CN111598793A - Method and system for defogging image of power transmission line and storage medium - Google Patents
Method and system for defogging image of power transmission line and storage medium Download PDFInfo
- Publication number
- CN111598793A CN111598793A CN202010330363.7A CN202010330363A CN111598793A CN 111598793 A CN111598793 A CN 111598793A CN 202010330363 A CN202010330363 A CN 202010330363A CN 111598793 A CN111598793 A CN 111598793A
- Authority
- CN
- China
- Prior art keywords
- network model
- neural network
- defogging
- transmission line
- flow
- 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
- 230000005540 biological transmission Effects 0.000 title claims abstract description 64
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000003062 neural network model Methods 0.000 claims abstract description 158
- 238000013140 knowledge distillation Methods 0.000 claims abstract description 31
- 238000012549 training Methods 0.000 claims description 65
- 230000006870 function Effects 0.000 claims description 20
- 238000012360 testing method Methods 0.000 claims description 9
- 238000010276 construction Methods 0.000 claims description 8
- 238000011176 pooling Methods 0.000 claims description 5
- 230000004927 fusion Effects 0.000 claims description 4
- 238000012545 processing Methods 0.000 abstract description 9
- 238000010586 diagram Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 6
- 230000007547 defect Effects 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Chemical compound O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 239000000428 dust Substances 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000003344 environmental pollutant Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 125000001475 halogen functional group Chemical group 0.000 description 1
- 238000010438 heat treatment Methods 0.000 description 1
- 230000009191 jumping Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 231100000719 pollutant Toxicity 0.000 description 1
- 238000007670 refining Methods 0.000 description 1
Images
Classifications
-
- G06T5/73—
-
- 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
- G06T2207/20—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]
Abstract
The application discloses a method, a system and a storage medium for defogging an image of a power transmission line, the application realizes end-to-end processing of the image of the power transmission line by using a teacher flow depth neural network model, a student flow shallow neural network model and a knowledge distillation neural network model, reduces the steps of artificially estimating parameters, saves time, greatly reduces the parameters of directly using the teacher flow depth neural network model by using the student flow shallow neural network model for learning the teacher flow depth neural network model, and performs defogging processing on the image of the fog through the student flow shallow neural network model with defogging capability, so that a clear image can be obtained.
Description
Technical Field
The present application relates to the field of image processing, and in particular, to a method, a system, and a storage medium for defogging an image of a power transmission line.
Background
With the development of science and technology, mobile phones or cameras have become necessities of life of people. In addition, and to a large scale, many photographs are being taken for high-level computer vision tasks, such as object detection, surveillance systems, identification fields, and so forth. However, the environment in which the photograph is taken may not be good, and the taken photograph may not be clearly visible. The most common causes of deterioration of the picture image quality include the influence of haze, the influence of noise, the influence of blur, and the like. The most important of the above several factors is the effect of haze and fog. The fog mainly refers to fog formed in nature, for example, at night, the capacity of water vapor in the air is reduced, and a part of the water vapor can be condensed into fog. For haze, because of the non-flowing of air, tiny particles in the air are gathered and float in the air; dust on the ground; the discharged automobile exhaust and secondary pollution produced by factories; pollutants such as C O2 discharged in winter heating and the like are all the causes of the haze weather. These reasons not only degrade the quality of the image but also affect the visual effect, while hindering the development of high vision tasks. Thus removing the effect of haze, improving the quality and visual effect of images, and restoring the quality of degraded images, haze removal plays a crucial role in computer vision tasks.
Before the defects of the power transmission line are identified, the collected images cannot well identify the defects in the power transmission line due to the interference of the haze days, so that a method is needed to be adopted, and the haze in the images is removed before the defects are identified.
