CN110349105B - Dual-polarization SAR image speckle filtering method combining context covariance matrix - Google Patents

Dual-polarization SAR image speckle filtering method combining context covariance matrix Download PDF

Info

Publication number
CN110349105B
CN110349105B CN201910614332.1A CN201910614332A CN110349105B CN 110349105 B CN110349105 B CN 110349105B CN 201910614332 A CN201910614332 A CN 201910614332A CN 110349105 B CN110349105 B CN 110349105B
Authority
CN
China
Prior art keywords
context
pixel
dual
covariance matrix
ccm
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201910614332.1A
Other languages
Chinese (zh)
Other versions
CN110349105A (en
Inventor
陈思伟
曾晖
陶臣嵩
崔兴超
吴国庆
李郝亮
肖顺平
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
National University of Defense Technology
Original Assignee
National University of Defense Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by National University of Defense Technology filed Critical National University of Defense Technology
Priority to CN201910614332.1A priority Critical patent/CN110349105B/en
Publication of CN110349105A publication Critical patent/CN110349105A/en
Application granted granted Critical
Publication of CN110349105B publication Critical patent/CN110349105B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/13—Edge detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10032—Satellite or aerial image; Remote sensing
    • G06T2207/10044—Radar image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20024—Filtering details

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Radar Systems Or Details Thereof (AREA)
  • Image Processing (AREA)

Abstract

The invention discloses a dual-polarization SAR image speckle filtering method combining a context covariance matrix, which comprises the following steps that 1, a dual-polarization SAR image to be filtered is input; 2. constructing a context scattering vector in the neighborhood of a pixel point in the dual-polarization SAR image, and constructing a context covariance matrix; 3. solving a similarity parameter; 4. calculating a judgment threshold of the similarity parameter; 5. selecting a similar sample pixel set, and carrying out filtering processing on pixel points to be filtered; step 6: and traversing each pixel in the dual-polarized SAR image to be filtered, and repeating the steps 2 to 5 to obtain a filtering result graph. The invention effectively utilizes the context information of each pixel by constructing the context covariance matrix, calculates the similarity by calculating the context covariance matrix of each pixel and the pixels in the neighborhood of each pixel, and improves the selection precision of similar pixels, so that the similarity of a sample set of similar pixels is higher, the coherent speckle filtering is carried out on the image, and the performance of the filter is superior.

