CN111539886A - Defogging method based on multi-scale feature fusion - Google Patents
Defogging method based on multi-scale feature fusion Download PDFInfo
- Publication number
- CN111539886A CN111539886A CN202010318381.3A CN202010318381A CN111539886A CN 111539886 A CN111539886 A CN 111539886A CN 202010318381 A CN202010318381 A CN 202010318381A CN 111539886 A CN111539886 A CN 111539886A
- Authority
- CN
- China
- Prior art keywords
- features
- scale
- dff
- module
- decoder
- 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
- 230000004927 fusion Effects 0.000 title claims abstract description 40
- 238000000034 method Methods 0.000 title claims abstract description 30
- 238000005516 engineering process Methods 0.000 claims abstract description 10
- 238000005070 sampling Methods 0.000 claims description 15
- 239000000654 additive Substances 0.000 claims 1
- 230000000996 additive effect Effects 0.000 claims 1
- 239000000126 substance Substances 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000008713 feedback mechanism Effects 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- 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]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a defogging method based on multi-scale feature fusion, which comprises the following steps of: a multi-scale feature fusion defogging network is designed based on the framework and the reprojection technology of a U-shaped network, and comprises an encoder G with a multi-scale feature fusion module (DFF)EncAnd a decoder GDecAnd a feature restoration module G consisting of residual blocksRes. Encoder GEncAnd a decoder GDecAnd the image features fused by the DFF module at each scale are directly connected to the DFF modules at all subsequent scales as input so as to realize the fusion of the features at different scales, and the fused features are utilized to carry out image defogging operation.
Description
Technical Field
The invention belongs to the field of computer vision and image processing, and relates to a defogging method based on multi-scale feature fusion.
Background
Image defogging is a classical image processing problem whose task is to repair a given foggy image to estimate a clear fogless image. The problem of image defogging is widely regarded in the computer vision field because the defogging process of the image is firstly needed to obtain a clear scene in various high-level computer vision tasks (detection, identification and the like). In computer vision and computer graphics, atmospheric scattering models have been widely used as a description of the process of generating hazy images, the process of fogging being typically modeled as:
I(x)＝T(x)J(x)+(1-T(x))A (1)
where I (x) represents the observed blurred image, J (x) represents the sharp image, T (x) represents the light transmission function, and A represents the atmospheric light intensity. The goal of image defogging is to recover T (x), J (x), and A from I (x).
In order to solve the problems that image prior is easy to fail and the operation efficiency is low in the traditional algorithm, a method based on deep learning is widely applied to various computer vision tasks, wherein in the field of image defogging, a neural network designed based on the framework of a U-type network (U-Net) achieves good effects. However, there are several limitations to this architecture, such as encoder GEncSpatial information of image features is lost during down-sampling of the image features and there is a lack of sufficient correlation between features of different scales between non-adjacent network layers.
Although the performance of the overall network is improved by extracting and utilizing features from different scales in the framework of the U-type network, the relationships among the features of different scales are not effectively fused. Converged features have proven to be an effective means of improving network performance in many deep learning architectures, so most networks do converged features by using dense connections, feature cascading, and weighted element progressive summation. However, most feature fusion modules attempt to fuse features with the same scale among network layers before, and therefore, the feature fusion modules cannot be used for solving the problem of feature fusion with different scales among non-adjacent network layers in the framework of the U-type network. Although some methods use strided convolutional layers in an attempt to fuse features of different scales. Although the method can fuse a plurality of features of different scales, the method combines a plurality of features which are enlarged/reduced to the same scale in a simple cascade mode, so that useful information among the features of different scales cannot be effectively extracted. Therefore, in order to better improve the performance of the U-type network architecture, a better method for solving the problem of fusion of different scale features between non-adjacent network layers needs to be provided.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a defogging method based on multi-scale feature fusion, which can effectively improve the performance of a U-shaped network architecture and realize defogging of images.
In order to achieve the above purpose, the defogging method based on the multi-scale feature fusion comprises the following steps:
a multi-scale feature fusion defogging network is designed based on the framework and the reprojection technology of a U-shaped network, and comprises an encoder G with a multi-scale feature fusion module (DFF)EncAnd a decoder GDecAnd a feature restoration module G consisting of residual blocksRes. Encoder GEncAnd a decoder GDecAnd the image features fused by the DFF module at each scale are directly connected to the DFF modules at all subsequent scales as input so as to realize the fusion of the features at different scales, and the fused features are utilized to carry out image defogging operation.
