SAR three-dimensional false target suppression method and system based on essential matrix and neighborhood consistency constraint

Abstract

The invention discloses an SAR three-dimensional false target suppression method and system based on an essential matrix and neighborhood consistency constraint, and relates to the field of false target suppression. The objective of the invention is to solve the problem that an existing false target elimination method is poor in false target discrimination capability. The method comprises the following steps: acquiring two-dimensional coordinates of to-be-estimated pixel points on a main image and an auxiliary image and a two-dimensional coordinate set of neighborhood pixel points, and acquiring three-dimensional coordinates of the to-be-estimated pixel points and three-dimensional coordinates of neighborhood pixel points by utilizing the two-dimensional coordinates of the to-be-estimated pixel points on the main image and the auxiliary image and the two-dimensional coordinate set of the neighborhood pixel points; constructing a measurement function by using the three-dimensional coordinates of the to-be-estimated pixel and the three-dimensional coordinates of the neighborhood pixel points so as to obtain a measurement value; and determining and storing the pixels of the real target according to the measurement function value. The method is used for suppressing false targets.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of false target suppression, in particular to a SAR three-dimensional false target suppression method and system based on an essential matrix and neighborhood consistency constraint. BACKGROUND

[0002] The false target of the forward-looking SAR often appears due to the imaging condition limitation, which causes the three-dimensional reconstruction error. Since the SAR image has the singularity function expansion characteristic, and the target projection point is within a SAR image pixel point in the height resolution, it is inevitable to estimate the target point with different heights for one pixel point. In view of the false target characteristics, the false target suppression method is developed to solve the problem of the false target point in the forward-looking SAR imaging process, and to improve the three-dimensional reconstruction performance, which is an important research content in the field.

[0003] The prior art mainly adopts the false point elimination method based on the height estimation of the neighborhood information. First, the measure function is determined for measuring whether the target height is estimated incorrectly. Then, after obtaining the error target point, the false point is re-registered, and finally the target height is estimated to complete the false point elimination. This method needs to meet certain prior conditions, that is, the amplitude change of the adjacent pixel points in the multi-angle SAR image is small, which indicates that the distribution of the target in the three-dimensional space is similar. The target distribution in the surface target is adjacent in the SAR image, and the normal vector of the surface target is the same, so the target scattering coefficient change in the multi-angle SAR is similar. However, this method is difficult to effectively cope with the inherent speckle noise in the SAR image, the overlay, and the nonlinear coupling effect caused by the complex three-dimensional structure, so the existing method has poor discrimination ability for the false target. SUMMARY

[0004] The present application is to solve the problem of poor discrimination ability of the existing false target elimination method for the false target, and proposes a SAR three-dimensional false target suppression method and system based on an essential matrix and neighborhood consistency constraint.

[0005] The SAR three-dimensional false target suppression method based on the essential matrix and the neighborhood consistency constraint comprises:

[0006] Step one, obtaining the two-dimensional coordinates of the pixel point to be estimated on the main image and the secondary image and the two-dimensional coordinate set of the neighborhood pixel point, and obtaining the three-dimensional coordinates of the pixel point to be estimated and the three-dimensional coordinates of the neighborhood pixel point by using the two-dimensional coordinates of the pixel point to be estimated on the main image and the secondary image and the two-dimensional coordinate set of the neighborhood pixel point;

[0007] Step two, constructing a measure function by using the three-dimensional coordinates of the pixel to be estimated and the three-dimensional coordinates of the neighborhood pixel point, so as to obtain the measure value

[0008] ​Step three, if is greater than the preset threshold value , then the three-dimensional coordinates of the current pixel to be estimated are considered as the pixel coordinates of the false target, and step one is returned to reacquire the two-dimensional coordinates of the pixel to be estimated and the two-dimensional coordinate set of the neighborhood pixel points in the secondary image; otherwise, it is considered that the three-dimensional coordinates of the current pixel to be estimated are the pixel coordinates of the real target, and the three-dimensional coordinates of the pixel to be estimated are saved.

[0009] Further, the two-dimensional coordinates of the pixel to be estimated and the two-dimensional coordinate set of the neighborhood pixel points in the primary image and the secondary image in step one are obtained, and the three-dimensional coordinates of the pixel to be estimated and the three-dimensional coordinates of the neighborhood pixel points are obtained using the two-dimensional coordinates of the pixel to be estimated and the two-dimensional coordinate set of the neighborhood pixel points in the primary image and the secondary image, specifically:

[0010] Step one, obtain the two-dimensional coordinates of the pixel to be estimated in the primary image and the two-dimensional coordinate set of the neighborhood pixel points ;

[0011] wherein, is the neighborhood coordinate point label, is the total number of neighborhood coordinate points, is the two-dimensional coordinate of the th neighborhood pixel point of the pixel to be estimated in the primary image;

[0012] Step two, register the primary image and the secondary image using the SIFT method to obtain the corresponding pixel point of the pixel to be estimated in the secondary image , and obtain the two-dimensional coordinate set of the neighborhood pixel points in the secondary image ;

[0013] wherein, is the two-dimensional coordinate of the th neighborhood pixel point in the secondary image;

[0014] Step three, obtain the three-dimensional coordinates of the pixel to be estimated using and , and obtain the three-dimensional coordinates of the neighborhood pixel points using the pixel coordinates in and .