Description

Dual-polarization SAR image speckle filtering method combining context covariance matrix
Technical Field
The invention belongs to the technical field of dual-polarized SAR (Synthetic Aperture Radar) imaging remote sensing, and relates to a filtering method for speckle of a dual-polarized SAR image by combining a context covariance matrix.
Background
The dual-polarization SAR imaging system is an imaging system with the performance between single-polarization SAR and full-polarization SAR by adding another polarization mode under the single-polarization mode. The phenomenon of coherent speckle is widely existed in images obtained by coherent imaging systems such as dual-polarized SAR. The presence of coherent speckle presents difficulties and challenges to dual polarized SAR image understanding and interpretation. When processing such as target detection, classification and recognition is performed, speckle filtering preprocessing is generally required to be performed on the dual-polarized SAR image. The coherent speckle filtering method with excellent performance requires that the ground feature detail is well protected while the coherent speckle is fully inhibited. As dual-polarization SAR image preprocessing, the coherent speckle filtering performance directly influences the effects of various subsequent processing and application. Therefore, the development of the high-precision dual-polarization SAR coherent speckle adaptive filtering method is of great significance.
Speckle filtering mainly comprises two steps: the method comprises the steps of selecting similar candidate sample pixels and constructing an unbiased estimator. The selection of similar candidate sample pixels is a key for determining the filtering performance of the speckle, and becomes a research focus in the field. Since the 80 s of the last century, the speckle filtering method has been extensively studied domestically and abroad, and representative filters include: boxcar filters that do not average pixels within a sliding window, modified Lee filters that use 8 edge windows for similar sample pixel selection ("Computer Graphics and Image Processing", vol.15, pp.380-389,1981.), modified Sigma filters that use sample scattering fluctuation characteristics within a sliding window and introduce strong point protection ("I.S.Lee, J.H.Wen, T.L.Ainsworth, K.S.Chen, and A.J.Chen," Improved signature filter for spectrum filtering of Image, "IEEE SAR. geosci. Remote, Immun.47, 202-213, Jan.," MMSE. 2009, Jan. ", and Wavelet-filter decomposition-3. sub.3. transform. Sequence, M.19. noise, neighborhood filter, version, III.V.S.52. vol., and version filter, III.V.52-213, III. F.S. III, V, III. In the process of selecting similar samples, the method mainly utilizes the amplitude information of each pixel or each pixel in each cell block, does not fully consider the context information of the pixel and the adjacent pixels at the periphery, and has defects in the aspects of the quantity and the accuracy of similar sample selection, thereby causing the defect of speckle filtering performance. Therefore, the method has important significance for constructing the dual-polarization SAR image speckle filtering method combining the context covariance matrix by constructing the context covariance matrix fully considering the target scattering context information and establishing the similarity test factor of the context covariance matrix for describing and selecting the similarity sample.
Disclosure of Invention
The invention aims to solve the technical problem of providing a dual-polarization SAR image speckle filtering method which can accurately and adaptively select a candidate sample pixel set so as to improve the filtering precision and combines a context covariance matrix.
The basic idea of the invention is as follows: according to the dual-polarization SAR imaging principle, the value of each pixel in the SAR image is closely related to the surrounding neighborhood pixels. This correlation implies rich information, i.e., context information. The selection precision of similar pixels can be improved by effectively utilizing the context information, and further the filtering performance of the coherent speckles is improved. Specifically, for each pixel in the dual-polarized SAR image, the method firstly constructs a context scattering vector and a context covariance matrix in a win multiplied by win neighborhood. On the basis, a context covariance matrix similarity parameter is established according to a matrix similarity inspection principle. And then, a similar sample pixel set is accurately selected in a sliding window by utilizing the context covariance matrix similarity parameter and according to threshold judgment, so that the self-adaptive speckle filtering of the dual-polarization SAR image is realized.
The technical scheme adopted by the invention is as follows:
a dual-polarization SAR image speckle filtering method combining a context covariance matrix comprises the following steps:
step 1: inputting a dual-polarized SAR image to be filtered;
step 2: for pixel point S in dual-polarized SAR imagen,mAt the pixel point Sn,mConstructing a context scattering vector in the win multiplied by win neighborhood, and constructing a context covariance matrix C of the pixel point according to the context scattering vectorCCM-(n,m)，n＝1,2,…,N，m＝1,2,…,M，N,MRespectively representing the total number of row and column pixel points of the dual-polarized SAR image, wherein win is an odd number which is more than or equal to 3;
and step 3: according to the pixel point Sn,mContext covariance matrix C ofCCM-(n,m)Calculating a pixel point Sn,mContext covariance matrix C ofCCM-(n,m)And with the pixel point Sn,mContext covariance matrix C of each pixel within a sliding window I J as the centerCCM-(i,j)Similarity parameter lnQij-nmTo obtain a similarity parameter matrix lnQnm-IJI is 1,2, …, I, J is 1,2, …, J, I, J respectively represent the total number of pixels in the row and column of the sliding window, I, J is an odd number;
and 4, step 4: calculating a judgment threshold of the similarity parameter;
and 5: according to the judgment threshold of the similarity parameter, a pixel point S is usedn,mSelecting a similar sample pixel set in a sliding window I multiplied by J as a center, and treating a filtering pixel Sn,mCarrying out filtering treatment;
step 6: and traversing each pixel in the dual-polarized SAR image to be filtered, and repeating the steps from 2 to 5 to obtain an SAR speckle filtering result image.
For further optimizing the scheme, the following improvements are made:
further, the method for constructing the context scattering vector and the context covariance matrix in step 2 comprises:
step 2.1: constructing a rule by using a seed pixel point S according to a context scattering vectorn,mSelecting a certain number of pixel points in each vector construction direction in the neighborhood of the core, wherein the pixel points comprise seed pixel points Sn,mAnd neighborhood pixel points thereof, each vector constructing direction context scattering vector
Figure GDA0002881802280000031
The elements in the dual-polarized image are composed of pixel values of pixel points selected from each vector construction direction on a first channel and a second channel of the dual-polarized image, the dimensions of the context scattering vectors in each vector construction direction are consistent, and each context scattering vector contains a seed pixel point Sn,mIn all directionsThe context scattering vector of (1) except the seed pixel point Sn,mBesides, other neighborhood pixels are not repeatedly used,
Figure GDA0002881802280000032
representing seed pixel Sn,mA vth context scattering vector;
step 2.2: contextual scatter vectors in each vector construction direction
Figure GDA0002881802280000033
Multiplying the obtained data by the conjugate transpose of the data to obtain a context covariance matrix corresponding to the number of the context scattering vectors, and averaging the obtained context covariance matrix to obtain a seed pixel point Sn,mIn this way the context covariance matrix is constructed,
Figure GDA0002881802280000034
v is the total number of constructed context scattering vectors, superscript
Figure GDA0002881802280000035
Representing a conjugate transpose.
Furthermore, the context scattering vector construction rule refers to that the construction direction of the context scattering vector is determined according to the pixel point Sn,mThe seed pixels are in central-meter-shaped and end-point radial shapes, and the neighborhood pixel points in each direction in the two types of construction rules are selected according to the neighborhood pixel points relative to the pixel point Sn,mThe positions of the two types of the three-dimensional optical fiber are divided into a symmetrical type and an asymmetrical type.
Further, when the building direction is a central shape like a Chinese character 'mi', and the pixels in each building direction are selected to be symmetrical, the context scattering vector is expressed as:
Figure GDA0002881802280000041
and the following constraints need to be satisfied:
Figure GDA0002881802280000042
Figure GDA0002881802280000043
i1≥j1,i2≥j2,...,id≥jd (4)
i1, j 1; i2, j 2; ...; id, jd and not simultaneously zero (5)
Figure GDA0002881802280000044
D is the number of pixel points selected in the construction direction of the context scattering vector, superscript T represents transposition, channel-1 and channel-2 respectively represent a first channel and a second channel of the dual-polarization SAR image,
Figure GDA0002881802280000045
representing a pixel S located in the n + id row, m + jd column of a fully polarised SAR imagen+id,m+jdId denotes the pixel value on channel 1, i1, i2n,mJ1, j2,. jd represents the step value of the row on the basis of the coordinates (n, m) ofn,mStep values listed on the basis of the coordinates (n, m), d being an integer;
