Moving target-oriented low-light super-resolution imaging method

Abstract

A low-light super-resolution imaging method for a moving target comprises the following steps: carrying out continuous frame imaging on a ground scene by adopting a space-based remote sensing low-light area-array camera with a micro-scanning super-resolution imaging mode; carrying out pixel merging on the obtained continuous frame images to obtain a down-sampling image with an improved signal-to-noise ratio; performing point target detection, target positioning and motion information extraction on the obtained downsampling image, and calculating to obtain target coordinate information; mapping the target coordinates to an image with an original resolution according to a coordinate corresponding relation, and extracting an image slice containing a moving target according to a set neighborhood value; performing fine matching of a local area by using the obtained image slices of each frame to obtain a sub-pixel displacement matrix; and taking the obtained sub-pixel displacement matrix as prior input, and performing image reconstruction on the multi-frame image slices to obtain a super-resolution imaging result.

Description

Technical Field

[0001] The present invention belongs to the field of optical remote sensing detection technology and relates to a low-light super-resolution imaging method for moving targets. Background Art

[0002] With the particularity of the staring imaging working mode of the low-light area array camera, combined with the multi-frame image fusion enhancement technology, super-resolution imaging of the low-light camera can be achieved. The camera mechanism uses closed-loop control to make the optical image plane perform two-dimensional sub-pixel stepping and translation on the imaging focal plane, collect multiple frames of area array sequence images with sub-pixel dislocations, and then combine on-board registration and super-resolution reconstruction technology to enable the system to break through the sampling frequency of the area array detector and achieve super-resolution imaging of the system. In the process of using the micro-scanning mechanism to perform multi-frame super-resolution reconstruction, since the swing of the micro-scanning mechanism and the low-light integral imaging both require time, if there is a moving target in the picture, for the moving target itself, the sub-pixel displacement between frames does not conform to the micro-scanning displacement value returned by the micro-scanning mechanism and its control loop, but is the superposition result of the displacement of the micro-scanning mechanism and the motion of the moving target itself on the focal plane projection vector. If the moving target itself moves at a low speed (within 0.1 pixel difference between frames), the restoration effect will be basically unaffected. If the moving target itself moves at a high speed, the registration between the two frames will fail in the area where the moving target is located, causing the multi-frame super-resolution method to make errors in the local area of ​​the image, and thus making the camera super-resolution design capability fail when performing imaging and recognition functions for moving targets. Therefore, an adaptive super-resolution imaging method must be proposed for the case where there are moving targets in the scene. Summary of the invention

[0003] The technical problem solved by the present invention is: to overcome the shortcomings of the prior art and propose a low-light-level super-resolution imaging method for moving targets, which solves the problem that the existing low-light-level array super-resolution imaging has moving targets in the frame and cannot effectively reconstruct the moving targets with super-resolution.

[0004] The technical solution of the present invention is: a low-light level super-resolution imaging method for moving targets, comprising:

[0005] Use a space-based remote sensing low-light area array camera with micro-scanning super-resolution imaging mode to perform continuous frame imaging of ground scenes;

[0006] Merge pixels of acquired continuous frame images to obtain downsampled images with improved signal-to-noise ratio;

[0007] Point target detection, target positioning and motion information extraction are performed on the obtained downsampled image, and the target coordinate information is calculated; the target coordinates are mapped to the image of the original resolution according to the coordinate correspondence, and the image slice containing the moving target is extracted according to the set neighborhood value;

[0008] Using the obtained image slices of each frame to perform precise matching of local areas, a sub-pixel displacement matrix is ​​obtained;

[0009] The obtained sub-pixel displacement matrix is ​​used as a priori input to reconstruct multiple image slices to obtain super-resolution imaging results.

[0010] The continuous frame imaging of ground scenes comprises: the space-based remote sensing low-light area array camera with micro-scanning super-resolution imaging mode is mounted on a satellite platform with a step-and-stare function, and performs multi-frame staring low-light imaging of ground targets within a continuous integration time.

[0011] The pixel merging of the acquired continuous frame images to obtain a downsampled image with an improved signal-to-noise ratio includes: performing a pixel merging and downsampling operation on each frame of the acquired low-light area array image containing a moving target according to a rule of weighted averaging of adjacent n×n pixels to obtain a downsampled image with an improved signal-to-noise ratio; wherein n is an integer between 1 and 8 and is proportional to the target size.

