CN116578821A - Prediction method of atmospheric turbulence phase screen based on deep learning - Google Patents
Prediction method of atmospheric turbulence phase screen based on deep learning Download PDFInfo
- Publication number
- CN116578821A CN116578821A CN202310631745.7A CN202310631745A CN116578821A CN 116578821 A CN116578821 A CN 116578821A CN 202310631745 A CN202310631745 A CN 202310631745A CN 116578821 A CN116578821 A CN 116578821A
- Authority
- CN
- China
- Prior art keywords
- atmospheric turbulence
- phase screen
- neural network
- phase
- atmospheric
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 28
- 238000013135 deep learning Methods 0.000 title claims abstract description 13
- 230000006870 function Effects 0.000 claims abstract description 23
- 238000013528 artificial neural network Methods 0.000 claims abstract description 15
- 238000012549 training Methods 0.000 claims description 28
- 238000013527 convolutional neural network Methods 0.000 claims description 16
- 230000003287 optical effect Effects 0.000 claims description 12
- 230000004913 activation Effects 0.000 claims description 6
- 238000001228 spectrum Methods 0.000 claims description 6
- 230000005540 biological transmission Effects 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 4
- 230000003595 spectral effect Effects 0.000 claims description 4
- 238000012805 post-processing Methods 0.000 claims description 2
- 238000002360 preparation method Methods 0.000 claims description 2
- 238000000691 measurement method Methods 0.000 claims 1
- 238000004422 calculation algorithm Methods 0.000 description 7
- 238000000605 extraction Methods 0.000 description 7
- 230000003044 adaptive effect Effects 0.000 description 5
- 238000004891 communication Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 3
- 238000011176 pooling Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 239000000523 sample Substances 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000021615 conjugation Effects 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000010845 search algorithm Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/15—Correlation function computation including computation of convolution operations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Mathematical Optimization (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a prediction method of an atmospheric turbulence phase screen based on deep learning, wherein the method comprises the following steps: firstly, according to the internal relation between distorted light beams, undistorted light beams and ideal output phases, a convolution neural network with low complexity and a mean square error loss function with pixel penalty term are designed, any nonlinear function can be approximated with arbitrary precision in theory according to the neural network, and the corresponding atmospheric turbulence phase screen can be predicted through the distorted light beams when model parameters are fixed after parameter adjustment is continuously optimized. The network model has the advantages of less parameter quantity, higher speed and higher precision.
Description
Technical Field
The invention belongs to the field of free space optical communication, and particularly relates to prediction of an atmospheric turbulence phase screen based on deep learning.
Background
Atmospheric transport distortion is one of the major challenges that hinders practical applications of vortex beams carrying orbital angular momentum. How to accurately predict or capture the phase distortion caused by the atmospheric turbulence in the atmospheric channel is an important basis for realizing the atmospheric turbulence compensation. There have been many studies to propose corresponding solutions to improve the robustness of the swirling beam to atmospheric turbulence. Adaptive optics are the most common method of acquiring the wavefront phase of a distorted beam using a Shack-Hartmann wavefront sensor and then correcting the distorted phase using deformable mirrors or spatial light modulators, which are expensive and costly to maintain.
In recent years, adaptive optics without wavefront sensors has received a certain attention, and such systems differ from conventional adaptive optics in that they do not employ wavefront sensors, but rather employ random, local or global search algorithms to extract the wavefront phase of the beam, and then compensate for phase distortions based on phase conjugation. Common algorithms are the Gerchberg-Saxton phase extraction algorithm and the random parallel gradient descent phase extraction algorithm. However, both of the above phase extraction algorithms require multiple iterations, and therefore the system requires a long processing time to extract the wavefront phase. In addition, the algorithms have no learning and memory capacity, and convergence stagnation is often caused by the fact that the convergence is in a local minimum in the iterative calculation process, so that accurate wave-front phase information cannot be obtained.
In summary, the speed and effect balance cannot be achieved with either the wavefront sensor based adaptive optics system or the wavefront sensor free adaptive optics system.
