A star sensor image restoration method based on blind deconvolution

Abstract

The invention discloses a star sensor image restoration method based on blind deconvolution, and relates to the technical field of star sensor image processing. The method includes collecting a real-time star map through an imaging system, reading the entire star-sensing image of the real-time star map, estimating the position of a star point according to the entire star-sensing image, opening a window on the entire star-sensing image with each star point as the center to obtain a sub-image, restoring the sub-image to a target image, calculating the horizontal and vertical coordinate positions of the centroid of the star point according to the restored target image, and outputting the position of the star point to a star map recognition module, so as to solve the problem of inaccurate judgment of the position of the star point under aerodynamic turbulence and improve the centroid extraction accuracy of the target.

Description

Technical Field

[0001] The invention belongs to the technical field of star sensor image processing, and in particular relates to a star sensor image restoration method based on blind deconvolution. Background Art

[0002] Star sensors are widely used in the aerospace field. They can be used in multiple application scenarios such as navigation, mission planning, positioning, orbit control, etc. In addition, in deep space exploration missions, star sensors can also help spacecraft to perform precise attitude control and navigation to ensure the successful execution of the mission.

[0003] With the continuous development of aerospace technology, star sensors are also constantly upgraded and improved. Star sensors are gradually developing into miniaturization, intelligence and multi-functions, so that they can be used in conjunction with flight platforms.

[0004] The flight platform needs to constantly determine its own position during flight, including the flight of missiles, the driving of aircraft, and the cruising of near-Earth satellites. Therefore, a star sensor is installed on the flight platform to determine the position in combination with the star map. When the flight platform flies at high speed in near-Earth orbit, the observed star point will be affected by aerodynamic turbulence.

[0005] Therefore, under aerodynamic turbulence conditions, it is extremely necessary to perform a certain degree of image restoration on the images taken by the star sensor so as to better compare them with the star map. Summary of the invention

[0006] The purpose of the present invention is to provide a star sensor image restoration method based on blind deconvolution to solve the problem of inaccurate star point position judgment under aerodynamic turbulence.

[0007] To achieve the above object, the present invention provides a star sensor image restoration method based on blind deconvolution, the method comprising the following steps:

[0008] S1: Collect the real-time star map through the imaging system and read the entire star-sensitive image of the real-time star map;

[0009] S2: Estimate the positions of star points based on the entire star-sensitive image, and open a window on the entire star-sensitive image with each star point as the center to obtain a sub-image;

[0010] S3: restore the sub-image to the target image;

[0011] S4: Calculate the horizontal and vertical coordinates of the centroid of the star point according to the restored target image;

[0012] Star point centroid horizontal coordinate The calculation formula is: ,

[0013] Star point centroid ordinate The calculation formula is: ,

[0014] is the image horizontal coordinate, is the image ordinate, For the In the image Gray value at the coordinate;

[0015] S5: Output the positions of the star points to the star map recognition module.

[0016] Further, restoring the sub-image to the target image includes:

[0017] The initial value of the target image corresponding to the sub-image As the initial input for iterative calculation, the iterative process includes The result obtained after the iteration is calculated by fast Fourier transform to get the Fourier transform of iterative target ,in , The sub-image obtained by the windowed imaging system, is the original target image, is the point spread function, is the image noise, are the coordinates of the star point in two-dimensional space, and the symbol “*” represents convolution. Indicates the number of iterations;

[0018] No. Fourier transform of iterative target Add frequency domain constraints and calculate the frequency domain The optical transfer function of the iteration ( ;

[0019] For frequency domain The optical transfer function of the iteration ( ) is used to perform inverse fast Fourier transform to obtain the spatial domain The point spread function of the iteration , and then add the spatial domain constraint to get the spatial domain constraint The point spread function of the iteration ;

[0020] The spatial domain constraints The point spread function of the iteration Perform fast Fourier transform to obtain the frequency domain The optical transfer function of the iteration ; The frequency domain The optical transfer function of the iteration Add the frequency domain constraint and get the first Fourier transform of the target of the iteration ;

[0021] The frequency domain constraints Fourier transform of the target of the iteration Perform inverse fast Fourier transform to obtain the target image in the spatial domain ;

[0022] The target image in the spatial domain Add spatial domain constraints to obtain the target image constrained by spatial domain ;

[0023] Iterative calculation, when the iterative error is less than the set threshold The restored target image is obtained.

