Atmospheric turbulence degradation simulation method

Abstract

The invention discloses an atmospheric turbulence degradation simulation method. The method comprises the following steps: calculating Zernike coefficients under different temperatures and air pressures; calculating the spatial correlation of the Zernike coefficient, and performing tilt processing on the input image according to the spatial correlation to obtain a distorted image; constructing a non-tilt point spread function data set based on a Zernike coefficient; and based on the non-tilt point spread function data set and the Zernike coefficient, performing fuzzy processing under the turbulence influence on the distorted image by using a preset lightweight network based on a channel attention mechanism to obtain a turbulence degraded image. According to the method, a Tatarski formula is introduced into a Kolmogorov turbulence model, so that the purpose of image degradation simulation by atmospheric turbulence at different temperatures and air pressures is achieved; and by adopting a processing sequence of firstly inclining and then blurring, the shape integrity of blurring is reserved, and the actual turbulence degradation effect can be described more accurately.

Description

Technical Field

[0001] The present invention belongs to the technical field of meteorology, and particularly relates to a method for simulating atmospheric turbulence degradation. Background Art

[0002] Atmospheric turbulence is one of the main challenges in long-range imaging applications in astronomy, surveillance, and navigation. The medium in the atmosphere is non-uniform. Due to changes in temperature, pressure, etc., the refractive index of the medium in the atmosphere will change randomly. The random perturbation of atmospheric turbulence will change the properties of the light field transmitted through the atmosphere. Imaging of a distant target through atmospheric turbulence will result in blurred and distorted images. Therefore, simulating the image degradation caused by atmospheric turbulence is the key to high-quality reconstruction of turbulence-degraded images.

[0003] With the in-depth study of turbulence models and the promotion of emerging technologies, the simulation of turbulence degradation has seen rapid development. In the technical solutions of the prior art, a turbulence degradation simulation method based on the statistical characteristics of the turbulence field is designed. This method uses a statistical model to describe the characteristics of the turbulence field, thereby realizing the simulation of the turbulence degradation effect. First, calculate the statistical characteristics of the turbulence field, such as turbulence intensity, turbulence scale, etc. Secondly, use the statistical model to calculate the point spread function (PSF). Then, use the PSF to calculate the intensity distribution of the light after passing through the turbulence field. Finally, use the calculated intensity distribution to generate the image after turbulence degradation. This method can quickly realize the simulation of turbulence degradation without complex flow field calculations and a large amount of data processing. The simulation effect of this solution is not very good. Therefore, the reconstruction quality in the training of the subsequent high-quality reconstruction model for turbulence-degraded images is not very good. It lacks the physical characteristics of atmospheric turbulence, and the existing models cannot accurately describe the impact of atmospheric turbulence on images. Summary of the Invention

[0004] In order to solve the above problems existing in the prior art, the present invention provides a method for simulating atmospheric turbulence degradation. The technical problems to be solved by the present invention are realized through the following technical solutions:

[0005] The present invention provides a method for simulating atmospheric turbulence degradation, and the method includes:

[0006] Introduce the Tatarski formula into the Kolmogorov turbulence model to calculate the Zernike coefficients at different temperatures and pressures;

[0007] Calculate the spatial correlation of the Zernike coefficients, and perform tilt processing on the input image according to the spatial correlation to obtain a distorted image;

[0008] Construct a tilt-free point spread function data set based on the Zernike coefficients;

[0009] Based on the non-tilted point spread function dataset and the Zernike coefficients, the distorted image is subjected to blurring under the influence of turbulence by using a preset lightweight network based on the channel attention mechanism to obtain a turbulence-degraded image.

[0010] In one embodiment of the present invention, the Tatarski formula is introduced into the Kolmogorov turbulence model to calculate the Zernike coefficients at different temperatures and air pressures, including:

[0011] According to the Zernike polynomial and the Kolmogorov turbulence model introduced with the Tatarski formula, calculate the 36th-order Zernike coefficients at different temperatures and air pressures; where,

[0012] The phase structure function in the Kolmogorov turbulence model is as follows:

[0013]

[0014] and represent the position coordinates of any two points in the atmospheric turbulence field, r 0 represents the atmospheric coherence length, and the atmospheric coherence length r 0 has the following expression:

[0015]

[0016] k represents the wave number, γ represents the zenith angle, L represents the total propagation distance, z represents the propagation distance, represents the atmospheric refractive index structure constant.

