An autonomous detection method for non-visible, unknown, and dimly lit targets in a single frame

Abstract

This invention discloses an autonomous detection method for single-frame non-visible, unknown, and faint targets. The method involves preprocessing the acquired observation image and setting a set of trial-position stacking parameters, including velocity and direction parameters, based on the preprocessed image. Visible stars in a single frame of the observation image are subtracted and compensated to form a trial-position image. Based on the trial-position stacking parameter set, the trial-position images are stacked in parallel to obtain a stacked trial-position image. The stacked trial-position image is then fitted with a one-dimensional distribution projected in the X and Y directions. The parameters within the trial-position stacking parameter set are corrected based on the fitting results. Information from the optimal stacked image is measured and output. This method enables autonomous detection of single-frame non-visible, unknown, and faint targets within the field of view. It addresses the limitation of large-field-of-view photoelectric telescopes in detecting single-frame non-visible, unknown, and faint targets, autonomously achieving signal accumulation and signal-to-noise ratio enhancement for unknown targets, ultimately realizing autonomous and effective detection and high-precision positioning and photometry of faint targets.

Description

Technical Field

[0001] This invention relates to the fields of astronomical image processing and space target observation technology, specifically to an autonomous detection method for single-frame non-visible, unknown, and faint targets. Background Technology

[0002] Ground-based photogrammetry remains irreplaceable in the observation of moving targets such as near-Earth small objects and artificial satellites due to its low cost and ease of deployment. In recent years, low-cost, ultra-large field-of-view field-based photoelectric telescopes have been widely used in this field due to their extremely high observation efficiency. Large field-of-view observation equipment generally employs a fast focal ratio, small-aperture optical system. While possessing high observation efficiency, its detection capability is limited by the small aperture, resulting in insufficient detection capability for faint targets, especially those with unknown apparent motion speeds. In such cases, increasing exposure time is insufficient for detection. Therefore, an autonomous detection method for single-frame, non-visible, unknown, faint targets is urgently needed to address this problem. Summary of the Invention

[0003] The purpose of this invention is to provide an autonomous detection method for single-frame non-visible, unknown, and weak targets, which can effectively solve the problems existing in the prior art.

[0004] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: an autonomous detection method for single-frame non-visible, unknown, and weak targets, which preprocesses the acquired observation images and performs the following operations based on the preprocessed observation images:

[0005] Set up the test position superposition parameter group, including velocity parameters and direction parameters;

[0006] The visible stars in a single frame of the observed image are removed and compensated for to form the test position image;

[0007] Based on the trial position stacking parameter set, parallel trial position stacking is performed on the trial position images to obtain a stacked trial position image; and the one-dimensional distribution of the projection of the stacked trial position image in the X and Y directions is fitted; and

[0008] The parameters within the test site superposition parameter group were adjusted based on the fitting results;

[0009] Repeat the above steps to measure and output information about the optimal overlay image.

[0010] Preferably, the sky region where faint space targets or near-Earth small celestial objects may appear is selected as the target, and high frame rate continuous image acquisition is performed to obtain observation images;

[0011] Preferably, the acquired observation images are preprocessed to remove the effects of skylight background, instrument background, and fixed-mode noise; the preprocessing includes:

[0012] Median filtering is applied to the observed image. A square window with an odd side length that is more than twice the radius of all stars is selected. The window is moved row by row and column by column, and the median value of all pixels in the window is taken as the value of the center pixel of the window to obtain the image after median filtering.

[0013] The filtered image will contain low-frequency noise such as sky background and instrument background. Dividing the original image by the filtered image will remove its uneven background.

[0014] Based on the dark field images captured by the camera, the fixed-pattern noise caused by the non-uniformity of dark current is removed, thus completing all preprocessing.

[0015] Preferably, the time span of the observed image sequence is defined as T seconds, the target motion velocity range is V1-V2, the full width at half maximum (FWHM) of the imaging point spread function of the imaging system is N pixels, and the rules for setting the test position overlay parameter group are as follows:

[0016] Starting from V1, iteratively accumulate the velocity step size on the velocity parameters, and simultaneously calculate the direction step size; stop when the accumulated velocity parameters exceed V2;

[0017] Wherein, the speed step size is , directional step size .

[0018] Preferably, the test position parameters of the test position superposition parameter group are such that the difference in position between the first and last frames of the image sequence is ≤2 pixels.

[0019] Preferably, the composition of the test image specifically includes:

[0020] The background mean and background noise of the observed image are statistically analyzed, and global binarization segmentation of the observed image is performed based on the statistical values.

[0021] Connected component labeling is performed on the resulting binarized image;

[0022] The processed connected component labeling results are statistically analyzed, and the connected regions that meet the threshold are output as star detection results.

[0023] Fill the pixel area containing the star with the average background value to complete the bright star compensation.

[0024] Preferably, the number of test site stacking parameter groups is defined as m, and parallel test site stacking is performed, including:

[0025] Based on m sets of test site superposition parameters, m image matrices are set accordingly;

[0026] The time-series observation images are traversed, and the translation amount of each observation image is calculated based on the set of parameters for each test position. The calculation method is as follows:

[0027] ;

[0028] Based on the translation amount, each frame of the observed image is shifted at a subpixel scale to obtain the translated image;

[0029] The translation images corresponding to each set of parameters are accumulated into the corresponding image matrix. After traversing the time-series observation images, the images are averaged.

[0030] Preferably, the subpixel-scale shift includes both "integer pixel shift" and "subpixel shift";

[0031] Let P(i,j) be the pixel value of the image, where i and j represent the row and column numbers of the pixel, respectively; let the displacement to be made in the observed image be (dx,dy), where dx and dy are in pixels;

[0032] The displacement for "integer pixel shift" is the integer parts of dx and dy, int(dx) and int(dy), respectively.

[0033] Q(i,j)=P(i+int(dx),j+int(dy));

[0034] The displacement of "subpixel shift" is the fractional part of dx and dy: dx' = dx - int(dx) and dy' = dy - int(dy).

[0035] Perform a "subpixel shift" in the row direction, that is:

[0036] R(i,j)=(1- dx') ×Q(i,j)+ dx' ×Q(i,j+1);

[0037] Perform a "subpixel shift" in the column direction, that is:

[0038] S(i,j)=(1- dy') ×R(i,j)+ dy'×R(i+1,j).

[0039] Preferably, a one-dimensional distribution is defined as follows: Where i = 1, 2, ..., n, and n is the number of pixels occupied by the stars, the one-dimensional distribution of the projections in the X and Y directions is fitted with a one-dimensional Gaussian function, including:

[0040] Construct the objective function ;

[0041] Where A is the amplitude and u is the mean. It is the standard deviation;

[0042] Calculate the parameters A and u that minimize S. And take the partial derivative of S with respect to each parameter, and set it equal to zero:

[0043] ;

[0044] The parameters are solved iteratively using numerical methods until the convergence condition is met, such as the sum of squares of the residuals no longer decreasing significantly.

[0045] Preferably, the fitting results in the X and Y directions and It can characterize the width of the point spread function projected in two directions. It can represent the magnitude of the star pattern elongation, and its relationship with the original point spread function. The difference can correct the positional parameter V. ; and The ratio can characterize the directional deviation of celestial superposition, and the angle it reflects can correct the position parameters. , .

[0046] Preferably, the optimal superimposed image is measured to calculate the centroid value, flux value, and signal-to-noise ratio value.

[0047] Preferably, the centroid value is calculated based on the moment method:

[0048] ;

[0049] in, For pixel values, B is the pixel position, and B is the background value.

[0050] Count all pixels belonging to the corresponding constellation, subtract the background value from all pixel values, and then sum them up to obtain the flow value;

[0051] Overall signal-to-noise ratio of the stars:

[0052] ;

[0053] Where F is the star flux value, and n is the number of pixels occupied by the stars. It is background noise in the image.

[0054] Beneficial effects: This invention provides an autonomous detection method for single-frame non-visible, unknown, and faint targets, which "freezes" (does not stretch) all moving targets and reference stars in the observed image; detects and compensates for single-frame visible bright stars to avoid interference with subsequent trial-position superimposed images; rationally sets the trial-position parameter group according to the apparent motion velocity of the target to ensure completeness and simplicity; achieves the detection of single-frame non-visible, unknown, and faint targets through parallel trial-position superimposition at the sub-pixel scale; and corrects the trial-position velocity and direction parameters by enhancing the fitting of the one-dimensional projection distribution of stars in the X and Y directions to achieve the best superimposition effect, realizing effective detection and high-precision measurement of single-frame non-visible, unknown, and faint targets within the field of view, and solving the problem of insufficient detection capability of commonly used small-aperture, large-field-of-view observation equipment. Attached Figure Description

[0055] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0056] In the attached diagram:

[0057] Figure 1 This is a leaked image of the autonomous detection method for single-frame non-visible, unknown, and dim targets of the present invention;

[0058] Figure 2 This is a visual velocity distribution diagram of a space target in a synchronous orbit according to the present invention;

[0059] Figure 3 This is a schematic diagram of subpixel image shifting according to the present invention;

[0060] Figure 4 This is a comparison image of the original single-frame image and the image after bright star compensation according to the present invention;

[0061] Figure 5 This is a comparison image of the target star outline before and after convergence in this invention. Detailed Implementation

[0062] To make the objectives and advantages of this invention clearer, the invention will be specifically described below with reference to embodiments. It should be understood that the following text is merely used to describe an autonomous detection method for single-frame non-visible, unknown, and weakly lit targets, or several specific implementation methods of this invention, and does not strictly limit the scope of protection specifically claimed by this invention.

[0063] Example: Figure 1 As shown, an autonomous detection method for single-frame non-visible unknown faint targets includes: high frame rate star image acquisition and preprocessing, test position parameter group setting, test position image preparation, parallel test position superposition of time-series images, star profile convergence, star parameter measurement and target information output.

[0064] In this embodiment, the telescope used has an aperture of 150mm and a focal length of 160mm; the detector is a Dhyana 4040 camera with a target size of 4096×4096×9um; wide-area short-exposure images are acquired at high frequency by a small-aperture ultra-large field-of-view optical telescope and an sCMOS camera, and the shorter exposure time can suppress the influence of the sky background.

[0065] Based on the above, the autonomous detection method for single-frame non-visible, unknown, and dimly lit targets is as follows:

[0066] Select sky regions where faint space targets or near-Earth small celestial objects may appear as objects, and perform high frame rate continuous image acquisition (generally not less than 10Hz) to obtain observation images;

[0067] Preferably, the acquired observation images are preprocessed to remove the effects of skylight background, instrument background, and fixed-mode noise; the preprocessing includes:

[0068] Median filtering is applied to the observed image. A square window with an odd side length that is more than twice the radius of all stars is selected. The window is moved row by row and column by column, and the median value of all pixels in the window is taken as the value of the center pixel of the window to obtain the image after median filtering.

[0069] The filtered image will contain low-frequency noise such as sky background and instrument background. Dividing the original image by the filtered image will remove its uneven background.

[0070] Based on the dark field images captured by the camera, the fixed-pattern noise caused by the non-uniformity of dark current is removed, thus completing all preprocessing.

[0071] Set up the test position superposition parameter group, including velocity parameters and direction parameters;

[0072] Among them, reasonable parameter values ​​and number of parameters are set for the test position superposition parameter group based on the resolution of the observation system, the full width at half maximum (FWHM) of the point spread function (PSF), the approximate range of the target's apparent motion velocity, and the computing power resources of the computer.

[0073] The resolution of the observation system is calculated based on the focal length of the optical system and the pixel scale of the detector.

[0074] resolution=arcsin(pixelscale / focallength);

[0075] For a 14°×14° field of view device and a 4K×4K detector, the resolution is approximately 12.3 arcseconds per pixel;

[0076] Estimate the FWHM of the point spread function of the observation system. Select stars with a suitable signal-to-noise ratio (around 10-20) in the observation image and estimate their FWHM. For large field-of-view observation images, the default FWHM is generally 2 pixels.

[0077] Estimate the target's maximum apparent velocity. When observing a target in a geosynchronous orbit (semi-major axis = 42164 ± 200 km, orbital inclination < 15°), the target's apparent velocity is estimated to be no more than 5 arcseconds / second (reference). Figure 2 As shown in the figure, when observing MEO targets, their apparent motion velocity is generally no more than 40 arcseconds / second; when observing LEO targets, their apparent motion velocity is in the range of hundreds to thousands of arcseconds per second.

[0078] If the difference in the center of the superimposed PSF is less than the FWHM, it is considered a valid superposition. That is, a difference in the center of the star is within 2 pixels (when the FWHM is 2 pixels). The space target can be considered to be moving in a uniform linear motion for a short period of time. Therefore, if a set of trial parameters makes the difference in the position of the target signal at the beginning and end of the image sequence ≤ 2 pixels, then this set of parameters is considered valid. Assuming the time span of the image sequence is T seconds, the target's motion speed range is V1~V2, and the FWHM of the imaging system PSF is N pixels, the rules for setting the trial superposition parameter group are as follows:

[0079] Starting from V1, iteratively accumulate the velocity step size on the velocity parameters, and simultaneously calculate the direction step size; stop when the accumulated velocity parameters exceed V2;

[0080] Wherein, the speed step size is , directional step size .

[0081] In a specific case: taking N=2 pixels, V1=0 pixels / second, V2=0.5 pixels / second, and T=10 seconds as an example, to calculate the trial position parameter set, we can first calculate... The pixel-per-second coverage parameters for V2 are shown in the table below.

[0082]

[0083] To cover targets with greater speeds of motion, the number of test parameters needs to be increased;

[0084] The test positioning image is constructed by subtracting visible stars from a single frame of the observed image and then compensating for them, thus avoiding interference and influence from bright stars visible in a single frame on subsequent shifting and superposition. The test positioning image specifically includes:

[0085] The background mean and background noise of the observed image are statistically analyzed, and global binarization segmentation of the observed image is performed based on the statistical values.

[0086] Connectivity component labeling is performed on the processed binarized image, such as using the 8-connectivity method;

[0087] The processed connected component labeling results are statistically analyzed, and the connected regions that meet the threshold are output as star detection results. For conventional astronomical observation images, three times the background standard deviation is generally taken as the threshold, and regions higher than three times the background standard deviation are considered star regions.

[0088] Fill the pixel area containing the star with the background mean to complete the bright star compensation, such as... Figure 4 As shown, Figure 4 The image on the left is the original star map, and the image on the right is the image after bright star compensation.

[0089] Define m sets of parameter data for trial position stacking, and perform parallel trial position stacking, including:

[0090] Based on m sets of test position superposition parameters, define m image matrices in computer memory (if computer memory is insufficient, process in batches).

[0091] The time-series observation images are traversed, and the translation amount of each observation image is calculated based on the set of parameters for each test position. The calculation method is as follows:

[0092] ;

[0093] Based on the translation amount, each frame of the observed image is shifted at a subpixel scale to obtain the translated image;

[0094] The translation images corresponding to each set of parameters are accumulated into the corresponding image matrix. After traversing the time-series observation images, the images are averaged to complete the parallel trial-position superposition enhancement processing.

[0095] Subpixel-scale shifting includes "integer pixel shifting" and "subpixel shifting";

[0096] Let P(i,j) be the pixel value of the image, where i and j represent the row and column numbers of the pixel, respectively; let the displacement to be made in the observed image be (dx,dy), where dx and dy are in pixels;

[0097] The displacement for "integer pixel shift" is the integer parts of dx and dy, int(dx) and int(dy), respectively.

[0098] Q(i,j)=P(i+int(dx),j+int(dy));

[0099] The displacement of "subpixel shift" is the fractional part of dx and dy: dx' = dx - int(dx) and dy' = dy - int(dy).

[0100] like Figure 3As shown, the first step is to perform a "sub-pixel shift" in the row direction, that is:

[0101] R(i,j)=(1- dx') ×Q(i,j)+ dx' ×Q(i,j+1);

[0102] Perform a "subpixel shift" in the column direction, that is:

[0103] S(i,j)=(1- dy') ×R(i,j)+ dy'×R(i+1,j).

[0104] After the above operations, a translated image is obtained;

[0105] Perform the following operations again on the superimposed images of the test subjects:

[0106] The background mean and background noise of the image are statistically analyzed, and the image is then segmented globally based on the statistical values.

[0107] Connected component labeling is performed on the resulting binarized image;

[0108] The processed connected component labeling results are statistically analyzed, and the connected regions that meet the relevant thresholds are output as star detection results.

[0109] A window larger than the star image and including part of the background is defined. The windowed image is cropped with the centroid of the star as the center. The summation is performed in the X and Y directions of the windowed image, yielding two one-dimensional distributions. These can be considered as projections of the star point spread function in the X and Y directions. Fitting these two one-dimensional distributions to a low signal-to-noise ratio results in faster convergence and lower computational cost compared to direct two-dimensional fitting.

[0110] Define a one-dimensional distribution as Where i = 1, 2, ..., n, n is the number of pixels occupied by the image. A one-dimensional distribution of the projections in the X and Y directions is fitted using a one-dimensional Gaussian function, including:

[0111] Construct the objective function ;

[0112] Where A is the amplitude and u is the mean. It is the standard deviation; define the residual function. ;

[0113] Calculate the parameters A and u that minimize S. And take the partial derivative of S with respect to each parameter, and set it equal to zero:

[0114] ;

[0115] The parameters are solved iteratively using numerical methods until the convergence condition is met, such as when the sum of squares of the residuals no longer decreases significantly, at which point the best estimate of the parameters can be obtained.

[0116] Correction: Fitting results in the X and Y directions and It can characterize the width of the point spread function projected in two directions. It can represent the magnitude of the star pattern elongation, and its relationship with the original point spread function. The difference can correct the positional parameter V. ; and The ratio can characterize the directional deviation of celestial superposition, and the angle it reflects can correct the position parameters. , .

[0117] After correcting the positioning parameters, the image sequence after bright star compensation is re-superimposed and enhanced to obtain the superimposed enhanced result with the highest signal-to-noise ratio. This is the optimal superimposed image, which not only solves the detection problem but also improves the measurement accuracy of celestial phenomena. Figure 5 As shown, Figure 5 The left and middle images show the outlines of the stars in the overlay image under different positional parameters. Figure 5 The right-hand figure shows the star profile after the trial position parameters were corrected, resulting in effective convergence and enhancement;

[0118] The optimal overlay image is measured, and the centroid value, flux value, and signal-to-noise ratio value are calculated.

[0119] Calculation of centroid value based on moment method:

[0120] ;

[0121] in, For pixel values, B is the pixel position, and B is the background value.

[0122] Count all pixels belonging to the corresponding constellation, subtract the background value from all pixel values, and then sum them up to obtain the flow value;

[0123] Overall signal-to-noise ratio of the stars:

[0124] ;

[0125] Where F is the star flux value, and n is the number of pixels occupied by the stars. It is background noise in the image;

[0126] The enhanced information will be output and saved for later processing.

[0127] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. For those skilled in the art, after learning the contents described in the present invention, several equivalent changes and substitutions can be made without departing from the principle of the present invention. These equivalent changes and substitutions should also be considered to fall within the protection scope of the present invention.

Claims

1. An autonomous detection method of single frame non-visible unknown dim and weak targets, characterized in that: The collected observation images are preprocessed, and based on the preprocessed observation images, the following operations are performed: Set the test position superposition parameter group, including the speed parameter and the direction parameter; Subtract the star image visible in a single frame of the observation image and perform compensation to form a test position image; Based on the test position superposition parameter group, perform parallel test position superposition on the test position image to obtain a test position superposition image; and perform fitting on the one-dimensional distribution of the test position superposition image in the X and Y direction projection; and Based on the fitting result, correct the parameters in the test position superposition parameter group; Repeat the above operations to measure and output the information of the optimal superposition image.

2. The method according to claim 1, wherein the method is characterized in that: The observation image sequence time span is defined as T seconds, the target motion speed interval is V1-V2, the imaging system imaging point spread function full width at half maximum is N pixels, and the test position superposition parameter group setting rule is: Start from V1, iteratively accumulate the speed parameter by a speed step size and calculate the direction step size at the same time; stop when the speed parameter accumulation exceeds V2; wherein the velocity step size is and the direction step size is ; FWHM is full width at half maximum.

3. The method according to claim 2, wherein the method is characterized in that: The test position parameters of the test position superposition parameter group make the target signal have a position difference of ≤2 pixels at the beginning and end frames of the image sequence.

4. The method of claim 1, wherein the method is characterized by: The test position image formation specifically includes: Statistically analyze the background mean and background noise of the observation image, and perform global binaryzation segmentation on the observation image based on the statistical values; Perform connected domain labeling on the binaryzation image obtained by processing; Statistically analyze the connected domain labeling result obtained by processing, and output the connected regions meeting the threshold value as the star image detection result; Fill the pixel region where the star image is located with the background mean value to complete the bright star compensation.

5. The method according to claim 2, wherein the method is characterized in that: The test position superposition parameter group data is defined as m, and parallel test position superposition is performed, including: Based on the m test position superposition parameter groups, m image matrices are set correspondingly; Traverse the time sequence images, and calculate each frame of observation according to each test position superposition parameter group. The calculation method is as follows: ； For the translation amount of the image, according to the translation amount, each frame of the observed image is sub-pixel scale shifted to obtain a translation image; Add the translated image corresponding to each group of parameters to the corresponding image matrix. After traversing the time sequence observation images, take the average of the images.

6. The autonomous detection method of a single frame of non-visible unknown dim and weak target according to claim 5, characterized in that: The sub-pixel scale shift includes "integer pixel shift" and "sub-pixel shift".

7. The method according to claim 5, wherein the method is characterized by: The one-dimensional distribution is defined as where i = 1, 2, … n, n is the number of pixels occupied by the star image, and the one-dimensional distribution of the X and Y direction projections of the Gaussian function is fitted. Constructing the objective function ; where A is the amplitude, u is the mean, is the standard deviation; Compute the parameters A, u that minimize S, and take the partial derivatives of S with respect to each parameter and set them equal to zero: ； Iteratively solve the parameters based on the numerical method until the convergence condition is met.

8. The autonomous detection method of a single frame of non-visible unknown dim and weak target according to claim 7, characterized in that: The X and Y direction fitting results are and wherein: based on and the difference correction parameter V; ； Based on and the ratio of the modified parameters ; 。 9. The method of claim 1, wherein the method is characterized by: Measure the optimal superposition image, calculate the centroid value, flow value, and signal-to-noise ratio value.

10. The autonomous detection method of a single frame of non-visible unknown dim and weak target according to claim 9, characterized in that: Calculate the centroid value based on the moment method: ； wherein, is a pixel value, is a pixel position, B is a background value; Statistically analyze all the pixels belonging to the corresponding star image, subtract the background from all the pixel values, and sum them up to obtain the flow value; The overall signal-to-noise ratio of the star image is: ； where F is the star image flux value, n is the number of pixels occupied by the star image, is the background noise of the image.

Images