CN115526200A - Attention mechanism-based low-coherence interference signal denoising method and system - Google Patents
Attention mechanism-based low-coherence interference signal denoising method and system Download PDFInfo
- Publication number
- CN115526200A CN115526200A CN202210516792.2A CN202210516792A CN115526200A CN 115526200 A CN115526200 A CN 115526200A CN 202210516792 A CN202210516792 A CN 202210516792A CN 115526200 A CN115526200 A CN 115526200A
- Authority
- CN
- China
- Prior art keywords
- low
- signal
- coherence
- layer
- noise
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/11—Complex mathematical operations for solving equations, e.g. nonlinear equations, general mathematical optimization problems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
- G06F3/01—Input arrangements or combined input and output arrangements for interaction between user and computer
- G06F3/048—Interaction techniques based on graphical user interfaces [GUI]
- G06F3/0481—Interaction techniques based on graphical user interfaces [GUI] based on specific properties of the displayed interaction object or a metaphor-based environment, e.g. interaction with desktop elements like windows or icons, or assisted by a cursor's changing behaviour or appearance
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Operations Research (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a low coherence interference signal denoising method and system based on an attention mechanism, which comprises the following steps: s1, preprocessing a low-coherence signal, and constructing a low-coherence signal training set containing noise; s2, constructing a depth one-dimensional convolution denoising neural network consisting of a residual error module, a context information extraction module and a self-attention module; s3, inputting training data into the trained network to eliminate noise in signals; and S4, outputting a clean low-coherence signal. The system comprises: the system comprises a low-coherence measuring instrument, a data acquisition card, a calculation server, an interactive control device and a display screen. The low-coherence signal denoising method and system based on context information and self-attention provided by the invention have the advantages of simple parameter adjustment, good signal noise elimination effect and the like.
Description
Technical Field
The invention belongs to the technical field of low-coherence signal processing, and particularly relates to a low-coherence signal denoising method and system based on context information and self-attention.
Background
The low coherence signal is the signal acquired by the low coherence interferometer. Low coherence interferometers are commonly used in the fields of optical coherence tomography, object surface topography, roughness measurement, mirror flatness, lens set spacing measurement, and the like. The presence of noise in the low coherence signal can greatly affect the low coherence interferometer and the measurement accuracy and imaging quality. Particularly, as a precision instrument, the low coherence interferometer generates random noises due to the influence of external factors such as ambient air vibration and temperature variation, and the noises have the characteristics of low frequency, high energy, strong randomness and the like.
Therefore, in order to obtain a high-quality image and a high-quality measurement accuracy, noise in the low-coherence signal must be denoised, and the original signal should be prevented from being damaged as much as possible during denoising. The traditional low coherent signal denoising method comprises methods such as wavelet transformation, fourier transformation, polynomial fitting and the like, however, the methods all need complex parameter adjusting processes, and parameter adjusting personnel have abundant engineering experience; in addition, the noise elimination effect of the method is not ideal due to the strong randomness of the noise.
Disclosure of Invention
In order to overcome the limitation of the traditional denoising method, the invention provides a low coherence interference signal denoising method and system based on an attention mechanism, which have the advantages of simple parameter adjustment, good signal noise elimination effect and the like, and can realize the denoising of the low coherence signal.
The technical scheme adopted by the invention for realizing the purpose is as follows: the low-coherence interference signal denoising method based on the attention mechanism comprises the following steps:
s1, carrying out normalization processing on low coherent signals; constructing a low coherence signal data set containing noise;
s2, constructing a depth one-dimensional convolution denoising neural network;
s3, training a depth one-dimensional convolution denoising neural network by using training set data, verifying whether the set data verifying network is converged, and testing the set data to test the noise signal eliminating effect;
and S4, processing the to-be-processed noise-containing signal by using the trained one-dimensional convolution denoising neural network, and outputting a low-coherence signal.
The pretreatment in the S1 specifically comprises the following steps:
s11, collecting and normalizing low-coherence signals: acquisition of a data set containing M sequences of low-coherence signals, X = { X, using a low-coherence interferometer 1 ,…X m ,…,X M In which the signal sequence X m ＝{x m1 ,…,x mN For each low coherence signal sequence X m Normalization is performed according to equation (1):
in the formula
For each low coherence signal X m Sample data x in mi The maximum value of (a) is,
for each low coherence signal X m Sample data x in mi I is less than or equal to N, and N is the number of sampling points of the low coherent signal.