[0017] In one embodiment of the present invention, according to the Tatarski formula, the expression of the atmospheric refractive index structure constant is as follows:

[0018]

[0019] where, L 0 represents the outer scale of turbulence, T represents the absolute temperature, P represents the air pressure, γ a represents the dry air adiabatic lapse rate, h represents the height, and the expression of the Coulman mode of the outer scale of turbulence L 0 is as follows:

[0020]

[0021] In one embodiment of the present invention, the expression of the Zernike coefficient is as follows:

[0022]

[0023] Among them, a j represents the j-th order Zernike coefficient, W represents the normalized window function, ρ = [τ, θ] T , τ represents the scaling coefficient, θ represents the polar angle, R represents the aperture radius of the camera optical system, and Z j represents the j-th order Zernike polynomial.

[0024] In an embodiment of the present invention, calculating the spatial correlation of Zernike coefficients includes:

[0025] Based on the second-order and third-order Zernike coefficients in the Zernike coefficients, the covariance corresponding to the second-order and third-order Zernike coefficients is obtained according to the Noll theory;

[0026] Based on the expected value of the corresponding product evaluated by two different apertures and the covariance corresponding to the second-order and third-order Zernike coefficients, the spatial correlation of the Zernike coefficients is obtained by using the first formula; wherein, the first formula is as follows:

[0027] C 2 (ξ) = C 3 (ξ) = I 0 (s) + I 2 (s);

[0028] C 2 (ξ) represents the spatial correlation corresponding to the second-order Zernike coefficient, and C 3 (ξ) represents the spatial correlation corresponding to the third-order Zernike coefficient, and I 0 (s) and I 2 (s) are defined by Bessel functions.

[0029] In an embodiment of the present invention, according to the spatial correlation, extracting the tilt and performing tilt processing on the input image to obtain a distorted image includes:

[0030] Based on the spatial correlation corresponding to the second-order and third-order Zernike coefficients, according to the relationship between the tilt angle and the Zernike coefficient, the horizontal tilt angle corresponding to the second-order Zernike coefficient and the vertical tilt angle corresponding to the third-order Zernike coefficient are obtained;

[0031] Using 2D Fourier transform to process the horizontal tilt angle and the vertical tilt angle, and respectively obtaining the power spectral density corresponding to the horizontal tilt angle and the vertical tilt angle;

[0032] Multiply the power spectral densities corresponding to the horizontal tilt angle and the vertical tilt angle by white noise respectively to obtain the motion vector fields in the x direction and the y direction respectively;

[0033] Perform motion compensation on the input image using the motion vector fields in the x direction and the y direction to obtain the distorted image.

[0034] In an embodiment of the present invention, constructing a non-tilted point spread function dataset from Zernike coefficients includes:

[0035] Obtain the corresponding random phase according to the 4th - 36th order Zernike coefficients in the Zernike coefficients;

[0036] Multiply the random phase by the initial wavefront to obtain a complex wavefront;

[0037] Perform an inverse Fourier transform on the complex wavefront to obtain a non-tilted point spread function;

[0038] Obtain several groups of non-tilted point spread functions according to the Zernike coefficients under different temperatures and air pressures as the non-tilted point spread function dataset.

[0039] In an embodiment of the present invention, a preset lightweight network based on a channel attention mechanism includes:

[0040] A channel attention mechanism and three fully connected layers connected in sequence; wherein,

[0041] The channel attention mechanism includes a global average pooling layer and two fully connected layers.

[0042] In an embodiment of the present invention, the loss function of the preset lightweight network based on the channel attention mechanism includes a mean square error function; wherein,

[0043] The expression of the mean square error function is as follows:

[0044]

[0045] Among them, α represents the total number of basis coefficients, y μ represents the μth predicted basis coefficient, represents the μth true basis coefficient.