when the construction direction is a central Chinese character 'mi' shape and the pixels in each construction direction are selected to be asymmetric, the context scattering vector is expressed as:
Figure GDA0002881802280000046
and the following constraints need to be satisfied:
Figure GDA0002881802280000047
Figure GDA0002881802280000048
i1≥j1，...id1≥jd1，i2≥j2，...id2≥jd2 (10)
i1, j1 … id1, jd1 and i2, j2 … id2, jd2 are not zero at the same time (11)
i1 ═ i 2., id1 ═ id2, j1 ═ j 2., jd1 ═ jd2 at the same time, which is not true (12)
1≤d1≤D-2，1≤d2≤D-2，d1+d2≤D-2 (13)
i 1.. id 1; i 2.. id2 denotes at the pixel point Sn,mThe step value of the row, j1,. jd1, on the basis of the coordinates (n, m); j 2.. jd2 denotes at the pixel point Sn,mStep values listed on the basis of the coordinates (n, m), d1, d2 being integers;
when the build direction is radial to the endpoint, the context scatter vector is expressed as:
Figure GDA0002881802280000051
and the following constraints need to be satisfied:
i＝0,1,2,···,(win-1) (15)
j＝0,1,2,···,(win-1) (16)
0＜k1＜k2＜...＜kd≤1 (17)
i, j, k1. i, k1. j, … kd. i, kd. j are integers (18)
i+j-(win-1)＞0 (19)
i, j are not zero at the same time (20)
d＝D-1 (21)
k1 · i, k1 · j, … kd · i, kd · j respectively represent at the pixel point Sn,mBased on the coordinate (n, m), the step values of the rows and the columns are k1... kd are step growth coefficients of the rows and the columns, i and j respectively represent the horizontal and vertical coordinates of the selected D-th pixel point in the construction direction of each context scattering vector relative to the pixel point Sn,mIs offset from the coordinate (n, m).
Further, a similarity parameter lnQ is calculated in step 3ij-nmAnd a similarity parameter matrix lnQnm-IJThe method comprises the following steps:
1) calculating CCCM-(n,m)And CCCM-(i,j)Similarity parameter oflnQij-nm：
lnQij-nm＝2qln2+ln[Det(CCCM-(i,j))]+ln[Det(CCCM-(n,m))]-2ln[Det(CCCM-(i,j)+CCCM-(n,m))] (22)
Wherein q is a context covariance matrix CCCM-(n,m)D, Det (-) denotes the determinant of the matrix, the symbol ln denotes the natural logarithm, when CCCM-(i,j)＝CCCM-(n,m)lnQij-nm0; when C isCCM-(i,j)≠CCCM-(n,m)lnQij-nm＜0；
2) Similarity parameter matrix lnQnm-IJ
Traversing context covariance matrix C of each pixel within sliding window I × JCCM-(i,j)Each pixel S in the sliding window I × J is calculated by the formula (22)i,jContext covariance matrix and pixel point S to be filteredn,mlnQ of the context covariance matrixij-nmEach pixel in the sliding window I multiplied by J and the pixel S to be filteredn,mThe similarity parameters of (a) form a similarity parameter matrix lnQnm-IJ，lnQij-nmIs a similarity parameter matrix lnQnm-IJRow i and column j.
Further, the method for calculating the decision threshold in step 4 is
Figure GDA0002881802280000061
Wherein E is an adjusting parameter, the inhibition of balanced coherent speckles and the image detail maintenance are balanced, L is the multi-view number of the SAR image, and q is a context covariance matrix CCCM-(n,m)Of (c) is calculated.
Further, the method for selecting the similar sample pixel set in step 5 is as follows: at the pixel S to be filteredn,mIn the sliding window of (I) x J, the similarity parameter matrix lnQnm-IJEach similarity parameter lnQij-nmCompared with the decision threshold Th, the similarity parameter matrix lnQnm-IJTaking the pixel corresponding to the similarity parameter with the value greater than the threshold Th, determining the pixel as a candidate sample pixel set, and recording the pixel as:
{SCCM-g}＝{Si,j|lnQnm-IJ≥Th}。
further, the method for performing filtering processing on the pixel to be filtered in step 5 is as follows: let { SCCM-gIf the number of elements in the pixel is G, then the pixel S to be filtered is obtainedn,mResult of filtering processing of
Figure GDA0002881802280000062
Comprises the following steps:
Figure GDA0002881802280000063
furthermore, the value of I, J in the sliding window I × J is 15-25.
Further, the parameter of the decision threshold Th is win-3, and q-6, when the image data is ADTS data, L takes a value of 1, and E takes a value of E
Figure GDA0002881802280000064
When the image data is Radarsat2 data, L is 2 and E is 2
Figure GDA0002881802280000065
Compared with the prior art, the invention has the following beneficial effects:
the invention combines the dual-polarized SAR image speckle filtering method of the context covariance matrix, utilizes the relationship between the pixel values of the adjacent positions in the dual-polarized SAR image to construct the dual-polarized context covariance matrix, and the context covariance matrix fully utilizes the information among all channels in the dual-polarized image, so that a pixel sample set with higher similarity is selected according to the similarity parameter between the context matrices of all pixels in a sliding window, thereby leading the accuracy of the estimated value of the true value of the dual-polarized SAR data to be higher, and leading the filtering performance of the image for speckle filtering to be better. The invention has simple principle and excellent filter performance, and can carry out self-adaptive processing on various dual-polarized SAR data.
Drawings
FIG. 1 is a flow chart of an embodiment of the present invention;
fig. 2 is a schematic diagram of a neighborhood size of 3 × 3, a dimension of a context scattering vector of 3, and a central m-shaped + symmetric context scattering vector construction, where (a) shows a point-taking rule, and (b) - (e) are neighborhood pixel positions corresponding to three elements in 4 context scattering vectors in the construction method; wherein the black pixels represent selected seed pixels;
FIG. 3 is a schematic diagram of a context scattering vector construction with a neighborhood size of 3 × 3, a context scattering vector dimension of 3, and radial endpoints; graph (a) represents a point-taking rule, and graphs (b) - (d) are neighborhood pixel positions corresponding to three elements in 3 context scattering vectors under the construction mode; wherein the black pixels represent selected seed pixels;
fig. 4 is a schematic diagram of a neighborhood size of 5 × 5, a dimension of a context scattering vector of 3, and a central m-shaped + symmetric context scattering vector construction, where a diagram (a) shows a point-taking rule, and diagrams (b) - (m) show neighborhood pixel positions corresponding to three elements in 12 context scattering vectors in the construction method; wherein the black pixels represent selected seed pixels;
fig. 5 is a schematic diagram of a neighborhood size of 5 × 5, a dimension of a context scattering vector of 3, and a central m-shaped + asymmetric context scattering vector construction, where (a) shows a point-taking rule, and (b) - (i) are neighborhood pixel positions corresponding to three elements in 8 context scattering vectors in the construction method; wherein the black pixels represent selected seed pixels;
FIG. 6 is a schematic diagram of a construction of a central Mi-shaped + symmetric context scattering vector with a neighborhood size of 5 × 5 and a context scattering vector dimension of 5; graph (a) represents a point-taking rule, and graphs (b) - (e) are neighborhood pixel positions corresponding to three elements in 4 context scattering vectors under the construction mode; wherein the black pixels represent selected seed pixels;
FIG. 7 is a schematic diagram of a construction of context scattering vectors with neighborhood size of 5 × 5, context scattering vector dimension of 5, and radial endpoints; graph (a) represents a point-taking rule, and graphs (b) - (d) are neighborhood pixel positions corresponding to three elements in 3 context scattering vectors under the construction mode; wherein the black pixels represent selected seed pixels;
fig. 8 is a comparison graph of speckle filtering results of satellite-borne Radarsat2 HH + VV dual-polarized SAR data, (a) an original image, (b) a 7 × 7Boxcar method filtering result graph, (c) a 9 × 9Boxcar method filtering result graph, (d) a 7 × 7 improved Lee method filtering result graph, (e) a 9 × 9 improved Lee method filtering result graph, (f) a SAR-BM3D method filtering result graph, (g) a 9 × 9 improved Sigma method filtering result graph, (h) a 11 × 11 improved Sigma method filtering result graph, (i) a method filtering result graph of the present invention;
fig. 9 is a comparison graph of speckle filtering results of spaceborne Radarsat2 HH + HV dual-polarized SAR data, (a) an original image, (b) a 7 × 7Boxcar method filtering result graph, (c) a 9 × 9Boxcar method filtering result graph, (d) a 7 × 7 improved Lee method filtering result graph, (e) a 9 × 9 improved Lee method filtering result graph, (f) a SAR-BM3D method filtering result graph, (g) a 9 × 9 improved Sigma method filtering result graph, (h) a 11 × 11 improved Sigma method filtering result graph, (i) a method filtering result graph of the present invention;
fig. 10 is a comparison graph of speckle filtering results of onboard ADTS HH + VV dual-polarized SAR data, (a) an original image, (b) a 7 × 7Boxcar method filtering result graph, (c) a 9 × 9Boxcar method filtering result graph, (d) a 7 × 7 improved Lee method filtering result graph, (e) a 9 × 9 improved Lee method filtering result graph, (f) a 9 × 9 improved Sigma method filtering result graph, (g) a 11 × 11 improved Sigma method filtering result graph, (h) a SAR-BM3D method filtering result graph, (i) a method filtering result graph of the present invention;