Encoder GEncDFF module of middle nth layer
The re-projection technology is used for fusing features of different scales in the encoder, and the operation can be described as follows:
wherein the content of the first and second substances,inrepresentation encoder GEncThe features in the n-th layer are,
representing the fused features obtained by the DFF module, L representing the encoder GEncThe number of total layers (dimensions) of,
is indicated in encoder GEncAnd (5) obtaining the characteristics after the first n-1 in the population are fused by a DFF module.
Decoder GDecDFF module of middle nth layer
The re-projection technique is used to fuse features of different scales in the decoder, and its operation can be described as:
wherein j isnRepresentation decoder GDecThe features in the n-th layer are,
representing the fused features obtained by the DFF module, L representing the decoder GDecThe number of total layers (dimensions) of,
is shown in the decoder GDecAnd the middle and front L-n characteristics are obtained after DFF module fusion.
And the DFF module sequentially performs multiple times of fusion on the features with different scales by utilizing a reprojection technology. The DFF module firstly up-samples or down-samples the feature to be fused under the scale to the scale of the previous feature, down-samples or up-samples the difference value of the two features to the original scale, and adds and fuses the feature to be fused. Encoder GEncDFF module of middle nth layer
And decoder GDecMiddle L-n layer DFF module
Has a similar network structure, but the down-sampling operation is positioned opposite to the up-sampling operation.
The invention has the following beneficial effects:
when the defogging network with the multi-scale feature fusion designed based on the framework of the U-shaped network and the reprojection technology is specifically operated, the DFF module can effectively correct the encoder GEncUnder sampling image features and decoder GDecInformation lost in the process of up-sampling image features is fused with different scale features between non-adjacent network layers to improve the performance of the multi-scale network. Compared with other sampling and cascading fusion methods, the DFF module can better extract high-frequency information in the image features from high-resolution features of previous network layers due to a feedback mechanism, and spatial information lost by the image features can be corrected by gradually fusing the differences into potential features of down-sampling. In addition, the DFF module can utilize all the previously obtained image high-frequency characteristics and utilize an error correction feedback mechanism to perfectly fuse the image characteristics extracted by the current network layer so as to obtain a better image defogging effect.
Drawings
FIG. 1 is a block diagram of the present invention;
FIG. 2a shows an encoder GEncSchematic diagram of the DFF module of (a);
FIG. 2b shows a decoder GDecSchematic diagram of the DFF module of (a);
FIG. 3a is an image before defogging;
FIG. 3b is an image after defogging by the U-NET method;
FIG. 3c is an image after defogging according to the present invention.
Detailed Description
The invention is described in further detail below with reference to the accompanying drawings:
referring to fig. 1, the defogging method based on multi-scale feature fusion of the invention comprises the following steps:
multiscale networks designed based on a U-network (U-Net) architecture have some limitations of the network itself. For example, encoder GEncSpatial information lost during down-sampling of image features and lack of sufficient correlation between different scale features between non-adjacent network layers. The re-projection technique in the task of image super-resolution is an effective solution to this problem, aiming at minimizing the estimated high-resolution results
And errors between multiple observed low resolution inputs to reconstruct the generated high resolution content, an iterative reprojection technique was developed for the case of a single low resolution input, the algorithm of which can be described as:
wherein the content of the first and second substances,
representing the high resolution output, L, estimated in the t-th iterationobDenotes a low resolution image acquired by using the down-sampling operation f, and h denotes a re-projection operation.
