CN114972802A - Hyperspectral image recovery method and system based on hybrid supervision model - Google Patents
Hyperspectral image recovery method and system based on hybrid supervision model Download PDFInfo
- Publication number
- CN114972802A CN114972802A CN202210401601.8A CN202210401601A CN114972802A CN 114972802 A CN114972802 A CN 114972802A CN 202210401601 A CN202210401601 A CN 202210401601A CN 114972802 A CN114972802 A CN 114972802A
- Authority
- CN
- China
- Prior art keywords
- hyperspectral
- network
- hyperspectral image
- degraded
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000011084 recovery Methods 0.000 title claims abstract description 47
- 238000000034 method Methods 0.000 title claims abstract description 31
- 230000006870 function Effects 0.000 claims abstract description 29
- 238000006731 degradation reaction Methods 0.000 claims abstract description 22
- 230000015556 catabolic process Effects 0.000 claims abstract description 20
- 239000000126 substance Substances 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 6
- 239000000654 additive Substances 0.000 claims description 4
- 230000000996 additive effect Effects 0.000 claims description 4
- 238000000342 Monte Carlo simulation Methods 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000005457 optimization Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 230000003595 spectral effect Effects 0.000 description 2
- 150000001875 compounds Chemical class 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 229910052500 inorganic mineral Inorganic materials 0.000 description 1
- 239000011707 mineral Substances 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/58—Extraction of image or video features relating to hyperspectral data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A40/00—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production
- Y02A40/10—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production in agriculture
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a hyperspectral image recovery method and system based on a hybrid supervision model, and belongs to the technical field of computational camera shooting. Firstly, a supervised hyperspectral image data set is constructed according to a hyperspectral image degradation model and parameters. And then constructing a hyperspectral recovery network, and training the network by using a supervised hyperspectral image dataset. An unsupervised loss function based on the noise estimate is then established. And finally, using the network parameters obtained by supervised learning as initial values of the network, inputting the degraded hyperspectral image into the network, and optimizing by using an unsupervised loss function. The method recovers the degraded hyperspectral image through the depth network to obtain a clear and complete hyperspectral image, and can effectively solve the problem that the network is inconsistent in the performance of training data and test data. According to the method, the noise in the hyperspectral degraded process is evaluated through Stent unbiased estimation, so that the network has robustness on noise input, and high-quality recovery of degraded hyperspectral images is realized.
Description
Technical Field
The invention relates to a hyperspectral image recovery method and system based on a hybrid supervision model, and belongs to the technical field of computational camera shooting.
Background
The hyperspectral image contains abundant spatial information and spectral information and is widely applied to the fields of vegetation ecology, atmospheric environment, geological mineral products, oceans, military affairs and the like. However, due to the influence of the external environment and the limitation of the equipment of the instrument, the hyperspectral image often suffers from various degradation problems, such as noise, undersampling or data missing. These degradation problems severely affect the subsequent utilization of the hyperspectral image.
The hyperspectral recovery method can recover a high-quality hyperspectral image from a low-quality hyperspectral image, and becomes a necessary preprocessing means for hyperspectral image application.
The hyperspectral image recovery method generally comprises two types, one type is a traditional image recovery method, and the other type is an image recovery method based on deep learning. The traditional hyperspectral image recovery method utilizes various priori knowledge to optimize and solve the high-quality hyperspectral image. For example, Yongyong Chen et al uses a low-rank matrix to solve the hyperspectral denoising problem; lizhi Wang et al restore a hyperspectral image from a compressed sensing image by using non-local sparse representation; xiaolin Han et al restored high resolution hyperspectral images from low resolution hyperspectral images using non-factorized sparse representation and dictionary learning. These traditional methods rely on manually extracted features, which, while effectively taking into account spectral characteristics, often require iterative solution of complex optimization problems, resulting in a lengthy recovery process.
The hyperspectral recovery method based on deep learning has a good effect in recent years. However, although the supervised learning based deep network can automatically learn deep features from a large amount of data to perform image restoration, due to the distribution gap between training data and test data, the hyperspectral restoration network learned under the data driving is not as good as the training data in the test data. In addition, the presence of random additive noise makes this performance gap large. On the other hand, although the feature representation can also be learned from a single hyperspectral image in the unsupervised deep network, the training result depends on the initial value of the network and has instability, and meanwhile, the method usually needs to consume a large amount of time for iterative optimization.
