CN110349095A - Learn the adaptive optics wavefront compensation method of prediction wavefront zernike coefficient based on depth migration - Google Patents
Learn the adaptive optics wavefront compensation method of prediction wavefront zernike coefficient based on depth migration Download PDFInfo
- Publication number
- CN110349095A CN110349095A CN201910513458.XA CN201910513458A CN110349095A CN 110349095 A CN110349095 A CN 110349095A CN 201910513458 A CN201910513458 A CN 201910513458A CN 110349095 A CN110349095 A CN 110349095A
- Authority
- CN
- China
- Prior art keywords
- layer
- wavefront
- network
- source domain
- aiming field
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 37
- 230000003044 adaptive effect Effects 0.000 title claims abstract description 27
- 238000013508 migration Methods 0.000 title claims abstract description 26
- 230000005012 migration Effects 0.000 title claims abstract description 26
- 238000012549 training Methods 0.000 claims abstract description 18
- 238000012937 correction Methods 0.000 claims abstract description 10
- 210000002569 neuron Anatomy 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 8
- 230000006978 adaptation Effects 0.000 claims description 6
- 238000013480 data collection Methods 0.000 claims description 6
- 238000012512 characterization method Methods 0.000 claims description 3
- 238000003475 lamination Methods 0.000 claims description 3
- 238000013527 convolutional neural network Methods 0.000 abstract description 5
- 238000013461 design Methods 0.000 abstract description 4
- 238000013526 transfer learning Methods 0.000 abstract description 3
- 230000003287 optical effect Effects 0.000 description 14
- 238000002474 experimental method Methods 0.000 description 10
- 238000003384 imaging method Methods 0.000 description 10
- 230000006872 improvement Effects 0.000 description 6
- 230000008901 benefit Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000000386 microscopy Methods 0.000 description 3
- 230000021615 conjugation Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G02—OPTICS
- G02B—OPTICAL ELEMENTS, SYSTEMS OR APPARATUS
- G02B27/00—Optical systems or apparatus not provided for by any of the groups G02B1/00 - G02B26/00, G02B30/00
- G02B27/0012—Optical design, e.g. procedures, algorithms, optimisation routines
-
- G—PHYSICS
- G02—OPTICS
- G02F—OPTICAL DEVICES OR ARRANGEMENTS FOR THE CONTROL OF LIGHT BY MODIFICATION OF THE OPTICAL PROPERTIES OF THE MEDIA OF THE ELEMENTS INVOLVED THEREIN; NON-LINEAR OPTICS; FREQUENCY-CHANGING OF LIGHT; OPTICAL LOGIC ELEMENTS; OPTICAL ANALOGUE/DIGITAL CONVERTERS
- G02F1/00—Devices or arrangements for the control of the intensity, colour, phase, polarisation or direction of light arriving from an independent light source, e.g. switching, gating or modulating; Non-linear optics
- G02F1/01—Devices or arrangements for the control of the intensity, colour, phase, polarisation or direction of light arriving from an independent light source, e.g. switching, gating or modulating; Non-linear optics for the control of the intensity, phase, polarisation or colour
- G02F1/0121—Operation of devices; Circuit arrangements, not otherwise provided for in this subclass
-
- 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
- G06N3/045—Combinations of networks
-
- 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
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/067—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using optical means
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/80—Geometric correction
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Nonlinear Science (AREA)
- Optics & Photonics (AREA)
- Neurology (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The present invention provides a kind of adaptive optics wavefront compensation method for learning prediction wavefront zernike coefficient based on depth migration: including: to obtain the source domain sample of mark;Obtain the aiming field sample of no mark;Obtain the source domain sample of mark;Based on source domain data set and aiming field data set training network, model parameter is optimized by back-propagation algorithm, obtains the network for completing training;By the network of the obtained completion training of image input S35, the zernike coefficient of prediction is exported, Wave-front phase is calculated according to the zernike coefficient of prediction, and carry out wavefront compensation using wavefront correction device and realize correction.Present invention employs the double-flow designs of migration convolutional neural networks, can utilize wavefront coefficient prediction ability of the transfer learning improved model in different medium.
Description
Technical field
The present invention relates to a kind of optical image technologies, and in particular to one kind is based on depth migration study prediction wavefront Ze Nike
The adaptive optics wavefront compensation method of coefficient.
Background technique
Adaptive optical technique is the approach for solving optical wavefront error, passes through real-time detection-control-to money error is touched
Correction, enables optical system to overcome external disturbance automatically, overcomes the scattering of complicated observation medium to influence, keeps system good
Imaging performance.The adaptive optical technique of indirect formula based on image calculates wave by the scattering speckle analysis for point light source
Preceding parameter avoids the direct detection for medium Wave-front phase, the case where for living body medium, especially leads in micro-imaging etc.
