CN110097543A - Surfaces of Hot Rolled Strip defect inspection method based on production confrontation network - Google Patents
Surfaces of Hot Rolled Strip defect inspection method based on production confrontation network Download PDFInfo
- Publication number
- CN110097543A CN110097543A CN201910338448.7A CN201910338448A CN110097543A CN 110097543 A CN110097543 A CN 110097543A CN 201910338448 A CN201910338448 A CN 201910338448A CN 110097543 A CN110097543 A CN 110097543A
- Authority
- CN
- China
- Prior art keywords
- generator
- image
- hot rolled
- branch
- arbiter
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- 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
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The present invention is a kind of Surfaces of Hot Rolled Strip defect inspection method based on production confrontation network, specific steps are as follows: (1) extract industry spot Surfaces of Hot Rolled Strip defect image, and carry out image preprocessing;(2) Maker model and arbiter model of building production confrontation network G AN.Specific practice is that conditional tag vector c is added in the input of generator, is used for output category image;Pixel loss L is introduced in generator trainingp, improve the quality for generating image;Setting differentiates branch and more classification branches in arbiter, realizes more classification features and improves nicety of grading;(3) it is optimized with production confrontation network parameter of the particle swarm algorithm PSO to construction;(4) image will be generated and true picture merges into Surfaces of Hot Rolled Strip defect sample collection.The present invention proposes that method is able to solve the problem of sample data deficiency, improves the speed and accuracy of defect image feature extraction, new effective ways are provided for Surfaces of Hot Rolled Strip defects detection.
Description
Technical field
The invention belongs to Computer Vision Detection Technique fields, and in particular to a kind of hot rolling based on production confrontation network
Steel strip surface defect detection method.
Technical background
Steel strip surface defect seriously affects the performances such as appearance, fatigue resistance, corrosion resistance and the wearability of steel product, shadow
The subsequent use for ringing steel product, causes immeasurable industrial loss.Therefore the defects detection of belt steel product is in industrial production
Extremely important step.Steel strip surface defect detection method is that machine detects from artificial development, improves speed and accuracy of identification.
The current most common steel defect detection method is that defect characteristic is extracted and processed using means of different, then uses classifier pair
Defect is classified.T Ojala in 2002 et al. describes a kind of classification method for being based on uniform local binary patterns (ULBP),
This method obtains preferable performance at several aspects such as spatial resolution, grey scale change, rotation.Santanu in 2013
Ghorai et al. proposes a kind of Automatic Visual Inspection system with wavelet transform (DWT) feature and support vector machines (SVM)
System.Northeastern University's Song Ke minister in 2013 etc. describe it is a kind of based on local binary patterns (LBP) being capable of effective antimierophonic band
Steel defect identification method, mean accuracy 98%, and actual production can be effectively applied to.2016 Nian Wanglei et al. are adopted
The feature vector that hot rolled steel plate surface image obtains higher-dimension is decomposed in Tetrole transformation, is sent into SVM after dimensionality reduction and is completed to hot-strip
The Classification and Identification of surface defect, this method classification 8 kinds of defects such as including transversal crack, defect classification accuracy is 97.38%.
It is a kind of production model that production, which fights network (Generative Adversarial Network, GAN),
Most basic application is exactly to model the distribution of truthful data and generate sample data, such as generate image and video etc..Inside GAN
Using dual training mechanism, there is outstanding learning ability, sample encountered in conventional machines study can be effectively treated not
The problem of foot, improves the speed and accuracy of feature extraction, new thinking is provided for Surfaces of Hot Rolled Strip defect recognition.
Summary of the invention
The purpose of the present invention is to propose to a kind of Surfaces of Hot Rolled Strip defect inspection methods based on production confrontation network, will
GAN is applied to Surfaces of Hot Rolled Strip defects detection and is able to solve the problems such as data sample is few, degree of fitting is low, improves accuracy of identification.