Disclosure of Invention
The invention aims to overcome the defects and shortcomings of the prior art and creatively provides a hyperspectral image recovery method and system based on a hybrid supervision model. According to the invention, through the deep network, external information from a large-scale hyperspectral data set and internal information from a target degraded hyperspectral can be effectively utilized, and a better recovery effect is obtained.
The invention is realized by adopting the following technical scheme.
A hyperspectral image recovery method based on a hybrid supervision model comprises the following steps:
step 1: and constructing a supervised hyperspectral image data set according to the hyperspectral image degradation model and the parameters.
Specifically, the hyperspectral image training data set is constructed according to hyperspectral degradation model parameters and noise parameters aiming at different hyperspectral degradation tasks.
The hyperspectral image training dataset comprises: and the pair data set is formed by the noise degraded hyperspectral image and the noise-free original hyperspectral image. The hyperspectral degradation model is uniformly expressed as:
y＝Φx+η (1)
wherein y is the degraded hyperspectral image, x is the original hyperspectral image, phi is the degradation matrix, and eta is additive Gaussian noise.
Step 2: and constructing a hyperspectral recovery network, and training the network by using a supervised hyperspectral image data set.
In particular, the present invention trains the network using pairwise hyperspectral datasets, learning the network to external prior knowledge. Training function
As follows:
wherein the subscript ex indicates that the data information is from non-recoveryHyperspectrum of the compound target.
Represents the ith pair of hyperspectral data,
representing the ith original hyperspectral image,
is shown to pass through
Adding noise to obtain an ith degraded hyperspectral image; n represents the total number of hyperspectral datasets;
representing a hyperspectral recovery network with a parameter w ex 。
And step 3: an unsupervised loss function based on the noise estimate is established.
Specifically, the unsupervised loss function uses a loss function based on a Stent Unbiased Risk Estimate (SURE) and a Total Variation (TV). And evaluating the noise suffered by the hyperspectral degraded process through the Stent unbiased risk estimation.
Wherein the Stent unbiased estimation
Expressed as:
wherein the content of the first and second substances,
an estimation function for a hyperspectral degraded image y is represented, and σ is represented as the noise intensity. N represents the total number of hyperspectral datasets.
Estimate, y, representing the position of the ith degraded hyperspectral pixel i Representing the true value of the i-th degraded hyperspectral pixel,
representing a differential operator.
Total variation loss function
Expressed as:
wherein the content of the first and second substances,
is a differential operator;
representing a hyperspectral recovery network with a parameter w in 。y in And representing a degraded hyperspectral image, and in representing that data information comes from a recovery target hyperspectral.
Joint unsupervised loss functions
Is shown as
Wherein λ is a weighting factor.
And 4, step 4: and (3) using the network parameters obtained by supervised learning as initial values of the network, inputting the degraded hyperspectral image into the network, and optimizing by using an unsupervised loss function.
Specifically, the partial derivatives of the estimation function are obtained by the monte carlo method, and are expressed as:
wherein b represents a zero mean random vector unit variance; e represents a constant, and in practical application, can be set to 1 × 10 -4 (ii) a Φ is the degradation matrix.
And aiming at each degraded hyperspectral image, taking a parameter value obtained by supervised learning as a network initial value, using an unsupervised loss function, and iteratively optimizing the network initial value to obtain a restored hyperspectral image.
In order to achieve the purpose of the invention, according to the method, the invention further provides a hyperspectral image recovery system based on a hybrid supervised model, which comprises a supervised training subsystem, an unsupervised training subsystem and a recovery subsystem.
The supervised training subsystem is used for carrying out preliminary training on the network, and using the paired hyperspectral data sets to enable the network to learn supervised information from external hyperspectral data.
And the unsupervised training subsystem is used for further optimizing the initial network value obtained by the supervised training subsystem so that the network learns the essential information of the target hyperspectral data.
And the recovery subsystem recovers the target hyperspectral image by using the network trained by the unsupervised training subsystem.
The connection relationship among the subsystems is as follows: the output end of the supervised training subsystem is connected with the input end of the unsupervised training subsystem, and the output end of the unsupervised training subsystem is connected with the input end of the recovery subsystem.
Advantageous effects
Compared with the prior art, the invention has the following advantages:
1. the method can effectively recover the degraded hyperspectral image to obtain a clear and complete hyperspectral image. The method can be applied to various hyperspectral degradation problems, and degraded hyperspectrum is recovered in a targeted manner by combining different hyperspectral recovery networks.
