CN111062445A - Polarimetric SAR Image Classification Method Based on Co-regularization and Superpixels - Google Patents

Abstract

The invention belongs to the technical field of image processing, and in particular relates to a polarization SAR image classification method, system and device based on collaborative regularization and superpixels, and aims to solve the problem of low classification accuracy of the existing polarization SAR image classification methods. The system method includes obtaining a plurality of superpixels and their corresponding coherence matrices through a superpixel generation method based on the acquired polarimetric SAR images; extracting the polarization characteristics of each superpixel preset dimension; Calculate the Wishart distance and Euclidean distance between the features and other superpixels, and construct the first and second weight maps of each superpixel; obtain the first low-dimensional feature and the second low-dimensional feature through the dimensionality reduction model based on collaborative regularization ; Obtain the classification results of polarimetric SAR images through the nearest neighbor classifier. According to the spatial information of pixel points, the invention improves the classification accuracy of polarimetric SAR images by combining Wishart distance and polarimetric features.

Description

Polarimetric SAR Image Classification Method Based on Co-regularization and Superpixels

technical field

The invention belongs to the technical field of image processing, and in particular relates to a polarization SAR image classification method, system and device based on collaborative regularization and superpixels.

Background technique

Polarimetric SAR can not only obtain large-area high-resolution images in all-day and all-weather, but also obtain detailed and rich ground object information by using a variety of polarization combinations. Therefore, it is widely used in military and civilian fields. Among them, the classification of ground objects is an important part of polarimetric SAR image interpretation.

Lee et al. proposed a Wishart distance-based classification method in 1994, which became one of the most classic methods (see references: J.S.Lee, M.R.Grunes, R.Kwok, "Classification of multi-lookpolarimetric SAR imagery based on complex Wishart distribution", Int. J. RemoteSensing, vol. 15, no. 11, 1994.). The Wishart distance used in this method is derived from the distribution of polarimetric SAR data through maximum likelihood, so it is suitable for polarimetric SAR image classification and is widely used.

In 1999, Lee et al. combined H/A/α polarimetric target decomposition and Wishart classifier for polarimetric SAR image classification (see References: Jong-Sen Lee, M.R.Grunes, T.L.Ainsworth, Li-Jen Du, D.L. Schuler and S.R. Cloude, "Unsupervised classification using polarimetric decomposition and the complex Wishart classifier," in IEEE Transactions on Geoscience and Remote Sensing, vol. 37, no. 5, pp. 2249-2258, Sept. 1999.). Anfinsen et al. proposed a spectral clustering method based on Wishart distance. As the initial value of Wishart classifier, the paper proposed a modified version of Wishart distance that satisfies non-negativity and symmetry to construct a weight map (see Reference: S.N.Anfinsen , R. Jenssen, and T. Eltoft, "Spectral clustering of polarimetric SAR data with Wishart-derived distance measures," in Proc. POLInSAR 2007, Esrin, Italy, 22-26 January 2007, Esrin, Italy, 2007.).

In addition, the existing polarimetric SAR image classification includes two elements: feature extraction and classifier design. Therefore, whether it is possible to extract suitable features greatly affects the classification effect of polarimetric SAR images. Among them, the polarimetric feature is the most commonly used feature for polarimetric SAR image classification, which can be extracted from polarimetric SAR data, various polarimetric target decomposition methods, etc. Tu et al. summarized 42-dimensional polarimetric features and used Laplacian feature maps (LE) to extract low-dimensional features for polarimetric SAR image classification (see References: S.T.Tu, J.Y.Chen, W.Yang, and H. . Sun, "Laplacian eigenmaps-based polarimetric dimensionality reduction for SAR image classification," IEEE Transactions on Geoscience and Remote Sensing, vol. 50, no. 1, pp. 170–179, Jan. 2012.). Yang et al. summarized the polarimetric features generated by the decomposition of various polarimetric targets, and proposed a CNN-based feature selection method to select polarimetric features for polarimetric SAR image classification (see Reference: C. Yang,B.Hou,B.Ren,Y.Hu and L.Jiao,"CNN-Based Polarimetric Decomposition FeatureSelection for PolSAR Image Classification,"in IEEE Transactions on Geoscienceand Remote Sensing,vol.57,no.11,pp.8796 -8812, Nov. 2019.).

However, the above methods all have the following problems: (1) Wishart distance and polarization characteristics both play a crucial role in the classification of polarimetric SAR images, but they do not consider a reasonable and sufficient combination of Wishart distance and polarization characteristics; ( 2) Based on a single pixel, the spatial information is not considered, it is sensitive to noise, and the classification effect is not ideal.

SUMMARY OF THE INVENTION

In order to solve the above problems in the prior art, that is, in order to solve the problem that the existing polarimetric SAR image classification method does not consider the spatial information of the pixel points and does not fully combine the Wishart distance and polarization features, resulting in low classification accuracy, the present invention first On the one hand, a method for polarimetric SAR image classification based on co-regularization and superpixels is proposed, which includes:

Step S100, based on the acquired polarimetric SAR image, obtain a plurality of superpixels and their corresponding coherence matrices by a superpixel generation method;

Step S200, obtaining the covariance matrix of each superpixel; decomposing the coherence matrix and the covariance matrix respectively by presetting multiple types of polarization target decomposition methods, and combining the coherence matrix and the covariance matrix matrix to extract the polarization characteristics of each superpixel preset dimension;

Step S300, in the d × d area centered on each superpixel, calculate its Wishart distance and Euclidean distance with other superpixels based on its coherence matrix and polarization characteristics, and sort in ascending order; The method obtains the weight values corresponding to the sorted Wishart distance and Euclidean distance respectively, and constructs the first weight map and the second weight map corresponding to each superpixel;

Step S400, for each superpixel, construct a corresponding first matrix and a second matrix based on its first weight map and second weight map; The co-regularized dimensionality reduction model obtains the first low-dimensional feature and the second low-dimensional feature; the first matrix and the second matrix are Laplacian matrices of the normalized graph;

Step S500: Average the first low-dimensional feature and the second low-dimensional feature, obtain the average feature corresponding to each superpixel, and obtain the classification result of the polarimetric SAR image through the nearest neighbor classifier.

In some preferred embodiments, the coherence matrix is an average value of the coherence matrix corresponding to each pixel in each superpixel.

In some preferred embodiments, the preset multiple types 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, "respectively obtain the weight values corresponding to the sorted Wishart distance and the Euclidean distance through the preset weight assignment method", and the method is:

Based on the sorted Wishart distance and Euclidean distance, select the first k Wishart distances and Euclidean distances, and obtain the corresponding weight values through the following formulas respectively;

Among them, w _ij is the weight value, σ=max(d)-min(d), and d is the Wishart distance or the Euclidean distance;

The weight values corresponding to the remaining Wishart distance and Euclidean distance are set to zero.

In some preferred embodiments, in step S400, "constructing a corresponding first matrix and a second matrix based on its first weight map and second weight map", the method is as follows:

Step S410, based on the first weight map and the second weight map of each superpixel, construct a corresponding first diagonal matrix and a second diagonal matrix;

Step S411 , based on the first diagonal matrix and the second diagonal matrix of each superpixel, and combining the first weight map and the second weight map respectively, construct a first matrix and a second matrix corresponding to each superpixel.

In some preferred embodiments, the dimensionality reduction model based on co-regularization is expressed as:

Among them, U ⁽¹⁾ is the first low-dimensional feature, L ⁽¹⁾ is the first matrix, U ⁽²⁾ is the second low-dimensional feature, L ⁽²⁾ is the second matrix, α is the parameter, and N is the superpixel The number of , M is the dimension after dimensionality reduction, tr( ) is the trace of the matrix, T is the transpose, and R is a real number.

In some preferred embodiments, in step S400 "respectively obtain the first low-dimensional feature and the second low-dimensional feature through a pre-built dimensionality reduction model based on collaborative regularization", the method is as follows:

Step S420, respectively sort the eigenvalues in the first matrix and the second matrix in descending order, and select the eigenvectors corresponding to the first M eigenvalues as the first low-dimensional feature and the second low-dimensional feature after sorting;

Step A421, based on the first low-dimensional feature, perform eigenvalue decomposition on the matrix L ⁽²⁾ +αU ⁽¹⁾ U ^(1)T to obtain multiple eigenvalues, sort them in descending order, and select the first M features after sorting The feature vector corresponding to the value is used as the updated second low-dimensional feature;

Step S422, based on the second low-dimensional feature, perform eigenvalue decomposition on the matrix L ⁽¹⁾ +αU ⁽²⁾ U ^(2)T to obtain multiple eigenvalues, sort them in descending order, and select the first M features after sorting The feature vector corresponding to the value is used as the updated first low-dimensional feature;

Step S423, obtaining the current number of iterations and the F-norm of the difference between the first low-dimensional feature and the updated first low-dimensional feature, the second low-dimensional feature and the updated second low-dimensional feature. The sum of the F-norms of the difference of the dimensional features, if the sum is greater than the preset threshold or the current number of iterations is greater than the preset maximum number of iterations, output the first low-dimensional feature and the second low-dimensional feature, otherwise loop The method of step S421-step S422 is performed.

In the second aspect of the present invention, a system for classification of polarimetric SAR images based on collaborative regularization and superpixels is proposed. The system includes a superpixel generation module, a polarimetric feature extraction module, a weight map acquisition module, and a feature dimension reduction module. , output classification result module;

The superpixel generation module is configured to obtain a plurality of superpixels and their corresponding coherence matrices through a superpixel generation method based on the obtained polarimetric SAR image;

The polarization feature extraction module is configured to obtain the covariance matrix of each superpixel; decompose the coherence matrix and the covariance matrix respectively by presetting multiple types of polarization target decomposition methods, and combine the The coherence matrix and the covariance matrix are used to extract the polarization characteristics of each superpixel preset dimension;

The described acquisition weight map module is configured to calculate the Wishart distance and Euclidean distance with other superpixels based on its coherence matrix and polarization characteristics in a d×d area centered on each superpixel, and sort them in ascending order; The weight values corresponding to the sorted Wishart distance and the Euclidean distance are obtained respectively through the preset weight assignment method, and the first weight map and the second weight map corresponding to each superpixel are constructed;

The feature dimension reduction module is configured to construct a corresponding first matrix and a second matrix based on the first weight map and the second weight map for each superpixel; based on the first matrix and the second matrix, Obtain the first low-dimensional feature and the second low-dimensional feature respectively through the pre-built dimensionality reduction model based on collaborative regularization; the first matrix and the second matrix are Laplacian matrices of the normalized graph;

The output classification result module is configured to average the first low-dimensional feature and the second low-dimensional feature, obtain the average feature corresponding to each superpixel, and obtain the polarimetric SAR image through the nearest neighbor classifier. Classification results.

In a third aspect of the present invention, a storage device is provided, in which a plurality of programs are stored, and the program applications are loaded and executed by a processor to realize the above-mentioned collaborative regularization and superpixel-based polarization SAR image classification method.

In a fourth aspect of the present invention, a processing device is proposed, including a processor and a storage device; the processor is adapted to execute various programs; the storage device is adapted to store multiple programs; the programs are adapted to be loaded by the processor And execute to realize the above-mentioned collaborative regularization and superpixel based polarimetric SAR image classification method.

Beneficial effects of the present invention:

According to the spatial information of pixel points, the invention improves the classification accuracy of polarimetric SAR images by combining Wishart distance and polarimetric features. By utilizing the spatial information of the pixel points, the present invention performs superpixel segmentation on the polarimetric SAR image, and uses the superpixel as a processing unit, thereby greatly reducing the computational burden and the influence of noise.

At the same time, the present invention obtains the coherence matrix and covariance matrix of a plurality of superpixels, and decomposes the matrix through a plurality of polarization target decomposition methods to fully extract polarization features. By using the Wishart distance of the coherence matrix and the Euclidean distance of the polarization feature respectively, the weight map is constructed by locally searching for the adjacent samples, and the feature extraction is carried out by using the collaborative regularization method, which improves the classification accuracy.

Description of drawings

Other features, objects and advantages of the present application will become more apparent upon reading the detailed description of non-limiting embodiments taken with reference to the following drawings.

1 is a schematic flowchart of a method for classifying polarimetric SAR images based on collaborative regularization and superpixels according to an embodiment of the present invention;

2 is a schematic diagram of a framework of a polarimetric SAR image classification system based on collaborative regularization and superpixels according to an embodiment of the present invention;

3 is a pseudo-color schematic diagram of a polarimetric SAR image according to an embodiment of the present invention;

FIG. 4 is a schematic diagram of a polarimetric SAR image real feature marker according to an embodiment of the present invention;

5 is a schematic diagram of a classification result of a pixel-based Wishart classification method according to an embodiment of the present invention;

6 is a schematic diagram of a classification result of the Wihart classification method based on superpixels according to an embodiment of the present invention;

7 is a schematic diagram of a classification result of a polarimetric SAR image classification method based on collaborative regularization and superpixels according to an embodiment of the present invention.

Detailed ways

In order to make the objectives, 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. Obviously, the described embodiments are part of the embodiments of the present invention, not All examples. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

The present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the related invention, but not to limit the invention. In addition, it should be noted that, for the convenience of description, only the parts related to the related invention are shown in the drawings.

It should be noted that the embodiments in the present application and the features of the embodiments may be combined with each other in the case of no conflict.

The polarimetric SAR image classification method based on collaborative regularization and superpixels of the present invention, as shown in Figure 1, includes the following steps:

Step S100, based on the acquired polarimetric SAR image, obtain a plurality of superpixels and their corresponding coherence matrices by a superpixel generation method;

Step S200, obtaining the covariance matrix of each superpixel; decomposing the coherence matrix and the covariance matrix respectively by presetting multiple types of polarization target decomposition methods, and combining the coherence matrix and the covariance matrix matrix to extract the polarization characteristics of each superpixel preset dimension;

Step S300, in the d × d area centered on each superpixel, calculate its Wishart distance and Euclidean distance with other superpixels based on its coherence matrix and polarization characteristics, and sort in ascending order; The method obtains the weight values corresponding to the sorted Wishart distance and Euclidean distance respectively, and constructs the first weight map and the second weight map corresponding to each superpixel;

Step S400, for each superpixel, construct a corresponding first matrix and a second matrix based on its first weight map and second weight map; The co-regularized dimensionality reduction model obtains the first low-dimensional feature and the second low-dimensional feature; the first matrix and the second matrix are Laplacian matrices of the normalized graph;

Step S500: Average the first low-dimensional feature and the second low-dimensional feature, obtain the average feature corresponding to each superpixel, and obtain the classification result of the polarimetric SAR image through the nearest neighbor classifier.

In order to more clearly describe the polarization SAR image classification method based on synergistic regularization and superpixels of the present invention, each step in an embodiment of the method of the present invention will be described in detail below with reference to the accompanying drawings.

Step S100, based on the acquired polarimetric SAR image, obtain a plurality of superpixels and their corresponding coherence matrices through a superpixel generation method.

Synthetic Aperture Radar (SAR) is an all-weather, all-weather active microwave remote sensing imaging radar, which is widely used in geological survey, disaster control, parameter inversion and military fields. Polarimetric SAR can record the complete polarimetric scattering information of ground targets, which greatly improves the accuracy of image interpretation and analysis. The classification of polarimetric images is the basis and premise of object-oriented polarimetric SAR image interpretation.

In this embodiment, based on the acquired polarimetric SAR image, a plurality of superpixels are acquired through an adaptive superpixel generation method. The adaptive superpixel generation method can refer to the literature: "D.Xiang,Y.Ban,W.Wang and Y.Su,"Adaptive Superpixel Generation for Polarimetric SAR Images With LocalIterative Clustering and SIRV Model,"in IEEE Transactions on Geoscience andRemote Sensing, vol. 55, no. 6, pp. 3115-3131, June 2017.”. According to the obtained multiple superpixels, the average number of coherence matrices including all pixel points in each superpixel is calculated as the coherence matrix T _i of each superpixel, where i is a subscript.

Step S200, obtaining the covariance matrix of each superpixel; decomposing the coherence matrix and the covariance matrix respectively by presetting multiple types of polarization target decomposition methods, and combining the coherence matrix and the covariance matrix matrix to extract the polarization features of each superpixel preset dimension.

In this embodiment, the covariance matrix of each superpixel is obtained first, and the covariance is analyzed by various polarization target decomposition methods, such as Pauli decomposition, H/A/α decomposition, Freeman decomposition, Kroggar decomposition, and Huynen decomposition. The matrix and coherence matrix are decomposed, and combined with the coherence matrix and covariance matrix, the polarization characteristics of the preset dimension of each pixel in each superpixel are extracted, and normalized to between 0 and 1. In the present invention, the preset dimension is preferably set to 30 dimensions.

Calculate the average number of polarization features of all pixels contained in each superpixel, as the polarization feature x _i of the superpixel, assuming that the number of superpixels is N, the polarization feature corresponding to the polarization SAR image is

Step S300, in the d × d area centered on each superpixel, calculate its Wishart distance and Euclidean distance with other superpixels based on its coherence matrix and polarization characteristics, and sort in ascending order; The method obtains the weight values corresponding to the sorted Wishart distance and Euclidean distance respectively, and constructs the first weight map and the second weight map corresponding to each superpixel.

In this embodiment, in the d×d area centered on each _superpixel , search for the k nearest _superpixels with Wishart distance dW and Euclidean distance dP respectively, if superpixel j is a neighboring sample of superpixel i , obtain the corresponding weight value through formula (1), and construct the sparse weight map W ⁽¹⁾ , W ⁽²⁾ . Formula (1) is as follows:

Wherein, w _ij is the weight value, σ=max(d)-min(d), and d is the Wishart distance or the Euclidean distance.

The specific process of constructing the weight map is as follows:

In the d × d area centered on each superpixel, based on its coherence matrix, its Wishart distance from other superpixels is calculated. In this embodiment, the Wishart distance satisfying non-negativity and symmetry is adopted, and its calculation is shown in formula (2):

Among them, j is the subscript, and tr( ) is the trace of the matrix.

Sort in ascending order according to the obtained Wishart distance, select the first k Wishart distances, and obtain the corresponding weight value through formula (3). Formula (3) is as follows:

为Wishart距离对应的权重值。in,

is the weight value corresponding to Wishart distance.

The weight values corresponding to other Wishart distances are set to 0, and the first weight map corresponding to each superpixel is constructed based on the obtained weight values.

In the same way, in the d × d area centered on each superpixel, based on its polarization characteristics, calculate its Euclidean distance from other superpixels, sort in ascending order according to the obtained Euclidean distance, select the first k Euclidean distances, and pass Formula (4) obtains the corresponding weight value. Formula (4) is as follows:

为欧式距离对应的权重值。in,

is the weight value corresponding to the Euclidean distance.

The weight values corresponding to other Euclidean distances are set to 0, and a second weight map corresponding to each superpixel is constructed based on the obtained weight values.

Step S400, for each superpixel, construct a corresponding first matrix and a second matrix based on its first weight map and second weight map; The co-regularized low-dimensional model obtains the first low-dimensional feature and the second low-dimensional feature; the first and second matrices are Laplacian matrices of the normalized graph.

In this embodiment, for each superpixel, a corresponding first matrix and a second matrix are constructed based on its first weight map and second weight map. The specific processing steps are as follows:

Step S410, based on the first weight map and the second weight map of each superpixel, construct a corresponding first diagonal matrix D ⁽¹⁾ and 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 superpixel, and in combination with the first weight map and the second weight map, respectively, construct the first matrix L ⁽¹⁾ corresponding to each superpixel, The second matrix L ⁽²⁾ . Among them, L ⁽¹⁾ and L ⁽²⁾ are the normalized graph Laplacian matrices corresponding to W ⁽¹⁾ and W ⁽²⁾ .

The construction process is shown in formula (7) (8):

L ⁽¹⁾ = ID ^(1)-1/2 W ⁽¹⁾ D ^(1)-1/2 (7)

L ⁽²⁾ = ID ^(2)-1/2 W ⁽²⁾ D ^(2)-1/2 (8)

where I is the identity matrix.

Through the first matrix and the second matrix, a dimensionality reduction model based on cooperative regularization is constructed. Among them, the constructed dimensionality reduction model based on collaborative regularization is shown in formula (9):

为F-范数的平方。本公式中的第三项为正则项。Among them, U ⁽¹⁾ is the first low-dimensional feature, L ⁽¹⁾ is the first matrix, U ⁽²⁾ is the second low-dimensional feature, L ⁽²⁾ is the second matrix, α is a parameter, and M is dimensionality reduction After the dimension, T represents the transpose, R represents the real number,

is the square of the F-norm. The third term in this formula is the regular term.

After simplification, the above formula can be expressed as formula (10):

The first low-dimensional feature and the second low-dimensional feature are obtained through the constructed dimensionality reduction model based on co-regularization. The specific processing steps are as follows:

Step S420, sorting the eigenvalues in the first matrix and the second matrix in descending order respectively, and selecting the eigenvectors corresponding to the first M eigenvalues as the first low-dimensional feature and the second low-dimensional feature after sorting;

Step A421, based on the first low-dimensional feature, perform eigenvalue decomposition on the matrix L ⁽²⁾ +αU ⁽¹⁾ U ^(1)T to obtain a plurality of eigenvalues, sort them in descending order, and select the first M eigenvalues corresponding to them after sorting. The feature vector of is used as the updated second low-dimensional feature; that is, the first low-dimensional feature is fixed, and the second low-dimensional feature is solved;

Step S422, based on the second low-dimensional feature, perform eigenvalue decomposition on the matrix L ⁽¹⁾ + αU ⁽²⁾ U ^{(2) T} to obtain a plurality of eigenvalues, perform descending sorting, and select the first M eigenvalues corresponding to the sorting. The feature vector of is used as the updated first low-dimensional feature;

Step S423, obtaining the current iteration number and 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. If the sum is less than the preset threshold or the current iteration number is less than the preset maximum number of iterations, the method of step S421-step S422 is executed cyclically. Among them, the preset threshold is judged as shown in formula (11):

为未更新的第一、第二低维特征，

为更新后的第一、第二低维特征，ε为预设的阈值。in,

are the unupdated first and second low-dimensional features,

are the updated first and second low-dimensional features, and ε is a preset threshold.

Otherwise, output the final first low-dimensional feature and the second low-dimensional feature.

Step S500: Average the first low-dimensional feature and the second low-dimensional feature, obtain the average feature corresponding to each superpixel, and obtain the classification result of the polarimetric SAR image through the nearest neighbor classifier.

In this embodiment, the first low-dimensional feature and the second low-dimensional feature of each superpixel are averaged to obtain the average feature as a feature for classification. Among them, the average process is shown in formula (12):

U=(U ⁽¹⁾ +U ⁽²⁾ )/2 (12)

where U is the average feature.

According to the average feature of each superpixel, the classification result of the polarimetric SAR image is obtained through the nearest neighbor classifier.

In the present invention, the method of the present invention is compared with the Wishart classification method based on pixel points and the Wishart classification method based on superpixels through simulation experiments, so as to prove the effectiveness of the present invention in classifying polarimetric SAR images. 10% of each class is randomly selected as the training set, and the rest is used as the test set to obtain the classification accuracy.

The simulation experiments are carried out in the server hardware environment of Intel(R) Core(TM) i9-8950HK CPU 2.90GHz 32G RAM and the software environment of Matlab R2016a. The experimental data used are farmland data in the Flevoland (plain) area, with a size of 200 × 320.

Table 1 shows the classification accuracy obtained by the three methods in the simulation:

Table 1

As can be seen from Table 1, the classification accuracy obtained by using the present invention is significantly higher than the classical Wishart classification method, and the classification accuracy of the Wishart classification method based on superpixels is higher than the accuracy rate of the Wishart classification method based on a single pixel point. The advantages of the superpixel-based method are shown, and it is fully demonstrated that the method of the present invention has a good classification effect on polarimetric SAR image classification. Among them, in Table 1, the English on the left is the name of the crops grown in each farmland, Potatoes is potatoes, Grass is grass, Beet is sugar beet, Lucerne is alfalfa, Wheat is wheat field, Stem beans is dry beans, Bare soil is bare land , Rapeseed for Rapeseed. Total accuracy is the total correct rate.

In order to better illustrate the classification effect, the present invention is shown by a legend. Fig. 3 is a pseudo-color image of the polarimetric SAR image used for the simulation experiment, Fig. 4 is a real ground object labeling graph of the polarimetric SAR image, Fig. 5 is a classification result diagram of the Wishart classification method based on pixel points, and Fig. 6 is a A classification result diagram of the Wishart classification method based on superpixels, and FIG. 7 is a classification result diagram of the method of the present invention. By comparison, it can also be seen that the method of the present invention has a good classification effect.

A polarimetric SAR image classification system based on collaborative regularization and superpixels according to the second embodiment of the present invention, as shown in FIG. 2 , includes: a superpixel generation module 100 , a polarimetric feature extraction module 200 , and a weight map acquisition module 300 , feature dimension reduction module 400, output classification result module 500;

The superpixel generation module 100 is configured to obtain a plurality of superpixels and their corresponding coherence matrices through a superpixel generation method based on the obtained polarimetric SAR image;

The polarization feature extraction module 200 is configured to obtain the covariance matrix of each superpixel; the coherence matrix and the covariance matrix are decomposed respectively by preset multiple types of polarization target decomposition methods, and combined with all the polarization target decomposition methods. The coherence matrix and the covariance matrix are used to extract the polarization characteristics of each superpixel preset dimension;

The acquisition weight map module 300 is configured to calculate the Wishart distance and Euclidean distance with other superpixels based on its coherence matrix and polarization characteristics in a d×d area centered on each superpixel, and perform ascending sorting. ; Obtain the weight values corresponding to the sorted Wishart distance and the Euclidean distance through the preset weight assignment method, and construct the first weight map and the second weight map corresponding to each superpixel;

The feature dimension reduction module 400 is configured to construct a corresponding first matrix and a second matrix based on its first weight map and second weight map for each superpixel; based on the first matrix and the second matrix , respectively obtain the first low-dimensional feature and the second low-dimensional feature through a pre-built dimensionality reduction model based on collaborative regularization; the first matrix and the second matrix are Laplacian matrices of the normalized graph;

The output classification result module 500 is configured to average the first low-dimensional feature and the second low-dimensional feature, obtain the average feature corresponding to each superpixel, and obtain the polarimetric SAR image through a nearest neighbor classifier classification results.

Those skilled in the technical field can clearly understand that, for the convenience and brevity of description, for the specific working process and related description of the system described above, reference may be made to the corresponding process in the foregoing method embodiments, which will not be repeated here.

It should be noted that the polarimetric SAR image classification system based on collaborative regularization and superpixels provided by the above embodiments is only illustrated by the division of the above functional modules. In practical applications, the above functions can be allocated as required. It is completed by different functional modules, that is, the modules or steps in the embodiments of the present invention are decomposed or combined. For example, the modules in the above embodiments can be combined into one module, and can also be further split into multiple sub-modules to complete the above description. all or part of the functions. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing each module or step, and should not be regarded as an improper limitation of the present invention.

A storage device according to the third embodiment of the present invention stores a plurality of programs, and the programs are suitable for being loaded by a processor and implementing the above-mentioned method for classifying polarimetric SAR images based on cooperative regularization and superpixels.

A processing device according to a fourth embodiment of the present invention includes a processor and a storage device; the processor is adapted to execute various programs; the storage device is adapted to store multiple programs; the programs are adapted to be loaded and executed by the processor In order to realize the above-mentioned co-regularization and superpixel based polarimetric SAR image classification method.

Those skilled in the technical field can clearly understand that the undescribed convenience and brevity are not described. Repeat.

Those skilled in the art should be able to realize that the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software or a combination of the two, and the programs corresponding to the software modules and method steps Can be placed in random access memory (RAM), internal memory, read only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or as known in the art in any other form of storage medium. In order to clearly illustrate the interchangeability of electronic hardware and software, the components and steps of each example have been described generally in terms of functionality in the foregoing description. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may use different methods of implementing the described functionality for each particular application, but such implementations should not be considered beyond the scope of the present invention.

The terms "first," "second," etc. are used to distinguish between similar objects, and are not used to describe or indicate a particular order or sequence.

The term "comprising" or any other similar term is intended to encompass a non-exclusive inclusion such that a process, method, article or device/means comprising a list of elements includes not only those elements but also other elements not expressly listed, or Also included are elements inherent to these processes, methods, articles or devices/devices.

So far, the technical solutions of the present invention have been described with reference to the preferred embodiments shown in the accompanying drawings, however, those skilled in the art can easily understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims (10)

1. a polarization SAR image classification method based on collaborative regularization and superpixel, is characterized in that, this method comprises: Step S100, based on the acquired polarimetric SAR image, obtain a plurality of superpixels and their corresponding coherence matrices by a superpixel generation method; Step S200, obtaining the covariance matrix of each superpixel; decomposing the coherence matrix and the covariance matrix respectively by presetting multiple types of polarization target decomposition methods, and combining the coherence matrix and the covariance matrix matrix to extract the polarization characteristics of each superpixel preset dimension; Step S300, in the d × d area centered on each superpixel, calculate its Wishart distance and Euclidean distance with other superpixels based on its coherence matrix and polarization characteristics, and sort in ascending order; The method obtains the weight values corresponding to the sorted Wishart distance and Euclidean distance respectively, and constructs the first weight map and the second weight map corresponding to each superpixel; Step S400, for each superpixel, construct a corresponding first matrix and a second matrix based on its first weight map and second weight map; The co-regularized dimensionality reduction model obtains the first low-dimensional feature and the second low-dimensional feature; the first matrix and the second matrix are Laplacian matrices of the normalized graph; Step S500: Average the first low-dimensional feature and the second low-dimensional feature, obtain the average feature corresponding to each superpixel, and obtain the classification result of the polarimetric SAR image through the nearest neighbor classifier. 2 . The method for classifying polarimetric SAR images based on collaborative regularization and superpixels according to claim 1 , wherein the coherence matrix is an average value of the coherence matrices corresponding to each pixel in each superpixel. 3 . 3. the polarization SAR image classification method based on collaborative regularization and superpixel according to claim 1, is characterized in that, the polarization target decomposition method of described preset multiple types comprises Pauli decomposition, H/A/α Decomposition, Freeman Decomposition, Kroggar Decomposition, Huynen Decomposition. 4. the polarimetric SAR image classification method based on collaborative regularization and superpixels according to claim 1, is characterized in that, in step S300, "respectively obtain the Wishart distance after sorting, the Euclidean distance corresponding to by the preset weight assignment method. The weight value of ", the method is: Based on the sorted Wishart distance and Euclidean distance, select the first k Wishart distances and Euclidean distances, and obtain the corresponding weight values through the following formulas:

Among them, w _ij is the weight value, σ=max(d)-min(d), and d is the Wishart distance or the Euclidean distance; The weight values corresponding to the remaining Wishart distance and Euclidean distance are set to zero. 5. the polarimetric SAR image classification method based on collaborative regularization and superpixel according to claim 1, is characterized in that, in step S400 " build the corresponding first matrix based on its first weight map, the second weight map, The second matrix", whose method is: Step S410, based on the first weight map and the second weight map of each superpixel, construct a corresponding first diagonal matrix and a second diagonal matrix; Step S411 , based on the first diagonal matrix and the second diagonal matrix of each superpixel, and combining the first weight map and the second weight map respectively, construct a first matrix and a second matrix corresponding to each superpixel. 6. The polarimetric SAR image classification method based on collaborative regularization and superpixels according to claim 1, wherein the described dimensionality reduction model based on collaborative regularization is expressed as:

Among them, U ⁽¹⁾ is the first low-dimensional feature, L ⁽¹⁾ is the first matrix, U ⁽²⁾ is the second low-dimensional feature, L ⁽²⁾ is the second matrix, α is the parameter, and N is the superpixel The number of , M is the dimension after dimensionality reduction, tr( ) is the trace of the matrix, T is the transpose, and R is a real number. 7. the polarimetric SAR image classification method based on collaborative regularization and superpixels according to claim 6, is characterized in that, in step S400, "respectively obtain the first low-dimensional dimensionality reduction model based on pre-built collaborative regularization. feature, the second low-dimensional feature", the method is: Step S420, respectively sort the eigenvalues in the first matrix and the second matrix in descending order, and select the eigenvectors corresponding to the first M eigenvalues as the first reduced-dimensional feature and the second low-dimensional feature after sorting; Step A421, based on the first low-dimensional feature, perform eigenvalue decomposition on the matrix L ⁽²⁾ +αU ⁽¹⁾ U ^(1)T to obtain multiple eigenvalues, sort them in descending order, and select the first M features after sorting The feature vector corresponding to the value is used as the updated second low-dimensional feature; Step S422, based on the second low-dimensional feature, perform eigenvalue decomposition on the matrix L ⁽¹⁾ +αU ⁽²⁾ U ^(2)T to obtain multiple eigenvalues, sort them in descending order, and select the first M features after sorting The feature vector corresponding to the value is used as the updated first low-dimensional feature; Step S423, obtaining the current number of iterations and the F-norm of the difference between the first low-dimensional feature and the updated first low-dimensional feature, the second low-dimensional feature and the updated second low-dimensional feature. The sum of the F-norms of the difference of the dimensional features, if the sum is greater than the preset threshold or the current number of iterations is greater than the preset maximum number of iterations, output the first low-dimensional feature and the second low-dimensional feature, otherwise loop The method of step S421-step S422 is performed. 8. A polarimetric SAR image classification system based on collaborative regularization and superpixels, characterized in that the system comprises a superpixel generation module, a polarimetric feature extraction module, an acquisition weight map module, a feature dimension reduction module, and an output classification result. module; The superpixel generation module is configured to obtain a plurality of superpixels and their corresponding coherence matrices through a superpixel generation method based on the obtained polarimetric SAR image; The polarization feature extraction module is configured to obtain the covariance matrix of each superpixel; decompose the coherence matrix and the covariance matrix respectively by presetting multiple types of polarization target decomposition methods, and combine the The coherence matrix and the covariance matrix are used to extract the polarization characteristics of each superpixel preset dimension; The described acquisition weight map module is configured to calculate the Wishart distance and Euclidean distance with other superpixels based on its coherence matrix and polarization characteristics in a d×d area centered on each superpixel, and sort them in ascending order; The weight values corresponding to the sorted Wishart distance and the Euclidean distance are obtained respectively through the preset weight assignment method, and the first weight map and the second weight map corresponding to each superpixel are constructed; The feature dimension reduction module is configured to construct a corresponding first matrix and a second matrix based on the first weight map and the second weight map for each superpixel; based on the first matrix and the second matrix, Obtain the first low-dimensional feature and the second low-dimensional feature respectively through the pre-built dimensionality reduction model based on collaborative regularization; the first matrix and the second matrix are Laplacian matrices of the normalized graph; The output classification result module is configured to average the first low-dimensional feature and the second low-dimensional feature, obtain the average feature corresponding to each superpixel, and obtain the polarimetric SAR image through the nearest neighbor classifier. Classification results. 9. A storage device, wherein storing a plurality of programs, is characterized in that, described program application is loaded and executed by processor to realize the polar based on cooperative regularization and superpixel described in any one of claim 1-7. SAR image classification method. 10. A processing device, comprising a processor and a storage device; the processor is adapted to execute each program; the storage device is adapted to store a plurality of programs; characterized in that the program is adapted to be loaded and executed by the processor to The method for classifying polarimetric SAR images based on cooperative regularization and superpixels according to any one of claims 1 to 7 is implemented.

Images