A method for improving detection signal-to-noise ratio of area array staring camera

Abstract

The present invention discloses a method for improving the detection signal-to-noise ratio of an area array staring camera, comprising the following steps: (1) obtaining a non-uniformity correction coefficient based on K consecutive frames of light and dark images collected by the area array staring camera; (2) the area array staring camera continuously collects multiple frames of target images, and according to the non-uniformity correction coefficient obtained in step (1), performs non-uniformity correction on each frame of the collected target image to obtain multiple frames of target images with reduced spatial noise; (3) performs image registration on the multiple frames of target images with reduced spatial noise in step (2) to obtain multiple frames of target images after registration; (4) performs pixel merging on each frame of the registered target image to obtain each frame of target image after pixel merging; and (5) performs superposition processing on each frame of the target image after pixel merging to obtain a final image with improved signal-to-noise ratio.

Description

Technical Field

[0001] The invention relates to a method for improving the detection signal-to-noise ratio of a planar array staring camera, and belongs to the technical field of remote sensor photoelectric imaging. Background Art

[0002] In the fields of ocean exploration, deep space exploration, early warning detection and low-light imaging, the reflective or radiated energy of the detectable target is low, which belongs to the imaging of dim targets. Using traditional camera detection methods, the detection signal-to-noise ratio is low, and it is difficult to meet the application requirements of high signal-to-noise ratio. The signal strength of the detected target can be increased to a certain extent by increasing the camera aperture and extending the integration time, but the camera aperture and detector performance are limited by current technology, and the resulting signal-to-noise ratio increment still cannot meet the high signal-to-noise ratio requirements. Therefore, it is necessary to seek new technical ways to improve the signal-to-noise ratio. Summary of the invention

[0003] The technical problem solved by the present invention is: to overcome the deficiencies of the above-mentioned prior art and provide a method for improving the detection signal-to-noise ratio of an area array staring camera, which adopts the method of multiple sampling superposition in the time domain and pixel merging in the spatial domain to improve the detection signal-to-noise ratio of dim targets.

[0004] The technical solution provided by the present invention is: a method for improving the detection signal-to-noise ratio of an area array staring camera, and the preferred solution steps are as follows:

[0005] (1) Obtaining a non-uniformity correction coefficient based on K consecutive frames of light and dark images collected by the area array staring camera;

[0006] (2) The area array staring camera continuously collects multiple frames of target images, and performs non-uniformity correction on each frame of the collected target image according to the non-uniformity correction coefficient obtained in step (1), so as to obtain multiple frames of target images with reduced spatial noise;

[0007] (3) performing image registration on the multiple frames of target images after reducing spatial noise in step (2) to obtain the multiple frames of target images after registration;

[0008] (4) performing pixel merging on each frame of the target image after the registration in step (3) to obtain each frame of the target image after pixel merging;

[0009] (5) The target images of each frame after the pixel merging in step (4) are superimposed to obtain a final image with an improved signal-to-noise ratio.

[0010] Preferably, (1) a non-uniformity correction coefficient is obtained according to K consecutive frames of light and dark images collected by the area array staring camera, specifically:

[0011] The non-uniformity correction coefficient is obtained by the two-point method. The area array camera images the integrating sphere, and the integrating sphere is set to light and dark brightness respectively. The curve S determined by the light and dark images is considered to be the standard response curve of the detector. The correction coefficient of each pixel can be obtained from the mean value of the light and dark images and the response value of each pixel.

[0012] Preferably, (2) the area array staring camera continuously collects multiple frames of target images, and according to the non-uniformity correction coefficient obtained in step (1), non-uniformity correction is performed on each frame of the collected target image to obtain multiple frames of target images with reduced spatial noise; specifically,

[0013] The obtained non-uniformity correction coefficients are substituted into the detection unit response values ​​of each frame image, and pixel-by-pixel correction is performed to obtain multi-frame target images with reduced spatial noise.