The technical solution adopted by the present invention is that:
A kind of Surfaces of Hot Rolled Strip defect inspection method based on production confrontation network, includes the following steps:
Step 1: in conjunction with the case where industrial site, industry spot Surfaces of Hot Rolled Strip defect map is extracted in certain steel mill
Picture, and pre-processed.
Step 2: the generation model and discrimination model of building production confrontation network G AN.
Following improve is made to the Maker model in production confrontation network.First, condition is added in generator input
Label vector c, the i.e. input of generator are made of random noise z and conditional tag c.Second, picture is introduced in generator training
Element loss Lp, improve the quality for generating image.Improve the network architecture of generator are as follows: input as by random vector and label
The vector that vector connects and composes successively passes through convolutional layer Conv, active coating Relu, residual error structure Residual Block and warp
Lamination Deconv.
Arbiter model in production confrontation network is made of differentiation branch and more classification branches.In sentencing for original GAN
A conditional tag c is introduced in other device as branch is differentiated, increases the arbiter that branch of classifying one collectively forms GAN more.Its
Middle to differentiate that branch is used to judge the image that input picture is true picture or is generated by generator, the effects for classifying branch have more
Two, first is that the classification of input picture is judged, second is that generating Classification Loss Lcls.Differentiate the network mould of branch and more classification branches
Type is similar, is made of convolutional layer Conv, active coating LeakyRelu and warp lamination Deconv.
Step 3: it is optimized with production confrontation network parameter of the particle swarm algorithm PSO to construction.
Production fights the training of network using alternating training method, and the training majorized function of GAN model is
Since the generator and arbiter structure of GAN model used are complex, propose a kind of based on particle swarm algorithm
(PSO) production fights network optimized approach.Specific optimization process is using the parameter trained in model as in PSO algorithm
One particle, then particle length n is exactly that trained number of parameters is participated in this network.Use loss function L as PSO algorithm
Fitness function obtains the locally optimal solution and globally optimal solution of particle.Again with the position and speed of iterative manner more new particle,
Obtained particle is exactly updated network weight, and with this, iteration continues, until adaptive value, i.e. training error, converge to threshold value
The minimum of range completes the generator of production confrontation network and the parameter optimization of arbiter.
Step 4: using the true Surfaces of Hot Rolled Strip defect image of acquisition and conditional tag c as the input of generator G,
Output image is mixed with true picture as Surfaces of Hot Rolled Strip defect sample collection.Take part of samples pictures as test specimens
This collection, remaining is as training sample set.
Step 5: the arbiter D after extracting training removes the last layer, and carries out structure fine tuning, to hot-strip defect
Sample image data is effectively identified and is classified.
Further, the pixel loss L introduced in generator trainingpThe clarity of generator output image can be increased,
The recognition accuracy of Surfaces of Hot Rolled Strip defect is improved with the high pixel image processing of generation.Wherein
Lp=| G (z)-x |
The differentiation branch of arbiter is fed back in full articulamentum differentiates loss Ladv, it is true for characterizing the input of arbiter
Image still generates image.And the full articulamentum output for branch of more classifying is N-dimensional vector, N is Surfaces of Hot Rolled Strip defective data
Collect classification number, the purpose of training is the Classification Loss L of optimization feedbackcls.On differentiating that branch and more classification branches are shared when training
One layer of weight differentiates the differentiation loss L of branchadvWith the Classification Loss L of more classification branch feedbacksclsCommon guidance generator is raw
At the image of high quality.
To sum up, the loss function of GAN model is
L=Ladv+Lcls+Lp
The advantages of the present invention over the prior art are that:
1, the present invention constructs and production is trained to fight network G AN, can generate hot-strip defect image, enrich instruction
The diversity and randomness for practicing sample set, have carried out effective enhancing to data sample, have solved Surfaces of Hot Rolled Strip defect sample
Data scarcity problem, avoids over-fitting.