fig. 11 is an edge detection result diagram in the onboard ADTS HH + VV dual-polarized SAR data, (a1) an original image, (b1) a 7 × 7Boxcar method filtering result diagram, (c1) a 9 × 9Boxcar method filtering result diagram, (d1) a 7 × 7 improved Lee method filtering result diagram, (e1) a 9 × 9 improved Lee method filtering result diagram, and (f1) an edge true value diagram; (g1)9 × 9 modified Sigma method filter result graph, (h1)11 × 11 modified Sigma method filter result graph, (i1) SAR-BM3D method filter result graph, (j1) the method filter result graph of the present invention, (a2) edge detection result graph based on original image, (b2) edge detection result graph based on 7 × 7Boxcar method filter image, (c2) edge detection result graph based on 9 × 9Boxcar method filter image, (d2) edge detection result graph based on 7 × 7 modified Lee method filter image, (e2) edge detection result graph based on 9 × 9 modified Lee method filter image, (f2) edge true value graph; (g2) an edge detection result graph based on a 9 × 9 improved Sigma method filtered image, (h2) an edge detection result graph based on an 11 × 11 improved Sigma method filtered image, (i2) an edge detection result graph based on a SAR-BM3D method filtered image, (j2) an edge detection result graph based on a filtered image of the method of the present invention;
fig. 12 is a comparison graph of speckle filtering results of airborne ADTS HH + HV dual-polarization SAR data, (a) an original image, (b) a 7 × 7Boxcar method filtering result graph, (c) a 9 × 9Boxcar method filtering result graph, (d) a 7 × 7 improved Lee method filtering result graph, (e) a 9 × 9 improved Lee method filtering result graph, (f) a 9 × 9 improved Sigma method filtering result graph, (g) a 11 × 11 improved Sigma method filtering result graph, (h) a SAR-BM3D method filtering result graph, (i) a method filtering result graph of the present invention;
fig. 13 is a diagram of edge detection results in airborne ADTS HH + HV dual-polarization SAR data, (a1) an original image, (b1) a 7 × 7Boxcar method filtering result diagram, (c1) a 9 × 9Boxcar method filtering result diagram, (d1) a 7 × 7 improved Lee method filtering result diagram, (e1) a 9 × 9 improved Lee method filtering result diagram, and (f1) an edge true value diagram; (g1)9 × 9 modified Sigma method filter result graph, (h1)11 × 11 modified Sigma method filter result graph, (i1) SAR-BM3D method filter result graph, (j1) the invention method filter result graph; (a2) an edge detection result map based on an original image, (b2) an edge detection result map based on a 7 × 7Boxcar method filtered image, (c2) an edge detection result map based on a 9 × 9Boxcar method filtered image, (d2) an edge detection result map based on a 7 × 7 improved Lee method filtered image, (e2) an edge detection result map based on a 9 × 9 improved Lee method filtered image, (f2) an edge true value map; (g2) an edge detection result graph based on a 9 × 9 improved Sigma method filtered image, (h2) an edge detection result graph based on an 11 × 11 improved Sigma method filtered image, (i2) an edge detection result graph based on an SAR-BM3D method filtered image, and (j2) an edge detection result graph based on a filtered image of the method of the present invention.
Detailed Description
For better understanding of the technical solution of the present invention, fig. 1 to fig. 13 show a specific embodiment of the dual-polarized SAR image speckle filtering method in combination with a context covariance matrix according to the present invention, as shown in a flowchart of fig. 1, including the following steps:
step 1: inputting a single-polarized SAR image to be filtered;
step 2: for pixel point S in single-polarized SAR imagen,mAt the pixel point Sn,mConstructing a context scattering vector in the win multiplied by win neighborhood, and constructing a context covariance matrix C of the pixel point according to the context scattering vectorCCM-(n,m)N is 1,2, …, N, M is 1,2, …, M, N, M respectively represent the total number of row and column pixels of the SAR image, win is an odd number greater than or equal to 3;
in this embodiment, the method for constructing the context scattering vector and the context covariance matrix is as follows:
step 2.1: constructing a rule by using a seed pixel point S according to a context scattering vectorn,mSelecting a certain number of pixel points in each vector construction direction in the neighborhood of the core, wherein the pixel points comprise seed pixel points Sn,mAnd neighborhood pixel points thereof, each vector constructing direction context scattering vector
Figure GDA0002881802280000091
The elements in the dual-polarized image are composed of pixel values of pixel points selected from each vector construction direction on a first channel and a second channel of the dual-polarized image, the dimensions of the context scattering vectors in each vector construction direction are consistent, and each context scattering vector contains a seed pixel point Sn,mSeed-removing pixel point S in context scattering vector in all directionsn,mBesides, other neighborhood pixels are not repeatedly used,
Figure GDA0002881802280000101
representing seed pixel Sn,mA vth context scattering vector;
in this embodiment, the context scattering vector construction rule means that the construction direction of the context scattering vector is determined by the pixel point Sn,mThe core is in a central shape like a Chinese character 'mi' and in an end point radiation shape, and the neighborhood pixel points in each direction in the two types of construction rules are selected according to the neighborhood pixel points relative to the pixel point Sn,mPosition ofThe method is divided into a symmetrical type and an asymmetrical type.
Step 2.2: contextual scatter vectors in each vector construction direction
Figure GDA0002881802280000102
Multiplying the obtained data by the conjugate transpose of the data to obtain a context covariance matrix corresponding to the number of the context scattering vectors, and averaging the obtained context covariance matrix to obtain a seed pixel point Sn,mIn this way the context covariance matrix is constructed,
Figure GDA0002881802280000103
v is the total number of constructed context scattering vectors, superscript
Figure GDA0002881802280000104
Representing conjugate transpose, in this embodiment, V is not less than 3, that is, a context covariance matrix can be constructed by taking more than 3 context scattering vectors, and the more the context scattering vectors are, the more the seed pixel point S isn,mThe richer the neighborhood information is, the higher the filtering accuracy is.
Fig. 2, fig. 4, and fig. 6 are schematic diagrams showing a pixel point fetching rule and a construction direction when the construction direction is a central shape of a Chinese character 'mi' and pixels in each construction direction are selected as symmetrical pixels, where fig. 2(a) shows the point fetching rule, and fig. 2(b) - (e) are neighborhood pixel positions corresponding to three elements in 4 context scattering vectors in the construction mode; wherein the black pixels represent selected seed pixels; fig. 4 shows a schematic diagram of a construction of a central m-shaped + symmetric context scattering vector with a neighborhood size of 5 × 5 and a context scattering vector dimension of 3. FIG. 4(a) shows a point-taking rule, and FIGS. 4(b) - (m) show neighborhood pixel positions corresponding to three elements in 12 context scattering vectors in the construction method; wherein the black pixels represent selected seed pixels; FIG. 6 is a schematic diagram showing a neighborhood size of 5 × 5, a dimension of a context scattering vector of 5, and a central m-shaped + symmetric context scattering vector construction; FIG. 6(a) shows a point-taking rule, and FIGS. 6(b) - (e) show neighborhood pixel positions corresponding to three elements in 4 context scattering vectors in the construction method; wherein the black pixels represent selected seed pixels; therefore, when the building direction is a central shape of a meter and the selection of the pixel points in each building direction is a symmetric type, the context scattering vector is expressed as:
Figure GDA0002881802280000105
and the following constraints need to be satisfied:
Figure GDA0002881802280000111
Figure GDA0002881802280000112
i1≥j1,i2≥j2,...,id≥jd (4)
i1, j 1; i2, j 2; ...; id, jd are not zero at the same time (5)
Figure GDA0002881802280000113
D is the number of pixel points selected in the construction direction of the context scattering vector, and the mark is addedTRepresenting transposition, channel-1, channel-2 representing respectively a first channel and a second channel of a dual polarized SAR image, HH channel and VV channel respectively,
Figure GDA0002881802280000114
representing a pixel S located in the n + id row, m + jd column of a fully polarised SAR imagen+id,m+jdId denotes the pixel value on channel 1, i1, i2n,mJ1, j2,. jd represents the step value of the row on the basis of the coordinates (n, m) ofn,mStep values listed on the basis of the coordinates (n, m), d being an integer;
fig. 5 shows a schematic diagram of a point-fetching rule and a construction direction of a neighborhood of 5 × 5, a dimension of a context scattering vector of 3, and a central m-shaped + asymmetric context scattering vector. FIG. 5(a) shows a point-taking rule, and FIGS. 5(b) - (i) show neighborhood pixel positions corresponding to three elements in 8 context scattering vectors in the construction mode; where the black pixels represent the selected seed pixels. Therefore, when the building direction is a central shape like a Chinese character 'mi', and the pixels in each building direction are selected to be asymmetric, the context scattering vector is expressed as:
Figure GDA0002881802280000115
and the following constraints need to be satisfied:
Figure GDA0002881802280000116
Figure GDA0002881802280000117
i1≥j1，...id1≥jd1，i2≥j2，...id2≥jd2 (10)
i1, j1 … id1, jd1 and i2, j2 … id2, jd2 are not zero at the same time (11)