2. According to the invention, through the information of the external data set of supervised learning and the internal information of the hyperspectral target of unsupervised learning, the problem that the network is inconsistent in the training data and the test data can be effectively solved.
3. According to the invention, the noise in the hyperspectral degradation process is evaluated through the Stent unbiased estimation, so that the network has robustness to noise input, and high-quality recovery of degraded hyperspectral images is realized.
Drawings
FIG. 1 is a schematic flow diagram of the process of the present invention.
FIG. 2 is a schematic diagram of a model training mode of the method of the present invention;
FIG. 3 is a schematic diagram of the system of the present invention.
Detailed Description
For better illustrating the objects and advantages of the present invention, the following description will be made with reference to the accompanying drawings and examples.
Examples
As shown in fig. 1, a hyperspectral image restoration method based on a hybrid supervised model includes the following steps:
step 1: and constructing a supervised hyperspectral image data set according to the hyperspectral image degradation model and the parameters.
Step 2: constructing a hyperspectral recovery network, and training the network by using a supervised hyperspectral image dataset;
and step 3: establishing an unsupervised loss function based on noise estimation;
and 4, step 4: and (3) using the network parameters obtained by supervised learning as initial values of the network, inputting the degraded hyperspectral image into the network, and optimizing by using an unsupervised loss function.
According to the hyperspectral image recovery method based on the hybrid supervision model, a supervision hyperspectral image data set is constructed according to a hyperspectral image degradation model and parameters; constructing a hyperspectral recovery network, and training the network by using a supervised hyperspectral image dataset; establishing an unsupervised loss function based on noise estimation; and (3) using the network parameters obtained by supervised learning as initial values of the network, inputting the degraded hyperspectral image into the network, and optimizing by using an unsupervised loss function. Therefore, the problem that the training data and the test data have performance differences in the data-driven hyperspectral image recovery method can be solved, the hyperspectral image recovery method based on the hybrid supervision model is provided, and the noise influence is considered for the target hyperspectral image on the basis of supervised learning through the combined loss function of Stent unbiased estimation and variation, so that a better recovery effect is obtained.
Further, in an embodiment of the application, a hyperspectral image training data set is constructed according to hyperspectral degradation model parameters and noise parameters for different hyperspectral degradation tasks. The dataset comprises a large-scale paired dataset consisting of noisy degraded hyperspectral images and noiseless raw hyperspectral images. The hyperspectral degradation model can be uniformly expressed as:
y＝Φx+η
wherein y is a degraded hyperspectral image, x is an original hyperspectral image, phi is a degradation matrix, and eta is additive Gaussian noise.
Further, in one embodiment of the application, the network is trained using large-scale pairwise hyperspectral datasets to learn external prior knowledge. The training function is as follows:
wherein the content of the first and second substances,
representing the ith pair of hyperspectral data, N representing the total number of hyperspectral datasets,
representing a hyperspectral recovery network with a parameter w ex 。
Further, in one embodiment of the present application, the unsupervised loss function uses a loss function based on a Stent Unbiased Risk Estimate (SURE) and a Total Variation (TV). And evaluating the noise suffered by the hyperspectral degraded process through the Stent unbiased risk estimation.
Further, in one embodiment of the present application, the Stent unbiased estimate is expressed as:
wherein the content of the first and second substances,
an estimation function for a hyperspectral degraded image y is represented, and σ is represented as the noise intensity.
The total variation loss function is expressed as:
wherein the content of the first and second substances,
in order to be a differential operator, the system is,
representing a hyperspectral recovery network with a parameter w in 。
The joint unsupervised loss function is expressed as
Wherein λ is a weighting factor.
Further, in one embodiment of the present application, the partial derivative of the estimation function is expressed by a monte carlo method as:
where b represents the zero mean random vector unit variance and e represents a very small constant.
Further, in an embodiment of the application, for each degraded hyperspectral image, a parameter value obtained by supervised learning is used as a network initial value, an unsupervised loss function is used, the network initial value is iteratively optimized, and a restored hyperspectral image is obtained
Fig. 2 is a block diagram of a method training according to an example of the application, and a hyperspectral recovery network learns priori knowledge from the outside of a hyperspectral image of a target and essential information inside hyperspectral of the target through large-scale hyperspectral dataset training and single hyperspectral optimization.