[0012] The method performs point target detection, target positioning and motion information extraction on the obtained downsampled image, calculates target coordinate information; maps the target coordinates to the image of the original resolution according to the coordinate correspondence, and extracts image slices containing moving targets according to the set neighborhood value, including: performing moving target detection on the sequence frames of the downsampled image, returning the target coordinate value; mapping the returned coordinate value to the image of the original resolution to obtain the target center coordinate value; setting a slice parameter N to extract image slices containing moving targets according to the N×N pixel neighborhood value with the target center coordinate value as the center, and storing the image slice sequence in a temporary storage; the setting of the slice parameter N is greater than the target envelope size.

[0013] The target coordinate information can also be calculated by accessing the infrared point target detection result data with external time synchronization.

[0014] The infrared point target detection result data accessed with external time synchronization includes: being equipped with another medium / long wave infrared channel imaging camera, whose field of view includes the space-based remote sensing low-light area array camera; detecting, detecting and locating the target in the infrared channel, and converting the coordinates into the image plane coordinate system of the space-based remote sensing low-light area array camera after matching in the time and space domains, to obtain the infrared point target detection result data.

[0015] When using infrared point target detection result data connected to external time synchronization, extracting image slices containing moving targets, including: equipping another medium / long wave infrared channel imaging camera, the field of view of which includes the space-based remote sensing low-light area array camera; detecting, detecting, and positioning the target in the infrared channel, and converting the coordinates to the image plane coordinate system of the space-based remote sensing low-light area array camera after matching in the time and space domains, infrared point target detection result data; mapping the returned coordinate values ​​to the image of the original resolution to obtain the target center coordinate value; setting a slice parameter N to extract image slices containing moving targets according to the N×N pixel neighborhood value with the target center coordinate value as the center, and storing the image slice sequence in a temporary storage; the setting of the slice parameter N is greater than the target envelope size.

[0016] The method of using the obtained image slices of each frame to perform precise matching of the local area to obtain the sub-pixel displacement matrix includes: for each frame F in the image slice i Relative to the reference frame F1, i.e., the first frame in time order, the image matching algorithm is executed respectively, and the sub-pixel displacement matrix {H i}.

[0017] The method uses the obtained sub-pixel displacement matrix as a priori input to reconstruct multiple image slices to obtain super-resolution imaging results, including:

[0018] Each frame image slice and its transformation matrix relative to the reference frame {H i} is used as input and a multi-frame super-resolution reconstruction algorithm is used to calculate the super-resolution imaging result.

[0019] The multi-frame super-resolution reconstruction algorithm specifically uses a convex set projection algorithm.

[0020] The advantages of the present invention compared with the prior art are:

[0021] 1. The present invention can take into account the super-resolution reconstruction of dynamic targets and static targets, and can also achieve parallel processing of multiple targets with the help of multi-target parallel short-term tracking. The modular algorithm has strong plasticity and great upgrade potential;

[0022] 2. Through post-processing, the original frames with low signal-to-noise ratio and resolution are matched and fused to synthesize output images with improved signal-to-noise ratio and resolution. By handing over the workload to post-processing, the complexity and cost of implementing the low-light imaging mode are simplified;

[0023] 3. While completing super-resolution reconstruction, the present invention can also obtain additional information such as target motion vector, thereby improving the accuracy of multi-modal target recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is the flow chart of the moving target super-resolution algorithm.

[0025] Figure 2 Schematic diagram of the forward and backward trajectory optical flow prediction principle.

[0026] Figure 3 Schematic diagram of pyramid-shaped multi-scale space.

[0027] Figure 4 This is the flow chart of the POCS algorithm.

[0028] Figure 5 Schematic diagram of moving target segmentation results.