2, the present invention improves the generator and arbiter of original GAN, and pixel loss is introduced in generator, is drawn
It leads and generates high-resolution defect image, replace low-quality image to carry out defect recognition with high quality graphic;It is same in arbiter
When be added differentiate branch and classification branch, judge that image is true and false and carry out effective Classification and Identification, can be improved Surfaces of Hot Rolled Strip
The average nicety of grading of defect, obtains preferable recognition effect.
3, the present invention optimizes the parameter of GAN model with particle swarm algorithm (PSO), and particle swarm algorithm has relatively strong
Global optimizing ability, avoid occur in traditional optimal way gradient decline the problems such as, make production fight network acceleration
Convergence improves training effect.
Detailed description of the invention
Fig. 1 is a kind of process of Surfaces of Hot Rolled Strip defect inspection method that network is fought based on production of the invention
Figure.
Fig. 2 is the structural schematic diagram that production of the invention fights network.
Fig. 3 is the training process schematic diagram that production of the invention fights network.
Specific embodiment
The present invention is further described with reference to the accompanying drawings and detailed description.
It is as follows for implementation process of the invention as shown in Figure 1:
Step 1 acquires hot-strip defect image in industry spot, and does preliminary image preprocessing.
Step 2, building production confrontation network model simultaneously optimize, will treated true steel strip surface defect image and
Conditional tag is input in model, observes the generation image of output.
Step 3 generates image and mixes with the Surfaces of Hot Rolled Strip defect image of acquisition, by image procossing as hot rolling
Steel strip surface defect sample set.
Step 4 extracts discrimination model as classifier with particle swarm algorithm to GAN training is improved, adjustment parameter, effectively
Identification hot-strip defect sample image data is simultaneously able to carry out classification.
It is illustrated in figure 2 the structural schematic diagram of production confrontation network of the invention.
Original GAN is made of generator G and arbiter D.Wherein, generator G is according to random noise generation and truthful data
Similar puppet data, and arbiter D then judges that input is the pseudo- data of truthful data or generation.Due to the structure of original GAN
Simply, control is difficult and training is too free, therefore introduces in generating model and discrimination model in the GAN model of this research
Conditional tag c, contains defect classification information in the label, and the data of generator is instructed to generate and the identification of arbiter point
Class.In arbiter model, arbiter is divided into two branches, respectively differentiation branch and more classification branches, realizes sentence respectively
Other data are true and false and identify the function of classification.
It is illustrated in figure 3 the training process schematic diagram of production confrontation network of the invention.
The training optimization process of original GAN is exactly the minimax gambling process of generator G Yu arbiter D.Therefore model
Optimization formula can be written as
Arbiter D is in order to better discriminate between the data that data are truthful datas or are generated by generator G, it is necessary to accurate
Both identification simultaneously breaks up the probability as far as possible two that exports after judgement, i.e., so that D (x) is big as far as possible and D (G (x)) is small as far as possible;And it generates
Device G then needs to be continuously improved generation data and arbiter cannot be distinguished in the similarity degree of truthful data.So two modules
Training process be exactly the process vied each other, confronted with each other, the performance of the two is improved in constantly iteration, is finally reached
Nash banlance (i.e. generation data identical as the differentiation probability of truthful data) into game theory, at this time arbiter D and generator G
It is reached optimal.
During the improvement GAN actual optimization of this research, the training of the GAN method trained using substep alternating, specifically
Training step are as follows:
(1) training generator network.By in the real defect data x mixing condition label c input generator of acquisition, generate
High quality graphic.Calculate pixel loss Lp, pixel loss is bigger, illustrates that difference is bigger between high quality graphic and truthful data.
The process of training generator is exactly to minimize pixel loss L with particle swarm algorithmp;
(2) parameter of fixed generator, with the differentiation branch for the high quality graphic sample training arbiter that generator generates
With classification branch.
(3) the differentiation branch parameters of fixed arbiter, truthful data x are input in generator, outputting high quality image,
The loss L of computational discrimination branchadv, feed back to generator and optimize the parameter of generator.
(4) the classification branch parameters of fixed arbiter, truthful data x are input in generator, outputting high quality image,
Calculate the loss L of classification branchcls, feed back to generator and optimize the parameter of generator.
Repetition above-mentioned (2)~(4) step, repetitive exercise 10000 times.
Specific embodiments of the present invention are described above.It is to be appreciated that the invention is not limited to above-mentioned
Particular implementation, those skilled in the art can make a variety of changes or modify within the scope of the claims, this not shadow
Ring substantive content of the invention.In the absence of conflict, the feature in embodiments herein and embodiment can any phase
Mutually combination.
Claims (2)
1. a kind of Surfaces of Hot Rolled Strip defect inspection method based on production confrontation network, which is characterized in that including following step
It is rapid:
Step 1: extracting industry spot Surfaces of Hot Rolled Strip defect image in conjunction with the case where industrial site in steel mill, going forward side by side
Row pretreatment;
Step 2: the Maker model and arbiter model of building production confrontation network G AN;
Following improve is made to the Maker model in production confrontation network;
(1), conditional tag vector c is added in generator input, i.e., the input of generator is by random noise z and conditional tag c
Composition;
(2), pixel loss L is introduced in generator trainingp, improve the quality for generating image;Improve the network model knot of generator
Structure are as follows: input the vector to be connected and composed by random vector and label vector, successively by convolutional layer Conv, active coating Relu,
Residual error structure Residual Block and warp lamination Deconv;
Following improve is made to the arbiter model in production confrontation network:
A conditional tag c is introduced in the arbiter of original GAN as branch is differentiated, increases the common structure of branch of classifying one more
At the arbiter of GAN;Wherein differentiate branch be used to judge input picture be true picture or by generator generate image, it is more
There are two the effects of classification branch, first is that the classification of input picture is judged, second is that generating Classification Loss Lcls;Differentiate branch and more
The network model of classification branch is similar, is made of convolutional layer Conv, active coating LeakyRelu and warp lamination Deconv;
Step 3: being optimized with production confrontation network parameter of the particle swarm algorithm PSO to construction;
Production fights the training of network using alternating training method, and the training majorized function of GAN model is
Since the generator and arbiter structure of GAN model used are complex, a kind of life based on particle swarm algorithm PSO is proposed
An accepted way of doing sth fights network optimized approach;Specific optimization process is the parameter that will train in model as a particle in PSO algorithm,
Then particle length n is exactly that trained number of parameters is participated in this network;Use loss function L as the fitness letter of PSO algorithm
Number, obtains the locally optimal solution and globally optimal solution of particle;Again with the position and speed of iterative manner more new particle, obtained grain
Son is exactly updated network weight, and with this, iteration continues, until adaptive value, i.e. training error, converge to the pole of threshold range
Small value, determines according to data sample, completes the generator of production confrontation network and the parameter optimization of arbiter;
Step 4: using the true Surfaces of Hot Rolled Strip defect image of acquisition and conditional tag c as the input of generator G, output
Image is mixed with true picture as Surfaces of Hot Rolled Strip defect sample collection;Take part of samples pictures as test sample
Collection, remaining is as training sample set;
Step 5: extracting the arbiter D after training, the last layer is removed, and carry out structure fine tuning, to hot-strip defect sample
Image data is effectively identified and is classified.
2. a kind of Surfaces of Hot Rolled Strip defect inspection method based on production confrontation network according to claim 1,
It is characterized in that, the pixel loss L introduced in generator trainingpThe clarity that generator output image can be increased, with generation
The recognition accuracy of high pixel image processing raising Surfaces of Hot Rolled Strip defect;Wherein
Lp=| G (z)-x |
The differentiation branch of arbiter is fed back in full articulamentum differentiates loss Ladv, it is true picture for characterizing the input of arbiter
Or generate image;And the full articulamentum output for branch of more classifying is N-dimensional vector, N is Surfaces of Hot Rolled Strip defective data collection class
Shuo not;Its purpose trained is the Classification Loss L of optimization feedbackcls;Differentiate that branch and more classification branches share one layer when training
Weight, differentiate branch differentiation loss LadvWith the Classification Loss L of more classification branch feedbacksclsCommon guidance generator generates high
The image of quality;
To sum up, the loss function of GAN model are as follows:
L=Ladv+Lcls+Lp。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910338448.7A CN110097543B (en) | 2019-04-25 | 2019-04-25 | Hot-rolled strip steel surface defect detection method based on generation type countermeasure network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910338448.7A CN110097543B (en) | 2019-04-25 | 2019-04-25 | Hot-rolled strip steel surface defect detection method based on generation type countermeasure network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110097543A true CN110097543A (en) | 2019-08-06 |
CN110097543B CN110097543B (en) | 2023-01-13 |
Family
ID=67445823
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910338448.7A Active CN110097543B (en) | 2019-04-25 | 2019-04-25 | Hot-rolled strip steel surface defect detection method based on generation type countermeasure network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110097543B (en) |
Cited By (28)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110706308A (en) * | 2019-09-07 | 2020-01-17 | 创新奇智(成都)科技有限公司 | GAN-based steel coil end face edge loss artificial sample generation method |
CN110796174A (en) * | 2019-09-29 | 2020-02-14 | 郑州金惠计算机系统工程有限公司 | Multi-type virtual sample generation method and device, electronic equipment and storage medium |
CN111062918A (en) * | 2019-12-10 | 2020-04-24 | 歌尔股份有限公司 | Anomaly detection method and device based on computer vision |
CN111126566A (en) * | 2019-11-29 | 2020-05-08 | 江苏艾佳家居用品有限公司 | Abnormal furniture layout data detection method based on GAN model |
CN111126446A (en) * | 2019-11-29 | 2020-05-08 | 西安工程大学 | Method for amplifying defect image data of robot vision industrial product |
CN111126433A (en) * | 2019-11-19 | 2020-05-08 | 佛山市南海区广工大数控装备协同创新研究院 | Positive and negative sample data balancing method in factory PCB defect detection |
CN111223093A (en) * | 2020-03-04 | 2020-06-02 | 武汉精立电子技术有限公司 | AOI defect detection method |
CN111652864A (en) * | 2020-05-29 | 2020-09-11 | 沈阳铸造研究所有限公司 | Casting defect image generation method for generating countermeasure network based on conditional expression |
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 |
CN112036626A (en) * | 2020-08-21 | 2020-12-04 | 燕山大学 | Online forecasting method for linear defects at edge of hot-rolled strip steel |
CN112085734A (en) * | 2020-09-25 | 2020-12-15 | 西安交通大学 | GAN-based image restoration defect detection method |
CN112132181A (en) * | 2020-08-20 | 2020-12-25 | 黑龙江大学 | Image authenticity identification method based on generation type countermeasure network |
CN112215055A (en) * | 2020-08-15 | 2021-01-12 | 天津大学 | Method for detecting conductor defects in power transmission line based on single classification method |
CN112418310A (en) * | 2020-11-20 | 2021-02-26 | 第四范式(北京)技术有限公司 | Text style migration model training method and system and image generation method and system |
CN112488984A (en) * | 2019-09-11 | 2021-03-12 | 中信戴卡股份有限公司 | Method and device for acquiring defect picture generation network and defect picture generation method |
WO2021082467A1 (en) * | 2019-10-28 | 2021-05-06 | 南京智欧智能技术研究院有限公司 | Method and device for detecting surface defects of hot rolled steel material |
CN112990418A (en) * | 2019-12-02 | 2021-06-18 | 国网能源研究院有限公司 | Particle swarm-linear programming algorithm-based multi-energy cross-region optimization configuration method |
CN113298190A (en) * | 2021-07-05 | 2021-08-24 | 四川大学 | Weld image recognition and classification algorithm based on large-size unbalanced samples |
CN113450344A (en) * | 2021-07-19 | 2021-09-28 | 江南大学 | Strip steel surface defect detection method and system |
CN113469913A (en) * | 2021-07-06 | 2021-10-01 | 中南大学 | Hot-rolled strip steel surface water drop removing method based on gradual cycle generation countermeasure network |
CN113506239A (en) * | 2021-05-21 | 2021-10-15 | 冶金自动化研究设计院 | Strip steel surface defect detection method based on cross-stage local network |
WO2021217852A1 (en) * | 2020-04-30 | 2021-11-04 | 平安科技(深圳)有限公司 | Damage detection method and apparatus, and electronic device and medium |
US20210358095A1 (en) * | 2020-02-05 | 2021-11-18 | Google Llc | Image Transformation Using Interpretable Transformation Parameters |
CN113706477A (en) * | 2021-08-10 | 2021-11-26 | 南京旭锐软件科技有限公司 | Defect type identification method, device, equipment and medium |
CN113838208A (en) * | 2021-09-02 | 2021-12-24 | 桂林电子科技大学 | Strip steel defect detection method based on improved residual shrinkage network |
CN114511570A (en) * | 2022-04-20 | 2022-05-17 | 深圳市信润富联数字科技有限公司 | Pseudo defect library establishing method, defect detecting method, device and storage medium |
CN115943429A (en) * | 2020-08-03 | 2023-04-07 | 科磊股份有限公司 | Depth-generating models for optical or other mode selection |
CN117372438A (en) * | 2023-12-08 | 2024-01-09 | 湘江实验室 | Hot rolled steel plate defect detection method based on class unbalance and related equipment |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018028255A1 (en) * | 2016-08-11 | 2018-02-15 | 深圳市未来媒体技术研究院 | Image saliency detection method based on adversarial network |
CN108388925A (en) * | 2018-03-06 | 2018-08-10 | 天津工业大学 | The anti-pattern collapse robust image generation method for generating network is fought based on New Conditions |
WO2019015466A1 (en) * | 2017-07-17 | 2019-01-24 | 广州广电运通金融电子股份有限公司 | Method and apparatus for verifying person and certificate |
CN109544555A (en) * | 2018-11-26 | 2019-03-29 | 陕西师范大学 | Fine cracks dividing method based on production confrontation network |
-
2019
- 2019-04-25 CN CN201910338448.7A patent/CN110097543B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018028255A1 (en) * | 2016-08-11 | 2018-02-15 | 深圳市未来媒体技术研究院 | Image saliency detection method based on adversarial network |
WO2019015466A1 (en) * | 2017-07-17 | 2019-01-24 | 广州广电运通金融电子股份有限公司 | Method and apparatus for verifying person and certificate |
CN108388925A (en) * | 2018-03-06 | 2018-08-10 | 天津工业大学 | The anti-pattern collapse robust image generation method for generating network is fought based on New Conditions |
CN109544555A (en) * | 2018-11-26 | 2019-03-29 | 陕西师范大学 | Fine cracks dividing method based on production confrontation network |
Cited By (41)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110706308A (en) * | 2019-09-07 | 2020-01-17 | 创新奇智(成都)科技有限公司 | GAN-based steel coil end face edge loss artificial sample generation method |
CN112488984A (en) * | 2019-09-11 | 2021-03-12 | 中信戴卡股份有限公司 | Method and device for acquiring defect picture generation network and defect picture generation method |
CN110796174A (en) * | 2019-09-29 | 2020-02-14 | 郑州金惠计算机系统工程有限公司 | Multi-type virtual sample generation method and device, electronic equipment and storage medium |
WO2021082467A1 (en) * | 2019-10-28 | 2021-05-06 | 南京智欧智能技术研究院有限公司 | Method and device for detecting surface defects of hot rolled steel material |
CN111126433A (en) * | 2019-11-19 | 2020-05-08 | 佛山市南海区广工大数控装备协同创新研究院 | Positive and negative sample data balancing method in factory PCB defect detection |
CN111126566B (en) * | 2019-11-29 | 2022-07-08 | 江苏艾佳家居用品有限公司 | Abnormal furniture layout data detection method based on GAN model |
CN111126566A (en) * | 2019-11-29 | 2020-05-08 | 江苏艾佳家居用品有限公司 | Abnormal furniture layout data detection method based on GAN model |
CN111126446B (en) * | 2019-11-29 | 2023-04-07 | 西安工程大学 | Method for amplifying defect image data of robot vision industrial product |
CN111126446A (en) * | 2019-11-29 | 2020-05-08 | 西安工程大学 | Method for amplifying defect image data of robot vision industrial product |
CN112990418B (en) * | 2019-12-02 | 2024-05-07 | 国网能源研究院有限公司 | Multi-energy cross-region optimal configuration method based on particle swarm-linear programming algorithm |
CN112990418A (en) * | 2019-12-02 | 2021-06-18 | 国网能源研究院有限公司 | Particle swarm-linear programming algorithm-based multi-energy cross-region optimization configuration method |
CN111062918A (en) * | 2019-12-10 | 2020-04-24 | 歌尔股份有限公司 | Anomaly detection method and device based on computer vision |
CN111062918B (en) * | 2019-12-10 | 2023-11-21 | 歌尔股份有限公司 | Abnormality detection method and device based on computer vision |
US12002195B2 (en) | 2019-12-10 | 2024-06-04 | Goertek Inc. | Computer vision-based anomaly detection method, device and electronic apparatus |
US20210358095A1 (en) * | 2020-02-05 | 2021-11-18 | Google Llc | Image Transformation Using Interpretable Transformation Parameters |
US11599980B2 (en) * | 2020-02-05 | 2023-03-07 | Google Llc | Image transformation using interpretable transformation parameters |
US11908115B2 (en) | 2020-02-05 | 2024-02-20 | Google Llc | Image transformation using interpretable transformation parameters |
CN111223093A (en) * | 2020-03-04 | 2020-06-02 | 武汉精立电子技术有限公司 | AOI defect detection method |
WO2021217852A1 (en) * | 2020-04-30 | 2021-11-04 | 平安科技(深圳)有限公司 | Damage detection method and apparatus, and electronic device and medium |
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 |
CN111652864A (en) * | 2020-05-29 | 2020-09-11 | 沈阳铸造研究所有限公司 | Casting defect image generation method for generating countermeasure network based on conditional expression |
CN115943429B (en) * | 2020-08-03 | 2024-04-05 | 科磊股份有限公司 | Depth generation model for optical or other mode selection |
CN115943429A (en) * | 2020-08-03 | 2023-04-07 | 科磊股份有限公司 | Depth-generating models for optical or other mode selection |
CN112215055A (en) * | 2020-08-15 | 2021-01-12 | 天津大学 | Method for detecting conductor defects in power transmission line based on single classification method |
CN112132181A (en) * | 2020-08-20 | 2020-12-25 | 黑龙江大学 | Image authenticity identification method based on generation type countermeasure network |
CN112036626B (en) * | 2020-08-21 | 2023-08-22 | 燕山大学 | Online forecasting method for edge linear defects of hot rolled strip steel |
CN112036626A (en) * | 2020-08-21 | 2020-12-04 | 燕山大学 | Online forecasting method for linear defects at edge of hot-rolled strip steel |
CN112085734A (en) * | 2020-09-25 | 2020-12-15 | 西安交通大学 | GAN-based image restoration defect detection method |
CN112418310A (en) * | 2020-11-20 | 2021-02-26 | 第四范式(北京)技术有限公司 | Text style migration model training method and system and image generation method and system |
CN113506239B (en) * | 2021-05-21 | 2024-06-04 | 冶金自动化研究设计院 | Strip steel surface defect detection method based on cross-stage local network |
CN113506239A (en) * | 2021-05-21 | 2021-10-15 | 冶金自动化研究设计院 | Strip steel surface defect detection method based on cross-stage local network |
CN113298190A (en) * | 2021-07-05 | 2021-08-24 | 四川大学 | Weld image recognition and classification algorithm based on large-size unbalanced samples |
CN113469913A (en) * | 2021-07-06 | 2021-10-01 | 中南大学 | Hot-rolled strip steel surface water drop removing method based on gradual cycle generation countermeasure network |
CN113469913B (en) * | 2021-07-06 | 2022-06-03 | 中南大学 | Hot-rolled strip steel surface water drop removing method based on gradual cycle generation countermeasure network |
CN113450344A (en) * | 2021-07-19 | 2021-09-28 | 江南大学 | Strip steel surface defect detection method and system |
CN113706477A (en) * | 2021-08-10 | 2021-11-26 | 南京旭锐软件科技有限公司 | Defect type identification method, device, equipment and medium |
CN113706477B (en) * | 2021-08-10 | 2024-02-13 | 南京旭锐软件科技有限公司 | Defect category identification method, device, equipment and medium |
CN113838208B (en) * | 2021-09-02 | 2024-02-02 | 桂林电子科技大学 | Strip steel defect detection method based on improved residual shrinkage network |
CN113838208A (en) * | 2021-09-02 | 2021-12-24 | 桂林电子科技大学 | Strip steel defect detection method based on improved residual shrinkage network |
CN114511570A (en) * | 2022-04-20 | 2022-05-17 | 深圳市信润富联数字科技有限公司 | Pseudo defect library establishing method, defect detecting method, device and storage medium |
CN117372438A (en) * | 2023-12-08 | 2024-01-09 | 湘江实验室 | Hot rolled steel plate defect detection method based on class unbalance and related equipment |
Also Published As
Publication number | Publication date |
---|---|
CN110097543B (en) | 2023-01-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110097543A (en) | Surfaces of Hot Rolled Strip defect inspection method based on production confrontation network | |
CN107610087B (en) | Tongue coating automatic segmentation method based on deep learning | |
Chen et al. | A new ensemble approach based on deep convolutional neural networks for steel surface defect classification | |
CN108564115A (en) | Semi-supervised polarization SAR terrain classification method based on full convolution GAN | |
Abu et al. | The performance analysis of transfer learning for steel defect detection by using deep learning | |
CN110188824A (en) | A kind of small sample plant disease recognition methods and system | |
CN108985170A (en) | Transmission line of electricity hanger recognition methods based on Three image difference and deep learning | |
Liu et al. | Real-time classification of steel strip surface defects based on deep CNNs | |
Zhan et al. | Semi-supervised classification of hyperspectral data based on generative adversarial networks and neighborhood majority voting | |
CN108932712A (en) | A kind of rotor windings quality detecting system and method | |
CN109934204A (en) | A kind of facial expression recognizing method based on convolutional neural networks | |
Aslam et al. | Ensemble convolutional neural networks with knowledge transfer for leather defect classification in industrial settings | |
CN114022469A (en) | Cigarette appearance defect image classification method and system | |
CN112749675A (en) | Potato disease identification method based on convolutional neural network | |
CN113505810A (en) | Pooling vision-based method for detecting weed growth cycle by using Transformer | |
Chen et al. | X-ray of tire defects detection via modified faster R-CNN | |
Wan et al. | Side-path FPN-based multi-scale object detection | |
Li et al. | Target detection in color sonar image based on YOLOV5 network | |
CN116468083A (en) | Transformer-based network generation countermeasure method | |
CN116129215A (en) | Long-tail target detection method based on deep learning | |
CN115810109A (en) | Tongue image feature extraction method, system and storage medium | |
Huo et al. | Traffic sign recognition based on resnet-20 and deep mutual learning | |
CN111046861B (en) | Method for identifying infrared image, method for constructing identification model and application | |
Ahmed et al. | Fresh or Stale: Leveraging Deep Learning to Detect Freshness of Fruits and Vegetables | |
Wang et al. | Fusion of infrared and visible images based on improved generative adversarial networks |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |