Polarimetric SAR speckle filtering method based on joint similarity metric

Abstract

The present invention discloses a polarimetric SAR speckle filtering method based on a joint similarity metric, comprising: solving a polarimetric coherence matrix based on a polarimetric scattering matrix corresponding to polarimetric SAR image data; determining an irregular filtering morphological window to rapidly segment the polarimetric SAR image to obtain a plurality of filtering windows; calculating the weighted Euclidean distance and weighted Wishart distance between pixels within the filtering windows based on the polarimetric coherence matrix; establishing a joint similarity metric based on the weighted Euclidean distance and weighted Wishart distance, and selecting filtering sample pixels based on the joint similarity metric; and filtering the polarimetric coherence matrix using the filtered sample pixels to obtain polarimetric SAR data after polarimetric speckle filtering. This method not only effectively suppresses speckle noise, but also has better texture information and polarimetric scattering characteristic preservation capabilities. It also has a simple algorithm and high computational efficiency, making it suitable for large-scale airborne or satellite-borne data processing.

Description

Technical Field

[0001] The invention belongs to the technical field of radar signal processing, and in particular relates to a polarization SAR speckle filtering method based on a joint similarity measurement criterion. Background Art

[0002] Polarimetric Synthetic Aperture Radar (PolSAR) exploits the polarization properties of electromagnetic waves, using different SAR channels to transmit and receive electromagnetic waves of different polarizations. PolSAR is highly sensitive to the scattering and structural characteristics of targets and is widely used in areas such as object decomposition, object classification and interpretation, scattering mechanism analysis, and hidden target monitoring. However, due to the unique coherent imaging mechanism of SAR systems, polarimetric SAR data inherently suffers from speckle noise, significantly increasing the difficulty of image interpretation and processing. Therefore, research on polarimetric SAR speckle reduction techniques is crucial for improving polarimetric SAR data processing and analysis capabilities.

[0003] Early research on polarimetric SAR speckle filtering involved the polarimetric whitening filter (PWF) proposed by Novak et al. This filter generates de-noised polarimetric SAR images by optimizing the combination of elements in the polarimetric covariance. Liu et al. later extended the PWF filter to multi-look polarimetric SAR data, primarily by performing multi-look processing on the covariance matrix. However, this method suffers from the drawback that it only produces a single de-noised image using the covariance matrix and cannot filter individual elements within the covariance matrix. Lee et al. proposed a linear filter based on a multiplicative speckle noise model that can reduce speckle for data from the HH, HV, and VV polarization channels, but does not filter the off-diagonal elements of the single-look covariance matrix. These filters all exploit the statistical correlation between the three polarization channels. Their drawback is that they cannot address crosstalk between polarization channels. To address these issues, Lee et al. proposed a minimum mean square error (MMSE) filter. This filter, for the first time, utilizes edge-aligned windows to preserve image detail information and adaptively selects edge windows in different directions. This filter is known as the refined Lee filter. The greatest advantage of this filter is that it preserves image edge information while reducing the effects of speckle noise. However, due to the fixed edge window pattern, its edge preservation capability in complex scenes is very limited. Therefore, when the scene becomes complex, the refined Lee filter cannot fully match the edge characteristics of the ground object. Subsequently, Lee et al. proposed an improved sigma filter, which first utilizes the statistical characteristics of distributed targets. By combining the intensity information of three polarization scattering mechanisms (surface scattering, secondary scattering, and volume scattering) with their probability distribution curves, pixels within the window are selected and then averaged. This filter largely preserves pixels whose statistical characteristics are consistent with the target pixels, while also preserving some polarization scattering characteristics. However, it cannot preserve polarization information within the off-diagonal elements of the covariance matrix or coherence matrix.

[0004] In summary, traditional polarimetric SAR speckle filtering methods either cannot effectively filter out speckle noise or have poor performance in preserving image edge texture features and polarization scattering characteristics. In addition, the related algorithms are too complex and computationally inefficient, making them less practical for large-scale airborne or satellite-borne data preprocessing. Summary of the Invention

[0005] In order to solve the problem that the traditional polarimetric SAR speckle filtering method has defects in maintaining the performance of image edge texture features and polarimetric scattering characteristics, the present invention proposes a polarimetric SAR speckle filtering method based on a joint similarity metric.