i1 ═ i 2., id1 ═ id2, j1 ═ j 2., jd1 ═ jd2 at the same time, which is not true (12)
1≤d1≤D-2，1≤d2≤D-2，d1+d2≤D-2 (13)
i 1.. id 1; i 2.. id2 denotes at the pixel point Sn,mThe step value of the row, j1,. jd1, on the basis of the coordinates (n, m); j 2.. jd2 denotes at the pixel point Sn,mThe step values listed on the basis of the coordinates (n, m), d1, d2 being integers.
Fig. 3 and 7 show the construction schematic diagram of the context scattering vector when the construction direction is end point radial, and fig. 3 shows the construction schematic diagram of the context scattering vector when the construction neighborhood size is 3 × 3, the dimension of the context scattering vector is 3, and the end point radial; FIG. 3(a) shows a point-taking rule, and FIGS. 3(b) - (d) show neighborhood pixel positions corresponding to three elements in 3 context scattering vectors in the construction method; where the black pixels represent the selected seed pixels. FIG. 7 is a schematic diagram of a construction of context scattering vectors with neighborhood size of 5 × 5, context scattering vector dimension of 5, and radial endpoints; FIG. 7(a) shows a point-taking rule, and FIGS. 7(b) - (d) show neighborhood pixel positions corresponding to three elements in 3 context scattering vectors in this construction mode; where the black pixels represent the selected seed pixels. Thus when the build direction is radial to the endpoint, the context scatter vector is expressed as:
Figure GDA0002881802280000121
and the following constraints need to be satisfied:
i＝0,1,2,···,(win-1) (15)
j＝0,1,2,···,(win-1) (16)
0＜k1＜k2＜...＜kd≤1 (17)
i, j, k1. i, k1. j, … kd. i, kd. j are integers (18)
i+j-(win-1)＞0 (19)
i, j are not zero at the same time (20)
d＝D-1 (21)
k1 · i, k1 · j, … kd · i, kd · j respectively represent at the pixel point Sn,mBased on the step values of the rows and columns, k1,. kd is the step growth coefficient of the rows and columns, for example, for win being 3, i, j may take 2, k1 may take 1/2; in the case of 5 × 5, i, j may be 4 at maximum, k1 is 1/4, k2 is 2/4, and k3 is 3/4.
In this embodiment, win is equal to 3, and the pixel point S isn,mThe constructed context scattering vector in the 3 × 3 neighborhood of (a) is:
Figure GDA0002881802280000122
Figure GDA0002881802280000123
Figure GDA0002881802280000131
Figure GDA0002881802280000132
on this basis, the constructed context covariance matrix is:
Figure GDA0002881802280000133
when win is 3, V takes 4. Upper labelTIndicating transposition, superscript
Figure GDA0002881802280000134
Representing a conjugate transpose.
And step 3: calculating a pixel point S to be filtered according to the context covariance matrixn,mContext covariance matrix C ofCCM-(n,m)And with the pixel point Sn,mContext covariance matrix C of each pixel within a sliding window I J as the centerCCM-(i,j)Similarity parameter lnQ betweenij-nmTo obtain a similarity parameter matrix lnQnm-IJI is 1,2, …, I, J is 1,2, …, J, I, J respectively represent the total number of pixels in the row and column of the sliding window, I, J is an odd number; in this embodiment, I, J takes on a value of 15-25.
1) Similarity parameter lnQij-nm：
lnQij-nm＝2qln2+ln[Det(CCCM-(i,j))]+ln[Det(CCCM-(n,m))]-2ln[Det(CCCM-(i,j)+CCCM-(n,m))] (22)
Wherein q is a context covariance matrix of CCCM-(n,m)D, Det (-) denotes the determinant of the matrix, the symbol ln denotes the natural logarithm, when CCCM-(i,j)＝CCCM-(n,m)lnQij-nm0; when C isCCM-(i,j)≠CCCM-(n,m)lnQij-nm＜0；
2) Similarity parameter matrix lnQnm-IJ：
Traversing each pixel point in the sliding window I multiplied by J, and calculating by the formula (22)To the pixel S to be filteredn,mSimilarity parameter lnQ with all pixels in sliding window I × Jij-nmForm a similarity parameter matrix lnQ of size I × Jnm-IJ，lnQij-nmIs a similarity parameter matrix lnQnm-IJRow i and column j.
And 4, step 4: calculating a judgment threshold of the similarity parameter;
Figure GDA0002881802280000135
wherein E is an adjusting parameter, the inhibition of balanced coherent speckles and the image detail maintenance are balanced, L is the multi-view number of the SAR image, and q is a context covariance matrix CCCM-(n,m)Of (c) is calculated. In this embodiment, win is equal to 3, the number D of pixel points taken by each context scattering vector is 3, q is equal to 6 because the number of channels of the dual-polarized image is 2, when the image data is ADTS data, L is equal to 1, and E is equal to E
Figure GDA0002881802280000141
When the image data is Radarsat2 data, L is 2 and E is 2
Figure GDA0002881802280000142
Of course, the decision threshold can also be directly assigned according to the filtering effect requirement, and the range is usually (-20, -0.1), and typical values of the decision threshold can be-0.5, -1, -1.5, -2, -2.5 and-3.
And 5: according to the judgment threshold of the similarity parameter, a pixel point S is usedn,mSelecting a similar sample pixel set in a sliding window I multiplied by J as a center, and treating a filtering pixel Sn,mCarrying out filtering treatment;
the method for selecting the similar sample pixel set comprises the following steps:
traversal similarity parameter matrix lnQnm-IJIf the value of the similarity parameter is greater than the threshold Th, the pixel corresponding to the similarity parameter is determined as the similarity sample of the pixel to be filtered, so as to determine the similarity sample set similar to the pixel to be filtered in the neighborhood, and the similarity sample set is recorded as：
{SCCM-g}＝{Si,j|lnQnm-IJ≥Th}。
The method for filtering the pixel to be filtered comprises the following steps: let { SCCM-gIf the number of elements in the pixel is G, then the pixel S to be filtered is obtainedn,mResult of filtering processing of
Figure GDA0002881802280000143
Comprises the following steps:
Figure GDA0002881802280000144
step 6: and traversing each pixel in the single-polarized SAR image to be filtered, and repeating the steps 2 to 5 to obtain an SAR speckle filtering result image.
Fig. 8 to 13 are graphs comparing speckle filtering results of HH + VV dual polarized SAR data, HH + HV dual polarized data, and onboard ADTS HH + VV dual polarized SAR data, HH + HV dual polarized SAR data using satellite-borne Radarsat 2. Wherein 7 × 7Boxcar method, 9 × 9Boxcar method, 7 × 7 improved Lee method, 9 × 9 improved Sigma method, 11 × 11 improved Sigma method and SAR-BM3D method, 7 most commonly used SAR coherent speckle filtering methods at present are selected as comparison methods.
Fig. 8 is a comparison graph of speckle filtering results of satellite-borne Radarsat2 HH + VV dual-polarized SAR data. Fig. 8(a) is an original image, fig. 8(b) is a 7 × 7Boxcar method filtering result graph, fig. 8(c) is a 9 × 9Boxcar method filtering result graph, fig. 8(d) is a 7 × 7 improved Lee method filtering result graph, fig. 8(e) is a 9 × 9 improved Lee method filtering result graph, fig. 8(f) is a SAR-BM3D method filtering result graph, fig. 8(g) is a 9 × 9 improved Sigma method filtering result graph, fig. 8(h) is a 11 × 11 improved Sigma method filtering result graph, and fig. 8(i) is a filtering result graph of the method of the present invention. The method has the advantage that the comprehensive performance of the HHVV dual-polarized SAR data in the aspects of speckle suppression, target detail protection and the like is better than that of other methods.
Fig. 9 is a comparison graph of speckle filtering results of satellite-borne Radarsat2 HH + HV dual-polarized SAR data. Fig. 9(a) is an original image, fig. 9(b) is a 7 × 7Boxcar method filter result graph, fig. 9(c) is a 9 × 9Boxcar method filter result graph, fig. 9(d) is a 7 × 7 modified Lee method filter result graph, fig. 9(e) is a 9 × 9 modified Lee method filter result graph, fig. 9(f) is a SAR-BM3D method filter result graph, fig. 9(g) is a 9 × 9 modified Sigma method filter result graph, fig. 9(h) is a 11 × 11 modified Sigma method filter result graph, and fig. 9(i) is a filter result graph of the method of the present invention. As can be seen from the vision, for HHHV dual-polarized SAR data, the comprehensive performance of the method in the aspects of speckle suppression, target detail protection and the like is superior to that of other methods.
Tables 1 and 2 show the speckle filtering results of spaceborne Radarsat2 HH + VV and HH + HV dual-polarized SAR data in quantitative comparison. In order to objectively and quantitatively compare the performance of these speckle filtering methods, 3 clutter regions were arbitrarily selected for analysis in FIG. 8(a) and FIG. 9(a), and respectively labeled as ROI-1, ROI-2, and ROI-3. The selected quantitative evaluation index is equivalent visual Number ENL (equivalent Number of looks), and the definition formula is as follows:
Figure GDA0002881802280000151
wherein: μ represents the mean of the amplitudes of all pixels in the ROI region, and σ represents the standard deviation of the amplitudes of all pixels in the ROI region.
TABLE 1 quantitative comparison of speckle filtering results of spaceborne Radarsat2 HH + VV dual-polarized SAR data
Figure GDA0002881802280000152
TABLE 2 quantitative comparison of speckle filtering results of spaceborne Radarsat2 HH + HV dual-polarized SAR data
Figure GDA0002881802280000161
The larger the value of the equivalent vision ENL is, the better the speckle suppression performance is. As can be seen from tables 1 and 2, for HH + VV dual-polarized SAR data, for 3 ROI regions, the ENLs of the HH channel after speckle filtering of the present invention are 561.39, 301.65, and 926.78, respectively, and the ENLs of the VV channel are 764.83, 303.38, and 727.99, respectively; for HH + HV dual polarized SAR data, for 3 ROI regions, the ENLs of the HH channel after speckle filtering of the invention are 681.68, 312.37, and 857.40, respectively, and the ENLs of the HV channel are 249.76, 150.65, and 518.48, respectively. The ENL index of the invention is obviously superior to other comparison methods, and the advantages of the invention are also proved from objective evaluation.
