CN109509159A - A kind of end-to-end restored method of UAV Fuzzy image based on deep learning - Google Patents
A kind of end-to-end restored method of UAV Fuzzy image based on deep learning Download PDFInfo
- Publication number
- CN109509159A CN109509159A CN201811381302.2A CN201811381302A CN109509159A CN 109509159 A CN109509159 A CN 109509159A CN 201811381302 A CN201811381302 A CN 201811381302A CN 109509159 A CN109509159 A CN 109509159A
- Authority
- CN
- China
- Prior art keywords
- image
- deep learning
- fuzzy
- uav
- unmanned plane
- 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
- 238000013135 deep learning Methods 0.000 title claims abstract description 38
- 238000000034 method Methods 0.000 title claims abstract description 34
- 238000013136 deep learning model Methods 0.000 claims abstract description 6
- 238000012549 training Methods 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 7
- 238000013178 mathematical model Methods 0.000 claims description 2
- 238000011084 recovery Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 4
- 238000005457 optimization Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 3
- VMXUWOKSQNHOCA-UKTHLTGXSA-N ranitidine Chemical compound [O-][N+](=O)\C=C(/NC)NCCSCC1=CC=C(CN(C)C)O1 VMXUWOKSQNHOCA-UKTHLTGXSA-N 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/73—Deblurring; Sharpening
-
- 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
- G06T2207/10016—Video; Image sequence
-
- 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
- G06T2207/20024—Filtering details
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The present invention discloses a kind of end-to-end restored method of UAV Fuzzy image based on deep learning, belongs to image restoration field.Building deep learning first emulates data set, obtain motion blur blurred picture corresponding with defocusing blurring and clear image pair, construct dense connection coding and decoding deep learning network model, deep learning network model is trained using emulation data set, for the practical blurred picture of unmanned plane shooting, without classifying to vague category identifier, can directly obtain restoring clear image using the model.The present invention gives full play to the self-teaching advantage of deep learning, consider motion blur and defocusing blurring simultaneously in deep learning model, unmanned plane, which can effectively be solved, to be influenced during actual photographed by various situations, and the vague category identifier for shooting image is complicated, the uncertain problem of fog-level.
Description
Technical field
The present invention relates to image restoration fields, refer in particular to a kind of end-to-end recovery of UAV Fuzzy image based on deep learning
Method.
Background technique
In recent years, the range that unmanned plane uses is more and more extensive.Unmanned plane from Large-scale professional unmanned plane to compact entertainment,
It is mainly used for aerial photographing image.Due to unmanned plane lighter weight, it is easy affected by environment, and the requirement to operator
It is higher, therefore the attitude stability that unmanned plane is shot in the sky is poor, the image of shooting can inevitably obscure.If directly nobody
The blurred picture of machine shooting, which is applied to professional domains, the blurred pictures such as investigation, detection, can seriously affect subsequent image treatment effect.
Although the obscure portions image-erasing in the great amount of images captured by unmanned plane can only be retained clear image application,
In actual operation, it is influenced by weather landform etc., clear image can not be taken sometimes or key message is deposited in certain moulds
It pastes in image, therefore, after flight shooting task, it is necessary first to which image restoration is carried out to the blurred picture of unmanned plane shooting
Processing.
Cause the blurred image factor of unmanned plane mainly to have 4 classes: weather reason, unmanned aerial vehicle platform reason, camera itself are former
Cause and atural object environment reason.Weather reason, which is primarily referred to as wind and rainfall, can make unmanned plane during flying unsteady attitude, cause to shoot
Image it is fuzzy；Unmanned aerial vehicle platform reason is primarily referred to as unmanned plane to be influenced by GPS signal difference, and location information floating is larger, is made
At unmanned plane mobile amendment on a large scale in a short time, and the image generated in makeover process is fuzzy；Camera self reason
Image caused by the parameter settings such as focal length of camera lens are incorrect when being primarily referred to as camera shooting obscures；Atural object environment reason is mainly
Refer to that the region of shooting is changed greatly because of light reason or relief so that the camera carried on unmanned plane can not exact focus or
The range of hypsography has exceeded the field depth of camera, so that part atural object in image be caused to be in the fuzzy feelings outside focus
Condition.These four types of blurred image reasons can be divided into motion blur and two kinds of defocus blur according to the fuzzy mode formed.
