CN107563510A - A kind of WGAN model methods based on depth convolutional neural networks - Google Patents
A kind of WGAN model methods based on depth convolutional neural networks Download PDFInfo
- Publication number
- CN107563510A CN107563510A CN201710690212.0A CN201710690212A CN107563510A CN 107563510 A CN107563510 A CN 107563510A CN 201710690212 A CN201710690212 A CN 201710690212A CN 107563510 A CN107563510 A CN 107563510A
- Authority
- CN
- China
- Prior art keywords
- neural networks
- arbiter
- convolutional neural
- mrow
- depth convolutional
- 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
Abstract
The invention discloses a kind of WGAN model methods based on depth convolutional neural networks, belong to deep learning field of neural networks, this method comprises the following steps：S1, construction Wo Sesitan production confrontation network WGAN models；S2, arbiter is configured to depth convolutional neural networks structure；S3, maker is configured to transposition convolutional neural networks structure；S4, the loss function to arbiter take the loss function of Wasserstein distances；S5, prepare data set, the network finished to construction is trained.The characteristics of this method resists network model according to generation, creatively propose the make that depth convolutional neural networks are combined with WGAN, maker and arbiter are configured to the form of depth convolutional neural networks, simultaneously using WGAN loss function form, characteristics of image can be learnt in the process of training, and can reflects the quality of generation picture quality according to the size of loss function.
Description
Technical field
The present invention relates to deep learning nerual network technique field, and in particular to a kind of based on depth convolutional neural networks
WGAN model methods.
Background technology
Production confrontation network (Generative Adversarial Network, abbreviation GAN) is by Goodfellow
In the framework that 2014 propose, it is based on the thought of " game theory ", construction maker (generator) and arbiter
(discriminator) two kinds of models, the former generates image by the Uniform noise or gaussian random noise for inputting (0,1), after
Person differentiates to the image of input, it is determined that being the image from data set or the image as caused by maker.Arbiter is every
Completion once judges, resultant error is returned into maker.
However, in traditional GAN models, the loss function of arbiter can not indicate the direction of network training, that is, do not have
One index can reflect the quality of generation image.As a result, network will ceaselessly train down caused by such case,
That is, the training of generation confrontation network does not have clear and definite end condition.
In the WGAN models of standard, maker and arbiter are traditional full connection Rotating fields, are worked as in the process of training
In, maker can only go to correct the image of itself generation by the loss function error-duration model of arbiter.In this case, net
The speed of network training is slow, and maker also has no idea to learn to the feature in data set.If by depth convolutional network with
WGAN is combined, then network can either learn the feature into data set, the enough sizes according to arbiter loss function of and can
Carry out the quality in the direction of network training, i.e. reflection generation picture quality.
The content of the invention
The invention aims to solve drawbacks described above of the prior art, construct one kind and be based on depth convolutional Neural
The WGAN model methods of network.
The purpose of the present invention can be reached by adopting the following technical scheme that：
A kind of WGAN model methods based on depth convolutional neural networks, the dynamic adjustment algorithm comprise the following steps：
S1, construction Wo Sesitan production confrontation network WGAN models, model include maker and arbiter；
S2, arbiter is configured to depth convolutional neural networks structure；
S3, maker is configured to transposition depth convolutional neural networks structure；
S4, the loss function to arbiter take the loss function of Wasserstein distances；
S5, prepare data set, the network finished to construction is trained.
Further, if the depth convolutional neural networks of arbiter are divided into dried layer in described step S2, each layer has phase
The convolution kernel answered, that is, there is corresponding weight parameter.
Further, the transposition depth convolutional network number of plies of maker and the depth convolution of arbiter in described step S3
The neutral net number of plies is identical, and each layer of convolution kernel is the transposition of each layer of convolution kernel in arbiter in maker.
Further, the loss letter of Wasserstein distances is taken to the loss function of arbiter in described step S4
Number, is described as follows：
Arbiter loss function is different from traditional generation confrontation network losses functional form；
The loss function of arbiter can reflect the quality of network generation image.
Further, in described step S5, data set is prepared, the network finished to construction is trained.It is specific as follows：
Prepare data set, wherein, described data set is the set with same type or similar features picture, data set
Scale should be as big as possible, the network finished to construction is trained, and maker is exported into generation image to arbiter, arbiter
Judge.
The present invention is had the following advantages relative to prior art and effect：
The form of present invention arbiter loss function in WGAN according to the conceptual constructs of Wasserstein distances, can
The similarity degree of reflection generation sample distribution and data set sample distribution, so as to indicate the training process of whole network, i.e.,
The quality of reflection generation picture quality, has guidance quality.
Brief description of the drawings
Fig. 1 is the overall structure diagram of WGAN depth convolutional neural networks.
Embodiment
To make the purpose, technical scheme and advantage of the embodiment of the present invention clearer, below in conjunction with the embodiment of the present invention
In accompanying drawing, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is
Part of the embodiment of the present invention, rather than whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art
The every other embodiment obtained under the premise of creative work is not made, belongs to the scope of protection of the invention.
Embodiment
Present embodiment discloses a kind of WGAN model methods based on depth convolutional neural networks, following step is specifically included
Suddenly：
Step S1, Wo Sesitan productions confrontation network WGAN models are constructed, model includes maker and arbiter.
Step S2, arbiter is configured to depth convolutional neural networks structure；
Arbiter is configured to the form of depth convolutional neural networks.If it is divided into dried layer, each layer has corresponding convolution
Core, that is, there is corresponding weight parameter.
Step S3, maker is configured to transposition convolutional neural networks structure；
The convolutional network number of plies of maker is identical with arbiter, and convolution kernel is the transposition of arbiter convolution kernel.
Step S4, the loss function of Wasserstein distances is taken to the loss function of arbiter.In WGAN, differentiate
The loss function of device is：
Wherein, D (x) represents differentiation of the arbiter to image, and pr represents the distribution of data images, and pg represents generation image
Distribution, λ is hyper parameter,For gradient.
The loss function of arbiter is different from traditional generation confrontation network losses functional form；
The loss function of arbiter can reflect the quality of network generation image.
Step S5, data set is prepared, the network finished to construction is trained, and is described as follows：
Wherein, data set should be the set with same type or similar features picture；Also, the scale of data set should
It is as big as possible.
During training, maker output generation image to arbiter, arbiter judges.
In summary, present embodiment discloses a kind of WGAN model methods based on depth convolutional neural networks, the model
Algorithm adds depth convolutional neural networks on the basis of traditional WGAN models, to maker and arbiter, so that generation
Device can work as learning to the characteristics of image of data set sample in the process of network training.In addition, the loss function of arbiter
The direction of network training can be instructed.
Above-described embodiment is the preferable embodiment of the present invention, but embodiments of the present invention are not by above-described embodiment
Limitation, other any Spirit Essences without departing from the present invention with made under principle change, modification, replacement, combine, simplification,
Equivalent substitute mode is should be, is included within protection scope of the present invention.
Claims (5)
1. a kind of WGAN model methods based on depth convolutional neural networks, it is characterised in that described method includes following step
Suddenly：
S1, construction Wo Sesitan production confrontation network WGAN models, the model include maker and arbiter；
S2, arbiter is configured to depth convolutional neural networks structure；
S3, maker is configured to transposition depth convolutional neural networks structure；
S4, the loss function to arbiter are the loss function for taking Wasserstein distances；
S5, prepare data set, the network finished to construction is trained.
A kind of 2. WGAN model methods based on depth convolutional neural networks according to claim 1, it is characterised in that institute
If the depth convolutional neural networks of arbiter are divided into dried layer in the step S2 stated, each layer has corresponding convolution kernel, that is, has corresponding
Weight parameter.
A kind of 3. WGAN model methods based on depth convolutional neural networks according to claim 1, it is characterised in that institute
The transposition depth convolutional network number of plies of maker and the depth convolutional neural networks number of plies of arbiter are identical in the step S3 stated, and
And each layer of convolution kernel is the transposition of each layer of convolution kernel in arbiter in maker.
A kind of 4. WGAN model methods based on depth convolutional neural networks according to claim 1, it is characterised in that institute
That states take, and the loss function of Wasserstein distances is：
<mrow>
<mi>L</mi>
<mrow>
<mo>(</mo>
<mi>D</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mo>-</mo>
<msub>
<mi>E</mi>
<mrow>
<mi>x</mi>
<mo>~</mo>
<mi>pr</mi>
</mrow>
</msub>
<mo>[</mo>
<mi>D</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
<mo>]</mo>
<mo>+</mo>
<msub>
<mi>E</mi>
<mrow>
<mi>x</mi>
<mo>~</mo>
<mi>pg</mi>
</mrow>
</msub>
<mo>[</mo>
<mi>D</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
<mo>]</mo>
<mo>+</mo>
<mi>&lambda;</mi>
<msub>
<mi>E</mi>
<mrow>
<mi>x</mi>
<mo>~</mo>
<mi>X</mi>
</mrow>
</msub>
<msub>
<mo>&dtri;</mo>
<mi>x</mi>
</msub>
</mrow>
Wherein, D (x) represents differentiation of the arbiter to image, and pr represents the distribution of data images, and pg represents point of generation image
Cloth, λ are hyper parameter,For gradient.
A kind of 5. WGAN model methods based on depth convolutional neural networks according to claim 1, it is characterised in that institute
The step S5 stated is specially：
Prepare data set, wherein, described data set is the set with same type or similar features picture, and construction is finished
Network be trained, by maker export generation image to arbiter, arbiter judge.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710690212.0A CN107563510A (en) | 2017-08-14 | 2017-08-14 | A kind of WGAN model methods based on depth convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710690212.0A CN107563510A (en) | 2017-08-14 | 2017-08-14 | A kind of WGAN model methods based on depth convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107563510A true CN107563510A (en) | 2018-01-09 |
Family
ID=60975450
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710690212.0A Pending CN107563510A (en) | 2017-08-14 | 2017-08-14 | A kind of WGAN model methods based on depth convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107563510A (en) |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108090521A (en) * | 2018-01-12 | 2018-05-29 | 广州视声智能科技有限公司 | A kind of image interfusion method and arbiter of production confrontation network model |
| CN108197700A (en) * | 2018-01-12 | 2018-06-22 | 广州视声智能科技有限公司 | A kind of production confrontation network modeling method and device |
| CN108257100A (en) * | 2018-01-12 | 2018-07-06 | 北京奇安信科技有限公司 | A kind of image repair method and server |
| CN108399422A (en) * | 2018-02-01 | 2018-08-14 | 华南理工大学 | A kind of image channel fusion method based on WGAN models |
| CN108460720A (en) * | 2018-02-01 | 2018-08-28 | 华南理工大学 | A method of changing image style based on confrontation network model is generated |
| CN108470208A (en) * | 2018-02-01 | 2018-08-31 | 华南理工大学 | It is a kind of based on be originally generated confrontation network model grouping convolution method |
| CN108763857A (en) * | 2018-05-29 | 2018-11-06 | 浙江工业大学 | A kind of process soft-measuring modeling method generating confrontation network based on similarity |
| CN109002686A (en) * | 2018-04-26 | 2018-12-14 | 浙江工业大学 | A kind of more trade mark chemical process soft-measuring modeling methods automatically generating sample |
| CN109120652A (en) * | 2018-11-09 | 2019-01-01 | 重庆邮电大学 | It is predicted based on difference WGAN network safety situation |
| CN109146868A (en) * | 2018-08-27 | 2019-01-04 | 北京青燕祥云科技有限公司 | 3D Lung neoplasm generation method, device and electronic equipment |
| CN109144987A (en) * | 2018-08-03 | 2019-01-04 | 天津相和电气科技有限公司 | Electric system based on deep learning measures missing values method for reconstructing and its application |
| CN109168003A (en) * | 2018-09-04 | 2019-01-08 | 中国科学院计算技术研究所 | A method of generating the neural network model for being used for video estimation |
| CN109360159A (en) * | 2018-09-07 | 2019-02-19 | 华南理工大学 | A kind of image completion method based on generation confrontation network model |
| CN109410148A (en) * | 2018-11-06 | 2019-03-01 | 电子科技大学 | It is a kind of based on circulation consistency generation confrontation type network photo and traditional Chinese painting between image style transfer method |
| CN109635748A (en) * | 2018-12-14 | 2019-04-16 | 中国公路工程咨询集团有限公司 | The extracting method of roadway characteristic in high resolution image |
| CN109660206A (en) * | 2018-12-20 | 2019-04-19 | 福州大学 | A kind of diagnosing failure of photovoltaic array method based on Wasserstein GAN |
| CN109813542A (en) * | 2019-03-15 | 2019-05-28 | 中国计量大学 | The method for diagnosing faults of air-treatment unit based on production confrontation network |
| CN110119746A (en) * | 2019-05-08 | 2019-08-13 | 北京市商汤科技开发有限公司 | A kind of characteristic recognition method and device, computer readable storage medium |
| CN110874819A (en) * | 2018-08-29 | 2020-03-10 | 中国电信股份有限公司 | Video image restoration method, device and storage medium |
| CN111582348A (en) * | 2020-04-29 | 2020-08-25 | 武汉轻工大学 | Method, device, equipment and storage medium for training condition generating type countermeasure network |
| CN111986142A (en) * | 2020-05-23 | 2020-11-24 | 冶金自动化研究设计院 | Unsupervised enhancement method for surface defect image data of hot-rolled plate coil |
| CN112070209A (en) * | 2020-08-13 | 2020-12-11 | 河北大学 | Stable controllable image generation model training method based on W distance |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106951919A (en) * | 2017-03-02 | 2017-07-14 | 浙江工业大学 | A kind of flow monitoring implementation method based on confrontation generation network |
| CN106997380A (en) * | 2017-03-21 | 2017-08-01 | 北京工业大学 | Imaging spectrum safe retrieving method based on DCGAN depth networks |
| CN107016406A (en) * | 2017-02-24 | 2017-08-04 | 中国科学院合肥物质科学研究院 | The pest and disease damage image generating method of network is resisted based on production |
-
2017
- 2017-08-14 CN CN201710690212.0A patent/CN107563510A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107016406A (en) * | 2017-02-24 | 2017-08-04 | 中国科学院合肥物质科学研究院 | The pest and disease damage image generating method of network is resisted based on production |
| CN106951919A (en) * | 2017-03-02 | 2017-07-14 | 浙江工业大学 | A kind of flow monitoring implementation method based on confrontation generation network |
| CN106997380A (en) * | 2017-03-21 | 2017-08-01 | 北京工业大学 | Imaging spectrum safe retrieving method based on DCGAN depth networks |
Non-Patent Citations (1)
| Title |
|---|
| ISHAAN GULRAJANI ET AL: "Improved Training of Wasserstein GANs", 《MARCHINE LEARNING》 * |
Cited By (29)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108090521B (en) * | 2018-01-12 | 2022-04-08 | 广州视声智能科技股份有限公司 | Image fusion method and discriminator of generative confrontation network model |
| CN108197700A (en) * | 2018-01-12 | 2018-06-22 | 广州视声智能科技有限公司 | A kind of production confrontation network modeling method and device |
| CN108257100A (en) * | 2018-01-12 | 2018-07-06 | 北京奇安信科技有限公司 | A kind of image repair method and server |
| CN108090521A (en) * | 2018-01-12 | 2018-05-29 | 广州视声智能科技有限公司 | A kind of image interfusion method and arbiter of production confrontation network model |
| CN108399422A (en) * | 2018-02-01 | 2018-08-14 | 华南理工大学 | A kind of image channel fusion method based on WGAN models |
| CN108460720A (en) * | 2018-02-01 | 2018-08-28 | 华南理工大学 | A method of changing image style based on confrontation network model is generated |
| CN108470208A (en) * | 2018-02-01 | 2018-08-31 | 华南理工大学 | It is a kind of based on be originally generated confrontation network model grouping convolution method |
| CN109002686A (en) * | 2018-04-26 | 2018-12-14 | 浙江工业大学 | A kind of more trade mark chemical process soft-measuring modeling methods automatically generating sample |
| CN109002686B (en) * | 2018-04-26 | 2022-04-08 | 浙江工业大学 | Multi-grade chemical process soft measurement modeling method capable of automatically generating samples |
| CN108763857A (en) * | 2018-05-29 | 2018-11-06 | 浙江工业大学 | A kind of process soft-measuring modeling method generating confrontation network based on similarity |
| CN109144987A (en) * | 2018-08-03 | 2019-01-04 | 天津相和电气科技有限公司 | Electric system based on deep learning measures missing values method for reconstructing and its application |
| CN109146868A (en) * | 2018-08-27 | 2019-01-04 | 北京青燕祥云科技有限公司 | 3D Lung neoplasm generation method, device and electronic equipment |
| CN110874819A (en) * | 2018-08-29 | 2020-03-10 | 中国电信股份有限公司 | Video image restoration method, device and storage medium |
| CN110874819B (en) * | 2018-08-29 | 2022-06-17 | 中国电信股份有限公司 | Video image restoration method, device and storage medium |
| CN109168003B (en) * | 2018-09-04 | 2020-11-03 | 中国科学院计算技术研究所 | Method for generating neural network model for video prediction |
| CN109168003A (en) * | 2018-09-04 | 2019-01-08 | 中国科学院计算技术研究所 | A method of generating the neural network model for being used for video estimation |
| CN109360159A (en) * | 2018-09-07 | 2019-02-19 | 华南理工大学 | A kind of image completion method based on generation confrontation network model |
| CN109410148A (en) * | 2018-11-06 | 2019-03-01 | 电子科技大学 | It is a kind of based on circulation consistency generation confrontation type network photo and traditional Chinese painting between image style transfer method |
| CN109120652A (en) * | 2018-11-09 | 2019-01-01 | 重庆邮电大学 | It is predicted based on difference WGAN network safety situation |
| CN109635748A (en) * | 2018-12-14 | 2019-04-16 | 中国公路工程咨询集团有限公司 | The extracting method of roadway characteristic in high resolution image |
| CN109635748B (en) * | 2018-12-14 | 2021-09-03 | 中国公路工程咨询集团有限公司 | Method for extracting road characteristics in high-resolution image |
| CN109660206A (en) * | 2018-12-20 | 2019-04-19 | 福州大学 | A kind of diagnosing failure of photovoltaic array method based on Wasserstein GAN |
| CN109813542A (en) * | 2019-03-15 | 2019-05-28 | 中国计量大学 | The method for diagnosing faults of air-treatment unit based on production confrontation network |
| CN110119746A (en) * | 2019-05-08 | 2019-08-13 | 北京市商汤科技开发有限公司 | A kind of characteristic recognition method and device, computer readable storage medium |
| CN111582348A (en) * | 2020-04-29 | 2020-08-25 | 武汉轻工大学 | Method, device, equipment and storage medium for training condition generating type countermeasure network |
| CN111582348B (en) * | 2020-04-29 | 2024-02-27 | 武汉轻工大学 | Training method, device, equipment and storage medium for condition generation type countermeasure network |
| CN111986142A (en) * | 2020-05-23 | 2020-11-24 | 冶金自动化研究设计院 | Unsupervised enhancement method for surface defect image data of hot-rolled plate coil |
| CN112070209A (en) * | 2020-08-13 | 2020-12-11 | 河北大学 | Stable controllable image generation model training method based on W distance |
| CN112070209B (en) * | 2020-08-13 | 2022-07-22 | 河北大学 | Stable controllable image generation model training method based on W distance |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107563510A (en) | A kind of WGAN model methods based on depth convolutional neural networks | |
| CN105046277B (en) | Robust mechanism study method of the feature significance in image quality evaluation | |
| CN106204468A (en) | A kind of image de-noising method based on ReLU convolutional neural networks | |
| CN107679522B (en) | Multi-stream LSTM-based action identification method | |
| CN106204467A (en) | A kind of image de-noising method based on cascade residual error neutral net | |
| CN107886169A (en) | A kind of multiple dimensioned convolution kernel method that confrontation network model is generated based on text image | |
| CN107016406A (en) | The pest and disease damage image generating method of network is resisted based on production | |
| CN103035135B (en) | Children cognitive system based on augment reality technology and cognitive method | |
| CN107767384A (en) | A kind of image, semantic dividing method based on dual training | |
| CN105160678A (en) | Convolutional-neural-network-based reference-free three-dimensional image quality evaluation method | |
| CN106067161A (en) | A kind of method that image is carried out super-resolution | |
| CN107590518A (en) | A kind of confrontation network training method of multiple features study | |
| CN108230278A (en) | A kind of image based on generation confrontation network goes raindrop method | |
| CN109063724A (en) | A kind of enhanced production confrontation network and target sample recognition methods | |
| CN107767413A (en) | A kind of image depth estimation method based on convolutional neural networks | |
| CN106022380A (en) | Individual identity identification method based on deep learning | |
| CN107689034A (en) | A kind of training method of neutral net, denoising method and device | |
| CN110334580A (en) | The equipment fault classification method of changeable weight combination based on integrated increment | |
| CN108765512B (en) | Confrontation image generation method based on multi-level features | |
| CN106991666A (en) | A kind of disease geo-radar image recognition methods suitable for many size pictorial informations | |
| CN102902981B (en) | Violent video detection method based on slow feature analysis | |
| CN106339753A (en) | Method for effectively enhancing robustness of convolutional neural network | |
| CN110349087A (en) | RGB-D image superior quality grid generation method based on adaptability convolution | |
| CN107729901A (en) | Method for building up, device and the image processing method and system of image processing model | |
| CN112668623B (en) | Method and device for generating binaural pin defect sample based on generation countermeasure network |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180109 |