It is used widely in domain.
Adaptive optics method based on image divides image by equipment optical path acquisition distortion light spot image
Analysis calculates wavefront parameter, Wave-front phase is adjusted by spatial light modulator, to correct wavefront distortion.Existing method passes through at present
Deep learning method obtains neural network using manually generated distortion hot spot data training, utilizes the network query function Wave-front phase
Zernike coefficient expression, have a capability for processing actual wavefront distortion, but for different appointed conditions, observe sample
This property, it is difficult to obtain good effect.For the appointed condition of variation, observation sample, the property and training of the hot spot that distorts
Sample has larger difference, it is difficult to obtain the mark sample under the conditions of current experiment for training, this makes the adaptive optics side
Method is difficult to apply under the conditions of actual experiment.
Therefore, it is necessary to improve to the prior art.
Summary of the invention
The technical problem to be solved in the present invention is to provide a kind of efficiently based on depth migration study prediction wavefront Ze Nike
The adaptive optics wavefront compensation method of coefficient.
In order to solve the above technical problems, the present invention provides and a kind of learns prediction wavefront zernike coefficient based on depth migration
Adaptive optics wavefront compensation method: the following steps are included:
S1, the source domain sample { X for obtaining marks,Ys};Execute step S2;
S2, the aiming field sample for obtaining no markObtain the aiming field sample of markExecute step
S3;
S3, basis have the source domain sample { X of marks,YsSource domain data set is obtained, according to the aiming field sample of no markWith the aiming field sample for having markObtain aiming field data set;It is assembled for training based on source domain data set and aiming field data
Practice network, model parameter is optimized by back-propagation algorithm, obtains the network for completing training;Execute step S4;
The trained network of S4, the completion for obtaining image input S35, exports the zernike coefficient of prediction, executes step S5;
S5, Wave-front phase is calculated according to the zernike coefficient of step S4 prediction, and carries out wavefront benefit using wavefront correction device
Repay realization correction.
As to the present invention is based on the adaptive optics wavefront compensation sides that depth migration learns prediction wavefront zernike coefficient
The improvement of method: step S3 the following steps are included:
S31. using the source domain network of shared parameter and aiming field network as binary-flow network, judge source domain network and aiming field
Whether network meets the condition of convergence or reaches maximum number of iterations, otherwise executes S32, is, executes S35;
S32. it will be divided into multiple small-sized lot data collection, shape after source domain data set and aiming field data set respectively random disorder
At source domain batch data and aiming field batch data;Execute step S33;
S33. source domain batch data and aiming field batch data are inputted into source domain network and aiming field network respectively, passed using reversed
It broadcasts algorithm combination loss function and updates each network parameter;Execute step S34;
S34. model parameter is updated using gradient descent algorithm, and judges whether parameter restrains, be to execute S35,
No execution S32;
S35. the network of training is completed in output;Execute step S4.
As to the present invention is based on the adaptive optics wavefront compensation sides that depth migration learns prediction wavefront zernike coefficient
The further improvement of method: step S33 includes:
Loss function is defined as:
L=MSE+ λ lCORAL
Wherein MSE is mean square deviation error;MSE is shown below:
Wherein:The respectively predicted value of source domain network and aiming field network;To there is the aiming field sample of mark
Quantity;nsTo there is the source domain sample size of mark;
Covariance loses lCORALIt is shown below:
Wherein,Indicate square matrices Frobenius norm;Cs,CtRespectively indicate the association side of source domain and target domain characterization
Poor matrix, d are sample vector space dimensionality;
λ is the hyper parameter for balancing zernike coefficient prediction and the intensity being distributed between being aligned.
As to the present invention is based on the adaptive optics wavefront compensation sides that depth migration learns prediction wavefront zernike coefficient
The further improvement of method: step S34 includes:
Parameter form renewal are as follows:
Wherein, α is learning rate,For partial derivative operator,It indicates in t time step, i-th of parameter of kth layer
Value.
As to the present invention is based on the adaptive optics wavefront compensation sides that depth migration learns prediction wavefront zernike coefficient
The further improvement of method:
Source domain network and aiming field network include input layer-convolutional layer-pond layer-convolutional layer-pond layer-convolutional layer-
Full articulamentum-the adaptation layer of convolutional layer-convolutional layer-pond layer-.