Based on the re-projection technology in the image super-resolution task, the invention provides a DFF module which is used for effectively correcting a coder G of a U-shaped network architectureEncThe spatial information lost in the process of down-sampling the image features can be better fused with the different scale features between non-adjacent network layers, and the performance of a multi-scale network is improved. The DFF module is intended to further fuse the features of the current network layer through an error feedback mechanism and at the encoder GEncAnd a decoder GDecIs used in the preparation of the medicament. Encoder GEncAnd a decoder GDecThe image features fused by the DFF module at each scale are directly connected to the DFF modules at all subsequent scales as input so as to realize the fusion of the features at different scales. The multi-scale feature fused defogging network comprises a multi-scale feature fusion moldEncoder G for block (DFF)EncAnd a decoder GDecAnd a feature restoration module G consisting of residual blocksRes。
FIG. 2a and FIG. 2b are encoders G for a U-type network, respectivelyEncAnd a decoder GDecTo implement the structure of the DFF module. Encoder GEncDFF module of middle nth layer
The re-projection technology is used for fusing features of different scales in the encoder, and the operation can be described as follows:
wherein inRepresentation encoder GEncThe features in the n-th layer are,
representing the fused features obtained by the DFF module, L representing the encoder GEncThe number of total layers (dimensions) of,
is indicated in encoder GEncAnd (5) obtaining the characteristics after the first n-1 in the population are fused by a DFF module.
Decoder GDecDFF module of middle nth layer
The re-projection technique is used to fuse features of different scales in the decoder, and its operation can be described as:
wherein j isnRepresentation decoder GDecThe features in the n-th layer are,
representing fused features obtained by DFF moduleL denotes a decoder GDecThe number of total layers (dimensions) of,
is shown in the decoder GDecAnd the middle and front L-n characteristics are obtained after DFF module fusion.
The DFF module firstly up-samples or down-samples the feature to be fused under the scale to the scale of the previous feature, down-samples or up-samples the difference value of the two features to the original scale, and adds and fuses the feature to be fused. Wherein in the decoder GDecBy successively fusing one image feature which is fused by a DFF module
t ∈ {0,1, …, L-n-1} to progressively fuse the current features jnThe fusion of the multi-scale features in the decoder G can be defined in the following wayDecThe updating process in (1):
(1) calculating the feature to be fused at the scale of the t-th iteration
With features obtained after t-th DFF module fusion
Difference therebetween
Its operation can be described as:
wherein the content of the first and second substances,
representing features to be fused at that scale
Down-sampling to the tth pass beforeFeatures obtained after fusion of DFF modules
Reprojection operations with the same dimensions.
(2) Updating the feature needing to be fused at the scale through a reprojection operation
Its operation can be described as:
wherein the content of the first and second substances,
representing the difference of the t-th iteration
Upsampling to features to be fused at that scale
Reprojection operations with the same dimensions.
(3) Finally obtaining the fusion characteristics of the current network layer by fusing all the previous image characteristics fused by the DFF module under each scale
In the defogging network of the multi-scale feature fusion, the up-sampling and the down-sampling of the image features are respectively realized by adopting an deconvolution layer and a convolution layer with a convolution kernel of 3 × 3 and a step length of 2, a group of residual error blocks are used in an optimization unit and an encoder of each layer, each residual error block consists of three residual error sub-modules, each sub-module comprises two convolution layers with convolution kernels of 3 × 3 and a path directly connected with input and output, and a feature recovery module GResDecoder G consisting of 18 residual blocks, simultaneously in a multi-scale feature-fused defogging networkDecAnd encoder GEncEach of (1)DFF modules are respectively introduced into a network level, and a decoder GDecAnd encoder GEncWherein the DFF fusion module is implemented using deconvolution and convolution layers with convolution kernels of 4 × 4 and step size of 2.
In the training process of the defogging network with multi-scale feature fusion, a defogging training set with clear images is adopted, wherein 4000 indoor image pairs (a foggy image and a corresponding clear image) and 9000 outdoor image pairs are included, and a loss function used by the network is a minimum Mean Square Error (MSE).
As can be seen from fig. 3a, 3b and 3c, compared with the existing U-NET network architecture, the image defogging effect of the invention is better.
Claims (5)
1. A defogging method based on multi-scale feature fusion is characterized by comprising the following steps:
the defogging network with the fusion of the multi-scale features is designed based on the framework of the U-shaped network and the reprojection technology, and comprises an encoder G with a multi-scale feature fusion moduleEncAnd a decoder GDecAnd a feature restoration module G consisting of residual blocksResWherein the encoder GEncAnd a decoder GDecAnd the image features fused by the DFF module at each scale are connected to the DFF modules at all subsequent scales as input so as to realize the fusion of the features at different scales, and the fused features are utilized to carry out image defogging operation.
