CN104680184A - Polarization SAR terrain classification method based on deep RPCA - Google Patents
Polarization SAR terrain classification method based on deep RPCA Download PDFInfo
- Publication number
- CN104680184A CN104680184A CN201510112514.0A CN201510112514A CN104680184A CN 104680184 A CN104680184 A CN 104680184A CN 201510112514 A CN201510112514 A CN 201510112514A CN 104680184 A CN104680184 A CN 104680184A
- Authority
- CN
- China
- Prior art keywords
- pixel
- rpca
- principal component
- component analysis
- depth
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Abstract
The invention discloses a polarization SAR terrain classification method based on deep RPCA. The polarization SAR terrain classification method comprises the following steps: (1) reading a polarization SAR image; (2) filtering; (3) extracting characteristics; (4) normalizing characteristic groups; (5) selecting a training sample and a testing sample; (6) training a first layer of deep robust principal component analysis RPCA; (7) training a second layer of deep robust principal component analysis RPCA; (8) training a support vector machine SVM; (9) generating superpixels; (10) classifying; (11) calculating classification precision; (12) outputting a result. Compared with scattering characteristics of the polarization SAR image, the image characteristics extracted according to the method comprise relatively rich terrain information, when the terrain information is classified, the classification precision is effectively improved, and the polarization SAR terrain classification method can be applied to detection and recognition of polarization SAR image targets.
Description
Technical field
The invention belongs to technical field of image processing, further relate to a kind of polarization SAR terrain classification method based on degree of depth RPCA in Image Classfication Technology field.The method can be applicable to target detection to Polarimetric SAR Image and target identification, and effectively improves Classification of Polarimetric SAR Image accuracy.
Background technology
Polarimetric SAR Image describes by transmitting and receiving polarimetric radar ripple the land cover pattern thing and target observed, is one of sensor that remote sensing fields is the most advanced in recent years.Polarimetric SAR Image, as a kind of important remote sensing images obtaining means, has a wide range of applications in agriculture and forestry, military affairs, ocean, hydrology and geology etc.The object of Classification of Polarimetric SAR Image is the classification that the polarization measurement data utilizing airborne or borne polarization sensor to obtain determine belonging to each pixel.
Wuhan University discloses a kind of polarization SAR data classification method based on hybrid classifer in its patented claim " polarization SAR data classification method and system based on hybrid classifer " (number of patent application: 201310310179, publication number: CN103366184A).First the method obtains the initial polarization feature of the different class of Polarimetric SAR Image; Then, employing decision tree classifier selects the polarization characteristic for classifying from initial polarization feature; Finally, support vector machine classifier is adopted to classify to polarimetric SAR image data.Although the method is integrated with the advantage of decision tree classifier and support vector machine classifier.But the deficiency that the method still exists is, complicated operation, and owing to only considered the scattering signatures of the single pixel of Polarimetric SAR Image, thus be vulnerable to noise interference and can not the actual atural object of accurate characterization, cause in classification results, there is more wrong branch.
A kind of Classification of Polarimetric SAR Image method based on SDIT and SVM is proposed in patent " the Classification of Polarimetric SAR Image method based on SDIT and SVM " (number of patent application: 201410089692.1, the publication number: CN103824084A) of Xian Electronics Science and Technology University's application.First the method carries out exquisite Lee filtering to Polarimetric SAR Image to be sorted; Then, extract the scattering of image, polarization and textural characteristics, the feature obtained is carried out combination normalization; Finally with the features training sorter after normalization, prediction is classified and calculates nicety of grading.Although the method takes full advantage of abundant texture information and polarization information that Polarimetric SAR Image comprises, compensate for and only rely on scattering properties to the deficiency of Classification of Polarimetric SAR Image, improve the precision of Classification of Polarimetric SAR Image to a certain extent.But, the deficiency that the method still exists is, merely extracted scattering, polarization and texture information are carried out simply stacking, then directly input in support vector machines and classify, cause thus comprising more redundant information in the feature inputted, fail the essential characteristic of effecting reaction Polarimetric SAR Image, classification effectiveness is declined greatly.
Summary of the invention
The object of the invention is to the deficiency overcoming above-mentioned prior art, propose a kind of polarization SAR terrain classification method based on degree of depth RPCA.The present invention makes full use of the power of the pixel of Polarimetric SAR Image, data distribution characteristics parameter, relative peak, scattering signatures parameter, polarization characteristic and neighborhood information feature, by degree of depth Robust Principal Component Analysis RPCA, obtain the second order feature of sample, then pixel is replaced to classify to super-pixel with super-pixel, improve the validity of feature, improve nicety of grading and classification effectiveness.
The thinking realizing above-mentioned purpose of the present invention is: carry out exquisite Lee filtering to Polarimetric SAR Image to be sorted, the power of the pixel of extraction filtered image, data distribution characteristics parameter, relative peak, scattering signatures parameter, polarization characteristic, neighborhood information Feature Combination become feature group, using the original input data of feature group normalization result as degree of depth Robust Principal Component Analysis RPCA; Select training sample and test sample book; With training sample training degree of depth Robust Principal Component Analysis RPCA, obtain the second order feature of training sample; With the second order features training support vector machines of training sample; From filtered Polarimetric SAR Image, generate super-pixel, and with the support vector machines trained, super-pixel is classified; Carry out colouring to sorted Polarimetric SAR Image to export.
Step of the present invention comprises as follows:
(1) Polarimetric SAR Image to be sorted is read in;
(2) filtering:
Adopt exquisite polarization LEE filter method, filtering is carried out to all pixels in Polarimetric SAR Image, obtains the coherence matrix of filtered Polarimetric SAR Image pixel;
(3) feature is extracted:
(3a) from the coherence matrix of the pixel of filtered Polarimetric SAR Image, the power of each pixel, data distribution characteristics parameter and relative peak is extracted;
(3b) utilize Pauli Pauli decomposition method, 3 scattering signatures parameters characterizing Pauli Pauli and decompose are extracted to each pixel;
(3c) utilize freeman-Freeman-Durden decomposition method must be stepped on, 9 scattering signatures parameters characterizing freeman-must step on Freeman-Durden and decompose are extracted to each pixel;
(3d) utilize carat DS Cloude decomposition method, 6 scattering signatures parameters characterizing carat DS Cloude and decompose are extracted to each pixel;
(3e) utilize intelligent energy Huynen decomposition method, 9 scattering signatures parameters characterizing intelligent energy Huynen and decompose are extracted to each pixel;
(3f) utilize Crow to end Krogager decomposition method in distress, 3 are extracted to each pixel and characterizes Crows and to end the scattering signatures parameter that Krogager in distress decomposes;
(3g) 3 that each pixel are extracted characterize scattering signatures parameter that Pauli Pauli decompose, 9 characterize scattering signatures parameter that freeman-must step on Freeman-Durden decomposes, 6 characterize scattering signatures parameter that carat DS Cloude decomposes, 9 characterize the intelligent Huynen scattering signatures parameter of decomposing and 3 can sign Crow and to end the scattering signatures parameter that Krogager in distress decomposes, form one the 30 scattering signatures parameter tieed up;
(3h) from the coherence matrix of filtered Polarimetric SAR Image pixel, the polarization characteristic of each pixel is extracted;
(3k) from each pixel of filtered Polarimetric SAR Image and the coherence matrix of 8 neighborhood territory pixel points of this some correspondence, extract the value of real part, imaginary values and the mould that are arranged in three elements at the upper triangle place of coherence matrix, and in coherence matrix, be positioned at the value of real part of three elements on diagonal line; By the value of real part of three elements on the extracted value of real part being arranged in three elements at the upper triangle place of coherence matrix, imaginary values and mould and diagonal line, as the neighborhood information feature of pixel;
(4) feature group normalization:
(4a) power of pixel, data distribution characteristics parameter, relative peak, scattering signatures parameter, polarization characteristic, neighborhood information Feature Combination are become feature group;
(4b) feature group is normalized to the numerical value between 0 ~ 1, obtains normalized feature group;
(4c) by normalized feature group, as the original input data of degree of depth Robust Principal Component Analysis RPCA;
(5) training sample and test sample book is selected:
According to the substance markers truly of Polarimetric SAR Image, respectively from the markd original input data of each atural object classification random selecting 5% as training sample, the markd original input data of remaining 95% is as test sample book;
(6) degree of depth Robust Principal Component Analysis RPCA ground floor is trained:
(6a) utilize matlab software, a training sample is converted to square formation;
(6b) according to order from top to bottom, from left to right, get vector block successively to each element in square formation, the size of vector block is set to 3 × 3 pixels;
(6c) respectively each vector block is averaged, deduct average by the data in vector block, the result obtained is removed result as the average of this vector block;
(6d) judge whether the vector block of all training samples all carries out average and remove, and if so, then performs step (6e), otherwise, perform step (6a);
(6e) remove result to the average of all training samples to sort top to bottom, obtain the matrix after sorting;
(6f) to the matrix after sequence, the method for approaching with low-rank, solves the low-rank part of the rear matrix of sequence;
(6g) with constructive formula, the Robust Principal Component Analysis wave filter of construction depth Robust Principal Component Analysis RPCA ground floor;
(6h) two-dimensional convolution is carried out, using the single order feature of the result of convolution as training sample with the Robust Principal Component Analysis wave filter of degree of depth Robust Principal Component Analysis RPCA ground floor and training sample;
(7) degree of depth Robust Principal Component Analysis RPCA second layer is trained:
(7a) utilizing matlab software, is square formation by the single order Feature Conversion of a training sample;
(7b) according to order from top to bottom, from left to right, get vector block successively to each element in square formation, the size of vector block is set to 3 × 3 pixels;
(7c) respectively each vector block is averaged, deduct average by the data in vector block, the result obtained is removed result as the average of this vector block;
(7d) judge whether the vector block of the single order feature of all training samples all carries out average and remove, and if so, then performs step (7e), otherwise, perform step (7a);
(7e) remove result to the average of the single order feature of all training samples to sort top to bottom, obtain the matrix after sorting;
(7f) to the matrix after sequence, the method for approaching with low-rank, solves the low-rank part of the rear matrix of sequence;
(7g) with constructive formula, the Robust Principal Component Analysis wave filter of the construction depth Robust Principal Component Analysis RPCA second layer;
(7h) two-dimensional convolution is carried out, using the second order feature of the result of convolution as training sample by the Robust Principal Component Analysis wave filter of the degree of depth Robust Principal Component Analysis RPCA second layer and the single order feature of training sample;
(8) with the second order features training support vector machines of training sample, the support vector machines trained is obtained;
(9) super-pixel is generated:
(9a) from filtered Polarimetric SAR Image, selected pixels point equally spacedly, as initial seed point;
(9b) initial seed point is collided, obtain super-pixel;
(10) classify:
(10a) adopt eigenwert learning method, eigenwert study is carried out to a super-pixel in degree of depth Robust Principal Component Analysis RPCA, obtains the eigenwert of super-pixel;
(10b) by the eigenwert of super-pixel input support vector machines, with the support vector machines after training, super-pixel is classified, obtain the classification of atural object belonging to super-pixel;
(10c) by atural object category division atural object classification belonging to this super-pixel belonging to all pixels in super-pixel, atural object classification belonging to all pixels in this super-pixel is obtained;
(10d) judge whether to obtain atural object classification belonging to all pixels in whole super-pixel, if so, perform step (11), otherwise, perform step (10a);
(11) nicety of grading is calculated:
(11a) atural object classification belonging to the test sample book pixel of Polarimetric SAR Image and true atural object classification are contrasted, by the ratio of the whole pixel number of this classification in pixel number consistent for classification and test sample book, as such other accuracy;
(11b) judge whether the accuracy obtaining whole classification, if so, perform step (11c), otherwise, perform step (11a);
(11c) using the nicety of grading of the ratio of pixel number whole in pixel number consistent for classification and test sample book as Polarimetric SAR Image;
(12) Output rusults:
According to the red, green, blue principle of three primary colours, to atural object classification belonging to each pixel, mark similar atural object by same color, obtain the Polarimetric SAR Image after painting and output image.
The present invention compared with prior art has the following advantages:
First, because the present invention adopts, the power of each pixel, data distribution parameter, relative peak, scattering signatures parameter, polarization characteristic, neighborhood information Feature Combination are become feature group, overcome prior art owing to only considering the scattering signatures of single pixel, therefore be vulnerable to noise interference and can not the actual atural object of accurate characterization, cause the problem that there is more wrong branch in classification results, the present invention can be made full use of scattering properties and statistical property that polarimetric SAR image data is different from other data, better characterizes the scattering mechanism of actual atural object.
Second, because the present invention adopts degree of depth Robust Principal Component Analysis RPCA to obtain the second order feature of sample, overcoming prior art is merely undertaken simply stacking by extracted scattering, polarization and texture information, then directly input in support vector machines and classify, cause thus comprising more redundant information in the feature inputted, fail the essential characteristic of effecting reaction Polarimetric SAR Image, make the problem that classification effectiveness declines greatly, make the present invention can the advanced features of learning data better, be more conducive to improving classification results and nicety of grading.
3rd, because the present invention adopts the technology of classifying to super-pixel, overcome the problem of the classification results homogeneous region consistance difference of prior art, make the spatial coherence that the present invention can keep in Polarimetric SAR Image between pixel and pixel, substantially increase nicety of grading, improve the classification performance of Polarimetric SAR Image.
Accompanying drawing explanation
Fig. 1 is process flow diagram of the present invention;
Fig. 2 is the Polarimetric SAR Image used in emulation experiment of the present invention;
Fig. 3 is the real atural object signature of Polarimetric SAR Image used in emulation experiment of the present invention;
Fig. 4 is the result figure adopting existing wishart sorting technique to classify to the Polarimetric SAR Image in Fig. 2;
Fig. 5 adopts support vector machines sorting technique to the result figure of the Classification of Polarimetric SAR Image in Fig. 2;
Fig. 6 is the result figure adopting the present invention to classify to the Polarimetric SAR Image in Fig. 2.
Embodiment
Below in conjunction with accompanying drawing, the present invention will be further described.
With reference to Fig. 1, specific embodiment of the invention step is as follows.
Step 1. reads in a Polarimetric SAR Image to be sorted.
Step 2. filtering.
Adopt exquisite polarization LEE filter method, carry out filtering to all pixels in Polarimetric SAR Image, the edge windows size of exquisite polarization LEE filter method is 3 × 3 pixels, obtains the coherence matrix of filtered Polarimetric SAR Image pixel.
Step 3. extracts feature.
First, from the coherence matrix of filtered Polarimetric SAR Image, the power of each pixel, data distribution characteristics parameter and relative peak is extracted.
Then, utilize Pauli Pauli decomposition method, 3 scattering signatures parameters characterizing Pauli Pauli and decompose are extracted to each pixel; Utilize freeman-Freeman-Durden decomposition method must be stepped on, 9 scattering signatures parameters characterizing freeman-must step on Freeman-Durden and decompose are extracted to each pixel; Utilize carat DS Cloude decomposition method, 6 scattering signatures parameters characterizing carat DS Cloude and decompose are extracted to each pixel; Utilize intelligent energy Huynen decomposition method, 9 scattering signatures parameters characterizing intelligent energy Huynen and decompose are extracted to each pixel; Utilize Crow to end Krogager decomposition method in distress, 3 are extracted to each pixel and characterizes Crows and to end the scattering signatures parameter that Krogager in distress decomposes.
Secondly, 3 that each pixel is extracted characterize scattering signatures parameter that Pauli Pauli decompose, 9 characterize scattering signatures parameter that freeman-must step on Freeman-Durden decomposes, 6 characterize scattering signatures parameter that carat DS Cloude decomposes, 9 characterize the intelligent Huynen scattering signatures parameter of decomposing and 3 can sign Crow and to end the scattering signatures parameter that Krogager in distress decomposes, form one the 30 scattering signatures parameter tieed up.
Again, from the coherence matrix of filtered Polarimetric SAR Image pixel, the polarization characteristic of each pixel is extracted.
Finally, from each pixel of filtered Polarimetric SAR Image and the coherence matrix of 8 neighborhood territory pixel points of this some correspondence, extract the value of real part, imaginary values and the mould that are arranged in three elements at the upper triangle place of coherence matrix, and in coherence matrix, be positioned at the value of real part of three elements on diagonal line; By the value of real part of three elements on the extracted value of real part being arranged in three elements at the upper triangle place of coherence matrix, imaginary values and mould and diagonal line, as the neighborhood information feature of pixel.
The normalization of step 4. feature group.
The power of pixel, data distribution characteristics parameter, relative peak, scattering signatures parameter, polarization characteristic, neighborhood information Feature Combination are become feature group.Then, feature group is normalized to the numerical value between 0 ~ 1, obtains normalized feature group.Finally, by normalized feature group, as the original input data of degree of depth Robust Principal Component Analysis RPCA.
Step 5. selects training sample and test sample book.
According to the substance markers truly of Polarimetric SAR Image, respectively from the markd original input data of each atural object classification random selecting 5% as training sample, the markd original input data of remaining 95% is as test sample book
Step 6. trains degree of depth Robust Principal Component Analysis RPCA ground floor.
The first step, utilizes matlab software, and a training sample is converted to square formation.
Second step, according to order from top to bottom, from left to right, gets vector block to each element in square formation successively, and the size of vector block is set to 3 × 3 pixels.
3rd step, averages to the vector block of each element, and then, deduct average by the data in vector block, the result obtained removes result as the average of vector block.
4th step, judges whether the vector block of all training samples all carries out average and remove, and if so, then performs step 6 the 5th step, otherwise, perform step 6 first step.
5th step, removes result to the average of all training samples and sorts top to bottom, and obtains the matrix after sorting.
6th step, utilizes matlab software, and to the matrix after sequence, the method for approaching with low-rank, solves the low-rank part of the rear matrix of sequence.
7th step, with constructive formula, the Robust Principal Component Analysis wave filter of construction depth Robust Principal Component Analysis RPCA ground floor, constructive formula is as follows:
Wherein,
represent l the Robust Principal Component Analysis wave filter of degree of depth Robust Principal Component Analysis RPCA i-th layer, i represents the number of plies of degree of depth Robust Principal Component Analysis RPCA, herein i=1, l=1,2 ..., L
i, L
irepresent the number of i-th layer of Robust Principal Component Analysis wave filter, mat () represents map operation
,q
l() represents the operation asking proper vector, X
irepresent the low-rank part of the matrix after sequence, T represents matrix transpose operation.
8th step, carries out two-dimensional convolution, using the single order feature of the result of convolution as training sample with the Robust Principal Component Analysis wave filter of degree of depth Robust Principal Component Analysis RPCA ground floor and training sample.
Step 7. trains the degree of depth Robust Principal Component Analysis RPCA second layer.
The first step, utilizes matlab software, is square formation by the single order Feature Conversion of a training sample.
Second step, according to order from top to bottom, from left to right, gets vector block to each element in square formation successively, and the size of vector block is set to 3 × 3 pixels.
3rd step, averages to the vector block of each element, and then, deduct average by the data in vector block, the result obtained removes result as the average of vector block.
4th step, judges whether the vector block of the single order feature of all training samples all carries out average and remove, and if so, then performs step 7 the 5th step, otherwise, perform step 7 first step.
5th step, removes result to the average of the single order feature of all training samples and sorts top to bottom, and obtains the matrix after sorting.
6th step, to the matrix after sequence, the method for approaching with low-rank, solves the low-rank part of the rear matrix of sequence.
7th step, with constructive formula, the Robust Principal Component Analysis wave filter of the construction depth Robust Principal Component Analysis RPCA second layer, constructive formula is as follows:
Wherein,
represent l the Robust Principal Component Analysis wave filter of degree of depth Robust Principal Component Analysis RPCA i-th layer, i represents the number of plies of degree of depth Robust Principal Component Analysis RPCA, herein i=2, l=1,2 ..., L
i, L
irepresent the number of i-th layer of Robust Principal Component Analysis wave filter, mat () represents map operation
,q
l() represents the operation asking proper vector, X
irepresent the low-rank part of the matrix after sequence, T represents matrix transpose operation.
8th step, carries out two-dimensional convolution, using the second order feature of the result of convolution as training sample by the Robust Principal Component Analysis wave filter of the degree of depth Robust Principal Component Analysis RPCA second layer and the single order feature of training sample.
The step 8. second order features training support vector machines of training sample, obtains the support vector machines trained.
Step 9. generates super-pixel.
From filtered Polarimetric SAR Image, selected pixels point equally spacedly, as initial seed point, in the present invention, spacing is set to 10
3individual pixel, by colliding initial seed point, obtains super-pixel.
Step 10. is classified.
The first step, adopts eigenwert learning method, carries out eigenwert study, obtain the eigenwert of super-pixel to a super-pixel in degree of depth Robust Principal Component Analysis RPCA.
First, by the original input data of all pixels of each super-pixel of composition, be input in degree of depth Robust Principal Component Analysis RPCA, obtain the second order eigenwert of all pixels; Then, the second order feature of all pixels is averaged, obtains the eigenwert of super-pixel.
Second step, by the eigenwert of super-pixel input support vector machines, classifies to super-pixel with the support vector machines after training, obtains the classification of atural object belonging to super-pixel.
3rd step, by atural object category division atural object classification belonging to this super-pixel belonging to all pixels in super-pixel, obtains atural object classification belonging to all pixels in this super-pixel.
4th step, judges whether to obtain atural object classification belonging to all pixels in whole super-pixel, if so, performs step 11, otherwise, perform step 10 first step.
Step 11. calculates nicety of grading.
The first step, contrasts atural object classification belonging to the test sample book pixel of Polarimetric SAR Image and true atural object classification, by the ratio of the whole pixel number of this classification in pixel number consistent for classification and test sample book, as such other accuracy.
Second step, judges whether the accuracy obtaining whole classification, if so, performs step 11 the 3rd step, otherwise, perform step 11 first step.
3rd step, using the nicety of grading of the ratio of pixel number whole in pixel number consistent for classification and test sample book as Polarimetric SAR Image.
Step 12. Output rusults.
According to the red, green, blue principle of three primary colours, to atural object classification belonging to each pixel, mark similar atural object by same color, obtain the Polarimetric SAR Image after painting and output image.
Effect of the present invention further illustrates by following emulation.
1, simulated conditions:
Emulation of the present invention carries out under Core (TM) the i5-4200M CPU E6550@2.50GHZ of dominant frequency 2.5GHZ, the hardware environment of 8GB RAM and the software environment of MATLAB R2013b.
2, content is emulated:
The present invention's emulation carries out classification experiments to the Polarimetric SAR Image in Fig. 2.Fig. 2 is Polarimetric SAR Image to be sorted in the embodiment of the present invention, and this figure is the L-band Polarimetric SAR Image in the Flevoland area, Dutch Fu Laifulan area that NASA lab A IRSAR system obtains.Emulation experiment is classified to Polarimetric SAR Image by 15 classes.Fig. 3 is the image of substance markers truly of Polarimetric SAR Image to be sorted in the embodiment of the present invention.
The present invention emulates content: emulation 1, adopts existing wishart sorting technique to classify, the results detailed in Fig. 4 to the Polarimetric SAR Image in Fig. 2.Emulation 2, adopts support vector machines sorting technique to classify, the results detailed in Fig. 5 to the Polarimetric SAR Image in Fig. 2.Emulation 3, classifies to Fig. 2 with the present invention, the results detailed in Fig. 6.
3 simulated effect analyses
Respectively Fig. 4, Fig. 5 and Fig. 6 and the present invention are tested Polarimetric SAR Image to be sorted used truly substance markers Fig. 3 contrast and can find out, Fig. 6 is compared to control methods Fig. 4 and Fig. 5, Fig. 6 is more close to Fig. 3 Polarimetric SAR Image to be sorted substance markers image truly, and the differentiation of all kinds of atural object is more careful in Fig. 6, accurately.
Using the polarization SAR to be sorted in Fig. 3 truly substance markers image as precision evaluation standard, add up the nicety of grading of the Wishart supervised classification method of prior art, the method adopting support vector machines to classify, sorting technique of the present invention, result is as table 1.
Using the polarization SAR to be sorted in Fig. 3 truly substance markers image as precision evaluation standard, add up the nicety of grading of the Wishart supervised classification method of prior art, the method adopting support vector machines to classify, sorting technique of the present invention, result is as table 1.As can be seen from Table 1, the present invention is based on the two kind contrast experiments of polarization SAR terrain classification method compared to prior art of degree of depth Robust Principal Component Analysis RPCA, precision is greatly improved, this is mainly because the feature extracted of the present invention is compared to the coherence matrix of Polarimetric SAR Image itself and simple scattering signatures, contain more abundant terrestrial object information, and feature redundancy is little, have more representative, be conducive to classification, nicety of grading is higher, and classify by introducing super-pixel, make the spatial coherence that the present invention can keep in Polarimetric SAR Image between pixel and pixel, substantially increase nicety of grading, improve the classification performance of Polarimetric SAR Image.
Table 1
Claims (5)
1., based on a polarization SAR terrain classification method of degree of depth RPCA, comprise the steps:
(1) Polarimetric SAR Image to be sorted is read in;
(2) filtering:
Adopt exquisite polarization LEE filter method, filtering is carried out to all pixels in Polarimetric SAR Image, obtains the coherence matrix of filtered Polarimetric SAR Image pixel;
(3) feature is extracted:
(3a) from the coherence matrix of the pixel of filtered Polarimetric SAR Image, the power of each pixel, data distribution characteristics parameter and relative peak is extracted;
(3b) utilize Pauli Pauli decomposition method, 3 scattering signatures parameters characterizing Pauli Pauli and decompose are extracted to each pixel;
(3c) utilize freeman-Freeman-Durden decomposition method must be stepped on, 9 scattering signatures parameters characterizing freeman-must step on Freeman-Durden and decompose are extracted to each pixel;
(3d) utilize carat DS Cloude decomposition method, 6 scattering signatures parameters characterizing carat DS Cloude and decompose are extracted to each pixel;
(3e) utilize intelligent energy Huynen decomposition method, 9 scattering signatures parameters characterizing intelligent energy Huynen and decompose are extracted to each pixel;
(3f) utilize Crow to end Krogager decomposition method in distress, 3 are extracted to each pixel and characterizes Crows and to end the scattering signatures parameter that Krogager in distress decomposes;
(3g) 3 that each pixel are extracted characterize scattering signatures parameter that Pauli Pauli decompose, 9 characterize scattering signatures parameter that freeman-must step on Freeman-Durden decomposes, 6 characterize scattering signatures parameter that carat DS Cloude decomposes, 9 characterize the intelligent Huynen scattering signatures parameter of decomposing and 3 can sign Crow and to end the scattering signatures parameter that Krogager in distress decomposes, form one the 30 scattering signatures parameter tieed up;
(3h) from the coherence matrix of filtered Polarimetric SAR Image pixel, the polarization characteristic of each pixel is extracted;
(3k) from each pixel of filtered Polarimetric SAR Image and the coherence matrix of 8 neighborhood territory pixel points of this some correspondence, extract the value of real part, imaginary values and the mould that are arranged in three elements at the upper triangle place of coherence matrix, and in coherence matrix, be positioned at the value of real part of three elements on diagonal line; By the value of real part of three elements on the extracted value of real part being arranged in three elements at the upper triangle place of coherence matrix, imaginary values and mould and diagonal line, as the neighborhood information feature of pixel;
(4) feature group normalization:
(4a) power of pixel, data distribution characteristics parameter, relative peak, scattering signatures parameter, polarization characteristic, neighborhood information Feature Combination are become feature group;
(4b) feature group is normalized to the numerical value between 0 ~ 1, obtains normalized feature group;
(4c) by normalized feature group, as the original input data of degree of depth Robust Principal Component Analysis RPCA;
(5) training sample and test sample book is selected:
According to the substance markers truly of Polarimetric SAR Image, respectively from the markd original input data of each atural object classification random selecting 5% as training sample, the markd original input data of remaining 95% is as test sample book;
(6) degree of depth Robust Principal Component Analysis RPCA ground floor is trained:
(6a) utilize matlab software, a training sample is converted to square formation;
(6b) according to order from top to bottom, from left to right, get vector block successively to each element in square formation, the size of vector block is set to 3 × 3 pixels;
(6c) respectively each vector block is averaged, deduct average by the data in vector block, the result obtained is removed result as the average of this vector block;
(6d) judge whether the vector block of all training samples all carries out average and remove, and if so, then performs step (6e), otherwise, perform step (6a);
(6e) remove result to the average of all training samples to sort top to bottom, obtain the matrix after sorting;
(6f) to the matrix after sequence, the method for approaching with low-rank, solves the low-rank part of the rear matrix of sequence;
(6g) with constructive formula, the Robust Principal Component Analysis wave filter of construction depth Robust Principal Component Analysis RPCA ground floor;
(6h) two-dimensional convolution is carried out, using the single order feature of the result of convolution as training sample with the Robust Principal Component Analysis wave filter of degree of depth Robust Principal Component Analysis RPCA ground floor and training sample;
(7) degree of depth Robust Principal Component Analysis RPCA second layer is trained:
(7a) utilizing matlab software, is square formation by the single order Feature Conversion of a training sample;
(7b) according to order from top to bottom, from left to right, get vector block successively to each element in square formation, the size of vector block is set to 3 × 3 pixels;
(7c) respectively each vector block is averaged, deduct average by the data in vector block, the result obtained is removed result as the average of this vector block;
(7d) judge whether the vector block of the single order feature of all training samples all carries out average and remove, and if so, then performs step (7e), otherwise, perform step (7a);
(7e) remove result to the single order characteristic mean of all training samples to sort top to bottom, matrix after obtaining sorting;
(7f) to the matrix after sequence, the method for approaching with low-rank, solves the low-rank part of the rear matrix of sequence;
(7g) with constructive formula, the Robust Principal Component Analysis wave filter of the construction depth Robust Principal Component Analysis RPCA second layer;
(7h) by degree of depth robust major component, the single order feature of the Robust Principal Component Analysis wave filter and training sample of analyzing the RPCA second layer carries out two-dimensional convolution, using the second order feature of the result of convolution as training sample;
(8) with the second order features training support vector machines of training sample, the support vector machines trained is obtained;
(9) super-pixel is generated:
(9a) from filtered Polarimetric SAR Image, selected pixels point equally spacedly, as initial seed point;
(9b) initial seed point is collided, obtain super-pixel;
(10) classify:
(10a) adopt eigenwert learning method, eigenwert study is carried out to a super-pixel in degree of depth Robust Principal Component Analysis RPCA, obtains the eigenwert of super-pixel;
(10b) by the eigenwert of super-pixel input support vector machines, with the support vector machines after training, super-pixel is classified, obtain the classification of atural object belonging to super-pixel;
(10c) by atural object category division atural object classification belonging to this super-pixel belonging to all pixels in super-pixel, atural object classification belonging to all pixels in this super-pixel is obtained;
(10d) judge whether to obtain atural object classification belonging to all pixels in whole super-pixel, if so, perform step (11), otherwise, perform step (10a);
(11) nicety of grading is calculated:
(11a) atural object classification belonging to the test sample book pixel of Polarimetric SAR Image and true atural object classification are contrasted, by the ratio of the whole pixel number of this classification in pixel number consistent for classification and test sample book, as such other accuracy;
(11b) judge whether the accuracy obtaining whole classification, if so, perform step (11c), otherwise, perform step (11a);
(11c) using the nicety of grading of the ratio of pixel number whole in pixel number consistent for classification and test sample book as Polarimetric SAR Image;
(12) Output rusults:
According to the red, green, blue principle of three primary colours, to atural object classification belonging to each pixel, mark similar atural object by same color, obtain the Polarimetric SAR Image after painting and output image.
2. the polarization SAR terrain classification method based on degree of depth RPCA according to claim 1, is characterized in that: the edge windows size of the exquisiteness polarization LEE filter method described in step (2) is 3 × 3 pixels.
3. the polarization SAR terrain classification method based on degree of depth RPCA according to claim 1, is characterized in that: step (6g), constructive formula described in step (7g) are as follows:
W
l i＝mat(q
l(X
iX
i T)).
Wherein, W
l irepresent l the Robust Principal Component Analysis wave filter of degree of depth Robust Principal Component Analysis RPCA i-th layer, i represents the number of plies of degree of depth Robust Principal Component Analysis RPCA, i=1,2, l=1,2 ..., L
i, L
irepresent the number of i-th layer of Robust Principal Component Analysis wave filter, mat () represents map operation
,q
l() represents the operation asking proper vector, X
irepresent the low-rank part of the matrix after sequence, T represents matrix transpose operation.
4. the polarization SAR terrain classification method based on degree of depth RPCA according to claim 1, is characterized in that: the spacing described in step (9a) refers to [10
2, 10
3] individual pixel span in a numerical value choosing arbitrarily.
5. the polarization SAR terrain classification method based on degree of depth RPCA according to claim 1, is characterized in that: the concrete steps of step (10a) described eigenwert learning method are as follows:
1st step, by the original input data of all pixels of each super-pixel of composition, is input in degree of depth Robust Principal Component Analysis RPCA, obtains the second order eigenwert of all pixels;
2nd step, averages to the second order feature of all pixels, obtains the eigenwert of super-pixel.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510112514.0A CN104680184B (en) | 2015-03-14 | 2015-03-14 | Polarization SAR terrain classification method based on depth RPCA |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510112514.0A CN104680184B (en) | 2015-03-14 | 2015-03-14 | Polarization SAR terrain classification method based on depth RPCA |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104680184A true CN104680184A (en) | 2015-06-03 |
| CN104680184B CN104680184B (en) | 2018-03-06 |
Family
ID=53315201
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510112514.0A Active CN104680184B (en) | 2015-03-14 | 2015-03-14 | Polarization SAR terrain classification method based on depth RPCA |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104680184B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105069459A (en) * | 2015-08-18 | 2015-11-18 | 电子科技大学 | Surface feature type extracting method for high-resolution SAR image |
| CN105910827A (en) * | 2016-04-25 | 2016-08-31 | 东南大学 | Induction motor fault diagnosis method based on discriminant convolutional feature learning |
| CN105930846A (en) * | 2016-04-05 | 2016-09-07 | 西安电子科技大学 | Neighborhood information and SVGDL (support vector guide dictionary learning)-based polarimetric SAR image classification method |
| CN106203489A (en) * | 2016-07-01 | 2016-12-07 | 西安电子科技大学 | Classification of Polarimetric SAR Image method based on multiple dimensioned depth direction wave network |
| CN108319963A (en) * | 2018-02-06 | 2018-07-24 | 太原理工大学 | Urban road underground disease recognition method based on constraint Robust Principal Component Analysis |
| CN112561949A (en) * | 2020-12-23 | 2021-03-26 | 江苏信息职业技术学院 | Fast moving target detection algorithm based on RPCA and support vector machine |
| CN113344013A (en) * | 2021-07-15 | 2021-09-03 | 西安电子科技大学 | Polarimetric SAR image feature mining method based on graph low-rank sparse decomposition |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7200243B2 (en) * | 2002-06-28 | 2007-04-03 | The United States Of America As Represented By The Secretary Of The Army | Spectral mixture process conditioned by spatially-smooth partitioning |
| CN104123555A (en) * | 2014-02-24 | 2014-10-29 | 西安电子科技大学 | Super-pixel polarimetric SAR land feature classification method based on sparse representation |
| CN104331707A (en) * | 2014-11-02 | 2015-02-04 | 西安电子科技大学 | Polarized SAR (synthetic aperture radar) image classification method based on depth PCA (principal component analysis) network and SVM (support vector machine) |
-
2015
- 2015-03-14 CN CN201510112514.0A patent/CN104680184B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7200243B2 (en) * | 2002-06-28 | 2007-04-03 | The United States Of America As Represented By The Secretary Of The Army | Spectral mixture process conditioned by spatially-smooth partitioning |
| CN104123555A (en) * | 2014-02-24 | 2014-10-29 | 西安电子科技大学 | Super-pixel polarimetric SAR land feature classification method based on sparse representation |
| CN104331707A (en) * | 2014-11-02 | 2015-02-04 | 西安电子科技大学 | Polarized SAR (synthetic aperture radar) image classification method based on depth PCA (principal component analysis) network and SVM (support vector machine) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105069459B (en) * | 2015-08-18 | 2018-06-19 | 电子科技大学 | One kind is directed to High Resolution SAR Images type of ground objects extracting method |
| CN105069459A (en) * | 2015-08-18 | 2015-11-18 | 电子科技大学 | Surface feature type extracting method for high-resolution SAR image |
| CN105930846B (en) * | 2016-04-05 | 2019-05-03 | 西安电子科技大学 | Classification of Polarimetric SAR Image method based on neighborhood information and SVGDL |
| CN105930846A (en) * | 2016-04-05 | 2016-09-07 | 西安电子科技大学 | Neighborhood information and SVGDL (support vector guide dictionary learning)-based polarimetric SAR image classification method |
| CN105910827A (en) * | 2016-04-25 | 2016-08-31 | 东南大学 | Induction motor fault diagnosis method based on discriminant convolutional feature learning |
| CN105910827B (en) * | 2016-04-25 | 2017-04-05 | 东南大学 | Induction machine method for diagnosing faults based on identification convolution feature learning |
| CN106203489A (en) * | 2016-07-01 | 2016-12-07 | 西安电子科技大学 | Classification of Polarimetric SAR Image method based on multiple dimensioned depth direction wave network |
| CN106203489B (en) * | 2016-07-01 | 2019-02-15 | 西安电子科技大学 | Classification of Polarimetric SAR Image method based on multiple dimensioned depth direction wave network |
| CN108319963A (en) * | 2018-02-06 | 2018-07-24 | 太原理工大学 | Urban road underground disease recognition method based on constraint Robust Principal Component Analysis |
| CN108319963B (en) * | 2018-02-06 | 2021-10-01 | 太原理工大学 | Urban road underground disease identification method based on constrained robust principal component analysis |
| CN112561949A (en) * | 2020-12-23 | 2021-03-26 | 江苏信息职业技术学院 | Fast moving target detection algorithm based on RPCA and support vector machine |
| CN112561949B (en) * | 2020-12-23 | 2023-08-22 | 江苏信息职业技术学院 | Rapid moving object detection algorithm based on RPCA and support vector machine |
| CN113344013A (en) * | 2021-07-15 | 2021-09-03 | 西安电子科技大学 | Polarimetric SAR image feature mining method based on graph low-rank sparse decomposition |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104680184B (en) | 2018-03-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104680184A (en) | Polarization SAR terrain classification method based on deep RPCA | |
| CN103247059B (en) | A kind of remote sensing images region of interest detection method based on integer wavelet and visual signature | |
| CN104331707A (en) | Polarized SAR (synthetic aperture radar) image classification method based on depth PCA (principal component analysis) network and SVM (support vector machine) | |
| CN104680180B (en) | Classification of Polarimetric SAR Image method based on K mean values and sparse own coding | |
| CN103413146B (en) | Polarimetric SAR Image sophisticated category method based on Freeman entropys and self study | |
| CN106056157A (en) | Hyperspectral image semi-supervised classification method based on space-spectral information | |
| CN104036289A (en) | Hyperspectral image classification method based on spatial and spectral features and sparse representation | |
| CN102999762B (en) | Decompose and the Classification of Polarimetric SAR Image method of spectral clustering based on Freeman | |
| CN103824084A (en) | Polarimetric SAR (Synthetic Aperture Radar) image classification method based on SDIT (Secretome-Derived Isotopic Tag) and SVM (Support Vector Machine) | |
| CN103310230B (en) | Combine the hyperspectral image classification method separating mixed and self-adaptation Endmember extraction | |
| CN104077599A (en) | Polarization SAR image classification method based on deep neural network | |
| CN103886336A (en) | Polarized SAR image classifying method based on sparse automatic encoder | |
| CN104239900A (en) | Polarized SAR image classification method based on K mean value and depth SVM | |
| CN104299232B (en) | SAR image segmentation method based on self-adaptive window directionlet domain and improved FCM | |
| CN103413151A (en) | Hyperspectral image classification method based on image regular low-rank expression dimensionality reduction | |
| CN104463193A (en) | Polarization SAR image classifying method based on depth sparsity ICA | |
| CN105894013B (en) | Classification of Polarimetric SAR Image method based on CNN and SMM | |
| CN104156736A (en) | Polarized SAR image classification method on basis of SAE and IDL | |
| CN105956612A (en) | Hyperspectral image classification method based on active learning and semi-supervised learning | |
| CN105160353B (en) | Polarization SAR data terrain classification method based on multiple features collection | |
| CN105718942A (en) | Hyperspectral image imbalance classification method based on mean value drifting and oversampling | |
| CN105117736A (en) | Polarized SAR image classification method based on sparse depth stack network | |
| CN105160351B (en) | Semi-supervised hyperspectral classification method based on anchor point sparse graph | |
| CN107358214A (en) | Polarization SAR terrain classification method based on convolutional neural networks | |
| CN103824093A (en) | SAR (Synthetic Aperture Radar) image target characteristic extraction and identification method based on KFDA (Kernel Fisher Discriminant Analysis) and SVM (Support Vector Machine) |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |