CN103984746A - Semi-supervised classification and regional distance measurement based SAR (Synthetic Aperture Radar) image identification method - Google Patents

Abstract

The invention discloses a SAR image recognition method based on semi-supervised classification and region distance measure. The implementation steps are: establish an image library by segmenting the original SAR image, select a single target SAR image block from the image library; extract the feature vector of the image block in the library; divide the selected SAR image blocks into several categories, and use the corresponding The feature vector is used as a training sample to train a semi-supervised classifier, and use this classifier to classify the image library; for the query image block input by the user, use the trained classifier to obtain its category; obtain the category set of the query image block according to the confusion matrix , calculate the area similarity distance between the query image block and the image blocks belonging to the set in the image database, and return the number of image blocks required by the user in the order of the distance from small to large. The invention has the advantages of correctable classification errors and high information recognition accuracy, and can be used to simultaneously interpret multiple SAR images.

Description

SAR image recognition method based on semi-supervised classification and region distance measure

technical field

The invention belongs to the technical field of image processing, and relates to a method for recognizing SAR images, which can be applied to simultaneous interpretation of multiple SAR images.

Background technique

Because of the all-day and all-weather detection capability of SAR images, especially the fact that relative optical images are completely independent of weather factors, its application fields are gradually expanding, including military, agriculture, navigation, geographic surveillance, etc. The segmentation, denoising, and change detection of SAR images are all research hotspots, and an important basis of these research fields is SAR image recognition. Some traditional recognition technologies are mainly aimed at the problem of recognition accuracy, and most of them are applied to the small-scale area recognition problem of a single SAR image. However, these technologies are obviously not suitable for the current application environment where the number of SAR images is increasing massively. In order to overcome the above-mentioned technical shortcomings, the present invention proposes a SAR image recognition technology under the framework of content-based image retrieval technology in combination with graph theory semi-supervised learning method and improved region distance measure. The technology is simple to implement, which not only ensures the recognition accuracy but also meets the application scenario of simultaneous recognition of a large number of SAR images.

In practical problems, easily available labeled data are often much less than unlabeled data. In order to solve such problems, semi-supervised learning methods came into being. The semi-supervised learning method is a kind of learning method between supervised learning and unsupervised learning. This method uses both labeled samples and unlabeled samples, and uses the overall geometric structure to complete the class label propagation. In the field of semi-supervised learning in recent years, the semi-supervised learning method based on graph theory is the most active research direction. The most famous and widely used semi-supervised learning methods based on graph theory include: minimum cut method, Gaussian random field method, etc., see A.Blum and S.Chawla.Learning from labeled and unlabeled data using graph mincuts.In Prnceedings of the18th International Conference on Machine Learning. Morgan Kaufmann, San Francisco, CA, 2001.19-26; Zhu, Xiaojin, Zoubin Ghahramani, and John Lafferty."Semi-supervised learning using gaussian fields and harmonic functions."ICML.Vol.3.2003. These methods use labeled and unlabeled data as the vertices of the graph, and the similarity between the data as the edges and weights connecting the vertices, and use the geometric characteristics of the graph to complete the propagation of class labels from labeled data to unlabeled data. So as to achieve the purpose of classification. Due to the low computational complexity of the Gaussian random field method, the present invention adopts this method in consideration of the timeliness of the identification system.

Content-based image retrieval (CBIR) is based on the low-level visual features of the image to complete the process of retrieving images in the image database that are consistent or similar in content to the query image. This technology includes a series of image processing methods, including feature extraction, similarity measurement, user feedback, etc. So far, many mature and well-known retrieval systems have been proposed, such as SIMPLIcity retrieval system, see James Z.Wang, Jia Li, Gio Wiederhold.SIMPLIcity: Semantics-Sensitive Integrated Matching for Picture Llbraries.IEEE Trans.onPattern Analysis and Machine Intelligence, 2001, 23(9): 947-963, the SIMPLIcity retrieval system has been successfully applied to massive natural image retrieval problems, but due to technical limitations and the characteristics of SAR images, it is not effective to directly apply it to SAR image recognition. And ideally. Another example is the SAR image retrieval system combined with Gaussian mixture model classification proposed in 2009, that is, the GMM retrieval system, see Hou, B., Tang, X., Jiao, L., & Wang, S. (2009, October). SAR image retrieval based on Gaussian Mixture Model classification.In SyntheticAperture Radar,2009.APSAR2009.2nd Asian-Pacific Conference on(pp.796-799).IEEE, this method is oriented to SAR images and effectively uses texture features in the retrieval process, but Due to the use of supervised classification methods, its generalization ability in real problems is low, and because the similarity matching technology of this method does not consider the characteristics of SAR images, the retrieval effect is not ideal. Although the excellent experimental results are given in this article, these results rely on overlapping cutting original SAR images to build a gallery. The image blocks obtained by this strategy have a high degree of clustering characteristics, that is, the distance between samples in the same class is small, The distance between samples of different classes is very large, such data distribution is often quite different from the data distribution in practical applications, and the obtained experimental results cannot fully verify the effectiveness of the method.