[0015] Further, in step three, the three-dimensional coordinates of the pixel to be estimated are obtained using and , and the three-dimensional coordinates of the neighborhood pixel points are obtained using the pixel coordinates in and , specifically: ​

[0016] First, a matrix is constructed by using the two-dimensional coordinates of the to-be-estimated pixel and the neighborhood pixels in the primary image and the secondary image , and the matrix is normalized to obtain a normalized matrix ;

[0017] The matrix is as follows:

[0018] ,

[0019] wherein, is the two-dimensional coordinate of the pixel in the primary image, is the two-dimensional coordinate of the pixel in the secondary image, , and are the row vectors of the primary image corresponding to the essential matrix , , and are the row vectors of the secondary image corresponding to the essential matrix ;

[0020] Then, the singular value decomposition is performed on the normalized matrix to obtain a right singular value vector matrix , and the last column of is taken as the three-dimensional coordinate of the pixel and the three-dimensional coordinate of the neighborhood pixel.

[0021] Further, the singular value decomposition performed on the matrix is specifically as follows:

[0022] ,

[0023] wherein, is a left singular value vector matrix, is a singular value matrix, is a right singular value vector matrix.

[0024] Further, the step two of constructing a measure function by using the three-dimensional coordinate of the to-be-estimated pixel and the three-dimensional coordinate of the neighborhood pixel to obtain a measure value is specifically as follows:

[0025] ,

[0026] wherein, is the three-dimensional coordinate of the to-be-estimated pixel, is the label of the neighborhood pixel, is the first Coordinates of a neighborhood pixel point, is a variance representing a normal distribution.

[0027] Further, the variance representing the normal distribution Specifically,

[0028] ,

[0029] ,

[0030] ,

[0031] wherein, is a constant, is a pixel label, is a total number of pixels, is the pixel value in the th position, is the pixel value in the th position, , is the intermediate vector.

[0032] The SAR three-dimensional false target suppression system based on the essential matrix and neighborhood consistency constraint comprises a pixel three-dimensional coordinate acquisition module, a measure value acquisition module and a real target pixel point saving module.

[0033] The pixel three-dimensional coordinate acquisition module is used for acquiring two-dimensional coordinates of a to-be-estimated pixel point on a primary image and a secondary image and a two-dimensional coordinate set of neighborhood pixel points, and acquiring three-dimensional coordinates of the to-be-estimated pixel point and three-dimensional coordinates of neighborhood pixel points by using the two-dimensional coordinates of the to-be-estimated pixel point on the primary image and the secondary image and the two-dimensional coordinate set of the neighborhood pixel points.

[0034] The measure value acquisition module obtains a measure value by using the three-dimensional coordinates of the to-be-estimated pixel and the three-dimensional coordinates of the neighborhood pixel points.

[0035] The real target pixel point saving module judges whether the to-be-estimated pixel point is a real target pixel point coordinate by using the measure value , and if so, the real target pixel point is saved; otherwise, the pixel three-dimensional coordinate acquisition module is returned to.

[0036] Further, the pixel three-dimensional coordinate acquisition module is used for acquiring two-dimensional coordinates of a to-be-estimated pixel point on a primary image and a secondary image and a two-dimensional coordinate set of neighborhood pixel points, and acquiring three-dimensional coordinates of the to-be-estimated pixel point and three-dimensional coordinates of neighborhood pixel points by using the two-dimensional coordinates of the to-be-estimated pixel point on the primary image and the secondary image and the two-dimensional coordinate set of the neighborhood pixel points, and specifically,

[0037] a1, obtaining a pixel point to be estimated in a main image a set of two-dimensional coordinates of neighborhood pixel points in the main image ;

[0038] wherein, is a neighborhood coordinate point label, is a total number of neighborhood coordinate points, is a first neighborhood pixel point two-dimensional coordinate of the pixel point to be estimated in the main image ;

[0039] a2, using a SIFT method to register the main image and the secondary image, obtaining a corresponding pixel point of the pixel point to be estimated in the secondary image , and obtaining a set of two-dimensional coordinates of neighborhood pixel points in the secondary image ;

[0040] wherein, is a first neighborhood pixel point two-dimensional coordinate in the secondary image ;

[0041] a3, using and to obtain a three-dimensional coordinate of the pixel point to be estimated, using pixel coordinates in and to obtain a three-dimensional coordinate of the neighborhood pixel point, specifically:

[0042] First, using the two-dimensional coordinates of the pixel point to be estimated and the neighborhood pixel points in the main image and the secondary image to construct a matrix , and performing normalization processing on the matrix to obtain a normalized matrix ;

[0043] The matrix is as follows:

[0044] ,

[0045] wherein, is a two-dimensional coordinate of a pixel point in the main image, is a two-dimensional coordinate of a pixel point in the secondary image, , and are row vectors of the main image corresponding to the essential matrix , , and are row vectors of the secondary image corresponding to the essential matrix ;

[0046] ​Then, singular value decomposition is performed on the normalized matrix to obtain a right singular vector matrix , and the last column of is taken as the three-dimensional coordinates of the pixel point and the three-dimensional coordinates of the neighboring pixel points .

[0047] The singular value decomposition performed on the matrix is specifically:

[0048] ,

[0049] wherein is a left singular vector matrix, is a singular value matrix, is a right singular vector matrix.

[0050] Further, the measure value acquisition module obtains the measure value using the three-dimensional coordinates of the pixel to be estimated and the three-dimensional coordinates of the neighboring pixel points, and the measure value is specifically:

[0051] ,

[0052] ,

[0053] ,

[0054] ,

[0055] wherein is the three-dimensional coordinates of the pixel to be estimated, is the label of the neighboring pixel, is the coordinates of the th neighboring pixel point, is a variance representing a normal distribution, is a constant, is the pixel label, is the total number of pixels, is the th pixel value in , is the th pixel value in , , is the intermediate vector.

[0056] Further, the real target pixel point saving module judges whether the pixel point to be estimated is a real target pixel point coordinate using the measure value , and if it is a real target pixel point coordinate, it is saved; otherwise, it goes to the pixel point three-dimensional coordinates acquisition module, and the specific process is as follows:

[0057] will be compared with a preset threshold , if , the three-dimensional coordinates of the current pixel to be estimated are considered as the pixel coordinates of the false target, and the pixel three-dimensional coordinate acquisition module is returned to reacquire the two-dimensional coordinates of the pixel to be estimated and the two-dimensional coordinate set of the neighborhood pixel points in the secondary image; otherwise, the three-dimensional coordinates of the current pixel to be estimated are considered as the pixel coordinates of the real target, and the three-dimensional coordinates of the pixel to be estimated are saved.

[0058] The beneficial effects of the present application are:

[0059] The present application adopts a registration algorithm to obtain the corresponding pixel point coordinates and neighborhood coordinate information in the primary image and the secondary image, then obtains a projection expression based on the essential matrix and the relationship between the primary image and the secondary image, subsequently substitutes the homonymous point coordinates into the projection expression, and adopts a singular value decomposition algorithm to calculate the target three-dimensional coordinates. Then, the correlation of the pixel to be estimated and the neighborhood pixel points is estimated based on a measure function, so as to measure whether the target height is estimated incorrectly, and according to the correlation of the pixel to be estimated and the neighborhood pixel points, it is selected whether to save the three-dimensional coordinates of the pixel to be estimated. The present application adopts SAR image neighborhood information to eliminate the false points of height estimation, improves the discrimination ability of the false target, and further improves the ability of false target suppression. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 is the flowchart of the present application;

[0061] Figure 2 is the SAR primary image of the ship target;

[0062] Figure 3 is the SAR secondary image of the ship target;

[0063] Figure 4 is the three-dimensional reconstruction result of the ship target;

[0064] Figure 5 is the three-dimensional reconstruction false point of the ship target;

[0065] Figure 6 is the point error histogram of the three-dimensional reconstruction of the ship target;

[0066] Figure 7 is the SAR trajectory imaging diagram. DETAILED DESCRIPTION

[0067] Specific implementation one: as shown in the figure, the specific process of the SAR three-dimensional false target suppression method based on the essential matrix and neighborhood consistency constraint in the present embodiment is as follows: Figure 1

[0068] ​Step 1: Obtain the 2D coordinates of the pixel to be estimated in the main image and the set of 2D coordinates of its neighboring pixels. Obtain the 2D coordinates of the pixel to be estimated in the sub-image and the set of 2D coordinates of its neighboring pixels. Use the 2D coordinates of the pixel to be estimated in the main image and the sub-image to obtain the 3D coordinates of the pixel to be estimated. Use the set of 2D coordinates of the neighboring pixels in the main image and the sub-image to obtain the 3D coordinates of the neighboring pixels. Specifically:

[0069] Step 11: Obtain the pixel to be estimated In the main image Set of two-dimensional coordinates of middle neighbor pixels ;

[0070] in, It is the label of the neighborhood coordinate point. It is the total number of neighborhood coordinate points. It is the first pixel in the main image that needs to be estimated. Two-dimensional coordinates of a neighboring pixel;

[0071] Steps 1 and 2: Use the SIFT method to register the main image and the sub-image to obtain the corresponding pixel in the sub-image for the pixel to be estimated. and obtain The set of two-dimensional coordinates of neighboring pixels in the sub-image ;

[0072] in, yes The first in the sub-image Two-dimensional coordinates of neighboring pixels;

[0073] Step 13, Utilize and Obtain the 3D coordinates of the pixel to be estimated, and use and The pixel coordinates in the image are used to obtain the 3D coordinates of neighboring pixels, specifically:

[0074] Step 131: Construct the correspondence between the 3D coordinates of pixels and the 2D coordinates of the same pixels, specifically as follows:

[0075] First, the coordinates of the 3D pixels projected onto the main and sub-images are derived:

[0076] ,

[0077] ,

[0078] ,

[0079] ,

[0080] wherein, is the homogeneous coordinate of the pixel point in the primary image, is the homogeneous coordinate of the pixel point in the secondary image, is the essential matrix corresponding to the primary image, is the essential matrix corresponding to the secondary image, is the three-dimensional coordinate of the pixel point; take or ; take or ;

[0081] Then, the above formula is expanded to obtain:

[0082] ,

[0083] ,

[0084] ,

[0085] ,

[0086] wherein, , and are row vectors of , , and are row vectors of , is the two-dimensional coordinate of the pixel point in the primary image, is the two-dimensional coordinate of the pixel point in the secondary image.

[0087] Then, the relationship between the three-dimensional coordinate of the pixel point and the two-dimensional coordinate of the same pixel point in the primary image, and the relationship between the three-dimensional coordinate of the pixel point and the two-dimensional coordinate of the same pixel point in the secondary image are obtained, which are specifically:

[0088] ,

[0089] ,

[0090] Finally, the following relationship is obtained:

[0091] ,

[0092] Step two, the three-dimensional coordinates of the to-be-estimated pixel point and the neighborhood pixel points are obtained by using the relationship between the three-dimensional coordinate of the pixel point and the two-dimensional coordinate of the same pixel point, which are specifically:

[0093] First, construct a matrix , and normalize the matrix to obtain a normalized matrix ;

[0094] The matrix is as follows:

[0095] ,

[0096] Then, singular value decomposition is performed on the normalized matrix to obtain the basis of the output space , and the last column of the matrix is the three-dimensional coordinates of the pixel point , thereby obtaining the three-dimensional coordinates of the pixel point to be estimated and the three-dimensional coordinates of the neighborhood pixel point ;

[0097] The singular value decomposition of the normalized matrix is specifically as follows:

[0098] ,

[0099] wherein is the left singular value vector matrix, representing the basis of the input space, is the singular value matrix, representing the scale (i.e., singular value size) in each principal direction, is the right singular value vector matrix, representing the basis of the output space.

[0100] Step two, construct a measure function using the three-dimensional coordinates of the pixel to be estimated and the three-dimensional coordinates of the neighborhood pixel point, thereby obtaining a measure value , which is specifically as follows:

[0101] ,

[0102] ,

[0103] ,

[0104] ,

[0105] wherein is the three-dimensional coordinates of the pixel to be estimated, is the label of the neighborhood pixel, is the coordinates of the th neighborhood pixel point, is the variance representing the normal distribution, , is a constant, is a pixel label, is a total number of pixels, is is a pixel value in the th pixel value in the th pixel value in the th

[0106] Step three, comparing the with a preset threshold value , if , considering the pixel point to be estimated as a false target pixel point, returning to step one to re-register the primary image and the secondary image, obtaining the two-dimensional coordinates of the pixel point to be estimated in the secondary image and the two-dimensional coordinate set of the neighborhood pixel points; otherwise, considering the pixel point to be estimated as a real target pixel point, saving the three-dimensional coordinates of the pixel to be estimated .

[0107] In this step, and the array of pixel points corresponding to the same named points. Therefore, when the pixel values of and are similar, it indicates that and have strong correlation, and the obtained parameter value is small; when the pixel values of and are very different, it indicates that and have weak correlation, and the obtained parameter value is large.

[0108] When the pixel values of and are similar, the parameter value is small, at this time the normal distribution curve is thin and high, when the three-dimensional coordinates of the pixel point to be estimated and the three-dimensional coordinates of the neighborhood pixel points are very different, at this time the value decreases rapidly, and the threshold value can eliminate the point; when the three-dimensional coordinates of the pixel point to be estimated and the three-dimensional coordinates of the neighborhood pixel points are small, at this time the value is large. The above reflects that when the neighborhood pixel values are similar, the estimated three-dimensional coordinates are similar.

[0109] When the pixel values of and are very different, the parameter value is large, at this time the normal distribution curve is flat, when the three-dimensional coordinates of the pixel point to be estimated and the three-dimensional coordinates of the neighborhood pixel points are very different, at this time the value decreases slowly; when the three-dimensional coordinates of the pixel point to be estimated and the three-dimensional coordinates of the neighborhood pixel points are small, at this time the value is large. The above reflects that when the neighborhood pixel values are suddenly changed, A large parameter value allows for a greater tolerance for differences in estimated 3D coordinate distances.

[0110] when Less than the threshold At that time, the pixels will be... Re-evaluate the elevation, because Registration points The error caused an inaccurate elevation estimation. Therefore, it is necessary to adjust the pixel values. Re-register, then estimate the 3D information, and then calculate... , with threshold Compare them.

[0111] Specific implementation method 2: SAR three-dimensional false target suppression system based on essential matrix and neighborhood consistency constraint, including: pixel three-dimensional coordinate acquisition module, metric value acquisition module and real target pixel storage module;

[0112] The pixel 3D coordinate acquisition module is used to acquire the 2D coordinates of the pixel to be estimated on the main image and the sub-image, and the set of 2D coordinates of neighboring pixels. Specifically, it uses the 2D coordinates of the pixel to be estimated on the main image and the sub-image, and the set of 2D coordinates of neighboring pixels, to acquire the 3D coordinates of the pixel to be estimated and the 3D coordinates of neighboring pixels.

[0113] a1. Obtain the pixel to be estimated. In the main image Set of two-dimensional coordinates of middle neighbor pixels ;

[0114] in, It is the label of the neighborhood coordinate point. It is the total number of neighborhood coordinate points. It is the first pixel in the main image that needs to be estimated. Two-dimensional coordinates of a neighboring pixel;

[0115] a2. Use the SIFT method to register the main image and the sub-image to obtain the corresponding pixel in the sub-image for the pixel to be estimated. and obtain The set of two-dimensional coordinates of neighboring pixels in the sub-image ;

[0116] in, yes The first in the sub-image Two-dimensional coordinates of neighboring pixels;

[0117] a3. Utilize and Obtain the 3D coordinates of the pixel to be estimated, and use and The pixel coordinate acquisition module acquires the three-dimensional coordinates of the neighborhood pixel points in the pixel coordinates, and specifically acquires the three-dimensional coordinates of the neighborhood pixel points in the pixel coordinates in the primary image and the secondary image by using the two-dimensional coordinates of the to-be-estimated pixel and the neighborhood pixel points in the primary image and the secondary image.

[0118] First, a matrix is constructed by using the two-dimensional coordinates of the to-be-estimated pixel and the neighborhood pixel points in the primary image and the secondary image. The matrix is normalized to obtain a normalized matrix.

[0119] The matrix is as follows.

[0120]

[0121] wherein, x and y are the two-dimensional coordinates of the pixel in the primary image, x' and y' are the two-dimensional coordinates of the pixel in the secondary image, and a, b, c, d, e and f are the row vectors of the essential matrix corresponding to the primary image, and g, h, i, j, k and l are the row vectors of the essential matrix corresponding to the secondary image.

[0122] Then, singular value decomposition is performed on the normalized matrix to obtain a right singular value vector matrix, and the last column of the right singular value vector matrix is taken as the three-dimensional coordinates of the pixel and the three-dimensional coordinates of the neighborhood pixel points.

[0123] The singular value decomposition on the matrix is specifically as follows.

[0124]

[0125] wherein, U is a left singular value vector matrix, S is a singular value matrix, and V is a right singular value vector matrix.

[0126] The measure value acquisition module acquires the measure value by using the three-dimensional coordinates of the to-be-estimated pixel and the three-dimensional coordinates of the neighborhood pixel points, and specifically acquires the measure value by using the three-dimensional coordinates of the to-be-estimated pixel and the three-dimensional coordinates of the neighborhood pixel points.

[0127]

[0128]

[0129] ​​​​​​​​​​​​​​​​​​​​​​​​​​​

[0130]

[0131] wherein, is the three-dimensional coordinate of the pixel to be estimated, is the label of the neighborhood pixel, is the coordinate of the th neighborhood pixel point, is the variance representing the normal distribution, is a constant, is the pixel label, is the total number of pixels, is the th pixel value in the th neighborhood, is the th pixel value in the th neighborhood, , is the intermediate vector;

[0132] The real target pixel point saving module uses the measure value to determine whether the pixel point to be estimated is a real target pixel point coordinate, and if so, it is saved; otherwise, it goes to the pixel three-dimensional coordinate acquisition module, which is specifically:

[0133] The is compared with a preset threshold , and if , it is considered that the three-dimensional coordinate of the current pixel to be estimated is a false target pixel coordinate, and the pixel three-dimensional coordinate acquisition module is returned to reacquire the two-dimensional coordinate of the pixel to be estimated in the secondary image and the two-dimensional coordinate set of the neighborhood pixel points; otherwise, it is considered that the three-dimensional coordinate of the current pixel to be estimated is a real target pixel coordinate, and the three-dimensional coordinate of the pixel to be estimated is saved.

[0134] Embodiment: In order to verify the beneficial effects of the present application, the following simulation experiment is carried out in this embodiment:

[0135] This embodiment estimates the elevation of a ship target, uses neighborhood information to eliminate false points, and compares the results before and after. The SAR trajectory imaging trajectory is shown in Figure 7

[0136] ​​​​SAR parameters are shown in Table 1. Multi-angle SAR elevation extraction uses two SAR images with an azimuth span of 14.3° and a downward viewing angle span of 20.4°. The SAR image resolution is set to 0.5 meters, and the sampling frequency and bandwidth are calculated. Satellite parameters are shown in Table 2, where the parameters are values ​​transformed to the imaging coordinate system. Each parameter has two values ​​corresponding to the two SAR imaging parameters. The ship target size is 130m × 14.5m. During the simulation, the ship target is stationary and does not exhibit three-dimensional rotation or sway. The simulation considers the anisotropy of the target's scattering characteristics, and the ship target's RCS is recalculated when imaging the ship target at different angles. The results of the ship target SAR main image and sub-image are shown below. Figures 2-3 As shown.

[0137] Table 1 SAR Parameter Table

[0138]

[0139] Table 2 Satellite Parameter Table

[0140]

[0141] First, the three-dimensional coordinates of pixels in the main and secondary images are obtained, and then neighborhood information is used to remove false points. Figure 4 The image shows the 3D reconstruction results of the ship target. Black dots represent the coordinates of the re-estimated points after removing false points using neighborhood information, while other colors represent the 3D reconstruction results obtained through the neighborhood information removal algorithm. The color also indicates the height distribution of the target.

[0142] Figure 5 The histogram results for the 3D reconstruction of the spurious points of the ship target are shown. Black dots represent the distribution of spurious points removed using information filtering, while other colors represent the distribution of coordinate points after recalculation using these spurious points. The color also indicates the target's altitude distribution. Figure 4 The results show that the essential matrix transformation algorithm can estimate the 3D model of the ship target, but the estimation effect is poor; from Figure 5 The results show that the black dots are significantly deviated from the ship model. Using neighborhood information can effectively identify false dots and re-estimate their 3D coordinates.

[0143] Figure 6 This is a 3D reconstruction error map of a ship target, where the horizontal axis represents the error distribution and the vertical axis represents the number of error points. The red square curve represents the target point curve verified by the false point removal algorithm, the black plus sign curve represents the curve after false points have been removed, and the blue circular curve represents the set of points after false point correction. Figure 6It is found that the point target error distribution is mostly between 0 m and 5 m, the number of false points is large, and the false points can be effectively reduced. Table 3 is the average error of the above several types of point targets, the error of the found false points is 5.22 m, the root mean square error is 7.31 m; greater than the real point error 4.17 m, the root mean square error 6.62 m; the error of the false point after correction also decreases to 4.20 m; the root mean square error decreases to 6.59 m.

[0144] Table 3

[0145]

[0146] The neighborhood information is used to eliminate the false points in three-dimensional information estimation, and the algorithm principle is analyzed. Simulation experiments show that in the three-dimensional reconstruction of ship targets, due to the anisotropic scattering characteristics of the target and the existence of overlapping phenomenon, there are a large number of false points in the three-dimensional reconstruction, the use of neighborhood information can effectively eliminate false points, and correct false points.

Claims

1. A SAR three-dimensional false target suppression method based on an essential matrix and neighborhood consistency constraint, characterized in that The method specifically comprises the following steps: Step one, obtaining the two-dimensional coordinates of the pixel to be estimated on the primary image and the secondary image and the two-dimensional coordinate set of the neighborhood pixel points, and obtaining the three-dimensional coordinates of the pixel to be estimated and the three-dimensional coordinates of the neighborhood pixel points by using the two-dimensional coordinates of the pixel to be estimated on the primary image and the secondary image and the two-dimensional coordinate set of the neighborhood pixel points; Step two, using the three-dimensional coordinates of the pixel to be estimated and the three-dimensional coordinates of the neighborhood pixels to construct a measure function, thereby obtaining a measure value ; Step three, if the preset threshold is compared, if , the three-dimensional coordinates of the current pixel to be estimated are considered as the pixel coordinates of the false target, and step one is returned to reacquire the two-dimensional coordinates of the pixel point to be estimated and the two-dimensional coordinate set of the neighborhood pixel points in the secondary image; Otherwise, it is considered that the three-dimensional coordinates of the current pixel to be estimated are the pixel coordinates of the real target, and the three-dimensional coordinates of the pixel to be estimated are saved.

2. The SAR three-dimensional false target suppression method based on an essence matrix and neighborhood consistency constraint according to claim 1, characterized in that: In the step one, the two-dimensional coordinates of the pixel to be estimated on the primary image and the secondary image and the two-dimensional coordinate set of the neighborhood pixel points are obtained, and the three-dimensional coordinates of the pixel to be estimated and the three-dimensional coordinates of the neighborhood pixel points are obtained by using the two-dimensional coordinates of the pixel to be estimated on the primary image and the secondary image and the two-dimensional coordinate set of the neighborhood pixel points, specifically as follows: Step one, obtaining a pixel point to be estimated In the main image The two-dimensional coordinate set of the neighborhood pixel points ; in, It is the label of the neighborhood coordinate point. It is the total number of neighborhood coordinate points. It is the first pixel in the main image that needs to be estimated. Two-dimensional coordinates of a neighboring pixel; Step one two, adopt SIFT method to the main image and the sub image registration, obtain the pixel point in the sub image corresponding pixel point to be estimated , and obtain In the sub image neighborhood pixel point two-dimensional coordinate set ; wherein, is in the sub-image, neighborhood pixel points two-dimensional coordinates; Step one three, using and Obtain the three-dimensional coordinates of the pixel to be estimated, using and the three-dimensional coordinates of the neighborhood pixel points are obtained.

3. The SAR three-dimensional false target suppression method based on an essence matrix and neighborhood consistency constraint of claim 2, characterized in that: The step one three in using And Obtain the three-dimensional coordinates of the pixel to be estimated, using And The three-dimensional coordinates of the neighborhood pixel points are obtained by using the pixel coordinates in Firstly, a matrix is constructed by using the two-dimensional coordinates of the to-be-estimated pixel point and the neighborhood pixel points in the primary image and the secondary image , and the matrix is normalized to obtain a normalized matrix ; The matrix As follows: ， wherein is a two-dimensional coordinate of a pixel point in the primary image, is a two-dimensional coordinate of a pixel point in the secondary image, , and are row vectors of the essential matrix corresponding to the primary image, , and are row vectors of the essential matrix corresponding to the secondary image. Then, singular value decomposition is performed on the normalized matrix to obtain a right singular vector matrix , and the last column of is taken as the three-dimensional coordinates of the pixel point and the three-dimensional coordinates of the neighboring pixel points .

4. The SAR three-dimensional false target suppression method based on an essence matrix and neighborhood consistency constraint of claim 3, characterized in that: The pair of matrices The singular value decomposition is specifically ， wherein is a left singular vector matrix, is a singular value matrix, is a right singular vector matrix.

5. The SAR three-dimensional false target suppression method based on an essence matrix and neighborhood consistency constraint of claim 4, characterized in that: The step two uses the three-dimensional coordinates of the pixel to be estimated and the three-dimensional coordinates of the neighborhood pixels to construct a measurement function, thereby obtaining a measurement value , specifically: ， wherein, is the three-dimensional coordinate of the pixel to be estimated, is the index of the neighboring pixel, is the coordinate of the th neighboring pixel point, is the variance representing the normal distribution.

6. The SAR three-dimensional false target suppression method based on an essence matrix and neighborhood consistency constraint of claim 5, characterized in that: the variance of the normal distribution Specifically: ， ， ， wherein, is a constant, is a pixel index, is the total number of pixels, is the pixel value in the is the pixel value in the , is the intermediate vector.

7. A SAR three-dimensional false target suppression system based on an essential matrix and neighborhood consistency constraint, characterized in that: The system comprises a pixel three-dimensional coordinate acquisition module, a measure value acquisition module and a real target pixel point saving module. The pixel three-dimensional coordinate acquisition module is used for obtaining the two-dimensional coordinates of the pixel to be estimated on the primary image and the secondary image and the two-dimensional coordinate set of the neighborhood pixel points, and obtaining the three-dimensional coordinates of the pixel to be estimated and the three-dimensional coordinates of the neighborhood pixel points by using the two-dimensional coordinates of the pixel to be estimated on the primary image and the secondary image and the two-dimensional coordinate set of the neighborhood pixel points. The measure value obtaining module obtains the measure value by using the three-dimensional coordinates of the pixel to be estimated and the three-dimensional coordinates of the neighboring pixel points ; The real target pixel point saving module utilizes the measurement value The pixel point three-dimensional coordinate obtaining module obtains the three-dimensional coordinates of the pixel point to be estimated.

8. The system for SAR three-dimensional false target mitigation based on eigenmatrix and neighborhood consistency constraint of claim 7, wherein: The pixel three-dimensional coordinate acquisition module is used for obtaining the two-dimensional coordinates of the pixel to be estimated on the primary image and the secondary image and the two-dimensional coordinate set of the neighborhood pixel points, and obtaining the three-dimensional coordinates of the pixel to be estimated and the three-dimensional coordinates of the neighborhood pixel points by using the two-dimensional coordinates of the pixel to be estimated on the primary image and the secondary image and the two-dimensional coordinate set of the neighborhood pixel points. a1, obtaining a pixel point to be estimated in the main image a set of two-dimensional coordinates of the neighborhood pixel points ; wherein, is a neighborhood coordinate point label, is a total number of neighborhood coordinate points, is a two-dimensional coordinate of the i-th neighborhood pixel point of the to-be-estimated pixel point in the main image; and is a two-dimensional coordinate of the j-th neighborhood pixel point of the to-be-estimated pixel point in the main image. a2, using the SIFT method to register the main image and the secondary image, obtaining the corresponding pixel point of the pixel point to be estimated in the secondary image , and obtaining a neighborhood pixel point two-dimensional coordinate set in the secondary image ; wherein, is the first neighborhood pixel point two-dimensional coordinates; a3、Utilizing and obtaining the three-dimensional coordinates of the pixel point to be estimated, utilizing and the three-dimensional coordinates of the neighborhood pixel points are obtained by using the pixel coordinates in the and First, a matrix is constructed by using the two-dimensional coordinates of the to-be-estimated pixel point and the neighborhood pixel points in the primary image and the secondary image , and the matrix is normalized to obtain a normalized matrix ; The matrix As follows: ， wherein is a two-dimensional coordinate of a pixel point in the primary image, is a two-dimensional coordinate of a pixel point in the secondary image, , and is a row vector of the essential matrix corresponding to the primary image, , and is a row vector of the essential matrix corresponding to the secondary image. Then, singular value decomposition is performed on the normalized matrix to obtain a right singular vector matrix , and the last column of is taken as the three-dimensional coordinates of the pixel point and the three-dimensional coordinates of the neighboring pixel points ; The pair of matrices The singular value decomposition is specifically: ， wherein is a left singular vector matrix, is a singular value matrix, is a right singular vector matrix.

9. The system for SAR three-dimensional false target mitigation based on eigenmatrix and neighborhood consistency constraint of claim 8, wherein: The measure value obtaining module obtains the measure value by using the three-dimensional coordinates of the pixel to be estimated and the three-dimensional coordinates of the neighboring pixel points , and specifically comprises: ， ， ， ， wherein, is the three-dimensional coordinate of the pixel to be estimated, is the index of the neighboring pixel, is the coordinate of the th neighboring pixel point, is the variance representing the normal distribution, is a constant, is the pixel index, is the total number of pixels, is the th pixel value in the th pixel value in the th pixel value in the th pixel value in the th pixel value in the , is the intermediate vector.​ 10. The system for SAR three-dimensional false target mitigation based on eigenmatrix and neighborhood consistency constraint of claim 9, wherein: The real target pixel point saving module utilizes the measurement value The pixel point three-dimensional coordinate obtaining module judges whether the to-be-estimated pixel point is a real target pixel point coordinate. If it is a real target pixel point coordinate, it is saved; otherwise, it is transferred to the pixel point three-dimensional coordinate obtaining module. Will be compared with the preset threshold , if , the three-dimensional coordinates of the current pixel to be estimated are considered as the pixel coordinates of the false target, and the pixel three-dimensional coordinate acquisition module returns to acquire the two-dimensional coordinates of the pixel to be estimated and the two-dimensional coordinate set of the neighborhood pixel points in the secondary image. Otherwise, it is considered that the three-dimensional coordinates of the current pixel to be estimated are the pixel coordinates of the real target, and the three-dimensional coordinates of the pixel to be estimated are saved.

Images