CN108958000B - A kind of optical scanner holography self-focusing method based on classification learning and dichotomy - Google Patents
A kind of optical scanner holography self-focusing method based on classification learning and dichotomy Download PDFInfo
- Publication number
- CN108958000B CN108958000B CN201811079497.5A CN201811079497A CN108958000B CN 108958000 B CN108958000 B CN 108958000B CN 201811079497 A CN201811079497 A CN 201811079497A CN 108958000 B CN108958000 B CN 108958000B
- Authority
- CN
- China
- Prior art keywords
- hologram
- distance
- focusing
- model
- reconstruction
- 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.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 45
- 230000003287 optical effect Effects 0.000 title claims abstract description 32
- 238000001093 holography Methods 0.000 title claims abstract description 26
- 230000006870 function Effects 0.000 claims description 27
- 210000001747 pupil Anatomy 0.000 claims description 20
- 238000012549 training Methods 0.000 claims description 14
- 238000013527 convolutional neural network Methods 0.000 claims description 11
- 230000008569 process Effects 0.000 claims description 11
- 230000004913 activation Effects 0.000 claims description 10
- 210000002569 neuron Anatomy 0.000 claims description 9
- 230000021615 conjugation Effects 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 4
- 238000012546 transfer Methods 0.000 claims description 4
- 230000010354 integration Effects 0.000 claims description 3
- 230000000149 penetrating effect Effects 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 abstract description 7
- 238000010587 phase diagram Methods 0.000 abstract 1
- 239000010410 layer Substances 0.000 description 20
- 238000011160 research Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000009647 digital holographic microscopy Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000009877 rendering Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 239000002356 single layer Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G03—PHOTOGRAPHY; CINEMATOGRAPHY; ANALOGOUS TECHNIQUES USING WAVES OTHER THAN OPTICAL WAVES; ELECTROGRAPHY; HOLOGRAPHY
- G03H—HOLOGRAPHIC PROCESSES OR APPARATUS
- G03H1/00—Holographic processes or apparatus using light, infrared or ultraviolet waves for obtaining holograms or for obtaining an image from them; Details peculiar thereto
- G03H1/04—Processes or apparatus for producing holograms
- G03H1/0443—Digital holography, i.e. recording holograms with digital recording means
Landscapes
- Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Holo Graphy (AREA)
Abstract
The optical scanner holography self-focusing method based on classification learning and dichotomy that the invention proposes a kind of, belongs to optical scanner holography and deep learning field, mainly solves the problems, such as optical scanner holography self-focusing.The present invention is classified using phase diagram of the deep learning to the hologram of reconstruction, and the focal position of hologram is then constantly approached using dichotomy.The present invention effectively, intelligently realizes the self-focusing in optical scanner holography.The method of this self-focusing is suitable for every field.
Description
Technical field
The present invention relates to optical scanner holography fields and deep learning field, it particularly relates to which a kind of be based on taxology
Practise the optical scanner holography self-focusing method with dichotomy.
Background technique
Optical scanner holographic technique, abbreviation OSH, it is using the method for point by point scanning by three-dimension object with two-dimensional digital image
Form storage, compared to general Digital Holography, maximum advantage is the interference of not twin image.It is proposed from the technology
Since, it is applied in multiple fields, such as: the neck such as scanning holographic microscope, 3D rendering identification and 3D optical remote sensing
Domain.
In holographic field, the reconstruction of image is always a research hotspot of this field, and the reconstruction for hologram,
The focal position for finding hologram is committed step in reconstruction process again.In recent years, deep learning becomes Chinese and overseas scholars and grinds
The hot topic direction studied carefully, and the practical problem of every field is solved using deep learning, so that the related skill of every field
Art is more intelligent, efficient.Digital hologram is similarly subjected to similar influence, therefore, it is complete to solve optical scanner using deep learning
The practical problem of breath is also at research hotspot in recent years.
