CN112488130A - AI micro-pore wall detection algorithm - Google Patents

AI micro-pore wall detection algorithm Download PDF

Info

Publication number
CN112488130A
CN112488130A CN202011496848.XA CN202011496848A CN112488130A CN 112488130 A CN112488130 A CN 112488130A CN 202011496848 A CN202011496848 A CN 202011496848A CN 112488130 A CN112488130 A CN 112488130A
Authority
CN
China
Prior art keywords
image
network
function
micro
generated
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
Application number
CN202011496848.XA
Other languages
Chinese (zh)
Other versions
CN112488130B (en
Inventor
刘金建
随岁寒
黄伟国
张程
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Suzhou Juyue Information Technology Co ltd
Original Assignee
Suzhou Juyue Information Technology Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Suzhou Juyue Information Technology Co ltd filed Critical Suzhou Juyue Information Technology Co ltd
Priority to CN202011496848.XA priority Critical patent/CN112488130B/en
Publication of CN112488130A publication Critical patent/CN112488130A/en
Application granted granted Critical
Publication of CN112488130B publication Critical patent/CN112488130B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • 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
    • 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
    • G06N3/048—Activation functions
    • 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
    • G06N3/08—Learning methods
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Mathematical Physics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Multimedia (AREA)
  • Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
  • Image Processing (AREA)

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a micro-pore wall detection algorithm of AI. It includes S1: generating a generating network of a micro hole wall defect position image by using the difference of the pixel value of the defect position and the normal sample; s2: adding the generated image with the defects and the normal image to obtain a defective image generated by using the generation network; s3: designing a countermeasure network and sending the image obtained in the step S2 and the real image into the countermeasure network; s4: and calculating a defect value of the micro hole wall according to the loss function, and then transmitting the defect value back to the generation network and the countermeasure network. In the invention, by means of adding the image generated by the generated network and the normal sample, the difficulty of generating the network is reduced, the training difficulty is reduced, the training time is shortened, the possibility of gradient loss and non-convergence is reduced, and in addition, more samples close to actual defects can be generated by adopting the mode of generating the confrontation network.

Description

AI micro-pore wall detection algorithm
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a micro-pore wall detection algorithm of AI.
Background
At present, the defect detection method mostly adopts the traditional machine learning method, but only can solve the defect of a single mode, and the artificial intelligence method can realize the detection of the defect of multiple modes;
however, methods such as color transformation and random cutting are mainly adopted for data augmentation, the method generates a sample with a single defect condition and cannot generate a defect sample picture under a complex condition, and in addition, the conventional method for generating an image based on an antagonistic network generates images by generating a network with a defect, so that the generated content is large, the network is not easy to train, and the conditions of gradient loss and non-convergence are easy to occur.
Disclosure of Invention
The invention aims to provide a micro-hole wall detection algorithm for AI, so as to solve the problems in the background technology.
In order to achieve the above object, the present invention provides an AI micro-hole wall detection algorithm, which comprises the following steps:
s1: generating a generating network of a micro hole wall defect position image by using the difference of the pixel value of the defect position and the normal sample;
s2: adding the generated image with the defects and the normal image to obtain a defective image generated by using the generation network;
s3: designing a countermeasure network, and sending the image and the real image obtained in the step S2 into the countermeasure network for judging whether the generated image is true or false until the generated image is judged to be true;
s4: and calculating a defect value of the micro hole wall according to the loss function, and then transmitting the defect value back to the generation network and the countermeasure network.
As a further improvement of the present technical solution, the generation network in S1 adopts a full convolution network to generate a defect image with a size of 512 × 512, and the specific method thereof is as follows:
s1.1, decomposing a defect position image of a micro hole wall into R, G, B three channels, wherein R, G, B is designed according to the principle of color luminescence, the color mixing mode comprises three colors of red, green and blue, the three colors are respectively expressed as R, G, B, when the color values of the three colors are superposed, the colors are mixed, the brightness is equal to the sum of the brightness of the three colors, the mixed brightness is higher, the addition mixing is carried out, and each value is between 0 and 255;
s1.2, calculating the decomposed image;
s1.3, activating the full convolution network by using an activation function;
s1.4, extracting the features of the image in a pooling mode to reduce the data volume transmitted to the next stage;
and S1.5, classifying the extracted features by using a full-connection mode.
As a further improvement of the technical solution, the operation steps in S1.2 are as follows:
s2.1, expanding the scale of the image by deconvolution with the step length of 2 and the convolution kernel size of 5, and thus up-sampling the features with the original dimension of 512 to a defective pixel picture of 512x 512;
s2.2, calculating an input matrix of the image, wherein the calculation formula is as follows:
Figure BDA0002842427960000021
wherein, REC is an input matrix; k is a convolution kernel; r is an R channel; g is a G channel; and B is a B channel.
As a further improvement of the technical solution, the activation function of the last layer in S1.3 adopts a Sigmoid function, and the function formula thereof is as follows:
Figure BDA0002842427960000022
the Sigmoid function is unilateral inhibition, has a relatively wide excitation boundary and has the characteristic of sparse activation.
As a further improvement of the technical solution, the feature classification in S1.5 adopts a Softmax function, and the classification method thereof is as follows:
s3.1, extracting the characteristics of the m images of the positions of the defects of the micro hole wall to form a training set { (x)1，y1)，(xm，ym)}；
S3.2, training a model parameter theta, wherein the cost function is as follows:
Figure BDA0002842427960000023
as a further improvement of the present technical solution, the network generated in S2 employs a Relu function, and the Relu function makes outputs of a part of neurons be 0, thereby causing sparsity of the network, reducing interdependencies of parameters, and alleviating the occurrence of the overfitting problem of the Softmax function, and the functional formula is as follows:
Figure BDA0002842427960000024
wherein L is the step length.
As a further improvement of the present invention, in S3, the method for determining whether the generated image is true or false is as follows:
s4.1, generating cloth image data by using a generator and a real defect-free sample;
s4.2, sending the data generated in the S4.1 and the real data into a discriminator, calculating loss, fixing generator parameters at the moment, and updating parameters of the discriminator according to the loss;
s4.3, after the training of the discriminator is finished, fixing the parameters of the discriminator, and setting the label of the image generated by the generator as a real image, thereby calculating the loss and transmitting the loss back to the generator, so that the generator can generate a better defect picture;
and S4.4, repeating the S4.2 and the S4.3 until the picture generated by the generator is consistent with the real defect picture.
As a further improvement of the present technical solution, the data generating method of the generator in S4.1 is as follows:
s5.1, the generator is a function for generating one value at a time;
and S5.2, calling the function to return a Generator which can be used for generating continuous values, wherein in the execution process of the function, the yield statement returns the required value to the place where the Generator is called, then the function is quitted, when the Generator function is called next time, the execution is started from the place where the Generator is interrupted last time, and all variable parameters in the Generator are saved for the next time.
As a further improvement of the present technical solution, the loss function in S4 is a penalty loss function.
As a further improvement of the present technical solution, the functional expression of the countermeasure loss function is as follows:
Figure BDA0002842427960000031
wherein R is an R channel; g is a G channel; and B is a B channel.
Compared with the prior art, the invention has the beneficial effects that: in the AI micro-pore wall detection algorithm, the difficulty of generating the network is reduced, the training difficulty is reduced, the training time is shortened, the probability of gradient disappearance and non-convergence is reduced by adding the image generated by the generated network and a normal sample, and in addition, more samples close to actual defects can be generated by adopting a mode of generating a confrontation network.
Drawings
FIG. 1 is an overall algorithm flow diagram of example 1;
FIG. 2 is a flowchart of an algorithm for generating a defect image in the full convolution network of example 1;
FIG. 3 is a flowchart of an image calculation algorithm of embodiment 1;
FIG. 4 is a flowchart of the algorithm of the Softmax function of example 1;
FIG. 5 is a flowchart of an algorithm for generating image data of a minute hole wall in example 1;
FIG. 6 is a flowchart of an algorithm for data generation of the generator of embodiment 1;
FIG. 7 is a flowchart of the convolution method of the full convolution net image according to example 1;
fig. 8 is a Relu function image of example 1.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the 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.
Example 1
The invention provides a micro-hole wall detection algorithm of AI, please refer to FIGS. 1-8, comprising the following steps:
s1: generating a generating network of a micro hole wall defect position image by using the difference of the pixel value of the defect position and the normal sample;
s2: adding the generated image with the defects and the normal image to obtain the image with the defects generated by using the generated network, thereby reducing the difficulty of generating the network, reducing the training difficulty and shortening the training time;
s3: designing a countermeasure network, sending the image and the real image obtained in the step S2 into the countermeasure network, and generating more samples close to actual defects by using a mode of generating the countermeasure network for judging whether the generated image is true or false until the generated image is judged to be true;
s4: and calculating a defect value of the micro hole wall according to the loss function, and then transmitting the defect value back to the generation network and the countermeasure network.
In this embodiment, the generation network in S1 adopts a full convolution network to generate a defect image with a size of 512 × 512, and the specific method is as follows:
s1.1, decomposing a defect position image of a micro hole wall into R, G, B three channels, wherein R, G, B is designed according to the principle of color luminescence, the color mixing mode comprises three colors of red, green and blue, the three colors are respectively expressed as R, G, B, when the color values of the three colors are superposed, the colors are mixed, the brightness is equal to the sum of the brightness of the three colors, the mixed brightness is higher, the addition mixing is carried out, and each value is between 0 and 255;
s1.2, calculating the decomposed image;
s1.3, activating the full convolution network by using an activation function;
s1.4, extracting the features of the image in a pooling mode to reduce the data volume transmitted to the next stage;
and S1.5, classifying the extracted features by using a full-connection mode.
Further, in S1.2, the operation steps are as follows:
s2.1, expanding the scale of the image by deconvolution with the step length of 2 and the convolution kernel size of 5, and thus up-sampling the features with the original dimension of 512 to a defective pixel picture of 512x 512;
s2.2, calculating an input matrix of the image, wherein the calculation formula is as follows:
Figure BDA0002842427960000051
wherein, REC is an input matrix; k is a convolution kernel; r is an R channel; g is a G channel; and B is a B channel.
Specifically, the activation function of the last layer in S1.3 is a Sigmoid function, and the function formula is as follows:
Figure BDA0002842427960000052
the Sigmoid function is unilateral inhibition, has a relatively wide excitation boundary and has the characteristic of sparse activation.
In addition, the characteristic classification in S1.5 adopts a Softmax function, and the classification method is as follows:
s3.1, extracting the characteristics of the m images of the positions of the defects of the micro hole wall to form a training set { (x)1，y1)，(xm，ym)}；
S3.2, training a model parameter theta, wherein the cost function is as follows:
Figure BDA0002842427960000053
for a given feature input x of the image of the location of the micro-hole wall defect, then the cost function estimates a probability value p (y is J | x) for each class J, so we assume that the function is to output a k-dimensional vector to represent k estimated probability values, specifically, assume that the function h is to output a k-dimensional vectorθ(x) The form is as follows:
Figure BDA0002842427960000061
where θ is a parameter of the model, and further
Figure BDA0002842427960000062
This term normalizes the probability distribution so that the sum of all probabilities is 1.
In addition, the network generated in S2 employs a Relu function, which makes the output of some neurons 0, thereby resulting in sparsity of the network, and reduces interdependency of parameters, and alleviates the problem of overfitting the Softmax function, and the function formula is as follows:
Figure BDA0002842427960000063
wherein L is the step length.
Further, the method for judging whether the generated image is true or false in S3 is as follows:
s4.1, generating cloth image data by using a generator and a real defect-free sample;
s4.2, sending the data generated in the S4.1 and the real data to a discriminator, calculating loss, fixing generator parameters at the moment, and updating parameters of the discriminator according to the loss;
s4.3, after the training of the discriminator is finished, fixing the parameters of the discriminator, and setting the label of the image generated by the generator as a real image, thereby calculating the loss and transmitting the loss back to the generator, so that the generator can generate a better defect picture;
and S4.4, repeating S4.2 and S4.3 until the picture generated by the generator is consistent with the real defect picture.
Specifically, the data generation method of the generator in S4.1 is as follows:
s5.1, the generator is a function for generating one value at a time;
and S5.2, calling the function to return a Generator which can be used for generating continuous values, wherein in the execution process of the function, the yield statement returns the required value to the place where the Generator is called, then the function is quitted, when the Generator function is called next time, the execution is started from the place where the Generator is interrupted last time, and all variable parameters in the Generator are saved for the next time.
Further, the loss function in S4 is a penalty loss function.
In addition, the functional expression of the penalty function is as follows:
Figure BDA0002842427960000071
wherein R is an R channel; g is a G channel; and B is a B channel.
The foregoing shows and describes the general principles, essential features, and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, and the preferred embodiments of the present invention are described in the above embodiments and the description, and are not intended to limit the present invention. The scope of the invention is defined by the appended claims and equivalents thereof.

Claims (10)

1. An AI micro-pore wall detection algorithm, comprising the steps of:
s1: generating a generating network of a micro hole wall defect position image by using the difference of the pixel value of the defect position and the normal sample;
s2: adding the generated image with the defects and the normal image to obtain a defective image generated by using the generation network;
s3: designing a countermeasure network, and sending the image and the real image obtained in the step S2 into the countermeasure network for judging whether the generated image is true or false until the generated image is judged to be true;
s4: and calculating a defect value of the micro hole wall according to the loss function, and then transmitting the defect value back to the generation network and the countermeasure network.
2. The AI micro-aperture wall detection algorithm of claim 1, wherein: the generation network in S1 adopts a full convolution network to generate a defect image with a size of 512 × 512, and the specific method is as follows:
s1.1, decomposing a defect position image of the micro hole wall into R, G, B channels;
s1.2, calculating the decomposed image;
s1.3, activating the full convolution network by using an activation function;
s1.4, extracting the features of the image in a pooling mode;
and S1.5, classifying the extracted features by using a full-connection mode.
3. The AI micro-aperture wall detection algorithm of claim 2, wherein: the operation steps in S1.2 are as follows:
s2.1, enlarging the scale of the image by using deconvolution with the step length of 2 and the convolution kernel size of 5;
s2.2, calculating an input matrix of the image, wherein the calculation formula is as follows:
Figure FDA0002842427950000011
wherein, REC is an input matrix; k is a convolution kernel; r is an R channel; g is a G channel; and B is a B channel.
4. The AI micro-aperture wall detection algorithm of claim 2, wherein: the activation function of the last layer in the S1.3 adopts a Sigmoid function, and the function formula is as follows:
Figure FDA0002842427950000012
wherein the Sigmoid function is one-sided suppression.
5. The AI micro-aperture wall detection algorithm of claim 2, wherein: the characteristic classification in the S1.5 adopts a Softmax function, and the classification method is as follows:
s3.1, extracting the characteristics of the m images of the positions of the defects of the micro hole wall to form a training set { (x)1，y1)，(xm，ym)}；
S3.2, training a model parameter theta, wherein the cost function is as follows:
Figure FDA0002842427950000021
6. the AI micro-aperture wall detection algorithm of claim 1, wherein: the network generated in S2 adopts a Relu function, and the Relu function makes the output of some neurons be 0, thereby causing sparsity of the network, reducing interdependence of parameters, and alleviating the problem of overfitting the Softmax function, and the function formula is as follows:
Figure FDA0002842427950000022
wherein L is the step length.
7. The AI micro-aperture wall detection algorithm of claim 6, wherein: the method for judging whether the generated image is true or false in S3 is as follows:
s4.1, generating cloth image data by using a generator and a real defect-free sample;
s4.2, sending the data generated in the S4.1 and the real data into a discriminator, calculating loss, fixing generator parameters at the moment, and updating parameters of the discriminator according to the loss;
s4.3, after the training of the discriminator is finished, fixing the parameters of the discriminator, and setting the label of the image generated by the generator as a real image, thereby calculating the loss and transmitting the loss back to the generator;
and S4.4, repeating the S4.2 and the S4.3 until the picture generated by the generator is consistent with the real defect picture.
8. The AI micro-aperture wall detection algorithm of claim 7, wherein: the data generation method of the generator in S4.1 is as follows:
s5.1, the generator is a function for generating one value at a time;
s5.2, calling the function will return a Generator that can be used to generate continuous values.
9. The AI micro-aperture wall detection algorithm of claim 1, wherein: the loss function in S4 is a penalty loss function.
10. The AI micro-aperture wall detection algorithm of claim 9, wherein: the functional expression of the countermeasure loss function is as follows:
Figure FDA0002842427950000031
wherein R is an R channel; g is a G channel; and B is a B channel.
CN202011496848.XA 2020-12-17 2020-12-17 AI micro hole wall detection method Active CN112488130B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011496848.XA CN112488130B (en) 2020-12-17 2020-12-17 AI micro hole wall detection method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011496848.XA CN112488130B (en) 2020-12-17 2020-12-17 AI micro hole wall detection method

Publications (2)

Publication Number Publication Date
CN112488130A true CN112488130A (en) 2021-03-12
CN112488130B CN112488130B (en) 2023-08-15

Family

ID=74917376

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011496848.XA Active CN112488130B (en) 2020-12-17 2020-12-17 AI micro hole wall detection method

Country Status (1)

Country Link
CN (1) CN112488130B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113284086A (en) * 2021-03-31 2021-08-20 广东电力信息科技有限公司 Method and device for generating and detecting power scarcity defect image and related equipment

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2002010962A1 (en) * 2000-07-28 2002-02-07 Storymail, Inc. System, method and computer program product for device, operating system, and network transport neutral secure interactive multi-media messaging
US20120201454A1 (en) * 2011-02-07 2012-08-09 Genta Sato Signal processing device, signal processing method, imaging apparatus, and imaging processing method
AU2018100325A4 (en) * 2018-03-15 2018-04-26 Nian, Xilai MR A New Method For Fast Images And Videos Coloring By Using Conditional Generative Adversarial Networks
CN110189268A (en) * 2019-05-23 2019-08-30 西安电子科技大学 Underwater picture color correcting method based on GAN network
WO2020018154A1 (en) * 2018-07-19 2020-01-23 The Regents Of The University Of California Method and system for digital staining of label-free phase images using deep learning
CN111340791A (en) * 2020-03-02 2020-06-26 浙江浙能技术研究院有限公司 Photovoltaic module unsupervised defect detection method based on GAN improved algorithm
CN111445459A (en) * 2020-03-27 2020-07-24 广东工业大学 Image defect detection method and system based on depth twin 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
WO2020233564A1 (en) * 2019-05-21 2020-11-26 华为技术有限公司 Method and electronic device for detecting adversarial example

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2002010962A1 (en) * 2000-07-28 2002-02-07 Storymail, Inc. System, method and computer program product for device, operating system, and network transport neutral secure interactive multi-media messaging
US20120201454A1 (en) * 2011-02-07 2012-08-09 Genta Sato Signal processing device, signal processing method, imaging apparatus, and imaging processing method
AU2018100325A4 (en) * 2018-03-15 2018-04-26 Nian, Xilai MR A New Method For Fast Images And Videos Coloring By Using Conditional Generative Adversarial Networks
WO2020018154A1 (en) * 2018-07-19 2020-01-23 The Regents Of The University Of California Method and system for digital staining of label-free phase images using deep learning
WO2020233564A1 (en) * 2019-05-21 2020-11-26 华为技术有限公司 Method and electronic device for detecting adversarial example
CN110189268A (en) * 2019-05-23 2019-08-30 西安电子科技大学 Underwater picture color correcting method based on GAN network
CN111340791A (en) * 2020-03-02 2020-06-26 浙江浙能技术研究院有限公司 Photovoltaic module unsupervised defect detection method based on GAN improved algorithm
CN111445459A (en) * 2020-03-27 2020-07-24 广东工业大学 Image defect detection method and system based on depth twin 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

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113284086A (en) * 2021-03-31 2021-08-20 广东电力信息科技有限公司 Method and device for generating and detecting power scarcity defect image and related equipment

Also Published As

Publication number Publication date
CN112488130B (en) 2023-08-15

Similar Documents

Publication Publication Date Title
CN110210486B (en) Sketch annotation information-based generation countermeasure transfer learning method
CN107506695A (en) Video monitoring equipment failure automatic detection method
US20100228694A1 (en) Data Processing Using Restricted Boltzmann Machines
CN112766315B (en) Method and system for testing robustness of artificial intelligence model
CN112489164B (en) Image coloring method based on improved depth separable convolutional neural network
CN105718932A (en) Colorful image classification method based on fruit fly optimization algorithm and smooth twinborn support vector machine and system thereof
CN115272777B (en) Semi-supervised image analysis method for power transmission scene
CN112287941A (en) License plate recognition method based on automatic character region perception
CN111652297B (en) Fault picture generation method for image detection model training
CN112488130A (en) AI micro-pore wall detection algorithm
CN112131403A (en) Knowledge graph representation learning method in dynamic environment
CN114972216A (en) Construction method and application of texture surface defect detection model
CN115861246A (en) Product quality abnormity detection method and system applied to industrial Internet
CN116188516A (en) Training method of defect data generation model
CN114359387A (en) Bag cultivation mushroom detection method based on improved YOLOV4 algorithm
Fang et al. A New Method of Image Restoration Technology Based on WGAN.
CN115082435B (en) Defect detection method based on self-supervision momentum contrast
CN114723733B (en) Class activation mapping method and device based on axiom explanation
CN116433509A (en) Progressive image defogging method and system based on CNN and convolution LSTM network
CN115457015A (en) Image no-reference quality evaluation method and device based on visual interactive perception double-flow network
CN115908379A (en) Electric cooker liner image data enhancement method based on mask generation countermeasure network
CN115690574A (en) Remote sensing image ship detection method based on self-supervision learning
CN115223033A (en) Synthetic aperture sonar image target classification method and system
CN110889461B (en) Image multi-feature extraction and classification method for chassis assembly quality detection
CN113077525A (en) Image classification method based on frequency domain contrast 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
GR01 Patent grant
GR01 Patent grant