Existing image recovery method is broadly divided into traditional images restored method and the image based on deep learning frame
Restored method, wherein traditional images restored method is generally basede on the optimization algorithms frame such as blind convolution or non-blind convolution, needs to estimate
The fuzzy core of image is counted, algorithm is complicated, and consuming time is long for image procossing, and single image is needed to handle, and is not suitable for processing nothing
Man-machine captured high-volume blurred picture；The existing image recovery method based on deep learning is generally specific just for certain
The blurred picture of type, vague category identifier is single, high to data set quality requirement, and image captured by unmanned plane, practical fuzzy
Complex genesis, but also including two kinds of vague category identifiers of motion blur and defocus blur, existing deep learning model can not be effective
Vague category identifier is judged, to influence image recovery effect.Therefore, it for the actual complex situation of unmanned plane shooting image, needs
Design significantly more efficient Image Restoration Algorithm.
Summary of the invention
The object of the present invention is to provide a kind of end-to-end restored method of UAV Fuzzy image based on deep learning, is used for
The blurred picture that processing unmanned plane is shot in actual operation, for complicated practical blurred picture, it is clear quickly to recover
Image.
In order to solve the above technical problems, technical scheme is as follows:
A kind of end-to-end restored method of UAV Fuzzy image based on deep learning, which is characterized in that including following several
A step:
Step 1: building deep learning emulates data set: being directed to motion blur, generates motion blur degree uniformly increased M
A motion blur core；For defocusing blurring, defocusing blurring degree uniformly increased M defocusing blurring core is generated；With M generated
A motion blur core and M defocusing blurring core make convolution with N clear images respectively, obtain corresponding MN different fuzzy journeys
The defocus blurred image of the different fog-levels of the motion blur image and MN of degree, final emulation data set is altogether including 2MN
Blurred picture and corresponding 2MN clear images；
Motion blur in the step 1 is that unmanned plane flies along a certain fixed-direction, because flying during shooting image
Scanning frequency degree slightly has image caused by deviation and obscures, and wherein unmanned plane during flying speed is the major parameter for generating motion blur core.
Defocusing blurring in the step 1 be the camera that is carried on unmanned plane in shooting process can not exact focus or because
Hypsography range is fuzzy beyond image caused by camera field depth, generates defocusing blurring core with Gaussian blurring function,
Middle Gaussian Blur nuclear radius is the major parameter for generating defocusing blurring core.
The value range of M is M >=5 in the step 1, and the value range of N is N >=2000.
Step 2: the dense connection coding and decoding deep learning network model of building: the model includes 14 dense connection nets altogether
Network module, wherein it include 4 convolutional layers in each dense connection network module, the filter size in each convolutional layer is 3 ×
3, all exist in module between any two convolutional layer and is directly connected to；In the network, preceding 7 modules are coded portions, latter 7
Module is decoded portion, and each decoder module needs to call the characteristics of image of corresponding coding module in specific calculate；The model
Input be emulate data set in blurred picture, output be emulate data set in clear image；
The convolution operation of dense connection coding and decoding deep learning network model uses multichannel convolutive side in the step 2
Formula, activation primitive use ReLU function.
Step 3: deep to dense connection coding and decoding constructed in step 2 using the emulation data set generated in step 1
Degree learning network model is trained, and is constantly adjusted to the parameter of each network layer in training process, is obtained optimal depth
Spend learning model relevant parameter；
Training deep learning model use environment is Caffe frame in the step 3.
Step 4: obtaining the blurred picture of unmanned plane actual photographed, will directly be trained in the blurred picture input step 3
Deep learning network model in, the clear image after directly quickly being restored.
The invention has the benefit that
The present invention simulates the blurred picture of unmanned plane actual photographed using emulation data set as much as possible, makes full use of depth
The powerful learning ability of network allows deep learning network constantly to learn fuzzy graph in the training process by supervised learning mode
As with the characteristics of image of corresponding clear image.The end-to-end image recovery method of designed deep learning, can handle difference simultaneously
The blurred picture of vague category identifier and different fog-levels can effectively reduce the artificial workload for extracting feature and algorithm design.
Compared with traditional image recovery method based on iteration optimization algorithms, training pattern of the invention, can be with once training is completed
A large amount of different types of practical blurred pictures are quickly handled, manual sort's vague category identifier is not necessarily to, are not necessarily to long-time iteration optimization；With
Existing deep learning image recovery method is compared, and method designed by the present invention is directed to motion blur and defocusing blurring two simultaneously
Kind vague category identifier is trained, and to every kind of specific vague category identifier, can choose different fog-levels according to the actual situation
Range limits deep learning model by single image vague category identifier.Because unmanned plane is during actual photographed by each
The influence of kind situation, captured image vague category identifier is complicated, and fog-level is not fixed, and end designed by the present invention is arrived
End image recovery method can effectively solve the problems in unmanned plane actual photographed image.
Detailed description of the invention
Fig. 1 is the end-to-end restored method flow chart of UAV Fuzzy image based on deep learning；
Fig. 2 is motion blur core schematic diagram；
Fig. 3 is defocusing blurring core schematic diagram；
Fig. 4 is the clear image emulated in data set；
Fig. 5 is the motion blur image for emulating the different fog-levels in data set；
Fig. 6 is the defocus blurred image for emulating the different fog-levels in data set；
Fig. 7 is dense connection coding and decoding deep learning network model schematic diagram；
Fig. 8 is dense connection coding and decoding modular structure schematic diagram；
Fig. 9 is the blurred picture of unmanned plane actual photographed；
Figure 10 is the clear restored image directly obtained using deep learning model.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that the described embodiment is only a part of the embodiment of the present invention, instead of all the embodiments.Based on this
Embodiment in invention, every other reality obtained by those of ordinary skill in the art without making creative efforts
Example is applied, shall fall within the protection scope of the present invention.
As shown in Figure 1, the present embodiment provides a kind of end-to-end restored method of UAV Fuzzy image based on deep learning,
Specific embodiment is as follows:
Step 1: building deep learning emulates data set.
As shown in Fig. 2, being directed to motion blur, motion blur degree uniformly increased 5 motion blur core is generated, mould is moved
Paste is that unmanned plane flies along a certain fixed-direction, the image mould caused by slightly having deviation because of flying speed during shooting image
Paste, wherein unmanned plane during flying speed is the major parameter for generating motion blur core, in a particular embodiment, raw with Matlab software
At 5 motion blur core in Fig. 2, parameter is respectively as follows: 9,19,29,39,49, and unit is pixel number.
As shown in figure 3, being directed to defocusing blurring, defocusing blurring degree uniformly increased 5 defocusing blurring core is generated, mould is defocused
Paste is that the camera that carries in shooting process can not exact focus or because hypsography range is beyond camera depth of field model on unmanned plane
Image caused by enclosing is fuzzy, generates defocusing blurring core with Gaussian blurring function, and wherein Gaussian Blur nuclear radius is to generate to defocus
The major parameter of fuzzy core is respectively as follows: 11,21 with 5 defocusing blurring cores, radius parameter in Matlab Software Create Fig. 3,
31,41,51, unit is pixel number.
In this embodiment, it is demonstrated only with simplest motion blur core with Gaussian Blur core, in practical operation
Cheng Zhong can replace motion blur core or Gaussian Blur with other kinds of fuzzy core according to the practical vague category identifier of unmanned plane
Core, and the generation parameter of motion blur core can also increase the other parameters such as direction.
With 5 motion blur core generated and 5 defocusing blurring cores, make convolution with 2000 clear images respectively,
In a width clear image as shown in figure 4, the motion blur image of corresponding 10000 different fog-levels is obtained, such as Fig. 5 institute
Show and the defocus blurred image of 10000 different fog-levels, as shown in fig. 6, final emulation data set includes 20000 altogether
Open blurred picture and corresponding 20000 clear images.
Step 2: the dense connection coding and decoding deep learning network model of building.