Fig. 3 is a composition diagram of a hyperspectral image restoration system based on a hybrid supervised model according to an embodiment of the present application. Including the supervised training subsystem 10, the unsupervised training subsystem 20, the recovery subsystem 30:
the supervised training subsystem 10 is used for performing preliminary training on the network. And a large-scale pair-wise hyperspectral data set is used, so that the network learning is rich in supervised information from external hyperspectral data.
And the unsupervised training subsystem 20 is used for further optimizing the initial network value obtained by the supervised training subsystem so that the network learns the essential information of the target hyperspectral data.
And the recovery subsystem 30 recovers the target hyperspectral image by using the network trained by the unsupervised training subsystem.
The explanation of the hybrid supervision model-based hyperspectral image restoration method in the foregoing embodiment is also applicable to the hybrid supervision model-based hyperspectral image restoration system in this embodiment, and details are not repeated here.
Claims (2)
1. A hyperspectral image recovery method based on a hybrid supervision model is characterized by comprising the following steps:
step 1: constructing a supervised hyperspectral image data set according to the hyperspectral image degradation model and the parameters;
constructing a hyperspectral image training data set according to hyperspectral degraded model parameters and noise parameters aiming at different hyperspectral degraded tasks; the hyperspectral image training dataset comprises: a pair data set consisting of the noise-degraded hyperspectral image and the noise-free original hyperspectral image;
the hyperspectral degradation model is uniformly expressed as:
y＝Φx+η (1)
wherein y is a degraded hyperspectral image, x is an original hyperspectral image, phi is a degradation matrix, and eta is additive Gaussian noise;
step 2: constructing a hyperspectral recovery network, and training the network by using a supervised hyperspectral image dataset;
training the network by using a pair of hyperspectral data sets, so that the network learns external prior knowledge; training function
As follows:
the subscript ex represents that the data information comes from the hyperspectral of the non-recovery target;
represents the ith pair of hyperspectral data,
representing the ith original hyperspectral image,
is shown to pass through
Adding noise to obtain an ith degraded hyperspectral image; n represents the total number of hyperspectral datasets;
representing a hyperspectral recovery network with a parameter w ex ；
And step 3: establishing an unsupervised loss function based on noise estimation;
the unsupervised loss function uses a loss function based on the Stant unbiased risk estimation SURE and the total variation TV, and noise received in the hyperspectral degradation process is estimated through the Stant unbiased risk estimation;
wherein the Stent unbiased estimation
Expressed as:
wherein the content of the first and second substances,
representing an estimation function of the hyperspectral degraded image y, and expressing sigma as noise intensity; n represents the total number of hyperspectral datasets;
estimate, y, representing the position of the i-th degraded hyperspectral pixel i Representing the true value of the i-th degraded hyperspectral pixel,
representing a differential operator;
total variation loss function
Expressed as:
wherein the content of the first and second substances,
is a differential operator;
representing a hyperspectral recovery network with a parameter w in ；y in Representing a degraded hyperspectral image, and in representing that data information comes from a recovery target hyperspectral;
joint unsupervised loss functions
Expressed as:
wherein λ is a weighting factor;
and 4, step 4: using the network parameters obtained by supervised learning as initial values of the network, inputting the degraded hyperspectral image into the network, and optimizing by using an unsupervised loss function;
wherein the partial derivatives of the estimation function are obtained by a monte carlo method, and are represented as:
wherein b represents a zero mean random vector unit variance; e represents a constant; phi is a degradation matrix;
and aiming at each degraded hyperspectral image, taking a parameter value obtained by supervised learning as a network initial value, using an unsupervised loss function, and iteratively optimizing the network initial value to obtain a restored hyperspectral image.