As to the present invention is based on the adaptive optics wavefront compensation sides that depth migration learns prediction wavefront zernike coefficient
The further improvement of method:
The setting characteristic pattern number of 1st layer of input layer is 1;The characteristic pattern number of level 2 volume lamination is 32, and it is big that convolution kernel is arranged
Small is 5;3rd layer of pond layer setting down-sampling is having a size of 2;The characteristic pattern number of 4th layer of convolutional layer is 32, and convolution kernel size is arranged
It is 5;5th layer of pond layer setting down-sampling is having a size of 2;The characteristic pattern number of 6th layer of convolutional layer is 64, and convolution kernel size is arranged and is
3;The characteristic pattern number of 7th layer of convolutional layer is 64, and it is 3 that convolution kernel size, which is arranged,;The characteristic pattern number of 8th layer of convolutional layer is 64, and
It is 3 that convolution kernel size, which is arranged,;9th layer of pond layer setting down-sampling is having a size of 2;Neuron number is arranged in 10th layer of full articulamentum
512;Neuron number 512 is arranged in 11th layer adaptation layer;12nd layer of output layer setting neuron number is 21, corresponding 2-22 pool
Buddhist nun gram coefficient.
As to the present invention is based on the adaptive optics wavefront compensation sides that depth migration learns prediction wavefront zernike coefficient
The further improvement of method:
Hyper parameter λ is 7.
The present invention is based on the skills that depth migration learns the adaptive optics wavefront compensation method of prediction wavefront zernike coefficient
Art advantage are as follows:
(1) under the observation condition of change experiment condition, correction energy can be substantially improved compared to existing machine learning method
Power.
(2) for the experiment condition of change, the fault image for obtaining mark, the i.e. distortional wave without knowing them are not needed
Preceding parameter can realize being substantially improved for performance.
(3) for the experiment condition of change, the target numeric field data marked on a small quantity can more make correction performance be promoted, this makes
This method can preferably utilize the data that can be obtained.
(4) double-flow design of migration convolutional neural networks is used, it can be using transfer learning improved model in different medium
When wavefront coefficient prediction ability.
(5) the target numeric field data that mark is taken full advantage of at the design of loss function, can be realized faster convergence
Speed prediction accuracy.
Detailed description of the invention
Specific embodiments of the present invention will be described in further detail with reference to the accompanying drawing.
Fig. 1 is the adaptive optics wavefront compensation side of the invention for learning prediction wavefront zernike coefficient based on Transfer Depth
Method flow chart;
Fig. 2 is the semi-supervised second order alignment that its Wave-front phase zernike coefficient is predicted for optical imagery distortion light spot image
Depth migration learn the structural schematic diagram of double-current convolutional neural networks in embodiment;
Fig. 3 is the structural schematic diagram of laser microscopy imaging system.
Specific embodiment
The present invention is described further combined with specific embodiments below, but protection scope of the present invention is not limited in
This.
Embodiment 1, the adaptive optics wavefront compensation method for being learnt prediction wavefront zernike coefficient based on depth migration, are answered
For laser microscopy imaging system, as shown in Figure 3;
Laser microscopy imaging system includes laser 1 (OBIS 637nm LX 140mW), beam-expanding collimation system 2, polarizing film
3, optical beam-splitter 4, spatial light modulator 5 (SLM, PLUTO-NIR-011-A), conjugation microscope group 6, object lens 7 (RMS4X,
Olympus, 4X/0.10NA) and CMOS camera 8 (CMOS, DMK 23UV024,640*480Y800115fps).
The laser that laser 1 issues first passes through beam-expanding collimation system 2 and expands, then by polarizing film 3, subsequent laser passes through light
It learns beam splitter 4 and vertically injects spatial light modulator 5, after the modulation of spatial light modulator 5, after light beam fold-back optical beam-splitter 4
Reflection is then passed through sample and is received by CMOS camera 8 by being received after conjugation microscope group 6 by object lens 7.
Artificial distortion hot spot has labeled data sample to generate by the following method:
1) it, is uniformly distributed and is randomly generated 10000 groups zernike coefficient 2-22;
2) Wave-front phase, is calculated by zernike coefficient;
3) it is 128*128 that, Wave-front phase, which is loaded into SLM (spatial light modulator) through above-mentioned optical path acquisition resolution,
Distortion light spot image (sample is vacant at this time).
The distortion light spot image for passing through artificial dielectric sample with imaging system shooting laser, obtains 320 groups of artificials
The distortion light spot image of generation, wherein known its of only 20 groups of samples corresponds to wavefront zernike coefficient 2-22 and (pass through its other party
Method there is known 20 groups of samples and correspond to zernike coefficient), this 20 groups of samples, which constitute aiming field, labeled data collection, and other 300 groups
Sample constitutes aiming field without labeled data collection.
