CN112819829A - Visual defect detection method based on double-depth learning model - Google Patents
Visual defect detection method based on double-depth learning model Download PDFInfo
- Publication number
- CN112819829A CN112819829A CN202110417743.9A CN202110417743A CN112819829A CN 112819829 A CN112819829 A CN 112819829A CN 202110417743 A CN202110417743 A CN 202110417743A CN 112819829 A CN112819829 A CN 112819829A
- Authority
- CN
- China
- Prior art keywords
- deep learning
- learning model
- layer
- defect
- filter size
- 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
- 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
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- 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
- G06N3/047—Probabilistic or stochastic 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
- G06N3/08—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
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a visual defect detection method based on a double-depth learning model, which comprises the following steps (J1): intercepting a required detection image into a plurality of input images with preset sizes; further comprising the step (J2): detecting the input images intercepted in the step (J1) by using a first deep learning model, and screening out defect images from all the input images intercepted in the step (J1); step (J3): classifying the defect varieties of the defect images screened in the step (J2) by using a second deep learning model; the number of layers of the first deep learning model is less than the number of layers of the second deep learning model. The invention combines and uses two deep learning models, not only utilizes the advantages of deep learning, but also solves the problem of long time consumption of deep learning.
Description
Technical Field
The invention relates to the technical field of visual defect detection, in particular to a visual defect detection method based on a double-deep learning model.
Background
There is a great need for defect detection in the industrial field. With the rapid development of machine vision and industrial automation, defect detection work traditionally performed by humans has begun to be replaced by machine vision. The core of machine vision is an algorithm, but the traditional algorithm generally has the problem that defect characteristics are not easy to describe when the defect detection task in the industrial field is solved. The defective products defined by the customers are very difficult and complicated to be dequantized through the characteristics of color, area, roundness, angle and length, so that the defect detection problem is time-consuming and labor-consuming to solve by using the traditional algorithm.
Although the deep learning algorithm can automatically extract the defect characteristics through learning, the difficulty of algorithm development is reduced, and the algorithm development period is shortened. However, the deep learning also has the problems of large calculation amount and long algorithm time consumption, and the generation efficiency of the visual detection equipment cannot meet the customer expectation. The deep learning algorithm needs a large amount of calculation, an image filter acquired in industrial detection is large, and the running time of the deep learning algorithm is increased. Therefore, in an actual industrial detection scene, the deep learning algorithm is applied less and is used as an auxiliary of a traditional algorithm more often.
The existing deep learning algorithm is mainly applied to industrial defect detection in two ways: 1) the deep learning segmentation model is used for carrying out defect segmentation (such as FCN, Unet and the like) on the whole image, and the algorithm is long in time consumption; 2) the entire image is cut into small images of a fixed size, and then a region-segmented deep learning model or an image-classified deep learning model is applied to all the small images.
These methods of defect detection using deep learning have two major problems: 1) the size of an acquired image in defect detection is large, and the time consumption of deep learning detection is long; 2) the deep learning model is difficult to adapt to defect detection application in the industrial field, and the detection effect is poor.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: a visual defect detection method based on a double-deep learning model is provided.
The technical scheme adopted by the invention for solving the technical problems is as follows: a visual defect detection method based on a dual-deep learning model, the visual defect detection method having the step (J1): and cutting a required detection image into a plurality of input images with preset sizes.
Further comprising the step (J2): and (4) detecting the input images intercepted in the step (J1) by using a first deep learning model, and screening out defect images from all the input images intercepted in the step (J1).
Step (J3): classifying the defect varieties of the defect images screened in the step (J2) by using a second deep learning model; the number of layers of the first deep learning model is less than the number of layers of the second deep learning model.
Further, the training process of the first deep learning model and the second deep learning model comprises: step (X1): and intercepting a collected image into a plurality of input images with preset sizes.
Step (X2): all the input images of the step (X1) are subjected to category labeling, and all the input images of the step (X1) are classified into normal and multiple defect varieties.
Step (X3): performing classification training using a first deep learning model, wherein the first deep learning model classifies the input images labeled in the step (X2) into two types: a normal category and a defect category; the plurality of defect varieties labeled in the step (X2) are all classified into defect types.
Step (X4): training a second deep learning model by using the input images of the plurality of defect varieties labeled in the step (X2) as a sample library, wherein the second deep learning model classifies each defect variety from the input images of the defect types classified by the first deep learning model in the step (X3).
Further, the structure of the first deep learning model is as follows: the first layer of the first deep learning model is an input layer of an input image.
The second layer of the first deep learning model is a convolution layer with a filter size of 7 × 7, a channel number of 16 and a step size of 2.
And the third layer and the fourth layer of the first deep learning model are convolution layers with the filter size of 3 x3, the channel number of 16 and the step length of 2.
And the fifth layer and the sixth layer of the first deep learning model are convolution layers with the filter size of 3 x3, the channel number of 64 and the step length of 1.
Further, the second deep learning model has a structure in which: the first layer of the second deep learning model is an input layer of an input image. And the second layer to the seventeenth layer of the second deep learning model are convolution layers in sequence.
The structure of the second deep learning model is as follows: and the eighteenth layer of the second deep learning model is a global pooling layer. And the nineteenth layer of the second deep learning model is a full connection layer. The twentieth layer of the second deep learning model is a Softmax layer.
Further, the second deep learning model has a structure in which: the second layer of the second deep learning model is connected with the eighth layer of the second deep learning model through a first bypass, and the first bypass is a convolution layer with the filter size of 1 × 1 and the step length of 2.
And the tenth layer of the second deep learning model is connected with the sixteenth layer of the second deep learning model through a second bypass, and the second bypass is a convolution layer with the filter size of 1 x1 and the step length of 2.
Further, the second deep learning model has a structure in which: the first layer of the second deep learning model and the third layer of the second deep learning model, and, the third layer of the second deep learning model and the fifth layer of the second deep learning model, and, the seventh layer of the second deep learning model and the ninth layer of the second deep learning model, and, the eleventh layer of the second deep learning model and the thirteenth layer of the second deep learning model and the fifteenth layer of the second deep learning model and the seventeenth layer of the second deep learning model are connected through a branch respectively.
Further, the second deep learning model has a structure in which: and the second layer to the fifth layer of the second deep learning model are convolution layers with the filter size of 3 x3, the channel number of 64 and the step length of 1 in sequence.
The sixth layer of the second deep learning model is a convolution layer with a filter size of 3 × 3, a channel number of 128, and a step size of 2.
And the seventh layer to the ninth layer of the second deep learning model are convolution layers with the filter size of 3 x3, the channel number of 128 and the step length of 1 in sequence.
The tenth layer of the second deep learning model is a convolution layer with a filter size of 3 × 3, a channel number of 256, and a step size of 2.
The eleventh layer to the thirteenth layer of the second deep learning model are convolution layers with the filter size of 3 x3, the channel number of 256 and the step length of 1.
The fourteenth layer of the second deep learning model is a convolutional layer with a filter size of 3 × 3, a channel number of 512, and a step size of 2.
And the fifteenth layer to the seventeenth layer of the second deep learning model are convolution layers with the filter size of 3 x3, the channel number of 512 and the step length of 1 in sequence.
The invention has the beneficial effects that: the invention adopts two deep learning models, firstly, a first shallow deep learning model is used for screening out defect images from all input images, thereby filtering a large number of non-defect images and leaving a small number of defect images. And then, a deeper second deep learning model is used for further classifying the defect varieties of the defect images, and the classification is more accurate by using the second deep learning model. Through the recognition classification of two-step deep learning, the detection efficiency of the algorithm is improved, and the defect detection effect is improved. The invention combines and uses two deep learning models, not only utilizes the advantages of deep learning, but also solves the problem of long time consumption of deep learning.
Drawings
FIG. 1 is a model diagram of a first deep learning model in the present invention.
FIG. 2 is a model diagram of a second deep learning model in the present invention.
Detailed Description
The invention will now be described in further detail with reference to the drawings and preferred embodiments. These drawings are simplified schematic views illustrating only the basic structure of the present invention in a schematic manner, and thus show only the constitution related to the present invention.
A visual defect detection method based on a dual-deep learning model, the visual defect detection method having the step (J1): and cutting a required detection image into a plurality of input images with preset sizes.
Further comprising the step (J2): and (4) detecting the input images intercepted in the step (J1) by using a first deep learning model, and screening out defect images from all the input images intercepted in the step (J1).
Step (J3): classifying the defect varieties of the defect images screened in the step (J2) by using a second deep learning model; the number of layers of the first deep learning model is less than the number of layers of the second deep learning model.
Further, the training process of the first deep learning model and the second deep learning model comprises: step (X1): and intercepting a collected image into a plurality of input images with preset sizes.
Step (X2): all the input images of the step (X1) are subjected to category labeling, and all the input images of the step (X1) are classified into normal and multiple defect varieties. For example, the plurality of defect varieties are classified into defect a, defect B, defect C, defect D, defect E, and the like.
Step (X3): performing classification training using a first deep learning model, wherein the first deep learning model classifies the input images labeled in the step (X2) into two types: a normal category and a defect category; the plurality of defect varieties labeled in the step (X2) are all classified into defect types. For example, a plurality of defect varieties such as defect a, defect B, defect C, defect D, defect E, and the like are classified into defect categories.
Step (X4): training a second deep learning model by using the input images of the plurality of defect varieties labeled in the step (X2) as a sample library, wherein the second deep learning model classifies each defect variety from the input images of the defect types classified by the first deep learning model in the step (X3).
As shown in fig. 1, the structure of the first deep learning model is: the first layer of the first deep learning model is an input layer of an input image. The second layer of the first deep learning model is a convolution layer with a filter size of 7 × 7, a channel number of 16 and a step size of 2. And the third layer and the fourth layer of the first deep learning model are convolution layers with the filter size of 3 x3, the channel number of 16 and the step length of 2. And the fifth layer and the sixth layer of the first deep learning model are convolution layers with the filter size of 3 x3, the channel number of 64 and the step length of 1. And the seventh layer of the first deep learning model is a fully connected layer. The eighth layer of the first deep learning model is a Softmax layer.
The first deep learning model is used for performing classification of a normal image and a defect image. For example, the defect image may include one or more defect varieties of defect a, defect B, defect C, defect D, defect E, and the like. The defect images are usually smaller in proportion in all input images, so that the shallow first deep learning model is used, the calculation force of deep learning can be saved, the first deep learning model is high in running speed, the defect detection speed is increased, and the defect classification method is more suitable for defect classification in the industrial field.
As shown in fig. 2, the structure of the second deep learning model is: the first layer of the second deep learning model is an input layer of an input image. And the second layer to the seventeenth layer of the second deep learning model are convolution layers in sequence. And the eighteenth layer of the second deep learning model is a global pooling layer. And the nineteenth layer of the second deep learning model is a full connection layer. The twentieth layer of the second deep learning model is a Softmax layer.
The structure of the second deep learning model is as follows: the second layer of the second deep learning model is connected with the eighth layer of the second deep learning model through a first bypass, and the first bypass is a convolution layer with the filter size of 1 × 1 and the step length of 2. And the tenth layer of the second deep learning model is connected with the sixteenth layer of the second deep learning model through a second bypass, and the second bypass is a convolution layer with the filter size of 1 x1 and the step length of 2.
The second deep learning model is based on ResNet18, is improved in order to adapt to the application of visual inspection in the industrial field, and is additionally provided with a first side branch and a second side branch, so that the bottom layer features of the image can be conducted sufficiently, and a better classification effect is achieved. The second deep learning model is particularly suitable for the scenes of defect classification in the industrial field, and the classification of defect varieties by using the second deep learning model is more accurate.
The structure of the second deep learning model is as follows: the first layer of the second deep learning model and the third layer of the second deep learning model, and, the third layer of the second deep learning model and the fifth layer of the second deep learning model, and, the seventh layer of the second deep learning model and the ninth layer of the second deep learning model, and, the eleventh layer of the second deep learning model and the thirteenth layer of the second deep learning model and the fifteenth layer of the second deep learning model and the seventeenth layer of the second deep learning model are connected through a branch respectively.
The structure of the second deep learning model is as follows: the second layer to the fifth layer of the second deep learning model are convolution layers with the filter size of 3 x3, the channel number of 64 and the step length of 1 in sequence; the sixth layer of the second deep learning model is a convolution layer with the filter size of 3 x3, the channel number of 128 and the step length of 2; the seventh layer to the ninth layer of the second deep learning model are convolution layers with the filter size of 3 x3, the channel number of 128 and the step length of 1 in sequence; the tenth layer of the second deep learning model is a convolution layer with the filter size of 3 x3, the channel number of 256 and the step length of 2; the eleventh layer to the thirteenth layer of the second deep learning model are convolution layers with the filter size of 3 x3, the channel number of 256 and the step length of 1; the fourteenth layer of the second deep learning model is a convolution layer with the filter size of 3 x3, the channel number of 512 and the step length of 2; and the fifteenth layer to the seventeenth layer of the second deep learning model are convolution layers with the filter size of 3 x3, the channel number of 512 and the step length of 1 in sequence.
The invention adopts two deep learning models, firstly, a first shallow deep learning model is used for screening out defect images from all input images, thereby filtering a large number of non-defect images and leaving a small number of defect images. And then, a deeper second deep learning model is used for further classifying the defect varieties of the defect images, and the classification is more accurate by using the second deep learning model. Through the recognition classification of two-step deep learning, the detection efficiency of the algorithm is improved, and the defect detection effect is improved. The invention combines and uses two improved deep learning models, not only utilizes the advantages of deep learning, but also solves the problem of long time consumption of deep learning.
While particular embodiments of the present invention have been described in the foregoing specification, various modifications and alterations to the previously described embodiments will become apparent to those skilled in the art from this description without departing from the spirit and scope of the invention.
Claims (8)
1. A visual defect detection method based on a dual-deep learning model, the visual defect detection method having the step (J1): intercepting a required detection image into a plurality of input images with preset sizes;
the method is characterized in that: further comprising the step (J2): detecting the input images intercepted in the step (J1) by using a first deep learning model, and screening out defect images from all the input images intercepted in the step (J1);
step (J3): classifying the defect varieties of the defect images screened in the step (J2) by using a second deep learning model; the number of layers of the first deep learning model is less than the number of layers of the second deep learning model.
2. The visual defect detection method based on the dual-deep learning model as claimed in claim 1, wherein: the training process of the first deep learning model and the second deep learning model comprises the following steps: step (X1): intercepting a collected image into a plurality of input images with preset sizes;
step (X2): performing category labeling on all input images in the step (X1), and classifying all input images in the step (X1) into normal and multiple defect varieties;
step (X3): performing classification training using a first deep learning model, wherein the first deep learning model classifies the input images labeled in the step (X2) into two types: a normal category and a defect category; classifying the plurality of defect varieties labeled in the step (X2) into defect categories;
step (X4): training a second deep learning model by using the input images of the plurality of defect varieties labeled in the step (X2) as a sample library, wherein the second deep learning model classifies each defect variety from the input images of the defect types classified by the first deep learning model in the step (X3).
3. The visual defect detection method based on the dual-deep learning model as claimed in claim 1, wherein: the structure of the first deep learning model is as follows: the first layer of the first deep learning model is an input layer of an input image;
the second layer of the first deep learning model is a convolution layer with the filter size of 7 x 7, the channel number of 16 and the step length of 2;
the third layer and the fourth layer of the first deep learning model are convolution layers with the filter size of 3 x3, the channel number of 16 and the step length of 2;
and the fifth layer and the sixth layer of the first deep learning model are convolution layers with the filter size of 3 x3, the channel number of 64 and the step length of 1.
4. The visual defect detection method based on the dual-deep learning model as claimed in claim 1, wherein: the structure of the second deep learning model is as follows: the first layer of the second deep learning model is an input layer of an input image;
and the second layer to the seventeenth layer of the second deep learning model are convolution layers in sequence.
5. The visual defect detection method based on the dual-deep learning model as claimed in claim 4, wherein: the structure of the second deep learning model is as follows:
an eighteenth layer of the second deep learning model is a global pooling layer;
a nineteenth layer of the second deep learning model is a full connection layer;
the twentieth layer of the second deep learning model is a Softmax layer.
6. The visual defect detection method based on the dual-deep learning model as claimed in claim 4, wherein: the structure of the second deep learning model is as follows:
the second layer of the second deep learning model is connected with the eighth layer of the second deep learning model through a first bypass, and the first bypass is a convolution layer with the filter size of 1 × 1 and the step length of 2;
and the tenth layer of the second deep learning model is connected with the sixteenth layer of the second deep learning model through a second bypass, and the second bypass is a convolution layer with the filter size of 1 x1 and the step length of 2.
7. The visual defect detection method based on the dual-deep learning model as claimed in claim 4, wherein: the structure of the second deep learning model is as follows:
the first layer of the second deep learning model and the third layer of the second deep learning model, and, the third layer of the second deep learning model and the fifth layer of the second deep learning model, and, the seventh layer of the second deep learning model and the ninth layer of the second deep learning model, and, the eleventh layer of the second deep learning model and the thirteenth layer of the second deep learning model and the fifteenth layer of the second deep learning model and the seventeenth layer of the second deep learning model are connected through a branch respectively.
8. The visual defect detection method based on the dual-deep learning model as claimed in claim 4, wherein: the structure of the second deep learning model is as follows:
the second layer to the fifth layer of the second deep learning model are convolution layers with the filter size of 3 x3, the channel number of 64 and the step length of 1 in sequence;
the sixth layer of the second deep learning model is a convolution layer with the filter size of 3 x3, the channel number of 128 and the step length of 2;
the seventh layer to the ninth layer of the second deep learning model are convolution layers with the filter size of 3 x3, the channel number of 128 and the step length of 1 in sequence;
the tenth layer of the second deep learning model is a convolution layer with the filter size of 3 x3, the channel number of 256 and the step length of 2;
the eleventh layer to the thirteenth layer of the second deep learning model are convolution layers with the filter size of 3 x3, the channel number of 256 and the step length of 1;
the fourteenth layer of the second deep learning model is a convolution layer with the filter size of 3 x3, the channel number of 512 and the step length of 2;
and the fifteenth layer to the seventeenth layer of the second deep learning model are convolution layers with the filter size of 3 x3, the channel number of 512 and the step length of 1 in sequence.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110417743.9A CN112819829A (en) | 2021-04-19 | 2021-04-19 | Visual defect detection method based on double-depth learning model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110417743.9A CN112819829A (en) | 2021-04-19 | 2021-04-19 | Visual defect detection method based on double-depth learning model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112819829A true CN112819829A (en) | 2021-05-18 |
Family
ID=75863695
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110417743.9A Pending CN112819829A (en) | 2021-04-19 | 2021-04-19 | Visual defect detection method based on double-depth learning model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112819829A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114998192A (en) * | 2022-04-19 | 2022-09-02 | 深圳格芯集成电路装备有限公司 | Defect detection method, device and equipment based on deep learning and storage medium |
| EP4361615A4 (en) * | 2021-07-08 | 2024-10-23 | Jfe Steel Corp | Inspection method, classification method, management method, steel material manufacturing method, training model generation method, training model, inspection device, and steel material manufacturing facility |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107180419A (en) * | 2016-04-15 | 2017-09-19 | 北京理工大学 | A kind of medium filtering detection method based on PCA networks |
| CN109859171A (en) * | 2019-01-07 | 2019-06-07 | 北京工业大学 | A kind of flooring defect automatic testing method based on computer vision and deep learning |
| CN112419401A (en) * | 2020-11-23 | 2021-02-26 | 上海交通大学 | Aircraft surface defect detection system based on cloud edge cooperation and deep learning |
-
2021
- 2021-04-19 CN CN202110417743.9A patent/CN112819829A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107180419A (en) * | 2016-04-15 | 2017-09-19 | 北京理工大学 | A kind of medium filtering detection method based on PCA networks |
| CN109859171A (en) * | 2019-01-07 | 2019-06-07 | 北京工业大学 | A kind of flooring defect automatic testing method based on computer vision and deep learning |
| CN112419401A (en) * | 2020-11-23 | 2021-02-26 | 上海交通大学 | Aircraft surface defect detection system based on cloud edge cooperation and deep learning |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP4361615A4 (en) * | 2021-07-08 | 2024-10-23 | Jfe Steel Corp | Inspection method, classification method, management method, steel material manufacturing method, training model generation method, training model, inspection device, and steel material manufacturing facility |
| CN114998192A (en) * | 2022-04-19 | 2022-09-02 | 深圳格芯集成电路装备有限公司 | Defect detection method, device and equipment based on deep learning and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110992317B (en) | PCB defect detection method based on semantic segmentation | |
| CN112037219B (en) | Metal surface defect detection method based on two-stage convolutional neural network | |
| CN113298757A (en) | Metal surface defect detection method based on U-NET convolutional neural network | |
| CN113643268B (en) | Industrial product defect quality inspection method and device based on deep learning and storage medium | |
| CN112819829A (en) | Visual defect detection method based on double-depth learning model | |
| CN110555831B (en) | Deep learning-based drainage pipeline defect segmentation method | |
| CN112907561A (en) | Notebook appearance flaw detection method based on deep learning | |
| CN110135486B (en) | Chopstick image classification method based on adaptive convolutional neural network | |
| CN114332008B (en) | Unsupervised defect detection and positioning method based on multi-level feature reconstruction | |
| CN111382785A (en) | GAN network model and method for realizing automatic cleaning and auxiliary marking of sample | |
| CN112102281B (en) | Truck brake cylinder fault detection method based on improved Faster Rcnn | |
| CN115082401B (en) | SMT production line chip mounter fault prediction method based on improved YOLOX and PNN | |
| CN111079645A (en) | Insulator self-explosion identification method based on AlexNet network | |
| CN115880223A (en) | Improved YOLOX-based high-reflectivity metal surface defect detection method | |
| CN116128839A (en) | Wafer defect identification method, device, electronic equipment and storage medium | |
| CN111524120A (en) | Printed matter tiny scratch and overprint deviation detection method based on deep learning | |
| CN113887524A (en) | Magnetite microscopic image segmentation method based on semantic segmentation | |
| CN116310548A (en) | Method for detecting invasive plant seeds in imported seed products | |
| Jia | Fabric defect detection based on open source computer vision library OpenCV | |
| CN113256608B (en) | Workpiece defect detection method and device | |
| CN113870204A (en) | Method and device for detecting abnormality of optical glass | |
| CN116958073A (en) | Small sample steel defect detection method based on attention feature pyramid mechanism | |
| CN116542962A (en) | Improved Yolov5m model-based photovoltaic cell defect detection method | |
| CN116823717A (en) | Neural network model for defect detection and training method, system and equipment thereof | |
| CN113947567B (en) | Defect detection method based on multitask learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210518 |
|
| RJ01 | Rejection of invention patent application after publication |