The prediction of the atmospheric turbulence phase screen can be used for estimating an atmospheric optical link of the free space optical communication system, extracting wave front phase distortion of an optical signal and carrying out phase compensation, thereby obviously improving the performance of the free space optical communication system. The transmitting end of the free space optical communication system transmits the modulated optical signal to an atmosphere channel for transmission, and the wave front phase distortion of the optical signal is caused by the atmosphere turbulence in the atmosphere channel. When an optical signal arrives at a receiving end through a distance of atmospheric transmission, the accumulated phase distortion will cause a drastic drop in the system channel capacity. Therefore, predicting and compensating for phase distortions caused by atmospheric turbulence is one of the key issues that free-space optical communication systems need to address.
The atmospheric turbulence phase screen prediction method based on deep learning can effectively solve the problems.
Disclosure of Invention
The present invention aims to solve the above technical problems. For this purpose, the invention proposes a deep learning based atmospheric turbulence phase screen prediction.
The technical scheme adopted for solving the technical problems is as follows:
the prediction method of the atmospheric turbulence phase screen based on deep learning is characterized by comprising the following steps of:
s1: according to the inherent relation between distorted and undistorted light beams and a real phase screen and the characteristic that a function can be approximated with arbitrary precision according to the neural network, the neural network is adopted to predict the feasibility of the phase screen, and a convolution neural network with low complexity and a mean square error loss function with a pixel penalty term are designed;
s2: the method comprises the steps of obtaining the light intensity of two Gaussian beams which are distorted and not distorted and a corresponding real phase screen as training data, and adopting an atmospheric refractive index power spectral density Hill-Andrews model for enabling atmospheric turbulence to be more fit with the actual situation, wherein the expression is as follows:
wherein Indicating when the light beam is inzAtmospheric refractive index power spectrum at transmission in direction, wherein +.>Indicates an exponential constant based on the natural constant e, < ->Represents the atmospheric turbulence intensity, +.>Represents the inner and outer scale factors of the atmospheric turbulence, respectively,/->Respectively arexAndywavenumbers in the direction;
s3: training the designed convolutional neural network model, continuously optimizing parameters of the model after parameter adjustment, reading a distorted light beam intensity graph into the trained convolutional neural network, and realizing prediction of a corresponding atmospheric turbulence phase screen.
The loss function of the invention is a judgment standard for learning parameters by taking the mean square error with pixel penalty term, and the loss functionThe method comprises the following steps:
(2)
wherein ,represents weight parameters, N represents the number of training samples, +.>Representing the%>Intensity profile of individual gaussian beams,/->Representing the%>A personal atmospheric turbulence phase screen->Taking the whole number->Represents a predicted atmospheric turbulence phase screen, +.>Representing a nonlinear activation function +.>Represents penalty factors,/->Representing a summation operation; the predicted atmospheric turbulence phase screen is characterized by determining an optimal weight parameter +.>Then, according to the input dataXA predicted atmospheric turbulence phase screen can be obtained:
(3)
wherein ,represents a predicted atmospheric turbulence phase screen, +.>Representing a minimum operation-> and />Respectively representing the input data and the output phase, and +.>Can be regarded as an abstract nonlinear function approximated by a neural network.
The convolutional neural network with low complexity is characterized in that the model complexity is small, the parameters required to be trained are few, the computational effort required to be trained is small, the whole process is an end-to-end process, no post-processing is required for data preparation and prediction results, and precise and expensive optical instruments are not required. The prediction method further comprises the following steps: the implementation method is used for predicting the atmospheric turbulence phase screen.
The invention further discloses an application of the prediction method of the atmospheric turbulence phase screen based on deep learning in accurately and rapidly extracting phase distortion and phase compensation caused by atmospheric turbulence. The experimental results show that: the trained neural network model can accurately and rapidly predict the atmospheric turbulence phase screen, the time is in the order of milliseconds, and is irrelevant to the atmospheric turbulence intensity, and the method is superior to the traditional Gerchberg-Saxton algorithm in the aspects of the speed and the accuracy of phase extraction.
The invention is described in more detail below:
the atmospheric turbulence phase screen prediction method based on deep learning is characterized by comprising the following steps of:
(1) Based on the inherent correlation between the light intensity of the beam and its wavefront phase, a special convolutional neural network structure predicted by the phase of atmospheric turbulence and a proper loss function are designed. The model has simple structure and small parameter quantity, and is easy to train and infer in real time.
The loss function in this patent consists essentially of three parts:
wherein ,represents weight parameters, N represents the number of training samples, +.>Representing the%>Intensity profile of individual gaussian beams,/->Representing the%>A personal atmospheric turbulence phase screen->Taking the whole number->
Representing a predicted atmospheric turbulence phase screen,representing a nonlinear activation function +.>Represents penalty factors,/->
Representing a summation operation.
(2) And acquiring light intensity data of two distorted Gaussian beams and undistorted Gaussian beams and a corresponding real atmospheric turbulence phase screen as training data. In the patent, hill-Andrews atmospheric refractive index power spectral density model consistent with actual atmospheric turbulence behavior is adopted to numerically simulate an atmospheric turbulence phase screen.
Atmospheric refractive index power spectral density function defined by the modelHas the following expression:
wherein Indicating when the light beam is inzAtmospheric refractive index power spectrum at transmission in direction, wherein +.>Expressed as natural constant->An exponential constant as a base, wherein->Represents the atmospheric turbulence intensity, +.>Represents the inner and outer scale factors of the atmospheric turbulence, respectively,/->Respectively arexAndywavenumbers in the direction;
(3) And (3) taking the distorted and undistorted two Gaussian beam light intensity graphs and the corresponding real atmospheric turbulence phase screen obtained in the step (2) as a training data set and a testing data set, training the established deep convolutional neural network model, and predicting the atmospheric turbulence phase screen according to the input beam intensity distribution diagram after the parameters of the model are fixed.
The invention mainly solves the important problems that the acquisition of the existing atmospheric turbulence phase is difficult to meet the requirement, and the acquisition method is complicated, and mainly examines the feasibility of extracting the phase information by adopting the deep neural network to the light intensity of two Gaussian beams which are not distorted and distorted, optimizes the phase extraction process, and ensures that the phase extraction speed becomes faster and the accuracy meets the requirement.
Compared with the prior art, the atmospheric turbulence phase screen prediction based on the deep convolutional neural network has the following positive effects:
(1) The mapping relation between the original light beam, the distorted light beam and the real phase is fitted through a neural network, so that end-to-end atmospheric turbulence phase screen prediction is realized.
(2) The trained neural network has stronger generalization, and can theoretically realize the phase screen prediction of any turbulence intensity.
The invention further discloses a feasible scheme of atmospheric turbulence phase screen prediction based on deep learning. We describe the proposed CNN model working mechanism as a mathematical problem: for the light intensity distribution of the standard gaussian probe beam e_1 that is not affected by turbulence, the light intensity distribution of the gaussian probe beam e_2 that is affected by turbulence, and the ideal output turbulence phase P, there is the following relation: e_2=f (P, e_1), f representing the light field transfer function of e_1 affected by P. After the change p=g (e_1, e_2), where g is the transform function of f, the CNN can learn the mapping relation g by a large number of (e_1, e_2) and corresponding ideal output phase screens P. This means that the turbulence phase can be well predicted regardless of the turbulence intensity. The reason that the information extraction capability of CNN is not affected by global changes of the object from an algorithm point of view is that CNN can automatically extract intrinsic features of an image through a multi-layer structure, where convolution operations and sub-sampling locally correlate the features of the object. This enables CNN-based turbulent phase prediction to be adapted to a variety of turbulent environments.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention.
In the drawings:
FIG. 1 is a flow chart of deep learning based atmospheric turbulence phase screen prediction;
FIG. 2 is a block diagram of a convolutional neural network employed in this patent;
FIG. 3 is a training loss curve for a neural network;
FIG. 4 is a true atmospheric turbulence phase screen;
FIG. 5 is a predicted atmospheric turbulence phase screen after model parameters are fixed;
fig. 6 is a plot of the predicted atmospheric turbulence phase screen immediately after training of the model.
Detailed Description
The invention is described below by means of specific embodiments. The technical means used in the present invention are methods well known to those skilled in the art unless specifically stated. Further, the embodiments should be construed as illustrative, and not limiting the scope of the invention, which is defined solely by the claims. Reference will now be made in detail to the embodiments of the present invention, examples of which are illustrated in the accompanying drawings and are exemplary of the embodiments described below with reference to the drawings, which are intended to illustrate the invention and not to be construed as limiting the invention.
Example 1
Step one:
the convolutional neural network designed by the patent comprises 15 convolutional layers and 3 upsamples, and adopts an encoder-decoder (encoder-decoder) architecture. Wherein the first 8-layer constituent encoder of the network performs downsampling to obtain a feature map. The layer 9 to 18 layer composition decoder performs up-sampling to restore the feature map to the resolution of the original image.
In the encoder section, network inputsTwo-dimensional gray level light intensity images with the size are obtained after 9 convolutions and 3 pooling of the input image, and the number of channels is 128 +.>A feature map of size. In the convolution layer, the input matrix is convolved with a particular feature detector (also called a kernel). When the image is convolved with the kernel, a new matrix is generated, called a feature map. Each kernel may extract a particular feature of the image and different kernels may extract different features. The generated feature map enters the next layer as an output of the convolutional layer. Except for the first layer is +.>All convolution kernels except the convolution are +.>. The feature map is pooled into a pooling layer, we choose +.>Is a maximum pooling of (a). Pool element is from every non-overlapping ++in the convolution profile>The sub-regions of each sub-region are input and the maximum value of each sub-region is selected as output, so that the computational complexity is greatly reduced. In the decoder section, the feature map is output after 7 convolutions and 3 upsamplesThe image is predicted, and the up-sampling operation does not select deconvolution (can be regarded as the inverse operation of convolution) but is bilinear interpolation, and mainly adopts a combination mode of convolution and bilinear interpolation to avoid the chessboard effect caused by deconvolution so that the up-sampled image has better quality. The length and width of each up-sampled image are increased by two times, and the final output of the network is +.>Is used for predicting the phase image of the object. Throughout the network we split the training dataset equally into batches to enhance the stability of convergence. And a batch of normalization layers are connected behind each convolution layer except the last layer, so that the normalization of the input image is realized and the network training time is reduced. And the correction linear unit (ReLU function) is used for nonlinear activation, so that gradient disappearance or explosion is prevented, and the training speed is increased. A sigmoid activation function is used at the final convolutional layer to control the magnitude of the output value.
Step two:
the loss function of this patent can be divided intoThree parts;
(9)
where N represents the number of samples,representing weight parameters->Representing the input +.>Light intensity patterns->Express +.>The labels or phase screens corresponding to the light intensity patterns represent nonlinear activation functions +.>Representing the mean square error loss, is usually used in regression tasks to approximate the true value, here to make the predicted phase more approximate to the true phase, +.>Is a penalty factor to control the dynamics of the penalty term. />Is used to penalize errors between pixels. And then by minimizing the loss function:
the optimal neural network model parameters can be obtained:
step three:
two kinds of beam data and corresponding real phase screens are obtained, a classical and widely used model developed by Hill is adopted in the patent and is defined by Andrews analysis, in which the atmospheric turbulence is simulated by a random phase screen loaded with a refractive index fluctuation spectrum;
wherein Representing the atmospheric refractive index power spectrum when a light beam is transmitted in the z-directionWherein->Expressed as natural constant->An exponential constant as a base, wherein->Represents the atmospheric turbulence intensity, +.>Respectively representing the inside and outside dimension radii. />The wavenumbers in the x and y directions, respectively. Wave front phase fluctuations are further distributed by random with variance +.>To model.
In a random phase screen, the perturbation of the refractive index is approximately represented by the Kolmogorov spectrum;
the size of the grid spacing and the random phase are represented, respectively. Wave number->Representing wave numbers. In addition, Δz is the separation distance between successive phase screens.
For ease of computation, the random distribution of phase disturbances is described in a rectangular coordinate system, and then the phase screen is represented in the frequency domain by a fast fourier transform operation.
Where FFT represents the fast fourier transform, M is a complex random matrix with a mean of 0 and a variance of 1.
For example, according to the flow shown in fig. 1, after the neural network structure shown in fig. 2 is designed, numerical simulation is adopted to obtain data required for training and testing, and the numerical simulation parameters are parameters shown in table 1.
Table 1 simulation parameters
30000 grey scale pictures were generated for use as training data, 3000 of which were used as test sets, the test data being used mainly to verify the effect of the training. Because the corresponding verification image is generated during each round of verification, the training can be stopped timely after the effect is still bad for a plurality of rounds of training, corresponding parameters are modified, fine adjustment is not needed until the training is finished, and a plurality of training time can be saved. In the case of a better parameter setting, a relationship between the loss value and the training epochs can be obtained after the training is completed, as shown in fig. 3, and it can be seen that the loss starts to decrease from fast to slow and converges to a lower value as the epochs increase. In addition, when model training is completed, the output of models in different stages can be compared through fig. 4, 5 and 6, the sequence from top to bottom is respectively a real phase screen, the phase screen predicted after model training is completed and the phase screen predicted when the model starts training can be clearly found, and the predicted phase screen obtained along with model training is more close to the real phase screen.
Claims (5)
1. The prediction method of the atmospheric turbulence phase screen based on deep learning is characterized by comprising the following steps of:
s1: according to the inherent relation between distorted and undistorted light beams and a real phase screen and the characteristic that a function can be approximated with arbitrary precision according to the neural network, the neural network is adopted to predict the feasibility of the phase screen, and a convolution neural network with low complexity and a mean square error loss function with a pixel penalty term are designed;
s2: the method comprises the steps of obtaining the light intensity of two Gaussian beams which are distorted and not distorted and a corresponding real phase screen as training data, and adopting an atmospheric refractive index power spectral density Hill-Andrews model for enabling atmospheric turbulence to be more fit with the actual situation, wherein the expression is as follows:
wherein Indicating when the light beam is inzAtmospheric refractive index power spectrum at transmission in direction, +.>Represents the atmospheric turbulence intensity, +.>Represents the inner and outer scale factors of the atmospheric turbulence, respectively,/->Respectively arexAndywavenumbers in the direction;
s3: training a designed convolutional neural network model, continuously optimizing parameters of the model after parameter adjustment, reading a distorted light beam intensity graph into the trained convolutional neural network, and realizing prediction of a corresponding atmospheric turbulence phase screen; wherein the loss function is a judgment standard of taking the mean square error with pixel penalty term as parameter learning, and the loss functionThe method comprises the following steps:
(2)
wherein ,represents weight parameters, N represents the number of training samples, +.>Representing the%>Intensity profile of individual gaussian beams,/->Representing the%>A personal atmospheric turbulence phase screen->Taking the whole number->Represents a predicted atmospheric turbulence phase screen, +.>Representing a nonlinear activation function +.>Represents penalty factors,/->Representing a summation operation; determining optimal weight parameter->Then, according to the input dataXA predicted atmospheric turbulence phase screen can be obtained:
(3)
wherein Represents a predicted atmospheric turbulence phase screen, +.> The minimum operation is indicated to be taken,Lrepresenting a loss function-> and />Representing input data and output phase, respectively, +.>Represents the optimal weight parameter, and +.>Can be regarded as an abstract nonlinear function approximated by a neural network.
2. The prediction method of claim 1, wherein the convolutional neural network with low complexity has a small model complexity, requires few parameters to train, and requires little computational effort to train.
3. The predictive method of claim 1 wherein the measurement method is an end-to-end process, and no post-processing is required for data preparation and prediction results, and no elaborate and expensive optical instrumentation is required.
4. The prediction method of claim 1, further comprising: the implementation method is used for predicting the atmospheric turbulence phase screen.
5. The prediction method of the atmospheric turbulence phase screen based on deep learning of claim 1 is mainly used for accurately and rapidly extracting the atmospheric turbulence phase screen.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310631745.7A CN116578821A (en) | 2023-05-31 | 2023-05-31 | Prediction method of atmospheric turbulence phase screen based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310631745.7A CN116578821A (en) | 2023-05-31 | 2023-05-31 | Prediction method of atmospheric turbulence phase screen based on deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116578821A true CN116578821A (en) | 2023-08-11 |
Family
ID=87541259
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310631745.7A Pending CN116578821A (en) | 2023-05-31 | 2023-05-31 | Prediction method of atmospheric turbulence phase screen based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116578821A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117056732A (en) * | 2023-10-11 | 2023-11-14 | 山东科技大学 | Non-isotropic NARX troposphere delay grid prediction method |
-
2023
- 2023-05-31 CN CN202310631745.7A patent/CN116578821A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117056732A (en) * | 2023-10-11 | 2023-11-14 | 山东科技大学 | Non-isotropic NARX troposphere delay grid prediction method |
CN117056732B (en) * | 2023-10-11 | 2023-12-15 | 山东科技大学 | Non-isotropic NARX troposphere delay grid prediction method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113129247B (en) | Remote sensing image fusion method and medium based on self-adaptive multi-scale residual convolution | |
CN111353424B (en) | Remote sensing image spatial spectrum fusion method of depth recursion residual error network and electronic equipment | |
CN101819325A (en) | Optical system and method for producing optical system | |
CN110930439B (en) | High-grade product automatic production system suitable for high-resolution remote sensing image | |
CN116578821A (en) | Prediction method of atmospheric turbulence phase screen based on deep learning | |
CN112801904B (en) | Hybrid degraded image enhancement method based on convolutional neural network | |
Suárez Gómez et al. | Improving adaptive optics reconstructions with a deep learning approach | |
Gao et al. | Stacked convolutional auto-encoders for single space target image blind deconvolution | |
Zeng et al. | Cascade neural network-based joint sampling and reconstruction for image compressed sensing | |
CN111353939A (en) | Image super-resolution method based on multi-scale feature representation and weight sharing convolution layer | |
Wang et al. | Automated clustering method for point spread function classification | |
CN113237554B (en) | Method and device for generating surface temperature image under cloud and terminal equipment | |
CN111277809A (en) | Image color correction method, system, device and medium | |
CN113256733B (en) | Camera spectral sensitivity reconstruction method based on confidence voting convolutional neural network | |
CN116299247B (en) | InSAR atmospheric correction method based on sparse convolutional neural network | |
CN113705340A (en) | Deep learning change detection method based on radar remote sensing data | |
CN110895790B (en) | Scene image super-resolution method based on posterior degradation information estimation | |
CN116883799A (en) | Hyperspectral image depth space spectrum fusion method guided by component replacement model | |
Cang et al. | Research on hyperspectral image reconstruction based on GISMT compressed sensing and interspectral prediction | |
CN107085843A (en) | System and method for estimating the modulation transfer function in optical system | |
Wei et al. | High-quality blind defocus deblurring of multispectral images with optics and gradient prior | |
Wang et al. | Single-frame super-resolution for high resolution optical remote-sensing data products | |
Li et al. | Multi-sensor multispectral reconstruction framework based on projection and reconstruction | |
Lu et al. | Multi-Supervised Recursive-CNN for Hyperspectral and Multispectral Image Fusion | |
CN111223044A (en) | Method for fusing full-color image and multispectral image based on dense connection network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |