CN114004998B - Unsupervised polarization SAR image ground object classification method based on multi-vision tensor product diffusion - Google Patents
Unsupervised polarization SAR image ground object classification method based on multi-vision tensor product diffusion Download PDFInfo
- Publication number
- CN114004998B CN114004998B CN202111293497.7A CN202111293497A CN114004998B CN 114004998 B CN114004998 B CN 114004998B CN 202111293497 A CN202111293497 A CN 202111293497A CN 114004998 B CN114004998 B CN 114004998B
- Authority
- CN
- China
- Prior art keywords
- view
- tensor product
- sar image
- graph
- feature vectors
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 135
- 238000009792 diffusion process Methods 0.000 title claims abstract description 71
- 230000010287 polarization Effects 0.000 title claims abstract description 22
- 239000013598 vector Substances 0.000 claims abstract description 159
- 239000011159 matrix material Substances 0.000 claims abstract description 85
- 238000007499 fusion processing Methods 0.000 claims abstract description 16
- 230000003595 spectral effect Effects 0.000 claims abstract description 14
- 238000004364 calculation method Methods 0.000 claims description 12
- 230000011218 segmentation Effects 0.000 claims description 11
- 238000010586 diagram Methods 0.000 description 12
- 238000002474 experimental method Methods 0.000 description 12
- 238000004590 computer program Methods 0.000 description 10
- 238000003860 storage Methods 0.000 description 10
- 238000004422 calculation algorithm Methods 0.000 description 8
- 238000011156 evaluation Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000000354 decomposition reaction Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 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 2
- 238000007635 classification algorithm Methods 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 229910052500 inorganic mineral Inorganic materials 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 239000011707 mineral Substances 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- YREOLPGEVLLKMB-UHFFFAOYSA-N 3-methylpyridin-1-ium-2-amine bromide hydrate Chemical compound O.[Br-].Cc1ccc[nH+]c1N YREOLPGEVLLKMB-UHFFFAOYSA-N 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000005336 cracking Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000011065 in-situ storage Methods 0.000 description 1
- 238000003064 k means clustering Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000000691 measurement method 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
- 230000003287 optical effect Effects 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 230000000149 penetrating effect Effects 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000011524 similarity measure Methods 0.000 description 1
- 230000007480 spreading Effects 0.000 description 1
- 238000003892 spreading Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/29—Graphical models, e.g. Bayesian networks
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The application relates to an unsupervised polarization SAR image ground object classification method based on multi-view tensor product diffusion. The method comprises the following steps: dividing a polarized SAR image to be classified by adopting a rapid superpixel division method to obtain a plurality of superpixels, constructing 3 image models based on the divided superpixels by utilizing the 3 high-dimensional feature vectors, carrying out multi-view tensor product operation according to the 3 image models to obtain a plurality of multi-view tensor product images, carrying out linear fusion processing according to the multi-view tensor product images to obtain a fused multi-view tensor product image, diffusing according to a similarity matrix of the fused multi-view tensor product image to obtain a diffused similarity matrix, and carrying out spectral clustering on the diffused similarity matrix to realize classification of ground features in the polarized SAR image. By adopting the method, the interference of speckle noise can be reduced, and the classification precision is effectively improved.
Description
Technical Field
The application relates to the technical field of polarized SAR image classification, in particular to an unsupervised polarized SAR image ground object classification method based on multi-view tensor product diffusion.
Background
Synthetic aperture radar (synthetic aperture radar, SAR) is favored in the remote sensing field as a technical means with all-weather, all-day and high-resolution imaging capability in various remote sensing technologies, and becomes an indispensable important branch in the remote sensing information acquisition technology. The polarized SAR (Polarimetric synthetic aperture radar, polSAR) alternately transmits and receives radar signals in a horizontal polarization mode and a vertical polarization mode, and can obtain more abundant scattering information of a target object by means of the penetrating capacity of microwaves to the ground object. Therefore, interpretation research on polarized SAR images is particularly urgent, and feature classification is one of the most fundamental and critical tasks in polarized SAR image interpretation. The method can realize the monitoring of the growth condition of crops, can also be used for researching the distribution condition of geology and mineral products, the development transition of cities, the exploration of mineral resources, the evaluation of natural disasters and the like, and can be widely applied to the military and civil fields.
Because of the differences in imaging modes, polarized SAR images have different characterization modes from optical images and also contain more speckle noise, which makes polarized SAR image feature classification a very challenging task all the time. The existing method mainly has the following problems:
(1) The existing classification method based on supervision is low in automation degree and poor in robustness. The supervision classification method generally needs to use a completely labeled data set to train a method model, and needs more priori knowledge, which consumes a great deal of manpower and material resources and does not meet the requirements of practical application.
(2) The classification method based on the pixel points is large in calculated amount and poor in noise resistance. For a large-scale remote sensing image, the pixel-by-pixel classification method has heavy calculation load, and neglects the regional information in the image, so that the interference of speckle noise cannot be reduced well.
(3) A common way to construct high-dimensional feature vectors is feature stacking, which results in a reduced or lost classification capability for part of the single view feature vectors. The polarization characteristics can objectively represent the similarity among data points in the polarized SAR image, and are key elements of a polarized SAR image classification system. In general, when a high-dimensional feature vector is constructed in a polarized SAR image classification method, the extracted feature vector is normalized and then directly added, and the performance of the single-view feature vector is weakened and classification errors are introduced.
Disclosure of Invention
In view of the above, it is necessary to provide an unsupervised polarization SAR image ground object classification method based on multi-view tensor product diffusion, which can effectively improve classification accuracy.
An unsupervised polarized SAR image ground object classification method based on multi-view tensor product diffusion, the method comprising:
acquiring a polarized SAR image to be classified, and segmenting the polarized SAR image by adopting a rapid superpixel segmentation method to obtain a plurality of superpixels;
extracting 5 representative feature vectors according to the polarized SAR image, combining the 5 feature vectors to obtain 3 high-dimensional feature vectors, and constructing 3 graph models based on a plurality of segmented super pixels by utilizing the 3 high-dimensional feature vectors;
performing multi-view tensor product operation according to 3 graph models to obtain a plurality of multi-view tensor product graphs, performing linear fusion processing according to the plurality of multi-view tensor product graphs to obtain a fused multi-view tensor product graph, and performing diffusion based on a similarity matrix of the fused multi-view tensor product graph to obtain a diffused similarity matrix;
and performing spectral clustering according to the diffused similarity matrix to realize the classification of the ground features of the polarized SAR image.
In one embodiment, the combining according to the 5 feature vectors to obtain 3 high-dimensional feature vectors, and constructing 3 graph models based on the segmented multiple super pixels by using the 3 high-dimensional feature vectors specifically includes:
Extracting 5 feature vectors from each pixel point in the polarized SAR image, and combining the extracted 5 feature vectors according to the corresponding pixel points to obtain 3 high-dimensional feature vectors corresponding to the pixel points;
calculating an average high-dimensional feature vector of each pixel point in the same super pixel in the plurality of super pixels on 3 high-dimensional feature vectors as the feature vector of the super pixel, so that each super pixel corresponds to 3 feature vectors;
and respectively constructing a graph model according to the different 3 feature vectors to obtain 3 corresponding graph models.
In one embodiment, the 5 representative feature vectors include: yamaguchi4 eigenvectors, krogager eigenvectors, HSI color space eigenvectors, cloude-potter's eigenvectors, eigenvectors consisting of scattered power entropy and co-polarization.
In one embodiment, the combining the 5 feature vectors corresponding to each pixel to obtain 3 high-dimensional feature vectors corresponding to each pixel includes:
forming one high-dimensional characteristic vector by the Yamaguchi4 characteristic vector, the HSI color space characteristic vector, the Cloude-Pottier's characteristic vector and the characteristic vector consisting of scattering power entropy and same polarization rate;
One of the high-dimensional feature vectors is composed of a Yamaguchi4 feature vector, a Krogager feature vector, a Cloude-Pottier's feature vector and a feature vector composed of scattering power entropy and same polarization rate;
and forming one high-dimensional characteristic vector by the Yamaguchi4 characteristic vector, the Krogager characteristic vector, the HSI color space characteristic vector, the Cloude-Pottier's characteristic vector and the characteristic vector formed by scattering power entropy and the same polarizability.
In one embodiment, the graph model is composed of a plurality of nodes and edges between two adjacent nodes; wherein each of the nodes represents a different one of the plurality of superpixels and the edge represents a degree of similarity by superpixels at both ends of the edge.
In one embodiment, the performing the multi-view tensor product operation according to the 3 graph models to obtain a plurality of multi-view tensor product graphs includes:
and selecting any two graph models from the 3 graph models to construct a multi-view tensor product graph, and performing multi-view tensor product operation according to the 3 graph models to obtain 9 multi-view tensor product graphs.
In one embodiment, constructing a multi-view tensor product graph from two graph models includes: and (5) carrying out Kronecker product calculation on the two graph models to obtain a multi-view tensor product graph.
In one embodiment, the fused multi-view tensor product map is obtained by performing linear fusion processing based on a similarity matrix corresponding to each multi-view tensor product map.
In one embodiment, the diffusion is performed according to the similarity matrix of the fused multi-view tensor product graph to obtain a diffused similarity matrix, wherein an efficient iteration method is adopted when the calculated similarity matrix is calculated.
In one embodiment, the plurality of superpixels are obtained by dividing the polarized SAR image based on a regular hexagon initialized fast superpixel dividing method.
An unsupervised polarized SAR image ground object classification device based on multi-view tensor product diffusion, the device comprising:
the polarized SAR image acquisition module is used for acquiring polarized SAR images to be classified, and dividing the polarized SAR images by adopting a rapid super-pixel dividing method to obtain a plurality of super-pixels;
the image model obtaining module is used for extracting 5 representative feature vectors according to the polarized SAR image, combining the 5 feature vectors to obtain 3 high-dimensional feature vectors, and constructing 3 image models based on a plurality of segmented super pixels by utilizing the 3 high-dimensional feature vectors;
The diffused similarity matrix obtaining module is used for carrying out multi-view tensor product operation according to 3 graph models to obtain a plurality of multi-view tensor product graphs, carrying out linear fusion processing according to the plurality of multi-view tensor product graphs to obtain a fused multi-view tensor product graph, and carrying out diffusion based on the similarity matrix of the fused multi-view tensor product graph to obtain a diffused similarity matrix;
and the ground object classification module is used for carrying out spectral clustering according to the diffused similarity matrix so as to realize ground object classification of the polarized SAR image.
A computer device comprising a memory storing a computer program and a processor which when executing the computer program performs the steps of:
acquiring a polarized SAR image to be classified, and segmenting the polarized SAR image by adopting a rapid superpixel segmentation method to obtain a plurality of superpixels;
extracting 5 representative feature vectors according to the polarized SAR image, combining the 5 feature vectors to obtain 3 high-dimensional feature vectors, and constructing 3 graph models based on a plurality of segmented super pixels by utilizing the 3 high-dimensional feature vectors;
performing multi-view tensor product operation according to 3 graph models to obtain a plurality of multi-view tensor product graphs, performing linear fusion processing according to the plurality of multi-view tensor product graphs to obtain a fused multi-view tensor product graph, and performing diffusion based on a similarity matrix of the fused multi-view tensor product graph to obtain a diffused similarity matrix;
And performing spectral clustering according to the diffused similarity matrix to realize the classification of the ground features of the polarized SAR image.
A computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of:
acquiring a polarized SAR image to be classified, and segmenting the polarized SAR image by adopting a rapid superpixel segmentation method to obtain a plurality of superpixels;
extracting 5 representative feature vectors according to the polarized SAR image, combining the 5 feature vectors to obtain 3 high-dimensional feature vectors, and constructing 3 graph models based on a plurality of segmented super pixels by utilizing the 3 high-dimensional feature vectors;
performing multi-view tensor product operation according to 3 graph models to obtain a plurality of multi-view tensor product graphs, performing linear fusion processing according to the plurality of multi-view tensor product graphs to obtain a fused multi-view tensor product graph, and performing diffusion based on a similarity matrix of the fused multi-view tensor product graph to obtain a diffused similarity matrix;
and performing spectral clustering according to the diffused similarity matrix to realize the classification of the ground features of the polarized SAR image.
According to the non-supervision polarized SAR image ground object classification method based on multi-view tensor product diffusion, a rapid superpixel segmentation method is adopted to segment a polarized SAR image to be classified to obtain a plurality of superpixels, 5 representative feature vectors are extracted according to the polarized SAR image, 3 high-dimensional feature vectors are obtained by combining the 5 feature vectors, 3 image models are constructed by utilizing the 3 high-dimensional feature vectors based on the segmented plurality of superpixels, multi-view tensor product operation is carried out according to the 3 image models to obtain a plurality of multi-view tensor product images, linear fusion processing is carried out according to the plurality of multi-view tensor product images to obtain a fused multi-view tensor product image, the similarity matrix after the diffusion is obtained according to the similarity matrix of the fused multi-view tensor product image, and spectral clustering is carried out on the similarity matrix after the diffusion, so that ground objects in the polarized SAR image are classified. By adopting the method, the interference of speckle noise can be reduced, and the classification precision is effectively improved.
Drawings
FIG. 1 is a flow chart of an unsupervised polarized SAR image clutter classification method based on multi-view tensor product diffusion in one embodiment;
FIG. 2 is a schematic flow chart of an algorithm based on the method in the present application in one embodiment;
FIG. 3 is a simplified schematic diagram of a multi-view tensor product graph in one embodiment;
FIG. 4 is a schematic representation of a Flevoland measured data image in one embodiment;
FIG. 5 is a diagram of an image of the obenpfaffenhofen measured data in one embodiment;
FIG. 6 is a schematic diagram of classification results of 5 methods based on Flevoland measured polarized SAR images in one embodiment;
FIG. 7 is a schematic diagram of evaluation results of 5 methods based on Flevoland measured polarized SAR images in one embodiment;
FIG. 8 is a schematic diagram of classification results of 6 methods based on Obenpfaffenhofen measured data in one embodiment;
FIG. 9 is a block diagram of an unsupervised polarized SAR image clutter classification device based on multi-view tensor product diffusion in one embodiment;
fig. 10 is an internal structural view of a computer device in one embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the present application.
1-2, an unsupervised polarized SAR image ground object classification method based on multi-view tensor product diffusion is provided, which comprises the following steps:
step S100, obtaining a polarized SAR image to be classified, and dividing the polarized SAR image by adopting a rapid super-pixel dividing method to obtain a plurality of super-pixels;
step S110, extracting 5 representative feature vectors according to the polarized SAR image, combining the 5 feature vectors to obtain 3 high-dimensional feature vectors, and constructing 3 graph models based on a plurality of segmented super pixels by utilizing the 3 high-dimensional feature vectors;
step S120, performing multi-view tensor product operation according to 3 graph models to obtain a plurality of multi-view tensor product graphs, performing linear fusion processing according to the plurality of multi-view tensor product graphs to obtain a fused multi-view tensor product graph, and performing diffusion according to a similarity matrix of the fused multi-view tensor product graph to obtain a diffused similarity matrix;
and step S130, performing spectral clustering according to the diffused similarity matrix to realize feature classification of the polarized SAR image.
In step S100, a series of pixel points with similar positions and similar low-level features in the polarized SAR image are locally clustered to form superpixels by using a superpixel segmentation method, and the polarized SAR image after superpixel segmentation can obtain a plurality of superpixels. After the polarized SAR image is subjected to super-pixel segmentation, the influence of inherent speckle noise in the polarized SAR image can be overcome by utilizing local information, so that the operation efficiency of subsequent processing is improved.
In this embodiment, the multiple superpixels are obtained by dividing the polarized SAR image based on the fast superpixel dividing method of regular hexagon initialization, and the dividing process sequentially includes regular hexagon initialization, local k-means clustering, updating, and post-processing, and finally the superpixels are obtained.
In step S110, 3 high-dimensional feature vectors are obtained by combining the 5 feature vectors, and 3 graph models are constructed based on the segmented multiple super pixels by using the 3 high-dimensional feature vectors, which specifically includes: and 5 feature vectors are extracted from each pixel point in the polarized SAR image, 3 high-dimensional feature vectors corresponding to each pixel point are obtained by combining the extracted 5 feature vectors corresponding to each pixel point, the average high-dimensional feature vector of each pixel point in the same super pixel on the 3 high-dimensional feature vectors in a plurality of super pixels is calculated and used as the feature vector of the super pixel, so that each super pixel corresponds to 3 feature vectors, and finally, the corresponding 3 graph models are respectively constructed according to different 3 feature vectors.
Furthermore, the polarized SAR image comprises a plurality of different ground objects, and the sensitivity of each ground object to different polarized scattering characteristics is also different. Therefore, a certain number of features with strong discrimination capability for different ground object categories need to be selected, and through reference to related documents and experimental verification, 5 feature vectors, namely a Yamaguchi4 feature vector, a Krogager feature vector, an HSI color space feature vector, a Cloude-potter's feature vector and a feature vector composed of scattering power entropy and co-polarization rate, are selected in the embodiment, as shown in fig. 2.
The multi-view tensor product diffusion can effectively combine the classifying ability of the data of a plurality of different view angles, generally, better experimental results can be obtained when the number of view angles is a plurality of, and when each single view angle has stronger classifying ability, the multi-view tensor product graph can fully utilize the classifying ability of the data of the different view angles to strengthen the effective diffusion of the similarity information. Therefore, in order to secure the classification capability for each view angle, the following combination method is also adopted in the present embodiment to combine five kinds of feature vectors into 3 kinds of high-dimensional feature vectors.
The combination mode comprises the following steps: forming one high-dimensional characteristic vector by using a Yamaguchi4 characteristic vector, an HSI color space characteristic vector, a Cloude-Pottier's characteristic vector and a characteristic vector consisting of scattering power entropy and same polarization rate; one of the high-dimensional feature vectors is composed of a Yamaguchi4 feature vector, a Krogager feature vector, a Cloude-Pottier's feature vector and a feature vector composed of scattering power entropy and same polarization rate; one of the high-dimensional feature vectors is composed of Yamaguchi4 feature vector, krogager feature vector, HSI color space feature vector, cloude-potter's feature vector, and feature vector composed of scattering power entropy and co-polarization ratio, as shown in fig. 2.
Specifically, in order to better perform feature representation on each superpixel in the plurality of superpixels, each pixel point in the original polarized image is represented by adopting the five feature vectors, and then the five feature vectors are combined according to the combination mode to obtain three high-dimensional feature vectors, that is, each pixel point in the polarized SAR image can be represented by the three high-dimensional feature vectors. And calculating the average feature vector of each pixel point after processing the same super pixel under each high-dimensional feature vector, and taking the average feature vector obtained by corresponding to each high-dimensional feature vector as the feature vector of the corresponding super pixel, namely, at the moment, each super pixel can be characterized by three different feature vectors.
Based on super-pixel segmentation, polarized SAR images can adopt weighted undirected graph modelsG= (V, E). Wherein node v= { V 1 ,v 2 ,...,v M And represents a superpixel set, edges (v i ,v j ) E-E connection node v i And v j . While the nodes represent superpixels, the edges between the nodes represent the similarity between the two superpixels, and each edge has a corresponding non-negative weight w ij (i, j=1, 2,., M) representing a neighboring node v i And v j M represents the number of super-pixels obtained after over-segmentation of the polarized SAR image. Thus, in this embodiment, corresponding 3 graph models can be constructed according to 3 different feature vectors.
Based on each model, the original similarity matrix W can be constructed by adopting Gaussian kernel, and matrix elements are corresponding non-negative weights W of each edge ij Matrix element w ij Is calculated as follows:
in formula (1), d ij Representing superpixel v i And super pixel v j The Euclidean distance between them, mu is a super-parameter, epsilon ij The scale parameters used to eliminate scaling problems are defined as:
in formula (2), d i, N i Representing superpixel v i Nearest-neighbor (k-Nearest Neighbors, k-NN) super-pixel N thereto i Euclidean distance between, mean i, N i Mean Euclidean distance, mean j, N j Definition and mean i, N i The same applies.
Since each superpixel in the image can be represented by N different types of average eigenvectors, N different graph models G can be constructed (n) ＝(V (n) ,E (n) ) Wherein N is more than or equal to 1 and N is more than or equal to N.
In the present invention3 feature vectors are constructed, and other feature vectors with different numbers can be extracted and combined according to different requirements of practical application conditions. Thus, each polarized SAR image can be represented by 3 graph models G (n) ＝(V (n) ,E (n) ) (1. Ltoreq.n. Ltoreq.N) and where the graph model is a single view model, G (n) ＝(V (n) ,E (n) ) Representing a single view model constructed with the nth feature vector, N is the number of view models constructed for different features and combinations (n=3 in this embodiment), and W n The similarity matrix corresponding to each graph model is obtained by constructing edges in the graph models.
In step S120, multi-view tensor product diffusion is performed based on the graph model, and finally a similarity matrix after cross diffusion is obtained. In conventional unsupervised polarized SAR image classification systems, the similarity matrix is typically determined by pairwise similarity between data points (pixels or superpixels), ignoring global feature information of the polarized SAR image. Therefore, a method of spreading the similarity measure around on the graph structure is proposed, and learning of the similarity is realized by considering the relationship between each data point and the points in its neighborhood. In general, when a high-dimensional feature vector is constructed in a polarized SAR image classification method, the extracted feature vector is directly stacked head and tail, and although the method is simple and feasible and fuses various feature information, the method can weaken the performance of the feature vector with strong partial feature discrimination capability and introduce classification errors. The multi-view tensor product can fully utilize the complementary information among different polarized scattering characteristics, and further generate a similarity matrix with stronger classification capability after cross diffusion based on the internal relation among data points (super pixels).
In this embodiment, similarity information is propagated and diffused on the multi-view tensor product graph, and compared with the original graph, the multi-view tensor product graph considers higher-order upper and lower invention relationships, introduces the thought of multi-view learning, and can better reveal the internal manifold structure of data so as to construct a similarity matrix with stronger discrimination capability and classification capability, thereby improving the classification precision of ground features.
Before performing multi-view tensor product diffusion explanation based on the graph model, a diffusion process based on the original graph is introduced, that is, diffusion is directly performed based on the original graph model, so that subsequent understanding is facilitated.
The graph-theory based diffusion process can reveal the inherent geometric relationships between data points. Graph-based diffusion is the simplest understanding of the product W of graph similarity matrices t (t is the number of iterations). From this, it is known that the diffusion of the graph is based on a similarity matrix. However, this classical diffusion process is constrained by the number of iterations t. If the row sums of the similarity matrix W are all smaller than 1, where "row sums" represent the sum of elements of each row in the matrix, the classical diffusion process based on W eventually converges to a 0 matrix, and thus the setting of the iteration number is crucial. In order to reduce the influence of the iteration number t on the diffusion process, a diffusion form in which the similarity matrix W is weighted may be adopted, as shown in the following formula:
In equation (3), in order to avoid convergence of the diffusion process to a 0 matrix, it is necessary to ensure that W is a non-negative matrix and that(M is the number of super pixels), i.e. the row sums are each less than 1; for any i, if there is +.>Equation (3) cannot converge. According to the matrix principle, a matrix w satisfying the above conditions ij May be generated by a random matrix transformation. After the condition is satisfied, the formula (3) may converge to:
in formula (4), I 1 Is an identity matrix of the same dimension as W.
In this embodiment, the three constructed graph models are converted into a plurality of multi-view tensor product graphs by multi-view tensor product operation, and then the multi-view tensor product graphs are expanded. The process of obtaining the multiple multi-view tensor product graphs comprises the following steps: and selecting any two graph models from the 3 graph models to construct a multi-view tensor product graph, and performing multi-view tensor product operation according to the 3 graph models to obtain 9 multi-view tensor product graphs. And then carrying out linear fusion processing on similarity matrixes corresponding to the 9 multi-view tensor product graphs to obtain a fused multi-view tensor product graph, and finally obtaining the similarity matrix after cross diffusion based on the fused multi-view tensor product graph.
Specifically, a multi-view tensor product graph (i.e., multi-view tensor product graph) is employedIs defined as:
Multi-view tensor product graphA single view model G constructed from different feature vectors k And G l Is composed of one node of->The similarity matrix is defined as +.>Wherein (1)>Represents the Kronecker product. In particular, for α, β, i, j=1, 2,..m, there is:
thus, ifThen->A simplified example of which is shown in figure 3.
Compared with other multi-view learning methods based on graph theory, in the method, kronecker product calculation is performed on the N feature vector-based graph models two by two, so that n×n multi-view tensor product graphs are obtained instead of N, and in this embodiment, 9 multi-view tensor product graphs are obtained finally.Compared with the original image, each node in the model contains richer information, a single-view model constructed based on different feature vectors captures higher-order similarity information, and the inherent relation between data points is effectively revealed.
After a plurality of multi-view tensor product graphs are obtained, a fused multi-view tensor product graph is obtained based on linear fusion processing, and expansion calculation is carried out based on a similar feature matrix of the multi-view tensor product graph.
Among a plurality of graph model fusion methods applied to multi-view clustering based on graph theory, linear addition of single graph models is the simplest and feasible processing mode. Therefore, after obtaining n×n multi-view tensor product graphs, in this embodiment, a linear fusion process is performed on the similarity matrix corresponding to each multi-view tensor product graph to implement a diffusion process based on the multi-view tensor product graph, so as to obtain a fused multi-view tensor product graph with the following formula:
According to the foregoing diffusion process, the diffusion process based on the multi-view tensor product graph is similar to the formula (3), as shown in the following formula:
similar to the convergence shown in equation (4), the diffusion process shown in equation (8) may also converge according to the following equation:
in formula (9), I 2 Is in combination withIdentity matrix of the same dimension. However, as described above, in order to avoid convergence of the diffusion process to a matrix 0, it is necessary to ensure +.>The sum of rows of (a) is less than 1.
In the diffusion process based on the similarity matrix expressed by the formula (9), the tensor product graph is generated due to multiple viewsCorresponding similarity matrix->The dimension is very high, the calculation efficiency of the diffusion and convergence process is very low, and the similarity matrix obtained after diffusion is +.>With original G (n) Corresponding similarity matrix W n Nor are the dimensions of (a). Therefore, in the present embodiment, it is desirable to be able to perform in-situ G (n) Learning similarity information by using multi-view tensor product, namely a diffused similarity matrix W * The dimension is M x M, and meanwhile, the similarity information of multiple views is combined:
in equation (10), vec is an operator that functions to transform each column of a matrix into a column vector by stacking the matrix -1 To convert a column vector into an inverse of a matrix.
The similarity matrix after cross diffusion can be obtained after the calculation according to the formula (10), but since the multi-view tensor product graph has higher order than the original graph, the calculation of the formula (10) consumes more calculation amount and storage space when processing some larger graphs. Therefore, in order to improve the diffusion efficiency of the multi-view tensor product graph, a new iteration method based on the original graph, namely a c-TPGD efficient iteration method is adopted, which is equivalent to the diffusion process based on the multi-view tensor product graph, but saves a large amount of calculation time and storage space, and the calculation process is as follows:
first, define:
and (3) repeating the iteration of the formula (11) until the Q converges, so that a similarity matrix after cross diffusion can be obtained:
for ease of understanding, the present invention will be briefly described below with respect to the proof thinking of equation (12). First, Q is (t +1) The unfolding steps are as follows:
due to(M is the number of super pixels), i.e. W (k) The sum of the rows of (2) is less than 1, and thus hasTherefore:
the following identities are known:
thus, it is possible to obtain:
wherein,the following was demonstrated:
from the combination of the formulas (13) to (17), it is possible to obtain:
thus, it is finally possible to:
therefore, this new iteration method based on the original graph is equivalent to the diffusion process based on the multi-view tensor product graph, but can save a lot of computation time and storage space. The new cross-diffused similarity matrix obtained contains higher-order internal information between data points, and generally, when the iteration number is 20, the classification capability of the diffused similarity matrix can reach a relatively stable level, so that the iteration number t=20 is set in the experiment of the method.
In step S130, based on the similarity matrix obtained after the cross diffusion through the above calculation, the result of feature classification in the polarized SAR image can be obtained by performing spectral clustering.
In particular, the clustering method is receiving more and more attention because it can obtain better clustering results in (ground object) feature space of any shape and has a more perfect mathematical framework. The spectral clustering method is based on feature decomposition of a similarity matrix and further clustering is performed by using a k-means method, and in the embodiment, a new similarity matrix generated by tensor product diffusion is used as input of the spectral clustering method for feature decomposition, so that a final classification result is obtained.
To evaluate the performance of the method, multiple sets of experiments were developed based on multiple data and analyzed in detail. The experimental part is described in detail as follows:
in the experiment, the first actually measured polarized SAR image is a partial region of the Flevoand test area shot by AirSAR, the area is 300×270 pixels, the truth diagram is shown in fig. 4 (a), and the corresponding Pauli-RGB image and the generated super-pixel diagram are shown in fig. 4 (b) and (c), respectively. The second actually measured polarized SAR image is an L-band image shot by ESAR, the shooting area is located in an obenpfaffenhofen test area, the image size is 700×1000 pixels, a truth diagram is shown in fig. 5 (a), and a Pauli-RGB image and a generated super-pixel diagram are shown in fig. 5 (b) and (c) respectively. The actually measured polarized SAR image mainly comprises 3 types of ground objects: woodland, open area 1 and open area 2.
The structure of this experimental part is as follows. To verify the effectiveness of the classification method in this application, 5 methods of comparison experiments were performed based on the fleveland measured polarized SAR images. The 5 methods involved in the comparison included: an Unsupervised k-means with pixels (UKWC-P) classification method based on the Unsupervised k-means Wishart Classification algorithm based on Pixels, an Unsupervised k-means with pixels (UKWC-S) classification method based on the Unsupervised k-means Wishart Classification algorithm based on Superpixels, UKWC-S), a pixel-level Unsupervised classification method based on the scattered power entropy and the same polarization rate (Unsupervised Classification based on Scattering power entropy and Copolarized ratio based on Pixels, UCSC-P), an Unsupervised classification method based on the scattered power entropy and the same polarization rate (Unsupervised Classification based on Scattering power entropy and Copolarized ratio based on Superpixels, UCSC-S) and a method (c-TPGD) of the invention.
Then, 6 methods of comparative analysis experiments are performed based on the Obenpfaffenhofen actually measured polarized SAR image, including: an unsupervised classification method (TPGD method) based on single vision tensor product diffusion, a UCSC-S method, a UKWC-S method, a superpixel level Wishart classification method (GDWC-S method) based on geodesic distance, a superpixel level Wishart classification method (CPWC-S method) based on polarization decomposition, and the method (c-TPGD) of the invention.
In order to ensure fairness of comparison experiments, the number of categories in all experiments is preset manually according to priori knowledge, and experimental parameters of the comparison method are set according to optimal parameters of the corresponding invention. 4 evaluation metrics were taken in the experiment to evaluate effectiveness: mixed User precision (UA, representing the proportion of pixels whose actual type is also the i-th class among all pixels classified as the i-th class), drawing precision (Producer Accuracy, PA, representing the proportion of pixels whose actual type is also the i-th class among all pixels whose actual type is the i-th class), overall precision (OA, representing the proportion of pixels whose actual type is correctly classified among all samples), and Kappa coefficient (Kappa coefficient, K, the UA and PA are integrated to evaluate the precision of classifying images).
Contrast experiments based on fleveland measured polarized SAR data:
in order to further evaluate the performance of the c-TPGD method, a comparison experiment was performed on the actually measured polarized SAR image by using a total of 5 unsupervised classification methods of UKWC-P, UKWC-S, UCSC-P, UCSC-S and the method of the present invention, the experimental results of which are shown in fig. 6, wherein fig. 6 (a) is an experimental result of UKWC-P, fig. 6 (b) is an experimental result of UKWC-S, fig. 6 (c) is an experimental result of UCSC-P, fig. 6 (d) is an experimental result of UCSC-S, and fig. 6 (e) is an experimental result of the method. Because these 5 methods are all unsupervised, the labels of the classification results are random, and the final result labels are re-labeled according to a truth chart for convenient observation and comparison, but this does not affect the result of the unsupervised classification, and some classification results may be mistakenly classified into one class, and the principle of re-labeling is to maximize the OA value.
As can be seen from fig. 6 (a) and (c), the experimental results of the two methods of UKWC-P and UCSC-P are greatly affected by the speckle noise, and in particular, the UCSC-P method has a better classifying ability for each category based on the visual angle, but has a Kappa coefficient of only 0.3195 due to the interference of the speckle noise. This indicates that the pixel-based classification method does not resist well against the interference of the speckle noise inherent to the polarized SAR image. Fig. 6 (b) and (d) are two methods based on the classification result of the super pixel respectively, and it can be obviously seen that compared with the classification result based on the pixel, the classification result can better overcome the influence caused by the speckle noise, and a lot of 'speckle' phenomena are reduced, which indicates that the classification method based on the super pixel can effectively improve the classification precision while overcoming the interference of the speckle noise.
Some superpixels in fig. 6 (b) and (d) are misclassified, probably because both the UKWC-S and UCSC-S methods employ insufficient features to describe all of the features. It is obvious that the experimental result in fig. 6 (e) is better than other experimental results, because the method selects and combines 3 effective feature vectors, and diffuses similarity information based on the multi-view tensor product graph, effectively combines the classification capability among polarization scattering features, and makes the similarity matrix after cross diffusion have stronger discrimination capability on ground objects.
To evaluate these 6 methods further quantitatively, the evaluation was performed using UA, PA, OA and K4 evaluation metrics, the corresponding results are shown in fig. 7, where fig. 7 (a) is the user precision of the 5 methods, fig. 7 (b) is the drawing precision of the 5 methods, fig. 7 (c) is the overall precision of the 5 methods, and fig. 7 (d) is the Kappa coefficient of the 5 methods. From the data in the graph, it can be seen that the c-TPGD method is generally higher than other methods for the drawing precision and user precision of 6 ground object categories. The overall accuracy and Kappa coefficient of the classification result of the 5 methods are shown in fig. 7 (c) and (d), respectively, and it can be seen that the overall accuracy and Kappa coefficient of the classification result of the method of the present invention are the highest. Meanwhile, the instability of the UCSC method can be seen, and the classification requirement on complex actually-measured polarized SAR data can not be well met because the selected characteristics are single. The method can show stronger ground object discrimination capability on complex measured data, further shows that the multi-view tensor product graph can better reveal the internal relation of the data, and effectively fuses similarity information of multiple views, thereby improving classification accuracy of the ground objects.
Comparison experiment based on Oberpfaffenhofen actual measurement polarization SAR data:
The results of comparative experiments of the 6 methods based on the obspfaffenhofen measured data are shown in fig. 8, in which fig. 8 (a) is the S1-TPGD method, fig. 8 (b) is the S2-TPGD method, fig. 8 (c) is the S3-TPGD method, fig. 8 (d) is the UCSC-S method, fig. 8 (e) is the UKWC-S method, fig. 8 (f) is the GDWC-S method, fig. 8 (g) is the CPWC-S method, fig. 8 (h) is the c-TPGD method, and PA values, kappa coefficients and OA values of the 6 methods are shown in table 1. Since the TPGD method is tensor product diffusion of single view feature vectors, the 3 original similarity matrices S1, S2 and S3 described in fig. 2 are used as inputs of the single view tensor product diffusion to obtain final classification results, so they are called S1-TPGD, S2-TPGD and S3-TPGD methods, respectively.
Table 1 6 3 evaluation metrics results of the methods based on Oberpfaffenhofen measured data
As shown in the region B in fig. 8 (d), the UCSC-S algorithm misclassifies a large number of open areas 2 into open areas 1, while the partial feature edges and homogeneous regions in fig. 8 (e) exhibit incomplete cracking, which indicates that the ability to discriminate features may be affected by the lack of extraction of more polarized scattering features. In fig. 8 (g), there are many isolated small regions, which illustrates that the classification result of the CPWC-S algorithm is severely affected by the inherent speckle noise in the polarized SAR image, as shown in region a, and that the algorithm cannot classify open area 2 in region B. The OA value of the GDWC-S algorithm can reach 80.74% higher than the above 4 algorithms, again proving the importance of finding the shortest distance between data points (superpixels). However, looking at region a in fig. 8 (c), there is still more noise, mainly also because it does not classify in combination with a variety of valid polarized SAR features.
In fig. 8, (a), (b), (c) and (e) all adopt methods based on tensor product diffusion, but the TPGD algorithm only considers single view characteristic information, so that the classification effect is not ideal. The algorithm combines various typical polarized SAR features, diffuses similarity information based on a multi-view tensor product graph, effectively combines the classification capability among polarized scattering features, and enables the similarity matrix after cross diffusion to have stronger discrimination capability on ground objects.
In the non-supervision polarized SAR image ground object classification method based on multi-view tensor product diffusion, the non-supervision classification method is adopted, a large amount of manpower and material resources are not required to be consumed to label a complete data set, the automation degree is high, and the actual requirements are met. The method also adopts a multi-vision tensor product diffusion technology, can effectively mine the intrinsic similarity information of the data, reduces the interference of speckle noise, and improves the classification precision. The method provides an unsupervised polarized SAR image ground object classification framework based on multi-vision tensor product diffusion, and the framework has universality and can achieve the purpose of ground object classification by combining with other similarity measurement methods. I.e. the method is mobile.
It should be understood that, although the steps in the flowchart of fig. 1 are shown in sequence as indicated by the arrows, the steps are not necessarily performed in sequence as indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in fig. 1 may include multiple sub-steps or stages that are not necessarily performed at the same time, but may be performed at different times, nor do the order in which the sub-steps or stages are performed necessarily performed in sequence, but may be performed alternately or alternately with at least a portion of other steps or sub-steps of other steps.
In one embodiment, as shown in fig. 9, there is provided an unsupervised polarized SAR image ground object classification device based on multi-view tensor product diffusion, including: a polarized SAR image acquisition module 200, a graph model obtaining module 210, a similarity matrix after diffusion obtaining module 220, and a ground object classification module 230, wherein:
the polarized SAR image acquisition module 200 is used for acquiring a polarized SAR image to be classified, and dividing the polarized SAR image by adopting a rapid super-pixel dividing method to obtain a plurality of super-pixels;
The map model obtaining module 210 is configured to extract 5 representative feature vectors according to the polarized SAR image, combine the 5 feature vectors to obtain 3 high-dimensional feature vectors, and construct 3 map models based on the segmented multiple superpixels by using the 3 high-dimensional feature vectors;
the diffused similarity matrix obtaining module 220 is configured to perform multi-view tensor product operation according to 3 graph models to obtain a plurality of multi-view tensor product graphs, perform linear fusion processing according to the plurality of multi-view tensor product graphs to obtain a fused multi-view tensor product graph, and perform diffusion based on the similarity matrix of the fused multi-view tensor product graph to obtain a diffused similarity matrix;
the feature classification module 230 is configured to perform spectral clustering according to the diffused similarity matrix, so as to implement feature classification of the polarized SAR image.
For specific limitation of the non-supervision polarized SAR image feature classification device based on multi-view tensor product diffusion, reference may be made to the limitation of the non-supervision polarized SAR image feature classification method based on multi-view tensor product diffusion hereinabove, and the description thereof is omitted here. The modules in the non-supervision polarization SAR image ground object classification device based on multi-view tensor product diffusion can be fully or partially realized by software, hardware and a combination thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.
In one embodiment, a computer device is provided, which may be a terminal, and an internal structure diagram thereof may be as shown in fig. 10. The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program when executed by the processor is used for realizing an unsupervised polarization SAR image ground object classification method based on multi-vision tensor product diffusion. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, can also be keys, a track ball or a touch pad arranged on the shell of the computer equipment, and can also be an external keyboard, a touch pad or a mouse and the like.
It will be appreciated by those skilled in the art that the structure shown in fig. 10 is merely a block diagram of some of the structures associated with the present application and is not limiting of the computer device to which the present application may be applied, and that a particular computer device may include more or fewer components than shown, or may combine certain components, or have a different arrangement of components.
In one embodiment, a computer device is provided comprising a memory and a processor, the memory having stored therein a computer program, the processor when executing the computer program performing the steps of:
acquiring a polarized SAR image to be classified, and segmenting the polarized SAR image by adopting a rapid superpixel segmentation method to obtain a plurality of superpixels;
extracting 5 representative feature vectors according to the polarized SAR image, combining the 5 feature vectors to obtain 3 high-dimensional feature vectors, and constructing 3 graph models based on a plurality of segmented super pixels by utilizing the 3 high-dimensional feature vectors;
performing multi-view tensor product operation according to 3 graph models to obtain a plurality of multi-view tensor product graphs, performing linear fusion processing according to the plurality of multi-view tensor product graphs to obtain a fused multi-view tensor product graph, and performing diffusion based on a similarity matrix of the fused multi-view tensor product graph to obtain a diffused similarity matrix;
And performing spectral clustering according to the diffused similarity matrix to realize the classification of the ground features of the polarized SAR image.
In one embodiment, a computer readable storage medium is provided having a computer program stored thereon, which when executed by a processor, performs the steps of:
acquiring a polarized SAR image to be classified, and segmenting the polarized SAR image by adopting a rapid superpixel segmentation method to obtain a plurality of superpixels;
extracting 5 representative feature vectors according to the polarized SAR image, combining the 5 feature vectors to obtain 3 high-dimensional feature vectors, and constructing 3 graph models based on a plurality of segmented super pixels by utilizing the 3 high-dimensional feature vectors;
performing multi-view tensor product operation according to 3 graph models to obtain a plurality of multi-view tensor product graphs, performing linear fusion processing according to the plurality of multi-view tensor product graphs to obtain a fused multi-view tensor product graph, and performing diffusion based on a similarity matrix of the fused multi-view tensor product graph to obtain a diffused similarity matrix;
and performing spectral clustering according to the diffused similarity matrix to realize the classification of the ground features of the polarized SAR image.
Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the various embodiments provided herein may include non-volatile and/or volatile memory. The nonvolatile memory can include Read Only Memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), memory bus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), among others.
The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The above examples merely represent a few embodiments of the present application, which are described in more detail and are not to be construed as limiting the scope of the invention. It should be noted that it would be apparent to those skilled in the art that various modifications and improvements could be made without departing from the spirit of the present application, which would be within the scope of the present application. Accordingly, the scope of protection of the present application is to be determined by the claims appended hereto.
Claims (8)
1. The unsupervised polarization SAR image ground object classification method based on multi-vision tensor product diffusion is characterized by comprising the following steps:
acquiring a polarized SAR image to be classified, and segmenting the polarized SAR image by adopting a rapid superpixel segmentation method to obtain a plurality of superpixels;
extracting 5 representative feature vectors according to the polarized SAR image, combining the 5 feature vectors to obtain 3 high-dimensional feature vectors, and constructing 3 graph models based on a plurality of segmented super pixels by using the 3 high-dimensional feature vectors, wherein the 5 representative feature vectors comprise: yamaguchi4 eigenvector, krogager eigenvector, HSI color space eigenvector, cloud-potter's eigenvector, and eigenvector composed of scattering power entropy and co-polarization, combining according to these 5 eigenvectors to obtain 3 high-dimensional eigenvectors comprising: one of the high-dimensional feature vectors is composed of the Yamaguchi4 feature vector, the HSI color space feature vector, the Cloude-Pottier's feature vector and the feature vector composed of scattering power entropy and the same polarization rate, one of the high-dimensional feature vectors is composed of the Yamaguchi4 feature vector, the Krogager feature vector, the Cloude-Pottier's feature vector and the feature vector composed of scattering power entropy and the same polarization rate, and one of the high-dimensional feature vectors is composed of the Yamaguchi4 feature vector, the Krogager feature vector, the HSI color space feature vector, the Cloude-Pottier's feature vector and the feature vector composed of scattering power entropy and the same polarization rate;
Performing multi-view tensor product operation according to 3 graph models to obtain a plurality of multi-view tensor product graphs, performing linear fusion processing according to the plurality of multi-view tensor product graphs to obtain a fused multi-view tensor product graph, and performing diffusion based on a similarity matrix of the fused multi-view tensor product graph to obtain a diffused similarity matrix;
and performing spectral clustering according to the diffused similarity matrix to realize the classification of the ground features of the polarized SAR image.
2. The method for classifying features of an unsupervised polarized SAR image according to claim 1, wherein said combining according to the 5 feature vectors to obtain 3 high-dimensional feature vectors, and constructing 3 graph models based on the segmented super pixels by using the 3 high-dimensional feature vectors specifically comprises:
extracting 5 feature vectors from each pixel point in the polarized SAR image, and combining the extracted 5 feature vectors according to the corresponding pixel points to obtain 3 high-dimensional feature vectors corresponding to the pixel points;
calculating an average eigenvector of 3 high-dimensional eigenvectors of each pixel point in the same super pixel in the plurality of super pixels as the eigenvector of the super pixel, so that each super pixel corresponds to 3 eigenvectors;
And respectively constructing a graph model according to the different 3 feature vectors to obtain 3 corresponding graph models.
3. The non-supervised polarized SAR image ground object classification method of any one of claims 1-2, wherein said graph model is comprised of a plurality of nodes and edges between two adjacent nodes; wherein each of the nodes represents a different superpixel in the superpixel segmented image and the edges represent a similarity of superpixels at both ends of the edge.
4. The method for classifying features of an unsupervised polarized SAR image according to claim 3, wherein said performing a multi-view tensor product operation according to 3 of said graph models to obtain a plurality of multi-view tensor product graphs comprises:
and selecting any two graph models from the 3 graph models to construct a multi-view tensor product graph, and performing multi-view tensor product operation according to the 3 graph models to obtain 9 multi-view tensor product graphs.
5. The method of non-supervised polarimetric SAR image feature classification as set forth in claim 4, wherein constructing a multi-view tensor product graph from two graph models comprises: and (5) carrying out Kronecker product calculation on the two graph models to obtain a multi-view tensor product graph.
6. The method for classifying features of an unsupervised polarized SAR image according to claim 5, wherein the fused multi-view tensor product map is obtained by performing linear fusion processing based on a similarity matrix corresponding to each multi-view tensor product map.
7. The method for classifying features of an unsupervised polarized SAR image according to claim 6, wherein the diffused similarity matrix is obtained based on the similarity matrix of the fused multi-view tensor product graph, and wherein an efficient iterative method is used in calculating the diffused similarity matrix.
8. The non-supervised polarized SAR image ground object classification method according to claim 1, wherein said plurality of superpixels are obtained by dividing said polarized SAR image by a fast superpixel division method based on regular hexagonal initialization.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111293497.7A CN114004998B (en) | 2021-11-03 | 2021-11-03 | Unsupervised polarization SAR image ground object classification method based on multi-vision tensor product diffusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111293497.7A CN114004998B (en) | 2021-11-03 | 2021-11-03 | Unsupervised polarization SAR image ground object classification method based on multi-vision tensor product diffusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114004998A CN114004998A (en) | 2022-02-01 |
| CN114004998B true CN114004998B (en) | 2024-01-19 |
Family
ID=79927196
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111293497.7A Active CN114004998B (en) | 2021-11-03 | 2021-11-03 | Unsupervised polarization SAR image ground object classification method based on multi-vision tensor product diffusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114004998B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114898193A (en) * | 2022-07-11 | 2022-08-12 | 之江实验室 | Manifold learning-based image feature fusion method and device and image classification system |
| CN116400356B (en) * | 2023-06-07 | 2023-08-18 | 中山大学 | Chromatographic SAR three-dimensional imaging method based on homogeneous region combination |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2767849A2 (en) * | 2014-01-13 | 2014-08-20 | Institute of Electronics, Chinese Academy of Sciences | Method and apparatus for processing polarimetric synthetic aperture radar image |
| CN111062445A (en) * | 2019-12-23 | 2020-04-24 | 中国科学院自动化研究所 | Polarimetric SAR image classification method based on collaborative regularization and superpixel |
-
2021
- 2021-11-03 CN CN202111293497.7A patent/CN114004998B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2767849A2 (en) * | 2014-01-13 | 2014-08-20 | Institute of Electronics, Chinese Academy of Sciences | Method and apparatus for processing polarimetric synthetic aperture radar image |
| CN111062445A (en) * | 2019-12-23 | 2020-04-24 | 中国科学院自动化研究所 | Polarimetric SAR image classification method based on collaborative regularization and superpixel |
Non-Patent Citations (3)
| Title |
|---|
| Huanxin Zou ; Meilin Li.Superpixel-Oriented Unsupervised Classification for Polarimetric SAR Images Based on Consensus Similarity Network Fusion.IEEE Access.2019,第7卷全文. * |
| Unsupervised classification of PolSAR image based on tensor product graph diffusion;李美霖 邹焕新;Fourth International Workshop on Pattern Recognition * |
| 一种基于测地线距离的极化SAR图像快速超像素分割算法;邹焕新 李美霖;Journal of Radars;第10卷(第1期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114004998A (en) | 2022-02-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111860612B (en) | Unsupervised hyperspectral image hidden low-rank projection learning feature extraction method | |
| Zhang et al. | Simultaneous spectral-spatial feature selection and extraction for hyperspectral images | |
| Poggi et al. | Supervised segmentation of remote sensing images based on a tree-structured MRF model | |
| Bencherif et al. | Fusion of extreme learning machine and graph-based optimization methods for active classification of remote sensing images | |
| Marpu et al. | Classification of hyperspectral data using extended attribute profiles based on supervised and unsupervised feature extraction techniques | |
| CN114004998B (en) | Unsupervised polarization SAR image ground object classification method based on multi-vision tensor product diffusion | |
| Ducournau et al. | A reductive approach to hypergraph clustering: An application to image segmentation | |
| CN107491734B (en) | Semi-supervised polarimetric SAR image classification method based on multi-core fusion and space Wishart LapSVM | |
| Liu et al. | An efficient matrix factorization based low-rank representation for subspace clustering | |
| Gorricha et al. | Improvements on the visualization of clusters in geo-referenced data using Self-Organizing Maps | |
| CN103258324B (en) | Based on the method for detecting change of remote sensing image that controlled kernel regression and super-pixel are split | |
| Zhang et al. | Road recognition from remote sensing imagery using incremental learning | |
| Marinoni et al. | Unsupervised data driven feature extraction by means of mutual information maximization | |
| Du et al. | Multiple graph unsupervised feature selection | |
| Kim et al. | Deep learning-based monitoring of overshooting cloud tops from geostationary satellite data | |
| Wang et al. | Dynamic graph-level neural network for SAR image change detection | |
| Zhen et al. | Kernel truncated regression representation for robust subspace clustering | |
| CN114119621A (en) | SAR remote sensing image water area segmentation method based on depth coding and decoding fusion network | |
| Li et al. | Variational Bayesian independent component analysis-support vector machine for remote sensing classification | |
| CN105005987A (en) | SAR image superpixel generating method based on general gamma distribution | |
| Li et al. | Enhanced bird detection from low-resolution aerial image using deep neural networks | |
| CN104050489B (en) | SAR ATR method based on multicore optimization | |
| Pan et al. | GRU with spatial prior for hyperspectral image classification | |
| Alam et al. | Combining unmixing and deep feature learning for hyperspectral image classification | |
| CN115578599A (en) | Polarized SAR image classification method based on superpixel-hypergraph feature enhancement network |
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 |