CN111476125A - Three-dimensional fluorescence microscopic signal denoising method based on generation countermeasure network - Google Patents
Three-dimensional fluorescence microscopic signal denoising method based on generation countermeasure network Download PDFInfo
- Publication number
- CN111476125A CN111476125A CN202010227291.3A CN202010227291A CN111476125A CN 111476125 A CN111476125 A CN 111476125A CN 202010227291 A CN202010227291 A CN 202010227291A CN 111476125 A CN111476125 A CN 111476125A
- Authority
- CN
- China
- Prior art keywords
- data
- denoising
- convolution
- network
- layer
- 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.)
- Withdrawn
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Signal Processing (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Investigating, Analyzing Materials By Fluorescence Or Luminescence (AREA)
Abstract
The invention discloses a three-dimensional fluorescence microscopic signal denoising method based on a generation countermeasure network, which comprises the following steps: building a denoising generation network, wherein the denoising generation network comprises a plurality of first convolution layers, a first maximum pooling layer, an up-sampling layer, a jump connection layer, and a first normalization operation and a first excitation operation which are connected behind each convolution layer; constructing a denoising discrimination network, wherein the denoising discrimination network comprises a plurality of convolution layers, a maximum pooling layer, a full-connection layer, and a second normalization operation, a second excitation operation and two classified outputs which are connected behind each convolution layer; establishing training data and test sample data; training a denoising generation network and a denoising discrimination network by taking a minimum cost function as a target according to the training data and the test sample data; and inputting the fluorescence microscopic data into a denoising generation network to obtain a denoising result. The method can effectively improve the denoising capability of the three-dimensional fluorescence microscopic signal with low signal-to-noise ratio, and is more reliable and accurate.
Description
Technical Field
The invention relates to the technical field of computational imaging, microscopic imaging and machine learning, in particular to a three-dimensional fluorescence microscopic signal denoising method based on a generation countermeasure network.
Background
Fluorescence signal three-dimensional imaging plays a very important role in microscopic imaging, including three-dimensional structural imaging and three-dimensional functional imaging. However, these fluorophores in vivo are limited by the maximum photon exposure, and the excessive light intensity can cause bleaching or inactivation of the fluorophores in vivo, so that the observation of the fluorescence signal must be performed under limited illumination intensity, so that the observation results need to be balanced between imaging speed, spatial resolution and imaging depth.
It is a very significant task how to obtain high signal-to-noise ratio fluorescence data similar to that obtained at long exposures from low signal-to-noise ratio fluorescence data obtained at short exposures. The method does not need additional hardware equipment, and only needs to carry out denoising and analysis on the acquired low signal-to-noise ratio data, so that a fluorescence result with a high signal-to-noise ratio is obtained.
In the related technology, with the rapid development of the deep learning technology, the network training is performed through the training data, so that tasks such as classification, regression and the like can be better solved, and ideal effects can be obtained in tasks such as segmentation, tracking and the like. In the field of microscopic imaging, the convolutional neural network technology has achieved certain effects in the fields of cell classification, cancer cell segmentation, electron microscope chromatography and the like, and provides certain directions and ways for the application of the technology in other fields and tasks.
Disclosure of Invention
The present invention is directed to solving, at least to some extent, one of the technical problems in the related art.
Therefore, the invention aims to provide a three-dimensional fluorescence microscopic signal denoising method based on a generation countermeasure network, which can effectively improve the denoising capability of a three-dimensional fluorescence microscopic signal with a low signal-to-noise ratio and is more reliable and accurate.
In order to achieve the above object, an embodiment of the present invention provides a three-dimensional fluorescence microscopic signal denoising method based on a generated countermeasure network, including the following steps: step S1: building a denoising generation network, wherein the denoising generation network comprises a plurality of first convolution layers, a first maximum pooling layer, an up-sampling layer, a jump connection layer, and a first normalization operation and a first excitation operation which are connected behind each convolution layer; step S2: building a denoising discrimination network, wherein the denoising discrimination network comprises a plurality of convolution layers, a maximum pooling layer, a full-connection layer, and a second normalization operation, a second excitation operation and two classified outputs which are connected behind each convolution layer; step S3: establishing training data and test sample data; step S4: training the denoising generation network and the denoising discriminant network according to the training data and the test sample data by taking a minimum cost function as a target; step S5: and inputting the fluorescence microscopic data into the denoising generation network to obtain a denoising result.
According to the three-dimensional fluorescence microscopic signal denoising method based on the generation countermeasure network, the three-dimensional fluorescence data with low signal-to-noise ratio can be denoised to obtain the data with high signal-to-noise ratio by training the denoising generation network and the denoising judgment network, the effect of the three-dimensional fluorescence signal is obviously improved, the denoising capability of the three-dimensional fluorescence microscopic signal with low signal-to-noise ratio can be effectively improved, and the method is more reliable and accurate.
In addition, the three-dimensional fluorescence microscopic signal denoising method based on the generation countermeasure network according to the above embodiment of the present invention may further have the following additional technical features:
optionally, in an embodiment of the present invention, the step S1 includes: step S11: building a first convolution part, wherein the first convolution part is 2 layers of 3D convolution layers, and a maximum pooling layer is correspondingly connected behind each first convolution layer; step S12: building an upper convolution part, wherein the upper convolution part is 2 layers of 3D convolution layers, an upper sampling layer is correspondingly connected behind each convolution layer, and a jump connection layer is simultaneously connected behind each convolution layer to combine data before corresponding pooling; step S13: and constructing an output part of the whole generated network model, wherein the output part comprises 1 layer of convolution layer, so as to obtain an output result by adding the convolution layer with input data, the output result is data with the same size as the input data, and the input data is a denoised result.
Optionally, in an embodiment of the present invention, the step S2 includes: step S21: building a second convolution part, wherein the second convolution part is 4 layers of 3D convolution layers, and a maximum pooling layer is correspondingly connected behind each second convolution layer; step S22: building a full connection part; step S23: and constructing an output part of the whole convolutional neural network model, wherein the output is one number, and the current input data is original high signal-to-noise ratio data.
Further, in an embodiment of the present invention, the input in step S1 is three-dimensional low snr fluorescence data of 64x64x16, the three-dimensional low snr fluorescence is acquired randomly or through a sliding window from a sequence stack of low snr fluorescence images captured by a camera, and the target output is three-dimensional high snr fluorescence data of 64x64x16, the three-dimensional high snr fluorescence data is acquired from a corresponding input data region in the sequence stack of high snr fluorescence images captured by the camera.
Further, in an embodiment of the present invention, the input in step S2 is 64x64x16 raw high snr data or denoised low snr data, the denoised data is three-dimensional fluorescence data output by the denoising network in step S1, and the raw data is obtained from a corresponding input data region in the sequence stack of high snr fluorescence images captured by the camera.
Further, in an embodiment of the present invention, in the step S3, the training data includes 10000 fluorescence region data and has a size of 64x64x16, wherein different signal-to-noise ratio data in the same region are obtained by using different illumination intensities; and acquiring different signal-to-noise ratio data in the same area by using different exposure time, wherein the test sample data is low signal-to-noise ratio data in different areas.
Further, in an embodiment of the present invention, the denoising discriminant network cost function in step S4 is as follows:
L1(z,z*)＝-z*·log(z)-(1-z*)·log(1-z)，
wherein z is a probability value obtained by the low signal-to-noise ratio data through the output of the denoising discrimination network, and z is*Is whether the training sample belongs to the original high signal-to-noise ratio data.
Further, in an embodiment of the present invention, the denoising generating network cost function in step S4 is as follows:
L2(y,y*)＝sum(y-y*)2+log(1-z)，
wherein, the low signal-to-noise ratio data is denoised data y obtained by the denoising generation network output*And the data is the corresponding high signal-to-noise ratio true value data in the training sample.
Further, in an embodiment of the present invention, the network weight initialization in step S4 employs a one-dimensional gaussian distribution, and utilizes an adammoptimizer method to minimize the loss function.
Further, in an embodiment of the present invention, the network weight value in step S4 is obtained by minimizing the loss function until no more decrease.
Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
The foregoing and/or additional aspects and advantages of the present invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:
fig. 1 is a flowchart of a three-dimensional fluorescence microscopic signal denoising method based on a generation countermeasure network according to an embodiment of the invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the drawings are illustrative and intended to be illustrative of the invention and are not to be construed as limiting the invention.
The three-dimensional fluorescence microscopic signal denoising method based on the generation countermeasure network proposed by the embodiment of the invention is described below with reference to the accompanying drawings.
FIG. 1 is a flowchart of a three-dimensional fluorescence microscopy signal denoising method based on generation of a countermeasure network according to an embodiment of the present invention.
As shown in FIG. 1, the three-dimensional fluorescence microscopic signal denoising method based on the generation countermeasure network comprises the following steps:
step S1: and building a denoising generation network, wherein the denoising generation network comprises a plurality of first convolution layers, a first maximum pooling layer, an up-sampling layer, a jump connection layer, and a first normalization operation and a first excitation operation which are connected behind each convolution layer.
It can be understood that a denoising generating network is firstly built, the generating network comprises a plurality of convolutional layers, a max pooling layer, an upsampling layer, a jump connection layer, and a Normalization operation and an excitation operation which are connected behind each convolutional layer, wherein the Normalization operation adopts Batch Normalization, and the excitation operation adopts a Re L U function.
Optionally, in an embodiment of the present invention, step S1 includes:
step S11: and building a first convolution part, wherein the first convolution part is 2 layers of 3D convolution layers, and a maximum pooling layer is correspondingly connected behind each first convolution layer.
Step S12: and constructing an upper convolution part, wherein the upper convolution part is 2 layers of 3D convolution layers, an upper sampling layer is correspondingly connected behind each convolution layer, and a jump connection layer is simultaneously connected to combine data before corresponding pooling.
Step S13: and constructing an output part of the whole generated network model, wherein the output part comprises 1 layer of convolution layer to be added with input data to obtain an output result, the output result is data with the same size as the input data, and the input data is a denoised result.
Specifically, the step S1 of building the denoising generating network comprises the steps of S11 building a convolution part which is a 2-layer 3D convolution layer, each convolution layer comprises convolution operation, Batchnormalization operation and Re L U excitation operation, a maximum pooling layer is connected behind each convolution layer, S12 building a rear half part of the whole generating network model, namely building an upper convolution part which is a 2-layer 3D convolution layer, each convolution layer comprises convolution operation, Batchnormalization operation and Re L U excitation operation, an upper sampling layer is connected behind each convolution layer and a jump connection layer is connected to each convolution layer to combine data before the corresponding pooling, and S13 building an output part of the whole generating network model, wherein the output part comprises 1-layer of convolution layer, each convolution layer comprises convolution operation, Batchnormalization operation and Re L U excitation operation, finally, the output result and input data are obtained, and the output result is the same as the input data, namely the input data and the denoising result is added to the input data.
Further, in an embodiment of the present invention, the input in step S1 is three-dimensional low snr fluorescence data of 64x64x16, the three-dimensional low snr fluorescence is acquired randomly or through a sliding window from a sequence stack of low snr fluorescence images captured by a camera, the target output is three-dimensional high snr fluorescence data of 64x64x16, and the three-dimensional high snr fluorescence data is acquired from a corresponding input data region in the sequence stack of high snr fluorescence images captured by the camera.
That is, the input in step S1 is three-dimensional low signal-to-noise ratio fluorescence data of 64x64x16, which is acquired randomly or through a sliding window from a sequence stack of low signal-to-noise ratio fluorescence images captured by a camera, and the target output is three-dimensional high signal-to-noise ratio fluorescence data of 64x64x16, which is acquired from a corresponding input data region in a sequence stack of high signal-to-noise ratio fluorescence images captured by a camera.
Step S2: and constructing a denoising discrimination network, wherein the denoising discrimination network comprises a plurality of convolution layers, a maximum pooling layer, a full-connection layer, and a second normalization operation, a second excitation operation and two classified outputs which are connected behind each convolution layer.
It can be understood that a denoising discriminant network is constructed, the discriminant network comprises a plurality of convolutional layers, a max-pooling layer, a full-link layer, and a Normalization operation, an excitation operation and a two-class output, which are connected behind each convolutional layer, the Normalization operation adopts Batch Normalization, and the excitation operation adopts a Re L U function.
Optionally, in an embodiment of the present invention, step S2 includes:
step S21: and constructing a second convolution part, wherein the second convolution part is 4 layers of 3D convolution layers, and each second convolution layer is correspondingly connected with one maximum pooling layer behind.
Step S22: and building a full connection part.
Step S23: and constructing an output part of the whole convolutional neural network model, wherein the output is one number, and the current input data is original high signal-to-noise ratio data.
Specifically, the step S2 of building the denoising discrimination network comprises the steps of S21 building a convolution part, namely the front half part of the whole discrimination network model, wherein the convolution part is 4 layers of 3D convolution layers, each convolution layer comprises convolution operation, BatchNormal operation and Re L U excitation operation, a maximum pooling layer is connected behind each convolution layer, S22 building a rear half part of the whole discrimination network model, namely the full connection part is 2 layers of full connection layers, and S23 outputting the whole convolution neural network model, namely outputting two categories, namely outputting one number, which represents the probability that the current input data is original high signal-to-noise ratio data instead of denoised low signal-to-noise ratio data.
Further, in an embodiment of the present invention, the input in step S2 is 64x64x16 raw high snr data or denoised low snr data, the denoised data is three-dimensional fluorescence data output by the denoising network in step S1, and the raw data is obtained from a corresponding input data region in a sequence stack of high snr fluorescence images captured by a camera.
That is, the input in step S2 is 64x64x16 raw high snr data or denoised low snr data, the denoised data is three-dimensional fluorescence data output by the denoising network in step S1, and the raw data is obtained from a corresponding input data region in a sequence stack of high snr fluorescence images captured by a camera.
Step S3: and establishing training data and test sample data, namely establishing the required training and test sample data.
Further, in an embodiment of the present invention, in step S3, the training data includes 10000 fluorescence region data and has a size of 64 × 16, wherein different signal-to-noise ratio data in the same region are obtained by using different illumination intensities; and acquiring different signal-to-noise ratio data in the same area by using different exposure time, wherein the test sample data is low signal-to-noise ratio data in different areas.
It is understood that, in step S3, the fluorescence micro-training sample includes 10000 fluorescence regions data, which are 64x64x16, and the data acquisition methods include but are not limited to: acquiring data with different signal-to-noise ratios in the same area by using different illumination light intensities; and acquiring data with different signal-to-noise ratios in the same area by using different exposure times. The test sample is low signal-to-noise ratio data under different areas.
Step S4: and training a denoising generation network and a denoising discrimination network by taking a minimum cost function as a target according to the training data and the test sample data.
That is, the generator network and discriminant network are trained with the goal of minimizing a cost function.
Further, in an embodiment of the present invention, the denoising discriminant network cost function in step S4 is as follows:
L1(z,z*)＝-z*·log(z)-(1-z*)·log(1-z)，
wherein z is a probability value obtained by outputting low signal-to-noise ratio data through a denoising discrimination network, and z is*Is whether the training sample belongs to the original high signal-to-noise ratio data.
Further, in an embodiment of the present invention, the denoising generating network cost function in step S4 is as follows:
L2(y,y*)＝sum(y-y*)2+log(1-z)，
y is denoised data obtained by denoising and generating network output for low signal-to-noise ratio data, and y*The corresponding high snr true data in the training samples.
Further, in one embodiment of the present invention, the initialization of network weights in step S4 employs a one-dimensional gaussian distribution, using the adammoptimizer method to minimize the loss function.
Further, in one embodiment of the present invention, the network weight value in step S4 is obtained by minimizing the loss function until it does not decrease any more.
In summary, in step S3, training samples are selected from the data set, and in step S4, the denoising generation network and the denoising discrimination network formed by the networks respectively trained by all the training samples are input to the denoising generation network model in the next step, so as to output the denoised data.
Step S5: and inputting the fluorescence microscopic data into a denoising generation network to obtain a denoising result.
Namely, the fluorescence microscopic data is input into a denoising generation network, a denoising result is output,
specifically, the method comprises the steps of building a denoising generation network, building a denoising discrimination network, inputting fluorescence microscopic data into the denoising generation network, outputting a denoising result, wherein the generation network comprises a plurality of convolution layers, a maximum pooling layer, an upper sampling layer, a jump connection layer, and a Normalization operation and an excitation operation which are connected behind each convolution layer, the Normalization operation adopts Batch Normalization, the excitation operation adopts a Re L U function, the discrimination network comprises a plurality of convolution layers, a maximum pooling layer and a full connection layer, and the Normalization operation, the excitation operation and a two-classification output which are connected behind each convolution layer, the Normalization operation adopts Batch Normalization, the excitation operation adopts a Re L U function, required training and test sample data are built, the minimum cost function is used as a target to train the generation network and the discrimination network, the fluorescence microscopic data are input into the denoising generation network, and the denoising result is output, and accordingly the denoising capability of the three-dimensional fluorescence microscopic signal with low signal-to-noise ratio can be effectively improved.
According to the three-dimensional fluorescence microscopic signal denoising method based on the generation countermeasure network, the three-dimensional fluorescence data with low signal-to-noise ratio can be denoised to obtain data with high signal-to-noise ratio by training the denoising generation network and the denoising judgment network, the effect of the three-dimensional fluorescence signal is obviously improved, the denoising capability of the three-dimensional fluorescence microscopic signal with low signal-to-noise ratio can be effectively improved, and the method is more reliable and accurate.
Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "a plurality" means at least two, e.g., two, three, etc., unless specifically limited otherwise.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention, and that variations, modifications, substitutions and alterations can be made to the above embodiments by those of ordinary skill in the art within the scope of the present invention.
Claims (10)
1. A three-dimensional fluorescence microscopic signal denoising method based on a generation countermeasure network is characterized by comprising the following steps:
step S1: building a denoising generation network, wherein the denoising generation network comprises a plurality of first convolution layers, a first maximum pooling layer, an up-sampling layer, a jump connection layer, and a first normalization operation and a first excitation operation which are connected behind each convolution layer;
step S2: building a denoising discrimination network, wherein the denoising discrimination network comprises a plurality of convolution layers, a maximum pooling layer, a full-connection layer, and a second normalization operation, a second excitation operation and two classified outputs which are connected behind each convolution layer;
step S3: establishing training data and test sample data;
step S4: training the denoising generation network and the denoising discriminant network according to the training data and the test sample data by taking a minimum cost function as a target;
step S5: and inputting the fluorescence microscopic data into the denoising generation network to obtain a denoising result.
2. The method according to claim 1, wherein the step S1 includes:
step S11: building a first convolution part, wherein the first convolution part is 2 layers of 3D convolution layers, and a maximum pooling layer is correspondingly connected behind each first convolution layer;
step S12: building an upper convolution part, wherein the upper convolution part is 2 layers of 3D convolution layers, an upper sampling layer is correspondingly connected behind each convolution layer, and a jump connection layer is simultaneously connected behind each convolution layer to combine data before corresponding pooling;
step S13: and constructing an output part of the whole generated network model, wherein the output part comprises 1 layer of convolution layer, so as to obtain an output result by adding the convolution layer with input data, the output result is data with the same size as the input data, and the input data is a denoised result.
3. The method according to claim 1, wherein the step S2 includes:
step S21: building a second convolution part, wherein the second convolution part is 4 layers of 3D convolution layers, and a maximum pooling layer is correspondingly connected behind each second convolution layer;
step S22: building a full connection part;
step S23: and constructing an output part of the whole convolutional neural network model, wherein the output is one number, and the current input data is original high signal-to-noise ratio data.
4. The method as claimed in claim 2, wherein the input in step S1 is three-dimensional low signal-to-noise ratio fluorescence data of 64x64x16, the data three-dimensional low signal-to-noise ratio fluorescence is acquired randomly or through a sliding window from a stack of low signal-to-noise ratio fluorescence image sequences captured by a camera, and the target output is three-dimensional high signal-to-noise ratio fluorescence data of 64x64x16, the three-dimensional high signal-to-noise ratio fluorescence data is acquired from a corresponding input data region in a stack of high signal-to-noise ratio fluorescence image sequences captured by the camera.
5. The method according to claim 3, wherein the input in step S2 is 64x64x16 raw high signal-to-noise ratio data or denoised low signal-to-noise ratio data, the denoised data is three-dimensional fluorescence data output by the denoising network in step S1, and the raw data is obtained from a corresponding input data region in a stack of high signal-to-noise ratio fluorescence image sequences captured by the camera.
6. The method according to claim 1, wherein in step S3, the training data comprises 10000 fluorescence region data and the size is 64x64x16, wherein different signal-to-noise ratio data in the same region are obtained by using different illumination intensities; and acquiring different signal-to-noise ratio data in the same area by using different exposure time, wherein the test sample data is low signal-to-noise ratio data in different areas.
7. The method according to claim 1, wherein the denoising discriminant network cost function in step S4 is as follows:
L1(z，z*)＝-z*·log(z)-(1-z*)·log(1-z)，
wherein z is a probability value obtained by the low signal-to-noise ratio data through the output of the denoising discrimination network, and z is*Is whether the training sample belongs to the original high signal-to-noise ratio data.
8. The method according to claim 1, wherein the denoising generating network cost function in step S4 is as follows:
L2(y,y*)＝sum(y-y*)2+log(1-z)，
wherein, the low signal-to-noise ratio data is denoised data y obtained by the denoising generation network output*And the data is the corresponding high signal-to-noise ratio true value data in the training sample.
9. The method according to claim 7 or 8, wherein the network weight initialization in step S4 employs a one-dimensional gaussian distribution, and the adammoptimizer method is used to minimize the loss function.
10. The method according to claim 7, wherein the network weight value in step S4 is obtained by minimizing a loss function until no more decrease is obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010227291.3A CN111476125A (en) | 2020-03-27 | 2020-03-27 | Three-dimensional fluorescence microscopic signal denoising method based on generation countermeasure network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010227291.3A CN111476125A (en) | 2020-03-27 | 2020-03-27 | Three-dimensional fluorescence microscopic signal denoising method based on generation countermeasure network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111476125A true CN111476125A (en) | 2020-07-31 |
Family
ID=71748437
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010227291.3A Withdrawn CN111476125A (en) | 2020-03-27 | 2020-03-27 | Three-dimensional fluorescence microscopic signal denoising method based on generation countermeasure network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111476125A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP4174758A1 (en) * | 2021-10-26 | 2023-05-03 | Leica Microsystems CMS GmbH | Training a denoising model for a microscope |
| US12008737B2 (en) | 2020-08-07 | 2024-06-11 | Nanotronics Imaging, Inc. | Deep learning model for noise reduction in low SNR imaging conditions |
-
2020
- 2020-03-27 CN CN202010227291.3A patent/CN111476125A/en not_active Withdrawn
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US12008737B2 (en) | 2020-08-07 | 2024-06-11 | Nanotronics Imaging, Inc. | Deep learning model for noise reduction in low SNR imaging conditions |
| EP4174758A1 (en) * | 2021-10-26 | 2023-05-03 | Leica Microsystems CMS GmbH | Training a denoising model for a microscope |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP7249326B2 (en) | Method, Apparatus, and Computer Program for Improved Reconstruction of High Density Super-Resolved Images from Diffraction Limited Images Acquired by Single Molecule Localization Microscopy | |
| JP7422825B2 (en) | Focus-weighted machine learning classifier error prediction for microscope slide images | |
| George et al. | Deep Transfer Learning: A new deep learning glitch classification method for advanced LIGO | |
| CN111524064B (en) | Fluorescence microscopic image super-resolution reconstruction method based on deep learning | |
| JP2019110120A (en) | Method, device, and system for remote deep learning for microscopic image reconstruction and segmentation | |
| US20240094151A1 (en) | Adaptive specimen image acquisition using an artificial neural network | |
| JP2022516467A (en) | Two-dimensional fluorescence wave propagation system and method to the surface using deep learning | |
| JP2018180635A (en) | Image processing device, image processing method, and image processing program | |
| CN110070517B (en) | Blurred image synthesis method based on degradation imaging mechanism and generation countermeasure mechanism | |
| CN113256772B (en) | Double-angle light field high-resolution reconstruction system and method based on visual angle conversion | |
| CN111476125A (en) | Three-dimensional fluorescence microscopic signal denoising method based on generation countermeasure network | |
| Lebbink et al. | Template matching as a tool for annotation of tomograms of stained biological structures | |
| Boland et al. | Improving axial resolution in Structured Illumination Microscopy using deep learning | |
| Fazel et al. | Analysis of super-resolution single molecule localization microscopy data: A tutorial | |
| EP3994516A1 (en) | System and method for generating an image | |
| Li et al. | Fast confocal microscopy imaging based on deep learning | |
| Duchesne et al. | Evolutionary Map of the Universe (EMU): a pilot search for diffuse, non-thermal radio emission in galaxy clusters with the Australian SKA Pathfinder | |
| TW201910929A (en) | Generating high resolution images from low resolution images for semiconductor applications | |
| CN111524078A (en) | Dense network-based microscopic image deblurring method | |
| Bertin | Mining pixels: The extraction and classification of astronomical sources | |
| Boland et al. | Improving axial resolution in SIM using deep learning | |
| Zhang et al. | Entropy-Based re-sampling method on SAR class imbalance target detection | |
| Biggs et al. | Subpixel deconvolution of 3D optical microscope imagery | |
| US11238566B2 (en) | Image processing device, system, and method for improving signal-to-noise of microscopy images | |
| Li et al. | Automatic detection of circulating tumor cells with very deep residual networks |
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 | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20200731 |
|
| WW01 | Invention patent application withdrawn after publication |