2. The multi-scale feature fusion based defogging method according to claim 1, wherein an encoder GEncDFF module of middle nth layer
And (3) fusing the features of different scales in the encoder by utilizing a reprojection technology, namely:
wherein inRepresentation encoder GEncThe features in the n-th layer are,
representing the fused features obtained by the DFF module, L representing the encoder GEncThe total number of layers of (a) and (b),
is indicated in encoder GEncAnd (5) obtaining the characteristics after the first n-1 in the population are fused by a DFF module.
3. The multi-scale feature fusion based defogging method according to claim 1, wherein a decoder GDecDFF module of middle nth layer
And (3) fusing features of different scales in the decoder by utilizing a re-projection technology, namely:
wherein j isnRepresentation decoder GDecThe features in the n-th layer are,
representing the fused features obtained by the DFF module, L representing the decoder GDecThe total number of layers of (a) and (b),
is shown in the decoder GDecAnd the middle and front L-n characteristics are obtained after DFF module fusion.
4. The defogging method based on multi-scale feature fusion of claim 1, wherein the DFF module utilizes a reprojection technique to successively perform multiple fusion on features of different scales, the DFF module first up-samples or down-samples the feature to be fused at the scale to the scale of the previous feature, and down-samples or up-samples the difference between the two to the original scale, and performs additive fusion with the feature to be fused.
5. The multi-scale feature fusion based defogging method according to claim 1, wherein an encoder GEncDFF module of middle nth layer
And decoder GDecMiddle L-n layer DFF module
Has a similar network structure, and the down-sampling operation is opposite to the up-sampling operation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010318381.3A CN111539886B (en) | 2020-04-21 | 2020-04-21 | Defogging method based on multi-scale feature fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010318381.3A CN111539886B (en) | 2020-04-21 | 2020-04-21 | Defogging method based on multi-scale feature fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111539886A true CN111539886A (en) | 2020-08-14 |
| CN111539886B CN111539886B (en) | 2023-01-03 |
Family
ID=71980000
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010318381.3A Active CN111539886B (en) | 2020-04-21 | 2020-04-21 | Defogging method based on multi-scale feature fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111539886B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112150395A (en) * | 2020-10-15 | 2020-12-29 | 山东工商学院 | Encoder-decoder network image defogging method combining residual block and dense block |
| CN112967272A (en) * | 2021-03-25 | 2021-06-15 | 郑州大学 | Welding defect detection method and device based on improved U-net and terminal equipment |
| CN113034413A (en) * | 2021-03-22 | 2021-06-25 | 西安邮电大学 | Low-illumination image enhancement method based on multi-scale fusion residual error codec |
| CN113034445A (en) * | 2021-03-08 | 2021-06-25 | 桂林电子科技大学 | Multi-scale connection image defogging algorithm based on UNet3+ |
| CN113240589A (en) * | 2021-04-01 | 2021-08-10 | 重庆兆光科技股份有限公司 | Image defogging method and system based on multi-scale feature fusion |
| WO2023046136A1 (en) * | 2021-09-27 | 2023-03-30 | 北京字跳网络技术有限公司 | Feature fusion method, image defogging method and device |
| CN116416248A (en) * | 2023-06-08 | 2023-07-11 | 杭州华得森生物技术有限公司 | Intelligent analysis system and method based on fluorescence microscope |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110570371A (en) * | 2019-08-28 | 2019-12-13 | 天津大学 | image defogging method based on multi-scale residual error learning |
| CN110782399A (en) * | 2019-08-22 | 2020-02-11 | 天津大学 | Image deblurring method based on multitask CNN |
| US20200082219A1 (en) * | 2018-09-07 | 2020-03-12 | Toyota Research Institute, Inc. | Fusing predictions for end-to-end panoptic segmentation |
-
2020
- 2020-04-21 CN CN202010318381.3A patent/CN111539886B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200082219A1 (en) * | 2018-09-07 | 2020-03-12 | Toyota Research Institute, Inc. | Fusing predictions for end-to-end panoptic segmentation |
| CN110782399A (en) * | 2019-08-22 | 2020-02-11 | 天津大学 | Image deblurring method based on multitask CNN |
| CN110570371A (en) * | 2019-08-28 | 2019-12-13 | 天津大学 | image defogging method based on multi-scale residual error learning |
Non-Patent Citations (2)
| Title |
|---|
| TONG CUI 等，: "Multi-scale Densely Connected Dehazing Network", 《INTERNATIONAL CONFERENCE ON INTELLIGENT ROBOTICS AND APPLICATIONS》 * |
| 陈永 等，: "基于多尺度卷积神经网络的单幅图像去雾方法", 《光学学报》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112150395A (en) * | 2020-10-15 | 2020-12-29 | 山东工商学院 | Encoder-decoder network image defogging method combining residual block and dense block |
| CN113034445A (en) * | 2021-03-08 | 2021-06-25 | 桂林电子科技大学 | Multi-scale connection image defogging algorithm based on UNet3+ |
| CN113034445B (en) * | 2021-03-08 | 2022-11-11 | 桂林电子科技大学 | Multi-scale connection image defogging algorithm based on UNet3+ |
| CN113034413A (en) * | 2021-03-22 | 2021-06-25 | 西安邮电大学 | Low-illumination image enhancement method based on multi-scale fusion residual error codec |
| CN113034413B (en) * | 2021-03-22 | 2024-03-05 | 西安邮电大学 | Low-illumination image enhancement method based on multi-scale fusion residual error coder-decoder |
| CN112967272A (en) * | 2021-03-25 | 2021-06-15 | 郑州大学 | Welding defect detection method and device based on improved U-net and terminal equipment |
| CN112967272B (en) * | 2021-03-25 | 2023-08-22 | 郑州大学 | Welding defect detection method and device based on improved U-net and terminal equipment |
| CN113240589A (en) * | 2021-04-01 | 2021-08-10 | 重庆兆光科技股份有限公司 | Image defogging method and system based on multi-scale feature fusion |
| WO2023046136A1 (en) * | 2021-09-27 | 2023-03-30 | 北京字跳网络技术有限公司 | Feature fusion method, image defogging method and device |
| CN116416248A (en) * | 2023-06-08 | 2023-07-11 | 杭州华得森生物技术有限公司 | Intelligent analysis system and method based on fluorescence microscope |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111539886B (en) | 2023-01-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111539886B (en) | Defogging method based on multi-scale feature fusion | |
| CN109345449B (en) | Image super-resolution and non-uniform blur removing method based on fusion network | |
| CN110120011B (en) | Video super-resolution method based on convolutional neural network and mixed resolution | |
| CN108596841B (en) | Method for realizing image super-resolution and deblurring in parallel | |
| CN112508083B (en) | Image rain and fog removing method based on unsupervised attention mechanism | |
| CN110349087B (en) | RGB-D image high-quality grid generation method based on adaptive convolution | |
| CN104657962B (en) | The Image Super-resolution Reconstruction method returned based on cascading linear | |
| CN109034198B (en) | Scene segmentation method and system based on feature map recovery | |
| CN109410144A (en) | A kind of end-to-end image defogging processing method based on deep learning | |
| CN116777764A (en) | Diffusion model-based cloud and mist removing method and system for optical remote sensing image | |
| CN112241939A (en) | Light-weight rain removing method based on multi-scale and non-local | |
| CN111539885B (en) | Image enhancement defogging method based on multi-scale network | |
| CN112200752A (en) | Multi-frame image deblurring system and method based on ER network | |
| CN116721033A (en) | Single image defogging method based on random mask convolution and attention mechanism | |
| CN111986121A (en) | Based on Framellet l0Norm-constrained fuzzy image non-blind restoration method | |
| CN115578638A (en) | Method for constructing multi-level feature interactive defogging network based on U-Net | |
| CN116433516A (en) | Low-illumination image denoising and enhancing method based on attention mechanism | |
| CN115187775A (en) | Semantic segmentation method and device for remote sensing image | |
| CN113256528B (en) | Low-illumination video enhancement method based on multi-scale cascade depth residual error network | |
| CN113450267B (en) | Transfer learning method capable of rapidly acquiring multiple natural degradation image restoration models | |
| CN115496764A (en) | Dense feature fusion-based foggy image semantic segmentation method | |
| CN115660984A (en) | Image high-definition restoration method and device and storage medium | |
| Wang et al. | A CBAM‐GAN‐based method for super‐resolution reconstruction of remote sensing image | |
| CN111046740B (en) | Classification method for human action video based on full tensor cyclic neural network | |
| CN113240589A (en) | Image defogging method and system based on multi-scale feature fusion |
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 |