[0029] Figure 6 Schematic diagram of the before and after comparison of moving target super-resolution. DETAILED DESCRIPTION

[0030] The present invention is oriented to the application needs of space-based nighttime low-light imaging and moving target extraction, and provides a low-light level super-resolution imaging method for moving targets. The camera mechanism uses closed-loop control to make the optical image plane perform two-dimensional sub-pixel-level stepping translation on the imaging focal plane, collect multiple frames of array sequence images with sub-pixel dislocations, and then combine the registration and super-resolution reconstruction technology to enable the system capability to break through the sampling frequency of the array detector and achieve super-resolution imaging of the system. In the process of performing the above-mentioned multi-frame super-resolution reconstruction using the micro-scanning mechanism, the movement and imaging of the micro-scanning mechanism both require time. If the picture contains a moving target of interest, for the moving target itself, the sub-pixel-level displacement between frames does not conform to the micro-scanning displacement value returned by the micro-scanning mechanism and its control loop, but is the superposition result of the displacement of the micro-scanning mechanism and the movement of the moving target itself on the focal plane projection vector.

[0031] Based on the above-mentioned micro-scanning super-resolution imaging technology, the present invention discusses a low-light level super-resolution imaging method for moving targets, which is specifically divided into low-light level moving target detection - moving target image slice extraction - moving target image registration - super-resolution reconstruction, wherein low-light level moving target detection realizes the discovery, detection and motion information acquisition of moving targets through multi-modal means such as low-resolution point target detection or infrared point target detection, realizes the slice extraction of the moving target neighborhood image with the help of real-time motion information, performs sub-pixel level registration on the moving target neighborhood slices and completes the moving target super-resolution reconstruction. This method can effectively and adaptively solve the failure of multi-frame super-resolution processing technology when there are moving targets in the image and the problem of insufficient signal-to-noise ratio under low-light conditions, and significantly improves the image resolution of the moving target area.

[0032] The specific process of the present invention is as follows Figure 1As shown. The array camera with micro-scanning super-resolution imaging mode carries out continuous frame imaging and merges the pixels of the continuous frame images with n×n templates to improve the signal-to-noise ratio performance of target detection, or accesses the multi-modal point target detection data of the external time-synchronized medium-wave infrared; detects, locates, extracts and estimates the motion information of the point target in the continuous frame images, expresses and extracts the vector information in its motion, maps the positioning coordinate information to the image of the original resolution according to the set N×N neighborhood value, and extracts the image slices containing the moving target, and stores the slice sequence in the temporary storage; performs the image precise matching algorithm for each frame slice relative to the reference frame (usually the first frame in time sorting), and calculates the transformation matrix H of each frame relative to the reference frame through matrix operation. The obtained H matrix is ​​used as input to perform super-resolution reconstruction on multiple frame slices, while improving the resolution and signal-to-noise ratio performance within the neighborhood of the moving target.

[0033] The obtained downsampled image is subjected to point target detection, target positioning and motion information extraction, and the target coordinate information is calculated; the target coordinate is mapped to the image of the original resolution according to the coordinate correspondence, and the image slice containing the moving target is extracted according to the set neighborhood value, including: performing moving target detection on the sequence frames of the downsampled image, returning the target coordinate value; mapping the returned coordinate value to the image of the original resolution to obtain the target center coordinate value; setting the slice parameter N to extract the image slice containing the moving target according to the N×N pixel neighborhood value with the target center coordinate value as the center, and storing the image slice sequence in the temporary storage; the setting of the slice parameter N is greater than the target envelope size. In the above steps, it is necessary to detect the moving target in the sequence frames of the stationary track array image, and have the short-term tracking capability. For the description of the motion in the sequence image, the corresponding motion field can be used to express the three-dimensional motion. Optical flow is defined as the instantaneous velocity field reflected by the target point with grayscale when it performs a specific motion in the image, which satisfies the properties of continuity and differentiability in space and time in the image area, and this assumption is usually used as a more important condition for related calculations. There is a sequence of images. Define (x, y) as a specific pixel in the image, and define I(x, y, t) as the specific value of this point at time t. Define u(x, y) and v(x, y) as the specific movement components of the optical flow in the x and y directions, respectively. For time t, it is very small, that is, dt. At this time, for the pixel, it moves from the initial position to the point (x+dx, y+dy), and dx=udt, dy=vdt. Know that there is constant brightness, that is, for pixels that make related movements in a certain trajectory, they have the same grayscale values, which is also the so-called constant brightness. At this time, we have:

[0034] I(x+dx,y+dy,t+dt)=I(x,y,t) (1)

[0035] Now perform Taylor expansion on the above formula

[0036]

[0037] In the formula, As dx, dy, dt are traces and are of second order or greater, they can be ignored. Since dx = udt, dy = vdt, we can get

[0038] I x u+I y v+I t =0 (3)

[0039] In the formula, I x ,I y ,I t Decibel represents the partial derivative of the pixel in different directions, corresponding to x, y, and t. At this time, the above formula is called the optical flow constraint equation. From this formula, it can be seen that the relationship between the optical flow speed and the spatial gradient is expressed in a relevant way. There are two unknowns, namely u and v. Since there is only one corresponding equation, the unique solution of the equation cannot be obtained. In order to solve the equation, other corresponding conditions must be added.

[0040] The Lucas-Kanada LK optical flow method assumes that the motion vector has an invariant property in a small neighborhood in space. In addition, when estimating the optical flow, the method used is the weighted least squares method. The fundamental reason for choosing this specific method is that it has a more convenient input of a set of points in the image. Based on this point, it is widely used in sparse optical flow fields. For this algorithm, it is also based on certain assumptions, specifically:

[0041] 1) Constant brightness. For the target pixel, it moves in different frames and has the property of constant appearance. For the grayscale image, it is assumed that the brightness is constant during the whole tracking process.

[0042] 2) It has temporal continuity. Since the image is constantly moving with time during the entire motion process, but its motion is relatively slow relative to time, in actual situations, the former has a smaller ratio than the temporal change and image motion, which results in very small motion in adjacent frames.

[0043] 3) Spatial consistency: For the same surface in the same scene, the adjacent pixels have similar motions, and the projections of adjacent points are also adjacent.

[0044] For very small areas, corresponding changes can be made when estimating motion-related information:

[0045]

[0046] Where W 2 (X) specifically represents the window weight function. The existence of this function will make the center of the neighborhood have a larger weight ratio than the surrounding area. n It is the n specific points in the Ω set, where V = (u, v) is defined T , ΔI(X)=(I x , I y ) T At this time, the solution to the above equation is performed by the least squares method:

[0047] A T W 2 AV=A T W 2 b (5)

[0048] in,

[0049]

[0050] W = diag(W(X1), ..., W(X N )) (7)

[0051]

[0052] V=(A T W 2 A) -1 A T W 2 b (9)

[0053] The reliability of the estimated value of V is directly affected by A T W 2 The influence of the eigenvalue of A. Assuming that the eigenvalues ​​exist λ1, λ2, λ1≥λ2, τ is used as a hypothetical threshold. When λ1≥λ2≥τ exists, V can be completely obtained. When λ2=0, the broken jade matrix becomes a singular matrix, so its optical flow cannot be solved; when λ1≥τ and λ2<τ, it is impossible to obtain V information, and only the discovery component of the optical flow is obtained. In the TLD algorithm, the forward and backward optical flow method divides the starting tracking frame evenly according to the current frame I t Predict the next frame I t+1 The midpoint position. At the same time, reverse prediction is performed, that is, I t+1 The point position detected in the prediction is It Original position, when the displacement deviation between the reverse predicted feature point and the original feature point exceeds the threshold, it is removed to maintain the relative stability of the short-term tracker. Figure 2 shown.

[0054] If at time k = 1, X t+1 and The two overlap. Then for I t and I t+1 As two consecutive frames, forward prediction can be performed at this time to t Predicted, and obtained The forward and backward optical flow method is applied. Due to the use of relative motion field, it is adaptable in the process of extracting moving targets in sequential frame images with sub-pixel micro-scanning, and has a certain target short-term tracking prediction capability, which can improve the execution efficiency of subsequent algorithms.

[0055] The multi-frame sequence slices are matched, including: according to the relative displacement estimate obtained in the above motion estimation, the pixel-level matching position is roughly located in the fine matching stage, and then the sub-pixel interpolation (Sub-pixel Interpolation) is performed on the adjacent area of ​​the correlation function with the position as the center, so as to reduce the amount of calculation and improve the matching accuracy to the sub-pixel level. After the matching is completed, the matching point pairs are screened according to the epipolar constraint to eliminate the wrong matching points. Combined with the invariance of the SIFT feature point extraction method to rotation, scale scaling, affine transformation, and the stability to noise, viewing angle and illumination changes, a sub-pixel-level inter-frame precise matching method suitable for the research content of the present invention and with excellent performance is proposed.

[0056] The image matching algorithm used in remote sensing imaging needs to be stable to noise, viewing angle and illumination, and invariant to rotation, scale scaling and affine transformation. The spatial feature point extraction and characterization methods that meet the conditions and have excellent performance mainly include SIFT (Scale Invariant Feature Transform) and SURF (Speeded Up Robust Feature). Both SIFT and SURF algorithms are based on Gaussian pyramids to find local extreme points to determine the scale-invariant stable points. In the feature point description stage, the feature point is the center, and the features are statistically analyzed in the surrounding neighborhood. SIFT calculates the histogram of the amplitude of the statistical gradient in the square neighborhood to find the direction corresponding to the maximum amplitude; SURF calculates the haar wavelet response along the two coordinate axes of the image space in each sector in the circular domain area, and finds the sector direction with the largest modulus to characterize. In terms of performance, SURF runs at only one-third of SIFT, but there is a gap in description accuracy compared to SIFT. Taking into account the application background of this method and weighing the resource efficiency allocation, the present invention selects SIFT as the theoretical basis for sub-pixel feature point extraction and characterization, and makes adaptive improvements. Feature point extraction and description are specifically divided into the following five steps:

[0057] 1) Constructing scale space

[0058] Since the Gaussian convolution kernel is the only linear kernel that can achieve scale transformation, the original image is convolved with a two-dimensional scale-variable Gaussian function.

[0059] L(x,y,σ)=G(x,y,σ)*I(x,y) (10)

[0060]

[0061] Where I(x,y) is the original image; (x,y) represents the coordinates in the image space coordinate system, and σ is the scale coordinate.

[0062] The maximum and minimum values ​​of the Laplace function can produce the most stable image features compared with other commonly used feature extraction functions, such as Hessian or Harris corner features. To simplify the calculation, the Differential of Gussian (DOG) operator is usually used to approximate the Laplace function to construct a pyramid-shaped multi-scale space, such as Figure 3 shown.

[0063] D(x,y,σ)=(G(x,y,kσ)-G(x,y,σ))·I(x,y)=L(x,y,kσ)-L(x,y,σ) (12)

[0064] Where k is a constant factor.

[0065] There are O groups of image pyramids, each with S layers, and the images of the next group are obtained by downsampling the images of the previous group; Figure 3 Each layer on the right is formed by subtracting two adjacent layers of Gaussian images.

[0066] 2) Rough positioning of feature points

[0067] Each sampling point is compared with all its adjacent points in the multi-resolution scale space, that is, compared with its 8 adjacent points at the same scale and 9×2 points corresponding to the upper and lower adjacent scales, a total of 26 points, to find the extreme points in the scale space and image space as the image coarse positioning feature points.

[0068] 3) Precise positioning of feature points

[0069] In order to obtain accurate and robust feature points and improve the positioning accuracy to the sub-pixel level, it is necessary to perform binary quadratic fitting on the coarse positioning space and use discrete space point interpolation to obtain continuous space extreme point information. This method is called sub-pixel interpolation. At the same time, the edge points generated by the edge response of the DOG operator are removed.

[0070] 4) Determination of the main direction of feature points

[0071] Calculate the image gradient of the key point, and use the histogram to count the gradient and direction of the pixels in the neighborhood. The gradient histogram divides the direction range of 0° to 360° into 36 columns. The peak direction of the histogram is taken as the main direction of the feature point.

[0072] 5) Feature point description

[0073] For each feature point, there are three descriptive information: position, scale and direction. The purpose of feature point description is to describe the feature point with a unique feature vector. The descriptor uses the gradient information of 8 directions calculated in a 4×4 window in the key point scale space, a total of 4×4×8=128-dimensional vector representation.

[0074] The method of the cosine of the angle between feature point vectors is used to calculate and match the 128-dimensional feature descriptor vector extracted by the SIFT algorithm. First, for each feature point P0 in the reference frame, the inverse cosine function is used to calculate the similarity of all feature points P in the image to be matched. k The angle between the feature descriptor vector and P0

[0075]

[0076] In the formula and P0 and P k Normalized 128-dimensional feature descriptor vector.

[0077] After the calculation is completed, the 128 angles are sorted. If the ratio of the smallest angle to the second smallest angle is less than a certain threshold, P0 and P k is a pair of matching feature points. Regarding the selection of the threshold, 0.4 to 0.6 is generally the best. The smaller the threshold, the stricter the identification of the matching relationship. This step performs precise matching on the areas of different frames of the same target, and the matching results are averaged, with an accuracy of up to 0.1 pixel. The displacement matrix H of the two frames of images can be directly calculated based on the coordinate correspondence results of the image matching point pairs.

[0078] The matched multi-frame sequence slices are reconstructed with super resolution, including: using the target slice super resolution reconstruction based on the convex set projection method. The convex set projection (POCS) algorithm takes into account the nonlinear degradation model. The algorithm flow is as follows: Figure 4 As shown in the figure, a high-resolution image is restored by using multiple low-resolution images. The displacement matrix H between multiple frames is used as a priori, and the rich information contained in the multiple low-resolution images is super-resolution processed to obtain a high-resolution image with a higher resolution than the camera resolution used in the recorded image. Corresponding to the frequency domain algorithm, the POCS algorithm is a spatial domain high-resolution image reconstruction method. It starts with the guessed initial image and corrects the image successively according to the simulation error. The unknown image is assumed to be an element in a suitable Hilbert space. Each prior knowledge or constraint about the unknown image restricts the solution of a closed convex set in the Hilbert space. The amplitude boundary restriction (0 to the maximum grayscale value) is introduced to derive the iterative formula for solving the unknown image. The super-resolution image is iteratively calculated from the initial estimate, and the super-resolution image is restored using set theory methods. It effectively utilizes the spatial range observation model while allowing the inclusion of prior information. The image space solved by the super-resolution image restoration method is intersected with a restricted set of features that describe the ideal super-resolution image, including certainty, bounded energy metrics, data fidelity, smoothness, etc., and then a simplified solution space is formed.

[0079] The POCS algorithm is an iterative process that, given any point in the super-resolution image space, locates a point that satisfies all convex restriction sets.

[0080] The convex set of any observed low-resolution image can be written as:

[0081]

[0082] Where 0≤n1,n2≤N-1,k=1,2,…,L

[0083] In this formula, The value of can be given by:

[0084]

[0085] Where δ0 represents the confidence in the observation and is set equal to cσ ν where σ ν is the standard noise deviation c>0 determined by an appropriate confidence boundary. These parameters define a super-resolution image, and these sets define a super-resolution image related to a low-resolution frame within a certain boundary. This confidence boundary

[0086] The limit is proportional to the noise deviation.

[0087] For any x i (m1,m2) to Projection on: The definition is as follows:

[0088]

[0089] In the above formula equal:

[0090]

[0091] Additional constraints, such as amplitude, can be used to improve the results.

[0092] Based on amplitude limitation P A is defined as:

[0093]

[0094] After the above projection is given, the super-resolution image s i Estimate of (m1,m2) It can be obtained by iteration, and the formula is:

[0095]

[0096] j=0,1,…,0≤m1≤M-1, 0≤m2≤M-1.

[0097] Here T refers to the projection operator. The initial It can be obtained by bilinear interpolation of the low-resolution reference frame. Generally speaking, the more iterations, the better the result. In addition, the more restriction sets are used, the better the effect. Figure 5 and Figure 6 They are schematic diagrams of the moving target slicing results and the image super-resolution reconstruction effects before and after.

Claims

1. A low-light level super-resolution imaging method for moving targets, characterized in that: include: Use a space-based remote sensing low-light area array camera with micro-scanning super-resolution imaging mode to perform continuous frame imaging of ground scenes; Merge pixels of acquired continuous frame images to obtain downsampled images with improved signal-to-noise ratio; Point target detection, target positioning and motion information extraction are performed on the obtained downsampled image, and the target coordinate information is calculated; the target coordinates are mapped to the image of the original resolution according to the coordinate correspondence, and the image slice containing the moving target is extracted according to the set neighborhood value; Using the obtained image slices of each frame to perform precise matching of local areas, a sub-pixel displacement matrix is ​​obtained; The obtained sub-pixel displacement matrix is ​​used as a priori input to reconstruct multiple image slices to obtain super-resolution imaging results.

2. The low-light level super-resolution imaging method for moving targets according to claim 1, characterized in that: The continuous frame imaging of ground scenes comprises: the space-based remote sensing low-light area array camera with micro-scanning super-resolution imaging mode is mounted on a satellite platform with a step-and-stare function, and performs multi-frame staring low-light imaging of ground targets within a continuous integration time.

3. The low-light level super-resolution imaging method for moving targets according to claim 1, characterized in that: The pixel merging of the acquired continuous frame images to obtain a downsampled image with an improved signal-to-noise ratio includes: performing a pixel merging and downsampling operation on each frame of the acquired low-light area array image containing a moving target according to a rule of weighted averaging of adjacent n×n pixels to obtain a downsampled image with an improved signal-to-noise ratio; wherein n is an integer between 1 and 8 and is proportional to the target size.

4. The low-light level super-resolution imaging method for moving targets according to claim 3, characterized in that: The method performs point target detection, target positioning and motion information extraction on the obtained downsampled image, calculates target coordinate information; maps the target coordinates to the image of the original resolution according to the coordinate correspondence, and extracts image slices containing moving targets according to the set neighborhood value, including: performing moving target detection on the sequence frames of the downsampled image, returning the target coordinate value; mapping the returned coordinate value to the image of the original resolution to obtain the target center coordinate value; setting a slice parameter N to extract image slices containing moving targets according to the N×N pixel neighborhood value with the target center coordinate value as the center, and storing the image slice sequence in a temporary storage; the setting of the slice parameter N is greater than the target envelope size.

5. The low-light level super-resolution imaging method for moving targets according to claim 1, characterized in that: The target coordinate information can also be calculated by accessing the infrared point target detection result data with external time synchronization.

6. The low-light level super-resolution imaging method for moving targets according to claim 5, characterized in that: The infrared point target detection result data accessed with external time synchronization includes: being equipped with another medium / long wave infrared channel imaging camera, whose field of view includes the space-based remote sensing low-light area array camera; detecting, detecting and locating the target in the infrared channel, and converting the coordinates into the image plane coordinate system of the space-based remote sensing low-light area array camera after matching in the time and space domains, to obtain the infrared point target detection result data.

7. The low-light level super-resolution imaging method for moving targets according to claim 6, characterized in that: When using infrared point target detection result data connected to external time synchronization, extracting image slices containing moving targets, including: equipping another medium / long wave infrared channel imaging camera, the field of view of which includes the space-based remote sensing low-light area array camera; detecting, detecting, and positioning the target in the infrared channel, and converting the coordinates to the image plane coordinate system of the space-based remote sensing low-light area array camera after matching in the time and space domains, infrared point target detection result data; mapping the returned coordinate values ​​to the image of the original resolution to obtain the target center coordinate value; setting a slice parameter N to extract image slices containing moving targets according to the N×N pixel neighborhood value with the target center coordinate value as the center, and storing the image slice sequence in a temporary storage; the setting of the slice parameter N is greater than the target envelope size.

8. The low-light level super-resolution imaging method for moving targets according to claim 1, characterized in that: The method of using the obtained image slices of each frame to perform precise matching of the local area to obtain the sub-pixel displacement matrix includes: for each frame F in the image slice i Relative to the reference frame F1, i.e., the first frame in time order, the image matching algorithm is executed respectively, and the sub-pixel displacement matrix {H i }.

9. The low-light level super-resolution imaging method for moving targets according to claim 8, characterized in that: The method uses the obtained sub-pixel displacement matrix as a priori input to reconstruct multiple image slices to obtain super-resolution imaging results, including: Each frame image slice and its transformation matrix relative to the reference frame {H i } is used as input and a multi-frame super-resolution reconstruction algorithm is used to calculate the super-resolution imaging result.

10. The low-light level super-resolution imaging method for moving targets according to claim 9, characterized in that: The multi-frame super-resolution reconstruction algorithm specifically uses a convex set projection algorithm.

Images