Document " Performance of Autofocus Capability of Deep Convolutional Neural
Networks in Digital Holographic Microscopy " takes the lead in proposing to do hologram using deep neural network
Self-focusing in digital hologram is realized in classification.
Document " Convolutional neural network-based regression for depth
Prediction in digital holography " and document " Learning-based nonparametric
Autofocusing for digital holography " discloses a kind of by knowing in method for distinguishing realization digital hologram
Self-focusing.
But the above method all has that generalization ability is low, can only realize the self-focusing of the hologram of certain positions.
Summary of the invention
It is an object of the invention to realize the self-focusing in optical scanner holography, propose a kind of based on classification learning and two points
The optical scanner holography autohemagglutination set method of method, by the training to sample, is realized complete to rebuilding using depth convolutional neural networks
Cease the reconstruction distance and the size relation of focusing distance of figure;Then, the self-focusing of optical scanner holography is realized by dichotomy.
The technical solution adopted by the present invention is that:
A kind of optical scanner holography self-focusing method based on classification learning and dichotomy, comprising the following steps:
Step 1. is firstly, it is ω that laser, which emits a branch of frequency,0Laser, be divided into two beams by the first beam splitter BS1
Light;Then, wherein light beam passes through the first reflecting mirror M1 and the first pupil p1(x, y) and the first convex lens L1;Meanwhile it is another
Shu Guangbo is promoted to ω by acousto-optic modulator AOFS frequency0+ Ω, and pass through the second reflecting mirror M2 and the second pupil p2(x,
And the second convex lens L2 y);Wherein the first pupil is Di Li carats of δ (x, y) functions, and the second pupil is 1 function of rectangle;
Step 2: two-beam wave is interfered at the second beam splitter BS2 forms Fresnel single-slit diffraction, then anti-through X-Y scanning galvanometer
Scanning object is penetrated, wherein the light wave through object is converged through third convex lens L3, and is received by photoelectric detector PD, it is demodulated
After obtain hologram;
Step 3: repeating step 1 and step 2, repeat to obtain N width hologram, and find focusing by the method for traditional reconstruction
Distance;Then, it is rebuild under different distances using these holograms, record rebuilds the size relation of distance and focusing distance;
Finally, using the size relation of reconstructions distance and the focusing distance of the hologram and record rebuild as trained training data,
Training one can judge to rebuild the model of distance and focusing distance size relation;
Step 4: firstly, a width hologram is reacquired, and initialize two reconstruction distances, two according to step 1 and 2
It is a apart from upper reconstruction hologram;Then, the reconstruction distance for the hologram rebuild by step 3 judgement is closed with the size of focusing distance
System;Finally, obtaining the focusing distance of hologram by dichotomy iteration.
Wherein, detailed process is as follows for acquisition hologram in step 2:
For step 2-1. two-beam wave after BS2 is converged, interference forms Fresnel single-slit diffraction, and anti-by X-Y scanning galvanometer
Penetrate scanning object;The optical transfer function of the process are as follows:
Wherein, x and y indicates the position of object under test, and x' and y' are integration variable, and z indicates x-y scanning galvanometer to determinand
The distance of body,The focal length of expression wave number, λ expression optical wavelength, the first convex lens and the second convex lens is f, kxWith
kyIndicate frequency domain coordinates, p1(x, y) and p2(x, y) respectively indicates the first pupil function and the second pupil function, and subscript * is indicated altogether
Yoke,
Herein, using traditional pupil function, p1(x, y)=δ (x, y), p2(x, y)=1, according to (1) formula, then optics passes
Delivery function may be expressed as:
Step 2-2. interferes the Fresnel single-slit diffraction to be formed to scan object, the device wherein light wave for penetrating object is photoelectrically converted
PD receiving is converted to electric signal, passes to the end computer PC, obtains hologram, which may be expressed as:
G (x, y)=F-1{F[O(x,y;z0)]×OTF(kx,ky;z0)} (3)
Wherein, z0For the position where object, i.e. focusing distance, F-1Inverse Fourier transform and Fourier are respectively indicated with F
Transformation, O (x, y;z0) indicate object amplitude function.
For step 3 using hologram obtained in step 2, training one may determine that rebuild distance closes with focusing distance size
The model of system;The specific steps of which are as follows:
Step 3-1. repeats step 1 and step 2, obtains N width hologram, and found by traditional reconstructing method focus away from
From;Wherein traditional reconstructing method are as follows:
H (x, y)=F-1{F{g(x,y)}×OTF*(kx,ky,zi)} (4)
Wherein, H (x, y) indicates that the image rebuild, subscript * indicate conjugation, ziIt indicates to rebuild distance;By observing every width weight
The hologram built, when the hologram of reconstruction clearly displays object information, the as focusing distance z of hologrami=z0;
The N width hologram that step 3-2. utilizes step 3-1 to obtain, in different reconstruction distance ziIt is lower to rebuild M times, thus
Hologram is rebuild to N × M width;During reconstruction, compares reconstruction distance and records as the following formula:
Wherein, ε is the focusing distance of model prediction and the error size that true focusing distance allows;Here ε=1;
N × M width that step 3-3. is obtained using step 3-2 rebuilds the label of hologram and record as training data,
Training depth convolutional neural networks obtain the model that can judge to rebuild distance and the size relation of focusing distance, are denoted as letter
Number Model;Wherein depth convolutional neural networks structure is as follows:
Input: input layer, the hologram as rebuild;Layer1: convolutional layer 1, includes 64 7 × 7 convolution kernels, 2 ×
2 pond layer and ReLU activation primitive;Layer2: convolutional layer 2, includes 128 5 × 5 convolution kernels, 2 × 2 pond layer,
And ReLU activation primitive;Layer3: convolutional layer 3, includes 128 5 × 5 convolution kernels, and 2 × 2 pond layer and ReLU swashs
Function living;Layer4: convolutional layer 4 includes 256 3 × 3 convolution kernels, 2 × 2 pond layer and ReLU activation primitive;
FC1: full articulamentum 1 includes 1024 neurons;FC2: full articulamentum 2 includes 1024 neurons;Output: output layer,
Comprising 3 neurons, 3 classes that as mark.
Step 4. is judged to rebuild the size of distance and focusing distance using the model that step 3 obtains, be looked for using dichotomy
To focal position;The specific steps of which are as follows:
Step 4-1. reacquires a width hologram according to step 1 and 2;
Step 4-2. finds focusing distance using dichotomy;
Step 4-2-1. is firstly, original reconstruction distance z1、z2, respectively in z1、z2It rebuilds hologram and obtains H1And H2;
If step 4-2-2. Model (H1)*Model(H2) > 0, return step 4-2-1;If Model (H1)*Model
(H1) < 0, enter step 4-2-3;If Model (H1)=0 or Model (H2)=0, then export z1Or z2, and exit step 4-
2;
Step 4-2-3.And in z3It rebuilds hologram and obtains H3;
If step 4-2-4. Model (H3)=0 exports z3, and exit step 4-2;If Model (H3)*Model
(H1) < 0, then enable z1=z1, z2=z3, return step 4-2-3;If Model (H3)*Model(H2) < 0, then enable z1=z3, z2
=z2, return step 4-2-3;When | z1-z2| < thresh exports z3, exit step 4-2, wherein thresh is a threshold value,
For stopping criterion for iteration.
The distance of step 4-3. step 4-2 output is focusing distance.
The beneficial effects of the present invention are:
(1) present invention uses depth convolutional neural networks to carry out classification learning, and is focused using dichotomy close approximation
Distance, to realize the self-focusing under optical scanner holography.
(2) present invention efficiently solves oneself under optical scanner holography using the method for classification learning and dichotomy
Focus issues.
(3) compared to the prior art, method is more simple and efficient this method, and effect is more excellent, and more intelligent.
(4) not only implementation is simple, is convenient for operation by the present invention, while having very strong usability.
Detailed description of the invention
Fig. 1 is the basic block diagram that the method for the invention uses;
Fig. 2 is the object of scanning used in the embodiment of the present invention;
Fig. 3 is that the embodiment of the present invention obtains obtaining holographic legend;
Fig. 4 is that the embodiment of the present invention rebuilds hologram in the different phase distribution samples rebuild under distance, wherein (a) is rebuild
Distance is less than focusing distance, (b) rebuilds distance and is greater than focusing distance;
Fig. 5 is depth convolutional neural networks model used in the embodiment of the present invention;
Fig. 6 is that the embodiment of the present invention uses the present invention to obtain the reconstruction image after focusing distance.
Specific embodiment
With reference to the accompanying drawing with embodiment to the present invention into further explanation, embodiments of the present invention include but is not limited to
The following example.
Embodiment:
Basic structure used by the embodiment of the present invention is as shown in Figure 1, wherein wavelength X=632.8nm of light wave, convex lens
(L1And L2) focal length be all 75mm, the monolayer slices z that X-Y scanning galvanometer arrives0=40mm, scanned object sample such as Fig. 2 institute
Show, for the slice of use having a size of 1mm × 1mm, sampling pixel points are 256 × 256.
Specific step is as follows for the present embodiment:
Step 1. is firstly, it is ω that laser, which emits a branch of frequency,0Laser, be divided into two beams by the first beam splitter BS1
Light;Then, wherein light beam passes through the first reflecting mirror M1 and the first pupil p1(x, y) and the first convex lens L1;Meanwhile it is another
Shu Guangbo is promoted to ω by acousto-optic modulator AOFS frequency0+ Ω, and pass through the second reflecting mirror M2 and the second pupil p2(x,
And the second convex lens L2 y);Wherein the first pupil is Di Li carats of δ (x, y) functions, and the second pupil is 1 function of rectangle;
Step 2: two-beam wave is interfered at the second beam splitter BS2 forms Fresnel single-slit diffraction, then anti-through X-Y scanning galvanometer
Scanning object is penetrated, wherein the light wave through object is converged through third convex lens L3, and is received by photoelectric detector PD, it is demodulated
After obtain hologram;
Wherein, detailed process is as follows for step 2:
For step 2-1. two-beam wave after BS2 is converged, interference forms Fresnel single-slit diffraction, and anti-by X-Y scanning galvanometer
Penetrate scanning object;The optical transfer function of the process are as follows:
Wherein, x and y indicates the position of object under test, and x' and y' are integration variable, and z indicates x-y scanning galvanometer to determinand
The distance of body,The focal length of expression wave number, λ expression optical wavelength, the first convex lens and the second convex lens is f, kxWith
kyIndicate frequency domain coordinates, p1(x, y) and p2(x, y) respectively indicates the first pupil function and the second pupil function, and subscript * is indicated altogether
Yoke,
Herein, using traditional pupil function, p1(x, y)=δ (x, y), p2(x, y)=1, according to (1) formula, then optics passes
Delivery function may be expressed as:
Step 2-2. interferes the Fresnel single-slit diffraction to be formed to scan object, the device wherein light wave for penetrating object is photoelectrically converted
PD receiving is converted to electric signal, passes to the end computer PC, obtains hologram, which may be expressed as:
G (x, y)=F-1{F[O(x,y;z0)]×OTF(kx,ky;z0)} (3)
Wherein, z0For the position where object, i.e. focusing distance, F-1Inverse Fourier transform and Fourier are respectively indicated with F
Transformation, O (x, y;z0) indicate object amplitude function, determined by object.
Step 3: repeating step 1 and step 2, repeat to obtain N width hologram, and find focusing by the method for traditional reconstruction
Distance;Then, it is rebuild under different distances using these holograms, record rebuilds the size relation of distance and focusing distance;
Finally, using the size relation of reconstructions distance and the focusing distance of the hologram and record rebuild as trained training data,
Training one can judge to rebuild the model of distance and focusing distance size relation;
Detailed process is as follows for step 3:
Step 3-1. repeats step 1 and step 2, obtains N width hologram, and wherein holographic pattern is for example shown in Fig. 3, and passes through
Traditional reconstructing method finds focusing distance;Wherein traditional reconstructing method are as follows:
H (x, y)=F-1{F{g(x,y)}×OTF*(kx,ky,zi)} (4)
Wherein, H (x, y) indicates that the image rebuild, subscript * indicate conjugation, ziIt indicates to rebuild distance;By observing every width weight
The hologram built, when the hologram of reconstruction clearly displays object information, the as focusing distance z of hologrami=z0;
The N width hologram that step 3-2. utilizes step 3-1 to obtain, in different reconstruction distance ziIt is lower to rebuild M times, thus
Hologram is rebuild to N × M width;During reconstruction, compares reconstruction distance and records as the following formula:
Wherein, ε is the focusing distance of model prediction and the error size that true focusing distance allows;Here ε=1;Weight
Holographic sample figure is built as shown in figure 4, (a) indicates to rebuild master drawing when distance is less than focusing distance, (b) expression is rebuild apart from university
Master drawing when focusing distance.
N × M width that step 3-3. is obtained using step 3-2 rebuilds the label of hologram and record as training data,
Training depth convolutional neural networks obtain the model that can judge to rebuild distance and the size relation of focusing distance, are denoted as letter
Number Model;Wherein depth convolutional neural networks are as shown in figure 5, structure is as follows:
Input: input layer, the hologram as rebuild;Layer1: convolutional layer 1, includes 64 7 × 7 convolution kernels, 2 ×
2 pond layer and ReLU activation primitive;Layer2: convolutional layer 2, includes 128 5 × 5 convolution kernels, 2 × 2 pond layer,
And ReLU activation primitive;Layer3: convolutional layer 3, includes 128 5 × 5 convolution kernels, and 2 × 2 pond layer and ReLU swashs
Function living;Layer4: convolutional layer 4 includes 256 3 × 3 convolution kernels, 2 × 2 pond layer and ReLU activation primitive;
FC1: full articulamentum 1 includes 1024 neurons;FC2: full articulamentum 2 includes 1024 neurons;Output: output layer,
Comprising 3 neurons, 3 classes that as mark.
Step 4: firstly, a width hologram is reacquired, and initialize two reconstruction distances, two according to step 1 and 2
It is a apart from upper reconstruction hologram;Then, the reconstruction distance for the hologram rebuild by step 3 judgement is closed with the size of focusing distance
System;Finally, obtaining the focusing distance of hologram by dichotomy iteration.
Detailed process is as follows for step 4:
Step 4-1. reacquires a width hologram according to step 1 and 2;
Step 4-2. finds focusing distance using dichotomy;
Step 4-2-1. is firstly, original reconstruction distance z1、z2, respectively in z1、z2It rebuilds hologram and obtains H1And H2;
If step 4-2-2. Model (H1)*Model(H2) > 0, return step 4-2-1;If Model (H1)*Model
(H1) < 0, enter step 4-2-3;If Model (H1)=0 or Model (H2)=0, then export z1Or z2, and exit step 4-
2;
Step 4-2-3.And in z3It rebuilds hologram and obtains H3;
If step 4-2-4. Model (H3)=0 exports z3, and exit step 4-2;If Model (H3)*Model
(H1) < 0, then enable z1=z1, z2=z3, return step 4-2-3;If Model (H3)*Model(H2) < 0, then enable z1=z3, z2
=z2, return step 4-2-3;When | z1-z2| < thresh exports z3, exit step 4-2, wherein thresh is a threshold value,
For stopping criterion for iteration.
The distance of step 4-3. step 4-2 output is focusing distance z=40.078mm, in the figure that the focal position is rebuild
As shown in Figure 6.
The present embodiment has been implemented in combination with the self-focusing in optical scanner holography using classification learning and dichotomy, compared to existing
Some conventional methods are more intelligent, more efficient, realize that the method generalization ability of self-focusing is more preferable using deep learning compared to other,
With more universality.
Claims (7)
1. a kind of optical scanner holography self-focusing method based on classification learning and dichotomy, which is characterized in that including following step
It is rapid:
Step 1. is firstly, it is ω that laser, which emits a branch of frequency,0Laser, be divided into two-beam by the first beam splitter;Then,
Wherein light beam is by the first reflecting mirror and the first pupil and the first convex lens;Meanwhile another Shu Guangbo passes through acousto-optic modulation
Device frequency is promoted to ω0+ Ω, and by the second reflecting mirror and the second pupil and the second convex lens;
Step 2: two-beam wave interferes to form Fresnel single-slit diffraction at the second beam splitter, then reflects and scan through X-Y scanning galvanometer
Object wherein the light wave through object is converged through third convex lens, and is received by photodetector, obtains holography after demodulated
Figure;
Step 3: repeat step 1 and step 2, repeat to obtain N width hologram, and found by the method for traditional reconstruction focus away from
From;Then, it is rebuild under different distances using these holograms, record rebuilds the size relation of distance and focusing distance;Most
Afterwards, using the size relation of the reconstruction distance of the hologram of reconstruction and record and focusing distance as training training data, instruction
Practice the model that can judge to rebuild distance and focusing distance size relation;
Step 4: firstly, according to step 1 and 2, reacquire a width hologram, and initialize two reconstruction distances, two away from
From upper reconstruction hologram;Then, the reconstruction distance and the size relation of focusing distance for the hologram rebuild by step 3 judgement;
Finally, obtaining the focusing distance of hologram by dichotomy iteration.
2. the optical scanner holography self-focusing method according to claim 1 based on classification learning and dichotomy, feature
It is, wherein detailed process is as follows for step 2:
Step 2-1. two-beam wave is after the convergence of the second beam splitter, and interference forms Fresnel single-slit diffraction, and by X-Y scanning galvanometer
Reflection scanning object;The optical transfer function of the process are as follows:
Wherein, x and y indicates the position of object under test, and x' and y' are integration variable, and z indicates x-y scanning galvanometer to object under test
Distance,The focal length of expression wave number, λ expression optical wavelength, the first convex lens and the second convex lens is f, kxAnd kyTable
Show frequency domain coordinates, p1(x, y) and p2(x, y) respectively indicates the first pupil function and the second pupil function, and subscript * indicates conjugation,
Step 2-2. interferes the Fresnel single-slit diffraction to be formed to scan object, the device receiving wherein the light wave for penetrating object is photoelectrically converted
Electric signal is converted to, computer end is passed to, obtains hologram, which may be expressed as:
G (x, y)=F-1{F[O(x,y;z0)]×OTF(kx,ky;z0)} (3)
Wherein, z0For the position where object, i.e. focusing distance, F-1Inverse Fourier transform and Fourier transformation are respectively indicated with F,
O(x,y;z0) indicate object amplitude function.
3. the optical scanner holography self-focusing method according to claim 2 based on classification learning and dichotomy, feature
It is, detailed process is as follows for step 3:
Step 3-1. repeats step 1 and step 2, obtains N width hologram, and find focusing distance by traditional reconstructing method;Its
Middle traditional reconstructing method are as follows:
H (x, y)=F-1{F{g(x,y)}×OTF*(kx,ky,zi)} (4)
Wherein, H (x, y) indicates that the image rebuild, subscript * indicate conjugation, ziIt indicates to rebuild distance;It is rebuild by observing every width
Hologram, when the hologram of reconstruction clearly displays object information, the as focusing distance z of hologrami=z0;
The N width hologram that step 3-2. utilizes step 3-1 to obtain, in different reconstruction distance ziIt is lower rebuild M times, thus obtain N ×
M width rebuilds hologram;During reconstruction, compares reconstruction distance and records as the following formula:
Wherein, ε is the focusing distance of model prediction and the error size that true focusing distance allows;
Step 3-3. rebuilds the label of hologram and record as training data, training using N × M width that step 3-2 is obtained
Depth convolutional neural networks obtain the model that can judge to rebuild distance and the size relation of focusing distance, are denoted as function
Model;Wherein depth convolutional neural networks structure is as follows:
Input: input layer, the hologram as rebuild;Layer1: convolutional layer 1, includes 64 7 × 7 convolution kernels, 2 × 2
Pond layer and ReLU activation primitive;Layer2: convolutional layer 2, includes 128 5 × 5 convolution kernels, 2 × 2 pond layer, with
And ReLU activation primitive;Layer3: convolutional layer 3 includes 128 5 × 5 convolution kernels, 2 × 2 pond layer and ReLU activation
Function;Layer4: convolutional layer 4 includes 256 3 × 3 convolution kernels, 2 × 2 pond layer and ReLU activation primitive;FC1:
Full articulamentum 1 includes 1024 neurons;FC2: full articulamentum 2 includes 1024 neurons;Output: output layer includes 3
A neuron, 3 classes as marked.
4. the optical scanner holography self-focusing method according to claim 3 based on classification learning and dichotomy, feature
It is, detailed process is as follows for step 4:
Step 4-1. reacquires a width hologram according to step 1 and 2;
Step 4-2. finds focusing distance using dichotomy;
Step 4-2-1. is firstly, original reconstruction distance z1、z2, respectively in z1、z2It rebuilds hologram and obtains H1And H2;
If step 4-2-2. Model (H1)*Model(H2) > 0, return step 4-2-1;If Model (H1)*Model(H1)
< 0, enters step 4-2-3;If Model (H1)=0 or Model (H2)=0, then export z1Or z2, and exit step 4-2;
Step 4-2-3.And in z3It rebuilds hologram and obtains H3;
If step 4-2-4. Model (H3)=0 exports z3, and exit step 4-2;If Model (H3)*Model(H1) < 0,
Then enable z1=z1, z2=z3, return step 4-2-3;If Model (H3)*Model(H2) < 0, then enable z1=z3, z2=z2, return
Return step 4-2-3;When | z1-z2| < thresh exports z3, exit step 4-2, wherein thresh is a threshold value, is that iteration is whole
Only condition;
The distance of step 4-3. step 4-2 output is focusing distance.
5. the optical scanner holography self-focusing method according to claim 1 based on classification learning and dichotomy, feature
It is, in step 1, the first pupil is Di Li carats of δ (x, y) functions, and the second pupil is 1 function of rectangle.
6. the optical scanner holography self-focusing method according to claim 5 based on classification learning and dichotomy, feature
It is, in step 2-1, p1(x, y)=δ (x, y), p2(x, y)=1, according to (1) formula, then optical transfer function may be expressed as:
7. the optical scanner holography self-focusing method according to claim 3 based on classification learning and dichotomy, feature
It is, in step 3-2, ε=1.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811079497.5A CN108958000B (en) | 2018-09-17 | 2018-09-17 | A kind of optical scanner holography self-focusing method based on classification learning and dichotomy |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811079497.5A CN108958000B (en) | 2018-09-17 | 2018-09-17 | A kind of optical scanner holography self-focusing method based on classification learning and dichotomy |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108958000A CN108958000A (en) | 2018-12-07 |
CN108958000B true CN108958000B (en) | 2019-06-21 |
Family
ID=64476764
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811079497.5A Expired - Fee Related CN108958000B (en) | 2018-09-17 | 2018-09-17 | A kind of optical scanner holography self-focusing method based on classification learning and dichotomy |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108958000B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109884869A (en) * | 2019-04-22 | 2019-06-14 | 重庆邮电大学 | One kind calculating holographic imaging systems and method based on dicoria heterodyne |
Family Cites Families (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20060282275A1 (en) * | 2005-06-08 | 2006-12-14 | Pineda Douglass D | Survey Method for Facilitating Real Estate Transactions |
CN103713097B (en) * | 2013-07-30 | 2015-04-15 | 山东建筑大学 | Large-area water body sediment heavy metal contamination situation investigation point position layout method |
EP2887322B1 (en) * | 2013-12-18 | 2020-02-12 | Microsoft Technology Licensing, LLC | Mixed reality holographic object development |
KR101558235B1 (en) * | 2014-01-17 | 2015-10-13 | 전자부품연구원 | Holographic wave-front Recording Apparatus and Method for Seamless Color Holographic Image Display |
CN105023016B (en) * | 2015-06-25 | 2018-08-28 | 中国计量学院 | Target apperception method based on compressed sensing classification |
CN105204311B (en) * | 2015-07-06 | 2018-05-18 | 电子科技大学 | A kind of optical scanner holography edge detection method based on Gauss apodization |
CN106303481B (en) * | 2016-09-09 | 2019-03-19 | 深圳市Tcl高新技术开发有限公司 | A kind of method and system of projection TV focusing |
TWI644098B (en) * | 2017-01-05 | 2018-12-11 | 國立臺灣師範大學 | Method and apparatus for defect inspection of transparent substrate |
CN108089425B (en) * | 2018-01-16 | 2019-09-24 | 电子科技大学 | A method of the elimination optical scanner holography defocus noise based on deep learning |
-
2018
- 2018-09-17 CN CN201811079497.5A patent/CN108958000B/en not_active Expired - Fee Related
Also Published As
Publication number | Publication date |
---|---|
CN108958000A (en) | 2018-12-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Fienup et al. | Phase retrieval for a complex-valued object by using a low-resolution image | |
Rivenson et al. | Compressive multiple view projection incoherent holography | |
CN109597291B (en) | Optical scanning holographic image identification method based on convolutional neural network | |
CN108089425B (en) | A method of the elimination optical scanner holography defocus noise based on deep learning | |
Yin et al. | Digital holographic reconstruction based on deep learning framework with unpaired data | |
Wenchao et al. | Research on strawberry disease diagnosis based on improved residual network recognition model | |
Sarmis et al. | High resolution reflection tomographic diffractive microscopy | |
Jiang et al. | Fourier single pixel imaging reconstruction method based on the U-net and attention mechanism at a low sampling rate | |
CN108958000B (en) | A kind of optical scanner holography self-focusing method based on classification learning and dichotomy | |
Chen et al. | DH-GAN: a physics-driven untrained generative adversarial network for holographic imaging | |
Jiao et al. | Enhanced autofocusing in optical scanning holography based on hologram decomposition | |
Zhao et al. | Photon-efficient 3D reconstruction employing a edge enhancement method | |
Leihong et al. | Restoration of single pixel imaging in atmospheric turbulence by Fourier filter and CGAN | |
Wang et al. | Resolution enhancement in microscopic imaging based on generative adversarial network with unpaired data | |
CN105184295A (en) | Holographic scanning space distance extracting method based on wavelet transform and connected domain | |
CN107967703B (en) | A kind of self-focusing of the optical scanner holography based on connected domain and method for reconstructing | |
Fu et al. | Multi‐style Chinese art painting generation of flowers | |
He et al. | Further improvements to iterative off-axis digital holography | |
Balasubramani et al. | Influence of noise-reduction techniques in sparse-data sample rotation tomographic imaging | |
Terbe et al. | Deep-learning-based bright-field image generation from a single hologram using an unpaired dataset | |
Mahesh et al. | Deep convolutional neural network for binary regression of three-dimensional objects using information retrieved from digital Fresnel holograms | |
Belaid et al. | Application of two-dimensional wavelet transform to hologram analysis: visualization of glass fibers in a turbulent flame | |
Zhang et al. | Lens-free auto-focusing imaging algorithm for the ultra-broadband light source | |
Chen et al. | Quantitative phase imaging based on model transfer learning | |
Ramaiah et al. | Dynamic displacement measurement in digital holographic interferometry using eigenspace analysis |
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 | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190621 |