A method for detecting space targets against a stellar background

Abstract

This invention discloses a method for detecting space targets against a stellar background. After background removal from a star map, local gray-level maxima are extracted. These maxima are then filtered using a sub-image of the target and a Gaussian kernel function. After star map registration, the target is extracted from the star map. Before filtering, background noise in the star map is removed, achieving effective separation of the background and the target. This invention uses three indicators: the similarity between the Gaussian kernel function and the sub-image, the degree of energy concentration at the center of the sub-image, and the trend of gray-level change relative to the center of the sub-image. Suspected targets are filtered from local gray-level maxima, resulting in highly targeted filtering with a low probability of false positives and false negatives. This invention uses star map registration to remove stars contained within suspected targets, completing the separation of the target from the stars. This invention updates the motion step size based on the spacing between suspected targets in different frames, and associates the trajectories of the same target in each frame's star map based on the motion step size, supplementing the falsely removed suspected targets.

Description

Technical Field

[0001] This invention relates to the field of space target detection technology, and more specifically to a method for detecting space targets against a stellar background. Background Technology

[0002] Currently, the number of space targets is growing exponentially, and the probability of collisions with spacecraft in orbit is also increasing. Tens of thousands of space targets pose a serious threat to the normal operation of spacecraft in orbit, making the monitoring and tracking of space targets essential. Extracting space targets is a crucial step in this monitoring and tracking process. Therefore, there is an urgent need for a method to detect faint space targets against a stellar background, which can effectively extract these targets. Summary of the Invention

[0003] In view of this, the present invention provides a method for detecting space targets against a stellar background, which can separate the background, stars and targets to achieve effective target extraction.

[0004] To achieve the above-mentioned objectives, the technical solution of this invention is as follows:

[0005] A method for detecting space targets against a stellar background, comprising the following steps:

[0006] S1. The satellite payload continuously captures images of the target's location in space to obtain star maps; background noise in the star maps is removed, and local gray-level maxima are extracted from each frame; sub-images of the local gray-level maxima are extracted, and a Gaussian kernel function of the same scale as the sub-image is created; the similarity E between the Gaussian kernel function and the sub-image is calculated respectively. R The degree to which the energy of a subgraph is concentrated at its center, E C The trend of gray level change of the sub-image relative to the center of the sub-image, E T ; Filter out E R E C E T Local gray-scale maxima that meet the corresponding set thresholds are considered as suspected targets.

[0007] S2. Perform star map registration on two adjacent star maps and remove star points in suspected targets.

[0008] Furthermore, it also includes S3: performing trajectory correlation on the suspected targets obtained in S2 frame by frame, and adding targets that were removed due to overlap with star points.

[0009] Furthermore, in S1, energy similarity is used to measure the similarity E between the subgraph and the Gaussian kernel function. R Its expression is:

[0010]

[0011] Among them, ER (x,y) is the energy similarity function of the subgraph; f(x,y) is the subgraph; ω(x,y) is the weight function; g(x,y) is the Gaussian kernel function; x is the x-coordinate of a point in the subgraph f(x,y), and y is the y-coordinate of a point in the subgraph f(x,y).

[0012] In S1, energy concentration is used to measure the degree to which the energy of a subgraph is concentrated at its center. C Its expression is:

[0013]

[0014] Among them, E C (x,y) is the energy concentration function of the subgraph; S sub Let f(x,y) be a subregion that is 2σ away from the local maximum point (x0,y0), where σ is the variance of the subgraph f(x,y). norm (x,y) is the normalization function of f(x,y).

[0015] In S1, the energy transfer degree is used to measure the trend of the gray value of the sub-image relative to the center of the sub-image. T Its expression is:

[0016]

[0017] Among them, E T (x,y) is the energy transfer function of the subgraph, and D(x,y) is the local deviation judgment value corresponding to the point (x,y).

[0018] Furthermore, the specific method of S2 is as follows:

[0019] The star map is divided into multiple regions, and the suspected target with the highest brightness value in each region is selected as the registration point. The RANSAC algorithm is used to calculate the affine transformation matrix of two adjacent star maps. The registration point of the later star map is transformed onto the previous star map through the affine transformation matrix. The position where the registration point appears in two adjacent star maps is marked as the position of the star, and the suspected target at the position is removed.

[0020] Furthermore, in S1, the background histogram mode estimation method is used to remove background noise from the star map. The specific method is as follows:

[0021] The star map is divided into multiple local regions. Within each local region, the pixel mean (Mean), median (Med), and standard deviation (λ) are calculated multiple times. Pixel values ​​that exceed Med±λ are removed each time until all pixel values ​​do not exceed Med±λ. At this point, all pixels form the background histogram of the star map.

[0022] The background value for the local area is selected based on the variation range of the standard deviation λ of the background histogram. The pixel mean (Mean) or the mode estimate of the histogram (Mode = 2.5 × Med - 1.5 × Mean) is used. When the variation range is greater than 20%, the star map is highly uneven and the background value is the mode estimate of the histogram (Mode). When the variation range is no greater than 20%, the star map is not highly uneven and the background value is the pixel mean (Mean).

[0023] Median filtering is applied to the background values ​​of the star map to obtain the background noise; the background noise is then subtracted from the star map.

[0024] Furthermore, the specific method of S3 is as follows:

[0025] S31. Starting from the first frame of the star map, estimate the area where a suspected target x1 is located on the second frame of the star map based on the target's velocity range, and search for suspected targets within this area; if no suspected target that has not been matched or has been matched is found, the trajectory association process for suspected target x1 ends.

[0026] S32. After matching the suspected target x2 on the second frame star map, calculate the distance between the suspected targets on the first and second frame star maps as the motion step size, use the motion step size as the distance between the suspected targets on the second and third frame star maps, and predict the position of the suspected target on the third frame star map.

[0027] S33. Using the predicted position of the suspected target on the third frame star map as the center, draw a circular area on the third frame star map with a preset value k as the radius to search for the suspected target; if no unmatched or suspected matched suspected targets are found, return to S31 to continue execution until all suspected targets in the second frame star map have been matched, and then end the trajectory association process.

[0028] S34. In the subsequent frame star map, update the motion step size according to the spacing of suspected targets between frames, and predict the position of suspected targets in the next frame star map; with the predicted position of the suspected target in the previous frame star map as the center and a preset value k as the radius, draw a circular area on the next frame star map to search for unmatched or suspected matched suspected targets; when a suspected target is found, return to the beginning of S34 to continue execution; when no suspected target is found, and no suspected target is found in the previous two frames, supplement candidate points on the next frame star map, and return to the beginning of S34 to continue execution; when no suspected target is found, and no suspected target is found in the previous two frames, end the trajectory association process.

[0029] Furthermore, the centroid of the trailing target among the suspected targets obtained in S1 is located, and then S2 is executed. The specific method of centroid location is as follows: construct its point spread function with the centroid of the trailing target as the center, take the logarithm of the point spread function to calculate the coefficients, and solve the centroid coordinates based on the coefficients.

[0030] Beneficial effects:

[0031] 1. This invention proposes a method for detecting space targets against a stellar background. After background removal from a star map, local gray-level maxima are extracted. These are then filtered using a sub-image of the target and a Gaussian kernel function. The target is extracted from the star map after registration. Before filtering, this invention removes background noise from the star map, achieving effective separation between the background and the target. This invention sets the similarity E between the Gaussian kernel function and the sub-image. R The degree to which the energy of a subgraph is concentrated at its center, E C The trend of gray level change of the sub-image relative to the center of the sub-image, E T Three indicators are used to screen for suspected targets from local gray-scale maxima, resulting in highly targeted screening with a low probability of false positives and false negatives. This invention employs a star map registration method to remove stars contained within suspected targets, thus separating the target from the stars.

[0032] 2. To further improve extraction accuracy and prevent missed detections, this invention also performs trajectory association on targets separated from stars, and supplements targets that were eliminated due to overlap with stars based on their motion characteristics.

[0033] 3. Based on the characteristics of the large field of view of the star map, the present invention divides it into multiple regions for separate registration, which further improves the detection accuracy and speed.

[0034] 4. This invention uses the mode estimation method of background histogram to remove background noise in star maps. The background value is calculated by observing the standard deviation of the background histogram of the star map. It is applicable to star maps with different degrees of background non-uniformity. When calculating the background value, this invention divides the star map into multiple regions for separate calculation based on the characteristics of the large field of view of the star map, which further improves the detection accuracy and speed.

[0035] 5. This invention uses the motion characteristics of the target as a reference and uses the motion step size to associate the trajectory of the same target in each frame of the star map. This can avoid matching errors caused by differences in motion speed and direction, obtain accurate motion trajectories, and correct the motion characteristics of the target. At the same time, it supplements the rejected suspected targets. These suspected targets coincide with the positions of stars and were mistakenly removed during star map registration.

[0036] 6. Due to camera movement and stray light, some suspected targets may exhibit a trailing phenomenon. Before performing star map registration, this invention will use the derivative of the point spread function to locate the centroid of suspected targets in order to improve the subsequent registration accuracy and achieve effective removal of stars. Attached Figure Description

[0037] Figure 1 This is a flowchart of the method in this embodiment.

[0038] Figure 2This is a schematic diagram of the trajectory association method used in this invention. Detailed Implementation

[0039] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0040] Star maps are fundamental for target extraction, primarily composed of the target, stars, and background noise, and can be acquired from satellite payloads (such as imaging devices like cameras). The imaging characteristic of star maps is the discrete distribution of approximately Gaussian-distributed point-like light spots against a dark background. Some of these point-like spots represent the target to be extracted, while others represent stars. This invention has found that star maps are affected by factors such as illumination, stray light, and camera thermal noise, resulting in significant and uneven background noise. Furthermore, both stars and targets are faint, with low signal-to-noise ratios, making them easily overwhelmed by background noise. When background noise interference is severe, extracting faint targets becomes very difficult. Therefore, to achieve accurate and effective target extraction, it is necessary not only to remove background noise but also to separate the target from the stars. Figure 1 As shown, this invention provides a method for detecting space targets against a stellar background, comprising the following steps:

[0041] Step 1: Remove background noise using a block-based background histogram mode estimation method.

[0042] After acquiring the star map, its background value is calculated to remove background noise. Because star maps have a large field of view and uneven noise distribution, this invention divides them into multiple local regions and calculates the background value for each region. Specifically, this invention calculates the pixel average value (Mean), median value (Med), and standard deviation (λ) multiple times within each local region, and removes pixel values ​​exceeding Med±λ each time, until all pixel values ​​do not exceed Med±λ; based on the difference in the standard deviation variation Δλ, the pixel average value (Mean) is selected. At this point, the star chart unevenness is low, or Mode = 2.5 × Med - 1.5 × Mean. At this point, the star map shows a high degree of unevenness, which is used as the background value for that local area. Median filtering is applied to the background value of the entire star map to obtain the background noise, which is then subtracted from the star map.

[0043] Step 2: Use an adaptive threshold method to screen potential targets:

[0044] To extract as many targets as possible, this invention treats both the target and stars as potential targets, separating them in step 4. First, let's analyze the imaging characteristics of potential targets: Noise from stray light in the star image can affect stars and targets with lower magnitudes, reducing their signal-to-noise ratio or even causing them to be completely overwhelmed by noise, becoming faint targets. Furthermore, during star image acquisition, the camera and its detection object in the space-based detection system may be in motion, causing image displacement of stars and targets within the exposure time, forming trailing targets and affecting positioning accuracy. These two reasons result in the imaging of potential targets appearing as a bright spot with an approximately Gaussian distribution, the highest central gray value, and energy concentrated at the center.

[0045] Based on these imaging characteristics, this invention proposes an adaptive threshold method to screen suspected targets, thereby reducing the probability of false positives and false negatives:

[0046] Step 21: Extract local gray-scale maxima on the star map.

[0047] Step 22: Using the local gray-level maxima as the center, extract its sub-image f(x,y) at a set scale, and generate a Gaussian kernel function g(x,y) representing the gray-level changes of this sub-image at the same scale. Its expression is:

[0048]

[0049]

[0050]

[0051] Where g(x,y) is the gray value at point (x,y), (x,y) is the coordinate of any point in the subimage, (x0,y0) is the center of the bright spot (i.e., the local maximum point), k is a constant, and σ is the variance of the gray values ​​in the subimage.

[0052] Step 23: Calculate the energy similarity, energy concentration, and energy transfer of local gray-level maxima. Compare these three metrics with a set threshold to select local gray-level maxima that meet the imaging characteristics as potential targets. The expressions for the three metrics are:

[0053] Energy similarity is used to characterize the degree of similarity between a subgraph and a Gaussian kernel function. R Its expression is:

[0054]

[0055] Among them, E R (x,y) is the energy similarity function of the subgraph, and ω(x,y) is the weight function.

[0056] Energy concentration is used to characterize the degree to which the energy of a subgraph is concentrated at its center.C Its expression is:

[0057]

[0058] Among them, E C (x,y) is the energy concentration function; S sub Let f(x,y) be a subregion that is 2σ away from the local maximum point (x0,y0), where σ is the variance of the subgraph f(x,y). norm (x,y) is the normalization function of f(x,y).

[0059] Energy transfer measure is the trend of the gray values ​​of a subimage relative to the local maxima (x0, y0). T Its expression is:

[0060]

[0061]

[0062] Among them, E T (x,y) is the energy transfer function, and D(x,y) is the local deviation judgment value corresponding to the point (x,y). φ The threshold for energy transfer; φ diff (x,y) is the gradient function.

[0063] Step 3: Locate the centroid of the suspected target:

[0064] To eliminate the impact of trailing targets among suspected targets on positioning accuracy, this invention constructs a point spread function centered on the current centroid of the trailing target. The logarithm of the point spread function is taken to calculate various coefficients, and the centroid coordinates are corrected based on these coefficients to improve positioning accuracy. The expression for the corrected centroid coordinates is as follows:

[0065]

[0066] Where t0, t1, t2, t3, t4, and t5 are the constant term, x-coefficient, y-coefficient, xy-coefficient, and x, respectively, of the logarithm of the point spread function. 2 Term coefficient, y 2 Term coefficient.

[0067] Step 4: Separate the target from the stars using star map registration:

[0068] Since the positions of stars and targets change between adjacent star chart frames, it is difficult to separate them using trajectories. Therefore, stars need to be removed through star chart matching before trajectory association. Because the star chart has a large field of view, to improve registration accuracy, this invention divides the star chart into multiple regions (e.g., 16×16 squares). Within each region, the point with the highest DN value (luminance value) is selected as the registration point. An affine transformation matrix is ​​used to transform the registration point from the later frame onto the earlier frame. The positions where consecutive registration points appear are marked as star positions, and suspected targets at those positions are removed. The affine transformation matrix is ​​calculated as follows: registration points within the same region of the two consecutive star chart frames are considered as a pair. The RANSAC algorithm (Random Sample Consensus) is used to determine the matching relationship of each registration point in the pair, thus obtaining the affine transformation matrix between adjacent star chart frames.

[0069] Step 5: Associate the trajectory of the target:

[0070] Targets move in different directions and at different speeds, but share similar motion characteristics. Within a short timeframe, their motion can be equated to uniform linear motion, which in a star map manifests as moving the same number of pixels each time. Utilizing this characteristic, this invention, while establishing the target's trajectory, also supplements targets that were previously excluded due to overlap with stars, further improving detection accuracy. For example... Figure 2 As shown. The specific steps for trajectory association are as follows:

[0071] S51. Starting from the first frame of the star map, estimate the area where a suspected target x1 is located on the second frame of the star map based on the target's velocity range, and search for suspected targets within this area; if no suspected target that has not been matched or has been matched is found, the trajectory association process for suspected target x1 ends.

[0072] S52. After matching the suspected target x2 on the second frame star map, calculate the distance between the suspected targets on the first and second frame star maps as the motion step size, use the motion step size as the distance between the suspected targets on the second and third frame star maps, and predict the position of the suspected target on the third frame star map.

[0073] S53. Using the predicted position of the suspected target on the third frame star map as the center, draw a circular area on the third frame star map with a preset value k as the radius to search for the suspected target; if no unmatched or potentially matched suspected target is found, return to S51 to continue execution until all suspected targets in the second frame star map have been matched, then the trajectory association process ends. In this embodiment, the preset value k can be 3.

[0074] S54. In the subsequent frame star map, update the motion step size based on the distance between the suspected target and the two frames, and predict the position of the suspected target in the next frame star map; using the predicted position of the suspected target in the previous frame star map as the center, draw a circular area on the next frame star map with a preset value k as the radius to search for suspected targets that have not been matched or have been matched before; when a suspected target is found, return to the beginning of S54 to continue execution; when no suspected target is found, and no suspected target is found in the previous two frames, supplement candidate points on the next frame star map, and return to the beginning of S54 to continue execution; when no suspected target is found, and no suspected target is found in the previous two frames, end the trajectory association process.

[0075] Based on the prior target motion model, the target's velocity range in each frame is estimated, and the search domain is divided on the star map of the next frame according to the velocity range. For example... Figure 2 As shown, the search domain searches for potential targets that have not been matched with the target (if there are no unmatched targets, potential matches are also acceptable), representing the position of the target in the next frame; when all potential targets in the search domain have been matched, the target is changed and execution returns to the beginning of S51 to continue.

[0076] S52. Calculate the distance between the target and the two frames, and use it as the movement step length to predict the target's position in the next frame; with the predicted position as the center, set a radius (e.g., a radius of 3 times the movement step length) to divide the search area on the star map of the next frame, search for suspected targets within the search area, and represent the position of the target in the next frame; when there are no suspected targets within the search area, use the predicted position to represent the position of the target in the next frame.

[0077] S53. Repeat S52 to obtain the target's position in subsequent frames and form a trajectory; when no suspected target is found for three consecutive frames, the trajectory association process for that target ends.

[0078] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting a space object against a stellar background, characterized by the steps of include: S1, the satellite payload continuously photographs the space where the target is located to obtain a star map; background noise in the star map is removed, and local gray maximum points in each frame of the star map are extracted; a subgraph of the local gray maximum points is extracted, and a Gaussian kernel function with the same scale as the subgraph is established; similarity of the Gaussian kernel function and the subgraph is calculated respectively , degree of energy concentration of the subgraph in the center , change trend of the gray of the subgraph relative to the center of the subgraph ; local gray maximum points that satisfy corresponding set thresholds are screened out as suspected targets , , ​ S2. Perform star map registration on two adjacent star maps and remove star points in suspected targets; It also includes S3: performing trajectory correlation on a frame-by-frame basis for suspected targets and adding targets that were removed due to overlap with star points; The centroid of the trailing target among the suspected targets obtained in S1 is located, and then S2 is executed. The specific method of centroid location is as follows: construct its point spread function with the centroid of the trailing target as the center, take the logarithm of the point spread function to calculate the coefficients, and solve the centroid coordinates based on the coefficients.

2. The method as described in claim 1, characterized in that, In S1, energy similarity is used to measure the similarity between the subgraph and the Gaussian kernel function. Its expression is: in, The energy similarity function for the subgraph; For subgraphs; For weighting functions; The Gaussian kernel function; For subgraph The x-coordinate of the midpoint For subgraph The ordinate of the midpoint; In S1, energy concentration is used to measure the degree to which the energy of a subgraph is concentrated at its center. Its expression is: in, The energy concentration function of the subgraph; For local maxima distance subregions, For subgraph The variance; for The normalization function; In S1, energy transfer degree is used to measure the trend of gray value change of the sub-image relative to the center of the sub-image. Its expression is: in, The energy transfer function of the subgraph. For point The corresponding local deviation judgment value.

3. The method as described in claim 1, characterized in that, The specific method of S2 is as follows: The star map is divided into multiple regions, and the suspected target with the highest brightness value in each region is selected as the registration point. The RANSAC algorithm is used to calculate the affine transformation matrix of two adjacent star maps. The registration point of the later star map is transformed onto the previous star map through the affine transformation matrix. The position where the registration point appears in two adjacent star maps is marked as the position of the star, and the suspected target at the position is removed.

4. The method as described in claim 1, characterized in that, In S1, the background noise in the star map is removed using the background histogram mode estimation method. The specific method is as follows: The star map is divided into multiple local regions, and the pixel average value is calculated multiple times within each local region. Median and standard deviation And will exceed each time Remove pixel values ​​until all pixel values ​​do not exceed [the specified value]. At this point, all pixels form the background histogram of the star map; Based on the standard deviation of the background histogram The change range is selected by the pixel average value. Or histogram mode estimate As a background value for this local area, when the change is greater than 20%, the star map shows high unevenness, and the background value is the histogram mode estimate. When the change is no greater than 20%, the star map has low unevenness, and the background value is the pixel average. ; Median filtering is applied to the background values ​​of the star map to obtain the background noise; the background noise is then subtracted from the star map.

5. The method as described in claim 1, characterized in that, The specific method for S3 is as follows: S31. Starting from the first frame of the star map, estimate a suspected target based on the target's velocity range. x 1. In the region where the target is located on the second frame of the star map, search for potential targets within that region; if no unmatched or potentially matched potential targets are found, terminate the search for potential targets. x The trajectory association process of 1; S32, A suspected target was matched on the second frame of the star map. x 2. Then, the distance between the suspected targets on the first and second frame star maps is calculated as the motion step size. The motion step size is used as the distance between the suspected targets on the second and third frame star maps to predict the position of the suspected targets on the third frame star map. S33. Using the predicted position of the suspected target on the third frame star map as the center, draw a circular area on the third frame star map with a preset value k as the radius to search for the suspected target; if no unmatched or suspected matched targets are found, return to S31 to continue execution until all suspected targets in the second frame star map are matched and the trajectory association process ends. S34. In the subsequent frame star map, update the motion step size according to the spacing of suspected targets between frames, and predict the position of suspected targets in the next frame star map; with the predicted position of the suspected target in the previous frame star map as the center and a preset value k as the radius, draw a circular area on the next frame star map to search for unmatched or suspected matched suspected targets; when a suspected target is found, return to the beginning of S34 to continue execution; when no suspected target is found, and no suspected target is found in the previous two frames, supplement candidate points on the next frame star map, and return to the beginning of S34 to continue execution; when no suspected target is found, and no suspected target is found in the previous two frames, end the trajectory association process.

Images