Adding noise in S1 includes:
s12, constructing a noise-containing low-coherence signal sample: constructing white noise n with Gaussian distribution; different proportions of noise are added to the M low coherence signals in data set X according to equation (2):
y mi ＝z mi +αn (2)
obtaining a low coherent signal sequence added with white noise: y is m ＝{y m1 ,…y mi …,y mN }；
To obtain MData set of low coherent signal sequence with white noise added: y = { Y 1 ,…Y i …,Y N }；
Where α is a parameter controlling the proportion of noise added, y mi Is a low coherence signal with white noise added.
The partitioning of the data set in S1 includes:
s13, dividing a neural network training set, a verification set and a test set: uniformly mixing the data set Y of the low coherent signal sequence added with the Gaussian white noise with the data set X of the low coherent signal sequence without the noise, and mixing according to the ratio of 5:3: and 2, dividing the training set, the verification set and the test set.
The S2 specifically comprises the following steps:
s21, establishing a network model, which comprises the following steps: the system comprises a feature extraction module, a context feature enhancement module, a self-attention module and a residual error module; the characteristic extraction module is combined with the context characteristic enhancement module to process the received original low-coherence signal and extract the nonlinear signal characteristic;
s25, the self-attention module is used for further enhancing the noise extraction capability of the network for the nonlinear characteristics and outputting extracted noise signals;
and S26, the residual error module is used for subtracting the extracted noise signal from the original low coherent signal to obtain a clean low coherent signal.
The feature extraction module combines with the context feature enhancement module to process the original low-coherence signal received by the feature extraction module, and includes:
s22, the feature extraction module comprises a 1 st layer, a 2 nd layer, a 3 rd layer, a 4 th layer, a 5 th layer, a 6 th layer, a 7 th layer, an 8 th layer, a 9 th layer, a 10 th layer and an 11 th layer of convolutional neural network; wherein the operations of layer 1, layer 3, layer 5, layer 6, layer 8, layer 10 include convolution, batch normalization, and activation; the convolution adopts one-dimensional convolution, the convolution kernel size is 1 multiplied by 3, the activation adopts a linear rectification function, and the step length is 2; operations of the 2 nd layer, the 7 th layer and the 9 th layer comprise expanding convolution, batch normalization and activation; the expansion convolution expansion factor is 2, the convolution kernel size is 1 multiplied by 3, the function adopted by activation is a linear rectification function, and the step length is 2; layer 11 has only convolution operation, the convolution adopts one-dimensional convolution, and the convolution kernel size is 1 × 3.
S23, the characteristics output by the 5 th network are sent to a context characteristic enhancement module so as to enhance the capability of the network for obtaining the context information of the whole network; after the characteristic enhancement, the context characteristic enhancement module inputs the enhanced characteristic into a 6 th layer network, and the enhanced characteristic is processed by a 7 th layer, an 8 th layer, a 9 th layer, a 10 th layer and an 11 th layer network.
And S24, fusing the characteristics output by the 11 th layer network with the original low-coherence signal containing noise through cascade operation, further enhancing the characteristic extraction capability of the network, converting the fused characteristics into nonlinear characteristics by using a hyperbolic tangent (Tanh) activation function, and simultaneously carrying out normalization.
S3 specifically comprises the following steps:
s31, training a deep one-dimensional convolution denoising neural network by using a training set, taking a mean square error as a loss function, and training parameters of the network by using a back propagation algorithm;
s32, in the training process, using a verification set to check whether the model is converged;
and S33, after the model is converged, evaluating the capability of the network for eliminating the noise signal by using the test set, and finishing the training.
S4 specifically comprises the following steps:
and S41, inputting the low-coherence signal with noise into the trained depth one-dimensional convolution denoising neural network in the S3 to obtain a clean low-coherence signal.
A system of the low coherent signal denoising method based on self-supervised learning according to claim 1, the system comprising;
a signal acquisition device that acquires a low coherence signal;
the data acquisition card is used for receiving the low coherence signals acquired by the low coherence measuring instrument;
the computing server is loaded with a program module, and when the program is executed, the computing server is used for processing the acquired low coherence signals according to the steps of the method, so that the denoising processing of the low coherence signals is realized, and the processing result is output to the display screen;
the interactive control device is used for inputting instructions to the computing server;
and the display screen is used for visually displaying the low coherent signals before and after the denoising processing and a graphical interface in the processing process.
