A Multi-Layer Atmospheric Turbulence Reconstruction Method Based on Pupil Surface Complex Amplitude Detection

Abstract

This invention discloses a multi-layer atmospheric turbulence reconstruction method based on pupil plane complex amplitude detection, belonging to the field of multi-layer atmospheric turbulence detection technology. The method involves stratifying the atmosphere and simulating a multi-layer turbulent phase screen. The initial light field undergoes multiple diffraction transmissions to obtain pupil plane complex amplitude information on the final plane. Pupil plane intensity images and Shaker-Hartmann images are calculated to characterize pupil plane amplitude and phase information, respectively. An Atmospheric Turbulence 3D Reconstruction Network (AT-3DRN) is constructed, incorporating coordinate attention and linear attention mechanisms to enhance its modeling capabilities for local and global spatial information. AT-3DRN can accurately reconstruct multi-layer turbulent wavefront aberrations based on the input pupil plane intensity image and Shaker-Hartmann image. This invention achieves high-precision reconstruction of multi-layer turbulence, providing an effective solution for stratified correction and high-precision wavefront detection under strong turbulence.

Description

Technical Field

[0001] This invention belongs to the field of multi-layer atmospheric turbulence detection technology, specifically relating to a multi-layer atmospheric turbulence reconstruction method based on pupil plane complex amplitude detection. Background Technology

[0002] Ground-based optical telescopes inevitably suffer from atmospheric turbulence when conducting high-resolution observations of celestial targets. Atmospheric turbulence causes random fluctuations in the refractive index, resulting in wavefront distortion of incident light waves and severely reducing imaging resolution. Adaptive Optics (AO) technology, by detecting and correcting wavefront distortion in real time, can effectively overcome the effects of atmospheric turbulence and has become an indispensable key technology for large-aperture telescopes. However, when the turbulence intensity is strong, especially in the presence of upper-level turbulence, the light beam, after being propagated through multiple layers of turbulent diffraction to reach the pupil plane, not only experiences phase distortion but also causes significant uneven fluctuations in the intensity distribution at the pupil plane, i.e., intensity scintillation. This severely affects the wavefront detection accuracy of the AO system, thus causing it to malfunction.

[0003] To overcome the effects of scintillation, some researchers have used deep learning algorithms to recover the sub-aperture loss in Shaker-Hartmann wavefront detectors caused by scintillation through post-processing. Other studies have used additional correctors to correct pupil intensity and compensate for pupil intensity loss on top of traditional AO, achieving simultaneous correction of pupil phase and intensity. However, since actual atmospheric turbulence is distributed in multiple layers in space, the scintillation effect is a result of these multiple layers of turbulence, especially upper-level turbulence. To fundamentally solve the scintillation effect, it is necessary to reconstruct multiple aberrations and correct them layer by layer. The algorithms mentioned above do not fundamentally solve the scintillation effect and are therefore difficult to use for practical strong turbulence detection.

[0004] The Shaker-Hartmann wavefront sensor is currently the mainstream wavefront sensor in AO systems. It primarily characterizes the phase information of the pupil plane, but cannot effectively characterize the amplitude information. Under strong turbulence conditions, wavefront distortion exhibits severe nonlinearity and amplitude-phase coupling characteristics, making accurate reconstruction difficult relying solely on the local slope information provided by the Hartmann wavefront sensor. The pupil plane amplitude is jointly affected by the turbulence phase and altitude; introducing pupil plane light intensity information can provide additional amplitude information, thus enabling the reconstruction of multiple aberrations. Therefore, how to fully utilize pupil plane amplitude and phase information to achieve high-precision reconstruction of multi-layer atmospheric turbulence, fundamentally overcoming the influence of strong turbulence light intensity scintillation on wavefront detection, has become a pressing technical problem to be solved. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a multi-layer atmospheric turbulence reconstruction method based on pupil plane complex amplitude detection. It integrates pupil plane light intensity images and Shack-Hartmann images as inputs to a deep learning network, and reconstructs multi-layer atmospheric turbulence wavefronts by combining amplitude and phase information, thereby fundamentally solving the strong turbulence light intensity scintillation effect.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A multi-layer atmospheric turbulence reconstruction method based on pupil plane complex amplitude detection includes:

[0008] Step 1: Stratify the atmosphere and simulate multi-layered turbulent phase screens at different heights; diffract parallel light through the multi-layered turbulent phase screens to obtain pupil plane complex amplitude information; calculate the pupil plane light intensity image representing amplitude information and the Shaker-Hartmann image representing phase information based on the pupil plane complex amplitude information; construct a dataset using the pupil plane light intensity image and the Shaker-Hartmann image.

[0009] Step 2: Construct the Atmospheric Turbulence 3D Reconstruction Network AT-3DRN. In the AT-3DRN, a coordinate attention mechanism is introduced into the residual block to extract local spatial features, and two layers of linear attention mechanism are introduced at the back end of the residual block to model the global spatial dependency.

[0010] Step 3: Using the dataset as input, train the AT-3DRN to obtain the mapping relationship from pupil detection information to multi-layer turbulent phase screen wavefront aberrations, thereby reconstructing multi-layer turbulent wavefront aberrations based on the real-time acquired pupil light intensity image and Shaker-Hartmann image.

[0011] Furthermore, in step 1, the wavefront of the multi-layer turbulent phase screen is described by the Zernike coefficient, which conforms to the Kolmogorov distribution; the pupil plane complex amplitude information obtained after the parallel light is diffracted and transmitted through the multi-layer turbulent phase screen is obtained by sequentially convolving and multiplying the spatial impulse response of Fresnel diffraction with the phase of each layer of turbulence.

[0012] Furthermore, the multi-layer turbulent phase screen has seven layers, and the equivalent atmospheric coherence length of each phase screen from the lower to the upper layers ranges from 3 to 43 centimeters; the Zernike coefficient has an order from the third to the ninth order.

[0013] Furthermore, in step 2, the input dimension of the atmospheric turbulence 3D reconstruction network is a two-channel image, corresponding to the pupil light intensity image and the Shaker-Hartmann image respectively; the output of the atmospheric turbulence 3D reconstruction network is the Zernike coefficients of each order of the multi-layer turbulence phase screen.

[0014] Furthermore, in step 2, the atmospheric turbulence 3D reconstruction network AT-3DRN sequentially includes: a convolutional layer with a kernel size of 3×3, a max pooling layer with a size of 3×3 and a stride of 2, multiple cascaded residual blocks, and two linear attention mechanisms located at the rear end of the last residual block; wherein each residual block sequentially includes three convolutional blocks, and the coordinate attention mechanism is introduced between the second and third convolutional blocks.

[0015] Furthermore, in the two-layer linear attention mechanism, the first layer linear attention mechanism is used to aggregate global information to capture spatial dependencies across sub-apertures, and the second layer linear attention mechanism is used to adaptively reweight and refine features under the guidance of global information.

[0016] Furthermore, each residual block includes three convolutional blocks of sizes 1×1, 3×3, and 1×1, respectively.

[0017] Furthermore, in step 3, when training the AT-3DRN, the pupil light intensity image and the Shaker-Hartmann image are normalized, and the Adam optimizer is used to update the gradient using the root mean square error between the network output and the standard output as the loss function.

[0018] In a second aspect, the present invention provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned method for reconstructing multi-layer atmospheric turbulence based on pupil plane complex amplitude detection.

[0019] Thirdly, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enable the processor to implement the aforementioned method for reconstructing multi-layer atmospheric turbulence based on pupil plane complex amplitude detection.

[0020] The beneficial effects of this invention are as follows:

[0021] First, this invention achieves the joint utilization of pupil plane amplitude and phase information by fusing pupil plane intensity images and Shack-Hartmann probe images as input to a deep learning network. Compared to traditional methods that rely solely on Shack-Hartmann probe data, this invention fully exploits the turbulence information contained in the pupil plane intensity distribution under strong turbulence, effectively overcoming the problem of insufficient information from a single information source under strong turbulence conditions, and providing richer feature data for high-precision wavefront reconstruction.

[0022] Second, this invention directly reconstructs the wavefront aberrations of multi-layered atmospheric turbulence, rather than merely recovering the distorted wavefront of a single plane. By modeling the atmosphere in layers, the network learns the mapping relationship from pupil plane detection information to the wavefront aberrations of each phase screen, achieving accurate perception of the three-dimensional turbulence distribution. This characteristic allows for subsequent layered correction of turbulence, fundamentally solving the light intensity scintillation effect caused by strong turbulence from a physical mechanism perspective.

[0023] Third, the AT-3DRN constructed in this invention introduces an attention mechanism, enhancing the network's ability to extract key features such as non-uniform distribution of pupil light intensity and Hartmann spot distortion, while maintaining good reconstruction accuracy under strong turbulence conditions. Test results show that this invention can achieve accurate reconstruction of multi-layer wavefronts under various turbulence intensities, with small reconstruction residuals, meeting the accuracy requirements of precision optical systems for wavefront detection.

[0024] Fourth, this invention is based on an end-to-end deep learning framework, which can achieve millisecond-level wavefront reconstruction speed after training, meeting the real-time requirements of adaptive optics systems and providing reliable technical support for large field-of-view high-resolution imaging observation in strong turbulent environments. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of a multi-layer atmospheric turbulence reconstruction method based on pupil plane complex amplitude detection according to the present invention.

[0026] Figure 2 This is a schematic diagram of the AT-3DRN structure;

[0027] Figure 3 This is a graph showing a set of test set results obtained according to the method of the present invention. Detailed Implementation

[0028] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0029] like Figure 1 The diagram shown illustrates the principle of a multi-layer atmospheric turbulence reconstruction method based on pupil plane complex amplitude detection according to the present invention. Specifically, it includes:

[0030] Step 1: Stratify the atmosphere and simulate multi-layered turbulent phase screens at different heights; diffract parallel light through the multi-layered turbulent phase screens to obtain pupil plane complex amplitude information; calculate the pupil plane light intensity image representing amplitude information and the Shaker-Hartmann image representing phase information based on the pupil plane complex amplitude information; construct a dataset using the pupil plane light intensity image and the Shaker-Hartmann image.

[0031] Step 2: Construct the Atmospheric Turbulence 3D Reconstruction Network AT-3DRN, wherein a coordinate attention mechanism is introduced into the residual block of the AT-3DRN to extract local spatial features, and a linear attention mechanism is introduced at the back end of the residual block to model global spatial dependencies.

[0032] Step 3: Using the dataset as input, train the AT-3DRN to obtain the mapping relationship from pupil detection information to multi-layer turbulent phase screen wavefront aberrations, thereby reconstructing multi-layer turbulent wavefront aberrations based on the real-time acquired pupil light intensity image and Shaker-Hartmann image.

[0033] Furthermore, in step 1,

[0034] Assume the telescope aperture in the simulation The total atmospheric coherence length is 80 cm. The aperture is 3cm, and the Shaker-Hartmann sub-aperture arrangement is 27×27. In the case of strong turbulence, wavefront distortion exhibits severe nonlinearity and amplitude-phase coupling characteristics, making accurate reconstruction difficult relying solely on local slope information provided by a Hartmann wavefront sensor. Since pupil plane amplitude is jointly affected by turbulent phase and height, introducing pupil plane light intensity information can provide additional amplitude information, thereby enabling the reconstruction of multiple aberrations.

[0035] The atmosphere is stratified; in this example, it is divided into seven layers. The equivalent atmospheric coherence length of the phase screen for each layer, from the lower to the upper levels, is also considered. The range is from 3 to 43 cm, specifically 3.32, 15.21, 26.08, 40.67, 34.72, 34.25, and 42.09 cm. The multi-layered phase screens are set at different heights. Zernike coefficients conforming to the Kolmogorov distribution are randomly generated based on the set turbulence intensity to describe the phase of each phase screen.

[0036] ,

[0037] in, It is the first Wavefront of the phase screen, It is the highest Zernike mode order. These are the coefficients of each order of the Zernike model. These are the various Zernike modes. Represents the rectangular coordinates of the plane in which it is located.

[0038] Parallel light from the target object undergoes diffraction propagation through multiple phase screens, resulting in a distorted pupil plane complex amplitude.

[0039] ,

[0040] in, This indicates the total number of turbulent layers. Represents the complex amplitude distribution on the initial plane. Indicates the process Free space propagation and After subphase modulation, the complex amplitude distribution on the final plane This represents the rectangular coordinates of the plane in which it is located, with the subscript n corresponding to the nth turbulent layer, n∈[0,N]. This represents the spatial impulse response of Fresnel diffraction. This represents the turbulent phase of the nth layer. This represents the propagation distance of the light beam to reach the nth layer of turbulent phase. Indicates the height at which the turbulence phase of the nth layer is located. " " indicates convolution, and "·" indicates product. It represents the imaginary unit.

[0041] The pupil light intensity distribution image and Shack-Hartmann image are obtained by calculating the complex amplitude on the final plane.

[0042] Each dataset contains input data including pupil plane intensity images and Shack-Hartmann images, which represent the amplitude and phase information of the light field, respectively, and output data including Zernike coefficients of orders 3 to 90 from seven phase screens. A total of 50,000 datasets were generated, and the data were divided into training, validation, and test sets according to a ratio of 90%, 5%, and 5%, respectively. This allowed the network to learn the mapping relationship from pupil plane intensity images and Shack-Hartmann images to wavefront aberrations of multi-layer phase screen images. After training, the accuracy of the network was tested using the test set.

[0043] Furthermore, in step 2:

[0044] An Atmospheric turbulence three-dimensional reconstruction network (AT-3DRN) is constructed. The pupil light intensity image and the Shaker-Hartmann detector image are used as inputs, and the wavefront aberration of the multi-layer turbulent phase screen is used as output. The mapping relationship from the pupil detection information to the wavefront aberration of the multi-layer turbulent phase screen is obtained through training.

[0045] The nonlinear activation function in AT-3DRN is the ReLU function. Its inputs are the pupil intensity image and the Shak-Hartmann image, representing the amplitude and phase information in the final pupil complex amplitude, respectively. The input dimension is 2×300×300. After the pupil intensity image and the Shak-Hartmann image are input into the network, they first pass through a convolutional layer to extract features. The kernel size of this convolutional layer is 3×3. Then, a max pooling layer is used to downsample and filter significant features. The max pooling layer in the network is 3×3 in size with a stride of 2. The residual blocks, while extracting features, alleviate the gradient vanishing and model degradation problems during network training through residual connections. Each residual block used in the network contains three convolutional blocks with sizes of 1×1, 3×3, and 1×1, respectively. A coordinate attention mechanism is introduced between the second and third convolutional blocks to enhance the extraction of local sub-aperture and local intensity feature information. The network backend incorporates a two-layer linear attention mechanism. This mechanism aggregates information from the entire input sequence, modeling global spatial dependencies. This helps capture the overall spatial structural features of the wavefront, overcoming the limitation of convolutional layers that can only capture local information, thereby improving the representation ability of complex aberration patterns. Each layer of the two-layer linear attention mechanism has a different focus. The first layer enables the network to perceive global information, focusing on the aggregation of global information to capture spatial dependencies across sub-apertures. The second layer further fine-tunes the network based on the existing global information, adaptively reweighting and refining features under the guidance of global information, optimizing the network's ability to model overall wavefront aberration patterns, thus achieving accurate modeling of complex wavefront structures. The network outputs Zernike coefficients of orders 3 to 90 for seven phase screens through two fully connected layers, with an output dimension of 7×88. Figure 2 This is a schematic diagram of the AT-3DRN in this example.

[0046] Furthermore, in step 3:

[0047] The AT-3DRN was trained using a dataset. Before inputting the data into the network, the pupil light intensity images and Shack-Hartmann images were normalized to a value between 0 and 1. During training, the Adam optimizer was used, with the root mean square error (RMSE) between the network output and the standard output as the loss function, and the network parameters were updated using gradients based on this loss function. Training was completed when the loss function converged.

[0048] After training, the network's accuracy is tested using a test set. Figure 3The results of AT-3DRN on a set of test data are shown. Layers one through seven represent turbulence phase screens from low to high levels, with colors in the phase screens indicating phase magnitude. The results demonstrate that by combining amplitude and phase information, the network can accurately reconstruct multi-layered wavefronts of atmospheric turbulence based on pupil intensity images and Shack-Hartmann images. The predicted phase screens for each layer are close to the true phase screens, and the reconstruction residuals for each layer are relatively small. The peak-to-valley (PV) and root mean square (RMS) values ​​of each phase screen are effectively reduced, resulting in high reconstruction accuracy.

[0049] In practical applications, parameters such as telescope aperture, atmospheric coherence length, sub-aperture arrangement, number of atmospheric layers, and equivalent atmospheric coherence length of each layer can be adjusted according to the specific detection scenario and system configuration. Any parameter changes, equivalent substitutions, or adaptive modifications made by those skilled in the art based on the technical solutions disclosed in this invention for different turbulence intensities, different telescope apertures, or different layering schemes all fall within the protection scope of this invention.

[0050] In summary, this invention achieves the joint utilization of pupil plane amplitude and phase information by fusing pupil plane intensity images and Shaker-Hartmann detection images. The constructed atmospheric turbulence 3D reconstruction network incorporates coordinate attention and linear attention mechanisms, enabling accurate reconstruction of wavefront aberrations of multi-layered atmospheric turbulence from pupil plane detection information. This invention maintains good reconstruction accuracy under strong turbulence conditions, exhibits small reconstruction residuals, and boasts fast processing speed, meeting the real-time requirements of adaptive optics systems. It provides reliable technical support for large field-of-view, high-resolution imaging observations in strong turbulent environments.

[0051] In a second aspect, the present invention provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned method for reconstructing multi-layer atmospheric turbulence based on pupil plane complex amplitude detection.

[0052] Thirdly, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enable the processor to implement the aforementioned method for reconstructing multi-layer atmospheric turbulence based on pupil plane complex amplitude detection.

[0053] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-layer atmospheric turbulence reconstruction method based on pupil plane complex amplitude detection, characterized in that, include: Step 1: Stratify the atmosphere and simulate multi-layered turbulent phase screens at different heights; Parallel light is diffracted and transmitted through the multi-layer turbulent phase screen to obtain pupil plane complex amplitude information; based on the pupil plane complex amplitude information, a pupil plane light intensity image representing amplitude information and a Shaker-Hartmann image representing phase information are calculated respectively; a dataset is constructed using the pupil plane light intensity image and the Shaker-Hartmann image; Step 2: Construct the Atmospheric Turbulence 3D Reconstruction Network AT-3DRN. In the AT-3DRN, a coordinate attention mechanism is introduced into the residual block to extract local spatial features, and two layers of linear attention mechanism are introduced at the back end of the residual block to model the global spatial dependency. Step 3: Using the dataset as input, train the AT-3DRN to obtain the mapping relationship from pupil detection information to multi-layer turbulent phase screen wavefront aberrations, thereby reconstructing multi-layer turbulent wavefront aberrations based on the real-time acquired pupil light intensity image and Shaker-Hartmann image.

2. The method for reconstructing multi-layer atmospheric turbulence based on pupil plane complex amplitude detection according to claim 1, characterized in that, In step 1, the wavefront of the multi-layer turbulent phase screen is described by the Zernike coefficient, which conforms to the Kolmogorov distribution; the pupil plane complex amplitude information obtained after the parallel light is diffracted and transmitted through the multi-layer turbulent phase screen is obtained by sequential convolution and multiplication of the spatial impulse response of Fresnel diffraction with the turbulent phase of each layer.

3. The multi-layer atmospheric turbulence reconstruction method based on pupil plane complex amplitude detection according to claim 1, characterized in that, The multi-layered turbulent phase screen has seven layers, and the equivalent atmospheric coherence length of each phase screen, from the lower to the upper layers, ranges from 3 to 43 centimeters.

4. The multi-layer atmospheric turbulence reconstruction method based on pupil plane complex amplitude detection according to claim 1, characterized in that, In step 2, the input dimension of the atmospheric turbulence 3D reconstruction network is a two-channel image, corresponding to the pupil light intensity image and the Shaker-Hartmann image respectively; the output of the atmospheric turbulence 3D reconstruction network is the Zernike coefficients of each order of the multi-layer turbulence phase screen.

5. The multi-layer atmospheric turbulence reconstruction method based on pupil plane complex amplitude detection according to claim 1, characterized in that, In step 2, the atmospheric turbulence 3D reconstruction network AT-3DRN sequentially includes: a convolutional layer with a kernel size of 3×3, a max pooling layer with a size of 3×3 and a stride of 2, multiple cascaded residual blocks, and two linear attention mechanisms located at the rear end of the last residual block; wherein each residual block sequentially includes three convolutional blocks, and the coordinate attention mechanism is introduced between the second and third convolutional blocks.

6. The multi-layer atmospheric turbulence reconstruction method based on pupil plane complex amplitude detection according to claim 5, characterized in that, In the two-layer linear attention mechanism, the first layer linear attention mechanism is used to aggregate global information to capture spatial dependencies across sub-apertures, and the second layer linear attention mechanism is used to adaptively reweight and refine features under the guidance of global information.

7. The multi-layer atmospheric turbulence reconstruction method based on pupil plane complex amplitude detection according to claim 5, characterized in that, Each residual block consists of three convolutional blocks of sizes 1×1, 3×3, and 1×1, respectively.

8. The multi-layer atmospheric turbulence reconstruction method based on pupil plane complex amplitude detection according to claim 1, characterized in that, In step 3, when training the AT-3DRN, the pupil light intensity image and the Shaker-Hartmann image are normalized, and the Adam optimizer is used to update the gradient using the root mean square error between the network output and the standard output as the loss function.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When one or more programs are executed by the one or more processors, the one or more processors implement the multi-layer atmospheric turbulence reconstruction method based on pupil plane complex amplitude detection as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, It stores executable instructions that, when executed by a processor, enable the processor to implement the multi-layer atmospheric turbulence reconstruction method based on pupil plane complex amplitude detection as described in any one of claims 1-8.

Images