[0014] Preferably, (3) performing image registration on the multiple frames of target images after reducing spatial noise in step (2) to obtain multiple frames of target images after registration; specifically,

[0015] Due to the jitter of the satellite platform and unstable attitude control, there is image offset between consecutive multiple frames of images. Before the subsequent multi-frame superposition, the obtained K-frame images need to be aligned. In order to calculate the offset between different image frames, the gyroscope can measure the attitude changes of different frames of images. Through a strict imaging geometry model, the image pixel offset between adjacent frames can be calculated. Finally, according to the calculated offset of each frame of the image, they are aligned with the first frame of the image.

[0016] Preferably, (4) performing pixel merging on each frame of the target image after the registration in step (3) to obtain each frame of the target image after pixel merging; specifically:

[0017] Pixel merging uses the method of adding and averaging the DN values ​​of adjacent pixels in the same frame of image. After multiple pixels are merged, they become one pixel, the spatial resolution is reduced, and the image signal-to-noise ratio is improved.

[0018] Preferably, (5) performing superposition processing on each frame of the target image after the pixel merging in step (4) to obtain a final image with improved signal-to-noise ratio; specifically:

[0019] The pixels of K frames acquired continuously are combined and the same pixels of the image are added frame by frame to obtain the average. After superimposing and averaging multiple frames of images, a high signal-to-noise ratio image is obtained.

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

[0021] (1) The present invention proposes a method for improving the signal-to-noise ratio based on the time and space domains. The time domain signal-to-noise ratio improvement utilizes the characteristics of staring imaging to perform multiple sampling superpositions, and the space domain signal-to-noise ratio improvement utilizes adjacent sampling digital merging processing to meet the high signal-to-noise ratio requirements in fields such as ocean exploration.

[0022] (2) The present invention proposes a method for measuring the camera optical axis angle offset using a gyroscope angle measuring element and then performing image registration. This solution has high measurement accuracy, with an accuracy error of less than 1 / 3 pixel, and can simplify the image registration process. At the same time, image registration is not affected by observation time and observation target.

[0023] (3) The present invention proposes a two-point method for detecting pixel response non-uniformity correction, which determines the standard response curve and correction coefficient of the detector through two images, bright and dark. This non-uniformity correction scheme is simple, efficient, and saves onboard hardware resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Flow chart of the image registration algorithm of the present invention;

[0025] Figure 2 Schematic diagram of the effect of vertical optical axis rotation on image shift in the present invention;

[0026] Figure 3 Schematic diagram of bilinear interpolation calculation method of the present invention;

[0027] Figure 4 Schematic diagram of image shift caused by rotation around the optical axis of the present invention;

[0028] Figure 5 The multi-frame superposition of the present invention suppresses temporal noise. After the image is corrected, the effect of multi-frame superposition on the improvement of the signal-to-noise ratio is shown;

[0029] Figure 6 The pixel merging of the present invention suppresses spatial noise. After the image is corrected, the effect of multi-frame superposition on the improvement of the signal-to-noise ratio is shown;

[0030] Figure 7 This is the original image without the effect of improving the signal-to-noise ratio in the spatiotemporal domain;

[0031] Figure 8 This is a diagram showing the effect of improving the signal-to-noise ratio in the spatiotemporal domain according to the present invention. DETAILED DESCRIPTION

[0032] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0033] The present invention discloses a method for improving the detection signal-to-noise ratio of an area array staring camera, comprising the following steps: (1) obtaining a non-uniformity correction coefficient based on K consecutive frames of light and dark images collected by the area array staring camera; (2) the area array staring camera continuously collects multiple frames of target images, and according to the non-uniformity correction coefficient obtained in step (1), performs non-uniformity correction on each frame of the collected target image to obtain multiple frames of target images with reduced spatial noise; (3) performs image registration on the multiple frames of target images with reduced spatial noise in step (2) to obtain multiple frames of target images after registration; (4) performs pixel merging on each frame of the registered target image to obtain each frame of target image after pixel merging; and (5) performs superposition processing on each frame of the target image after pixel merging to obtain a final image with improved signal-to-noise ratio.

[0034] In the fields of ocean exploration, deep space exploration, early warning detection and low-light imaging, the reflective or radiated energy of the detectable target is low, which belongs to the imaging of dim targets. When using traditional camera detection methods, the detection signal-to-noise ratio is low, and it is difficult to meet the application requirements of high signal-to-noise ratio. The method for improving the signal-to-noise ratio based on the spatiotemporal domain proposed in the present invention can improve the target detection signal-to-noise ratio in the above application fields, meet the detection requirements of high signal-to-noise ratio in this field, and realize the detection of dim targets.

[0035] The present invention provides a method for improving the detection signal-to-noise ratio of an area array staring camera, and the preferred steps are as follows:

[0036] (1) Obtaining a non-uniformity correction coefficient based on K consecutive frames of images collected by the area array staring camera; the preferred solution is as follows:

[0037] The preferred solution is: the non-uniformity correction is performed using a two-point method, and the integrating sphere is set to two brightnesses ΦA and ΦB, respectively. The curve S determined by ΦA and ΦB is considered to be the standard response curve of the detector. Next, the response curves of the remaining detectors are corrected according to the curve S. The correction method is as follows:

[0038] First, the average response value of each detection unit under uniform irradiance ΦA and ΦB is obtained:

[0039]

[0040]

[0041] In general, the detector response curve is a linear model within the dynamic range of the detector, and the formula is as follows:

[0042] Y ij (Φ) = G ij (Φ)+O ij (3)

[0043] where Y ij(Φ) represents the response output value of the detection unit numbered (i, j), Φ represents the radiation input value of the detection unit (i, j), G ij is the multiplicative gain of the (i, j) detection unit, O ij is the additive bias of the (i, j) detection unit.

[0044] Substituting the average response value under bright and dark radiation into the above linear model, we can get

[0045]

[0046]

[0047] The correction gain coefficient and offset correction coefficient can be obtained by the above two equations:

[0048]

[0049]

[0050] (2) The area array staring camera continuously collects multiple frames of target images, and performs non-uniformity correction on each frame of the collected target image according to the non-uniformity correction coefficient obtained in step (1), so as to obtain multiple frames of target images with reduced spatial noise; the preferred scheme is specifically as follows:

[0051] The preferred solution is: according to step (1), two correction coefficients k are obtained for each detection unit: ij 、b ij Then the correction output of each detection unit can be obtained as

[0052]

[0053] By performing the above correction processing on K frames of images respectively, multiple frames of target images with reduced spatial noise can be obtained.

[0054] (3) performing image registration on the multiple target images after reducing spatial noise in step (2) to obtain the multiple target images after registration; Figure 1 As shown, the preferred solution is as follows:

[0055] The preferred solution is: in order to calculate the offset between different image frames, the gyroscope can measure the posture changes of different frame images, and through the strict imaging geometry model, the image pixel offset between adjacent frames can be calculated. The algorithm flow is as follows Figure 1 shown.

[0056] The rotation displacements in the roll and pitch axis directions can be collectively referred to as the rotation displacements in the direction perpendicular to the optical axis. They have similar effects on image shift, so they are put together. Figure 2 It is a schematic diagram showing the effect of the vertical optical axis rotation on the image shift of the present invention;

[0057] The preferred solution is: Figure 2 As shown, the angle between the target principal ray and the optical axis before the rotation is α, and the satellite has an angular displacement of θx (or θy) around the ox axis (or oy axis) during the integration time. Then the image displacement dim of the image point in the oy direction (or ox direction) after the rotation can be approximately expressed as:

[0058] dim=f·tan(α+θx)-tanα-≈f·tanθx (9)

[0059] Where f is the focal length of the camera, L is the straight-line distance between the target and the subsatellite point of the optical axis, l' is the image distance of L, and H is the satellite altitude.

[0060] The preferred solution is: according to the calculated image offset, the image shift of integer pixels is directly translated, and the image shift of sub-pixels is resampled by bilinear interpolation to adjacent frames, such as Figure 3 As shown in the figure, the specific algorithm is to calculate the new grid value of the sampling point by weighting the distance of the four neighboring pixels around the sampling point. The specific operation is to first interpolate once in the X direction, then interpolate once in the Y direction, and obtain the grid value of the pixel by distance weighted calculation.

[0061] The optimal calculation formula is:

[0062] I(P)=(1-Δx)(1-Δy)·I 11 +(1-Δx)·Δy·I 12 +Δx·(1-Δy)·I 21 +Δx·Δy·I 22 (10)

[0063] Among them, I(P) is the DN value of the sampling point pixel, P 11 , P 12 , P 13 , P 14 They are the four pixels around the sampling point, I 11 ,I 12 ,I 13 ,I 14 are the DN values ​​of the four surrounding pixels, and Δx is the distance P from the sampling point 11 and P 12 The straight-line distance of the center line in the x direction, Δy is the distance P from the sampling point 11 and P 21 The straight-line distance between the centers in the y direction.

[0064] The preferred solution is: the rotation around the optical axis is also the rotation around the yaw axis. Figure 4The image shift caused by the rotation displacement around the optical axis is shown in the figure. During the integration time, the image shift dim caused by the rotation displacement of the remote sensing satellite around the optical axis θz can be expressed as:

[0065] dim=l·θz (11)

[0066] Where l is the distance between the target image point and the center of the field of view.

[0067] The preferred solution is: according to the calculated image offset, the image shift of integer pixels is directly translated, and the image shift of sub-pixels is resampled by bilinear interpolation to adjacent frames. The specific algorithm is the same as formula (10).

[0068] (4) merging pixels of each frame of the target image after the registration in step (3) to obtain each frame of the target image after pixel merging; the preferred solution is as follows:

[0069] The preferred solution is: adopting the method of adding and averaging the DN values ​​of adjacent pixels in the same frame image, multiple pixels are merged into one pixel, the spatial resolution is reduced, and the image signal-to-noise ratio is improved. If n*n pixels are merged, the spatial noise of the image is suppressed, and the signal-to-noise ratio of each point in the image is increased by n times. As the number of merged pixels increases, the image signal-to-noise ratio also increases.

[0070] (5) performing superposition processing on each frame of the target image after the pixel merging in step (4) to obtain a final image with improved signal-to-noise ratio; the preferred solution is as follows:

[0071] The preferred solution is to use the method of adding and averaging the same pixel of multiple frames of images collected continuously, and then superimposing and averaging multiple frames to obtain a high signal-to-noise ratio image. The images used for multi-frame superposition must be images registered at the same integration time and location. If K frames of images are superimposed and averaged, the temporal noise of the image is suppressed, and the signal-to-noise ratio of each point in the image is improved. Times, as the number of superimposed image frames increases, the cumulative signal-to-noise ratio also increases.

[0072] The preferred solution is: multi-frame superposition suppresses temporal noise. After the image is corrected, the effect of multi-frame superposition on improving the signal-to-noise ratio is as follows: Figure 5 As shown (100, 300, 500, 1000 are the sizes of image statistical areas, and the signal-to-noise ratio improvement curves basically overlap).

[0073] Pixel merging suppresses spatial noise. After the image is corrected, the effect of multi-frame superposition on improving the signal-to-noise ratio is as follows: Figure 6 shown.

[0074] The image has been improved in the spatiotemporal domain. Figure 7 , Figure 8 As shown; ( Figure 7 For the original picture, Figure 8 It is a 4*4 merged image with 4 frames superimposed at the same time).

[0075] The present invention proposes a method for improving the signal-to-noise ratio based on the time and space domain. The time domain signal-to-noise ratio improvement utilizes the characteristics of staring imaging to perform multiple sampling superpositions, and the space domain signal-to-noise ratio improvement utilizes adjacent sampling digital merging processing to meet the high signal-to-noise ratio requirements in fields such as ocean exploration; the present invention proposes a method for measuring the camera optical axis angle offset using a gyroscope angle measurement element and then performing image registration. This solution has high measurement accuracy, with an accuracy error of less than 1 / 3 pixel, and can simplify the image registration process. At the same time, image registration is not affected by observation time and observation target.

[0076] The present invention proposes a two-point method for detecting pixel response non-uniformity correction, which determines the standard response curve and correction coefficient of the detector through two bright and dark images. The non-uniformity correction scheme is simple, efficient and saves onboard hardware resources.

Claims

1. A method for improving the detection signal-to-noise ratio of an area array staring camera, characterized in that Here are the steps: (1) Obtaining a non-uniformity correction coefficient based on K consecutive frames of light and dark images collected by the area array staring camera; (2) The area array staring camera continuously collects multiple frames of target images, and performs non-uniformity correction on each frame of the collected target image according to the non-uniformity correction coefficient obtained in step (1), so as to obtain multiple frames of target images with reduced spatial noise; (3) performing image registration on the multiple frames of target images after reducing spatial noise in step (2) to obtain the multiple frames of target images after registration; comprising: The rotation displacements in the roll axis and pitch axis directions are uniformly referred to as the rotation displacement in the direction of the vertical optical axis. Assuming that the angle between the target principal ray and the optical axis before rotation is α, the satellite undergoes an angular displacement of θx around the ox axis during the integration time. The image displacement dim of the image point in the oy direction after rotation is approximately expressed as: dim=f·tan(α+θx)-tanα≈f·tanθx Where f represents the focal length of the camera; According to the calculated image shift, the image shift of integer pixels is directly translated, and the image shift of sub-pixel pixels is resampled by bilinear interpolation to adjacent frames; During the integration time, the image shift dim caused by the angular displacement of the remote sensing satellite around the optical axis is expressed as: dim=l·θz Where l is the distance between the target image point and the center of the field of view; According to the calculated image shift, the image shift of integer pixels is directly translated, and the image shift of sub-pixel pixels is resampled by bilinear interpolation to adjacent frames; Due to the jitter of the satellite platform and the instability of attitude control, there is an image offset between consecutive multiple frames of images. Before the subsequent multi-frame superposition, the obtained K-frame images need to be registered. In order to calculate the offset between different image frames, the attitude changes of different frames are measured by the gyroscope, and the image pixel offset between adjacent frames is calculated by the strict imaging geometry model. Finally, according to the calculated offset of each frame image, it is registered with the first frame image. (4) performing pixel merging on each frame of the target image after the registration in step (3) to obtain each frame of the target image after pixel merging; specifically: Pixel merging uses the method of adding and averaging the DN values ​​of adjacent pixels in the same frame of image. Multiple pixels are merged into one pixel, which reduces the spatial resolution and improves the image signal-to-noise ratio. (5) The target images of each frame after the pixel merging in step (4) are superimposed to obtain a final image with improved signal-to-noise ratio, specifically: The pixels of K frames acquired continuously are combined and the same pixels of the image are added frame by frame to obtain the average. After superimposing and averaging multiple frames of images, a high signal-to-noise ratio image is obtained.

2. The method for improving the detection signal-to-noise ratio of an area array staring camera according to claim 1, characterized in that: (1) According to the K consecutive frames of light and dark images collected by the area array staring camera, the non-uniformity correction coefficient is obtained, which is: The non-uniformity correction coefficient is obtained by the two-point method. The area array camera images the integrating sphere, and the integrating sphere is set to light and dark brightness respectively. The curve S determined by the light and dark images is considered to be the standard response curve of the detector. The correction coefficient of each pixel can be obtained from the mean value of the light and dark images and the response value of each pixel.

3. The method for improving the detection signal-to-noise ratio of an area array staring camera according to claim 1, characterized in that: (2) The area array staring camera continuously collects multiple frames of target images, and performs non-uniformity correction on each frame of the collected target image according to the non-uniformity correction coefficient obtained in step (1), so as to obtain multiple frames of target images with reduced spatial noise; specifically: The obtained non-uniformity correction coefficients are substituted into the detection unit response values ​​of each frame image, and pixel-by-pixel correction is performed to obtain multi-frame target images with reduced spatial noise.

Images