Contents of the invention

The purpose of the present invention is to address the defects in the above-mentioned prior art, according to the special imaging characteristics of SAR images, under the framework of traditional content-based image retrieval, propose a SAR image recognition method based on semi-supervised classification and improved region distance measurement, In order to reduce the difference between the experimental data and the data distribution in the actual application, and improve the information recognition accuracy in the actual application.

The technical solution for realizing the object of the present invention is: using discrete wavelets, analyzing the texture information in the SAR image, using the texture information, utilizing the Gaussian random field semi-supervised learning method to complete the classification work of the SAR image library, and using the improved regional comprehensive feature similarity matching The algorithm completes the similarity matching of SAR images. Its specific implementation steps include the following:

1) Segment the original SAR image without overlapping to establish a SAR image library {p ₁ ,p ₂ ,…,p _N }, and select image blocks {p ₁ ,p ₂ ,… , p _l }, where l<<N, N represents the number of SAR image blocks in the gallery, l represents the number of selected SAR image blocks, and the principle of single target refers to the percentage of the total image area occupied by a certain target in the image block more than half;

2) Extract the subband energy of the discrete wavelet three-layer transform of all image blocks as the feature vector of the image block $\{f_1,f_2,\&Center\; Dot;\&Center\; Dot;\&Center\; Dot;,f_{\stackrel{\&OverBar;}{\mathrm{no}}}\},$ in, $\stackrel{\&OverBar;}{\mathrm{no}}=10;$

3) Divide the selected SAR image blocks {p ₁ ,p ₂ ,…,p _l } into { _ci ,1≤i≤k} categories according to the semantic content, where k represents the number of semantic categories, and use the corresponding feature The vector is used as a training sample to train a Gaussian random field semi-supervised classifier, and use the classifier to classify the entire SAR image library {p ₁ ,p ₂ ,…,p _N } to obtain a SAR image library with class labels;

4) For the query image block p' input by the user, use the same method as step 2) to extract its feature vector f', and use the same training samples as step 3) and the trained Gaussian random field semi-supervised classifier to obtain the query The category number c _i of the image block;

5) According to the number of categories c _i and the empirical confusion matrix obtained in step 4), calculate the category set {c} of the query image block:

5a) In the classified SAR image library, select the first K image blocks of each category to form a new image sample set, randomly select training samples from this sample set to train a Gaussian random field semi-supervised classifier, use the classifier Conduct 100 random classification experiments, and get the empirical confusion matrix Con∈R ^k×k , the confusion matrix Con is a square matrix, where row i and column j Con(i,j) indicates that samples belonging to class c _i are divided into c _j The number of classes, 1≤i≤k, 1≤j≤k;

5b) Carrying out column normalization to the empirical confusion matrix, that is, dividing each element in a column by the sum of the elements in the column to obtain the empirical posterior probability matrix ConP∈R ^k×k ;

5c) Set the threshold T, compare ConP(i,j) in the i-th row and column j in the posterior probability matrix with the threshold T, and when ConP(i,j)≤T, set ConP(i,j) is 0, otherwise ConP(i,j) remains unchanged, and the size of the threshold T is set according to the expected number of non-zero elements in each column;

5d) According to the category number c _i of the query image block, find the position of the non-zero element in the i-th column in the posterior probability matrix ConP, and finally obtain the category set {c};

6) Calculate the area distance between the query image block p' and all image blocks belonging to the category set {c} in the gallery;

7) According to the area distance obtained in step 6), the number of images required by the user is returned in order from small to large, and image recognition is completed.

Compared with the prior art, the present invention has the following advantages:

1. Since the present invention creates a SAR image database with a non-overlapping segmentation method, the gap between the test sample and the actual data is reduced, and the credibility of the recognition result is increased;

2. The present invention classifies the gallery by using the Gaussian random field semi-supervised learning method, which reduces the workload of manual selection of training samples and reduces the influence of human subjective factors on the classification results;

3. The present invention effectively reduces the impact of classification errors on similarity matching due to the adoption of the strategy of correcting classification errors;

4. Aiming at the particularity of the SAR image, the present invention improves the similarity measure of the comprehensive feature of the region, makes the matching result of the similarity of the SAR image more accurate, and improves the recognition accuracy.

Description of drawings

Fig. 1 is the realization flow schematic diagram of the present invention;

Fig. 2 is used to set up the original SAR image of SAR image storehouse among the present invention;

Fig. 3 is the SAR image block sample figure that the present invention selects in SAR gallery;

Fig. 4 is the overall performance comparison figure of the present invention and GMM retrieval system;

Fig. 5 is a performance comparison diagram between the present invention and the GMM retrieval system in each semantic category.

Detailed ways

With reference to Fig. 1, the concrete realization steps of the present invention are as follows:

Step 1. Establish the SAR image library {p ₁ ,p ₂ ,…,p _N }, and select SAR image blocks according to the principle of single target.

The specific implementation of this step is as follows:

1a) Select two large-scale SAR images with pixel sizes of 19035×7330 and 7082×7327 as the original SAR images for building the library, as shown in Figure 2(a) and Figure 2(b) respectively;

1b) Segment the two selected original SAR images without overlapping, and obtain 2828 SAR image blocks with a size of 256×256 after segmentation, and use this to build a SAR image library {p ₁ ,p ₂ ,…, p _N }, where N=2828;

1c) Select SAR image blocks {p ₁ , p ₂ ,...,p _l } in the image library according to the principle of single target, where l<<N, N represents the number of SAR image blocks in the image library, and l represents the selected The number of SAR image blocks, the single target principle means that a target in an image block accounts for more than half of the total area of the image. When the present invention is selected, a total of 28 SAR image blocks are selected, namely l=28.

Step 2, perform feature extraction on all image blocks in the gallery.

The subband energy ξ of discrete wavelet three-layer transform is selected as the feature vector of the image block in, Indicates the dimension of the feature vector, this example chooses But not limited to 10. For a subband, the energy is defined as:

$\mathrm{<span\; class="notranslate">\; \&xi;</span>}<span\; class="notranslate">\; =</span>{<span\; class="notranslate">\; (</span>\underset{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; b</span>}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span>}^{\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; /</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ></span>$

Among them, m ₁ ×m ₂ is the size of the subband, (n ₁ ,n ₂ ) indicates the index of the subband coefficient, b(n ₁ ,n ₂ ) indicates the coefficient of the n _1th row and n _2th column in the subband value, and the energies of the remaining 9 subbands are calculated according to formula <1>.

Step 3, classify the entire SAR image library {p ₁ , p ₂ ,...,p _N } to obtain a SAR image library with class labels.

The specific implementation of this step is as follows:

3a) Artificial semantic classification is performed in the selected l SAR image blocks. In this example, the strategy of area percentage is used to determine _the image category. 50% of the total area of the image, it is stipulated that the image block p is classified as c _i , and finally the l SAR image block is artificially divided into four types: mountain, ocean, urban area, and suburban area, with 7 image blocks in each type. The sample map is shown in Figure 3, where Figure 3(a) is a mountainous area, Figure 3(b) is an ocean, Figure 3(c) is an urban area, and Figure 3(d) is a suburb;

3b) According to each type of image block, corresponding to a set of sub-band energy features of discrete wavelet three-layer transform, 4 sets of feature vectors are obtained, and these 4 sets of feature vectors are used as training samples, and the Gaussian random field semi-supervised learning classifier is used. Classification of SAR image library:

3b1) Use the energy features of the 28 image blocks selected in step 1 as labeled samples {(x ₁ ,y ₁ ),…(x _l ,y _l )}, and use the remaining 2800 SAR image blocks in the SAR image library Energy features are used as unlabeled samples {x _l+1 ,…x _l+u }, and graph G=(V,E) is established with labeled samples and unlabeled samples, where l<<u, u represents the number of unlabeled samples Number, l+u=N, N represents the number of total samples, V represents the vertices of graph G, E represents the edge of graph G, and l=28, u=2800, N=2828;

3b2) Obtain the similarity matrix W of the graph G according to the formula <2>,

$\mathrm{<span\; class="notranslate">\; W</span>}<span\; class="notranslate">\; =</span>{<span\; class="notranslate">\; (</span>{\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}}_{\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; j</span>}}}<span\; class="notranslate">\; )</span>}_{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; N</span>}}<span\; class="notranslate">\; ,</span>{\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}}_{\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; j</span>}}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>\underset{\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; no</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\frac{{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; i</span>}}\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; d</span>}}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; j</span>}}\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; d</span>}}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; d</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; ></span>$

in, Represents the similarity matrix Wth row number elements of the column, Indicates the dimensionality of the image block feature vector, Indicates the sample First dimension component, Indicates the sample First dimension component, is the first dimension hyperparameters, $1\&le;\tilde{i}\&le;N,1\&le;\tilde{j}\&le;N,1\&le;\tilde{d}\&le;\stackrel{\&OverBar;}{\mathrm{no}},\stackrel{\&OverBar;}{\mathrm{no}}=10;$

3b3) Construct the energy function of graph G according to the formula <3>:

$\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span><span\; class="notranslate">\; \&infin;</span>\underset{\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; i</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}\underset{\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; j</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}}_{\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; j</span>}}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; ></span>$

Among them, E(f ^s ) represents the energy function of graph G, Represents the real function of labeled samples on the graph G, Represents the real function of unlabeled samples on the graph G, express function value, Represents the similarity matrix W of the graph G row number elements of the column, $1\&le;\tilde{i}\&le;N,1\&le;\tilde{j}\&le;N;$

3b4) Solve the optimal solution argminE(f ^s ) of the energy function E(f ^s ), and obtain the classification results, that is, the final classification results of the 2828 image blocks in the image library are 1143 mountain images, 748 ocean images, and 344 urban images. There are 593 suburban images, with a total of 2828 SAR image blocks. The goal of this step is to change the SAR image library into a SAR image library with class labels, so as to reduce the workload of similarity matching and improve the recognition speed.

Step 4, classify the query image block p' input by the user.

The specific implementation of this step is as follows:

4a) using the same method as step 2 to extract the feature vector f' of the query image block p';

4b) Using the same training samples and Gaussian random field semi-supervised classifier as in step 3, obtain the category number c _i of the query image block p′.

Step 5, in order to reduce the impact of classification errors on the recognition results, calculate the category set {c} of the query image block p′ according to the number of categories c _i and the empirical confusion matrix obtained in step 4.

The specific implementation of this step is as follows:

5a) From the classified SAR image library, select the first 100 image blocks of each category to form a new image sample set, randomly select training samples from this sample set to train a Gaussian random field semi-supervised classifier, use the classifier Carry out 100 random classification experiments, in which the ratio of training samples to test samples is 1:99, so as to obtain the empirical confusion matrix Con∈R ^k×k , the confusion matrix Con is a square matrix, where row i and column j Con(i, j) represents the number of samples belonging to class c _i divided into class c _j , 1≤i≤k, 1≤j≤k, k represents the number of semantic categories, R represents the real number field, Table 1 gives the empirical confusion An example of the matrix Con∈R ^k×k , where the number of semantic categories k=4, which are mountainous, oceanic, urban, and suburban respectively, and each value represents the number of samples belonging to class c _i divided into class c _j , for example Con(1,1) indicates that the number of samples belonging to mountainous areas is divided into 41;

Table 1 Confusion matrix obtained from 100 random classification experiments

Con∈R ^k×k mountains ocean urban area suburbs mountains 41 20 0 38 ocean 6 93 0 0 urban area 0 0 97 2 suburbs 26 0 13 60

5b) Perform column normalization on the empirical confusion matrix Con∈R ^k×k , that is, divide each element of one column by the sum of the elements in this column to obtain the empirical posterior probability matrix ConP∈R ^k×k . Table 2 gives an example of the posterior probability matrix ConP∈R ^k×k , where the number of semantic categories k=4, which are mountainous, oceanic, urban, and suburban respectively, and each value is the empirical confusion matrix given in Table 1 Con∈R ^k×k is obtained by column normalization;

Table 2 Posterior probability matrix of experience

ConP∈Rk ^×k mountains ocean urban area suburbs mountains 0.5616 0.1770 0 0.3800 ocean 0.0822 0.8230 0 0 urban area 0 0 0.8818 0.0200 suburbs 0.3562 0 0.1182 0.6000

5c) Set the threshold T, compare the i-th row ConP(i,j) in the j-th row ConP(i,j) in the posterior probability matrix ConP∈R ^k×k with the threshold T, when ConP(i,j)≤T, set ConP (i,j) is set to 0, otherwise ConP(i,j) remains unchanged, the size of the threshold T is set according to the expected number of non-zero elements in each column, this example sets T=0.1, Table 3 gives An example of the posterior probability matrix ConP∈R ^k×k after threshold processing, where the number of semantic categories k=4, which are mountainous, oceanic, urban, and suburban respectively, and each value is the posterior probability matrix given in Table 2 ConP∈R ^k×k is obtained by thresholding;

Table 3 Posterior probability matrix after thresholding

ConP∈Rk ^×k mountains ocean urban area suburbs mountains 0.5616 0.1770 0 0.3800 ocean 0 0.8230 0 0 urban area 0 0 0.8818 0 suburbs 0.3562 0 0.1182 0.6000

5d) According to the number of categories c _i of the query image block, find the position of the non-zero element in the i-th column of the posterior probability matrix ConP, and finally obtain the category set {c}.

Step 6, calculate the area distance between the query image block p' and all image blocks belonging to the category set {c} in the gallery.

The specific implementation of this step is as follows:

6a) For the image block p and the query image block p′, calculate the high-frequency sub-band energy of the discrete wavelet one-layer transform with a 4×4 area size as the unit and grayscale features and use As the segmentation features, the adaptive k-means algorithm is used to cluster the segmentation features to obtain the texture region set R1={r ₁ ,r ₂ ,…,r _h ,…r _m } of the image block p and the query image block p ′ texture region set R ₂ ={r′ ₁ ,r′ ₂ ,…,r′ _o ,…r′ _n }, r _h , r′ _o respectively denote image block p and query image block p′ using texture feature segmentation After each region, wherein 1≤h≤m, 1≤o≤n, m represents the number of texture regions of the image block p, n represents the number of texture regions of the query image block p';

In order to make the k-means algorithm work adaptively, this example uses the following two strategies to complete the adaptive k-means segmentation:

6a1) Set a threshold T _num =0.035 for the divergence D(num), where D(num) is defined as follows:

$\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; num</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; L</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; \&le;</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; num</span>}}{\mathrm{<span\; class="notranslate">\; min</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; -</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}}_{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; j</span>}}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; \&le;</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; \&le;</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; num</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; ></span>$

In the formula, num represents the number of clusters, L represents the number of segmentation features, Represents a segmentation feature vector, Indicates the clustering center of a certain class. The value of num increases from 2. When D(num)<T _num , num stops increasing, and the number of clusters num is determined. Otherwise, the value of num increases by 1;

6a2) Set the upper limit num _max =5 for the number of clusters num, when num>num _max , num stops increasing, and the number of clusters num is determined, otherwise the value of num increases by 1;

6b) Calculate the distance d(r _h , r′ _o ) between the image block p and the texture area of the query image block p′ according to the formula <5>,

in, is the average feature vector of the texture region r _h in the image block p, is the average feature vector of the texture region r' _o in the query image patch p', is the weight coefficient of each vector, where and

6c) For the image block p and the query image block p', use the Prewitt operator and the binary segmentation method to obtain the edge region set RE ₁ ={re _v } of the image block p and the edge region set RE ₂ = of the query image block p'{re' _z }, rev _and re' _z represent the regions of image block p and query image block p' segmented by edge features respectively, where 1≤v≤2, 1≤z≤2;

6c1) Set the Prewitt operator:

Wherein λ=4, g ^y represents the edge detection operator in the vertical direction, and g ^x represents the edge detection operator in the horizontal direction respectively;

6c2) According to the Prewitt operator, calculate the edge information in the vertical direction and the edge information in the horizontal direction of the image block p respectively:

${\mathrm{<span\; class="notranslate">\; G</span>}}^{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; \&CircleTimes;</span>{\mathrm{<span\; class="notranslate">\; g</span>}}^{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; G</span>}}^{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; \&CircleTimes;</span>{\mathrm{<span\; class="notranslate">\; g</span>}}^{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; 7</span>}<span\; class="notranslate">\; ></span>$

Among them, G ^y represents the edge information in the vertical direction of the image block p, G ^x represents the edge information in the horizontal direction of the image block p, Represents linear convolution;

6c3) Calculate the edge feature f ^E of the image block p according to the edge information G ^y in the vertical direction and the edge information G ^x in the horizontal direction of the image block p:

${\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; E.</span>}}<span\; class="notranslate">\; =</span>\sqrt{{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; G</span>}}^{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; G</span>}}^{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; 8</span>}<span\; class="notranslate">\; ></span>$

Using the binary segmentation method to segment the edge feature f ^E to obtain the edge region set;

6d) Calculate the distance d(re _v , re' _z ) between the image block p and the edge area of the query image block p' according to the formula <9>:

$\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; re</span>}}_{\mathrm{<span\; class="notranslate">\; v</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; re</span>}}_{\mathrm{<span\; class="notranslate">\; z</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; kk</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&omega;e</span>}}_{\mathrm{<span\; class="notranslate">\; kk</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; fe</span>}}_{\mathrm{<span\; class="notranslate">\; kk</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; fe</span>}}_{\mathrm{<span\; class="notranslate">\; kk</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; 9</span>}<span\; class="notranslate">\; ></span>$

Among them, kk=1,2, when kk=1, fe _kk is the mean value feature of the edge region _rev in the image block p, fe′ _kk is the mean value feature of the edge region re′ _z in the query image block p′; when kk = 2, fe _kk is the variance feature of the edge region _rev in the image block p, _fe'kk is the variance feature of the edge region _re'z in the query image block p', ωe _kk is the weight coefficient of each vector, and $\underset{\mathrm{kk}=1}{\overset{2}{\mathrm{\&Sigma;}}}{\mathrm{\&omega;e}}_{\mathrm{kk}}=1;$

6e) According to the formula <10>, calculate the significance factor s _h,o matching between the texture regions of the image block p and the query image block p′,

$\begin{array}{c}\underset{\mathrm{<span\; class="notranslate">\; o</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; o</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; m</span>}\\ \underset{\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; o</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; o</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}\end{array}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; ></span>$

Among them, P _h is the percentage of the texture area r _h in the image block p to the area of the image block, P' _o is the percentage of the texture area r' _o in the query image block p' to the area of the query image block, and m represents the texture of the image block p The number of regions, n represents the number of texture regions of the query image block p';

6f) According to the formula <11>, calculate the significance factor s _v,z matching between the edge regions of the image block p and the query image block p′,

$\begin{array}{c}\underset{\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; z</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; v</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 2</span>}\\ \underset{\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; z</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; z</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 2</span>}\end{array}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; 11</span>}<span\; class="notranslate">\; ></span>$

Among them, P _v is the area percentage of the edge region rev in the image block p, and P' _z is the area percentage of the edge region re' _z in the query image _block p' to the query image block;

6g) According to the significance factor s _h,o of matching between texture regions, the significance factor s _v,z of matching between edge regions, the distance d(r _h ,r′ _o ) between texture regions, the distance between edge regions The distance d(re _v , re′ _z ), according to the formula <12>, the distance d between the query image block p′ and the image block p is obtained,

d=ω ₁ ×dT(R ₁ ,R ₂ )+ω ₂ ×dE(RE ₁ ,RE ₂ ),ω ₁ +ω ₂ =1 <12>

in, Indicates the distance between the image block p and the query image block p′ based on the texture area, 1≤h≤m, 1≤o≤n; Indicates the distance between image block p and query image block p′ based on the edge region, 1≤v≤2, 1≤z≤2; ω ₁ represents the weight of dT(R ₁ , R ₂ ), ω ₂ represents dE(RE ₁ , RE ₂ ) weights;

6h) According to the method of steps 6a) to 6g), calculate the area distance between the query image block p' and all image blocks belonging to the category set {c} in the gallery.

In step 7, according to the area distance obtained in step 6, the number of images required by the user is returned in ascending order to complete the image recognition.

Effect of the present invention can be further illustrated by following simulation:

1. Simulation conditions

The simulation conditions of this example are as follows: CORE i5 3.2GHz PC Windows 7 operating system, Matlab2012 operating platform.

2. Simulation content and results

Simulation 1. Randomly select 100 SAR image blocks from the image library, use the present invention and the GMM retrieval system to carry out recognition simulation on the selected SAR image blocks respectively, and calculate the overall average precision of the simulation results, the results are as shown in Figure 4. Among them, the definition of precision is as follows:

precision=n _c /n _s <13>

In the formula, n _c represents the number of image blocks that meet the conditions among the image blocks returned by the system, and n _s represents the number of image blocks returned by the system. In Figure 4, the horizontal axis represents the number of image blocks returned by the method, and the vertical axis represents the average precision rate.

It can be seen from FIG. 4 that the overall average recognition accuracy of the present invention is better than that of the GMM retrieval system.

Simulation 2. Randomly select 100 SAR image blocks from the image library, use the present invention and the GMM retrieval system to carry out recognition simulation on the selected SAR image blocks respectively, and calculate the precision rate of each semantic category in the simulation results, the results are as shown in the figure 5, where the number of returned image blocks is required to be 35. The horizontal axis represents each semantic category, and the vertical axis represents the precision rate.

It can be seen from FIG. 5 that for each semantic category, the recognition accuracy of the present invention is better than that of the GMM retrieval system.

It can be shown from the above experiments that the present invention is superior to the existing GMM retrieval system in terms of the overall precision rate and the precision rate of each semantic category in terms of SAR image recognition.

Claims (6)

1. A SAR image recognition method based on semi-supervised classification and regional distance measure, comprising the steps: 1) Segment the original SAR image without overlapping to establish a SAR image library {p ₁ ,p ₂ ,…,p _N }, and select image blocks {p ₁ ,p ₂ ,… , p _l }, where l<<N, N represents the number of SAR image blocks in the gallery, l represents the number of selected SAR image blocks, and the principle of single target refers to the percentage of the total image area occupied by a certain target in the image block more than half; 2) Extract the subband energy of the discrete wavelet three-layer transform of all image blocks as the feature vector of the image block in, $\stackrel{\&OverBar;}{\mathrm{no}}=10;$ 3) Divide the selected SAR image blocks {p ₁ ,p ₂ ,…,p _l } into { _ci ,1≤i≤k} categories according to the semantic content, where k represents the number of semantic categories, and use the corresponding feature The vector is used as a training sample to train a Gaussian random field semi-supervised classifier, and use the classifier to classify the entire SAR image library {p ₁ ,p ₂ ,…,p _N } to obtain a SAR image library with class labels; 4) For the query image block p' input by the user, use the same method as step 2) to extract its feature vector f', and use the same training samples as step 3) and the trained Gaussian random field semi-supervised classifier to obtain the query The category number c _i of the image block; 5) According to the number of categories c _i and the empirical confusion matrix obtained in step 4), calculate the category set {c} of the query image block: 5a) In the classified SAR image library, select the first K image blocks of each category to form a new image sample set, randomly select training samples from this sample set to train a Gaussian random field semi-supervised classifier, use the classifier Conduct 100 random classification experiments, and get the empirical confusion matrix Con∈R ^k×k , the confusion matrix Con is a square matrix, where row i and column j Con(i,j) indicates that samples belonging to class c _i are divided into c _j The number of classes, 1≤i≤k, 1≤j≤k; 5b) Carrying out column normalization to the empirical confusion matrix, that is, dividing each element in a column by the sum of the elements in the column to obtain the empirical posterior probability matrix ConP∈R ^k×k ; 5c) Set the threshold T, compare ConP(i,j) in the i-th row and column j in the posterior probability matrix with the threshold T, and when ConP(i,j)≤T, set ConP(i,j) is 0, otherwise ConP(i,j) remains unchanged, and the size of the threshold T is set according to the expected number of non-zero elements in each column; 5d) According to the category number c _i of the query image block, find the position of the non-zero element in the i-th column in the posterior probability matrix ConP, and finally obtain the category set {c}; 6) Calculate the area distance between the query image block p' and all image blocks belonging to the category set {c} in the gallery; 7) According to the area distance obtained in step 6), the number of images required by the user is returned in order from small to large, and image recognition is completed. 2. The SAR image recognition method according to claim 1, wherein step 6) the described calculation query image block p ' and the regional distance belonging to all image blocks in the category collection {c} in the gallery are carried out as follows: 6a) For the image block p and the query image block p′, calculate the high-frequency sub-band energy of the discrete wavelet one-layer transform with a 4×4 area size as the unit and grayscale features and use As the segmentation features, the adaptive k-means algorithm is used to cluster the segmentation features to obtain the texture region set R ₁ ={r ₁ ,r ₂ ,…,r _h ,…r _m } of the image block p and the query image block The texture region set R ₂ of p′ ={r′ ₁ ,r′ ₂ ,…,r′ _o ,…r′ _n }, r _h , r′ _o represent image block p and query image block p′ using texture features Each region after segmentation, wherein 1≤h≤m, 1≤o≤n, m represents the number of texture regions of the image block p, and n represents the number of texture regions of the query image block p'; 6b) Calculate the distance d(r _h ,r' _o ) between the image block p and the texture area of the query image block p'; 6c) For the image block p and the query image block p', use the Prewitt operator and the binary segmentation method to obtain the edge region set RE ₁ ={re _v } of the image block p and the edge region set RE ₂ = of the query image block p'{re' _z }, rev _and re' _z represent the regions of image block p and query image block p' segmented by edge features respectively, where 1≤v≤2, 1≤z≤2; 6d) Calculate the distance d(re _v ,re' _z ) between the image block p and the edge area of the query image block p'; 6e) Calculate the significance factor s _h,o of the matching between the image block p and the texture regions of the query image block p′, and the significance factor s _{v, z} of the matching between the image block p and the edge regions of the query image block p′ ; 6f) According to the significance factor s _h,o of matching between texture regions, the significance factor s _v,z of matching between edge regions, the distance d(r _h ,r′ _o ) between texture regions, the distance between edge regions The distance d(re _v , re′ _z ) of the query image block p′ and the image block p is the area distance d: d=ω ₁ ×dT(R ₁ ,R ₂ )+ω ₂ ×dE(RE ₁ ,RE ₂ ), ω ₁ +ω ₂ =1, in, Indicates the distance between the image block p and the query image block p′ based on the texture area, 1≤h≤m, 1≤o≤n; Indicates the distance between image block p and query image block p′ based on the edge region, 1≤v≤2, 1≤z≤2; ω ₁ represents the weight of dT(R ₁ , R ₂ ), ω ₂ represents dE(RE ₁ , RE ₂ ) weights; 6g) According to the method of steps 6a) to 6f), calculate the area distance between the query image block p' and all image blocks belonging to the category set {c} in the gallery. 3. The SAR image recognition method according to claim 2, wherein the distance d(r _h , r' _o ) between the image block p and the query image block p' texture area is calculated in the step 6b), by the following formula calculate: in, is the average feature vector of the texture region r _h in the image block p, is the average feature vector of the texture region r' _o in the query image patch p', is the weight coefficient of each vector, where and 4. The SAR image recognition method according to claim 2, wherein in said step 6d), the distance d(re _v , re' _z ) between the image block p and the query image block p' edge region is calculated by the following formula : $\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; re</span>}}_{\mathrm{<span\; class="notranslate">\; v</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; re</span>}}_{\mathrm{<span\; class="notranslate">\; z</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; kk</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&omega;e</span>}}_{\mathrm{<span\; class="notranslate">\; kk</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; fe</span>}}_{\mathrm{<span\; class="notranslate">\; kk</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; fe</span>}}_{\mathrm{<span\; class="notranslate">\; kk</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ,</span>$ Among them, kk=1,2, when kk=1, fe _kk is the mean value feature of the edge region _rev in the image block p, fe′ _kk is the mean value feature of the edge region re′ _z in the query image block p′; when kk = 2, fe _kk is the variance feature of the edge region _rev in the image block p, _fe'kk is the variance feature of the edge region _re'z in the query image block p', ωe _kk is the weight coefficient of each vector, and 5. The SAR image recognition method according to claim 2, wherein the step 6e) calculates the significance factor _sh,o matching between the texture regions of the image block p and the query image block p', calculated by the following formula : $\begin{array}{c}\underset{\mathrm{<span\; class="notranslate">\; o</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; o</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; m</span>}\\ \underset{\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; o</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; o</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}\end{array}<span\; class="notranslate">\; ,</span>$ Among them, P _h is the percentage of the texture area r _h in the image block p to the area of the image block, P' _o is the percentage of the texture area r' _o in the query image block p' to the area of the query image block, and m represents the texture of the image block p The number of regions, n represents the number of texture regions of the query image block p'. 6. The SAR image recognition method according to claim 2, wherein in the step 6e), the significance factor s _{v, z} matching between the image block p and the query image block p' is calculated respectively, by the following formula calculate: $\begin{array}{c}\underset{\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; z</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; v</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 2</span>}\\ \underset{\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; z</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; z</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 2</span>}\end{array}<span\; class="notranslate">\; ,</span>$ Among them, P _v is the area percentage of the edge region _rev in the image block p to the image block, and P' _z is the area percentage of the edge region re' _z in the query image block p' to the query image block.