[0024] Furthermore, the spatial domain constraint adopts a constraint function of a point spread function :

[0025] ,

[0026] The spatial domain constraint uses the constraint function of the target image :

[0027] ,

[0028] The spatial domain constraint The point spread function of the iteration Multiplying with the constraint function of the point spread function forms a constraint;

[0029] The target image in the spatial domain The constraint is formed by multiplying it with the constraint function of the target image.

[0030] Furthermore, by adding frequency domain constraints, the frequency domain The optical transfer function of the iteration ( The iteration formula is:

[0031] ,

[0032] ,

[0033] The frequency domain Optical transfer function Add the frequency domain constraint to get the first Fourier transform of the target of the iteration The iteration formula is:

[0034] ,

[0035] ,

[0036] ,

[0037] in, is the Wiener filter coefficient, which is the square of the image signal-to-noise ratio; is the regularization coefficient, determined according to the image noise level, is the second-order difference operator Fourier transform after zero padding and extension; is the optical transfer function of the first iteration in the frequency domain, is the Fourier transform of the first iteration target in the frequency domain, for The Fourier transform of for The conjugated form of for The conjugated form of is the initial value of the optical transfer function in the frequency domain, for The conjugated form of The Fourier transform of the target The initial value of the conjugate form of .

[0038] Furthermore, the iteration error is less than the set threshold It is expressed as:

[0039] .

[0040] The present invention achieves the following beneficial effects:

[0041] 1. The present invention solves the problem of star sensor image restoration under aerodynamic turbulence conditions and improves the accuracy of target centroid extraction. Under aerodynamic turbulence conditions, the method of the present invention can improve the accuracy of star point extraction compared with traditional star sensor image processing technology. Through verification, the accuracy can be improved by more than 50%;

[0042] 2. The present invention has the performance of real-time processing. By opening a window on the star point target on the star sensor image and adding constraints to the blind deconvolution algorithm, the amount of calculation can be effectively reduced and the convergence speed of the algorithm can be improved. By selecting a high-performance embedded real-time processing platform, under the condition of opening a window of 8 pixels × 8 pixels, the estimated single-frame processing time is 8ms. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A flowchart of a method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0045] See also Figure 1 As shown, the present invention provides a star sensor image restoration method based on blind deconvolution, the method comprising the following steps:

[0046] S1: Collect the real-time star map through the imaging system and read the entire star-sensitive image of the real-time star map;

[0047] S2: Estimate the positions of star points based on the entire star-sensitive image, and open a window on the entire star-sensitive image with each star point as the center to obtain a sub-image;

[0048] S3: restore the sub-image to the target image;

[0049] S4: Calculate the horizontal and vertical coordinates of the centroid of the star point according to the restored target image;

[0050] Star point centroid horizontal coordinate The calculation formula is: ,

[0051] Star point centroid ordinate The calculation formula is: ,

[0052] is the image horizontal coordinate, is the image ordinate, For the In the image Grayscale value at coordinates.

[0053] S5: Output the positions of the star points to the star map recognition module.

[0054] In the linear system framework, the image obtained by the imaging system is the original target image and the point spread function The result of convolution. In actual systems, the image also includes noise The imaging process can be expressed as:

[0055] (1)

[0056] The symbol "*" represents convolution. By performing a fast Fourier transform on the above equation, we can get the representation in the spatial frequency domain:

[0057] (2)

[0058] In the above formula Indicates frequency, For images The Fourier transform of is the original target image The Fourier transform of is the point spread function The Fourier transform of , i.e. the optical transfer function, Image noise The Fourier transform of . Under the action of aero-optical effect, the point spread function It is impossible to obtain it more accurately, so blind deconvolution technology is a way to estimate the original target image. A feasible method. Apply blind deconvolution technology to solve the original target image When the original target image and the point spread function are all unknown, so an iterative algorithm needs to be designed to alternately estimate the original target image and the point spread function .

[0059] is the initial estimate of the target image, and the sub-image obtained in step S2 is taken. Initially, constraints are added alternately in the spatial domain and the frequency domain through fast Fourier transform and inverse fast Fourier transform, and a cyclic iteration is performed to obtain the final estimated value of the target.

[0060] Specifically,

[0061] Restoring the sub-image to the target image includes:

[0062] The initial value of the target image corresponding to the sub-image As the initial input for iterative calculation, the iterative process includes The result obtained after the iteration is calculated by fast Fourier transform to get the Fourier transform of the target of the iteration ,in , The sub-image obtained by the windowed imaging system, is the original target image, is the point spread function, is the image noise, are the coordinates of the star point in two-dimensional space, and the symbol “*” represents convolution. Indicates the number of iterations;

[0063] No. Fourier transform of the target of the iteration Add frequency domain constraints and calculate the frequency domain The optical transfer function of the iteration ( ;

[0064] For frequency domain The optical transfer function of the iteration ( ) is used to perform inverse fast Fourier transform to obtain the spatial domain The point spread function of the iteration , and then add the spatial domain constraint to get the spatial domain constraint The point spread function of the iteration ;

[0065] The spatial domain constraints The point spread function of the iteration Perform fast Fourier transform to obtain the frequency domain The optical transfer function of the iteration ; The frequency domain The optical transfer function of the iteration Add the frequency domain constraint and get the first Fourier transform of the target of the iteration ;

[0066] The frequency domain constraints Fourier transform of the target of the iteration Perform inverse fast Fourier transform to obtain the target image in the spatial domain ;

[0067] The target image in the spatial domain Add spatial domain constraints to obtain the point spread function of spatial domain constraints ;

[0068] Iterative calculation, when the iterative error is less than the set threshold The restored target image is obtained.

[0069] Applying blind deconvolution technology to restore star sensor images is a difficult problem. The following problems are expected: first, when the image signal-to-noise ratio is insufficient, the iteration process may lead to noise amplification and non-convergence of iterations; second, the target information may be lost during the iteration process. However, compared with remote sensing images with more complex targets and backgrounds, the targets in star sensor images are point targets, with fixed target shapes and less high-frequency information.

[0070] Since the point spread function The value of is only meaningful when it is non-negative, so we can use This feature adds a non-negative constraint and defines the constraint function of the point spread function:

[0071] (3)

[0072] For the original target image Similarly, define the constraint function of the target image:

[0073] (4)

[0074] In each iteration, the spatial domain constraint The point spread function of the iteration Multiplying with the constraint function of the point spread function forms a constraint;

[0075] The target image in the spatial domain The constraint is formed by multiplying it with the constraint function of the target image.

[0076] In order to improve the algorithm's anti-noise ability, Wiener filtering and regularization constraints are usually added in the frequency domain, and the resulting iterative formula is as follows:

[0077] (5)

[0078] (6)

[0079] (7)

[0080] (8)

[0081] (9),

[0082] in, is the Wiener filter coefficient, which is the square of the image signal-to-noise ratio; is the regularization coefficient, determined according to the image noise level, is the second-order difference operator Fourier transform after zero padding and extension; is the optical transfer function of the first iteration in the frequency domain, is the Fourier transform of the first iteration target in the frequency domain, for The Fourier transform of for The conjugated form of for The conjugated form of is the initial value of the optical transfer function in the frequency domain, for The conjugated form of The Fourier transform of the target The initial value of the conjugate form of .

[0083] As mentioned above, adding frequency domain constraints, the frequency domain is calculated The optical transfer function of the iteration ( The iteration formula is:

[0084] ,

[0085] ,

[0086] The frequency domain Optical transfer function Add the frequency domain constraint to get the first Fourier transform of the target of the iteration The iteration formula is:

[0087] ,

[0088] ,

[0089] ,

[0090] The image restoration of the present invention sets the following iterative convergence conditions:

[0091] The iterative error shown in formula (10) is less than a certain value ;

[0092] (10)

[0093] Or the specified number of iterations is reached, here 40.

[0094] Although the content of the present invention has been described in detail through the above preferred embodiments, it should be recognized that the above description should not be considered as limiting the present invention. After reading the above content, various modifications and substitutions of the present invention will be obvious to those skilled in the art. Therefore, the protection scope of the present invention should be defined by the attached claims.

Claims

1. A star sensor image restoration method based on blind deconvolution, characterized in that , the method comprises the following steps: S1: Collect the real-time star map through the imaging system and read the entire star-sensitive image of the real-time star map; S2: Estimate the positions of star points based on the entire star-sensitive image, and open a window on the entire star-sensitive image with each star point as the center to obtain a sub-image; S3: restore the sub-image to the target image; S4: Calculate the horizontal and vertical coordinates of the centroid of the star point according to the restored target image; Star point centroid horizontal coordinate The calculation formula is: , Star point centroid ordinate The calculation formula is: , is the image horizontal coordinate, is the image ordinate, For the In the image Gray value at the coordinate; S5: Output the position of the star point to the star map recognition module; Restoring the sub-image to the target image comprises: The initial value of the target image corresponding to the sub-image As the initial input for iterative calculation, the iterative process includes The result obtained after the iteration is calculated by fast Fourier transform to get the Fourier transform of iterative target ,in , The sub-image obtained by the windowed imaging system, is the original target image, is the point spread function, is the image noise, is the coordinate of the star point in two-dimensional space, the symbol "*" represents convolution, Indicates the number of iterations; No. Fourier transform of iterative target Add frequency domain constraints and calculate the frequency domain The optical transfer function of the iteration ( ; For frequency domain The optical transfer function of the iteration ( ) is used to perform inverse fast Fourier transform to obtain the spatial domain The point spread function of the iteration , and then add the spatial domain constraint to get the spatial domain constraint The point spread function of the iteration ; The spatial domain constraints The point spread function of the iteration Perform fast Fourier transform to obtain the frequency domain The optical transfer function of the iteration ; The frequency domain The optical transfer function of the iteration Add the frequency domain constraint and get the first Fourier transform of the target of the iteration ; The frequency domain constraints Fourier transform of the target of the iteration Perform inverse fast Fourier transform to obtain the target image in the spatial domain ; The target image in the spatial domain Add spatial domain constraints to obtain the target image constrained by spatial domain ; Iterative calculation, when the iterative error is less than the set threshold The restored target image is obtained.

2. The star sensor image restoration method based on blind deconvolution according to claim 1, characterized in that: The spatial domain constraint adopts the constraint function of the point spread function : ， The spatial domain constraint uses the constraint function of the target image : ， The spatial domain constraint The point spread function of the iteration Multiplying with the constraint function of the point spread function forms a constraint; The target image in the spatial domain The constraint is formed by multiplying it with the constraint function of the target image.

3. A star sensor image restoration method based on blind deconvolution as claimed in claim 1, characterized in that: Add frequency domain constraints and calculate the frequency domain The optical transfer function of the iteration ( The iteration formula is: ， ， The frequency domain Optical transfer function Add the frequency domain constraint to get the frequency domain constraint Fourier transform of the target of the iteration The iteration formula is: ， ， ， in, is the Wiener filter coefficient, which is the square of the image signal-to-noise ratio; is the regularization coefficient, determined according to the image noise level, is the second-order difference operator Fourier transform after zero padding and extension; is the optical transfer function of the first iteration in the frequency domain, is the Fourier transform of the first iteration target in the frequency domain, for The Fourier transform of for The conjugated form of for The conjugated form of is the initial value of the optical transfer function in the frequency domain, for The conjugated form of The Fourier transform of the target The initial value of the conjugate form of .

4. A star sensor image restoration method based on blind deconvolution as claimed in claim 1, characterized in that: The iteration error is less than the set threshold It is expressed as: 。

Images