The signal acquisition equipment is a low coherence measuring instrument.
The invention has the following beneficial effects and advantages:
1. the deep one-dimensional convolution denoising neural network designed by the invention comprises convolution layers connected in series, context feature enhancement, self-attention noise extraction enhancement, residual noise elimination and other modules, and has the advantage of strong low-coherence signal noise removal effect.
2. The depth one-dimensional convolution denoising neural network designed by the invention has a simple parameter adjusting process, can denoise various low-coherence signals after training is finished, and does not need to adjust parameters again for different low-coherence signals.
3. The present invention provides a new idea for noise processing of low coherence signals using deep learning techniques.
Drawings
FIG. 1 is a flow chart of a denoising method of the present invention;
FIG. 2 is a diagram of a depth one-dimensional convolution denoising neural network structure according to the present invention;
FIG. 3 is a block diagram of the system of the present invention.
Detailed Description
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in detail below. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. This invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein, but rather should be construed as modified in the spirit and scope of the present invention as set forth in the appended claims.
The specific embodiment is as follows:
referring to fig. 1, the present invention provides a low coherence interference signal denoising method based on attention mechanism, including:
s1: the low coherence signal is pre-processed to construct a low coherence signal data set containing noise.
S11, collecting a data set X = { X = (zero) containing low-coherence signal sequences with the number of M =10000 by using a Sorpedo low-coherence interferometer TEL211C1 1 ,…X m ,…,X 10000 In which the low-coherence signal sequence X m ＝{x m1 ,…x mi …,x mN For each low coherence signal sequence X m Normalization is performed according to equation (1):
in the formula
For each low coherence signal X m Sample data x in mi The maximum value of (a) is,
for each low coherence signal X m Sample data x in mi I is not less than i and not more than N, N is the number of sampling points of the low coherent signal, and the number of sampling points N is 100000.
S12, constructing a noise-containing low-coherence signal sample: constructing white noise n with Gaussian distribution; 10000 low coherence signals in dataset X are added with different proportions of noise according to equation (2):
y mi ＝z mi +αn (2)
obtaining a low coherent signal sequence added with white noise: y is m ＝{y m1 ,…y mi …,y mN }；
Obtaining M data sets of low coherent signal sequences added with white noise: y = { Y 1 ,…Y i …,Y N }；
Wherein white noise obeys Gaussian distribution N (0, 1), the value range of alpha is 0.00-0.2, the value interval is 0.01, y is mi To add white noiseA coherent signal. The final data sets X and Y contain 200000 low coherence signals with different signal to noise ratios.
S13, dividing a neural network training set, a verification set and a test set: uniformly mixing a data set Y of the low coherent signal sequence added with the Gaussian white noise with a data set X of the low coherent signal sequence without added noise, and dividing the mixed signal into a training set, a verification set and a test set according to the proportion of 50%, 30% and 20%.
S2, referring to FIG. 2, an 11-layer depth one-dimensional convolution denoising neural network composed of a residual error module, a context information extraction module and a self-attention module is constructed.
S22, the feature extraction module comprises a 1 st layer, a 2 nd layer, a 3 rd layer, a 4 th layer, a 5 th layer, a 6 th layer, a 7 th layer, an 8 th layer, a 9 th layer, a 10 th layer and an 11 th layer of convolutional neural network; the operations of the 1 st layer, the 3 rd layer, the 5 th layer, the 6 th layer, the 8 th layer and the 10 th layer comprise convolution, batch normalization (bath norm) and activation, wherein the convolution adopts one-dimensional convolution, the convolution kernel size is 1 x 3, the activation adopts a linear rectification function (ReLU), and the step length is 2; the operations of the 2 nd layer, the 7 th layer and the 9 th layer comprise expanding convolution, batch normalization and activation, wherein the expanding convolution factor is 2, the convolution kernel size is 1 multiplied by 3, the function RelU adopted by activation is 2 in step length; layer 11 has only convolution operation, the convolution adopts one-dimensional convolution, and the convolution kernel size is 1 × 3.
S23, the characteristics output by the 5 th network are sent to a context characteristic enhancement module so as to enhance the capability of the network for obtaining the context information of the whole network; after the feature enhancement, the context feature enhancement module inputs the enhanced features into a 6 th layer network, and the enhanced features are processed by a 7 th layer, an 8 th layer, a 9 th layer, a 10 th layer and an 11 th layer network.
S24, fusing the characteristics output by the 11 th layer network with the original low-coherence signal containing noise through cascade operation (Cat), further enhancing the characteristic extraction capability of the network, converting the fused characteristics into nonlinear characteristics by using a hyperbolic tangent (Tanh) activation function, and simultaneously carrying out normalization;
s25, the normalized nonlinear characteristics are sent to a self-attention module, and the noise extraction capability of the network is further enhanced; the noise signal extracted by the network is output from the attention module.
And S26, finally, subtracting the extracted noise signal from the original low-coherence signal by using a residual error module to obtain a clean low-coherence signal.
And S3, training the depth one-dimensional convolution denoising neural network by using a training set, verifying whether the network is converged or not by using a verification set, and testing the noise signal elimination effect by using a test set.
S31, training the deep one-dimensional convolution denoising neural network by using a training set, and verifying the network noise elimination capability in the training process by using a verification set.
And S32, adopting the mean square error as a loss function, and training parameters of the network by using a back propagation algorithm. The specific definition of the loss function is:
and theta is a parameter of the depth one-dimensional convolution denoising neural network. And in the training process, the verification set is used for checking whether the model is converged, after the model is converged, the test set is used for evaluating the noise signal eliminating capability of the network, and the training is finished.
And S4, outputting a clean low-coherence signal by using the trained one-dimensional convolution denoising neural network.
And S41, inputting the low-coherence signal with noise into the trained depth one-dimensional convolution denoising neural network in the S3 to obtain a clean low-coherence signal.
The low coherence interference signal denoising system based on the attention mechanism comprises hardware and software; as shown in fig. 3, the hardware includes a low coherence measuring instrument ATR206C1, a data acquisition card ATS9350, a high performance computing server wave NF5468M5, an interactive control device (mouse, keyboard, etc.), a high definition display screen T3252U; inputting a low coherent signal acquired by a low coherence tester ATR206C1 into a high performance computing server wave NF5468M5 through a data acquisition card ATS9350, wherein an Ubuntu16.04 operating system is deployed on the high performance computing server wave NF5468M 5; the high-performance computing server wave NF5468M5 controls the low-coherence measuring instrument ATR206C1 and the data acquisition card ATS9350 through an interactive control device (a keyboard, a mouse and the like); the high-definition display screen T3252U is used for displaying a graphical interface;
the software comprises a graphical interface, a deep learning framework TensorFlow 1.12.0, an operating system Ubuntu16.04 and a denoising module; a deep learning framework TensorFlow 1.12.0 is deployed on an operating system Ubuntu16.04, and a denoising module is realized on the framework; the user operates the graphical interface through the interactive control device, and controls the high-performance server wave NF5468M5 through the operating system Ubuntu 16.04. The denoising module is a denoising method and program step and is used for realizing signal denoising.
The above-mentioned embodiments are provided to further explain the objects, technical solutions and advantages of the present invention in detail, and it should be understood that the above-mentioned embodiments are only examples of the present invention and should not be construed as limiting the present invention, and any modifications, equivalents and the like made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. The method for denoising the low coherence interference signal based on the attention mechanism is characterized by comprising the following steps of:
s1, carrying out normalization processing on a low-coherence signal; constructing a low coherence signal data set containing noise;
s2, constructing a depth one-dimensional convolution denoising neural network;
s3, training a depth one-dimensional convolution denoising neural network by using training set data, verifying whether the set data verification network is converged, and testing the set data to test the noise signal elimination effect;
and S4, processing the to-be-processed noise-containing signal by using the trained one-dimensional convolution denoising neural network, and outputting a low-coherence signal.
2. The attention-based low-coherence interference signal denoising method according to claim 1, wherein the preprocessing in S1 specifically comprises:
s11, collecting and normalizing low-coherence signals: acquisition of a data set containing M sequences of low-coherence signals, X = { X, using a low-coherence interferometer 1 ，...X m ，...，X M In which the signal sequence X m ＝{x m1 ，...，x mN For each low coherence signal sequence X m Normalization is performed according to equation (1):
in the formula
For each low coherence signal X m Sample data x in mi The maximum value of (a) is,
for each low coherence signal X m Sample data x in mi I is less than or equal to N, and N is the number of sampling points of the low-coherence signal.
3. The attention-based low coherence interference signal denoising method of claim 1, wherein the adding noise in S1 comprises:
s12, constructing a noise-containing low-coherence signal sample: constructing white noise n with Gaussian distribution; different proportions of noise are added to the M low coherence signals in data set X according to equation (2):
y mi ＝z mi +αn (2)
obtaining a low coherent signal sequence added with white noise: y is m ＝{y m1 ，…y mi …，y mN }；
Obtaining M data sets of low coherent signal sequences added with white noise: y = { Y 1 ，…Y i …，Y N }；
Where α is a parameter controlling the proportion of noise added, y mi For adding white noise to low coherence signals。
4. The attention-based low coherence interference signal denoising method of claim 1, wherein dividing the data set in S1 comprises:
s13, dividing a neural network training set, a verification set and a test set: and uniformly mixing the data set Y of the low coherent signal sequence added with the Gaussian white noise and the data set X of the low coherent signal sequence without added noise, and dividing the data set into a training set, a verification set and a test set according to the ratio of 5:3: 2.
5. The attention-based low coherence interference signal denoising method according to claim 1, wherein S2 specifically comprises:
s21, establishing a network model, which comprises the following steps: the system comprises a feature extraction module, a context feature enhancement module, a self-attention module and a residual error module; the feature extraction module is combined with the context feature enhancement module to process the received original low coherent signal and extract the nonlinear signal feature;
s25, the self-attention module is used for further enhancing the noise extraction capability of the network for the nonlinear features and outputting extracted noise signals;
and S26, the residual error module is used for subtracting the extracted noise signal from the original low-coherence signal to obtain a clean low-coherence signal.
6. The attention-based low coherence interference signal denoising method of claim 5, wherein the feature extraction module in combination with the context feature enhancement module processes the original low coherence signal received by the feature extraction module, comprising:
s22, the feature extraction module comprises a 1 st layer, a 2 nd layer, a 3 rd layer, a 4 th layer, a 5 th layer, a 6 th layer, a 7 th layer, an 8 th layer, a 9 th layer, a 10 th layer and an 11 th layer of convolutional neural network; wherein the operations of the 1 st, 3 rd, 5 th, 6 th, 8 th and 10 th layers comprise convolution, batch normalization and activation; operations of the 2 nd layer, the 7 th layer and the 9 th layer comprise expanding convolution, batch normalization and activation; layer 11 has only convolution operations.
S23, the characteristics output by the 5 th layer network are sent to a context characteristic enhancement module so as to enhance the capability of the network for obtaining the global context information; after the feature enhancement, the context feature enhancement module inputs the enhanced features into a 6 th layer network, and the enhanced features are processed by a 7 th layer, an 8 th layer, a 9 th layer, a 10 th layer and an 11 th layer network.
And S24, fusing the characteristics output by the 11 th layer network with the original low coherent signal containing noise through cascade operation, further enhancing the characteristic extraction capability of the network, converting the fused characteristics into nonlinear characteristics by utilizing a hyperbolic tangent (Tanh) activation function, and simultaneously carrying out normalization.
7. The attention-based low-coherence interference signal denoising method of claim 1, wherein S3 specifically comprises:
s31, training a depth one-dimensional convolution denoising neural network by using a training set, taking a mean square error as a loss function, and training parameters of the network by using a back propagation algorithm;
s32, in the training process, using a verification set to check whether the model is converged;
and S33, after the model is converged, evaluating the capability of the network for eliminating the noise signal by using the test set, and finishing the training.
8. The attention-based low-coherence interference signal denoising method of claim 1, wherein S4 specifically comprises:
and S41, inputting the low-coherence signal with noise into the trained depth one-dimensional convolution denoising neural network in the S3 to obtain a clean low-coherence signal.
9. A system of the low coherent signal denoising method based on self-supervised learning according to claim 1, wherein the system comprises;
a signal acquisition device that acquires a low coherence signal;
the data acquisition card is used for receiving the low coherence signals acquired by the low coherence measuring instrument;
the computing server is loaded with a program module, and when the program is executed, the computing server is used for processing the acquired low coherent signal according to the method steps of any one of claims 1 to 8, realizing the denoising processing of the low coherent signal and outputting the processing result to a display screen;
the interactive control device is used for inputting instructions to the computing server;
and the display screen is used for visually displaying the low coherent signals before and after the denoising processing and a graphical interface in the processing process.
10. The system of the method for denoising low-coherence signals based on self-supervised learning as recited in claim 9, wherein the signal acquisition device is a low-coherence measuring instrument.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210516792.2A CN115526200A (en) | 2022-05-12 | 2022-05-12 | Attention mechanism-based low-coherence interference signal denoising method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210516792.2A CN115526200A (en) | 2022-05-12 | 2022-05-12 | Attention mechanism-based low-coherence interference signal denoising method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115526200A true CN115526200A (en) | 2022-12-27 |
Family
ID=84695677
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210516792.2A Pending CN115526200A (en) | 2022-05-12 | 2022-05-12 | Attention mechanism-based low-coherence interference signal denoising method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115526200A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116563173A (en) * | 2023-07-11 | 2023-08-08 | 成都理工大学 | Ground penetrating radar image denoising method based on multi-noise self-supervision learning |
-
2022
- 2022-05-12 CN CN202210516792.2A patent/CN115526200A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116563173A (en) * | 2023-07-11 | 2023-08-08 | 成都理工大学 | Ground penetrating radar image denoising method based on multi-noise self-supervision learning |
| CN116563173B (en) * | 2023-07-11 | 2023-09-08 | 成都理工大学 | Ground penetrating radar image denoising method based on multi-noise self-supervision learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Sun et al. | Compressed sensing MRI using a recursive dilated network | |
| CN109858408B (en) | Ultrasonic signal processing method based on self-encoder | |
| CN109003232B (en) | Medical MRI image denoising method based on frequency domain scale smoothing Shearlet | |
| CN113376600B (en) | Pedestrian radar echo denoising method based on RSDNet | |
| Yin et al. | Image denoising using trivariate prior model in nonsubsampled dual-tree complex contourlet transform domain and non-local means filter in spatial domain | |
| Torres et al. | Total-variation-based methods for gravitational wave denoising | |
| Qiusheng et al. | Compressed sensing MRI based on the hybrid regularization by denoising and the epigraph projection | |
| CN115526200A (en) | Attention mechanism-based low-coherence interference signal denoising method and system | |
| Gemechu et al. | Random noise attenuation using an improved anisotropic total variation regularization | |
| CN111768349A (en) | ESPI image noise reduction method and system based on deep learning | |
| Liu et al. | Patch-based denoising method using low-rank technique and targeted database for optical coherence tomography image | |
| CN112381731B (en) | Single-frame stripe image phase analysis method and system based on image denoising | |
| CN111965599B (en) | Sound source identification method for two-dimensional dynamic grid compression beam forming | |
| Hou et al. | Super-resolution reconstruction algorithm for terahertz images | |
| ur Rehman et al. | Multi-scale image denoising based on goodness of fit (GOF) tests | |
| Shang et al. | Seismic random noise suppression using an adaptive nonlocal means algorithm | |
| CN110118958B (en) | Broadband radar complex echo denoising method based on variational coding-decoding network | |
| CN106772573A (en) | Seismic wavelet method for extracting signal based on maximal correlation entropy | |
| Huang et al. | Application of a novel constrained wavelet threshold denoising method in ensemble-based background-error variance | |
| CN115830153A (en) | Method and device for unfolding network by magnetic resonance imaging algorithm and electronic equipment | |
| CN113034475B (en) | Finger OCT (optical coherence tomography) volume data denoising method based on lightweight three-dimensional convolutional neural network | |
| Dobrosotskaya et al. | A Quantitative Comparison of Background Oriented and Conventional Schlieren Visualization | |
| Han et al. | Time-frequency enhanced characterization method based on asymmetric image reconstruction autoencoder | |
| Wang et al. | Suppression method for strong interference from stray light in 3D imaging system of structured light | |
| Kato et al. | Validation of denoising system using non-harmonic analysis and denoising convolutional neural network for removal of Gaussian noise from gravitational waves observed by LIGO |
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 |