[0006] The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0007] A polarimetric SAR speckle filtering method based on a joint similarity metric comprises:

[0008] Step 1: Solve the polarization coherence matrix based on the polarization scattering matrix corresponding to the polarimetric SAR image data;

[0009] Step 2: Determine the irregular filtering morphological window to quickly segment the polarimetric SAR image and obtain several filtering windows;

[0010] Step 3: Calculating the Euclidean distance and Wishart distance between pixels in the filter window based on the polarization coherence matrix, and calculating the corresponding weighted Euclidean distance coefficient and weighted Wishart distance coefficient;

[0011] Step 4: establishing a joint similarity measurement criterion based on the weighted Euclidean distance coefficient and the weighted Wishart distance coefficient, and selecting filtering sample pixels according to the joint similarity measurement criterion;

[0012] Step 5: Filter the polarization coherence matrix using the filtered sample pixels to obtain polarization SAR data after polarization coherence speckle filtering.

[0013] Beneficial effects of the present invention:

[0014] The method provided by the present invention is based on the polarimetric coherence matrix of polarimetric SAR image data. It utilizes weighted Euclidean distance and weighted Wishart distance to jointly design a similarity metric to select filtering sample pixels, thereby realizing coherent speckle filtering of airborne and spaceborne polarimetric SAR data. This method can not only ensure the effective suppression of coherent speckle noise, but also has better texture information and polarization scattering characteristics preservation capabilities. The algorithm is simple and the computational efficiency is high, making it suitable for large-scale airborne or spaceborne data processing.

[0015] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of a polarimetric SAR speckle filtering method based on a joint similarity metric provided by an embodiment of the present invention;

[0017] Figure 2 This is another flowchart of a polarization SAR speckle filtering method based on a joint similarity metric provided by an embodiment of the present invention;

[0018] Figure 3 2 is a schematic diagram of a multi-directional edge detection filter window provided by an embodiment of the present invention;

[0019] Figure 4This is the optical image and polarimetric SAR data power diagram corresponding to the test scene in Meishan, Sichuan in simulation experiment 1;

[0020] Figure 5 This is the Pauli decomposition pseudo-color composite image of the airborne data under different filtering methods in simulation experiment 1;

[0021] Figure 6 This is the SSF statistical curve of the airborne data after filtering by different methods in simulation experiment 1;

[0022] Figure 7 This is the optical image and polarimetric SAR data power diagram corresponding to the GF-3 test scene in the San Francisco area of ​​the United States in simulation experiment 2;

[0023] Figure 8 This is the Pauli decomposition pseudo-color composite image of the spaceborne data under different filtering methods in simulation experiment 2;

[0024] Figure 9 This is the SSF statistical curve of the satellite data after filtering by different methods in simulation experiment 2. DETAILED DESCRIPTION

[0025] The present invention will be further described in detail below with reference to specific examples, but the embodiments of the present invention are not limited thereto.

[0026] Example 1

[0027] Please see the joint Figure 1 and Figure 2 , Figure 1 This is a flow chart of a polarimetric SAR speckle filtering method based on a joint similarity metric provided by an embodiment of the present invention; Figure 2 This is another flowchart of a polarimetric SAR speckle filtering method based on a joint similarity metric provided by an embodiment of the present invention. The polarimetric SAR speckle filtering method based on a joint similarity metric provided by this embodiment specifically includes the following steps:

[0028] Step 1: Solve the polarimetric coherence matrix based on the polarimetric scattering matrix corresponding to the polarimetric SAR image data.

[0029] 11) Obtain the polarization scattering matrix S of the polarization SAR image data in the backscattering coordinate system.

[0030] Specifically, in the backscatter coordinate system, the single-look polarimetric SAR data can be expressed using the polarimetric scattering matrix S as follows:

[0031]

[0032] Among them, S HH 、S HV 、SVH 、S VV Represent the complex scattering coefficients under HH, HV, VH, and VV channels respectively.

[0033] 12) Solve the polarization coherence matrix T based on the polarization scattering matrix S.