In order to remove haze and improve the quality of degraded images, researchers have proposed a number of defogging methods, which have achieved good results to some extent. The method comprises the steps of estimating a transmission map, refining the transmission map, estimating global atmospheric light and reconstructing a fog-free image by utilizing an atmospheric scattering model. However, such a method has certain limitations, for example, under some conditions, the estimation of the transmission map based on the a priori method is not accurate, and besides, the estimation method for the traditional atmospheric light is obviously not true in the highlight area and the shadow on the object which is too bright. In addition to such methods, there are some methods based on deep learning, however, these methods are learned on the indoor data set and the synthetic data set, and thus result in low contrast and single picture color. Moreover, these methods are time consuming and space consuming due to the large number of learning parameters. Therefore, how to design a method for defogging an image of a power transmission line, which has high picture quality, less time consumption, less storage space saving and less parameters, becomes an urgent problem to be solved.
Disclosure of Invention
The application provides a method, a system and a storage medium for defogging an image of a power transmission line, which are used for solving the problems of poor quality of a defogged image, long time, large storage space and more required parameters in the existing method.
In a first aspect, the present application provides a method for defogging an image of a power transmission line, including:
s1: selecting a network training data set, wherein the network training data set comprises a training data set and a test set;
s2: constructing a teacher flow depth neural network model, a student flow shallow neural network model and a knowledge distillation neural network model;
s3: carrying out defogging supervision training on the teacher flow deep neural network model through the training data set;
s4: constructing a loss function of the knowledge distillation neural network model through the knowledge distillation neural network model;
s5: training the student flow shallow neural network model through the loss function until the loss item of the student flow shallow neural network model does not decrease any more, and stopping training;
s6: the trained student flow shallow neural network model learns the trained teacher flow deep neural network model to obtain a student flow shallow neural network model with defogging capability;
s7: and inputting the foggy electric transmission line picture into the student flow shallow neural network model with the defogging capability, and carrying out defogging treatment on the foggy electric transmission line picture by the student flow shallow neural network model with the defogging capability and outputting the defogged electric transmission line picture.
Alternatively, the training data set is a RESIDE data set.
Optionally, the test set is a transmission line foggy image
Optionally, the RESIDE data set comprises a transmission line foggy image and a clear image, and the RESIDE data set further comprises a depth map, a synthetic foggy image and a real foggy image, wherein the depth map corresponds to the foggy image and the clear image.
Optionally, the teacher flow depth neural network model includes 121 convolutional layers, and the teacher flow depth neural network model includes a spatial pyramid pooling algorithm and a hole convolution algorithm.
Optionally, the student flow shallow neural network model comprises 16 convolutional layers, and the student flow shallow neural network model comprises a jump connection algorithm, a hole convolution algorithm and a multi-scale fusion algorithm.
In a second aspect, the application provides a system for defogging images of a power transmission line, which comprises a network training data set selection module, a network model construction module, a loss function construction module, a student flow shallow neural network model training module, a learning module and a defogging module, wherein:
the network training data set selection module is used for selecting a network training data set, and the network training data set comprises a training data set and a test set;
the network model building module is used for building a teacher flow depth neural network model, a student flow shallow neural network model and a knowledge distillation neural network model;
the defogging supervision training module is used for performing defogging supervision training on the teacher flow deep neural network model through the training data set;
the loss function building module is used for building a loss function of the knowledge distillation neural network model through the knowledge distillation neural network model;
the student flow shallow neural network model training module is used for training the student flow shallow neural network model through the loss function until the loss item of the student flow shallow neural network model does not decrease any more, and stopping training;
the learning module is used for learning the trained student flow shallow neural network model to the trained teacher flow deep neural network model to obtain a student flow shallow neural network model with defogging capability;
the defogging module is used for inputting the foggy electric transmission line picture into the student flow shallow neural network model with the defogging capability, and the student flow shallow neural network model with the defogging capability is used for defogging the foggy electric transmission line picture and outputting the defogged electric transmission line picture.
In a third aspect, the present application provides a storage medium for image defogging of a power transmission line, wherein the computer executable instructions, when executed by a computer processor, implement the method for image defogging of a power transmission line according to the first aspect.
According to the technical scheme, the end-to-end processing of the transmission line fogging images is realized by using the teacher flow depth neural network model, the student flow shallow neural network model and the knowledge distillation neural network model, the steps of manually estimating parameters are reduced, time is saved, the parameters of the teacher flow depth neural network model directly used are greatly reduced by using the student flow shallow neural network model to learn the teacher flow depth neural network model, and the fogging images are subjected to defogging processing through the student flow shallow neural network model with defogging capability, so that clear images can be obtained.
Drawings
In order to more clearly explain the technical solution of the present application, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious to those skilled in the art that other drawings can be obtained according to the drawings without any creative effort.
Fig. 1 is a flowchart of a method for defogging an image of a power transmission line provided by the present application;
FIG. 2 is a schematic diagram of hole convolutions with different convolution rates for a method for defogging an image of a power transmission line according to the present application;
fig. 3 is a block diagram of the defogging network of the present invention.
Detailed Description
The technical solutions in the embodiments of the present application are clearly and completely described below with reference to the drawings in the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application, but the present application may be practiced in other ways than those described again, and it will be apparent to those of ordinary skill in the art that the present application is not limited to the specific embodiments disclosed below.
In a first aspect, referring to fig. 1 and 3, the present application provides a method for defogging an image of a power transmission line, including:
s1: selecting a network training data set, wherein the network training data set comprises a training data set and a test set;
s2: constructing a teacher flow depth neural network model, a student flow shallow neural network model and a knowledge distillation neural network model;
the construction of the teacher flow depth neural network model is based on dense connection, and the construction of the student flow shallow neural network model is based on u-net.
S3: carrying out defogging supervision training on the teacher flow deep neural network model through the training data set;
s4: constructing a loss function of the knowledge distillation neural network model through the knowledge distillation neural network model;
s5: training the student flow shallow neural network model through the loss function until the loss item of the student flow shallow neural network model does not decrease any more, and stopping training;
s6: the trained student flow shallow neural network model learns the trained teacher flow deep neural network model to obtain a student flow shallow neural network model with defogging capability;
the trained student flow shallow neural network model learns the trained teacher flow deep neural network model, the trained student flow shallow neural network model can learn the defogging capacity of the trained teacher flow deep neural network model, and meanwhile parameters of the neural network model are greatly reduced under the condition that a certain effect is guaranteed.
S7: and inputting the foggy electric transmission line picture into the student flow shallow neural network model with the defogging capability, and carrying out defogging treatment on the foggy electric transmission line picture by the student flow shallow neural network model with the defogging capability and outputting the defogged electric transmission line picture.
This application has realized the end-to-end processing to transmission line's fog image through using teacher's class depth neural network model, student's class shallow neural network model and knowledge distillation neural network model, the step of artificially estimating the parameter has been reduced, the time is saved, through using student's class shallow neural network model to teacher's study of class depth neural network model, the parameter of the direct teacher's class depth neural network model that uses has significantly reduced, carry out defogging processing to the fog image through the student who has defogging ability class shallow neural network model, can obtain clear image.
Alternatively, the training data set is a RESIDE data set.
Optionally, the test set is a transmission line foggy image
Optionally, the RESIDE data set comprises a transmission line foggy image and a clear image, and the RESIDE data set further comprises a depth map, a synthetic foggy image and a real foggy image, wherein the depth map corresponds to the foggy image and the clear image.
Optionally, the teacher flow depth neural network model includes 121 convolutional layers (see table 1), and the teacher flow depth neural network model includes a spatial pyramid pooling algorithm and a hole convolution algorithm.
TABLE 1
A spatial pyramid pooling algorithm and a hole convolution method are added in the teacher flow deep neural network model, different convolution rates correspond to different receptive fields, and the receptive fields of convolution kernels can be increased under the condition that the same parameters are used.
The teacher flow depth neural network model adds pyramid pooling operation on the basis of a dense connection network, and simultaneously adds hole convolution operation, so that the receptive field of the neural network can be expanded on the basis of the original network, and more detailed characteristics can be extracted.
Optionally, the student flow shallow neural network model comprises 16 convolutional layers (see table 2), and the student flow shallow neural network model comprises a jump connection algorithm, a hole convolution algorithm and a multi-scale fusion algorithm.
TABLE 2
In the student flow shallow neural network model, algorithms of jump connection, hole convolution (see fig. 2) and multi-scale fusion are added.
Compared with the traditional convolution method, different convolution rates correspond to different receptive fields, and the receptive fields of the convolution kernels can be increased under the condition that the same parameters are used. In the student flow shallow neural network model, because the number of network layers is small and the number of convolution kernels is limited, the characteristic diagram output by the network cannot be well guaranteed to contain rich information, therefore, a jump connection method similar to that in a dense connection network is adopted to guarantee that the network can contain more information, and in order to further enlarge the receptive field, a multi-scale fusion method is added after the hole convolution is used to fuse the characteristic diagrams with different convolution rates, so that the characteristic diagram contains more information.
The method uses 16 convolutional layers as a student flow shallow neural network model, jumping connection is added on the basis of the original network, the network can extract more deep and shallow information, and meanwhile, the student flow shallow neural network model adopts cavity convolution, and the convolutional neural network can extract more scales of information.
The teacher flow deep neural network model is a pre-trained deep residual neural network, and can extract expensive high-frequency features from the decomposition image of LZM conversion. The student flow shallow neural network model is a convolutional neural network for extracting features, and the high-frequency information required by the output image is extracted by restricting the extraction of the student flow shallow neural network model to expensive features through knowledge distillation. Knowledge distillation is carried out by adopting regression loss, and the formula is as follows:
Ltz＝||ft(x)-fs(x)||2
wherein f ist(x) Is a feature extracted from the "teacher flow" network, fs(x) Is a feature obtained by a student flow network, | | | | | luminance2Is represented by2And (5) carrying out norm operation.
And measuring the distance of a characteristic level between the teacher flow depth neural network model and the student flow shallow neural network model. For the knowledge distillation method of the defogged image, the distance measurement is carried out between the characteristic diagram output by the teacher flow depth neural network model and the characteristic diagram output by the student flow shallow neural network model pixel by pixel, and common formulas such as mean square error can be adopted. The traditional knowledge distillation algorithm usually performs knowledge distillation on the feature map output at the last layer, but the method usually obtains less teacher flow information, and information is lost. The invention adopts a multilayer knowledge distillation method, can better retain the information of the teacher flow depth neural network model, and achieves better defogging effect.
The method can better solve the problem that the halo effect of the sky area is easy to appear in the traditional defogging and clearing algorithm, so that the defogging effect is obviously improved.
The invention can better solve the problem that the detail image is lost in the traditional defogging method and effectively improve the quality of the defogged image.
The invention can compress the defogging model into a smaller network, can directly run on the mobile equipment, can better meet the requirements on defogging of the image of the power transmission line, has lower realization complexity and has the characteristic of wide application.
In a second aspect, the application provides a system for defogging images of a power transmission line, which comprises a network training data set selection module, a network model construction module, a loss function construction module, a student flow shallow neural network model training module, a learning module and a defogging module, wherein:
the network training data set selection module is used for selecting a network training data set, and the network training data set comprises a training data set and a test set;
the network model building module is used for building a teacher flow depth neural network model, a student flow shallow neural network model and a knowledge distillation neural network model;
the defogging supervision training module is used for performing defogging supervision training on the teacher flow deep neural network model through the training data set;
the loss function building module is used for building a loss function of the knowledge distillation neural network model through the knowledge distillation neural network model;
the student flow shallow neural network model training module is used for training the student flow shallow neural network model through the loss function until the loss item of the student flow shallow neural network model does not decrease any more, and stopping training;
the learning module is used for learning the trained student flow shallow neural network model to the trained teacher flow deep neural network model to obtain a student flow shallow neural network model with defogging capability;
the defogging module is used for inputting the foggy electric transmission line picture into the student flow shallow neural network model with the defogging capability, and the student flow shallow neural network model with the defogging capability is used for defogging the foggy electric transmission line picture and outputting the defogged electric transmission line picture.
In a third aspect, the present application provides a storage medium for image defogging of a power transmission line, wherein the computer executable instructions, when executed by a computer processor, implement the method for image defogging of a power transmission line according to the first aspect.
The application provides a method, a system and a storage medium for defogging images of a power transmission line, wherein the method for defogging images of the power transmission line comprises the steps of selecting a network training data set; constructing a teacher flow depth neural network model, a student flow shallow neural network model and a knowledge distillation neural network model; carrying out defogging supervision training on the teacher flow deep neural network model through the training data set; constructing a loss function of the knowledge distillation neural network model through the knowledge distillation neural network model; training the student flow shallow neural network model through the loss function until the loss item of the student flow shallow neural network model does not decrease any more, and stopping training; the trained student flow shallow neural network model learns the trained teacher flow deep neural network model to obtain a student flow shallow neural network model with defogging capability; and inputting the foggy electric transmission line picture into the student flow shallow neural network model with the defogging capability, and carrying out defogging treatment on the foggy electric transmission line picture by the student flow shallow neural network model with the defogging capability and outputting the defogged electric transmission line picture.
This application has realized the end-to-end processing to transmission line's fog image through using teacher's class depth neural network model, student's class shallow neural network model and knowledge distillation neural network model, the step of artificially estimating the parameter has been reduced, the time is saved, through using student's class shallow neural network model to teacher's study of class depth neural network model, the parameter of the direct teacher's class depth neural network model that uses has significantly reduced, carry out defogging processing to the fog image through the student who has defogging ability class shallow neural network model, can obtain clear image.
The foregoing is merely a detailed description of the present application, and it should be noted that modifications and embellishments could be made by those skilled in the art without departing from the principle of the present application, and these should also be considered as the protection scope of the present application.
Claims (8)
1. A method for defogging an image of a power transmission line, the method comprising:
s1: selecting a network training data set, wherein the network training data set comprises a training data set and a test set;
s2: constructing a teacher flow depth neural network model, a student flow shallow neural network model and a knowledge distillation neural network model;
s3: carrying out defogging supervision training on the teacher flow deep neural network model through the training data set;
s4: constructing a loss function of the knowledge distillation neural network model through the knowledge distillation neural network model;
s5: training the student flow shallow neural network model through the loss function until the loss item of the student flow shallow neural network model does not decrease any more, and stopping training;
s6: the trained student flow shallow neural network model learns the trained teacher flow deep neural network model to obtain a student flow shallow neural network model with defogging capability;
s7: and inputting the foggy electric transmission line picture into the student flow shallow neural network model with the defogging capability, and carrying out defogging treatment on the foggy electric transmission line picture by the student flow shallow neural network model with the defogging capability and outputting the defogged electric transmission line picture.
2. The method for defogging images of an electric transmission line according to claim 1, wherein said training data set is a RESIDE data set.
3. The method for defogging an image of a power transmission line according to claim 1, wherein said test set is a fogging image of the power transmission line.
4. The method for defogging an image of a power transmission line according to claim 2, wherein said RESIDE dataset comprises a foggy image and a clear image of the power transmission line, and further comprises a depth map, a synthetic foggy image and a real foggy image corresponding to said foggy image and said clear image.
5. The method for defogging the image of the power transmission line according to claim 1, wherein the teacher flow depth neural network model comprises 121 convolutional layers, and the teacher flow depth neural network model comprises a spatial pyramid pooling algorithm and a hole convolution algorithm.
6. The method for defogging the image of the power transmission line according to claim 1, wherein the student flow shallow neural network model comprises 16 convolutional layers, and the student flow shallow neural network model comprises a jump connection algorithm, a hole convolution algorithm and a multi-scale fusion algorithm.
7. The utility model provides a system for transmission line image defogging, its characterized in that, the system includes that network training data set chooses module, network model construction module, loss function construction module, student flow shallow neural network model training module, learning module and defogging module, wherein:
the network training data set selection module is used for selecting a network training data set, and the network training data set comprises a training data set and a test set;
the network model building module is used for building a teacher flow depth neural network model, a student flow shallow neural network model and a knowledge distillation neural network model;
the defogging supervision training module is used for performing defogging supervision training on the teacher flow deep neural network model through the training data set;
the loss function building module is used for building a loss function of the knowledge distillation neural network model through the knowledge distillation neural network model;
the student flow shallow neural network model training module is used for training the student flow shallow neural network model through the loss function until the loss item of the student flow shallow neural network model does not decrease any more, and stopping training;
the learning module is used for learning the trained student flow shallow neural network model to the trained teacher flow deep neural network model to obtain a student flow shallow neural network model with defogging capability;
the defogging module is used for inputting the foggy electric transmission line picture into the student flow shallow neural network model with the defogging capability, and the student flow shallow neural network model with the defogging capability is used for defogging the foggy electric transmission line picture and outputting the defogged electric transmission line picture.
8. A storage medium for image defogging of an electric transmission line, wherein said computer executable instructions, when executed by a computer processor, implement the method for image defogging of an electric transmission line according to any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010330363.7A CN111598793A (en) | 2020-04-24 | 2020-04-24 | Method and system for defogging image of power transmission line and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010330363.7A CN111598793A (en) | 2020-04-24 | 2020-04-24 | Method and system for defogging image of power transmission line and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111598793A true CN111598793A (en) | 2020-08-28 |
Family
ID=72188961
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010330363.7A Pending CN111598793A (en) | 2020-04-24 | 2020-04-24 | Method and system for defogging image of power transmission line and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111598793A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112862095A (en) * | 2021-02-02 | 2021-05-28 | 浙江大华技术股份有限公司 | Self-distillation learning method and device based on characteristic analysis and readable storage medium |
| CN113066025A (en) * | 2021-03-23 | 2021-07-02 | 河南理工大学 | Image defogging method based on incremental learning and feature and attention transfer |
| CN114565539A (en) * | 2022-03-17 | 2022-05-31 | 中国人民解放军火箭军工程大学 | Image defogging method based on online knowledge distillation |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108830288A (en) * | 2018-04-25 | 2018-11-16 | 北京市商汤科技开发有限公司 | Image processing method, the training method of neural network, device, equipment and medium |
| CN108830813A (en) * | 2018-06-12 | 2018-11-16 | 福建帝视信息科技有限公司 | A kind of image super-resolution Enhancement Method of knowledge based distillation |
| CN109146810A (en) * | 2018-08-08 | 2019-01-04 | 国网浙江省电力有限公司信息通信分公司 | A kind of image defogging method based on end-to-end deep learning |
| CN109801232A (en) * | 2018-12-27 | 2019-05-24 | 北京交通大学 | A kind of single image to the fog method based on deep learning |
