CN111402196A - Bearing roller image generation method based on countermeasure generation network - Google Patents
Bearing roller image generation method based on countermeasure generation network Download PDFInfo
- Publication number
- CN111402196A CN111402196A CN202010083798.6A CN202010083798A CN111402196A CN 111402196 A CN111402196 A CN 111402196A CN 202010083798 A CN202010083798 A CN 202010083798A CN 111402196 A CN111402196 A CN 111402196A
- Authority
- CN
- China
- Prior art keywords
- bearing roller
- loss
- network
- image
- data
- 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
Images
Classifications
-
- 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
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30164—Workpiece; Machine component
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)
- Image Processing (AREA)
Abstract
A bearing roller image generation method based on a countermeasure generation network, comprising the steps of: 1) acquiring data, and shooting a bearing roller picture through experimental device equipment; 2) data processing, namely preprocessing the data through a digital image processing technology; 3) constructing DCGAN deep convolution to generate a countermeasure network; 4) and selecting different training sets, and generating a new bearing roller picture by using the DCGAN. The method utilizes the experimental device to shoot training data, trains the training set to generate a new bearing roller image by constructing and generating the confrontation network, and solves the problem of insufficient data set of the bearing roller defect detection algorithm because the new bearing roller image has various defects.
Description
Technical Field
The invention relates to an image processing, deep learning, computer vision and countermeasure generation network, in particular to a bearing roller image generation method based on the countermeasure generation network.
Background
In national economy, the rolling bearing element is a name of an industrial joint and is widely applied to the fields of automobile industry, military equipment, aerospace, instruments and meters and the like. Today, the market demand for high quality products is becoming more stringent, and rollers are important components of bearings, the quality of which plays a crucial role in the performance of bearings. However, in the production process of the bearing roller, due to the influence of equipment and processes, different kinds of defects, such as scratches, holes, cracks and the like, often appear on the surface of the product. The surface quality not only affects the appearance of the product, but also more likely affects the functional characteristics of the product, and causes great loss to enterprises, so that the surface defect detection of the bearing roller is very necessary.
The traditional bearing roller surface defect detection method is a manual detection method, and manual detection has many defects, such as limited space and time resolution of human eyes, and easy generation of false detection and missed detection. In addition, manual detection also occupies more human resources, and the production cost of enterprises is increased. In view of the problems of manual detection, an advanced and efficient detection method is needed to replace the conventional manual detection. With the development of technologies such as image processing, pattern recognition and the like, a surface defect detection method mainly based on machine vision is widely applied in a product quality control link. In recent years, artificial neural networks have also been gradually applied to defect detection and identification, and have achieved good results. The idea of deep learning, which can autonomously learn from multiple layers of characterizations of the underlying distribution of modeled data, is derived from an artificial neural network. Compared with the traditional machine vision method, the deep learning can reduce the influence of manually extracted features on the identification precision. Aiming at the problems of complex characteristics and difficult extraction of the target to be detected, the deep learning can also provide a good solution. Due to the excellent performance of deep learning, a lot of research results show that the deep learning has excellent performance in the tasks of feature learning and identification.
Compared with manual detection, the bearing roller surface defect detection method based on deep learning can effectively control the surface quality of products, improve the production efficiency of enterprises, reduce the labor intensity of workers and the production cost, can more accurately detect and identify the surface defects of the products, and has important significance for promoting the development of manufacturing industry, improving the production process of the enterprises and improving the competitiveness of the enterprises. However, deep learning requires a large amount of data, and a bearing roller factory produces an excessively small number of defective products, resulting in insufficient data, and therefore, it is necessary to expand the data set using a countermeasure generation network.
GAN is collectively referred to as a generic adaptive Networks, meaning that the network is generated against. The original GAN is an unsupervised learning method, which skillfully utilizes the thought of 'antagonism' to learn a generative model, and can generate a brand new data sample once training is completed. DCGAN expands the concept of GAN into a convolutional neural network, and can generate picture samples with higher quality.
Disclosure of Invention
In order to overcome the difficulty of insufficient photos of defects in the existing data set, the invention provides a bearing roller image generation method based on a countermeasure generation network for supplementing the data set, and the method combines a digital image processing technology and a DCGAN to generate a bearing roller image meeting the requirements.
The technical scheme adopted by the invention for solving the technical problems is as follows:
a bearing roller image generation method based on a countermeasure generation network, comprising the steps of:
s1: acquiring data, and shooting a bearing roller picture through experimental device equipment;
s2: data processing, namely preprocessing the data through a digital image processing technology;
s3: constructing DCGAN deep convolution to generate a countermeasure network;
s4: and selecting different training sets, and generating a new bearing roller picture by using the DCGAN.
Further, in step S1, the experimental device includes a transfer disc made of transparent material, the edge of the transfer disc is provided with 4 stations, and each station is provided with a CMOS area-array camera; the 4 stations are respectively a bearing roller outer cylindrical surface detection station, a bearing roller inner cylindrical surface detection station, a bearing roller upper end surface detection station and a bearing roller lower end surface detection station; wherein: a CMOS area-array shooting camera of a bearing roller outer cylindrical surface detection station is provided with a ring outer side unfolding lens, a CMOS area-array shooting camera of a bearing roller inner cylindrical surface detection station is provided with a hole inner wall detection lens, and a plane shooting lens is arranged on the CMOS area-array shooting camera of a bearing roller upper end surface detection station and a bearing roller lower end surface detection station.
Further, in the step S2, because only the defective photos of the bearing roller are needed and the good photos are not needed, and the appearances of the upper end surface and the lower end surface of the bearing roller are consistent, some defective photos are selected and put into three folders, and the three folders store photos of different stations respectively; the data preprocessing comprises cutting and rotating, the size of original data is 1920-1200, all pictures are cut to 900-900, in order to expand a data set, each photo is rotated anticlockwise for a plurality of times, the angles are 45 degrees, 90 degrees, 135 degrees, 180 degrees, 225 degrees, 270 degrees, 315 degrees and the like, and the result of each rotation is stored;
the step S3 includes the steps of:
s3-1: training the bearing roller images by using a DCGAN-tensorflow network, defining and initializing relevant parameters, wherein the iteration number is set to 10000, the learning rate of Adam is default to 0.0002, the number of batch images, namely batch _ size, is default to 64, the number is set to 1, the input image size is 900 × 900, and the output image size is 227 × 227;
s3-2: loss of true sample input (D _ loss _ real) is such that D _ locations is close to 1, i.e., D identifies a true sample as true, loss of false sample input (D _ loss _ fake) is such that D _ locations _ is close to 0, D identifies a false sample as false, g _ loss: for D to identify a false sample as a true sample, G aims to reduce the loss, and the target loss (D _ loss _ real + D _ loss _ fake) is to minimize the loss. Inputting the images into a discrimination network and a generation network for training, and normalizing all training sample images to be between [ -1,1] by using a tanh activation function, wherein the calculation formula is as follows:
d _ loss and g _ loss are minimized by Adam optimizer. Firstly, inputting noise z and batch _ images, and updating a D network by optimizing D _ optim; then, inputting noise z, and optimizing G _ optim to update the G network, wherein in the training process, the two networks resist against each other to finally form a dynamic balance; the loss function of the network population is defined by:
wherein the discriminant model loss function is as follows:
LOSS(D)＝-(log(D1(x))+log(1-D2(G(z)))
the model loss function is generated as follows:
LOSS(G)＝-(log(D2(G(z))))
in the most ideal case, G can generate a picture G (z) which is very similar to the real picture, and D has difficulty in determining whether the generated picture is true, and randomly guessing whether the picture is true or false, i.e., D (G (z)) is 0.5.
3-3, for the discrimination network, all layers except the last layer use relu as the activation function, and the last layer uses tanh as the activation function, its input is an image, and its output is the probability that the image is a real image in DCGAN, the structure of discriminator D is a convolution nerve network, the input image is convoluted by several layers to get a convolution character, the character is sent to L logistic function, and the probability value of the output is obtained, for the generation network, after each layer operation, the data output is passed through a relu function, then it is normalized by using batchnorm, the last layer uses sigmoid function as the activation function, and it is not normalized by using batchnorm, the generator G receives a 100-D random noise z, it is transformed into 4.4.1024 featrumeshaps by Project and resihape, then it is passed through several deconvolution layers, it generates 227, the image of the discrimination network is stored near 227, and the roller ratio of the discrimination network is 0.5.
In the step 4, for each station, the more the number of the generated bearing roller images with defects is, the better the images are, and the more the defect types are, the better the images are, so that different training sets are made, and roller images with different defects are generated through DCGAN, so as to expand the training set of the bearing roller defect detection algorithm; the first type of training set is characterized by comprising various flaws, the training sets of the three stations are independently trained and new photos are generated, and the newly generated photos not only have the existing flaws of the training sets, but also have new flaws which do not appear in the training sets; the second type of training set is characterized in that only one flaw is contained, the photos of each station are divided into individual training sets according to the types of the flaws, the training sets are independently trained and generated, and the newly generated photos have flaws similar to the flaws in the training sets; the third type of training set is characterized in that the flaws are located at the same positions, and the positions of the flaws are the same as those of the flaws in the training set, but the types of the flaws are variable.
The invention has the beneficial effects that: the method comprises the steps of shooting training data by using an experimental device, training a training set to generate a new bearing roller image by constructing and generating a confrontation network, wherein the new bearing roller image has various defects, and the problem of insufficient data set of the existing algorithm is solved.
Drawings
FIG. 1 is a flow chart of a bearing roller image generation method based on a countermeasure generation network in accordance with an embodiment of the present invention;
FIG. 2 is a network model and arbiter network model architecture diagram of a DCGAN deep network model generator according to an embodiment of the present invention;
FIG. 3 is source bearing roller image data;
FIG. 4 is bearing roller image data generated by the DCGAN depth network model.
Detailed Description
The invention will be further explained with reference to the drawings.
Referring to fig. 1 to 4, a bearing roller image generation method based on a countermeasure generation network uses data collected by hardware devices such as an industrial camera as a data set. The method comprises the steps of data set acquisition, data preprocessing, construction of a confrontation generation network, model training and image generation.
The invention comprises the following steps:
s1: acquiring data, and shooting a bearing roller picture through experimental device equipment;
s2: data processing, namely preprocessing the data through a digital image processing technology;
s3: constructing DCGAN deep convolution to generate a countermeasure network;
s4: and selecting different training sets, and generating a new bearing roller picture by using the DCGAN.
Further, in step S1, the experimental device includes a transfer disc made of transparent material, the edge of the transfer disc is provided with 4 stations, and each station is provided with a CMOS area-array camera; the 4 stations are respectively a bearing roller outer cylindrical surface detection station, a bearing roller inner cylindrical surface detection station, a bearing roller upper end surface detection station and a bearing roller lower end surface detection station; wherein: a ring outer side unfolded lens is mounted on a CMOS area array shooting camera of a bearing roller outer cylindrical surface detection station, a hole inner wall detection lens is mounted on the CMOS area array shooting camera of the bearing roller inner cylindrical surface detection station, and a plane shooting lens is mounted on the CMOS area array shooting camera of a bearing roller upper end surface detection station and a bearing roller lower end surface detection station; the CMOS area-array shooting camera of the lower end face detection station of the bearing roller is positioned below the transmission disc, and the rest CMOS area-array shooting cameras are positioned above the transmission disc; the device can realize the full-surface shooting of the bearing roller.
Further, in the step S2, because only the defective photos of the bearing roller are needed and the good photos are not needed, and the appearances of the upper end surface and the lower end surface of the bearing roller are consistent, some defective photos are selected and put into three folders, and the three folders store photos of different stations respectively; the data preprocessing comprises cutting and rotating, the size of original data is 1920-1200, only the middle part of the whole picture is a workpiece, and the rest part of the whole picture is black, so that all the pictures are cut into 900-900 sizes, each picture is rotated anticlockwise for a plurality of times for expanding a data set, the angles are respectively 45 degrees, 90 degrees, 135 degrees, 180 degrees, 225 degrees, 270 degrees, 315 degrees and the like, and the result of each rotation is stored;
the step S3 includes the steps of:
s3-1: training the bearing roller images by using a DCGAN-tensorflow network, defining and initializing relevant parameters, wherein the iteration number is set to 10000, the learning rate of Adam is default to 0.0002, the number of batch images, namely batch _ size, is default to 64, the number is set to 1, the input image size is 900 × 900, and the output image size is 227 × 227;
s3-2: loss of true sample input (D _ loss _ real) is such that D _ locations is close to 1, i.e., D identifies a true sample as true, loss of false sample input (D _ loss _ fake) is such that D _ locations _ is close to 0, D identifies a false sample as false, g _ loss: for D to identify a false sample as a true sample, G aims to reduce the loss, and the target loss (D _ loss _ real + D _ loss _ fake) is to minimize the loss. Inputting the images into a discrimination network and a generation network for training, and normalizing all training sample images to be between [ -1,1] by using a tanh activation function, wherein the calculation formula is as follows:
d _ loss and g _ loss are minimized by Adam optimizer. Firstly, inputting noise z and batch _ images, and updating a D network by optimizing D _ optim; then, inputting noise z, and optimizing G _ optim to update the G network, wherein in the training process, the two networks resist against each other to finally form a dynamic balance; the loss function of the network population is defined by:
wherein the discriminant model loss function is as follows:
LOSS(D)＝-(log(D1(x))+log(1-D2(G(z)))
the model loss function is generated as follows:
LOSS(G)＝-(log(D2(G(z))))
in the most ideal case, G can generate a picture G (z) which is very similar to the real picture, and D has difficulty in determining whether the generated picture is true, and randomly guessing whether the picture is true or false, i.e., D (G (z)) is 0.5.
3-3, for the discrimination network, all layers except the last layer use relu as the activation function, and the last layer uses tanh as the activation function, its input is an image, and its output is the probability that the image is a real image in DCGAN, the structure of discriminator D is a convolution nerve network, the input image is convoluted by several layers to get a convolution character, the character is sent to L logistic function, and the probability value of the output is obtained, for the generation network, after each layer operation, the data output is passed through a relu function, then it is normalized by using batchnorm, the last layer uses sigmoid function as the activation function, and it is not normalized by using batchnorm, the generator G receives a 100-D random noise z, it is transformed into 4.4.1024 featrumeshaps by Project and resihape, then it is passed through several deconvolution layers, it generates 227, the image of the discrimination network is stored near 227, and the roller ratio of the discrimination network is 0.5.
In the step 4, for each station, the more the number of the generated bearing roller images with defects is, the better the images are, and the more the defect types are, the better the images are, so that different training sets are made, and roller images with different defects are generated through DCGAN, so as to expand the training set of the bearing roller defect detection algorithm; the first kind of training set is characterized by comprising various flaws, wherein 500 pictures exist in a station 0, 300 pictures exist in a station 1, 200 pictures exist in a station 2, the training sets of the three stations are independently trained and new photos are generated, and the new photos not only have the existing flaws of the training set, but also have new flaws which do not appear in the training set; the second type of training set is characterized in that only one flaw is contained, the photos of each station are divided into individual training sets according to the types of the flaws, the training sets are independently trained and generated, and the newly generated photos have flaws similar to the flaws in the training sets; the third type of training set is characterized in that the flaws are located at the same positions, and the positions of the flaws are the same as those of the flaws in the training set, but the types of the flaws are variable.
As described above, in the embodiment of the method for generating a bearing roller image based on a countermeasure generation network, training data is captured by using an experimental device, the countermeasure network is constructed and generated, and a training set is trained to generate a new bearing roller image, so as to solve the problem of insufficient data set of the existing algorithm.
The above-mentioned embodiments are only preferred embodiments of the present invention, which are merely illustrative and not restrictive, and any person skilled in the art may substitute or change the technical solution of the present invention and the inventive concept thereof within the scope of the present invention.
Claims (5)
1. A bearing roller image generation method based on a countermeasure generation network, characterized by comprising the steps of:
s1: acquiring data, and shooting a bearing roller picture through experimental device equipment;
s2: data processing, namely preprocessing the data through a digital image processing technology;
s3: constructing DCGAN deep convolution to generate a countermeasure network;
s4: and selecting different training sets, and generating a new bearing roller picture by using the DCGAN.
2. The method for generating an image of a bearing roller based on a network for generating a countermeasure as claimed in claim 1, wherein in step S1, the experimental setup comprises a transfer disc made of transparent material, the edge of the transfer disc is provided with 4 stations, each station is provided with a CMOS area-array camera; the 4 stations are respectively a bearing roller outer cylindrical surface detection station, a bearing roller inner cylindrical surface detection station, a bearing roller upper end surface detection station and a bearing roller lower end surface detection station.
3. The method for generating an image of a bearing roller based on a countermeasure generation network according to claim 1 or 2, wherein in the step S2, the preprocessed data are defect pictures; the data preprocessing comprises cutting and rotating, namely cutting the original data into the same size, rotating each photo anticlockwise for a plurality of times in order to expand the data set, and storing the result of each rotation.
4. A bearing roller image generating method based on a countermeasure generation network according to claim 1 or 2, wherein said step S3 includes the steps of:
s3-1: training the bearing roller image by using a DCGAN-tensorflow network, defining and initializing related parameters, wherein the iteration number is set to 10000, the learning rate of Adam is 0.0002 by default, the number of batch images, batch _ size, is set to 1, the input image size is 900 × 900, and the output image size is 227 × 227;
s3-2: loss of true sample input (D _ loss _ real) is such that D _ locations is close to 1, i.e., D identifies a true sample as true, loss of false sample input (D _ loss _ fake) is such that D _ locations _ is close to 0, D identifies a false sample as false, g _ loss: let D identify the false sample as a true sample, G aims to reduce the loss, and the target loss (D _ loss _ real + D _ loss _ fake) is to minimize the loss, input the images into the discrimination network and the generation network for training, and normalize all training sample images to between [ -1,1] by using the tanh activation function, and the calculation formula is as follows:
minimizing D _ loss and g _ loss through an Adam optimizer, firstly inputting noise z and batch _ images, and updating a D network through optimizing D _ optim; then, inputting noise z, and optimizing G _ optim to update the G network, wherein in the training process, the two networks resist against each other to finally form a dynamic balance; the loss function of the network population is defined by:
wherein the discriminant model loss function is as follows:
LOSS(D)＝-(log(D1(x))+log(1-D2(G(z)))
the model loss function is generated as follows:
LOSS(G)＝-(log(D2(G(z))))
in the most ideal case, G can generate a picture G (z) which is extremely similar to the real picture, and D has difficulty in judging whether the generated picture is true or not, and randomly guesses whether the picture is true or false, namely D (G (z)) is 0.5;
s3-3, for the identification network, all layers except the last layer use relu as an activation function, and the last layer uses tanh as an activation function, the input of the identification network is an image, the output is the probability that the image is a real image, in DCGAN, the structure of a discriminator D is a convolution neural network, the input image is convoluted by a plurality of layers to obtain a convolution characteristic, the obtained characteristic is sent to L logistic function to obtain the probability value of the output, after the generation network operates on each layer, the data output passes through a relu function, then is normalized by using batch _ norm, the last layer uses sigmoid function as an activation function and does not use batch _ norm to perform normalization operation, a generator G receives a 100-D random noise z, is converted into 4.4.1024 featuremap through Project and reshape (fully connected layers), then passes through a plurality of deconvolution layers to generate a rolling bearing identification image with the size of 227, and the rolling ratio of the identification network is 0.5.
5. The method for generating an image of a bearing roller based on a network generated by antagonism as claimed in claim 1 or 2, wherein in step S4, different training sets are made, and images of rollers with different defects are generated by DCGAN to expand the training set of the algorithm for detecting the defects of the bearing roller; the first type of training set is characterized by comprising various flaws and consisting of all flaw photos; the second class of training set is characterized by containing only one flaw and consisting of photos with the same flaw; the third class of training set is characterized by flaws in the same location.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010083798.6A CN111402196A (en) | 2020-02-10 | 2020-02-10 | Bearing roller image generation method based on countermeasure generation network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010083798.6A CN111402196A (en) | 2020-02-10 | 2020-02-10 | Bearing roller image generation method based on countermeasure generation network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111402196A true CN111402196A (en) | 2020-07-10 |
Family
ID=71428369
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010083798.6A Pending CN111402196A (en) | 2020-02-10 | 2020-02-10 | Bearing roller image generation method based on countermeasure generation network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111402196A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114654315A (en) * | 2022-02-17 | 2022-06-24 | 杭州深度视觉科技有限公司 | Machine vision detection system and method for poor grinding of tapered roller base surface |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180075581A1 (en) * | 2016-09-15 | 2018-03-15 | Twitter, Inc. | Super resolution using a generative adversarial network |
| CN109345507A (en) * | 2018-08-24 | 2019-02-15 | 河海大学 | A kind of dam image crack detection method based on transfer learning |
| CN109598287A (en) * | 2018-10-30 | 2019-04-09 | 中国科学院自动化研究所 | The apparent flaws detection method that confrontation network sample generates is generated based on depth convolution |
| US20190197368A1 (en) * | 2017-12-21 | 2019-06-27 | International Business Machines Corporation | Adapting a Generative Adversarial Network to New Data Sources for Image Classification |
| CN110472696A (en) * | 2019-08-22 | 2019-11-19 | 昆明理工大学 | A method of Terahertz human body image is generated based on DCGAN |
| CN110516561A (en) * | 2019-08-05 | 2019-11-29 | 西安电子科技大学 | SAR image target recognition method based on DCGAN and CNN |
-
2020
- 2020-02-10 CN CN202010083798.6A patent/CN111402196A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180075581A1 (en) * | 2016-09-15 | 2018-03-15 | Twitter, Inc. | Super resolution using a generative adversarial network |
| US20190197368A1 (en) * | 2017-12-21 | 2019-06-27 | International Business Machines Corporation | Adapting a Generative Adversarial Network to New Data Sources for Image Classification |
| CN109345507A (en) * | 2018-08-24 | 2019-02-15 | 河海大学 | A kind of dam image crack detection method based on transfer learning |
| CN109598287A (en) * | 2018-10-30 | 2019-04-09 | 中国科学院自动化研究所 | The apparent flaws detection method that confrontation network sample generates is generated based on depth convolution |
| CN110516561A (en) * | 2019-08-05 | 2019-11-29 | 西安电子科技大学 | SAR image target recognition method based on DCGAN and CNN |
| CN110472696A (en) * | 2019-08-22 | 2019-11-19 | 昆明理工大学 | A method of Terahertz human body image is generated based on DCGAN |
Non-Patent Citations (1)
| Title |
|---|
| 黄真;陈家琦;: "基于生成对抗网络自动生成动漫人物形象的研究" * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114654315A (en) * | 2022-02-17 | 2022-06-24 | 杭州深度视觉科技有限公司 | Machine vision detection system and method for poor grinding of tapered roller base surface |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhao et al. | A visual long-short-term memory based integrated CNN model for fabric defect image classification | |
| CN106875373B (en) | Mobile phone screen MURA defect detection method based on convolutional neural network pruning algorithm | |
| Rahman et al. | Face recognition using gabor filters | |
| CN111445459B (en) | Image defect detection method and system based on depth twin network | |
| CN108061735A (en) | The recognition methods of component surface defect and device | |
| CN112837344B (en) | Target tracking method for generating twin network based on condition countermeasure | |
| CN106650721A (en) | Industrial character identification method based on convolution neural network | |
| CN106951856A (en) | Bag extracting method of expressing one's feelings and device | |
| CN109544522A (en) | A kind of Surface Defects in Steel Plate detection method and system | |
| CN107230203A (en) | Casting defect recognition methods based on human eye vision attention mechanism | |
| CN108985337A (en) | A kind of product surface scratch detection method based on picture depth study | |
| CN107609575A (en) | Calligraphy evaluation method, calligraphy evaluating apparatus and electronic equipment | |
| CN112700435B (en) | Wall defect detection method based on deep learning | |
| CN109753864A (en) | A kind of face identification method based on caffe deep learning frame | |
| CN113160136B (en) | Wood defect identification and segmentation method based on improved Mask R-CNN | |
| CN114092793B (en) | End-to-end biological target detection method suitable for complex underwater environment | |
| CN109975307A (en) | Bearing surface defect detection system and detection method based on statistics projection training | |
| Arikan et al. | Surface defect classification in real-time using convolutional neural networks | |
| CN112559791A (en) | Cloth classification retrieval method based on deep learning | |
| CN107194380A (en) | The depth convolutional network and learning method of a kind of complex scene human face identification | |
| CN113095156A (en) | Double-current network signature identification method and device based on inverse gray scale mode | |
| CN111402196A (en) | Bearing roller image generation method based on countermeasure generation network | |
| CN111709443A (en) | Calligraphy character style classification method based on rotation invariant convolution neural network | |
| EP3664021A1 (en) | Server and method for recognizing image using deep learning | |
| CN114549489A (en) | Carved lipstick quality inspection-oriented instance segmentation defect detection method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |