CN-121981913-A - Astronomical imaging stripe noise correction method and system based on low-frequency directivity characteristics

CN121981913ACN 121981913 ACN121981913 ACN 121981913ACN-121981913-A

Abstract

The invention discloses an astronomical imaging stripe noise correction method and system based on low-frequency directivity characteristics, which comprises the steps of firstly realizing column-level brightness balance through steady statistics and preliminarily eliminating global vertical stripes; and finally, under the constraint of the mask, generating a fringe template through star point suppression and frequency domain filtering, and compensating and correcting the image. The method combines airspace equalization and frequency domain template estimation, and effectively corrects the area by precisely restricting the trapezoidal mask, so that the method not only reserves true star point signals, but also can effectively inhibit residual stripe noise, and remarkably improves the signal-to-noise ratio of astronomical images and the reliability of subsequent celestial body detection.

Inventors

LIU PI
Zhao Guwenxuan
ZHOU HAO
LIU BINYANG
GUO YANGYANG
Fan Banghao
MEI XUE
HE YI
WANG XIAORONG
CHEN YUMING

Assignees

南京工业大学

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (10)

1. An astronomical imaging streak noise correction method based on low frequency directivity characteristics, characterized in that the method comprises the following steps: S1, acquiring two-dimensional astronomical image data acquired by an inspection camera, estimating representative brightness of each column by adopting steady statistics, performing column-level brightness translation, and performing vertical stripe brightness balance on the two-dimensional astronomical image to obtain a two-dimensional astronomical image with balanced column-level brightness; S2, sliding and sampling on the two-dimensional astronomical image with balanced column-level brightness according to set window size and step length, performing mean removal, windowing and two-dimensional Fourier transformation on each window, extracting low-frequency girdle energy, counting energy distribution in a direction according to angle bin division, calculating a low-frequency directivity coefficient, and weighting and averaging window results according to coverage areas to generate a low-frequency directivity characteristic thermodynamic diagram which is the same in size as an original image and used for representing the spatial position and relative strength of residual stripes; S3, after judging the contrast effectiveness of the low-frequency directional characteristic thermodynamic diagram, performing threshold segmentation, morphological closing operation and denoising, and reserving only a communication region with the largest area and conforming to the area proportion constraint as a main region mask; S4, performing star point inhibition or star point mask processing on the two-dimensional astronomical image with balanced column brightness, performing two-dimensional Fourier transformation on the processed two-dimensional astronomical image, reserving only a vertical frequency band near the center of a frequency domain to obtain a vertical stripe/low frequency component, performing inverse transformation to obtain an image domain two-dimensional stripe template, and directly acting the image domain two-dimensional stripe template on the two-dimensional astronomical image with balanced column brightness under the constraint of a fixed trapezoid mask to perform compensation correction and output an image with corrected stripe noise.
2. The astronomical imaging streak noise correction method based on the low frequency directivity characteristic as in claim 1 wherein in step S1, the process of performing vertical streak luminance equalization on the two-dimensional astronomical image includes the steps of: Dividing the two-dimensional astronomical image into a plurality of vertical bars along the column direction, for the first The vertical bars randomly sample a fixed number of sample points in a non-return mode to form a sample set, and firstly calculate the median of the sample set as a robust central value Calculating absolute intermediate potential difference and converting the absolute intermediate potential difference into Gaussian equivalent standard deviation Only remain falling within the main peak interval The sample in as the first Main peak set, parameters of individual bars Outliers including star highlighting pixels, dead pixels or extreme noise are removed; taking the average value of the main peak set of each vertical bar as a main peak brightness estimated value, and synthesizing main peak brightness estimated values of all columns into a main peak sequence Calculating to obtain the global average value of the main peak brightness of all the vertical bars , Represent the first Main peak brightness estimation values of the vertical bars; Will global mean As alignment targets for the individual bars, the first The correction offset for each vertical bar is: ; Wherein the method comprises the steps of In order to correct the intensity coefficient of the light, Indicating that the alignment is complete and that, And performing brightness shift correction on all pixels in each vertical bar based on the correction offset to obtain a brightness balanced image, and obtaining a two-dimensional astronomical image with balanced column brightness.
3. The astronomical imaging fringe noise correction method based on low-frequency directivity characteristics as recited in claim 1, wherein in step S2, the process of generating a low-frequency directivity characteristic thermodynamic diagram comprises the steps of: The sliding window is set to be Step size of , And Respectively representing the number of pixels of sliding window moving in vertical direction and horizontal direction, generating window left upper corner starting point sequence by using welt covering strategy, starting point for any window Taking sub-blocks Performing mean removal and windowing on the window to obtain a preprocessed sub-block : ; Wherein the method comprises the steps of As the mean value of the window, Representing an element-by-element multiplication, Is a two-dimensional Hann window based on the sub-blocks after pretreatment Calculating a frequency domain result after two-dimensional Fourier transform And calculate to obtain the power spectrum In the formula (I), in the formula (II), Representing a two-dimensional fourier transform operator; representing a two-dimensional astronomical image with balanced column-level brightness; centering the power spectrum, defining a radius by taking the center of the spectrum as the origin Selecting a low-frequency zone area As a statistical object, and will Evenly divided into The angles are divided into boxes, and the energy in the annular belt is polymerized according to the angles to obtain the directional energy And calculating the low-frequency directivity coefficient by adopting the following formula: ; Wherein the method comprises the steps of A stabilization term to prevent zero removal; the directional energy obtained by polymerization in the kth angle bin; initializing cumulative graphs Hit map For each window, a low frequency directivity coefficient is zero matrix Backfilling to all pixel locations within the window footprint to cause After all window processing is completed, normalizing each pixel to obtain a low-frequency directivity characteristic thermodynamic diagram H: ; In the formula, Representing pixel location Normalized values of the low frequency directivity characteristic thermodynamic diagram at, Representing pixel location On all the sliding windows covered by it Is used in the method of the present invention, Representing over-pixel locations The total number of sliding windows at.
4. The astronomical imaging fringe noise correcting method based on low frequency directionality characteristics of claim 3, wherein in step S2, multiple window patches are assembled in batches into tensors and FFT and directional energy aggregation are performed in parallel on GPU.
5. The astronomical imaging streak noise correction method based on the low frequency directivity characteristic as in claim 1, wherein step S3 further includes: Computing low frequency directivity characteristic thermodynamic diagrams Dynamic range of (2) And standard deviation of : ; In the formula, And Respectively represent low-frequency directivity characteristic thermodynamic diagrams The minimum and maximum values of all pixels in (a), Representing a standard deviation calculation function; determining low frequency directivity characteristic thermodynamic diagrams Whether the dynamic range and standard deviation of (2) are smaller than the respective preset threshold values, when Or (b) In the time-course of which the first and second contact surfaces, Respectively determining a dynamic range threshold and a standard deviation threshold, considering that the thermodynamic diagram lacks effective contrast, directly returning to the empty mask, otherwise, considering that the low-frequency directional characteristic thermodynamic diagram meets contrast effectiveness, smoothing the low-frequency directional characteristic thermodynamic diagram to inhibit local noise fluctuation, and then adopting threshold segmentation to verify candidate areas to obtain a candidate binary diagram : ; In the formula, Representing a threshold value; Representing pixel location Binary candidate region values at; For candidate binary image Performing morphological closing operation and hole filling operation, removing small connected domain, marking connected domain, and retaining only connected domain with largest area as main region mask Calculating a main region mask Area ratio of (2) : ; When (when) Or (b) The time is returned to the empty mask, And The height and width of the two-dimensional astronomical image, And A minimum threshold and a maximum threshold representing the area ratio of the main region respectively, Mask representing main area And when the empty mask is returned, directly outputting the two-dimensional astronomical image with balanced column brightness, and ending the flow.
6. The method for correcting streak noise in astronomical imaging based on low frequency directionality in accordance with claim 5, wherein the threshold value The determination is made using Otsu adaptive thresholds or using percentile thresholding.
7. The astronomical imaging fringe noise correcting method based on low frequency directionality characteristics of claim 1, wherein in step S3, the process of generating a fixed trapezoidal mask comprises the steps of: mask the main area Performing fixed trapezoid fitting, specifically searching a pixel column index set with each row of mask as true, and taking leftmost and rightmost indexes as left and right boundary observation points And (3) with , Representing the row coordinates of the image, And Respectively represent the first Main area mask in row Leftmost column coordinates and rightmost column coordinates of the pixel that are true; respectively performing linear fitting on the left boundary observation point and the right boundary observation point ; In the formula, And The slope and intercept of the straight line fitted for the left boundary observation point are respectively, And Respectively fitting a straight line slope and an intercept for the right boundary observation point; The ordinate of the vertex of the trapezoid is forcedly set as the upper boundary and the lower boundary of the image And (3) with Thereby obtaining four vertex abscissas ; And cut it to In the formula (I), in the formula (II), 、、 And Respectively represent the left upper corner vertex abscissa, the left lower corner vertex abscissa, the right upper corner vertex abscissa and the right lower corner vertex abscissa of the trapezoid, And The height and width of the two-dimensional astronomical image are respectively equal to the distance 0 or the abscissa of a certain vertex Less than In the time-course of which the first and second contact surfaces, To adsorb proportion parameter, the vertex is welted to 0 or Finally, constructing trapezoid polygons with four vertexes and generating a fixed trapezoid mask 。
8. The astronomical imaging fringe noise correcting method based on low-frequency directionality characteristics of claim 1, wherein in step S4, the process of obtaining an image-domain two-dimensional fringe template comprises the steps of: for a two-dimensional astronomical image with balanced column-level brightness, for a pixel set in a certain direction Calculating a robust center ; MAD equivalent standard deviation ; In the formula, Representing a median calculation function; When the pixel value of any pixel When the pixel is marked as a star point and written into a mask, and the pixel value of the pixel is replaced by the median or neighborhood statistical value of the corresponding direction, After iteration and morphological expansion, filling the mask region with neighborhood median or edge statistic to obtain spatially continuous star point suppression background map ; Suppressing background images at star points Removing global mean value to restrain direct current component, multiplying two-dimensional Hann window, calculating two-dimensional Fourier transform and centralizing to obtain centralized frequency domain image : ; In the formula, Representing the two-dimensional fourier transform operator, Representing a fourier transform centralising operator; Let the center line of the frequency domain be , Representing the width of a two-dimensional astronomical image, preserving column extent Wherein For determining the frequency range included in the template, constructing a frequency domain mask So that ; And performing frequency domain reservation to obtain a frequency domain image after frequency domain mask processing : ; Performing inverse centralization and inverse Fourier transformation to obtain an image domain two-dimensional fringe template: ; Wherein the method comprises the steps of The representation takes the real part operator, Representing a two-dimensional inverse fourier transform operator, Representing element-by-element multiplication; Representing the row and column coordinates of the frequency domain image respectively, Representing the coordinates of the frequency domain mask The pixel value at which it is located, Representing the fourier transform inverse centralisation operator.
9. The astronomical imaging stripe noise correction method based on low frequency directivity characteristics according to claim 1, wherein in step S4, under the constraint of a fixed trapezoid mask, an image domain two-dimensional stripe template is directly applied to a two-dimensional astronomical image with balanced column brightness to perform compensation correction, and an image with corrected stripe noise is output: ; Wherein the method comprises the steps of Compensating the intensity coefficient for the template for adjusting the correction amplitude, In order to fix the trapezoidal mask, Representing a two-dimensional fringe template of the image domain, And representing the two-dimensional astronomical image after column-level brightness equalization.
10. An astronomical imaging fringe noise correction system based on low frequency directivity characteristics, the system comprising: the vertical stripe brightness balancing module is used for estimating representative brightness of each column and performing column-level brightness translation by adopting robust statistics aiming at the acquired two-dimensional astronomical image data, and performing vertical stripe brightness balancing on the two-dimensional astronomical image to obtain a two-dimensional astronomical image with balanced column-level brightness; the low-frequency directivity characteristic thermodynamic diagram calculation module is used for sliding and sampling on the two-dimensional astronomical image with balanced column-level brightness according to set window size and step length, performing mean value removal, windowing and two-dimensional Fourier transformation on each window, extracting low-frequency girdle energy, counting the directional energy distribution according to angle bin division, calculating a low-frequency directivity coefficient, and weighting and averaging the window results according to coverage areas to generate a low-frequency directivity characteristic thermodynamic diagram which is the same in size as an original image and used for representing the spatial position and relative strength of residual stripes; The trapezoid fitting module is used for performing threshold segmentation, morphological closing operation and denoising after judging the contrast effectiveness of the low-frequency directional characteristic thermodynamic diagram, and only reserving a communication region with the largest area and conforming to the area proportion constraint as a main region mask; The system comprises a fringe template estimation and compensation correction module, a compensation correction module and a correction module, wherein the fringe template estimation and compensation correction module is used for performing star point inhibition or star point mask processing on an original two-dimensional astronomical image, performing two-dimensional Fourier transformation on the processed two-dimensional astronomical image, reserving only a vertical frequency band near the center of a frequency domain to obtain vertical fringes/low frequency components, and performing inverse transformation to obtain an image domain two-dimensional fringe template, and directly acting the image domain two-dimensional fringe template on the two-dimensional astronomical image with balanced column brightness under the constraint of a fixed trapezoid mask to perform compensation correction and output an image with corrected fringe noise.

Description

Astronomical imaging stripe noise correction method and system based on low-frequency directivity characteristics Technical Field The invention relates to the technical field of astronomical inspection data processing and image quality improvement, in particular to an astronomical imaging stripe noise correction method and system based on low-frequency directivity characteristics, which are suitable for automatic preprocessing of multichannel spliced CCD imaging data, and are particularly suitable for scientific images (SCIENCEIMAGE, SCI for short) acquired by an inspection camera. Background Astronomical night observation continuously acquires large-scale sky image data through a wide-field imaging system, and provides a key data basis for celestial body detection, photometry and positioning, transient source discovery and subsequent physical parameter inversion. In the data processing link of the night vision camera, a scientific image (SCI) generally refers to an original observed image (including celestial signals and background noise) directly obtained after exposure, and the quality of the original observed image directly affects the performance upper limit of key tasks such as source detection, background estimation, image superposition, difference and the like. Therefore, SCI-oriented image preprocessing and noise correction are important front links for guaranteeing the scientific output of the patrol data. With the increase of the detector scale and readout rate, the sky-following cameras increasingly adopt a multi-channel spliced CCD architecture to obtain a larger field of view and higher sampling efficiency. The system is usually formed by splicing a plurality of CCD devices to form a large area array, and multiple paths of parallel reading are adopted to meet the requirement of high-flux observation. In engineering practice, it is difficult to achieve perfect agreement between the gain, bias and noise characteristics of the different readout channels, and it is also possible to suffer from factors such as electronic crosstalk, clock interference, supply ripple and temperature drift, resulting in the occurrence of structured noise with significant directivity in SCI, one of the typical manifestations being streak noise extending in the row/column direction. The stripe noise has the characteristics of obvious directivity, prominent low-frequency components, non-stable spatial distribution and the like. On one hand, the stripes can be represented as integral brightness deviation or slow fluctuation superposition local periodic structures among columns, and on the other hand, the intensity and the morphology of the stripes can be changed along with the position, and the stripes are more remarkable in a boundary proximate area or a partial channel area. The structured noise can obviously interfere with the background statistical characteristics, so that the background RMS is raised, the noise correlation is enhanced, and further the problems of unstable threshold detection, light measurement zero point offset and the like are caused, and the structured noise is more prominent in weak signals, complex background fluctuation or star point dense scenes. In the existing data processing flow, conventional calibration correction (such as baseline offset correction) is usually performed first to reduce the influence of readout baseline drift, but a directional stripe structure may still remain in the actual SCI. For stripe noise, common engineering processing strategies can be roughly divided into two types, namely column-level correction based on statistical alignment, estimation of background representative values of each column (or narrow column band) and alignment to a unified reference to reduce overall inter-column offset, stripe processing based on frequency domain analysis, filtering or template estimation by positioning directional components corresponding to stripes in a frequency spectrum, and compensation correction of images by returning to an image domain. The above method can improve image quality in a specific scene, but still has disadvantages under the complex condition of the night-following SCI. For example, column level statistics are easily affected by bright stars, dense stars and abnormal bright spots to estimate instability, uniform frequency domain filtering of a full graph is difficult to achieve both positioning accuracy and background fidelity of local stripes, meanwhile, a starlight highlighting structure generates strong energy components in a frequency domain, and if an effective area constraint and interference suppression mechanism is lacking, stripe template estimation is easy to pollute, so that under correction, over correction or new artifact introduction are caused. In summary, although the existing method can reduce the fringe noise influence to a certain extent, the existing method still has defects in coping with scenes such as local non-stable fringes, close boun