-
2020
- 2020-04-24 CN CN202010330363.7A patent/CN111598793A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108830288A (en) * | 2018-04-25 | 2018-11-16 | 北京市商汤科技开发有限公司 | Image processing method, the training method of neural network, device, equipment and medium |
| CN108830813A (en) * | 2018-06-12 | 2018-11-16 | 福建帝视信息科技有限公司 | A kind of image super-resolution Enhancement Method of knowledge based distillation |
| CN109146810A (en) * | 2018-08-08 | 2019-01-04 | 国网浙江省电力有限公司信息通信分公司 | A kind of image defogging method based on end-to-end deep learning |
| CN109801232A (en) * | 2018-12-27 | 2019-05-24 | 北京交通大学 | A kind of single image to the fog method based on deep learning |
Non-Patent Citations (3)
| Title |
|---|
| 袁配配;张良;: "基于深度学习的行人属性识别", 激光与光电子学进展, no. 06 * |
| 陈永: "基于双域分解的多尺度深度学习单幅图像去雾", 《光学学报》, pages 1 - 11 * |
| 陈永;郭红光;艾亚鹏;: "基于双域分解的多尺度深度学习单幅图像去雾", 光学学报, no. 02 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112862095A (en) * | 2021-02-02 | 2021-05-28 | 浙江大华技术股份有限公司 | Self-distillation learning method and device based on characteristic analysis and readable storage medium |
| CN112862095B (en) * | 2021-02-02 | 2023-09-29 | 浙江大华技术股份有限公司 | Self-distillation learning method and device based on feature analysis and readable storage medium |
| CN113066025A (en) * | 2021-03-23 | 2021-07-02 | 河南理工大学 | Image defogging method based on incremental learning and feature and attention transfer |
| CN113066025B (en) * | 2021-03-23 | 2022-11-18 | 河南理工大学 | Image defogging method based on incremental learning and feature and attention transfer |
| CN114565539A (en) * | 2022-03-17 | 2022-05-31 | 中国人民解放军火箭军工程大学 | Image defogging method based on online knowledge distillation |
| CN114565539B (en) * | 2022-03-17 | 2023-04-07 | 中国人民解放军火箭军工程大学 | Image defogging method based on online knowledge distillation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106910175B (en) | Single image defogging algorithm based on deep learning | |
| CN110796009A (en) | Method and system for detecting marine vessel based on multi-scale convolution neural network model | |
| CN110378849B (en) | Image defogging and rain removing method based on depth residual error network | |
| CN111598793A (en) | Method and system for defogging image of power transmission line and storage medium | |
| CN112907479B (en) | Residual single image rain removing method based on attention mechanism | |
| CN109584188B (en) | Image defogging method based on convolutional neural network | |
| CN110517203B (en) | Defogging method based on reference image reconstruction | |
| CN109993804A (en) | A kind of road scene defogging method generating confrontation network based on condition | |
| CN110288550B (en) | Single-image defogging method for generating countermeasure network based on priori knowledge guiding condition | |
| CN111681180B (en) | Priori-driven deep learning image defogging method | |
| CN110097522B (en) | Single outdoor image defogging method based on multi-scale convolution neural network | |
| CN109410144A (en) | A kind of end-to-end image defogging processing method based on deep learning | |
| CN109978799B (en) | Maritime unmanned aerial vehicle video image defogging method based on deep learning | |
| CN103578083A (en) | Single image defogging method based on joint mean shift | |
| CN113436124B (en) | Single image defogging method applied to ocean foggy environment | |
| CN110443759A (en) | A kind of image defogging method based on deep learning | |
| Qian et al. | CIASM-Net: a novel convolutional neural network for dehazing image | |
| CN115497015A (en) | River floating pollutant identification method based on convolutional neural network | |
| CN109685735B (en) | Single picture defogging method based on fog layer smoothing prior | |
| CN113487509B (en) | Remote sensing image fog removal method based on pixel clustering and transmissivity fusion | |
| CN114119586A (en) | Intelligent detection method for aircraft skin defects based on machine vision | |
| CN112070691A (en) | Image defogging method based on U-Net | |
| CN112164010A (en) | Multi-scale fusion convolution neural network image defogging method | |
| CN112150395A (en) | Encoder-decoder network image defogging method combining residual block and dense block | |
| CN111767826A (en) | Timing fixed-point scene abnormity detection method |
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 |