[0046] In an embodiment of the present invention, based on the non-tilted point spread function dataset and the Zernike coefficients, use the preset lightweight network based on the channel attention mechanism to perform blurring processing on the distorted image under the influence of turbulence to obtain a turbulence-degraded image, including:

[0047] Perform principal component analysis on the non-tilted point spread function dataset to obtain the basis functions of the point spread function;

[0048] Use the preset lightweight network based on the channel attention mechanism to adjust the weights of the channels corresponding to the 4th - 36th order Zernike coefficients in the Zernike coefficients to obtain the basis coefficients of the point spread function;

[0049] Multiply the basis functions corresponding to the point spread function by the basis coefficients, and perform blurring processing on the distorted image to obtain a turbulence - degraded image.

[0050] Advantages of the present invention:

[0051] In the solution provided by the present invention, aiming at the problem that current atmospheric turbulence simulation only simulates the randomness of turbulence through statistical models but lacks the characteristics of turbulence, the Tatarski formula is introduced into the Kolmogorov turbulence model to achieve the purpose of simulating the degradation of images by atmospheric turbulence under different temperatures and pressures; the processing sequence of tilting first and then blurring is adopted, which retains the shape integrity of the blur and can more accurately describe the actual turbulence degradation effect; further, a channel attention mechanism is added to the preset lightweight network of the present invention to dynamically adjust the weights of the channels corresponding to each Zernike coefficient, suppressing the secondary Zernike coefficient channels, thereby improving the performance of the network. Description of the drawings

[0052] Figure 1 It is a schematic diagram of the steps of an atmospheric turbulence degradation simulation method provided by an embodiment of the present invention;

[0053] Figure 2 It is a schematic flow diagram of an atmospheric turbulence degradation simulation method provided by an embodiment of the present invention;

[0054] Figure 3 It is a schematic structural diagram of a preset lightweight network based on the channel attention mechanism in an atmospheric turbulence degradation simulation method provided by an embodiment of the present invention;

[0055] Figure 4 It is a schematic diagram of the principle of the channel attention mechanism in a preset lightweight network based on the channel attention mechanism provided by an embodiment of the present invention;

[0056] Figure 5 It is a schematic comparison diagram of the input - output results of an atmospheric turbulence degradation simulation method provided by an embodiment of the present invention. Detailed implementation manners

[0057] The following further describes the present invention in detail with specific embodiments, but the implementation manners of the present invention are not limited thereto.

[0058] To achieve the purpose of simulating atmospheric turbulence to generate degraded images, an embodiment of the present invention provides an atmospheric turbulence degradation simulation method, as Figure 1 shown, which may include:

[0059] S1. Introduce the Tatarski formula into the Kolmogorov turbulence model, and calculate the Zernike coefficients at different temperatures and pressures;

[0060] S2. Calculate the spatial correlation of the Zernike coefficients, extract the tilt according to the spatial correlation, and perform tilt processing on the input image to obtain a distorted image;

[0061] S3. Construct a non-tilted point spread function data set based on the Zernike coefficients;

[0062] S4. Based on the non-tilted point spread function data set and the Zernike coefficients, use a preset lightweight network based on the channel attention mechanism to perform blurring processing on the distorted image under the influence of turbulence to obtain a turbulence degraded image.

[0063] The main purpose of the atmospheric turbulence degradation simulation method provided by the embodiment of the present invention is to simulate atmospheric turbulence to generate degraded images, and provide a synthetic large-scale data set for high-quality reconstruction of turbulence degraded images based on deep learning. By simulating the behavior of atmospheric turbulence, the present invention can generate a large number of degraded images, which can be used to train deep learning models, thereby achieving high-quality reconstruction of turbulence degraded images. This method can provide important data support for research and applications in fields such as optical engineering and atmospheric physics.

[0064] For the flow schematic diagram of the atmospheric turbulence degradation simulation method provided by the embodiment of the present invention, please refer to Figure 2 , for ease of understanding, the following will combine Figure 1 and Figure 2 , and introduce each step of the atmospheric turbulence degradation simulation method provided by the embodiment of the present invention separately.

[0065] For S1, it may include:

[0066] According to the Zernike polynomial and the Kolmogorov turbulence model introduced with the Tatarski formula, calculate the 36th-order Zernike coefficients at different temperatures and pressures; where

[0067] Atmospheric turbulence can be regarded as composed of many vortices, and each vortex has its refractive index characteristic value. For the Kolmogorov turbulence model, the phase structure function can be used to describe the change of the phase at different spatial scales. The phase structure function in the Kolmogorov turbulence model is as follows:

[0068]

[0069] and represent the position coordinates of any two points in the atmospheric turbulence field. represents and the distance between two points, r 0 represents the atmospheric coherence length, and the atmospheric coherence length can be the characteristic length of Kolmogorov turbulence. The atmospheric coherence length r 0 has the following expression:

[0070]

[0071] k represents the wave number, γ represents the zenith angle, L represents the total propagation distance, z represents the propagation distance, represents the atmospheric refractive index structure constant.

[0072] The atmospheric coherence length is used to represent the intensity of the perturbation of the atmospheric turbulence on the optical wavefront. Since the perturbation intensity of the atmospheric turbulence cannot be directly measured, r 0 is usually used as an approximation. When r 0 is less than the diffraction limit of the optical system, the atmosphere will affect the imaging performance of the optical system. The smaller r 0 , the more serious the influence of the turbulence. It can be understood that the atmospheric coherence length is a parameter with practical significance. In the expression of the atmospheric coherence length r 0 , according to the Tatarski formula, the expression of the atmospheric refractive index structure constant is as follows:

[0073]

[0074] where, L 0 represents the outer scale of turbulence, T represents the absolute temperature, P represents the air pressure, γ a represents the dry air adiabatic lapse rate, h represents the height, and the expression of the Coulman mode of the outer scale of turbulence L 0 is as follows:

[0075]

[0076] The dry air adiabatic lapse rate γ a can take 9.8×10 -3 Km -1 .

[0077] It can be understood that the atmospheric refractive index structure constant is a measure of the turbulence intensity. At night and in the early morning, due to the low heat of the sun, has a small value. At noon, becomes unstable. This is mainly because the heat of the sun causes the hot air to rise, and the encounter with the descending cold air creates irregularities in the movement. describes the magnitude of the fluctuations of the atmospheric refractive index with space and time, represents the average value of the squared change amplitude of the atmospheric refractive index within a certain distance range, and is an important parameter for describing atmospheric turbulence. According to the Tatarski formula, the intensity of atmospheric turbulence can be changed by changing the values of temperature and pressure, so as to achieve the purpose of simulating the turbulence degradation under different temperatures and pressures. Aiming at the problem that the current atmospheric turbulence simulation only simulates the randomness of turbulence through statistical models but lacks the characteristics of turbulence, the Tatarski formula is introduced into the Kolmogorov turbulence model to achieve the purpose of simulating the degradation of images by atmospheric turbulence under different temperatures and pressures.

[0078] The expression of the Zernike coefficient is as follows:

[0079]

[0080] where a j represents the j-th order Zernike coefficient, W represents the normalized window function, ρ = [τ, θ] T , τ represents the scaling coefficient, θ represents the polar angle, R represents the aperture radius of the camera optical system, and Z j represents the j-th order Zernike polynomial.

[0081] Specifically, can be defined as the Zernike basis, where Z j represents the j-th order Zernike polynomial, and is defined as the Zernike coefficient, where a j represents the j-th order Zernike coefficient, and the phase can be expressed as: where R represents the aperture radius of the camera optical system, ρ = [τ, θ] T , τ represents the scaling coefficient, and θ represents the polar angle.

[0082] It can be understood that each Zernike coefficient has its own geometric definition and can be determined by orthogonality. The correlation between a j and a j′ is the inter-mode correlation, where a j and a j′ are any of the Zernike coefficients, and a j ≠a j′, the inter-mode correlation can be expressed by the expectation between these two coefficients, and the expectation between the two coefficients is as follows:

[0083]

[0084] where ρ and ρ′ are respectively the matrices composed of the scaling coefficient and the polar angle corresponding to a j and a j′ , Z j and Z j′ are the Zernike polynomials corresponding to the coefficient a j and a j′ . The expectation within the double integral is the autocorrelation of the phase, which can be replaced by the structure function. Based on the Wiener-Khinchin theorem, it can be obtained that:

[0085]

[0086] where ζ represents the integration variable, m and m′ are respectively the azimuthal frequencies corresponding to a j and a j′ , n and n′ are respectively the radial degrees corresponding to a j and a j′ . The closed-form expression of the integral can be obtained by integrating the Bessel function.

[0087] Form a covariance matrix C from the set of correlations of multiple Zernike coefficients. To extract relevant samples from this covariance matrix, use Cholesky decomposition to decompose the covariance matrix C into C = UU T , and utilize Gaussian white noise b to obtain the transformed vector a as a = Ub, which has the property of E[aa T = C, thereby obtaining a set of Zernike coefficients with inter-mode correlation. Among them, U T is the transpose of U, and a T is the transpose of a.

[0088] It can be understood that by step S1, the behavior of atmospheric turbulence can be simulated, and the Zernike coefficients describing the characteristics of the turbulence can be obtained.

[0089] For S2, it can include:

[0090] S21, calculating the spatial correlation of the Zernike coefficients.

[0091] Specifically, S21 can include:

[0092] S211, based on the 2nd and 3rd order Zernike coefficients in the Zernike coefficients, obtain the covariance corresponding to the 2nd and 3rd order Zernike coefficients according to the Noll theory.

[0093] Since all Zernike coefficients are Gaussian distributed with a mean of zero, the covariance corresponding to the second and third order Zernike coefficients is obtained according to Noll's theory as follows:

[0094]

[0095] where W represents the normalized window function, ρ and ρ′ are respectively a j and a j′ corresponding matrix composed of the scaling coefficient and the polar angle, θ and θ′ are respectively a j and a j′ corresponding polar angles, which vanish outside the aperture (so that all integrals are over all space), represents the phase covariance function related to Kolmogorov turbulence.

[0096] S212, the expected value of the corresponding product evaluated based on two different apertures and the covariance corresponding to the second and third order Zernike coefficients, the spatial correlation of the Zernike coefficients is obtained using the first formula; where the first formula is as follows:

[0097] C 2 (ξ) = C 3 (ξ) = I 0 (s) + I 2 (s);

[0098] C 2 (ξ) represents the spatial correlation corresponding to the second order Zernike coefficient, C 3 (ξ) represents the spatial correlation corresponding to the third order Zernike coefficient, I 0 (s) and I 2 (s) are defined by Bessel functions.

[0099] Specifically, the above derivation yields the expected value of the product of two coefficients and a j′ evaluated on the same aperture. The expected value of the product evaluated on two different apertures is as follows:

[0100]

[0101] where Dξ represents the vector displacement from the center of the first aperture to the center of the second aperture, a parameter ξ is defined such that Dξ = L(θ - θ′), and D represents the separation aperture diameter.

[0102] For j ∈ {2, 3}, its spatial correlation is as follows:

[0103]

[0104] The above formula can be further transformed into:

[0105]

[0106] where i represents the imaginary unit, ψ represents the angle between ξ and ρ 0 , v 2 = 7.7554. When j takes 2, the corresponding sign in the above formula is "-", and when j takes 3, the corresponding sign in the above formula is "+", and ψ 0 represents the specific value when j = 2 or 3.

[0107] I 0 (s) and I 2 (s) are defined by Bessel functions, where

[0108]

[0109] The normalized spatial correlation of the Zernike coefficients is as follows:

[0110]

[0111] It can be understood that when j takes 2, the corresponding sign in the above formula is "-", and when j takes 3, the corresponding sign in the above formula is "+". In order to maintain the isotropy of the correlation function, the angle ψ 0 can be restricted to: ξ is always aligned with ρ - ρ′, and in this case, the first formula is obtained.

[0112] The first formula is as follows:

[0113] C 2 (ξ) = C 3 (ξ) = I 0 (s) + I 2 (s);

[0114] C 2 (ξ) represents the spatial correlation corresponding to the second-order Zernike coefficient, and C 3 (ξ) represents the spatial correlation corresponding to the third-order Zernike coefficient. I 0 (s) and I 2 (s) are defined by Bessel functions.

[0115] S22. According to the spatial correlation, the tilt is extracted to perform tilt processing on the input image to obtain a distorted image.

[0116] For S22, it may include:

[0117] S221. Based on the spatial correlation corresponding to the second - order and third - order Zernike coefficients, according to the relationship between the tilt angle and the Zernike coefficients, obtain the horizontal - direction tilt angle corresponding to the second - order Zernike coefficient and the vertical - direction tilt angle corresponding to the third - order Zernike coefficient.

[0118] Specifically, the second - order Zernike coefficient can be used to generate the tilt angle in the horizontal direction, and the third - order Zernike coefficient can be used to generate the tilt angle in the vertical direction. Since appropriate scaling is required to draw the tilt, and the Nyquist sampling between two adjacent pixels in the object plane is Lλ / 2D, the relationship between the tilt angle and the Zernike coefficient is as follows:

[0119]

[0120]

[0121] Where L represents the total propagation distance, λ represents the wavelength, and D represents the aperture diameter.

[0122] Therefore, the non - normalized correlation coefficient of the tilt angle corresponding to the second - order Zernike coefficient is as follows:

[0123]

[0124] It can be understood that the non - normalized correlation coefficient of the tilt angle corresponding to the second - order Zernike coefficient is actually the covariance matrix, which represents the tilt angle in the horizontal direction. The way to obtain the tilt angle in the vertical direction is to use the same method for the third - order Zernike coefficient.

[0125] S222. Use 2D Fourier transform to process the tilt angles in the horizontal and vertical directions respectively, and obtain the power spectral densities corresponding to the tilt angles in the horizontal and vertical directions.

[0126] S223. Multiply the power spectral densities corresponding to the tilt angles in the horizontal and vertical directions by white noise respectively to obtain the motion vector fields in the x - direction and y - direction.

[0127] S224. Use the motion vector fields in the x - direction and y - direction to perform motion compensation on the input image to obtain the distorted image.

[0128] It can be understood that in step S2, the spatial correlation of the Zernike coefficients is calculated using the second - order and third - order Zernike coefficients, and the covariance matrix is set using the spatial correlation of the Zernike coefficients. The covariance matrix is used to extract random tilts and apply tilt processing to the image to obtain the distorted image.

[0129] For S3, constructing a dataset of non-tilted point spread functions based on Zernike coefficients may include:

[0130] S31, obtaining corresponding random phases according to the 4th - 36th order Zernike coefficients in the Zernike coefficients.

[0131] S32, multiplying the random phase by the initial wavefront to obtain a complex wavefront.

[0132] Specifically, the expression of the complex wavefront U is as follows:

[0133]

[0134] where i represents the imaginary unit.

[0135] S33, performing an inverse Fourier transform on the complex wavefront to obtain a non-tilted point spread function.

[0136] S34, obtaining several groups of non-tilted point spread functions according to the Zernike coefficients under different temperatures and pressures as the dataset of non-tilted point spread functions.

[0137] For S4, it may include:

[0138] S41, performing principal component analysis on the dataset of non-tilted point spread functions to obtain the basis functions of the point spread function.

[0139] Specifically, the observed image can be expressed as:

[0140]

[0141] where x represents the distorted image obtained by tilt processing, and q 1 ,…,q O represents O spatially varying point spread functions stored as the linear operator Q.

[0142] q o The expression of is as follows:

[0143]

[0144] where represents the basis function of the point spread function, and the coefficient β m,o represents the m-th basis of the o-th pixel.

[0145] Therefore, each pixel y o can be rewritten as:

[0146]

[0147] Therefore, as long as the basis functions and basis coefficients of the point spread function are obtained, the image can be blurred under the influence of turbulence.

[0148] S42. Use a preset lightweight network based on the channel attention mechanism to adjust the weights of the channels corresponding to the 4th - 36th order Zernike coefficients in the Zernike coefficients, so as to obtain the basis coefficients of the point spread function.

[0149] The preset lightweight network based on the channel attention mechanism, such as Figure 3 shown, may include:

[0150] The channel attention mechanism and three fully connected layers connected in sequence; among them,

[0151] The channel attention mechanism includes a global average pooling layer and two fully connected layers.

[0152] Specifically, different Zernike coefficients correspond to different aberration types. When multiple Zernike coefficients act on the image simultaneously, their cumulative effect will cause a complex blurring effect. Therefore, in the embodiments of the present invention, a channel attention mechanism is added to the lightweight network of deep learning to focus on the importance of the channels corresponding to each Zernike coefficient, dynamically adjust the weights of each channel, so that the lightweight network of deep learning can focus on the features more critical to the task, thereby improving the performance of the lightweight network.

[0153] For the schematic diagram of the channel attention mechanism, please refer to Figure 4 , it can be understood that in the channel attention mechanism, first, the global average pooling layer is used to perform global spatial compression on all pixel points of each channel of the input feature map corresponding to the 4th - 36th order Zernike coefficients, and compress the G×B feature map corresponding to each channel into a scalar of 1*1*R as the global statistical information of each channel. Then, two fully connected layers are used to perform non - linear mapping on the compressed global statistical information to generate the attention weights of each channel. Finally, the generated attention weights are weighted with the input feature map, so that important channels are enhanced and secondary channels are suppressed.

[0154] The preset lightweight network based on the channel attention mechanism, such as Figure 3 shown, is mainly composed of a channel attention mechanism and three fully connected layers. The input dimension is 33, and the output dimension is 100. Using the 4th - 36th order Zernike coefficients as the input, 100 basis coefficients are output by the three fully connected layers in the network. The loss function of the preset lightweight network based on the channel attention mechanism includes the mean square error function; among them,

[0155] The expression of the mean square error function is as follows:

[0156]

[0157] Among them, α represents the total number of basis coefficients, and y μ represents the μ-th predicted basis coefficient, represents the μ-th true basis coefficient.

[0158] S43. Multiply the basis function corresponding to the point spread function by the basis coefficients to blur the distorted image and obtain a turbulent degradation image.

[0159] It can be understood that the embodiment of the present invention adopts a processing sequence of tilting first and then blurring, retains the shape integrity of the blur, and can more accurately describe the actual turbulent degradation effect.

[0160] For a comparison schematic diagram of the input and output results of an atmospheric turbulence degradation simulation method provided by an embodiment of the present invention, please refer to Figure 5 , By executing steps S1 - S4, rapid simulation of atmospheric turbulence degradation is realized, and high-quality turbulent degradation images are provided. This atmospheric turbulence degradation simulation method can provide important data support for research and applications in fields such as optical engineering and atmospheric physics.

[0161] In view of the problem that current atmospheric turbulence simulation only simulates the randomness of turbulence through a statistical model but lacks the characteristics of turbulence, the embodiment of the present invention introduces the Tatarski formula into the Kolmogorov turbulence model to achieve the purpose of simulating the degradation of images by atmospheric turbulence at different temperatures and pressures; adopts a processing sequence of tilting first and then blurring, retains the shape integrity of the blur, and can more accurately describe the actual turbulent degradation effect; furthermore, a channel attention mechanism is added to the preset lightweight network proposed by the present invention to dynamically adjust the weights of the channels corresponding to each Zernike coefficient, suppress the channels of secondary Zernike coefficients, thereby improving the performance of the network.

[0162] It should be noted that in the description of the present invention, it should be understood that the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of these features. In the description of the present invention, "a plurality" means two or more unless otherwise specifically defined.

[0163] The above are only the preferred embodiments of the present invention and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A method for simulating atmospheric turbulence degradation, characterized in that: include: The Tatarski formula is introduced into the Kolmogorov turbulence model to calculate the Zernike coefficients at different temperatures and pressures; Calculating the spatial correlation of Zernike coefficients, and performing tilt processing on the input image according to the spatial correlation to obtain a distorted image; constructing a non-tilted point spread function data set based on the Zernike coefficients; Based on the non-tilted point spread function data set and the Zernike coefficient, a preset lightweight network based on a channel attention mechanism is used to perform blur processing on the distorted image under the influence of turbulence to obtain a turbulence-degraded image.

2. The atmospheric turbulence degradation simulation method according to claim 1, characterized in that: The Tatarski formula is introduced into the Kolmogorov turbulence model to calculate the Zernike coefficients at different temperatures and pressures, including: According to the Zernike polynomial and the Kolmogorov turbulence model with the introduction of the Tatarski formula, the 36th-order Zernike coefficients at different temperatures and pressures are calculated; Phase structure function in the Kolmogorov turbulence model as follows: and represents the position coordinates of any two points in the atmospheric turbulence field, r0 represents the atmospheric coherence length, and the expression of the atmospheric coherence length r0 is as follows: k represents the wave number, γ represents the zenith angle, L represents the total propagation distance, and z represents the propagation distance. Represents the atmospheric refractive index structure constant.

3. The atmospheric turbulence degradation simulation method according to claim 2, characterized in that: According to the Tatarski formula, the atmospheric refractive index structure constant The expression is as follows: Where L0 represents the outer scale of turbulence, T represents absolute temperature, P represents air pressure, γ a represents the dry air adiabatic lapse rate, h represents the height, and the expression of the Coulman model of the turbulent outer scale L0 is as follows:

4. The atmospheric turbulence degradation simulation method according to claim 3, characterized in that: The expression of the Zernike coefficient is as follows: Among them, a j represents the jth order Zernike coefficient, W represents the normalized window function, ρ=[τ,θ] T , τ represents the zoom factor, θ represents the polar angle, R represents the aperture radius of the camera optical system, and Z j represents the j-th order Zernike polynomial.

5. The atmospheric turbulence degradation simulation method according to claim 1, characterized in that: The calculation of the spatial correlation of the Zernike coefficients includes: Based on the second and third order Zernike coefficients in the Zernike coefficients, the covariance corresponding to the second and third order Zernike coefficients is obtained according to Noll theory; Based on the expected value of the corresponding product of two different aperture evaluations and the covariance corresponding to the second-order and third-order Zernike coefficients, the spatial correlation of the Zernike coefficients is obtained using the first formula; wherein the first formula is as follows: C2(ξ)=C3(ξ)=I0(s)+I2(s); C2(ξ) represents the spatial correlation corresponding to the second-order Zernike coefficient, C3(ξ) represents the spatial correlation corresponding to the third-order Zernike coefficient, and I0(s) and I2(s) are defined by Bessel functions.

6. The atmospheric turbulence degradation simulation method according to claim 1, characterized in that: Extracting the tilt according to the spatial correlation performs tilt processing on the input image to obtain a distorted image, including: Based on the spatial correlation corresponding to the second-order and third-order Zernike coefficients and the relationship between the tilt angle and the Zernike coefficient, the horizontal tilt angle corresponding to the second-order Zernike coefficient and the vertical tilt angle corresponding to the third-order Zernike coefficient are obtained; The horizontal tilt angle and the vertical tilt angle are processed by 2D Fourier transform to obtain the power spectrum density corresponding to the horizontal tilt angle and the vertical tilt angle respectively; The power spectrum density corresponding to the horizontal tilt angle and the vertical tilt angle is multiplied by white noise to obtain the motion vector field in the x direction and the y direction respectively; The input image is motion compensated using the motion vector fields in the x direction and the y direction to obtain the distorted image.

7. The atmospheric turbulence degradation simulation method according to claim 1, characterized in that: A non-tilted point spread function data set is constructed based on the Zernike coefficients, including: Obtaining a corresponding random phase according to the 4th to 36th order Zernike coefficients in the Zernike coefficients; Multiplying the random phase with the initial wavefront to obtain a complex wavefront; performing an inverse Fourier transform on the complex wavefront to obtain a non-tilted point spread function; According to the Zernike coefficients at different temperatures and pressures, several groups of non-tilted point spread functions are obtained as the non-tilted point spread function data sets.

8. The atmospheric turbulence degradation simulation method according to claim 1, characterized in that: The preset lightweight network based on the channel attention mechanism includes: Sequentially connected channel attention mechanism and 3 fully connected layers; among them, The channel attention mechanism includes a global average pooling layer and two fully connected layers.

9. The atmospheric turbulence degradation simulation method according to claim 1, characterized in that: The preset loss function of the lightweight network based on the channel attention mechanism includes a mean square error function; wherein, The expression of the mean square error function is as follows: Among them, α represents the total number of basis coefficients, y μ represents the basis coefficient of the μth prediction, represents the μth real basis coefficient.

10. The atmospheric turbulence degradation simulation method according to claim 1, characterized in that: The method of performing blurring processing on the distorted image under the influence of turbulence using a preset lightweight network based on a channel attention mechanism based on the non-tilted point spread function data set and the Zernike coefficient to obtain a turbulence-degraded image includes: Performing principal component analysis on the non-tilted point spread function data set to obtain basis functions of the point spread function; Using the preset lightweight network based on the channel attention mechanism, the weights of the channels corresponding to the 4th to 36th order Zernike coefficients in the Zernike coefficients are adjusted to obtain the basis coefficients of the point spread function; The basis function and basis coefficient corresponding to the point spread function are multiplied together, and blurring is performed on the distorted image to obtain a turbulence-degraded image.

Images