Application learns the adaptive optics wavefront of prediction wavefront zernike coefficient based on Transfer Depth in the optical system
Compensation method, comprising the following steps:
The zernike coefficient of S1, the distortion hot spot that artificial is generated and its Wave-front phase are as the source domain for having mark
Sample { Xs,Ys};Execute step S2;
It S2, by the distortion light spot image under the conditions of current experiment (with the identical experiment condition of source domain sample is obtained) is target
Distortion light spot image under the conditions of current experiment is unaware that the labeled data of its Wave-front phase as no mark by domain sample
Aiming field sampleIf there is its a small amount of known Wave-front phase data in the distortion hot spot under the conditions of current experiment, that is current
Distortion light spot image under experiment condition and know the labeled data of its Wave-front phase as the aiming field sample for having markExecute step S3;
S3, the semi-supervised second order pair that its Wave-front phase zernike coefficient is predicted for optical imagery distortion light spot image is established
Neat depth migration learns double-current convolutional neural networks and is trained;Execute step S4;
It includes the following contents that the depth migration, which learns double-current convolutional neural networks:
1) one by input layer-convolutional layer-pond layer-convolutional layer-pond layer-convolutional layer-convolutional layer-convolutional layer-pond
The source domain network of the full articulamentum-adaptation layer of layer-is used to extract the feature of the image of source domain sample and carries out the pre- of zernike coefficient
It surveys.
2) the aiming field network of one and source domain network share parameter is used to extract the feature of the image of aiming field sample simultaneously
Carry out the prediction of zernike coefficient.
Specific network structure is as shown in Figure 2 in this embodiment, left side be source domain network, right side be aiming field network, two
The consistent simultaneously shared parameter of person's structure.The setting characteristic pattern number of 1st layer of input layer is 1;The characteristic pattern number of level 2 volume lamination is 32,
And it is 5 that convolution kernel size, which is arranged,;3rd layer of pond layer setting down-sampling is having a size of 2;The characteristic pattern number of 4th layer of convolutional layer is 32, and
It is 5 that convolution kernel size, which is arranged,;5th layer of pond layer setting down-sampling is having a size of 2;The characteristic pattern number of 6th layer of convolutional layer is 64, and is set
Setting convolution kernel size is 3;The characteristic pattern number of 7th layer of convolutional layer is 64, and it is 3 that convolution kernel size, which is arranged,;The spy of 8th layer of convolutional layer
Levying figure number is 64, and it is 3 that convolution kernel size, which is arranged,;9th layer of pond layer setting down-sampling is having a size of 2;10th layer of full articulamentum is set
Setting neuron number is 512;Neuron number 512 is arranged in 11th layer adaptation layer;Neuron number is arranged in 12nd layer of output layer
21, corresponding 2-22 zernike coefficient.
3) as binary-flow network, the training of the binary-flow network is based on source domain data set and mesh for source domain network and aiming field network
Numeric field data collection is marked, model parameter is optimized by back-propagation algorithm.
Source domain data set and aiming field data set have following definition: { Xs,YsIt is the source domain sample for having mark, { Xs,Ys}
In all data as source domain data set;To there is label target domain sample,For the aiming field sample of no mark
This,WithAll data as aiming field data set.Source domain and aiming field have different edge distributions, there is mark
The aiming field sample size of noteFar less than the source domain sample size n for having marks, quantity be zero have no effect on network structure with
Training process.
Training process specifically comprises the following steps:
S31. judge whether source domain network and aiming field network meet the condition of convergence or reach maximum number of iterations, otherwise
S32 is executed, is to execute S35;
S32. it will be divided into multiple small-sized lot data collection, shape after source domain data set and aiming field data set respectively random disorder
At source domain batch data and aiming field batch data;Execute step S33;
S33. source domain batch data and aiming field batch data are inputted into source domain network and aiming field network respectively, passed using reversed
It broadcasts algorithm combination loss function and updates each network parameter.Execute step S34;
Wherein loss function is defined as:
L=MSE+ λ lCORAL
Wherein MSE is mean square deviation error, for describing prediction loss;MSE is shown below:
Wherein:The respectively predicted value of source domain network and aiming field network;To there is the aiming field sample of mark
Quantity;nsTo there is the source domain sample size of mark;
Shi Zhong first part is source domain prediction error, and the aiming field sample of mark is then utilized in second part, so that net
Network can make full use of available labeled data.
In order to realize the transfer learning from source domain to aiming field, the present invention calculates covariance loss on adaptation layer
(CORAL) it is distributed using second order distributed feature align data.Covariance loses lCORALIt is shown below:
Wherein,Indicate square matrices Frobenius norm;Cs,CtRespectively indicate the association side of source domain and target domain characterization
Poor matrix, d are sample vector space dimensionality.
Hyper parameter λ is the intensity for balancing zernike coefficient prediction with being distributed between being aligned.In the present invention, we use
The method of grid search determines hyper parameter λ, and hyper parameter λ is set as 7 in the present embodiment.
S34. model parameter is updated using gradient descent algorithm, and judges whether parameter restrains, be to execute S35,
No execution S32;
Wherein design parameter form renewal are as follows:
Wherein, α is learning rate,For partial derivative operator,It indicates in t time step, i-th of parameter of kth layer
Value.
S35. the network of training is completed in output;
Training hardware platform are as follows: Intel (R) Xeon (R) CPU E5-2667 v4 3.20GHz, 2 inside save as 256G,
GPU:Tesla P4, video memory 7G;Training software platform is Tensorflow.
S4, the distortion light spot image for forming the laser that the imaging system is shot by artificial dielectric sample are (on can be
The distortion light spot image that the 320 groups of artificials stated generate is also possible to the new images shot by imaging system) input S35
The network of obtained completion training, exports the zernike coefficient of prediction, executes step S5;
S5, Wave-front phase is calculated according to the zernike coefficient of step S4 prediction, and using wavefront correction device (in the present embodiment
Wavefront compensation, which is carried out, for SLM) realizes correction.
It should be noted that the optical system example in embodiment is micro imaging system, and rectified using SLM as wavefront
Positive device, and present invention thinking to be protected and technology can also be using the situations with more than SLM as rectifier, or even do not terminate in
Micro imaging system, such as other optical systems of telescope are applicable in for other.
The above list is only a few specific embodiments of the present invention for finally, it should also be noted that.Obviously, this hair
Bright to be not limited to above embodiments, acceptable there are many deformations.Those skilled in the art can be from present disclosure
All deformations for directly exporting or associating, are considered as protection scope of the present invention.
Claims (7)
1. learning the adaptive optics wavefront compensation method of prediction wavefront zernike coefficient based on depth migration, it is characterised in that:
The following steps are included:
S1, the source domain sample { X for obtaining marks,Ys};Execute step S2;
S2, the aiming field sample for obtaining no markObtain the aiming field sample of markExecute step S3;
S3, basis have the source domain sample { X of marks,YsSource domain data set is obtained, according to the aiming field sample of no markWith
There is the aiming field sample of markObtain aiming field data set;Based on source domain data set and aiming field data set training net
Network optimizes model parameter by back-propagation algorithm, obtains the network for completing training;Execute step S4;
The trained network of S4, the completion for obtaining image input S35, exports the zernike coefficient of prediction, executes step S5;
S5, Wave-front phase is calculated according to the zernike coefficient of step S4 prediction, and carry out wavefront compensation reality using wavefront correction device
Now correct.
2. the adaptive optics wavefront according to claim 1 for learning prediction wavefront zernike coefficient based on depth migration is mended
Compensation method, it is characterised in that: step S3 the following steps are included:
S31. using the source domain network of shared parameter and aiming field network as binary-flow network, judge source domain network and aiming field network
Whether meet the condition of convergence or reach maximum number of iterations, otherwise executes S32, be, execute S35;
S32. it will be divided into multiple small-sized lot data collection after source domain data set and aiming field data set respectively random disorder, form source
Domain batch data and aiming field batch data;Execute step S33;
S33. source domain batch data and aiming field batch data are inputted into source domain network and aiming field network respectively, calculated using backpropagation
Method combination loss function updates each network parameter;Execute step S34;
S34. model parameter is updated using gradient descent algorithm, and judges whether parameter restrains, be to execute S35, it is no to hold
Row S32;
S35. the network of training is completed in output;Execute step S4.
3. the adaptive optics wavefront according to claim 2 for learning prediction wavefront zernike coefficient based on depth migration is mended
Compensation method, it is characterised in that: step S33 includes:
Loss function is defined as:
L=MSE+ λ lCORAL
Wherein MSE is mean square deviation error;MSE is shown below:
Wherein:The respectively predicted value of source domain network and aiming field network;To there is the aiming field sample size of mark;
nsTo there is the source domain sample size of mark;
Covariance loses lCORALIt is shown below:
Wherein,Indicate square matrices Frobenius norm;Cs,CtRespectively indicate the covariance square of source domain and target domain characterization
Battle array, d are sample vector space dimensionality;
λ is the hyper parameter for balancing zernike coefficient prediction and the intensity being distributed between being aligned.
4. the adaptive optics wavefront according to claim 3 for learning prediction wavefront zernike coefficient based on depth migration is mended
Compensation method, it is characterised in that: step S34 includes:
Parameter form renewal are as follows:
Wherein, α is learning rate,For partial derivative operator,Indicate the value of i-th of parameter of kth layer in t time step.
5. the adaptive optics wavefront according to claim 4 for learning prediction wavefront zernike coefficient based on depth migration is mended
Compensation method, it is characterised in that:
Source domain network and aiming field network include input layer-convolutional layer-pond layer-convolutional layer-pond layer-convolutional layer-convolution
Full articulamentum-the adaptation layer of layer-convolutional layer-pond layer-.
6. the adaptive optics wavefront according to claim 5 for learning prediction wavefront zernike coefficient based on depth migration is mended
Compensation method, it is characterised in that:
The setting characteristic pattern number of 1st layer of input layer is 1;The characteristic pattern number of level 2 volume lamination is 32, and convolution kernel size is arranged and is
5;3rd layer of pond layer setting down-sampling is having a size of 2;The characteristic pattern number of 4th layer of convolutional layer is 32, and it is 5 that convolution kernel size, which is arranged,;
5th layer of pond layer setting down-sampling is having a size of 2;The characteristic pattern number of 6th layer of convolutional layer is 64, and it is 3 that convolution kernel size, which is arranged,;The
The characteristic pattern number of 7 layers of convolutional layer is 64, and it is 3 that convolution kernel size, which is arranged,;The characteristic pattern number of 8th layer of convolutional layer is 64, and is arranged
Convolution kernel size is 3;9th layer of pond layer setting down-sampling is having a size of 2;10th layer of full articulamentum setting neuron number is 512;
Neuron number 512 is arranged in 11th layer adaptation layer;12nd layer of output layer setting neuron number is 21, corresponding 2-22 Ze Nike
Coefficient.
7. the adaptive optics wavefront according to claim 6 for learning prediction wavefront zernike coefficient based on depth migration is mended
Compensation method, it is characterised in that:
Hyper parameter λ is 7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910513458.XA CN110349095B (en) | 2019-06-14 | 2019-06-14 | Self-adaptive optical wavefront compensation method for predicting wavefront Zernike coefficient based on deep migration learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910513458.XA CN110349095B (en) | 2019-06-14 | 2019-06-14 | Self-adaptive optical wavefront compensation method for predicting wavefront Zernike coefficient based on deep migration learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110349095A true CN110349095A (en) | 2019-10-18 |
CN110349095B CN110349095B (en) | 2021-08-27 |
Family
ID=68182048
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910513458.XA Active CN110349095B (en) | 2019-06-14 | 2019-06-14 | Self-adaptive optical wavefront compensation method for predicting wavefront Zernike coefficient based on deep migration learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110349095B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110795892A (en) * | 2019-10-23 | 2020-02-14 | 北京邮电大学 | Channel simulation method and device based on generation countermeasure network |
CN111580271A (en) * | 2020-06-24 | 2020-08-25 | 哈工大机器人(中山)无人装备与人工智能研究院 | Self-adaptive aberration correction method and light sheet microscopic imaging device based on same |
CN111986067A (en) * | 2020-07-31 | 2020-11-24 | 重庆邮电大学 | Robust digital watermarking algorithm for non-subsampled contourlet transform |
CN112733859A (en) * | 2021-01-25 | 2021-04-30 | 重庆大学 | Depth migration semi-supervised domain self-adaptive classification method for histopathology image |
CN112975162A (en) * | 2021-04-21 | 2021-06-18 | 武汉华工激光工程有限责任公司 | Ground glass cutting device and method based on adaptive optics |
CN113419342A (en) * | 2021-07-01 | 2021-09-21 | 重庆邮电大学 | Free illumination optical design method based on deep learning |
CN113627541A (en) * | 2021-08-13 | 2021-11-09 | 北京邮电大学 | Light path transmission quality prediction method based on sample migration screening |
CN114022730A (en) * | 2021-10-28 | 2022-02-08 | 中国科学院光电技术研究所 | Point target phase retrieval method based on self-supervision learning neural network |
CN114519309A (en) * | 2022-02-23 | 2022-05-20 | 中国科学院光电技术研究所 | Distorted wavefront prediction method based on deep learning |
CN115374712A (en) * | 2022-10-24 | 2022-11-22 | 中国航天三江集团有限公司 | Method and device for calibrating optical transmission simulation parameters under influence of laser internal channel thermal effect |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20040056174A1 (en) * | 2002-09-24 | 2004-03-25 | Specht Donald Francis | Fast phase diversity wavefront correction using a neural network |
CN101551517A (en) * | 2009-05-11 | 2009-10-07 | 长春理工大学 | Coherent laser communication system based on wavefront correction |
CN101609485A (en) * | 2009-07-24 | 2009-12-23 | 西安电子科技大学 | Medical image diagnostic system and diagnostic method based on the migration nuclear matching tracing |
CN105933060A (en) * | 2016-06-24 | 2016-09-07 | 温州大学 | Wavefront reconstruction method based on dynamics recurrent neural network |
CN106646867A (en) * | 2016-12-16 | 2017-05-10 | 中国科学院光电研究院 | Deep UV optical system confocal alignment device and method |
CN106873152A (en) * | 2017-01-11 | 2017-06-20 | 浙江大学 | A kind of high speed aberration correcting method based on machine learning |
CN107843982A (en) * | 2017-12-01 | 2018-03-27 | 长春理工大学 | Based on real-time phase difference technology without Wavefront detecting adaptive optics system |
CN108983412A (en) * | 2018-07-09 | 2018-12-11 | 北京邮电大学 | A kind of no Wave-front measurement adaptive optics system and beam phase method of adjustment |
CN109031654A (en) * | 2018-09-11 | 2018-12-18 | 安徽农业大学 | A kind of adaptive optics bearing calibration and system based on convolutional neural networks |
CN109212735A (en) * | 2018-10-10 | 2019-01-15 | 浙江大学 | High-speed adaptive optics ring-shaped light spot based on machine learning corrects system and method |
CN109523018A (en) * | 2019-01-08 | 2019-03-26 | 重庆邮电大学 | A kind of picture classification method based on depth migration study |
CN109528161A (en) * | 2018-11-06 | 2019-03-29 | 浙江大学 | A kind of high speed and high resoltuion scanning microscopy imaging System and method for based on machine learning |
-
2019
- 2019-06-14 CN CN201910513458.XA patent/CN110349095B/en active Active
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20040056174A1 (en) * | 2002-09-24 | 2004-03-25 | Specht Donald Francis | Fast phase diversity wavefront correction using a neural network |
CN101551517A (en) * | 2009-05-11 | 2009-10-07 | 长春理工大学 | Coherent laser communication system based on wavefront correction |
CN101609485A (en) * | 2009-07-24 | 2009-12-23 | 西安电子科技大学 | Medical image diagnostic system and diagnostic method based on the migration nuclear matching tracing |
CN105933060A (en) * | 2016-06-24 | 2016-09-07 | 温州大学 | Wavefront reconstruction method based on dynamics recurrent neural network |
CN106646867A (en) * | 2016-12-16 | 2017-05-10 | 中国科学院光电研究院 | Deep UV optical system confocal alignment device and method |
CN106873152A (en) * | 2017-01-11 | 2017-06-20 | 浙江大学 | A kind of high speed aberration correcting method based on machine learning |
CN107843982A (en) * | 2017-12-01 | 2018-03-27 | 长春理工大学 | Based on real-time phase difference technology without Wavefront detecting adaptive optics system |
CN108983412A (en) * | 2018-07-09 | 2018-12-11 | 北京邮电大学 | A kind of no Wave-front measurement adaptive optics system and beam phase method of adjustment |
CN109031654A (en) * | 2018-09-11 | 2018-12-18 | 安徽农业大学 | A kind of adaptive optics bearing calibration and system based on convolutional neural networks |
CN109212735A (en) * | 2018-10-10 | 2019-01-15 | 浙江大学 | High-speed adaptive optics ring-shaped light spot based on machine learning corrects system and method |
CN109528161A (en) * | 2018-11-06 | 2019-03-29 | 浙江大学 | A kind of high speed and high resoltuion scanning microscopy imaging System and method for based on machine learning |
CN109523018A (en) * | 2019-01-08 | 2019-03-26 | 重庆邮电大学 | A kind of picture classification method based on depth migration study |
Non-Patent Citations (4)
Title |
---|
ARTEM ROZANTSEV 等: "Beyond Sharing Weights for Deep Domain Adaptation", 《 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
YOHEI NISHIZAKI等: "Deep learning wavefront sensing", 《OSA PUBLISHING》 * |
周舟: "基于改进的神经网络的自适应波前重构算法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
戴国强 等: "《科技大数据》", 31 August 2018 * |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110795892B (en) * | 2019-10-23 | 2021-10-01 | 北京邮电大学 | Channel simulation method and device based on generation countermeasure network |
CN110795892A (en) * | 2019-10-23 | 2020-02-14 | 北京邮电大学 | Channel simulation method and device based on generation countermeasure network |
CN111580271B (en) * | 2020-06-24 | 2022-02-18 | 哈工大机器人(中山)无人装备与人工智能研究院 | Self-adaptive aberration correction method and light sheet microscopic imaging device based on same |
CN111580271A (en) * | 2020-06-24 | 2020-08-25 | 哈工大机器人(中山)无人装备与人工智能研究院 | Self-adaptive aberration correction method and light sheet microscopic imaging device based on same |
CN111986067A (en) * | 2020-07-31 | 2020-11-24 | 重庆邮电大学 | Robust digital watermarking algorithm for non-subsampled contourlet transform |
CN112733859A (en) * | 2021-01-25 | 2021-04-30 | 重庆大学 | Depth migration semi-supervised domain self-adaptive classification method for histopathology image |
CN112733859B (en) * | 2021-01-25 | 2023-12-19 | 重庆大学 | Depth migration semi-supervised domain self-adaptive classification method for histopathological image |
CN112975162A (en) * | 2021-04-21 | 2021-06-18 | 武汉华工激光工程有限责任公司 | Ground glass cutting device and method based on adaptive optics |
CN113419342A (en) * | 2021-07-01 | 2021-09-21 | 重庆邮电大学 | Free illumination optical design method based on deep learning |
CN113627541A (en) * | 2021-08-13 | 2021-11-09 | 北京邮电大学 | Light path transmission quality prediction method based on sample migration screening |
CN113627541B (en) * | 2021-08-13 | 2023-07-21 | 北京邮电大学 | Optical path transmission quality prediction method based on sample migration screening |
CN114022730A (en) * | 2021-10-28 | 2022-02-08 | 中国科学院光电技术研究所 | Point target phase retrieval method based on self-supervision learning neural network |
CN114022730B (en) * | 2021-10-28 | 2023-08-15 | 中国科学院光电技术研究所 | Point target phase retrieval method based on self-supervision learning neural network |
CN114519309A (en) * | 2022-02-23 | 2022-05-20 | 中国科学院光电技术研究所 | Distorted wavefront prediction method based on deep learning |
CN115374712A (en) * | 2022-10-24 | 2022-11-22 | 中国航天三江集团有限公司 | Method and device for calibrating optical transmission simulation parameters under influence of laser internal channel thermal effect |
Also Published As
Publication number | Publication date |
---|---|
CN110349095B (en) | 2021-08-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110349095A (en) | Learn the adaptive optics wavefront compensation method of prediction wavefront zernike coefficient based on depth migration | |
Guo et al. | Adaptive optics based on machine learning: a review | |
Xin et al. | Object-independent image-based wavefront sensing approach using phase diversity images and deep learning | |
CN106845024B (en) | Optical satellite in-orbit imaging simulation method based on wavefront inversion | |
CN110148157A (en) | Picture target tracking, device, storage medium and electronic equipment | |
Suárez Gómez et al. | Improving adaptive optics reconstructions with a deep learning approach | |
Tang et al. | RestoreNet: a deep learning framework for image restoration in optical synthetic aperture imaging system | |
CN111579097B (en) | High-precision optical scattering compensation method based on neural network | |
CN110378981A (en) | Fourier's lamination microscope pupil restoration methods neural network based | |
Lianghua et al. | A high speed model-based approach for wavefront sensorless adaptive optics systems | |
Mikhaylov et al. | Machine learning aided phase retrieval algorithm for beam splitting with an LCoS-SLM | |
González-Gutiérrez et al. | Multi‐GPU Development of a Neural Networks Based Reconstructor for Adaptive Optics | |
Zepp et al. | Optimization of the holographic wavefront sensor for open-loop adaptive optics under realistic turbulence. Part I: simulations | |
Riesgo et al. | Comparative study of Shack-Hartmann configurations for atmospheric turbulence reconstructions in solar adaptive optics | |
Archinuk et al. | Mitigating the nonlinearities in a pyramid wavefront sensor | |
Delli Veneri et al. | 3D detection and characterization of ALMA sources through deep learning | |
CN117268259A (en) | CNN-based large dynamic range secondary mirror position error detection method | |
WO2023144519A1 (en) | Determining optical aberration | |
Riesgo et al. | Early fully-convolutional approach to wavefront imaging on solar adaptive optics simulations | |
CN206930789U (en) | A kind of subject imaging system based on compressed sensing | |
Norouzi et al. | CNN to mitigate atmospheric turbulence effect on Shack-Hartmann Wavefront Sensing: A case study on the Magdalena Ridge Observatory Interferometer. | |
Krokberg | Reinforcement learning in multi-mirror adaptive optics | |
Taghina | Wavefront sensorless adaptive optics for astronomical applications. | |
Gu et al. | Optronic convolutional neural network for SAR target recognition | |
Barrett et al. | Neural networks for control of telescope adaptive optics |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |