | Exploiting machine learning algorithms for tree species classification in a semiarid woodland using RapidEye image | |
| Zhu et al. | Assessment of spectral, polarimetric, temporal, and spatial dimensions for urban and peri-urban land cover classification using Landsat and SAR data | |
| Hariharan et al. | A novel phenology based feature subset selection technique using random forest for multitemporal PolSAR crop classification | |
| Liu et al. | Unsupervised change detection in multispectral remote sensing images via spectral-spatial band expansion | |
| Maghsoudi et al. | Polarimetric classification of Boreal forest using nonparametric feature selection and multiple classifiers | |
| Du et al. | A novel binary tree support vector machine for hyperspectral remote sensing image classification | |
| Sonobe et al. | Discrimination of crop types with TerraSAR-X-derived information | |
| Chen et al. | Decision-level fusion of Sentinel-1 SAR and Landsat 8 OLI texture features for crop discrimination and classification: case of Masvingo, Zimbabwe | |
| CN102982338B (en) | Classification of Polarimetric SAR Image method based on spectral clustering | |
| Forkuor et al. | Evaluating the sequential masking classification approach for improving crop discrimination in the Sudanian Savanna of West Africa | |
| Izquierdo-Verdiguier et al. | Semisupervised kernel feature extraction for remote sensing image analysis | |
| Bagan et al. | Combination of AVNIR-2, PALSAR, and polarimetric parameters for land cover classification | |
| CN111680579B (en) | Remote sensing image classification method for self-adaptive weight multi-view measurement learning | |
| Kavzoglu et al. | Dimensionality reduction and classification of hyperspectral images using object-based image analysis | |
| Zeyada et al. | Evaluation of the discrimination capability of full polarimetric SAR data for crop classification | |
| Shendryk et al. | Leveraging high-resolution satellite imagery and gradient boosting for invasive weed mapping | |
| CN105138966B (en) | Classification of Polarimetric SAR Image method based on fast density peak value cluster | |
| CN108460402A (en) | Polarimetric SAR image supervised classification method and device | |
| Abdullahi et al. | An unsupervised two-stage clustering approach for forest structure classification based on X-band InSAR data—A case study in complex temperate forest stands | |
| Tokar et al. | Using the Random Forest classification for land cover interpretation of Landsat images in the Prykarpattya region of Ukraine | |
| Jawak et al. | Advancement in landcover classification using very high resolution remotely sensed 8-band WorldView-2 satellite 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 |