[0034] Specifically, under the reciprocity assumption, the polarization covariance matrix C and the polarization coherence matrix T are defined as:

[0035]

[0036]

[0037] Where k and Ω are the target scattering vectors under the Pauli basis and Lexicographic basis, respectively, and are expressed as follows:

[0038]

[0039] [·] * 、[·] H 、[·] T denote conjugate, conjugate transpose, and transpose operations, respectively. <·> is the spatial statistical average under the assumption of anisotropy of random scattering media.

[0040] It can be seen from Equations (3) and (4) that the polarization coherence matrix T and the polarization covariance matrix C are obtained from different basis vectors, and the two can be converted into each other through matrices. Therefore, this embodiment will subsequently mainly perform polarization coherence speckle filtering based on the coherence matrix T.

[0041] Step 2: Determine the irregular filtering morphological window to quickly segment the polarimetric SAR image and obtain several filtering windows.

[0042] This embodiment first performs multi-directional proportional edge detection on the polarimetric SAR power image to extract the proportional edge intensity map of the polarimetric SAR image. Then, a threshold processing method is used to obtain the image edge detection result. Finally, the edge detection result is segmented using the watershed algorithm to obtain the fast superpixel segmentation result of the polarimetric SAR image. The details are as follows:

[0043] 21) Multi-directional proportional edge detection is performed on the polarimetric SAR power image to extract the proportional edge intensity map of the polarimetric SAR image, and the image edge detection result is obtained using the threshold processing method.

[0044] 21a) Set the multi-directional proportional edge detector to K f ={l,w,d,θ f}; where l, w, d, θ fRepresents the length, width, width between two rectangles and detector direction of the edge detector respectively. Figure 3 , Figure 3 3 is a schematic diagram of a multi-directional edge detection filter window provided by an embodiment of the present invention.

[0045] 21b) For a particular detector direction θ f , calculate the mean value of the pixels in the rectangular area on both sides of the central pixel (x, y) and And calculate θ f The ratio edge strength map (RESM) map r(x,y,θ f ), the calculation formula is:

[0046]

[0047] 21c) Based on the proportional edge intensity map, the image edge detection result is constructed using the thresholding method.

[0048] Specifically, K is the number of directions of the multi-directional proportional edge detector. The RESM of the image is constructed using the thresholding method:

[0049]

[0050] Among them, the threshold T α is the value of the statistical histogram of g(x,y) at the upper α% percentile. The α value is positively correlated with the initial segmentation pixel size of the image and can be adaptively selected according to the filter window size. τ represents the value greater than or equal to the α% percentile, and N1 represents the number of image pixels.

[0051] 22) The edge detection results are segmented using the watershed algorithm to obtain the superpixel segmentation results of the polarimetric SAR image and obtain several filtering windows.

[0052] Step 3: Calculate the Euclidean distance and Wishart distance between pixels in the filter window based on the polarization coherence matrix, and calculate the corresponding weighted Euclidean distance coefficient and weighted Wishart distance coefficient.

[0053] In polarimetric SAR image speckle filtering, in addition to preserving the scene's texture information, it's also necessary to preserve the polarimetric scattering characteristics of the objects in the SAR image. Both the polarimetric coherence matrix and the covariance matrix obey a multivariate complex Wishart distribution. Furthermore, the Wishart distance is independent of the number of views and can accurately describe the polarimetric similarity between the pixel to be filtered and the pixels within the filtering window in a single-look polarimetric SAR image. Therefore, this embodiment constructs a joint similarity metric based on the weighted Euclidean distance and the Wishart distance metrics.

[0054] Specifically, assume that the pixel to be filtered is M(x0, y0), and the pixel to be selected for filtering is N(x n ,y n ), then the Euclidean distance between the two is d E It can represent:

[0055]

[0056] The weighted Euclidean distance coefficient after Gaussian weighting is:

[0057]

[0058] Where S ED ∈(0,1]; σ is the scale factor, which determines the distribution of the weighted Euclidean distance coefficient and can be selected according to the size of the filter window.

[0059] Correspondingly, M(x0,y0) and N(x n ,y n ) can be expressed as:

[0060]

[0061] Among them, T0 and T n Represent M(x0,y0) and N(x n ,y n ) is the single-look polarization coherence matrix at .

[0062] The Wishart distance is normalized and Gaussian weighted to obtain the weighted Wishart distance coefficient S WD :

[0063]

[0064] Step 4: Establish a joint similarity measurement criterion based on the weighted Euclidean distance coefficient and the weighted Wishart distance coefficient, and select the filtering sample pixels according to the joint similarity measurement criterion.

[0065] 4a) Combined with the weighted Euclidean distance coefficient S ED and weighted Wishart distance coefficient S WD , establish the joint similarity measurement criterion and obtain the joint similarity measurement coefficient S, which is expressed as:

[0066] S=S ED ·S WD (11)

[0067] As can be seen, the value range of S is (0, 1). When S is closer to 1, the similarity between pixels is higher; when S is closer to 0, the similarity between pixels is lower. Therefore, the filter samples can be effectively selected based on their S values.

[0068] 4b) Arrange the joint similarity measurement coefficients S in descending order, and select filtering sample pixels that meet the conditions according to the preset filtering view size.

[0069] In this embodiment, the joint similarity metric coefficient S includes the physical distance metric and the polarization distance metric between pixels, which can accurately describe the actual physical distance similarity and polarization scattering similarity between pixels. The present invention defines the sample selection constraint of the polarization coherent speckle filtering as a joint similarity distance metric criterion.

[0070] Calculate the weighted Euclidean distance and Wishart between the pixel to be filtered and all pixels in the morphological window, determine the joint similarity S of all pixels to the pixel to be filtered according to the joint similarity measurement criterion, sort S in descending order, and select the filtering sample pixels that meet the conditions according to the initially set filter view size.

[0071] Step 5: Filter the polarization coherence matrix using the filtered sample pixels to obtain polarization SAR data after polarization coherence spot filtering.

[0072] Furthermore, after completing step 5, the following steps may also be included:

[0073] Step 6: Evaluate the effect of polarimetric SAR speckle filtering.

[0074] In order to further demonstrate the performance of speckle noise removal and edge feature preservation among different filters, this embodiment uses two different evaluation indicators to quantitatively evaluate the performance of each filter. One is

[0075] The equivalent number of looks (ENL) and the edge preserving index (EPI) are used.

[0076] The equivalent number of views is defined as the ratio of the mean square of the distributed target intensity to the variance:

[0077]

[0078] Where E(I) represents the mean of the distributed target intensity, and var(I) represents the variance of the distributed target intensity. It characterizes the filter's ability to smooth speckle noise in homogeneous regions. A larger ENL indicates more pronounced smoothing and better speckle noise suppression.

[0079] The edge preservation index is defined as the ratio of the cumulative gradient changes in azimuth and range before and after filtering:

[0080]

[0081] Among them, p b (i, j) is the pixel value of the image before filtering, p a (i, j) represents the pixel value of the filtered image, M and N represent the number of pixels in the range and azimuth directions, respectively. The EPI value represents the degree of edge preservation in the filtered and smoothed image, and ranges from [0 to 1]. A larger EPI indicates that the filtered image is closer to the original image and has a stronger edge preservation capability.

[0082] In another embodiment of the present invention, the polarization scattering similarity factor SSF may be used to evaluate the ability to maintain the polarization scattering characteristics.

[0083] The SSF size represents the degree of preservation of the polarization scattering mechanism. Assume that the target pixel to be filtered is u and the pixel to be selected for filtering is t. Similar to the representation of the interference coherence coefficient, the SSF is expressed as:

[0084]

[0085] Where p is a vector containing all elements in the polarization coherence matrix, reflecting all polarization scattering information at the target pixel, conj(·) represents conjugation, ||·|| is the vector binorm, and * represents the inner product of two vectors. The value range of SSF is [0,1]. When SSF=1, it means that the polarization scattering information contained in the two pixels is exactly the same; when SSF=0, it means that the polarization scattering information is completely different. p is expressed as follows:

[0086]

[0087] The method provided by the present invention is based on the polarimetric coherence matrix of polarimetric SAR image data. It utilizes weighted Euclidean distance and weighted Wishart distance to jointly design a similarity metric to select filtering sample pixels, thereby realizing coherent speckle filtering of airborne and spaceborne polarimetric SAR data. This method can not only ensure the effective suppression of coherent speckle noise, but also has better texture information and polarization scattering characteristics preservation capabilities. The algorithm is simple and the computational efficiency is high, making it suitable for large-scale airborne or spaceborne data processing.

[0088] Example 2

[0089] The beneficial effects of the present invention are verified and explained through simulation experiments below.

[0090] Simulation Experiment 1: Airborne Data Verification

[0091] This airborne embodiment uses airborne polarimetric SAR data collected by Xidian University and the Institute of Electronics of the Chinese Academy of Sciences in Meishan, Sichuan Province to perform polarimetric coherent speckle filtering. Figure 4 This is the optical image and polarimetric SAR data power diagram corresponding to the test scene in Meishan, Sichuan Province in the simulation experiment; the ground objects in the scene are mainly large dense eucalyptus shrubs and regularly distributed economic crops citrus trees. In order to clearly display the texture and other details before and after filtering, the image is captured. Figure 4 The image within the rectangular frame is filtered and the data size is 1500 × 1500 pixels. The basic parameter information of the scene is shown in Table 1. Figure 4 Figure (a) is the regional optical map, which is used to perform multi-directional proportional edge detection to obtain the scene polarization SAR data power map, such as Figure 4 As shown in Figure (b).

[0092] Table 1 Basic parameters of data in Meishan, Sichuan

[0093] Scene Parameters Numerical Center frequency (GHz) 9.60 Polarization Full polarization Average elevation (m) 554.46 Resolution (azimuth × distance) 0.2m×0.2m Scene center position 29.94°N,103.53°E Data size (azimuth × distance) 1500×1500 pixels

[0094] Evaluation of airborne filtering effects. Figure 5 , Figure 5 This is a pseudo-color composite image of Pauli decomposition under different filtering methods of airborne data in the simulation experiment. Figure 5 Figure (a) is the Pauli decomposition diagram of the original data. It can be seen that the original SAR data has a high signal-to-noise ratio and the outlines between vegetation are clear, but the influence of coherent speckle noise still exists. Figure 5 Figure (b) shows that the rectangular window filter effectively removes speckle noise, but the vegetation texture is blurred and the gaps between trees are unclear. Furthermore, due to inherent flaws, a noticeable "blocking effect" appears around strong scattering points after the rectangular window filter. Figure 5 Figure (c) shows that the modified polarization Lee filter has a strong ability to suppress coherent speckles in polarimetric SAR images. Compared with Figure (a), the gaps between trees are significantly clearer after filtering. Figure 5 As can be seen from Figure (d), the filter not only has a strong ability to remove coherent speckle noise, but also can better maintain the edge features of the ground objects. In terms of maintaining the vegetation contour and ground texture features, it shows better performance than other filters.

[0095] To further demonstrate the performance of speckle noise removal and edge feature preservation among different filters, this experiment used the equivalent view number ENL and edge preservation index EPI to evaluate the filtering effect. The results are shown in Tables 2 and 3.

[0096] Table 2 Comparison of equivalent view counts for different filtering methods

[0097]

[0098] Table 3 Comparison of edge preservation index of different filtering methods

[0099]

[0100] The above filtering methods are quantitatively evaluated using equivalent view counts and edge preservation indices. It can be seen that the equivalent view count of the filtering method based on the joint similarity measurement criterion is significantly greater than that of the other filtering methods. Compared with the original data, the equivalent view count is increased by 17.63% and 42.13% compared with the mean filtering and refined LEE filtering, respectively. The edge preservation index is increased by 74.40% and 30.77% compared with the mean filtering and refined LEE filtering, respectively, which proves the effectiveness of the method of the present invention for coherent speckle suppression and texture information preservation.

[0101] In addition, this experiment also used the polarization scattering similarity factor (SSF) to evaluate the ability to maintain polarization scattering characteristics. Figure 6 , Figure 6 This is the SSF statistical curve of the airborne data after filtering by different methods in the simulation experiment.

[0102] It can be seen that the SSF value of the method of the present invention is significantly improved compared with the other two filtering methods, which proves that the polarization scattering mechanism preservation performance of the method of the present invention is better than that of the other methods.

[0103] Simulation Experiment 2: Onboard Data Verification

[0104] This satellite-borne embodiment is further verified using C-band polarimetric SAR data of the San Francisco area of ​​the United States acquired by the domestic Gaofen-3 satellite. Figure 7 The optical image and polarimetric SAR data power map corresponding to the scene include ocean, vegetation, urban buildings, streets and other landforms. The buildings are regularly distributed and have obvious texture features. Figure 7 The image within the rectangular frame is filtered and the data size is 1500 × 1500 pixels. The basic parameter information of the scene is shown in Table 4. Figure 7 (a) is the regional optical map, and the scene polarization SAR data power map is obtained for multi-directional proportional edge detection, such as Figure 7 As shown in Figure (b).

[0105] Table 4 Basic parameters of data for San Francisco, USA

[0106]

[0107]

[0108] Evaluation of spaceborne filtering effects. Figure 8 , Figure 8This is the Pauli decomposition pseudo-color composite image of the satellite data under different filtering methods in simulation experiment 2. Figure 8 Figure (a) is the Pauli decomposition diagram of the original data. It can be seen that this data is more seriously affected by coherent speckle noise than the airborne data in the previous section. Figure 8 Figure (b) shows that the rectangular window filter can effectively filter out the coherent speckle noise, but the filtering results in excessive smoothing of the image and the texture of the building appears blurred. Figure 8 Figure (c) shows that the modified polarization Lee filter maintains the building edge better than the rectangular window at the boundary between vegetation and building areas. However, the texture of streets and houses in the center of the building area is severely deformed, and there is still blurring in the dense building area after filtering. For the joint pixel metric criterion filtering, Figure 8 As can be seen from Figure (d), the gaps between urban building blocks after filtering are also very clear, and the coherent speckle noise removal effect is better.

[0109] Correspondingly, this experiment also used the equivalent view number ENL, edge preservation index EPI and polarization scattering similarity factor SSF to evaluate the filtering effect. The results are shown in Table 5, Table 6 and Figure 8 shown.

[0110] Table 2 Comparison of equivalent view counts for different filtering methods

[0111]

[0112] Table 3 Comparison of edge preservation index of different filtering methods

[0113]

[0114]

[0115] Tables 5 and 6 show the comparison results of the equivalent view count and edge preservation index of different filtering methods for San Francisco area data. Figure 9 Figure 2 shows the SSF statistics of spaceborne data filtered by different methods in Simulation Experiment 2. Comparative analysis shows that rectangular window filtering maximizes the ENL value, but due to oversmoothing, the image edge preservation index is significantly lower than that of other methods. The equivalent view count of the refined polarization Lee filter is slightly better than that of the filter based on the joint similarity metric. However, intuitively, the edge information of the building area is significantly destroyed, and its performance in preserving texture details is worse.

[0116] In addition, for the polarization scattering preservation ability, the SSF curves of the rectangular window filtering and the refined polarization Lee filtering are almost identical, with a peak value of 0.78, while the peak value of the method of the present invention is 0.93, indicating that the polarization scattering characteristics remain basically unchanged before and after filtering by the method of the present invention.

[0117] The above is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention should not be considered to be limited to these descriptions. For those skilled in the art of the present invention, without departing from the concept of the present invention, several simple deductions or substitutions can be made, which should be considered to fall within the scope of protection of the present invention.

Claims

1. A polarimetric SAR speckle filtering method based on a joint similarity metric, characterized in that: include: Step 1: Solve the polarization coherence matrix based on the polarization scattering matrix corresponding to the polarimetric SAR image data; Step 2: Determine the irregular filtering morphological window to quickly segment the polarimetric SAR image and obtain several filtering windows; include: 21) Perform multi-directional proportional edge detection on the polarimetric SAR power image to extract the proportional edge intensity map of the polarimetric SAR image, and obtain the image edge detection result using a threshold processing method; 22) Segmenting the edge detection result using a watershed algorithm to obtain a superpixel segmentation result of the polarimetric SAR image and obtain a plurality of filtering windows; Step 3: Calculating the Euclidean distance and Wishart distance between pixels in the filter window based on the polarization coherence matrix, and calculating the corresponding weighted Euclidean distance coefficient and weighted Wishart distance coefficient; Step 4: establishing a joint similarity measurement criterion based on the weighted Euclidean distance coefficient and the weighted Wishart distance coefficient, and selecting filtered sample pixels according to the joint similarity measurement criterion; including: 4a) Combined with the weighted Euclidean distance coefficient and the weighted Wishart distance coefficient , establish the joint similarity measurement criterion and obtain the joint similarity measurement coefficient , whose expression is: 4b) The joint similarity measurement coefficient Arrange in descending order and select the filtering sample pixels that meet the conditions according to the preset filtering view size; Step 5: Filter the polarization coherence matrix using the filtered sample pixels to obtain polarization SAR data after polarization coherence speckle filtering.

2. The polarimetric SAR speckle filtering method based on a joint similarity metric according to claim 1, wherein: Step 1 includes: 11) Obtain the polarization scattering matrix of polarization SAR image data in the backscattering coordinate system , whose expression is: in, 、 、 、 Respectively Complex scattering coefficient under the channel; 12) Based on the polarization scattering matrix Solving the polarization coherence matrix , the calculation formula is: in, represents the target scattering vector under the Pauli basis, and its expression is: 、 、 Represent conjugate, conjugate transpose, and transpose operations respectively, is the spatial statistical average under the assumption of anisotropy of random scattering media.

3. The polarimetric SAR speckle filtering method based on a joint similarity metric according to claim 1, wherein: Step 21) includes: 21a) Set the multi-directional proportional edge detector to ;in, They represent the length, width, width between two rectangles and detector direction of the edge detector respectively; 21b) For a particular detector direction , calculate its center pixel The mean value of the pixels in the rectangular area on both sides and , and calculate Directional proportional edge strength mapping , the calculation formula is: 21c) Based on the proportional edge intensity map, a thresholding method is used to construct an image edge detection result, which is expressed as: in, Indicates the center pixel calculate Image edge detection results in direction; Threshold for The statistical histogram of The value at the quantile, The value is positively correlated with the initial segmentation pixel size of the image and can be adaptively selected according to the filter window size. Indicates greater than or equal to The value at the quantile; N 1 represents the number of image pixels.

4. The polarimetric SAR speckle filtering method based on a joint similarity metric according to claim 1, wherein: In step 3, calculating the Euclidean distance and the Wishart distance between pixels in the filtering window based on the polarization coherence matrix includes: Assume that the pixel to be filtered is , the pixels to be selected for filtering are , then the weighted Euclidean distance is expressed as , and its calculation formula is: The weighted Wishart distance is expressed as , and its calculation formula is: in, and Respectively and The single-look polarimetric coherence matrix at .

5. The polarimetric SAR speckle filtering method based on a joint similarity metric according to claim 4, wherein: In step 3, the calculation formulas of the weighted Euclidean distance coefficient and the weighted Wishart distance coefficient are respectively: in, represents the weighted Euclidean distance coefficient, represents the weighted Euclidean distance, represents the weighted Wishart distance coefficient, represents the weighted Wishart distance, Represents the scale factor.

6. The polarimetric SAR speckle filtering method based on a joint similarity metric according to claim 1, wherein: After step 5, the following steps are also included: Step 6: Evaluate the effect of polarimetric SAR speckle filtering.

7. The polarimetric SAR speckle filtering method based on a joint similarity metric according to claim 6, wherein: Step 6 includes: The equivalent number of views (ENL) and edge preservation index (EPI) are used to quantitatively evaluate the effect of polarimetric SAR speckle filtering. The equivalent number ENL is defined as the ratio of the mean square of the distributed target intensity to the variance, that is: represents the mean of the distribution target intensity, represents the variance of the distribution target intensity; The edge preservation index EPI is defined as the ratio of the cumulative gradient changes in azimuth and distance before and after filtering, that is: in, is the pixel value of the image before filtering, is the pixel value of the filtered image, M, N Represents the number of pixels in the range and azimuth directions respectively.

8. The polarimetric SAR speckle filtering method based on a joint similarity metric according to claim 6, wherein: Step 6 also includes: The polarization scattering similarity factor (SSF) is used to evaluate the ability to maintain the polarization scattering characteristics. The expression of the polarization scattering similarity factor SSF is: in, is a vector containing all elements in the polarization coherence matrix, with subscript is the target pixel to be filtered, subscript is the pixel to be selected for filtering, represents conjugation, is the vector two-norm.

Images