Target object camera array imaging method based on multi-view-field deconvolution turbulence resistance

Abstract

The invention provides a target object camera array imaging method based on multi-view-field deconvolution turbulence resistance, and relates to the technical field of image deblurring or deconvolution image processing based on a physical model, and the method comprises the steps: S1, building a physical model in which atmospheric turbulence affects the imaging quality of a target object, building a camera array imaging system, and carrying out the imaging of the target object; obtaining a target object image in a turbulence environment; s2, establishing a multi-view-field affine transformation registration model, and performing image registration through affine transformation to obtain a plurality of same-view-field image sequences of the target object; s3, performing deconvolution by using a convolution kernel of a point spread function of a preset optical system, enhancing the contrast of the target object image, and estimating the target object image; and S4, performing Fourier ring correlation operation based on the contrast-enhanced target object image, and performing adaptive iteration deconvolution again to obtain a recovered anti-turbulence target object image. According to the method, pre-deconvolution is performed on the original image, background noise is suppressed, and adaptive iteration deconvolution is performed after the image autocorrelation cumulant is calculated, so that anti-turbulence imaging is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of anti-turbulence imaging technology, and particularly relates to a multi-viewfield deconvolution anti-turbulence target object camera array imaging method. BACKGROUND

[0002] Turbulence is a common form of fluid motion in nature, and its irregular refractive index fluctuation can affect light wave transmission and image acquisition. Atmospheric and underwater turbulence often leads to a decline in photoelectric imaging quality, and in severe cases, it can even lead to no imaging. One reason is that the random fluctuation of the density of the transmission medium such as air or water in the atmosphere layer causes the refractive index to be uneven, resulting in changes in optical phase and amplitude. The second reason is that the low detection rate of the imaging detector causes the integral effect in the light wave transmission path to be obvious. The influence of turbulence on optical imaging is described by random disturbance in wavefront phase and light intensity fluctuation. Atmospheric turbulence is the most likely environmental disturbance under high-temperature sunny weather conditions, and finding a method to solve or weaken the problem of atmospheric turbulence disturbance is of great significance to the development of the imaging field.

[0003] The existing anti-turbulence research mainly has three directions: one is to weaken the interference of atmospheric turbulence on light wave transmission by designing special structure light source. Because the coherent structured light field does not have the invariance of long-distance transmission, scholars introduce special structure light source into computational correlation imaging to improve the coherence of light field in far field transmission. The theory of multi-Gaussian incoherent source CGI and CGD under the action of turbulence is established by using beam shaping in CGI and CGD. It is found that the appropriate increase of the beam order of multi-Gaussian source can improve the turbulence resolution of CGI and CGD. The point spread function of hollow Gaussian beam in lensless ghost imaging system is found when it passes through atmospheric turbulence. It is found that with the increase of the order of hollow Gaussian beam, the width of PSF in imaging system is narrowed, and the resolution of reconstructed image is effectively improved compared with Gaussian light source. In addition, researchers carry out the anti-turbulence theory research of Bessel beam, Lorentz type incoherent light, vortex partial coherent light, quantum entangled light, thermal light and other light sources. The second is to use adaptive optics technology to correct the phase distortion caused by atmospheric turbulence in real time to reduce its influence. In recent years, adaptive optics system gradually gets rid of the dependence on traditional wavefront sensor, and turns to model-free algorithm based on objective function optimization. The improved particle swarm optimization algorithm generates the control voltage of deformable mirror by directly analyzing the spot intensity distribution feedback by photoelectric detector, realizes the correction of residual error of λ / 15 level under the condition of no wavefront coherence detection, and the convergence speed is improved by 30% compared with traditional SPGD algorithm. The real-time control framework based on P-U-net neural network directly maps the distorted pupil image to the driving voltage of deformable mirror, avoiding the linear response assumption and hysteresis effect error of traditional PI algorithm, and the processing speed is millisecond level under the scene of 1000 order aberration. The third is to use deconvolution algorithm to post-process the acquired image to improve the imaging quality. The traditional blind deconvolution relies on iterative optimization, which has large calculation amount and slow convergence. The spatial variation convolution simulation method based on deep neural network converts the multi-frame blind deconvolution problem into forward model training, avoiding the traditional splicing artifact, and the calculation efficiency is improved by more than 50%. In addition, the turbulence compensation technology integrated with improved gated recurrent unit GRU optimizes the deconvolution parameters through time series modeling, and reduces the bit error rate to 10 −9 level in the communication link of naval aviation field.

[0004] However, special structure light source needs to have high power to ensure signal strength and stability when transmitting at a long distance; adaptive optics anti atmospheric turbulence has remarkable effect, but is high in cost, complex in system and limited in effect on extreme weather; the deconvolution method can effectively restore image details and improve observation accuracy, but is highly dependent on initial estimation and has high computational complexity. With the wide application of deep learning technology in various fields, researches on anti atmospheric turbulence imaging using deep network have been proposed, but still face challenges in data set collection and turbulence model establishment.

[0005] To solve the above problems, the application provides a target object camera array imaging method based on multi-view field deconvolution anti turbulence, which performs pre-deconvolution on the original image sequence to suppress background noise, and further improves the resolution of the image by calculating the autocorrelation accumulation of the pre-deconvolved image and performing adaptive deconvolution. SUMMARY

[0006] To solve the above problems of the prior art, the application provides a target object camera array imaging method based on multi-view field deconvolution anti turbulence, which establishes a physical model of atmospheric turbulence affecting target object imaging quality, builds a camera array imaging system, and obtains target object images in a turbulence environment; a multi-view field affine transformation registration model is established, image registration is performed through affine transformation, and a plurality of same-view field image sequences of the target object are obtained; then pre-deconvolution is performed on the original image sequence to suppress background noise; finally, the autocorrelation accumulation of the pre-deconvolved image is calculated, and adaptive deconvolution is performed to further improve the resolution of the image, significantly improve the image quality in a turbulence environment, and realize anti turbulence imaging.

[0007] Specifically, the application provides a target object camera array imaging method based on multi-view field deconvolution anti turbulence, which includes the following steps: S1: a physical model of atmospheric turbulence affecting target object imaging quality is established, a camera array imaging system is built, a turbulence influence optical transfer function of target object imaging is determined, and target object images in a turbulence environment are obtained; S2: a multi-view field affine transformation registration model is established, and the target object images in the turbulence environment obtained in step S1 are subjected to image registration through affine transformation to obtain a same-view field image sequence of the target object; S3: performing adaptive pre-deconvolution processing on the sequence of the same field images of the target object obtained in step S2, performing Fourier ring correlation (FRC) operation on every two target object images, selecting half of the corresponding number of the minimum point value of the Fourier ring correlation (FRC) as the iteration number k1 of adaptive pre-deconvolution using the preset optical system point spread function convolution kernel, enhancing the contrast of the target object image, and retaining the focal plane signal, and the specific process is as follows: ; wherein, is the estimated image of the target object after the i+1th iteration; is the estimated image of the target object after the ith iteration; is the camera array imaging system function; is the degraded target object image; is the convolution operation; is the horizontal coordinate of the two-dimensional image; is the vertical coordinate of the two-dimensional image; i is a non-negative integer, ; Iteration initial condition is the plurality of same field images of the target object under turbulence output in step S2, and the output of k1 times of adaptive pre-deconvolution is , which is the enhanced contrast target object image; S4: performing adaptive deconvolution processing on the enhanced contrast target object image retaining the focal plane signal obtained in step S3, performing Fourier ring correlation (FRC) operation on every two target object images again, selecting half of the corresponding number of the minimum point value of the Fourier ring correlation (FRC) as the iteration number k2 of adaptive deconvolution, and obtaining the best prediction image of the recovered target object image of the target object image under turbulence , which is the recovered anti-turbulence target object image, and the specific process is as follows: ; Iteration initial is the enhanced contrast target object image retaining the focal plane signal output in step S3, and the output of k2 times of adaptive deconvolution is the best prediction image of the recovered target object image under turbulence .

[0008] Preferably, step S3 is specifically as follows: S31: performing deconvolution on the sequence of the same field images of the target object using the convolution kernel of the preset optical system point spread function, enhancing the contrast, and degrading the target object image , and the original clear target object image , and the camera array imaging system function satisfies the Poisson distribution; S32: Obtain the deconvolution iteration equation of the deconvolution algorithm according to step S31, and evaluate the resolution of the reconstructed image at each iteration using the Fourier ring correlation (FRC); S33: Set the initial condition of iteration as That is, the multiple same-view images of the target object under turbulence output by step S2, the degraded target object image After k1 iterations, the estimated image of the degraded target object image is .

[0009] Preferably, the convolution kernel of the preset optical system point spread function in step S31 is specifically: ; ; Wherein, is the convolution kernel of the preset optical system point spread function; is the point amplitude function; C is a normalized constant related to light intensity; is the wave number of the wavefront function; is the aperture of the wavefront function; is the focal length of the wavefront function; is the normalized horizontal coordinate; is the normalized vertical coordinate; is the parameter variable of the wavefront function; is the imaginary unit parameter.

[0010] Preferably, the degraded target object image and the original clear target object image and the camera array imaging system function satisfy the Poisson distribution, which is specifically: ; Wherein, is the degraded target object image; is the original clear target object image; is the Poisson distribution probability density function; is the factorial operation.

[0011] Preferably, the Fourier ring correlation (FRC) calculation formula in step S32 is: ; Wherein, is the Fourier ring correlation parameter of the target object image in the frequency domain; is the Fourier transform of the first image; is the Fourier transform of the second image; is the kth frequency ring; to calculate the radius of the ring where the frequency value of the taken pixel point is located; to perform Fourier transform on the second image to perform conjugate Fourier transform on the result of the Fourier transform; k is the frequency ring number.

[0012] Preferably, step S33 is specifically: The initial condition of iteration is to output a plurality of co-FOV images of the target object under turbulence in step S2, and then to solve iteratively in step S32. With the increase of the number of iterations, the estimated image of the target object image after the k1+k2th iteration will finally converge to the original clear target object image , so as to restore the original clear image of the target object, which is the result of anti-turbulence imaging.

[0013] Preferably, the camera array imaging system built in step S1 includes a camera array, an optical imaging lens, and a target object, wherein the camera array is composed of a plurality of charge-coupled cameras (CCD) arranged regularly to form a detector; the sunlight field is irradiated to the surface of the target object through atmospheric turbulence, the reflected light of the target object is received by the optical imaging lens after passing through the atmospheric turbulence, and then enters the camera array to be imaged and detected, so as to realize multi-FOV imaging of the imaging target, obtain a plurality of co-FOV images of the target object, and then register the images through affine transformation to obtain a plurality of co-FOV image sequences of the target object; the focal plane transformation relationship of the cameras in the camera array is a translation transformation relationship or a matrix transformation relationship.

[0014] Preferably, the translation transformation relationship of the camera array, the coordinates of the object P in the first camera and the coordinates of the object P in the second camera are transformed as follows: ; wherein, is the horizontal coordinate of the target object in the second camera; is the horizontal coordinate of the target object in the first camera; is the horizontal coordinate transformation relationship between the first camera and the second camera; is the vertical coordinate of the target object in the second camera; is the vertical coordinate of the target object in the first camera; is the vertical coordinate transformation relationship between the first camera and the second camera.

[0015] Preferably, the matrix transformation relationship of the camera array, the coordinates of the object P in the first camera and the coordinates of the object P in the second camera are transformed as follows: ; Preferably, the physical model of the gas turbulence affecting the imaging quality of the target object in step S1 is: ; Wherein, is the surface light field distribution of the charge coupled camera (CCD); is the reflectivity function of the object surface; is the optical point spread function from the light source to the target object to the detector charge coupled camera (CCD); is the optical point spread function from the target object to the detector charge coupled camera (CCD); is the target object reflection data; is the surface light field distribution of the target object; is the sunlight source data; is the derivative of the light field input data; is the derivative of the sunlight source data.

[0016] Compared with the prior art, the beneficial effects of the present application are as follows: (1) The present application establishes a continuous multi-frame imaging of the target object by the imaging system to obtain the original image sequence of the target object imaging; then performs pre-deconvolution on the original image sequence to suppress the background noise; finally, the autocorrelation accumulation of the pre-deconvolution image is calculated and deconvolution is performed, further improving the resolution of the image, realizing anti-turbulence imaging.

[0017] (2) The method proposed in the present application has more obvious advantages in improving the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) image quality evaluation indicators compared with typical anti-turbulence imaging algorithms, significantly improving the image quality in turbulence environment, and providing an effective technical solution for atmospheric turbulence imaging detection.

[0018] (3) The present application can realize target object image reconstruction in atmospheric turbulence without calibration, and is also applicable in haze scattering medium, significantly solving the problem of traditional imaging method affected by turbulence, and significantly improving the imaging quality of the target object in atmospheric turbulence environment. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is the flow chart of the camera array imaging method of the target object based on multi-view deconvolution anti-turbulence of the present application; Figure 2 is the schematic diagram of the camera array imaging system of the present application; Figure 3 is the flow chart of the camera array imaging method of the present application; Figure 4 is the overall scheme diagram of the camera array imaging of the present application; Figure 5 is a housing building sample remote sensing image taken by a satellite; Figure 6 An example image of a satellite shot under strong turbulence intensity for simulation; Figure 7 An example image of a satellite shot under strong turbulence intensity for simulation after processing using a lucky imaging method; Figure 8 An example image of a satellite shot under strong turbulence intensity for simulation after processing using a target object camera array imaging method based on multi-view field deconvolution anti-turbulence; Figure 9 An example image of a first view field of turbulence collection for a second case; Figure 10 An example image of a 25th view field of turbulence collection for a second case; Figure 11 An example image of a first view field of turbulence collection for a second case after registration; Figure 12 An example image of a 25th view field of turbulence collection for a second case after registration; Figure 13 A second case image after processing using a lucky imaging method; Figure 14 A second case image after processing using a target object camera array imaging method based on multi-view field deconvolution anti-turbulence; DETAILED DESCRIPTION

[0020] Hereinafter, embodiments of the present application will be described with reference to the accompanying drawings.

[0021] The embodiments of the present application propose a target object camera array imaging method based on multi-view field deconvolution anti-turbulence, as shown in the formula (1). Figure 1 A physical model of atmospheric turbulence affecting the imaging quality of a target object is established, a camera array imaging system is built, and a target object image is obtained; a multi-view field affine transformation registration model is established, image registration is performed through affine transformation, and a sequence of multiple same view field images of the target object is obtained; a deconvolution is performed using a convolution kernel of a preset optical system point spread function, the contrast of the target object image is enhanced, and the target object image is estimated; a Fourier ring correlation operation is performed, and a restored anti-turbulence target object image is obtained; and the specific steps include the following steps: Step S1: a camera array imaging system is established, including a camera array, an optical imaging lens, and a target object; the camera array is composed of multiple groups of charge-coupled cameras (CCD) arranged according to a rule to form a detector; sunlight passes through atmospheric turbulence and irradiates the surface of the target object, the reflected light of the target object passes through atmospheric turbulence, is received by the optical lens, and enters the camera array to be imaged and detected, realizing multi-view field imaging of the imaging target, and obtaining multiple view field images of the target object imaging. As shown in the formula (2). Figure 2The camera array imaging system is shown in the schematic diagram. The camera array imaging system is used for continuous multi-frame imaging of the imaging target, and the original image sequence of the target object imaging is obtained, that is, the target object image. A physical model of the influence of atmospheric turbulence on the imaging quality of the target object is established. The turbulence influence optical transfer function of the light field from the sun to the target object and then reflected to the charge coupled camera CCD is described as: ; wherein, is the light field distribution on the surface of the charge coupled camera CCD; is the reflectivity function of the object surface; is the optical point spread function from the light source to the target object to the detector charge coupled camera CCD; is the optical point spread function from the target object to the detector charge coupled camera CCD; is the target object reflection data; is the light field distribution on the surface of the target object; is the light field input data; is the sun light source data; is the derivative of the light field input data; is the derivative of the sun light source data.

[0022] According to the extended Huygens-Fresnel integral, the expressions of the optical point spread function from the light source to the target object to the detector charge coupled camera CCD and the optical point spread function from the target object to the detector charge coupled camera CCD are respectively: ; ; wherein, is the distance between the sun and the target object; is the distance between the target object and the detector charge coupled camera CCD; is the light field phase disturbance caused by atmospheric turbulence between the light source sun and the target object; is the light field phase disturbance caused by atmospheric turbulence between the target object and the detector charge coupled camera CCD; is the magnification of the image, ; is the aperture of the optical imaging lens; is the natural logarithm; is the imaginary unit; is the wavelength of the sun light; is the parameter of the circular constant; is the focal length of the optical imaging lens; The distance between the target object and the detector charge-coupled camera (CCD); The differential of the distance between the target object and the detector charge-coupled camera (CCD).

[0023] As Figure 3 The camera array imaging method flowchart of the present application is shown in the figure, and the camera array imaging is realized through multi-view images, RL pre-deconvolution, calculation of autocorrelation and sparse deconvolution. The anti-turbulence imaging method of the present application has a relatively remarkable anti-turbulence imaging effect, and needs to satisfy not less than 16 times of imaging of the target object, so that not less than 16 charge-coupled cameras (CCD) are adopted for one-time imaging of the camera array, or less than 16 charge-coupled cameras (CCD) are adopted for multiple imaging of the camera array, as shown in the figure. Figure 4 The overall scheme of the camera array imaging is shown in the figure, and in the embodiment, the camera array one-time imaging scheme of not less than 16 cameras is adopted, and multiple different view images of the target object are obtained.

[0024] Step S2: A multi-view affine transformation registration model is established, and the target object images obtained in step S1 are subjected to image registration through affine transformation. The affine transformation changes include scaling, translation, rotation, reflection, skewing and the like, so that the multi-view images of the target object are registered to obtain a sequence of multiple same-view images of the target object. The registration of the target object with a single same-depth field of view of the camera array imaging device in the multi-view image only needs translation and rotation operations; the registration of the multi-depth target object image needs scaling, translation, rotation, reflection and skewing operations.

[0025] After the camera array is arranged in the same plane, the first camera focal plane view image and the second camera focal plane view image are in a translation transformation relationship. A point P of the target object is imaged in the first camera and the second camera as P1 and P2 respectively, so that And The transformation relationship is: ; Wherein, is the horizontal coordinate of the target object in the second camera; is the horizontal coordinate of the target object in the first camera; is the horizontal coordinate transformation relationship between the first camera and the second camera; is the vertical coordinate of the target object in the second camera; is the vertical coordinate of the target object in the first camera; is the vertical coordinate transformation relationship between the first camera and the second camera.

[0026] The transformation relationship of the matrix transformation relationship is: ; The horizontal coordinate and vertical coordinate transformation relationship between the first camera and the second camera is two unknowns and Two unknowns are obtained through the calibration of the charge-coupled camera (CCD), so as to realize the image registration of different fields of view and obtain a sequence of images of the target object in the same field of view. Figure 4 As shown in Fig. 1, a camera array imaging overall scheme is shown, and in the embodiment, a plurality of different field of view images obtained by one imaging of a camera array of not less than 16 charge-coupled cameras (CCDs) are adopted, and the parameters of the calibrated charge-coupled camera (CCD) are inputted and registered through affine transformation, so as to realize the image registration of different fields of view and obtain the image of the target object.

[0027] Step S3: estimating the image of the target object by using the anti-turbulence imaging algorithm of multi-step deconvolution.

[0028] Step S31: based on the anti-turbulence imaging algorithm of multi-step deconvolution, the multi-step adaptive deconvolution is used for anti-turbulence imaging of the target object, the convolution kernel of the preset optical system point spread function is used for deconvolution on the sequence of images of the target object in the same field of view processed in step S2, the contrast of the target object is enhanced, the background noise of the out-of-focus plane is suppressed, and the signal of the in-focus plane is reserved; the convolution kernel of the preset optical system point spread function is calculated by Fourier transform of the wavefront function, and specifically is: ; wherein, is the point amplitude function; is the horizontal coordinate of the two-dimensional image; is the vertical coordinate of the two-dimensional image; C is a normalized constant related to the light intensity; is the wave number of the wavefront function, ; is the aperture of the wavefront function; is the focal length of the wavefront function; is the normalized horizontal coordinate; is the normalized vertical coordinate; is the parameter variable of the wavefront function; is the imaginary unit parameter.

[0029] The acquisition method of the parameter variable of the wavefront function is: ; wherein, is the Zernike coefficient; is the orthogonal basis function, which is used for quantifying the aberration type such as spherical aberration and coma.

[0030] The convolution kernel of the preset optical system point spread function is: ; wherein, is a preset convolution kernel of a point spread function of an optical system.

[0031] degraded target object image and original clear target object image and camera array imaging system function satisfy a Poisson distribution, specifically: ; wherein, is a degraded target object image; is an original clear target object image; is a camera array imaging system function; is a Poisson distribution probability density function; is a convolution operation; is a factorial operation.

[0032] Step S32: obtaining a deconvolution iteration equation based on the Richardson-Lucy deconvolution algorithm according to step S31 is: ; wherein, is a target object image estimation image after the i+1th iteration; is a target object image estimation image after the ith iteration; is a camera array imaging system function; is a degraded target object image; is a convolution operation; is a horizontal coordinate of a two-dimensional image; is a vertical coordinate of a two-dimensional image; i is a non-negative integer, ; initial condition of iteration that is, a plurality of same-view images of the target object under turbulence output in step S2.

[0033] In the deconvolution iteration process, the resolution of the reconstructed image under each iteration is evaluated using Fourier ring correlation (FRC), and the convolution kernel corresponding to the best resolution is determined by analyzing the lowest point of the Fourier ring correlation (FRC) curve, thereby realizing automatic updating of the convolution kernel in the deconvolution iteration process. The Fourier ring correlation (FRC) calculation formula is: ; wherein, is a Fourier transform of a first image; is a Fourier transform of a second image; is a kth frequency ring; k is a frequency ring number; is a Fourier ring correlation parameter of the target object image in the frequency domain; is a ring radius at which the frequency value of the pixel point taken for calculation is located. for the Fourier transform of the second image the conjugate Fourier transform of the Fourier transform of the second image.

[0034] Step S33: in the deconvolution iteration of the Richardson-Lucy deconvolution algorithm, the degraded target object image after i iterations is used to obtain the target object image estimate image after the i iteration of the original clear target object image. The initial condition of the iteration is that the multiple same-view images of the target object under turbulence output in step S2 are iteratively solved through step S32, and in the ideal case without noise, the target object image estimate value after the k1 iteration will eventually converge to the original clear target object image , thereby restoring the original clear image of the target object. At this time, the camera array imaging system function is the point spread function of the imaging system, and different imaging systems need to calibrate the point spread function of the system and preset the value to the algorithm. The same-view image sequence of the target object obtained in step S2 is adaptively pre-deconvolved. The half of the corresponding number at the Fourier ring correlation FRC minimum point value is selected as the convolution kernel of the preset optical system point spread function to perform the first deconvolution iteration number k1, thereby suppressing the out-of-focus background noise through deconvolution and retaining the in-focus signal.

[0035] Step S4: the original clear target object image of the target object is restored by using the output of step S3 to perform adaptive deconvolution iteration calculation again.

[0036] After the degraded target object image is adaptively pre-deconvolved for k1 times, the enhanced contrast target object image retaining the in-focus signal is obtained. The enhanced contrast target object image retaining the in-focus signal obtained in step S3 is adaptively deconvolved, and the Fourier ring correlation FRC operation is performed again on every two target object images. The half of the corresponding number at the Fourier ring correlation FRC minimum point value is selected again as the second deconvolution iteration number k2, and finally the best prediction image of the target object image under turbulence conditions is obtained, which is the prediction image for obtaining the restored anti-turbulence target object image. Specifically, ; wherein, is the target object image estimate image after the i+1 iteration; is the target object image estimate image after the i iteration; for adaptive updating of a system function; for convolution operation; for abscissa of a two-dimensional image; for ordinate of a two-dimensional image; i is a non-negative integer, ; initial for the enhanced contrast target object image of the reserved focal plane signal output by step S3, k2 times of adaptive deconvolution are performed to output a recovered target object image , which is the obtained anti-turbulence target object imaging image.

[0037] As Figure 4 shown, in the embodiment, the existence of atmospheric turbulence can cause phase distortion and intensity fluctuation of the light field reflected by the target object, thereby causing significant decline in imaging quality. Compared with traditional imaging under atmospheric turbulence, the imaging method of the present application can obtain clearer results and effectively realize anti-atmospheric turbulence imaging.

[0038] The first case in the embodiment is the analysis of simulation experiment results as follows: To verify the effectiveness of the method, the corresponding refractive index structure constants -2 / 3 of weak, medium and strong turbulence intensities in the simulation experiment are 5*10-17 m -2 / 3 , 5*10-15 m -2 / 3 and 5*10-13 m , respectively. The imaging effect of the present application is analyzed and compared with the lucky imaging method. The turbulence-degraded image is obtained, the power spectrum inversion method is used to generate an atmospheric turbulence random phase screen conforming to the Kolmogorov statistical law. By interacting the phase screen with the target object image, the target object image after transmission through different turbulence intensities is simulated and simulated.

[0039] The target object image of the simulation experiment is a 0.5m resolution black and white satellite image of a housing building sample remote sensing image taken by a satellite. Since the size is large, 998*998 pixels, the experiment extracts Figure 5 the part marked by the square in the middle as 200*200 pixels as the original image for simulation. In the simulation setting, it is assumed that the distance between the target object and the detector is 50 km, the receiving aperture is 0.6m, and 10 random phase screens are set on the transmission path. The turbulence model is used to simulate the target object image, and 20 degraded images are generated under each turbulence intensity. After processing through weak, medium and strong atmospheric turbulence intensities, the degraded image under strong turbulence intensity is shown in Figure 6 , it can be observed that the blurring degree of the image under strong turbulence intensity is large. The anti-turbulence effects of the present application and the lucky imaging method are compared in the simulation experiment, wherein the anti-turbulence imaging result of the lucky imaging method under strong turbulence intensity is shown in Figure 7As shown, the overall image is blurred and it is difficult to distinguish the object signs in the image. Figure 8 As shown, compared with Figure 6 and Figure 7 In the strong turbulence environment, the simulation results of the atmospheric turbulence imaging resistance show that the present application can restore more detailed information and the imaging effect is stronger than that of the classic lucky imaging algorithm.

[0040] To evaluate the effects of the present application and the lucky imaging method in restoring turbulence images, the peak signal-to-noise ratio (PSNR) and the structural similarity (SSIM) of the restored images are calculated. The turbulence degradation model generated in the simulation experiment is random, so the peak signal-to-noise ratio (PSNR) and the structural similarity (SSIM) of the three turbulence intensity degradation images are obtained by averaging. The calculation results are shown in Table 1. After using the present application, the PSNR under weak, medium and strong turbulence intensity is increased by 2.26 dB, 0.98 dB and 1.08 dB, respectively, and the SSIM is increased by 10%, 14% and 22%, respectively. The data performance is consistent with that in Figure 6 、 Figure 7 and Figure 8 The image visual performance shows that under different turbulence intensities, the present application still effectively reduces the turbulence distortion and improves the imaging quality.

[0041] Table 1 Comparison of simulation results of atmospheric turbulence imaging resistance The second example of the embodiment of the present application is analyzed. In the verification of the field experiment, the turbulence image acquisition device is an industrial camera. The resolution of the camera is 1024x1280 pixels, and only 300x300 pixels are used in the experiment, with each pixel size being 5 μm. The camera is equipped with a manual zoom lens with a focal length range of 5-50 mm, and the lens focal length is 40 mm in the experiment. The experimental data acquisition is performed on the roof of a certain residential area, and the imaging distance is 800 m. The data acquisition experiment uses a 5x5 camera array to acquire data, and a total of 25 turbulence images are acquired in 25 fields of view. The target object image under the influence of turbulence acquired in the first field of view is shown in Figure 9 As shown, the picture is blurred and the target object shape can be roughly seen; the target object image under the influence of turbulence acquired in the 25th field of view is shown in Figure 10 As shown, the target object image is similar to Figure 9 From the above two target object images, it can be seen that the turbulence effect is very significant. The multi-view affine transformation registration model is established, and the 25 field of view images of the target object are registered by affine transformation to obtain a 25 field of view image sequence of the target object. The registered target object image in the first field of view is shown in Figure 11The affine transformation result of the target object image under the influence of turbulence collected by the 1st field of view is shown. Figure 12 The affine transformation result of the target object image under the influence of turbulence collected by the 25th field of view is shown. Figure 13 The second case processed by using the lucky imaging method is shown, which is the anti-turbulence image result recovered by using the lucky imaging method. Figure 14 The second case processed by the target object camera array imaging method based on multi-field-of-view deconvolution anti-turbulence is shown, and the antenna distortion phenomenon is corrected after anti-turbulence reconstruction, the blur phenomenon is enhanced in detail, which embodies the advancement of the algorithm, the potential and robustness in practical application.

[0042] The beneficial effects of the present application are: the present application establishes an imaging system to perform multi-field-of-view imaging on a target object to obtain a multi-field-of-view image of the target object; a multi-field-of-view affine transformation registration model is established, and the multi-field-of-view image of the target object is registered by affine transformation to obtain a plurality of same-field-of-view image sequences of the target object; an anti-turbulence imaging algorithm based on multi-step deconvolution is established, which first deconvolves the image sequence using a convolution kernel of a preset optical system point spread function, enhances the target contrast, suppresses the out-of-focus background noise, and preserves the in-focus signal; then the autocorrelation function of each frame of image is calculated, the deconvolution iteration number is calculated, and the convolution kernel is automatically estimated and updated through Fourier ring correlation, and the deconvolution is performed again to further improve the resolution of the image, realizing anti-turbulence imaging. In the embodiment, compared with the typical anti-turbulence imaging algorithm, the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) image quality evaluation indicators are improved, which has more obvious advantages, and significantly improves the image quality in the turbulence environment, providing an effective technical solution for atmospheric turbulence imaging detection. The present application can realize target object image reconstruction in atmospheric turbulence without calibration, significantly solving the problem of traditional imaging methods affected by turbulence, significantly improving the target object imaging quality in atmospheric turbulence environment, and being also applicable in haze scattering medium.

[0043] The above-described embodiments are only preferred embodiments of the present application, and do not limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements to the technical solutions of the present application made by those skilled in the art shall fall within the protection scope of the claims of the present application.

Claims

1. A target object camera array imaging method based on multi-field deconvolution to resist turbulence, characterized in that: It includes: S1: Establish a physical model of how atmospheric turbulence affects the imaging quality of the target object, build a camera array imaging system, determine the optical transfer function of turbulence affecting the imaging of the target object, and obtain the image of the target object under turbulent conditions; S2: Establish a multi-field affine transformation registration model. Through affine transformation, register the target object image obtained in step S1 under turbulent conditions to obtain a sequence of images of the target object in the same field of view. S3: Perform adaptive pre-deconvolution processing on the same field-of-view image sequence of the target object obtained in step S2. Perform Fourier ring correlation (FRC) operation on every two target object images. Select half of the number of iterations corresponding to the lowest point value of the Fourier ring correlation (FRC) as the number of adaptive pre-deconvolution iterations k1 using the preset optical system point spread function convolution kernel. This enhances the contrast of the target object image and preserves the focal plane signal. Specifically: ； in, Estimate the image of the target object after the (i+1)th iteration; Estimate the image of the target object after the i-th iteration; For camera array imaging system functions; To degrade the target object image; This is a convolution operation; The x-coordinate of the two-dimensional image; Let be the ordinate of the two-dimensional image; i is a non-negative integer. ; Iteration initial conditions The output of k1 adaptive pre-deconvolutions is derived from multiple images of the target object under turbulence output in step S2, all within the same field of view. To enhance the contrast of the target object image; S4: Perform adaptive deconvolution processing on the enhanced contrast target object image obtained in step S3, which retains the focal plane signal. Perform Fourier ring correlation (FRC) operation again on every two target object images. Select half of the number of iterations corresponding to the lowest point of the Fourier ring correlation (FRC) as the number of adaptive deconvolution iterations k2 to obtain the best predicted image of the target object image under turbulent conditions. This predicted image is the recovered image of the anti-turbulence target object, specifically: ； Initialization of iteration The enhanced contrast target object image with preserved focal plane signal output from step S3 is subjected to k2 adaptive deconvolutions to output the best predicted image under turbulence for restoring the target object image. .

2. The target object camera array imaging method based on multi-field deconvolution anti-turbulence as described in claim 1, characterized in that: Step S3 is as follows: S31: Deconvolve the same-field-of-view image sequence of the target object using a convolution kernel with a preset optical system point spread function to enhance contrast and degrade the target object image. and original clear target object image and camera array imaging system functions The distribution between them follows a Poisson distribution; S32: Obtain the deconvolution iterative equation of the deconvolution algorithm according to step S31, and use Fourier ring correlation (FRC) to evaluate the resolution of the reconstructed image in each iteration; S33: Set the initial conditions for the iteration. Degraded target object image After k1 iterations, the estimated image of the degraded target object is: .

3. The target object camera array imaging method based on multi-field deconvolution anti-turbulence as described in claim 2, characterized in that: In step S31, the convolution kernel of the point spread function of the optical system is preset as follows: ； ； in, The convolution kernel is the point spread function of the preset optical system; Let C be the point amplitude function; C is the normalization constant related to light intensity. Let be the wavenumber of the wavefront function; The aperture is the wavefront function. The focal length is the wavefront function. Normalized x-axis; The ordinate is the normalized vertical axis; These are the parameter variables of the wavefront function; The parameter is an imaginary unit.

4. The target object camera array imaging method based on multi-field deconvolution anti-turbulence as described in claim 2, characterized in that: Degraded target object image in step S31 and original clear target object image and camera array imaging system functions Specifically, the Poisson distribution between them is as follows: ； in, To degrade the target object image; For original, clear images of the target object; Let be the probability density function of the Poisson distribution; This is a factorial operation.

5. The target object camera array imaging method based on multi-field deconvolution anti-turbulence as described in claim 2, characterized in that: The formula for calculating the Fourier ring correlation FRC in step S32 is as follows: ； in, The Fourier ring correlation parameters of the target object image in the frequency domain; The Fourier transform of the first image; The Fourier transform of the second image; This is the k-th frequency ring; To calculate the radius of the ring containing the frequency value of the selected pixel; Fourier transform of the second image The conjugate Fourier transform of ; k is the frequency ring number.

6. The target object camera array imaging method based on multi-field deconvolution anti-turbulence as described in claim 2, characterized in that: Step S33 is as follows: The initial condition for iteration is The output of step S2 consists of multiple images of the target object under turbulence with the same field of view. Then, step S32 is used for iterative solution. As the number of iterations increases, the estimated image of the target object after the (k1+k2)th iteration is obtained. It will eventually converge to the original, clear image of the target object. This allows for the recovery of the original clear image of the target object, resulting in anti-turbulence imaging.

7. The target object camera array imaging method based on multi-field deconvolution anti-turbulence as described in claim 1, characterized in that: The camera array imaging system constructed in step S1 includes a camera array, an optical imaging lens, and a target object. The camera array consists of multiple sets of charge-coupled cameras (CCDs) arranged in a regular pattern to form a detector. The sunlight field shines on the surface of the target object through atmospheric turbulence. The light reflected from the target object passes through atmospheric turbulence and is received by the optical imaging lens, then enters the camera array for imaging and detection, realizing multi-field imaging of the target object and obtaining multi-field images of the target object. These images are then registered into a sequence of multiple images of the target object with the same field of view through affine transformation. The camera focal plane transformation relationship in the camera array is a translation transformation relationship or a matrix transformation relationship.

8. The target object camera array imaging method based on multi-field deconvolution anti-turbulence as described in claim 7, characterized in that: Given the translation transformation relationship of the camera array, the coordinates of object P in the first camera are... and the coordinates of object P in the second camera The transformation relationship is as follows: ； in, The x-coordinate of the target object in the second camera; The x-coordinate of the target object in the first camera; This represents the horizontal coordinate transformation relationship between the first camera and the second camera. The vertical coordinate of the target object in the second camera; The vertical coordinate of the target object in the first camera; This represents the ordinate transformation relationship between the first camera and the second camera.

9. The target object camera array imaging method based on multi-field deconvolution anti-turbulence as described in claim 7, characterized in that: Given the matrix transformation relationship of the camera array, the coordinates of object P in the first camera are... and the coordinates of object P in the second camera The transformation relationship is as follows: 。 10. The target object camera array imaging method based on multi-field deconvolution anti-turbulence as described in claim 1, characterized in that: The physical model for how air turbulence affects the imaging quality of the target object in step S1 is as follows: ； in, The light field distribution on the surface of a charge-coupled camera (CCD). Let reflectance be the surface reflectance function of the object. The optical point spread function is given from the light source to the target object to the detector, the charge-coupled camera (CCD). The optical point spread function is the distance from the target object to the detector charge-coupled camera CCD. For the reflection data of the target object; The light field distribution on the surface of the target object; Data on solar light sources; Derivatives of input data for the light field; This is the derivative of the solar light source data.

Images