2. A hyperspectral image recovery system based on a hybrid supervision model is characterized by comprising a supervision training subsystem, an unsupervised training subsystem and a recovery subsystem;
the system comprises a supervision training subsystem, a hyperspectral data acquisition subsystem and a hyperspectral data processing subsystem, wherein the supervision training subsystem is used for carrying out primary training on a network, and using a pair of hyperspectral data sets to enable the network to learn supervised information from external hyperspectral data;
the unsupervised training subsystem is used for further optimizing the initial network value obtained by the supervised training subsystem so that the network learns the essential information of the target hyperspectral data;
the recovery subsystem recovers the target hyperspectral image by utilizing the network trained by the unsupervised training subsystem;
the connection relationship among the subsystems is as follows: the output end of the supervised training subsystem is connected with the input end of the unsupervised training subsystem, and the output end of the unsupervised training subsystem is connected with the input end of the recovery subsystem.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210401601.8A CN114972802B (en) | 2022-04-18 | 2022-04-18 | Hyperspectral image recovery method and system based on hybrid supervision model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210401601.8A CN114972802B (en) | 2022-04-18 | 2022-04-18 | Hyperspectral image recovery method and system based on hybrid supervision model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114972802A true CN114972802A (en) | 2022-08-30 |
| CN114972802B CN114972802B (en) | 2024-05-31 |
Family
ID=82978478
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210401601.8A Active CN114972802B (en) | 2022-04-18 | 2022-04-18 | Hyperspectral image recovery method and system based on hybrid supervision model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114972802B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112132760A (en) * | 2020-09-14 | 2020-12-25 | 北京大学 | Image recovery method based on learnable differentiable matrix inversion and matrix decomposition |
| CN112700370A (en) * | 2020-12-29 | 2021-04-23 | 南京理工大学 | Hyperspectral image super-resolution method for hyper-parameter fidelity and depth prior joint learning |
| KR102288645B1 (en) * | 2020-08-26 | 2021-08-10 | 한국해양과학기술원 | Machine learning method and system for restoring contaminated regions of image through unsupervised learning based on generative adversarial network |
| CN114092834A (en) * | 2022-01-24 | 2022-02-25 | 南京理工大学 | Unsupervised hyperspectral image blind fusion method and system based on space-spectrum combined residual correction network |
-
2022
- 2022-04-18 CN CN202210401601.8A patent/CN114972802B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR102288645B1 (en) * | 2020-08-26 | 2021-08-10 | 한국해양과학기술원 | Machine learning method and system for restoring contaminated regions of image through unsupervised learning based on generative adversarial network |
| CN112132760A (en) * | 2020-09-14 | 2020-12-25 | 北京大学 | Image recovery method based on learnable differentiable matrix inversion and matrix decomposition |
| CN112700370A (en) * | 2020-12-29 | 2021-04-23 | 南京理工大学 | Hyperspectral image super-resolution method for hyper-parameter fidelity and depth prior joint learning |
| CN114092834A (en) * | 2022-01-24 | 2022-02-25 | 南京理工大学 | Unsupervised hyperspectral image blind fusion method and system based on space-spectrum combined residual correction network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114972802B (en) | 2024-05-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109859147B (en) | Real image denoising method based on generation of antagonistic network noise modeling | |
| Gai et al. | New image denoising algorithm via improved deep convolutional neural network with perceptive loss | |
| CN107610069B (en) | DVS visual video denoising method based on shared K-SVD dictionary | |
| Ni et al. | Example-driven manifold priors for image deconvolution | |
| CN111028166B (en) | Video deblurring method based on iterative neural network | |
| CN116579945B (en) | Night image restoration method based on diffusion model | |
| CN112967210A (en) | Unmanned aerial vehicle image denoising method based on full convolution twin network | |
| CN114511480A (en) | Underwater image enhancement method based on fractional order convolution neural network | |
| CN107292855B (en) | Image denoising method combining self-adaptive non-local sample and low rank | |
| CN104376533A (en) | Image denoising method based on regularization principal component tracking | |
| CN110728728B (en) | Compressed sensing network image reconstruction method based on non-local regularization | |
| CN113724146B (en) | Single-pixel imaging method based on plug-and-play prior | |
| Zhao et al. | Boundary-constrained interpretable image reconstruction network for deep compressive sensing | |
| CN110136164B (en) | Method for removing dynamic background based on online transmission transformation and low-rank sparse matrix decomposition | |
| CN117592543A (en) | Aeroengine rolling bearing fault diagnosis method based on self-supervision learning | |
| CN117408991A (en) | Image anomaly detection method and device based on reconstruction resistance and storage medium | |
| CN114972802A (en) | Hyperspectral image recovery method and system based on hybrid supervision model | |
| Li et al. | Automatic modulation recognition based on a new deep K-SVD denoising algorithm | |
| Rafi et al. | Data driven measurement matrix learning for sparse reconstruction | |
| Fu et al. | Double enhanced residual network for biological image denoising | |
| CN111681156B (en) | Deep compressed sensing image reconstruction method applied to wireless sensor network | |
| CN113592738A (en) | Underwater distorted image restoration method | |
| CN110415190B (en) | Method, device and processor for removing image compression noise based on deep learning | |
| Zhao et al. | Local-and-nonlocal spectral prior regularized tensor recovery for Cauchy noise removal | |
| Liu et al. | Research on video noise-reduction based on low rank representation |
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 |