Fig. 10 is a comparison graph of speckle filtering results of airborne ADTS HH + VV dual-polarized SAR data. Fig. 10(a) is an original image, fig. 10(b) is a 7 × 7Boxcar method filter result graph, fig. 10(c) is a 9 × 9Boxcar method filter result graph, fig. 10(d) is a 7 × 7 modified Lee method filter result graph, fig. 10(e) is a 9 × 9 modified Lee method filter result graph, fig. 10(f) is a 9 × 9 modified Sigma method filter result graph, fig. 10(g) is a 11 × 11 modified Sigma method filter result graph, and fig. 10(h) is a SAR-BM3D method filter result graph, and (i) is a method filter result graph of the present invention. As can be seen from the vision, for the airborne dual-polarized SAR data, the comprehensive performance of the method in the aspects of speckle suppression, target detail protection and the like is also obviously superior to that of other methods.
Fig. 11 is a diagram of target edge detection results in airborne ADTS HH + VV dual-polarized SAR data. Fig. 11(a1) original image, fig. 11(b1)7 × 7Boxcar method filter result graph, fig. 11(c1)9 × 9Boxcar method filter result graph, fig. 11(d1)7 × 7 modified Lee method filter result graph, fig. 11(e1)9 × 9 modified Lee method filter result graph, fig. 11(f1) edge true value graph; fig. 11(g1)9 × 9 modified Sigma method filter result graph, fig. 11(h1)11 × 11 modified Sigma method filter result graph, fig. 11(i1) SAR-BM3D method filter result graph, fig. 11(j1) method filter result graph of the present invention. Fig. 11(a2) is a graph of an edge detection result based on an original image, fig. 11(b2) is a graph of an edge detection result based on a 7 × 7Boxcar method filtered image, fig. 11(c2) is a graph of an edge detection result based on a 9 × 9Boxcar method filtered image, fig. 11(d2) is a graph of an edge detection result based on a 7 × 7 modified Lee method filtered image, fig. 11(e2) is a graph of an edge detection result based on a 9 × 9 modified Lee method filtered image, and fig. 11(f2) is a graph of an edge true value; fig. 11(g2) is a graph of the edge detection result of the filtered image based on the 9 × 9 improved Sigma method, fig. 11(h2) is a graph of the edge detection result of the filtered image based on the 11 × 11 improved Sigma method, fig. 11(i2) is a graph of the edge detection result of the filtered image based on the SAR-BM3D method, and fig. 11(j2) is a graph of the edge detection result of the filtered image based on the method of the present invention. As can be seen from fig. 11(a1) - (j1), the overall performance of the present invention in terms of speckle suppression and target detail protection is also significantly better for edge-type targets than other approaches. Fig. 11(a2) - (j2) are graphs of edge detection results obtained by a conventional ROA edge detection method (r. touzi, a. lopes, and p. bouquet, "a static and geographic edge detector for SAR images," IEEE trans. geosci. remote sens., vol.26, pp.764-773, Nov 1988.). Wherein, the edge detection threshold is 0.5. Comparing with the truth diagram, the edge detection result of the filtered image based on the method of the invention is best.
Table 3 shows the quantitative comparison of speckle filtering results of airborne ADTS HH + VV dual-polarized SAR data. In order to objectively and quantitatively compare the performances of the coherent speckle filtering methods, 1 clutter region is arbitrarily selected for equivalent vision ENL analysis. The selected region is shown in FIG. 10(a) and is designated as ROI. The larger the value of the equivalent vision ENL is, the better the speckle suppression performance is. As can be seen from fig. 10, for the ROI region, the ENLs of HH and VV after speckle filtering according to the present invention are 192.67 and 785.05, respectively, which are significantly superior to other comparative methods, and the advantages of the present invention are also confirmed from objective evaluation.
TABLE 3 airborne ADTS HH + VV dual polarized SAR data speckle filtering result quantitative comparison
Figure GDA0002881802280000171
Figure GDA0002881802280000181
Fig. 12 is a comparison graph of speckle filtering results of airborne ADTS HH + HV dual-polarized SAR data. Wherein, fig. 12(a) original image, fig. 12(b)7 × 7Boxcar method filtering result diagram, fig. 12(c)9 × 9Boxcar method filtering result diagram, fig. 12(d)7 × 7 improved Lee method filtering result diagram, fig. 12(e)9 × 9 improved Lee method filtering result diagram, fig. 12(f)9 × 9 improved Sigma method filtering result diagram, fig. 12(g)11 × 11 improved Sigma method filtering result diagram, fig. 12(h) SAR-BM3D method filtering result diagram, fig. 12(i) invention method filtering result diagram, it can be seen from view that, for airborne dual polarization SAR data, the comprehensive performance of the invention in terms of speckle suppression and target detail protection is also significantly better than that of other methods.
Fig. 13 is a diagram of target edge detection results in airborne ADTS HH + HV dual-polarized SAR data. Fig. 13(a1) original image, fig. 13(b1)7 × 7Boxcar method filter result graph, fig. 13(c1)9 × 9Boxcar method filter result graph, fig. 13(d1)7 × 7 modified Lee method filter result graph, fig. 13(e1)9 × 9 modified Lee method filter result graph, fig. 13(f1) edge true value graph; fig. 13(g1)9 × 9 modified Sigma method filter result graph, fig. 13(h1)11 × 11 modified Sigma method filter result graph, fig. 13(i1) SAR-BM3D method filter result graph, fig. 13(j1) method filter result graph of the present invention. Fig. 13(a2) is a graph of an edge detection result based on an original image, fig. 13(b2) is a graph of an edge detection result based on a 7 × 7Boxcar method filtered image, fig. 13(c2) is a graph of an edge detection result based on a 9 × 9Boxcar method filtered image, fig. 13(d2) is a graph of an edge detection result based on a 7 × 7 modified Lee method filtered image, fig. 13(e2) is a graph of an edge detection result based on a 9 × 9 modified Lee method filtered image, and fig. 13(f2) is a graph of an edge true value; fig. 13(g2) is a graph of the edge detection result of the filtered image based on the 9 × 9 improved Sigma method, fig. 13(h2) is a graph of the edge detection result of the filtered image based on the 11 × 11 improved Sigma method, fig. 13(i2) is a graph of the edge detection result of the filtered image based on the SAR-BM3D method, and fig. 13(j2) is a graph of the edge detection result of the filtered image based on the method of the present invention. As can be seen from fig. 13(a1) - (j1), the overall performance of the present invention in terms of speckle suppression and target detail protection is also significantly better for edge-type targets than other approaches. Fig. 13(a2) - (j2) are graphs of edge detection results obtained by a conventional ROA edge detection method (r. touzi, a. lopes, and p. bouquet, "a static and geographic edge detector for SAR images," IEEE trans. geosci. remote sens., vol.26, pp.764-773, Nov 1988.). Wherein, the edge detection threshold is 0.5. Comparing with the truth diagram, the edge detection result of the filtered image based on the method of the invention is best.
Table 4 shows the quantitative comparison of speckle filtering results of airborne ADTS HH + HV dual-polarized SAR data. In order to objectively and quantitatively compare the performance of the speckle filtering methods, 1 clutter region is arbitrarily selected from fig. 10(a) to perform equivalent vision ENL analysis and is marked as ROI. The larger the value of the equivalent vision ENL is, the better the speckle suppression performance is. As can be seen from fig. 10, for the ROI region, the ENLs of HH and HV after speckle filtering according to the present invention are 184.80 and 102.31, respectively, which are significantly superior to other comparative methods, and the advantages of the present invention are also confirmed from objective evaluation.
TABLE 4 airborne ADTS HH + HV dual polarized SAR data speckle filtering result quantitative comparison
Figure GDA0002881802280000191
On the basis, an edge protection index FOM is introduced to carry out quantitative analysis on the edge detection result. Wherein FOM is defined as
Figure GDA0002881802280000192
Wherein N isground-truthIs the number of edge pixels in the edge true value map, NdetectionIs the number of edge pixels in the edge detection result map.
Figure GDA0002881802280000193
The Euclidean distance between the detected edge pixel and the edge pixel nearest to the detected edge pixel in the true value image is used. And alpha is a regulating parameter, and alpha is 1. From the above definitions, the larger the FOM value is, the better the edge detection performance is. When the detection result is completely matched with the truth diagram, the FOM value is 1.
For better contrast performance, consider the case where the detection thresholds of the ROA edge detector are 0.5, 0.6, and 0.7, respectively. As can be seen by comparison, for the vehicle target edge, the FOM of the edge detection result of the filtered image based on the method is superior to that of other methods, and the performance advantage of the method is further proved.
The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above-mentioned embodiments, and all technical solutions belonging to the idea of the present invention belong to the protection scope of the present invention. It should be noted that modifications and embellishments within the scope of the invention may be made by those skilled in the art without departing from the principle of the invention.

Claims (10)

1. A dual-polarization SAR image speckle filtering method combining a context covariance matrix is characterized by comprising the following steps:
step 1: inputting a dual-polarized SAR image to be filtered;
step 2: for pixel point S in dual-polarized SAR imagen,mAt the pixel point Sn,mConstructing a context scattering vector in the win multiplied by win neighborhood, and constructing a context covariance matrix C of the pixel point according to the context scattering vectorCCM-(n,m)N is 1,2, …, N, M is 1,2, …, M, N, M respectively represent the total number of row and column pixels of the dual-polarized SAR image, win is an odd number greater than or equal to 3;
and step 3: calculating a pixel point Sn,mContext covariance matrix C ofCCM-(n,m)And with the pixel point Sn,mContext covariance matrix C of each pixel within a sliding window I J as the centerCCM-(i,j)Similarity parameter lnQij-nmTo obtain a similarity parameter matrix lnQnm-IJI is 1,2, …, I, J is 1,2, …, J, I, J respectively represent the total number of pixels in the row and column of the sliding window, I, J is an odd number;
and 4, step 4: calculating a judgment threshold of the similarity parameter;
and 5: according to the judgment threshold of the similarity parameter, a pixel point S is usedn,mSelecting a similar sample pixel set in a sliding window I multiplied by J as a center, and treating a filtering pixel Sn,mCarrying out filtering treatment;
step 6: and traversing each pixel in the dual-polarization SAR image, and repeating the steps 2 to 5 to obtain an SAR speckle filtering result image.
2. The dual-polarized SAR image speckle filtering method in combination with context covariance matrix according to claim 1, characterized by: the method for constructing the context scattering vector and the context covariance matrix in the step 2 comprises the following steps:
step 2.1: constructing a rule by using a seed pixel point S according to a context scattering vectorn,mSelecting a certain number of pixel points in each vector construction direction in the neighborhood of the core, wherein the pixel points comprise seed pixel points Sn,mAnd neighborhood pixel points thereof, each vector constructing direction context scattering vector
Figure FDA0002881802270000011
The elements in the dual-polarized image are composed of pixel values of pixel points selected from each vector construction direction on a first channel and a second channel of the dual-polarized image, the dimensions of the context scattering vectors in each vector construction direction are consistent, and each context scattering vector contains a seed pixel point Sn,mSeed-removing pixel point S in context scattering vector in all directionsn,mBesides, other neighborhood pixels are not repeatedly used,
Figure FDA0002881802270000012
representing seed pixel Sn,mA vth context scattering vector;
step 2.2: contextual scatter vectors in each vector construction direction
Figure FDA0002881802270000021
Multiplying the obtained data by the conjugate transpose of the data to obtain a context covariance matrix corresponding to the number of the context scattering vectors, and averaging the obtained context covariance matrix to obtain a seed pixel point Sn,mIn this way the context covariance matrix is constructed,
Figure FDA0002881802270000022
v is the total number of constructed context scattering vectors, superscript
Figure FDA0002881802270000023
Representing a conjugate transpose.
3. The dual-polarized SAR image speckle filtering method in combination with context covariance matrix according to claim 2, characterized by:
the context scattering vector construction rule means that the construction direction of the context scattering vector is determined according to the pixel point Sn,mThe core is in a central shape like a Chinese character 'mi' and in an end point radiation shape, and the neighborhood pixel points in each direction in the two types of construction rules are selected according to the neighborhood pixel points relative to the pixel point Sn,mThe positions of the two types of the three-dimensional optical fiber are divided into a symmetrical type and an asymmetrical type.
4. The dual-polarized SAR image speckle filtering method in combination with context covariance matrix according to claim 3, characterized by:
when the construction direction is a central shape like a Chinese character 'mi', and the pixels in each construction direction are selected to be symmetrical, the context scattering vector is expressed as:
Figure FDA0002881802270000024
and the following constraints need to be satisfied:
Figure FDA0002881802270000025
Figure FDA0002881802270000026
i1≥j1,i2≥j2,...,id≥jd (4)
i1, j 1; i2, j 2; ...; id, jd are not zero at the same time (5)
Figure FDA0002881802270000027
D is the number of pixel points selected in the construction direction of the context scattering vector, the superscript T represents transposition,channel-1 and channel-2 represent a first channel and a second channel of a dual-polarized SAR image respectively,
Figure FDA0002881802270000028
representing a pixel S located in the n + id row, m + jd column of a fully polarised SAR imagen+id,m+jdId denotes the pixel value on channel 1, i1, i2n,mJ1, j2,. jd represents the value of the increase in the row on the basis of the coordinates (n, m) ofn,mThe growth values listed on the basis of the coordinates (n, m), d being an integer;
when the construction direction is a central Chinese character 'mi' shape and the pixels in each construction direction are selected to be asymmetric, the context scattering vector is expressed as:
Figure FDA0002881802270000031
and the following constraints need to be satisfied:
Figure FDA0002881802270000032
Figure FDA0002881802270000033
i1≥j1，...id1≥jd1，i2≥j2，...id2≥jd2 (10)
i1, j1 … id1, jd1 and i2, j2 … id2, jd2 are not zero at the same time (11)
i1 ═ i 2., id1 ═ id2, j1 ═ j 2., jd1 ═ jd2 at the same time, which is not true (12)
1≤d1≤D-2，1≤d2≤D-2，d1+d2≤D-2 (13)
i 1.. id 1; i 2.. id2 denotes at the pixel point Sn,mA growth value, j1,. jd1, running on the basis of the coordinates (n, m) of (a); j 2.. jd2 denotes at the pixel point Sn,mThe values of increase listed on the basis of the coordinates (n, m), d1, d2 being integers;
when the build direction is radial to the endpoint, the context scatter vector is expressed as:
Figure FDA0002881802270000034
and the following constraints need to be satisfied:
i＝0,1,2,…,(win-1) (15)
j＝0,1,2,…,(win-1) (16)
0＜k1＜k2＜...＜kd≤1 (17)
i, j, k1. i, k1. j, … kd. i, kd. j are integers (18)
i+j-(win-1)＞0 (19)
i, j are not zero at the same time (20)
d＝D-1 (21)
k1 · i, k1 · j, … kd · i, kd · j respectively represent at the pixel point Sn,mK1,. kd are step growth coefficients of rows and columns, i and j respectively represent the horizontal and vertical coordinates of a Dth pixel point selected in each context scattering vector construction direction relative to the pixel point Sn,mIs offset from the coordinate (n, m).
5. The dual-polarized SAR image speckle filtering method in combination with a context covariance matrix according to any one of claims 1 to 4, characterized by: similarity parameter lnQ is calculated in step 3ij-nmAnd a similarity parameter matrix lnQnm-IJThe method comprises the following steps:
1) calculating CCCM-(n,m)And CCCM-(i,j)Similarity parameter lnQij-nm：
lnQij-nm＝2qln2+ln[Det(CCCM-(i,j))]+ln[Det(CCCM-(n,m))]-2ln[Det(CCCM-(i,j)+CCCM-(n,m))] (22)
Wherein q is a context covariance matrix CCCM-(n,m)D, Det (-) denotes the determinant of the matrix, the symbol ln denotes the natural logarithm, when CCCM-(i,j)＝CCCM-(n,m)lnQij-nm0; when C isCCM-(i,j)≠CCCM-(n,m)lnQij-nm＜0；
2) Similarity parameter matrix lnQnm-IJ：
Traversing each pixel point in the sliding window I multiplied by J, and calculating by the formula (22) to obtain a pixel point S to be filteredn,mSimilarity parameter lnQ with all pixels in sliding window I × Jij-nmForm a similarity parameter matrix lnQ of size I × Jnm-IJ，lnQij-nmIs a similarity parameter matrix lnQnm-IJRow i and column j.
6. The dual-polarized SAR image speckle filtering method in combination with context covariance matrix according to claim 5, characterized by: the calculation method of the decision threshold in the step 4 is
Figure FDA0002881802270000041
Wherein E is an adjusting parameter, the inhibition of balanced coherent speckles and the image detail maintenance are balanced, L is SAR image multi-view, q is a context covariance matrix CCCM-(n,m)Of (c) is calculated.
7. The dual-polarized SAR image speckle filtering method in combination with context covariance matrix according to claim 6, characterized by: the method for selecting the similar sample pixel set in the step 5 comprises the following steps: at the pixel S to be filteredn,mWithin the sliding window of (I) x J, the similarity parameter matrix lnQ is traversednm-IJIf the value of the similarity parameter is greater than the decision threshold Th, the pixel corresponding to the similarity parameter is determined to be a similarity sample of the pixel to be filtered, so as to determine a similarity sample set similar to the pixel to be filtered in the neighborhood, and the similarity sample set is recorded as:
{SCCM-g}＝{Si,j|lnQnm-IJ≥Th}。
8. the dual-polarized SAR image speckle filtering method in combination with context covariance matrix according to claim 7, characterized by: the method for filtering the pixel to be filtered in the step 5 comprises the following steps: let { SCCM-gIf the number of elements in the filter is G, the filter is treatedWave pixel Sn,mResult of filtering processing of
Figure FDA0002881802270000051
Comprises the following steps:
Figure FDA0002881802270000052
9. the dual-polarized SAR image speckle filtering method in combination with context covariance matrix according to claim 5, characterized by: the sliding window I multiplied by J is I, J, and the value of the sliding window I multiplied by J is 15-25.
10. The dual-polarized SAR image speckle filtering method in combination with context covariance matrix according to claim 6, characterized by: and selecting win as 3 and q as 6 as the parameter of the judgment threshold Th, wherein when the image data is ADTS data, the value of L is 1, and the value of E is E
Figure FDA0002881802270000053
When the image data is Radarsat2 data, L is 2 and E is 2
Figure FDA0002881802270000054
CN201910614332.1A 2019-07-09 2019-07-09 Dual-polarization SAR image speckle filtering method combining context covariance matrix Active CN110349105B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910614332.1A CN110349105B (en) 2019-07-09 2019-07-09 Dual-polarization SAR image speckle filtering method combining context covariance matrix

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910614332.1A CN110349105B (en) 2019-07-09 2019-07-09 Dual-polarization SAR image speckle filtering method combining context covariance matrix

Publications (2)

Publication Number Publication Date
CN110349105A CN110349105A (en) 2019-10-18
CN110349105B true CN110349105B (en) 2021-02-26

Family

ID=68177900

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910614332.1A Active CN110349105B (en) 2019-07-09 2019-07-09 Dual-polarization SAR image speckle filtering method combining context covariance matrix

Country Status (1)

Country Link
CN (1) CN110349105B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117310705B (en) * 2023-11-28 2024-02-09 中国石油大学(华东) Flood disaster rapid detection method based on dual-polarized SAR image

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104166856A (en) * 2014-07-30 2014-11-26 西安电子科技大学 Polarization SAR image classification method based on neighbor propagation clustering and region growing
CN106530242A (en) * 2016-10-31 2017-03-22 中国人民解放军国防科学技术大学 Clustering-based polarimetric SAR (synthetic aperture radar) image coherence speckle adaptive filtering method
EP2947799B1 (en) * 2014-05-23 2018-12-05 Thales Method of widely linear turbo-equalisation in a multi-user system and for a multi-antenna receiver

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2947799B1 (en) * 2014-05-23 2018-12-05 Thales Method of widely linear turbo-equalisation in a multi-user system and for a multi-antenna receiver
CN104166856A (en) * 2014-07-30 2014-11-26 西安电子科技大学 Polarization SAR image classification method based on neighbor propagation clustering and region growing
CN106530242A (en) * 2016-10-31 2017-03-22 中国人民解放军国防科学技术大学 Clustering-based polarimetric SAR (synthetic aperture radar) image coherence speckle adaptive filtering method

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
A review of speckle filtering in the context of estimation theory;R.Touzi et al.;《IEEE》;20021231;第265-272页 *
Continuous Estimation Using Context-Dependent Discrete Measurements;Radoslav Ivanov et al;《IEEE》;20190131;第2392-2404页 *
自适应上下文感知相关滤波跟踪;刘波 等;《中国光学》;20190430;第12卷(第2期);第238-253页 *

Also Published As

Publication number Publication date
CN110349105A (en) 2019-10-18

Similar Documents

Publication Publication Date Title
CN108389188B (en) Sparse hyperspectral abnormal target detection method
CN102609701B (en) Remote sensing detection method based on optimal scale for high-resolution SAR (synthetic aperture radar)
CN110619263B (en) Hyperspectral remote sensing image anomaly detection method based on low-rank joint collaborative representation
CN103400388A (en) Method for eliminating Brisk key point error matching point pair by using RANSAC
CN104809698A (en) Kinect depth image inpainting method based on improved trilateral filtering
CN111783583B (en) SAR image speckle suppression method based on non-local mean algorithm
CN102222322A (en) Multiscale non-local mean-based method for inhibiting infrared image backgrounds
CN110349105B (en) Dual-polarization SAR image speckle filtering method combining context covariance matrix
Jeong et al. Improved multiple matching method for observing glacier motion with repeat image feature tracking
CN111008664A (en) Hyperspectral sea ice detection method based on space-spectrum combined characteristics
CN107358214A (en) Polarization SAR terrain classification method based on convolutional neural networks
CN107680061A (en) Dual-polarization SAR image speckle filtering method based on similarity test
CN110100262A (en) For removing image processing equipment, method and the storage medium of cloud from image
Havlen OB stars distribution in Puppis
CN110335214B (en) Full-polarization SAR image speckle filtering method combining context covariance matrix
CN110321904B (en) Single-polarization SAR image speckle filtering method combining context covariance matrix
CN110297241B (en) Context covariance matrix construction method for image processing
CN106199544B (en) Differentiate the Recognition of Radar Target Using Range Profiles method of local tangent space alignment based on core
CN106097292B (en) Space-time neighborhood Gaussian weighted median filtering method for remote sensing image processing
CN111062972A (en) Image tracking method based on image frequency domain conversion
Zhang et al. Feature-band-based unsupervised hyperspectral underwater target detection near the coastline
CN116137067A (en) Hyperspectral image anomaly detection method based on coefficient analysis collaborative representation
Dharani et al. Shadow detection using index-based principal component analysis of satellite images
Ersahin et al. Classification of polarimetric SAR data using spectral graph partitioning
Lee et al. Plant Leaf Area estimation via image segmentation

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant