Dark space target detection and track prediction method, system and device under dense fixed star background and medium

Abstract

A method, system, device and medium for detecting a dark space target and predicting a track under a dense fixed star background comprises the following steps: firstly, obtaining a reconstructed multi-pulse function by using a frequency spectrum phase difference, further estimating motion parameters of a dark star background and a space target, and then designing an image difference fusion coarse processing module to obtain a candidate target image sequence; the number of tracks is effectively reduced; and finally, according to the motion parameters and the candidate target image sequence, directional track association is carried out, an association area is reduced to a fan-shaped range, and redundant false alarms are removed while the target track is accurately and rapidly detected. The system, the equipment and the medium are used for realizing the invention. Based on the distribution characteristic of the space target and the time domain motion characteristic, the method not only achieves the precise suppression of fixed stars with different brightness, but also improves the detection robustness of targets with different star magnitudes, and has a better detection effect on dark space targets with different motion parameters under the background of dense fixed stars.

Description

Technical Field

[0001] The present invention belongs to the technical field of space target monitoring, and in particular relates to a method, system, equipment and medium for detecting and predicting a dark space target under a dense star background. Background Art

[0002] With the rapid development of the aerospace industry, the number of spacecraft launched into space by humans has continued to increase, resulting in a large number of failed flying objects and debris, which pose a serious threat to the safety of spacecraft in orbit. Therefore, it is urgent to identify, detect and track these targets to ensure the safety of effective spacecraft and adapt to the needs of future space development. Due to the long distance between the space target and the observation equipment, the target is only a small number of pixels in the image. As the magnitude of the stars in the field of view increases, the difficulties in detecting space targets are mainly in the following aspects: densely distributed bright and dark stars and space targets have similar shapes and strong interference; the target brightness is weak, without obvious features, and it is easy to be submerged in the deep space background; the space target has different directions and speeds, and the trajectory is discontinuous.

[0003] At present, traditional detection methods mainly use the spatial and temporal characteristics of the target for detection. The image filtering-based method uses pixel grayscale information to estimate the image background and is sensitive to noise. The human vision method that combines the spatial and temporal contrast features is difficult to distinguish stars similar to the target, and the false alarm is high. Low-rank sparse and optical flow methods require complex calculation processes and do not meet the real-time requirements. Matched filtering and other methods require understanding of the target morphology or motion priors and are not robust for dim target detection. Multi-frame image difference is prone to mismatching, target omission, and dark star residues in the background of dense stars. Multi-level hypothesis testing, fast track prediction and other methods require the establishment of a screening tree, and large-scale blind detection of each frame is performed through inter-frame motion prediction. There are problems such as slow detection speed, inaccurate elimination, and low detection accuracy. In short, the core of the detection process is target recognition first and then multi-frame positioning. There is still room for improvement in noise sensitivity, star suppression, robust detection, and real-time performance.

[0004] Reference 1 [Kong Sijie. Research on the extraction technology of faint targets under dense star background [D]. University of Chinese Academy of Sciences (Institute of Optoelectronics Technology, Chinese Academy of Sciences), 2019.] is a study on the extraction technology of faint targets under dense star background. The multi-frame mask algorithm based on star registration filters out the star background, and combines the target morphological characteristics with the velocity characteristics to perform multi-frame track association. However, due to the variability of dark stars and the single velocity parameter, dark star interference and blind detection problems arise.

[0005] Reference 2 [Zou Yunlong. Research on Automatic Search and Recognition Technology for High-Orbit Space Targets [D]. University of Chinese Academy of Sciences (Changchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences), 2021. DOI: 10.27522 / d.cnki.gkcgs.2021.000016.] conducts research on automatic search and recognition technology for high-orbit space targets, analyzes the motion characteristics of targets and backgrounds from the time domain and spatial domain perspectives, and proposes a spatial target stripe inverse process detection technology based on motion parameter estimation. However, due to the low signal-to-noise ratio of targets, inaccurate extraction of pulse peaks and errors in the estimation of the morphology of space targets occur, resulting in target missed detection and false alarm problems. Summary of the Invention

[0006] In order to overcome the deficiencies of the above-mentioned prior art, the purpose of the present invention is to provide a method, system, device and medium for detecting and predicting the track of faint space targets under a dense stellar background. By registering bright star sequences and performing spectral phase shift difference, effective suppression of bright and dark stars is achieved. Subsequently, multi-pulse displacement screening and reconstruction are used to estimate the motion parameters of the dark star background and space targets. Then, an image difference fusion rough processing module is designed to obtain candidate targets. Finally, effective detection and track association of faint targets within a small range are achieved through directional track association; the present invention realizes effective detection of faint space targets under a dense stellar background and accurate prediction of target tracks, and has the effects of precise suppression of multi-brightness stars, accurate estimation of different motion component parameters, effective reduction of the target track range, high detection rate of faint space targets, and low false alarm rate.

[0007] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0008] A method for detecting and predicting the track of faint space targets under a dense stellar background, comprising the following steps:

[0009] Step 1, register bright stars between adjacent frames, construct a mask template for difference, and obtain an image sequence P after eliminating bright stars i (i = 1,..., N);

[0010] Step 2, perform adaptive threshold enhancement and connected domain denoising post-processing on the image sequence P after eliminating bright stars in Step 1 i (i = 1,..., N) to obtain a post-processed image sequence B i (i = 1,..., N);

[0011] Step 3, perform spectral phase shift screening and reconstruction on the post-processed image sequence B in Step 2 i (i = 1,..., N), and use a multi-pulse function to estimate the motion parameters of the dark star background and space targets;

[0012] Step 4: Based on the dark star background and the motion parameters of the space target obtained in Step 3, perform sequential image directional displacement and differential fusion on the image sequence B i (i = 1, …, N) to obtain a candidate target image sequence M f ;

[0013] Step 5: Based on the dark star background and the motion parameters of the space target estimated in Step 3, perform directional track association on the candidate target image sequence M f obtained in Step 4 to remove false alarms and detect the target.

[0014] The specific method of the above Step 1 is as follows:

[0015] Perform bright star registration between adjacent frames to obtain an image sequence I i (i = 1, …, N). Process the i-th frame image and the (i + 1)-th frame image in sequence. Starting from i = 1, N is the total number of images. Align adjacent frames and take the largest common area as the mask template. Select a rectangular structural element for mask dilation. Completely filter out the bright stars on the two sub-images through the dilated mask template. Finally, map the processed sequence of sub-images back to the original coordinates to obtain an image sequence P i (i = 1, …, N) after removing bright stars.

[0016] The specific method of the above Step 2 is as follows:

[0017] Enhance dim targets adaptively by thresholding, calculate the mean and variance of the image sequence P i (i = 1, …, N) obtained in Step 1, perform binarization with T as the threshold to detect space targets and dark stars, and obtain and set the connected domain area threshold A T and the aspect ratio threshold C T , eliminate point noise and diffusion phenomena to obtain a post-processed image sequence B i (i = 1, …, N).

[0018] The specific method of the above Step 3 is as follows:

[0019] In the post-processed image sequence B i (i = 1, …, N) obtained in Step 2, there are displacements of both space targets and dark star backgrounds, which can be expressed as:

[0020]

[0021] where B(x, y) and B'(x, y) represent two adjacent frames of images, B0(x, y) represents the dark star background, and B1(x, y) represents the space target. Perform Fourier transforms on two adjacent frames of images to obtain functions F1(u, v) and F2(u, v):

[0022] F1(u, v) = F{B(x, y)} (2)

[0023] F2(u, v) = F{B'(x, y)} (3)

[0024] Where x and y are spatial coordinates, u and v are frequency domain coordinates, and F{*} represents the Fourier transform; according to the phase shift property of the Fourier transform, the conjugate multiplication of functions F1 and F2 is calculated to eliminate the amplitude spectrum information in the frequency domain and retain the phase spectrum information, and the phase difference R(u, v) between two frames of images is recorded:

[0025]

[0026] Where is the conjugate complex number of F2(u, v); take the unit amplitude spectrum of the phase difference R(u, v) to form a new frequency spectrum, and make its maximum absolute value 1:

[0027]

[0028] Where max u,v |R(u, v)| represents the maximum value of the absolute values of all elements in R(u, v), and R norm (u, v) represents the normalized phase difference; then R norm (u, v) is transformed back to the spatial domain through the inverse Fourier transform to obtain the multi-pulse function r(x, y):

[0029] r(x, y) = F -1 {R norm (u, v)} (6)

[0030] Where F -1 {*} represents the inverse Fourier transform. The obtained multi-pulse function r(x, y) contains one or more peaks, and the positions of the peaks are related to the displacements of different motion components; extracting several higher peaks in the multi-pulse function r(x, y) can obtain the displacement information of the dark star background and the space target;

[0031] Select the coordinates (x1, y1) of the highest pulse peak i (i = 1, …, N) to screen the displacements of the dark star background on the x-axis and y-axis in the image sequence, and select (x2, y2) i , (x3, y3) i (i = 1, …, N) to screen the displacements of the space target on the x-axis and y-axis in the image sequence; according to the distance threshold t d The displacement list P1 = {p1, p2, …, p i} composed of (x1, y1) n , (x2, y2) i , (x3, y3)i The list of spatial target displacements P23 = {q1, q2, …, q n} is grouped for similarity:

[0032] p i =(x1, y1) i (i = 1, …, N) (7)

[0033] q i =(x 2,3 , y 2,3 ) i (i = 1, …, N) (8)

[0034]

[0035] where |p i+1 - p i | means that the displacements of two points in the dark star background displacement list P1 on the x - axis and y - axis do not exceed the distance threshold t d , |q i+1 - q i | means that the displacements of two points in the spatial target displacement list P23 on the x - axis and y - axis do not exceed the distance threshold t d , the points in the dark star background displacement list P1 that belong to within the distance threshold t d are classified into the same set G i , otherwise they are placed separately in set G j ; the points in the spatial target displacement list P23 that belong to within the distance threshold t d are classified into the same set H i , otherwise they are placed separately in set H j , i = {1, 2, …, S}, j = {1, 2, …, A} both represent set subscripts, and S, A represent the number of sets;

[0036] Retain the subsets with the largest number of points in sets G and H, denoted as G max and H max , and calculate the mean of all points in the sets:

[0037] G max = max{G} (13)

[0038] H max = max{H} (14)

[0039]

[0040] where k and l both represent the number of points, represents the average displacement of the dark star background on the x - axis and y - axis, Indicates the average displacement of the space target on the x-axis and y-axis; therefore, the motion parameters of the dark star background and the space target are as follows:

[0041]

[0042] Among them, Deg b and Dis b respectively represent the motion direction and motion distance of the dark star background, and Deg t and Dis t respectively represent the motion direction and motion distance of the space target.

[0043] The specific method of step 4 is as follows:

[0044] Design an image difference fusion rough processing module to obtain candidate targets; according to the motion parameters of the dark star background and the space target obtained in step 3, adjust the dark star background displacement list P1 = {p1, p2,..., p n} and the space target displacement list P23 = {q1, q2,..., q n} to obtain a reconstructed displacement list. Among them, each point in P1 and P23 is respectively:

[0045]

[0046] pf = {p i} (i = 1, 2..., N) (22)

[0047]

[0048] qf = {q i} (i = 1, 2..., N) (24)

[0049] Among them, pf and qf respectively represent the reconstructed displacement lists of the dark star background and the space target. According to the reconstructed displacement list, perform adjacent image displacement fusion rough processing on the image sequence B i (i = 1,…, N) to obtain a candidate target image sequence M f :

[0050]

[0051] Among them, M f represents the candidate target image sequence, and M B is the sequence difference image after eliminating the dark star. (B) pf represents that the image sequence B performs pixel displacement according to the reconstructed dark star background displacement list pf. (B) qf Similarly, D1 and D2 are rectangular structuring elements of different sizes. Represents the dilation operation, - represents the image matrix difference operation, represents the image matrix dot product operation.

[0052] The specific method of step 5 is as follows:

[0053] According to the motion direction Deg of the spatial target estimated in step 3 t and the motion distance Dis t , for the candidate target image sequence M obtained in step 4 f perform target track prediction and false alarm removal, set the direction threshold T a and the distance threshold T d , using the candidate targets in the candidate target image sequence M f as track points, with the direction and distance thresholds as the judgment basis, detect and eliminate the track points in the sequence difference image M after removing dark stars where the motion direction and motion distance of the corresponding candidate targets are both within the direction threshold T B and the distance threshold T a , record the tracks, retain the real targets on the tracks, filter out the remaining false alarms, and finally obtain the image sequence M containing the targets d and the multi-frame track record file. w and the multi-frame track record file.

[0054] The present invention also provides a dim space target detection and track prediction system under a dense star background, including:

[0055] A bright star elimination module, used to achieve bright star registration between adjacent frames, construct a mask template for difference, and obtain the image sequence P after eliminating bright stars i (i = 1,..., N);

[0056] The image sequence P i (i = 1,..., N) post-processing module, used to perform adaptive threshold enhancement and connected component denoising post-processing on the image sequence P after eliminating bright stars i (i = 1,..., N) to obtain the post-processed image sequence B i (i = 1,..., N);

[0057] A dark star background and spatial target motion parameter estimation module, used to perform spectral phase shift screening and reconstruction on the post-processed image sequence B i (i = 1,..., N), and estimate the motion parameters of the dark star background and spatial targets using the multi-pulse function;

[0058] The candidate target image sequence M f acquisition module, used to obtain the candidate target image sequence M according to the motion parameters of the dark star background and spatial targets, for the post-processed image sequence B i(i = 1, …, N) perform sequence image directional displacement and differential fusion to obtain a candidate target image sequence M f ;

[0059] A target detection module, which is used to perform directional track association on the candidate target image sequence M according to the estimated dark star background and the motion parameters of the space target f to remove false alarms and detect the target.

[0060] The present invention also provides a dim space target detection and track prediction device under a dense star background, including:

[0061] A memory: storing a computer program of the above-mentioned dim space target detection and track prediction method under a dense star background, which is a computer-readable device;

[0062] A processor: used to implement the above-mentioned dim space target detection and track prediction method when executing the computer program.

[0063] The present invention also provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, it can implement the above-mentioned dim space target detection and track prediction method.

[0064] Compared with the prior art, the beneficial effects of the present invention are:

[0065] 1. Effective filtering of stars: The present invention performs multi-stage filtering based on the brightness of stars. By performing bright star registration and mask difference between adjacent frames in step 1, bright stars are effectively eliminated; according to step 3, the motion parameters of the dark star background are estimated, and then in step 4, an image differential fusion module is designed to remove the interference of dark stars, fully ensuring the precise suppression of stars with different brightnesses.

[0066] 2. Good detection effect for dim targets: After removing bright stars by performing bright star registration and mask difference between adjacent frames in step 1 of the present invention, step 2 is followed by adaptive threshold enhancement and connected component denoising post-processing. The order of first removing bright stars and then enhancing the threshold not only simplifies the background complexity, but also further improves the signal-to-noise ratio of space dim targets, and has stronger robustness for target detection with different magnitudes.

[0067] 3. Detection of multiple space targets: The present invention performs spectrum phase shift screening and reconstruction in step 3, converts the spectrum phase difference into a reconstructed multi-pulse function, can estimate the motion parameters of the dark star background and space targets, and then realizes the detection of multiple space targets.

[0068] 4. Directional track association: The present invention uses the motion parameters of the space target estimated in step 3 and the candidate target image sequence M obtained in step 4 f, in step 5, directional track association is performed. The motion parameters of the space target are used to accurately guide the directional track association process, narrow the association range to a fan-shaped area, accurately detect the target, and reduce false alarms.

[0069] In summary, through technologies such as inter-frame bright star registration, adaptive threshold enhancement and connected component denoising post-processing, spectral phase shift screening and reconstruction, image differential fusion rough processing, and directional track association, the present invention removes redundant false alarms while detecting the true target track, and has the advantages of accurate suppression of multi-stage stars, robust detection of space targets of different magnitudes, accurate estimation of various motion component parameters, effective reduction of the track range, high detection efficiency of faint targets, and low false alarm rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 is a flowchart of the method for detecting faint space targets and predicting tracks under a dense star background according to the present invention.

[0071] Figure 2 is a differential result diagram of inter-frame bright star registration of the present invention.

[0072] Figure 3 is an adaptive threshold target enhancement diagram of the present invention.

[0073] Figure 4 is a schematic diagram of the phase difference spectrum of the present invention.

[0074] Figure 5 is a result diagram of the image differential fusion rough processing module of the present invention.

[0075] Figure 6 is a schematic diagram of the directional track association of the present invention.

[0076] Figure 7 is a detection result diagram of the sequence images of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0077] The flow of the present invention is as Figure 1 shown. The method first uses inter-frame bright star registration of sequences to simplify the background; then uses adaptive threshold to enhance faint targets and sets the connected component threshold to remove diffuse noise; then designs spectral phase shift screening and reconstruction, uses the inter-frame phase difference to obtain a multi-pulse function, and estimates the motion parameters of the faint star background and space targets; furthermore, designs image differential fusion to obtain candidate targets; finally, with the candidate target as the center and the motion parameters of the space target as the threshold, designs directional track association, greatly narrowing the target prediction range, and removing redundant false alarms while detecting the true target. The following will explain the specific implementation process of the technical solution of the present invention in conjunction with the drawings.

[0078] As Figure 1As shown in the figure, in step 1, bright star registration is performed between adjacent frames, a mask template is constructed for differentiation, and the image sequence P after removing bright stars is obtained i (i = 1, …, N);

[0079] Image sequence I i (i = 1, …, N) is sensitive to gray-scale changes in inter-frame registration. When the light changes strongly, global star registration is prone to incorrect matching. In order to retain faint targets while improving the star suppression effect, the present invention adopts a strategy of removing bright stars in different brightness levels in stages

[0080] First, bright star registration is performed between adjacent frames to obtain the registered image sequence R i (i = 1, …, N), and the common area of adjacent frames is selected to form a mask template sequence M that contains all bright stars i (i = 1, …, N - 1). Then, the bright stars in the common area are removed by using template image differentiation operation to simplify the star map background. Finally, the image sequence after star removal is mapped back to the original position; in order to eliminate the bright star residue after differentiation, a 5×5 structural element d0 is selected to dilate the mask, and the image sequence after removing bright stars is denoted as P i (i = 1, …, N), as Figure 2 shown. The process is as follows

[0081]

[0082] Among them, represents matrix dot multiplication operation, I i+1 →I i represents the registration of I i+1 to I i , is the dilation operation, represents mapping the processed image back to I i+1 original position

[0083] In step 2, the image sequence P i (i = 1, …, N) after removing bright stars in step 1 is subjected to adaptive threshold enhancement and connected component denoising post-processing to obtain the post-processed image sequence B i (i = 1, …, N);

[0084] In order to enhance the faint targets in the image, the present invention performs post-processing on the image after removing bright stars. First, the mean and variance of the image sequence P i (i = 1, …, N) are calculated, and an adaptive threshold T = μ + 2.3σ is set to perform binarization operation on the image to enhance the signal-to-noise ratio of faint targets. Then, according to the connected component, an area threshold A T = 8 and an aspect ratio threshold C T= 0.35. If the area of the connected component is less than 8 or the aspect ratio threshold is less than 0.35, then remove the connected component to eliminate the diffusion phenomenon and point noise, and finally obtain the post-processed image sequence B i (i = 1, …, N), as Figure 3 shown:

[0085] T = μ + 2.3σ (29)

[0086]

[0087] where μ and σ respectively represent the mean and variance of the image, R i (x, y) represents a certain connected component, i is the pixel within the connected component, and w and h are the width and height of the circumscribed rectangle of the connected component.

[0088] Step 3, perform spectral phase shift screening and reconstruction on the post-processed image sequence B i (i = 1, …, N), and use the multi-pulse function to estimate the motion parameters of the dark star background and the space target;

[0089] To effectively distinguish the space target from the dark star background, the present invention processes using time-domain information, that is, the motion directions and speeds of the space target and the dark star background are very different. First, use Fourier transform to the frequency domain, extract the image phase spectrum information, and calculate the phase difference R(u, v) of the sequence images. The phase difference R(u, v) includes the displacement information of each motion component in the image. To facilitate comparing the phase spectra and extracting the displacement information of the space target and the dark star background, take the unit amplitude spectrum (normalized) R norm (u, v) of the phase difference R(u, v), and then perform inverse Fourier transform to obtain the multi-pulse function r(x, y). Each motion component is transformed into multiple pulse peaks in the image, as Figure 4 shown:

[0090]

[0091] r(x, y) = F -1 {R norm (u, v)} (34)

[0092] where the function F1(u, v) and as shown in formula (4), R(u, v) represents the inter-frame phase difference, max u,v |R(u, v)| represents the maximum value of the absolute values of all elements in R(u, v), R norm (u, v) represents the normalized phase difference, F -1{*} represents the inverse Fourier transform, and r(x, y) represents the multi-pulse function. The higher peaks in r(x, y) represent the displacement information of different motion components. The first three pulse peaks (x1, y1) with the highest peaks in r(x, y) i (x2, y2) i , (x3, y3) i (i = 1, …, N) constitute the dark star background displacement list P1 and the space target displacement list P23:

[0093] P1 = {(x1, y1)1, (x1, y1)2, ..., (x1, y1) n} (35)

[0094]

[0095] Subsequently, screen and reconstruct P1 and P23, and filter out the outlier displacement points. First, perform similarity classification. Classify the points within the distance threshold t d = 15 into the same set (most displacements in the dark star background or space target displacement list are similar, and a few interfering displacements are caused by noise), and select the set with the largest number of points from the similarity classification set and denote it as G max and H max . This represents the correct displacement sequences of the dark star background and the space target after filtering out the outliers on the x and y axes. From this, the average displacement can be estimated and the motion parameters can be calculated:

[0096]

[0097] Among them, represents the average displacement of the dark star background on the x-axis and y-axis, represents the average displacement of the space target on the x-axis and y-axis, Deg b and Dis b respectively represent the motion direction and motion distance of the dark star background, Deg t and Dis t respectively represent the motion direction and motion distance of the space target. Thus, the motion parameter information of the target and the dark star is obtained.

[0098] Step 4, according to the motion parameters of the dark star background and the space target obtained in Step 3, perform sequence image directional displacement and differential fusion on the image sequence B i (i = 1, …, N) after post-processing in Step 2 to obtain the candidate target image sequence M f ;

[0099] In order to distinguish the space target from the dark star background, the present invention designs an image differential fusion rough processing module according to the dark star background obtained in step 3 and the motion parameters of the space target to remove the dark star background and obtain candidate targets. First, according to the average motion parameters and adjust the dark star background displacement list P1 and the space target displacement list P23 (see formulas 21 - 24) to obtain the reconstructed displacement lists pf and qf, and perform image directional displacement and differential fusion operations between adjacent frames of the sequence image B according to pf and qf:

[0100]

[0101] where D1 is a 5×5 rectangular structuring element, D2 is a 15×15 rectangular structuring element, the image sequence B moves pixels according to the dark star background displacement list pf, and then performs a dilation operation, and then makes a difference with the subsequent frame image, which can not only ensure that the target is located at the original position but also eliminate the dark star residue. The displacement of the space target is different from that of the dark star background. After the previous frame is displaced according to the dark star background displacement list pf, it will not overlap with the target in the subsequent frame. Thus, the sequence difference image M after removing the dark star is obtained B ; at the same time, the image sequence B moves pixels according to the space target displacement list qf, performs a dilation operation to ensure intersection with the target in the subsequent frame image, obtains candidate targets, and then makes a dot product intersection with the sequence difference image M after removing the dark star B to further remove the dark star, and finally obtain the candidate target image sequence M f , as shown in Figure 5 . The circle represents the real target, and the triangle represents the false alarm.

[0102] Step 5, according to the dark star background estimated in step 3 and the motion parameters of the space target, perform directional track association on the candidate target image sequence M obtained in step 4 f to remove false alarms and detect the target.

[0103] The sequence image after rough processing contains targets and a small number of false alarms. The displacement of the target on the sequence frames has regular continuity, generally linear motion, while the noise has great randomness, and the appearance position and frame number are uncertain. Therefore, the present invention designs a directional track prediction according to the estimated target motion parameters. As shown in Figure 6 , the motion direction Deg t and the motion distance Dis t of the space target are used as thresholds to narrow the track range in the prediction frame to the fan-shaped association area, which can remove noise false alarms while accurately detecting the target track. First, obtain all candidate target connected regions in the candidate target image sequence M f and save the centroids:

[0104]

[0105] Among them, a, b, c,... represent all connected regions in the image, (x a , y a ), (x b , y b ), (x c , y c ),... represent the centroids of the corresponding connected regions; is the next frame image of d , y d ), (x e , y e ), (x f , y f ),... represent the centroids of the corresponding connected regions, and i = (1, 2,..., N) is the number of frames of the sequence image. Combine the connected regions in adjacent frames pairwise, and judge whether a track is formed according to the moving distance Dis t and the moving direction Deg t , that is, whether it conforms to the target motion state. Assume taking ad as an example:

[0106] T d = 30, T a = 20° (46)

[0107]

[0108] Among them, T d and T a are the distance threshold and the direction threshold respectively. A track is generated between two points within the threshold range, otherwise it is not a track. Finally, check whether all the generated tracks are real tracks:

[0109]

[0110] When the number of connected regions on the track is greater than or equal to 5, that is, when the target appears within the threshold range of greater than or equal to five frames, it is determined as a real track, otherwise it is a false track. Finally, retain the connected regions on the real track in the corresponding image frames, filter out the remaining connected regions, detect the real target, and obtain the detected result sequence image M w (as shown in Figure 7 ) and the multi-frame track record file, as shown in Table 1 below.

[0111] Table 1 is the multi-frame target track result

[0112] x_coord y_coord frame_number 194 1143 19 250 1135 20 304 1127 21 356 1120 22 413 1112 23 467 1103 24 519 1095 25 572 1087 26 626 1078 27 676 1069 28 730 1060 29 783 1051 30 834 1041 31 881 1032 32 932 1021 33 983 1012 34 1033 1001 35 1083 990 36 1132 979 37 1182 969 38

[0113] In the context of a dense stellar background, traditional multi-frame detection algorithms still suffer from problems such as a high false alarm rate and a low detection rate for weak targets. The present invention makes full use of the correlation information of space targets in the spatial and temporal domains to perform multi-stage stellar filtering, faint target detection, and directional trajectory extraction. Due to severe stellar interference, performing image preprocessing in the initial stage according to the traditional sequence will result in a complex background and missed target detection. Therefore, the present invention first performs bright star filtering and then performs image post-processing, which can simplify the star map background while enhancing the signal-to-noise ratio of faint targets and improving the target detection rate. To effectively detect real trajectories, the present invention first uses the spectral phase difference to obtain a reconstructed multi-pulse function, and then estimates the motion parameters of the dark star background and space targets. Then, an image differential fusion rough processing module is designed to obtain a candidate target image sequence, effectively reducing the number of trajectories. Finally, based on the motion parameters and the candidate target image sequence, directional trajectory association is performed, and the association area is reduced to a fan-shaped range, accurately and quickly detecting the target trajectory while removing redundant false alarms. Based on the distribution characteristics and temporal motion characteristics of space targets, the present invention not only achieves precise suppression of stars with different brightnesses but also improves the detection robustness of targets with different magnitudes, and has a good detection effect on faint space targets with different motion parameters in a dense stellar background.

[0114] The present invention also provides a faint space target detection and trajectory prediction system under a dense stellar background, including:

[0115] A bright star elimination module for realizing bright star registration between adjacent frames in step 1, constructing a mask template for difference, and obtaining an image sequence P after eliminating bright stars i (i = 1,..., N);

[0116] The image sequence P i (i = 1,..., N) post-processing module for realizing the adaptive threshold enhancement and connected component denoising post-processing of the image sequence P i (i = 1,..., N) after eliminating bright stars in step 1, and obtaining a post-processed image sequence B i (i = 1,..., N);

[0117] A motion parameter estimation module for the dark star background and space targets for realizing the spectral phase shift screening and reconstruction of the post-processed image sequence B i (i = 1,..., N) in step 3, and estimating the motion parameters of the dark star background and space targets using the multi-pulse function;

[0118] The candidate target image sequence M f An acquisition module for realizing step 4 to obtain the motion parameters of the dark star background and space targets according to step 3, and for the post-processed image sequence B i(i = 1, …, N) perform sequence image directional displacement and differential fusion to obtain a candidate target image sequence M f ;

[0119] A target detection module, which is used to implement, according to the dark star background estimated in step 3 and the motion parameters of the space target in step 5, the directional track association for the candidate target image sequence M obtained in step 4 f to remove false alarms and detect the target.

[0120] The present invention also provides a device for detecting a dim space target and predicting its track in a dense star background, including:

[0121] A memory: storing a computer program for the method for detecting a dim space target and predicting its track in a dense star background as described above, which is a computer-readable device;

[0122] A processor: used to implement the method for detecting a dim space target and predicting its track in a dense star background when executing the computer program.

[0123] The present invention also provides a computer-readable storage medium, which stores a computer program that can implement the method for detecting a dim space target and predicting its track in a dense star background when being executed by a processor.

Claims

1. A method for detecting faint space targets and predicting their trajectories under a dense stellar background, characterized in that, Including the following steps: Step 1, perform bright star registration between adjacent frames, construct a mask template for difference, and obtain an image sequence P after eliminating bright stars i (i = 1, …, N); Step 2: For the image sequence P after eliminating bright stars in Step 1 i (i = 1, …, N), perform adaptive threshold enhancement and connected component denoising post-processing to obtain the post-processed image sequence B i (i = 1, …, N); Step 3: Perform spectral phase-shift screening and reconstruction on the image sequence B after post-processing in Step 2 i (i = 1, …, N) to estimate the motion parameters of the dark star background and spatial targets using a multi-pulse function; Step 4: Based on the dark star background and the motion parameters of the space target obtained in Step 3, perform sequential image directional displacement and differential fusion on the post-processed image sequence B i (i = 1, …, N) to obtain a candidate target image sequence M f ; Step 5: Based on the dark star background estimated in Step 3 and the motion parameters of the space target, perform directional track association on the candidate target image sequence M obtained in Step 4 f to remove false alarms and detect the target.

2. The method for detecting and predicting the track of a dim space target under a dense star background according to claim 1, characterized in that The specific method of step 1 is as follows: Registration of bright stars between adjacent frames to obtain the image sequence I i (i = 1, …, N), process the i-th frame image and the (i + 1)-th frame image in sequence. i starts from 1 and N is the total number of images. Align adjacent frames and take the largest common area as the mask template. Select a rectangular structuring element for mask dilation. Completely filter out the bright stars on the two sub-images through the dilated mask template. Finally, map the processed sequence of sub-images back to the original coordinates to obtain the image sequence P after eliminating the bright stars i (i = 1, …, N).

3. A method for detecting faint space targets and predicting their trajectories in a dense stellar background according to claim 1, characterized in that, The specific method of step 2 is as follows: Adaptive threshold to enhance dim targets, calculate the mean and variance of the image sequence P obtained in calculation step 1 i (i = 1, …, N), perform binarization with T as the threshold, detect space targets and faint stars, obtain and set the connected domain area threshold A T and the aspect ratio threshold C T , eliminate dot noise and diffusion phenomenon, and obtain the post-processed image sequence B i (i = 1, …, N).

4. A method for detecting and predicting the trajectory of faint space targets in a dense stellar background according to claim 1, characterized in that The specific method of step 3 is as follows: Step 2: Post-processed image sequence B i (where \(i = 1,\ldots,N\)) There are both spatial targets and displacement of the dark star background, which is expressed as: Among them, B(x,y) and B'(x,y) represent two adjacent frames of images, B0(x,y) represents the dark star background, and B1(x,y) represents the space target. Fourier transforms are performed on the two adjacent frames of images to obtain functions F1(u,v) and F2(u,v): F1(u,v) = F{B(x,y)} (2) F2(u,v) = F{B'(x,y)} (3) Where x and y are spatial coordinates, u and v are frequency domain coordinates, and F{*} represents the Fourier transform; according to the phase shift property of the Fourier transform, the conjugate multiplication of functions F1 and F2 is calculated to eliminate the amplitude spectrum information in the frequency domain and retain the phase spectrum information, and the phase difference R(u,v) between the two frames of images is recorded: wherein, is the conjugate complex number of F2(u, v); taking the unit amplitude spectrum of the phase difference R(u, v) to form a new frequency spectrum such that its maximum absolute value is 1: Among them, max u,v |R(u, v)| represents the maximum value of the absolute values of all elements in R(u, v), and R norm (u, v) represents the normalized phase difference; then, the inverse Fourier transform is used to transform R norm (u, v) back to the spatial domain to obtain the multi-pulse function r(x, y): r(x,y) = F -1 {R norm (u,v)} (6) Among them, F -1 {*} represents the inverse Fourier transform. The obtained multi-pulse function r(x, y) contains one or more peaks, and the positions of the peaks are related to the displacements of different motion components. By extracting several relatively high peaks in the multi-pulse function r(x, y), the displacement information of the dark star background and the space target can be obtained. Select the highest pulse peak coordinates (x1, y1) i (i = 1, …, N) Screen the displacements of the dark star background in the x-axis and y-axis in the image sequence, and select (x2, y2) i , (x3, y3) i (i = 1, …, N) Screen the displacements of the space target in the x-axis and y-axis in the image sequence; according to the distance threshold t d Take (x1, y1) i to form the dark star background displacement list P1 = {p1, p2, ..., p n}, (x2, y2) i , (x3, y3) i to form the space target displacement list P23 = {q1, q2, ..., q n} for similarity grouping: p i =(x1, y1) i (i = 1, …, N) (7) q i = (x 2,3 , y 2,3 ) i (i = 1, …, N) (8) where, |p i+1 -p i | means that the displacements between two points on the x - axis and y - axis in the dark star background displacement list P1 do not exceed the distance threshold t d and, |q i+1 -q i | means that the displacements between two points on the x - axis and y - axis in the space target displacement list P23 do not exceed the distance threshold t d Points within the distance threshold t d in the dark star background displacement list P1 are classified into the same set G i otherwise, they are placed separately in the set G j Points within the distance threshold t d in the space target displacement list P23 are classified into the same set H i otherwise, they are placed separately in the set H j , i = {1, 2,..., S}, j = {1, 2,..., A} both represent set subscripts, and S and A represent the number of sets; Retain the subset with the largest number of points in sets G and H, denoted as G max and H max , and calculate the mean value of all points in the set: G max = max{G} (13) H max = max{H} (14) where k and l both represent the number of points, represents the average displacement of the dark star background along the x-axis and y-axis, represents the average displacement of the space target along the x-axis and y-axis; therefore, the motion parameters of the dark star background and the space target are as follows: Among them, Deg b and Dis b respectively represent the motion direction and motion distance of the dark star background, and Deg t and Dis t respectively represent the motion direction and motion distance of the space target.

5. A method for detecting and predicting the trajectory of a faint space target against a dense star background according to claim 1, characterized in that The specific method of step 4 is as follows: Design an image differential fusion rough processing module to obtain candidate targets; according to the dark star background and the motion parameters of the space target obtained in step 3, adjust the dark star background displacement list P1 = {p1, p2, …, p n}, and the space target displacement list P23 = {q1, q2, …, q n}, and obtain the reconstructed displacement list, where each point in P1 and P23 is respectively: pf = {p i}(i = 1, 2..., N) (22) qf = {q i}(i = 1, 2..., N) (24) wherein, pf and qf respectively represent the displacement lists of the reconstructed dark star background and the space target, and the post-processed image sequence B after step 2 is subjected to rough processing of adjacent image displacement fusion according to the reconstructed displacement lists i (i = 1, …, N) to obtain a candidate target image sequence M f : Among them, M f represents the candidate target image sequence, and M B is the sequence difference image after eliminating dark stars. (B) pf represents that the image sequence B is pixel-displaced according to the reconstructed dark star background displacement list pf. (B) qf Similarly, D1 and D2 are rectangular structuring elements of different sizes, representing the dilation operation, - represents the image matrix difference operation, represents the image matrix dot product operation.

6. A method for detecting and predicting the trajectory of a faint space target against a dense stellar background according to claim 1, characterized in that, The specific method of step 5 is as follows: The movement direction Deg of the space target estimated according to Step 3 t and the movement distance Dis t , perform target track prediction and false alarm removal on the candidate target image sequence M f obtained in Step 4, set the direction threshold T a and the distance threshold T d , take the candidate targets in the candidate target image sequence M f as track points, use the direction and distance thresholds as the judgment basis, detect and eliminate the track points in the sequence difference image M B corresponding to the candidate targets whose movement direction and movement distance are both within the direction threshold T a and the distance threshold T d , record the track, retain the real targets on the track, filter out the remaining false alarms, and finally obtain the image sequence M w containing the targets and the multi-frame track record file.

7. A dim space target detection and track prediction system based on the method according to any one of claims 1 to 6, characterized in that Including: A bright star elimination module, which is used to implement bright star registration between adjacent frames, construct a mask template for differentiation, and obtain an image sequence P after bright star elimination i (i = 1, …, N); Image sequence P i (i = 1, …, N) post - processing module for the image sequence P after bright star elimination i (i = 1, …, N) to perform adaptive threshold enhancement and connected component denoising post - processing to obtain the post - processed image sequence B i (i = 1, …, N); The dark star background and space target motion parameter estimation module is used for the image sequence B after post-processing i (i = 1, …, N) for spectral phase shift screening and reconstruction, and uses multi-pulse functions to estimate the motion parameters of the dark star background and space targets; Candidate target image sequence M f An acquisition module, configured to perform sequence image directional displacement and differential fusion on the post-processed image sequence B i (i = 1, …, N) to obtain the candidate target image sequence M f ; A target detection module, which is used to perform directional track association on the candidate target image sequence M according to the estimated dark star background and the motion parameters of the space target, remove false alarms, and detect the target. f Perform directional track association, remove false alarms, and detect the target.

8. An apparatus for detecting faint space targets and predicting their trajectories against a dense stellar background, characterized in that, Including: A memory: storing a computer program of a method for detecting and predicting the trajectory of a dim space target under a dense star background according to any one of claims 1-6, which is a computer-readable device; A processor: used to implement a method for detecting and predicting the trajectory of a dim space target under a dense star background according to any one of claims 1-6 when executing the computer program.

9. A computer-readable storage medium, characterized in that, A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it can implement a method for detecting and predicting the trajectory of a dim space target under a dense star background according to any one of claims 1-6.

Images