As shown in fig. 7, the model includes 14 dense connection network modules (Dense Block, DB), dense connection net altogether
The specific structure of network module is as shown in Figure 8, wherein include in each dense connection network module 4 convolutional layers (Conv-1,
Conv-2, Conv-3, Conv-4), the filter size in each convolutional layer is 3 × 3, in module between any two convolutional layer
All exist and is directly connected to；In the network, preceding 7 modules are coded portions, and rear 7 modules are decoded portions, each decoding mould
Block needs to call the characteristics of image of corresponding coding module in specific calculate；The input of the model is to emulate obscuring in data set
Image, output are the clear images emulated in data set, and convolution operation uses multichannel convolutive mode, and activation primitive uses ReLU
Function.
Step 3: training deep learning network model.
Using the emulation data set generated in step 1 to dense connection coding and decoding deep learning constructed in step 2
Network model is trained, and is constantly adjusted to the parameter of each network layer in training process, is obtained optimal deep learning
Parameters in Mathematical Model.In specific training process, depth is trained using the Caffe deep learning frame that Ubuntu has is mounted on
Model is practised, is trained using ADAGRAD optimization algorithm, initial learning rate is 0.001, and frequency of training is 500000 times, wherein
When frequency of training is 200000,300000 and 400000, learning rate is reduced by 1/2 respectively.
Step 4: direct end-to-end restored image.
The blurred picture for obtaining unmanned plane actual photographed, as shown in figure 9, the blurred picture is mainly motion blur, directly
By the clear figure in deep learning network model trained in the blurred picture input step 3, after directly quickly being restored
Picture, as shown in Figure 10.
If the image fog-level of unmanned plane actual photographed has exceeded the range that emulation data set is simulated, available depth
Learning model makees preliminary treatment, is then further processed further according to specific fog-level, obtains the blurred picture being more clear, from
And extract effective information.
Above disclosed is only a preferred embodiment of the present invention, cannot limit the power of the present invention with this certainly
Sharp range, therefore equivalent changes made in accordance with the claims of the present invention, are still within the scope of the present invention.
Claims (6)
1. a kind of end-to-end restored method of UAV Fuzzy image based on deep learning, which is characterized in that including following
Step:
Step 1: building deep learning emulates data set: being directed to motion blur, generates motion blur degree uniformly increased M fortune
Dynamic fuzzy core；For defocusing blurring, defocusing blurring degree uniformly increased M defocusing blurring core is generated；It is transported with M generated
Dynamic fuzzy core and M defocusing blurring core make convolution with N clear images respectively, obtain corresponding MN different fog-levels
The defocus blurred image of motion blur image and MN different fog-levels, final emulation data set include 2MN fuzzy altogether
Image and corresponding 2MN clear images；
Step 2: the dense connection coding and decoding deep learning network model of building: the model includes 14 dense connection network moulds altogether
Block, wherein include 4 convolutional layers in each dense connection network module, the filter size in each convolutional layer is 3 × 3, mould
All exist between any two convolutional layer in block and is directly connected to；In the network, preceding 7 modules are coded portions, rear 7 modules
It is decoded portion, each decoder module needs to call the characteristics of image of corresponding coding module in specific calculate；The model it is defeated
Entering is the blurred picture emulated in data set, and output is the clear image emulated in data set；
Step 3: using the emulation data set generated in step 1 to dense connection coding and decoding depth constructed in step 2
It practises network model to be trained, constantly the parameter of each network layer is adjusted in training process, obtains optimal depth
Practise Parameters in Mathematical Model；
Step 4: the blurred picture of unmanned plane actual photographed is obtained, directly by trained depth in the blurred picture input step 3
It spends in learning network model, the clear image after directly quickly being restored.
2. the UAV Fuzzy image end-to-end restored method according to claim 1 based on deep learning, feature exist
In: the motion blur in the step 1 flies for unmanned plane along a certain fixed-direction, because of flying speed during shooting image
Slightly image caused by deviation is fuzzy, and wherein unmanned plane during flying speed is the major parameter for generating motion blur core.
3. the UAV Fuzzy image end-to-end restored method according to claim 1 based on deep learning, feature exist
It is that the camera that carries on unmanned plane in shooting process can not exact focus or because of landform in: defocusing blurring in the step 1
Fluctuating range is fuzzy beyond image caused by camera field depth, defocusing blurring core is generated with Gaussian blurring function, wherein high
This fuzzy nuclear radius is the major parameter for generating defocusing blurring core.
4. the UAV Fuzzy image end-to-end restored method according to claim 1 based on deep learning, feature exist
In: the value range of M is M >=5 in the step 1, and the value range of N is N >=2000.
5. the UAV Fuzzy image end-to-end restored method according to claim 1 based on deep learning, feature exist
In: the convolution operation of dense connection coding and decoding deep learning network model uses multichannel convolutive mode in the step 2, swashs
Function living uses ReLU function.
6. the UAV Fuzzy image end-to-end restored method according to claim 1 based on deep learning, feature exist
In: training deep learning model use environment is Caffe frame in the step 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811381302.2A CN109509159A (en) | 2018-11-20 | 2018-11-20 | A kind of end-to-end restored method of UAV Fuzzy image based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811381302.2A CN109509159A (en) | 2018-11-20 | 2018-11-20 | A kind of end-to-end restored method of UAV Fuzzy image based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109509159A true CN109509159A (en) | 2019-03-22 |
Family
ID=65749066
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811381302.2A Pending CN109509159A (en) | 2018-11-20 | 2018-11-20 | A kind of end-to-end restored method of UAV Fuzzy image based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109509159A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110648291A (en) * | 2019-09-10 | 2020-01-03 | 武汉科技大学 | Unmanned aerial vehicle motion blurred image restoration method based on deep learning |
| CN114119428A (en) * | 2022-01-29 | 2022-03-01 | 深圳比特微电子科技有限公司 | Image deblurring method and device |
| CN114564882A (en) * | 2022-01-29 | 2022-05-31 | 内蒙古工业大学 | Construction and application of edge deep learning simulator based on discrete events |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8861884B1 (en) * | 2011-11-21 | 2014-10-14 | Google Inc. | Training classifiers for deblurring images |
| CN106530256A (en) * | 2016-11-18 | 2017-03-22 | 四川长虹电器股份有限公司 | Improved-deep-learning-based intelligent camera image blind super-resolution system |
| CN107680053A (en) * | 2017-09-20 | 2018-02-09 | 长沙全度影像科技有限公司 | A kind of fuzzy core Optimized Iterative initial value method of estimation based on deep learning classification |
| CN108537746A (en) * | 2018-03-21 | 2018-09-14 | 华南理工大学 | A kind of fuzzy variable method for blindly restoring image based on depth convolutional network |
| CN108765338A (en) * | 2018-05-28 | 2018-11-06 | 西华大学 | Spatial target images restored method based on convolution own coding convolutional neural networks |
| CN108810319A (en) * | 2017-04-26 | 2018-11-13 | 富士通株式会社 | Image processing apparatus and image processing method |
| CN108830801A (en) * | 2018-05-10 | 2018-11-16 | 湖南丹尼尔智能科技有限公司 | A kind of deep learning image recovery method of automatic identification vague category identifier |
-
2018
- 2018-11-20 CN CN201811381302.2A patent/CN109509159A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8861884B1 (en) * | 2011-11-21 | 2014-10-14 | Google Inc. | Training classifiers for deblurring images |
| CN106530256A (en) * | 2016-11-18 | 2017-03-22 | 四川长虹电器股份有限公司 | Improved-deep-learning-based intelligent camera image blind super-resolution system |
| CN108810319A (en) * | 2017-04-26 | 2018-11-13 | 富士通株式会社 | Image processing apparatus and image processing method |
| CN107680053A (en) * | 2017-09-20 | 2018-02-09 | 长沙全度影像科技有限公司 | A kind of fuzzy core Optimized Iterative initial value method of estimation based on deep learning classification |
| CN108537746A (en) * | 2018-03-21 | 2018-09-14 | 华南理工大学 | A kind of fuzzy variable method for blindly restoring image based on depth convolutional network |
| CN108830801A (en) * | 2018-05-10 | 2018-11-16 | 湖南丹尼尔智能科技有限公司 | A kind of deep learning image recovery method of automatic identification vague category identifier |
| CN108765338A (en) * | 2018-05-28 | 2018-11-06 | 西华大学 | Spatial target images restored method based on convolution own coding convolutional neural networks |
Non-Patent Citations (4)
| Title |
|---|
| BING TIAN 等: "Super-resolution Deblurring Algorithm for Generative Adversarial Networks", 《2017 SECOND INTERNATIONAL CONFERENCE ON MECHANICAL, CONTROL AND COMPUTER ENGINEERING》 * |
| GUANBIN LI 等: "Non-locally Enhanced Encoder-Decoder Network for Single", 《MM’18：PROCEEDINGS OF THE 26TH ACM INTERNATIONAL CONFERENCE ON MULTIMEDIA》 * |
| YIWEN HUANG 等: "Densely Connected High Order Residual Network for Single Frame Image Super Resolution", 《ARXIV》 * |
| 缪弘等: "基于深度卷积神经网络的视觉SLAM去模糊系统", 《中兴通讯技术》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110648291A (en) * | 2019-09-10 | 2020-01-03 | 武汉科技大学 | Unmanned aerial vehicle motion blurred image restoration method based on deep learning |
| CN110648291B (en) * | 2019-09-10 | 2023-03-03 | 武汉科技大学 | Unmanned aerial vehicle motion blurred image restoration method based on deep learning |
| CN114119428A (en) * | 2022-01-29 | 2022-03-01 | 深圳比特微电子科技有限公司 | Image deblurring method and device |
| CN114564882A (en) * | 2022-01-29 | 2022-05-31 | 内蒙古工业大学 | Construction and application of edge deep learning simulator based on discrete events |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20200151858A1 (en) | Image contrast enhancement method and device, and storage medium | |
| CN111223062B (en) | Image deblurring method based on generation countermeasure network | |
| CN108376392B (en) | Image motion blur removing method based on convolutional neural network | |
| CN109509159A (en) | A kind of end-to-end restored method of UAV Fuzzy image based on deep learning | |
| CN110969589B (en) | Dynamic scene blurred image blind restoration method based on multi-stream annotating countermeasure network | |
| CN109087255B (en) | Lightweight depth image denoising method based on mixed loss | |
| CN111091091A (en) | Method, device and equipment for extracting target object re-identification features and storage medium | |
| CN111507182B (en) | Skeleton point fusion cyclic cavity convolution-based littering behavior detection method | |
| CN108510451A (en) | A method of the reconstruction car plate based on the double-deck convolutional neural networks | |
| CN109635728A (en) | A kind of isomery pedestrian recognition methods again based on asymmetric metric learning | |
| CN109345609B (en) | Method for denoising mural image and generating line drawing based on convolutional neural network | |
| CN114125216B (en) | Imaging system and imaging method for software defined satellite | |
| CN111626960A (en) | Image defogging method, terminal and computer storage medium | |
| CN109410144A (en) | A kind of end-to-end image defogging processing method based on deep learning | |
| CN110443764A (en) | Video repairing method, device and server | |
| CN109829868A (en) | A kind of lightweight deep learning model image defogging method, electronic equipment and medium | |
| CN107833193A (en) | A kind of simple lens global image restored method based on refinement network deep learning models | |
| CN108830801A (en) | A kind of deep learning image recovery method of automatic identification vague category identifier | |
| CN115086516A (en) | Camera system automatic focusing method based on deep learning model | |
| CN116485646A (en) | Micro-attention-based light-weight image super-resolution reconstruction method and device | |
| CN116188342A (en) | Infrared light and visible light image fusion method combining target detection | |
| CN114913068A (en) | Unmanned aerial vehicle image splicing method and system based on computer vision | |
| CN113112439A (en) | Image fusion method, training method, device and equipment of image fusion model | |
| CN114092354A (en) | Face image restoration method based on generation countermeasure network | |
| CN114067225A (en) | Unmanned aerial vehicle small target detection method and system and storable medium |
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 | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Liu Wantai Inventor after: Liu Lixin Inventor after: Song Weiwei Inventor after: Yan Tao Inventor before: Liu Lixin Inventor before: Liu Wantai Inventor before: Song Weiwei Inventor before: Yan Tao |
|
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190322 |