CN110147842A - Bridge Crack detection and classification method based on condition filtering GAN - Google Patents
Bridge Crack detection and classification method based on condition filtering GAN Download PDFInfo
- Publication number
- CN110147842A CN110147842A CN201910430207.5A CN201910430207A CN110147842A CN 110147842 A CN110147842 A CN 110147842A CN 201910430207 A CN201910430207 A CN 201910430207A CN 110147842 A CN110147842 A CN 110147842A
- Authority
- CN
- China
- Prior art keywords
- gan
- label
- network
- crack
- bridge
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to a kind of Bridge Crack detection and classification method based on condition filtering GAN network, it is related to Crack Detection technical field, the following steps are included: S1: according to three classes bridge danger classes, respectively choose appropriate picture, and add corresponding label, image pixel is chosen according to the actual situation, establishes database, and the three classes bridge danger classes includes linear crack, blocky crack and chicken-wire cracking；S2: set up the condition filters GAN network；S3: training arbiter D and generator G uses fixed party during training, updates the network weight of another party, alternating iteration；S4: picture to be sorted is inputted to condition filtering GAN network, corresponding label is exported, obtains classification results.
Description
Technical field
The invention belongs to the sorting technique fields of Bridge Crack, are related to a kind of Bridge Crack inspection based on condition filtering GAN
Survey and classification method.
Background technique
As the national projects such as China's economic development, the quickening of urbanization process, high-speed rail develop rapidly, in highway, railway
Or in the rural water conservancy construction of city, all kinds of bridge numbers for the great-leap-forward obstacle built increasingly are increased sharply, and bridge is in national economy
Very important effect is played in development, while being also a kind of embodiment of China's comprehensive strength.Due to the universal existence of bridge,
The structural and persistence of pontic can not be ignored.Crack is as a kind of main crossstructure Disease Characters, to crossstructure
Durability and the harm that generates of safety it is maximum, therefore, crack makes one of primary evaluation index of its health status.
Current Crack Detection classification method is also various, and the essential characteristic that most of algorithm utilizes is consistent, and
And the process of algorithm is also roughly the same: pretreatment, and crack area detection and segmentation, post-processing are described with feature, calculates with classification
Method is classified, and existing Bridge Crack detection sorting algorithm is there are still more defect, far from meeting its demand.
Production confrontation network processes image has following four advantage:
1, according to it is actual as a result, they are seemed can be produced than other models better sample (image is sharper keen,
Clearly).
2, any generator network can be trained by generating confrontation type network frame.
3, the model that design follows any kind of factorization is not needed, any generator network and any discriminator are all
It may be useful for.
4, it is sampled repeatedly without using Markov Chain, without being inferred in learning process, has avoided approximate calculation
The problem of intractable probability.
However, this method for not needing to model in advance of CGAN is the disadvantage is that too free, it is more for biggish picture
Pixel situation, the mode based on simple GAN is just less controllable.In order to solve the too free this problem of GAN, it is exactly
To GAN plus some constraints, CGAN then there has been.The powerful place of CGAN is automatically learn original authentic specimen collection
Data distribution can simply and effectively classify for high-volume image data.
CGAN can be regarded to a special case of condition filtering GAN as, condition filtering GAN directly uses observable variable (example
Such as, binary indicator existing for attribute) indicate attribute, therefore its controllability is extremely strong, is to generate to have obvious geometrical characteristic
Label graphic provide convenience.When the variation of attribute is not optional, it has architecture identical with CGAN, because
This, condition filtering GAN can be counted as the natural extension of CGAN.
In short, the feature for detecting crack is varied, but simple and efficient detect carries out crack in turn
Classification is a difficult point.How to generate the disaggregated model of precise and high efficiency, thus quickly judge Bridge Crack type all right and wrong often with
Challenging problem.
Summary of the invention
In view of this, the Bridge Crack detection and classification method that the present invention provides a kind of based on condition filtering GAN network,
Condition filtering GAN network, not only rich in expressive force, with more the controllability of attribute, we, which can be generated or be edited by GAC, schemes
Picture, while the intuitively variation of controlled attribute.And it is so controllable, so that can intuitively find and obtain desired has
The label image of obvious geometrical characteristic.And then the carry out Bridge Crack classification of precise and high efficiency.
In order to achieve the above objectives, the invention provides the following technical scheme:
A kind of Bridge Crack detection and classification method based on condition filtering GAN network, comprising the following steps:
S1: according to three classes bridge danger classes, appropriate picture is respectively chosen, and add corresponding label, selected according to the actual situation
Image pixel is taken, database is established, the three classes bridge danger classes includes linear crack, blocky crack and chicken-wire cracking；
S2: set up the condition filters GAN network；
S3: training arbiter D and generator G uses fixed party during training, updates the network weight of another party
Weight, alternating iteration；
S4: picture to be sorted is inputted to condition filtering GAN network, corresponding label is exported, obtains classification results.
Further, in step S1, label uses the form of same pixel picture, and the label in linear crack is that a white background has
The figure of a black line, the label in blocky crack be a white background have two intersection black line figure, chicken-wire cracking
Label is the figure that a white background has the black line of four intersections and closed figure of encircling a city.
Further, in step S2, the target formula of the training of condition filtering GAN network is as follows:
In above-mentioned formula, x is a true input picture；One supplementary variable y is (having the same with the y in CGAN
Effect)；Noise variance zi~Pzi(zi)、za~Pza(za) and y~Py (y), (zaIt is generated by y, za'=fy(za) (f in formulay
It is one and z is mapped according to yaScreening function, z ' is a condition latent variable, is input into generator input terminal and influences defeated
Out))；ziIt is a unconditional latent variable (with the z effect having the same in CGAN).
Detailed process can be divided into two parts and describe: generate in network: by noise variance zi~Pzi(zi)、za~Pza
(za) and y~Py (y) be mapped to data space x=tt (zi,za, y) generator tt, x at this time is generated by generator.
Differentiate in network: when x is a true input picture, distributing p=D (x, y) ∈ [0,1] and p=D of a detectivity
(tt(zi,za, y), y) the arbiter D (p is determined as true probability) of ∈ [0,1] distributes a probability 1-p when x is generated by tt.
Pzi(zi)、Pza(za) and Py (y) is priori value on z, generally selects uniform [- 1,1] distribution.
Further, in step S4, using the property control function of condition filtering GAN conditional variable z ', convenient for strengthening simultaneously
Control critical eigenvalue region, and generate the label of corresponding geometric attribute.
The beneficial effects of the present invention are: the present invention is had more using condition filtering GAN network not only rich in expressive force
The controllability of attribute generates by GAC or edits image, while the intuitively variation of controlled attribute.And it is so controllable, so that
In can intuitively find and obtain the desired label image with obvious geometrical characteristic.And then the carry out bridge of precise and high efficiency
Classification of rifts.
Other advantages, target and feature of the invention will be illustrated in the following description to a certain extent, and
And to a certain extent, based on will be apparent to those skilled in the art to investigating hereafter, Huo Zheke
To be instructed from the practice of the present invention.Target of the invention and other advantages can be realized by following specification and
It obtains.
Detailed description of the invention
To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention is made below in conjunction with attached drawing excellent
The detailed description of choosing, in which:
Fig. 1 a is linear crack original image；
Fig. 1 b is linear crack label；
Fig. 1 c is blocky crack original image；
Fig. 1 d is blocky crack label；
Fig. 1 e is chicken-wire cracking original image；
Fig. 1 f is chicken-wire cracking label；
Fig. 2 is condition filtering GAN network mechanism of the present invention.
Specific embodiment
Illustrate embodiments of the present invention below by way of specific specific example, those skilled in the art can be by this specification
Other advantages and efficacy of the present invention can be easily understood for disclosed content.The present invention can also pass through in addition different specific realities
The mode of applying is embodied or practiced, the various details in this specification can also based on different viewpoints and application, without departing from
Various modifications or alterations are carried out under spirit of the invention.It should be noted that diagram provided in following embodiment is only to show
Meaning mode illustrates basic conception of the invention, and in the absence of conflict, the feature in following embodiment and embodiment can phase
Mutually combination.
Wherein, the drawings are for illustrative purposes only and are merely schematic diagrams, rather than pictorial diagram, should not be understood as to this
The limitation of invention；Embodiment in order to better illustrate the present invention, the certain components of attached drawing have omission, zoom in or out, not
Represent the size of actual product；It will be understood by those skilled in the art that certain known features and its explanation may be omitted and be in attached drawing
It is understood that.
The same or similar label correspond to the same or similar components in the attached drawing of the embodiment of the present invention；It is retouched in of the invention
In stating, it is to be understood that if there is the orientation or positional relationship of the instructions such as term " on ", "lower", "left", "right", "front", "rear"
To be based on the orientation or positional relationship shown in the drawings, be merely for convenience of description of the present invention and simplification of the description, rather than indicate or
It implies that signified device or element must have a particular orientation, be constructed and operated in a specific orientation, therefore is described in attached drawing
The term of positional relationship only for illustration, is not considered as limiting the invention, for the ordinary skill of this field
For personnel, the concrete meaning of above-mentioned term can be understood as the case may be.
A kind of Bridge Crack detection and classification method based on condition filtering GAN network, comprising the following steps:
S1: according to three classes bridge danger classes, appropriate picture is respectively chosen, and add corresponding label, selected according to the actual situation
Image pixel is taken, database is established, the three classes bridge danger classes includes linear crack, blocky crack and chicken-wire cracking；
S2: set up the condition filters GAN network, as shown in Figure 2；
S3: training arbiter D and generator G uses fixed party during training, updates the network weight of another party
Weight, alternating iteration；
S4: picture to be sorted is inputted to condition filtering GAN network, corresponding label is exported, obtains classification results.
Optionally, as shown in Fig. 1 a~f, in step S1, label uses the form of same pixel picture, the mark in linear crack
Label are the figures that a white background has a black line, and the label in blocky crack is the black line that a white background has two intersections
Figure, the label of chicken-wire cracking are the figures that a white background has the black line of four intersections and closed figure of encircling a city.
Optionally, in step S2, the target formula of the training of condition filtering GAN network is as follows:
In above-mentioned formula, x is a true input picture；One supplementary variable y is (having the same with the y in CGAN
Effect)；Noise variance zi~Pzi(zi)、za~Pza(za) and y~Py (y), (zaIt is generated by y, za'=fy(za) (f in formulay
It is one and z is mapped according to yaScreening function, z ' is a condition latent variable, is input into generator input terminal and influences defeated
Out))；ziIt is a unconditional latent variable (with the z effect having the same in CGAN).
Detailed process can be divided into two parts and describe: generate in network: by noise variance zi~Pzi(zi)、za~Pza
(za) and y~Py (y) be mapped to data space x=tt (zi,za, y) generator tt, x at this time is generated by generator.
Differentiate in network: when x is a true input picture, distributing p=D (x, y) ∈ [0,1] and p=D of a detectivity
(tt(zi,za, y), y) the arbiter D (p is determined as true probability) of ∈ [0,1] distributes a probability 1-p when x is generated by tt.
Pzi(zi)、Pza(za) and Py (y) is priori value on z, generally selects uniform [- 1,1] distribution.
Optionally, in step S4, using the property control function of condition filtering GAN conditional variable z ', convenient for strengthening simultaneously
Control critical eigenvalue region, and generate the label of corresponding geometric attribute.
Distinguishing feature of the condition filtering GAN compared with CGAN is:
1, filtering GAN first edits the attribute of given photo x, estimates latent variable z from xi.Then pass through training tool
There is the generator of antagonism arbiter, latent variable can be decomposed into conditional-variable za' and unconditional variable zi.And CGAN only has
For one supplementary variable y as input, the controlling for generating picture is not so good as condition filtering GAN.
2, condition filtering GAN can show the attribute of stronger control with z '.The property control function of conditional-variable z '
The feature exclusive as condition filtering GAN can preferably be applied to Bridge Crack by content proposed in this paper and detect.It can be with
Using the property control function of conditional-variable z ', strengthens simultaneously control critical eigenvalue region, generate the label of corresponding geometric attribute.
Finally, it is stated that the above examples are only used to illustrate the technical scheme of the present invention and are not limiting, although referring to compared with
Good embodiment describes the invention in detail, those skilled in the art should understand that, it can be to skill of the invention
Art scheme is modified or replaced equivalently, and without departing from the objective and range of the technical program, should all be covered in the present invention
Scope of the claims in.
Claims (4)
1. Bridge Crack detection and classification method based on condition filtering GAN network, it is characterised in that: the following steps are included:
S1: according to three classes bridge danger classes, respectively choosing appropriate picture, and add corresponding label, chooses figure according to the actual situation
As pixel, database is established, the three classes bridge danger classes includes linear crack, blocky crack and chicken-wire cracking；
S2: set up the condition filters GAN network；
S3: training arbiter D and generator G uses fixed party during training, updates the network weight of another party,
Alternating iteration；
S4: picture to be sorted is inputted to condition filtering GAN network, corresponding label is exported, obtains classification results.
2. the Bridge Crack detection and classification method, feature according to claim 1 based on condition filtering GAN network exists
In: in step S1, label uses the form of same pixel picture, and the label in linear crack is that a white background has a black line
Figure, the label in blocky crack be a white background have two intersection black line figure, the label of chicken-wire cracking is a Zhang Bai
There is the figure of the black line of four intersections and closed figure of encircling a city at bottom.
3. the Bridge Crack detection and classification method, feature according to claim 2 based on condition filtering GAN network exists
In: in step S2, the target formula of the training of condition filtering GAN network is as follows:
In above-mentioned formula, x is a true input picture；One supplementary variable y；Noise variance zi~Pzi(zi)、za~Pza
(za) and y~Py (y), zaIt is generated by y, za'=fy(za), fyIt is one and z is mapped according to yaScreening function, z ' is an item
Part latent variable is input into generator input terminal and influences to export；ziIt is a unconditional latent variable；
Detailed process is divided into two parts description: generating in network: by noise variance zi~Pzi(zi)、za~Pza(za) and y~Py
(y) it is mapped to data space x=tt (zi,za, y) generator tt, x at this time is generated by generator；Differentiate in network:
When x is a true input picture, p=D (x, y) ∈ [0,1] and p=D (tt (z of a detectivity are distributedi,za,
Y), y) ∈ [0,1] arbiter D, wherein p is determined as true probability, when x is generated by tt, distributes a probability 1-p；Pzi(zi)、
Pza(za) and Py (y) is priori value on z, selects uniform [- 1,1] distribution.
4. the Bridge Crack detection and classification method, feature according to claim 3 based on condition filtering GAN network exists
In: in step S4, using the property control function of condition filtering GAN conditional variable z ', convenient for strengthening simultaneously control critical eigenvalue region,
And generate the label of corresponding geometric attribute.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910430207.5A CN110147842A (en) | 2019-05-22 | 2019-05-22 | Bridge Crack detection and classification method based on condition filtering GAN |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910430207.5A CN110147842A (en) | 2019-05-22 | 2019-05-22 | Bridge Crack detection and classification method based on condition filtering GAN |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110147842A true CN110147842A (en) | 2019-08-20 |
Family
ID=67592716
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910430207.5A Pending CN110147842A (en) | 2019-05-22 | 2019-05-22 | Bridge Crack detection and classification method based on condition filtering GAN |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110147842A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111625697A (en) * | 2020-05-12 | 2020-09-04 | 深圳赋乐科技有限公司 | Network data identification filtering method for generating countermeasure network based on condition |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108680807A (en) * | 2018-05-17 | 2018-10-19 | 国网山东省电力公司青岛供电公司 | The Diagnosis Method of Transformer Faults and system of network are fought based on condition production |
| CN109166126A (en) * | 2018-08-13 | 2019-01-08 | 苏州比格威医疗科技有限公司 | A method of paint crackle is divided on ICGA image based on condition production confrontation network |
| US20190096056A1 (en) * | 2017-09-25 | 2019-03-28 | United Technologies Corporation | Automated material characterization system including conditional generative adversarial networks |
| CN109544555A (en) * | 2018-11-26 | 2019-03-29 | 陕西师范大学 | Fine cracks dividing method based on production confrontation network |
| CN109741328A (en) * | 2019-02-02 | 2019-05-10 | 东北大学 | A kind of automobile apparent mass detection method based on production confrontation network |
-
2019
- 2019-05-22 CN CN201910430207.5A patent/CN110147842A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190096056A1 (en) * | 2017-09-25 | 2019-03-28 | United Technologies Corporation | Automated material characterization system including conditional generative adversarial networks |
| CN108680807A (en) * | 2018-05-17 | 2018-10-19 | 国网山东省电力公司青岛供电公司 | The Diagnosis Method of Transformer Faults and system of network are fought based on condition production |
| CN109166126A (en) * | 2018-08-13 | 2019-01-08 | 苏州比格威医疗科技有限公司 | A method of paint crackle is divided on ICGA image based on condition production confrontation network |
| CN109544555A (en) * | 2018-11-26 | 2019-03-29 | 陕西师范大学 | Fine cracks dividing method based on production confrontation network |
| CN109741328A (en) * | 2019-02-02 | 2019-05-10 | 东北大学 | A kind of automobile apparent mass detection method based on production confrontation network |
Non-Patent Citations (2)
| Title |
|---|
| TAKUHIRO KANEKO等: "Generative Attribute Controller with Conditional Filtered Generative Adversarial Networks", 《2017 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)》 * |
| 唐贤伦等: "基于条件深度卷积生成对抗网络的图像识别方法", 《自动化学报》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111625697A (en) * | 2020-05-12 | 2020-09-04 | 深圳赋乐科技有限公司 | Network data identification filtering method for generating countermeasure network based on condition |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109829893A (en) | A kind of defect object detection method based on attention mechanism | |
| CN109359684A (en) | Fine granularity model recognizing method based on Weakly supervised positioning and subclass similarity measurement | |
| CN103383733B (en) | A kind of track based on half machine learning video detecting method | |
| CN106920243A (en) | The ceramic material part method for sequence image segmentation of improved full convolutional neural networks | |
| CN110097049A (en) | A kind of natural scene Method for text detection and system | |
| CN108550259A (en) | Congestion in road judgment method, terminal device and computer readable storage medium | |
| CN109299274A (en) | A kind of natural scene Method for text detection based on full convolutional neural networks | |
| CN107909010A (en) | A kind of road barricade object detecting method and device | |
| CN108052984A (en) | Method of counting and device | |
| CN109196514A (en) | Image classification and label | |
| CN109598279A (en) | Based on the zero sample learning method for generating network from coding confrontation | |
| CN109242829A (en) | Liquid crystal display defect inspection method, system and device based on small sample deep learning | |
| CN110110780A (en) | A kind of picture classification method based on confrontation neural network and magnanimity noise data | |
| CN101833664A (en) | Video image character detecting method based on sparse expression | |
| CN105678218B (en) | A kind of method of mobile object classification | |
| CN105536205A (en) | Upper limb training system based on monocular video human body action sensing | |
| CN104751175B (en) | SAR image multiclass mark scene classification method based on Incremental support vector machine | |
| CN106529415A (en) | Characteristic and model combined road detection method | |
| CN106611421A (en) | SAR image segmentation method based on feature learning and sketch line constraint | |
| CN105469111A (en) | Small sample set object classification method on basis of improved MFA and transfer learning | |
| CN107622274A (en) | Neural network training method, device and computer equipment for image procossing | |
| CN110288013A (en) | A kind of defective labels recognition methods based on block segmentation and the multiple twin convolutional neural networks of input | |
| CN109800785A (en) | One kind is based on the relevant data classification method of expression and device certainly | |
| CN108647702A (en) | A kind of extensive food materials image classification method based on transfer learning | |
| CN103903441A (en) | Road traffic state distinguishing method based on semi-supervised learning |
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: 20190820 |