CN111709920A - Template defect detection method - Google Patents
Template defect detection method Download PDFInfo
- Publication number
- CN111709920A CN111709920A CN202010483881.2A CN202010483881A CN111709920A CN 111709920 A CN111709920 A CN 111709920A CN 202010483881 A CN202010483881 A CN 202010483881A CN 111709920 A CN111709920 A CN 111709920A
- Authority
- CN
- China
- Prior art keywords
- network
- defect
- data
- products
- defect detection
- 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
- 230000007547 defect Effects 0.000 title claims abstract description 55
- 238000001514 detection method Methods 0.000 title claims abstract description 25
- 238000012549 training Methods 0.000 claims abstract description 13
- 230000004927 fusion Effects 0.000 claims abstract description 5
- 238000000605 extraction Methods 0.000 claims abstract description 4
- 238000000034 method Methods 0.000 claims description 7
- 238000010586 diagram Methods 0.000 claims description 6
- 238000004519 manufacturing process Methods 0.000 abstract description 11
- 238000002360 preparation method Methods 0.000 abstract description 3
- 230000002950 deficient Effects 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 4
- 238000011161 development Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000013480 data collection Methods 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000011282 treatment Methods 0.000 description 1
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
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
Abstract
A template defect detection method comprises the following steps: acquiring product defect data by fusing small defects and good products acquired from a defect library, and generating corresponding marking data; the first half of the network is a twin network based on a residual error network, and the second half is a feature fusion and extraction network; inputting good products and corresponding defect sample generation and outputting as a label by a network, and training a model; obtaining a comparison network through training and learning, wherein the comparison network has the capability of comparing the difference between two samples, and the output of the network is the data difference degree of the two input samples; and the defect detection is realized by comparing the difference degree between two products output by the network. According to the invention, defect detection can be realized by comparing the difference degree between two products output by the network, so that the time cost of artificial marking is saved, the early-stage preparation work of model training is reduced, the fast switching detection of the products in the work production is realized, and the production efficiency of the products is improved.
Description
Technical Field
The invention relates to the field of defect detection, in particular to a template defect detection method.
Background
For template defect detection, a traditional algorithm and a deep learning algorithm exist at present. The traditional algorithm has the defects that the development period of the algorithm is long, the generalization capability is poor, the development difficulty of the traditional algorithm is high for some complex samples, meanwhile, the current algorithm is not suitable any more due to slight changes of the shape, the size or the color and the like of a product, a professional algorithm person is required to develop a new algorithm again, and the long maintenance period and the high cost of the traditional algorithm are caused.
The basic operation flow of the current deep learning algorithm applied to template defect detection is as follows, firstly, a large amount of defective product data and non-defective product data need to be collected, then corresponding data label marking is carried out manually, and finally, the training model is applied to an industrial production line.
Meanwhile, for some products, the yield is extremely high, the defective rate is extremely low, collection of a large amount of defect data is unrealistic, otherwise serious errors of sample data unbalance can be caused, and time and labor input of data collection and marking are restricted, so that a part of industrial manufacturers apply deep learning algorithms.
Disclosure of Invention
The invention provides a template defect detection method, which aims to solve at least one technical problem.
To solve the above problem, as an aspect of the present invention, there is provided a template defect detecting method including:
step 1, collecting template good product data: acquiring product defect data by fusing small defects and good products acquired from a defect library, and generating corresponding marking data;
step 2, constructing a double-input network model: the first half of the network is a twin network based on a residual error network, and the second half is a feature fusion and extraction network;
step 3, putting training data into the network: inputting good products and corresponding defect sample generation and outputting as a label by a network, and training a model;
step 4, obtaining a comparison network through training and learning, wherein the comparison network has the capability of comparing the difference between two samples, and the output of the network is the data difference degree of the two input samples;
and 5, outputting the difference degree between the two products through a comparison network to further realize defect detection.
Preferably, the defect detection includes the presence or absence of a defect and the position, size, etc. of the defect.
Preferably, the comparison network outputs a thermodynamic diagram, wherein a larger value in the thermodynamic diagram represents a larger difference between the two samples.
Preferably, whether the product has defects, the corresponding defect size and the like are calculated through the process of setting a threshold value and the like.
By adopting the technical scheme, the defect detection can be realized by comparing the difference degree between two products output by the network, so that the time cost of artificial marking is saved for the production detection of the template product, the early preparation work of model training is reduced, the fast switching detection of the product in the working production is realized, and the production efficiency of the product is improved.
Drawings
Fig. 1 schematically shows a flow chart of the present invention.
Detailed Description
The embodiments of the invention will be described in detail below with reference to the drawings, but the invention can be implemented in many different ways as defined and covered by the claims.
When the deep learning algorithm in the prior art is applied to template defect detection, defect data needs to be collected and sorted and data labels need to be marked, so that high labor and time costs need to be invested. In order to overcome the problem, the invention constructs a double-input depth model comparison network according to the characteristics of the product template, and the scheme does not need to collect defect data samples and label the samples, thereby greatly reducing the production cost.
For the production of template products, the consistency between two good products is extremely high, and the good products and the defective products have certain difference in some places, and the inconsistency is the defect of the defective products. Therefore, the invention firstly collects good products, utilizes the traditional image method to manufacture defect samples and generate corresponding labels, then constructs a double-input convolution network, the first half part of the network is a twin network for extracting special certificates, the second half part of the network is a network characteristic fusion and outputs results, the output result is a thermodynamic diagram, the larger the numerical value of the thermodynamic diagram is, the larger the difference of the two samples at the position is, and finally, the traditional method is used for setting threshold values and other treatments to calculate whether the products have defects, the corresponding defect sizes and the like.
The specific operation method and steps of the invention are as follows:
step 1: collecting template good product data, fusing small defects and good products obtained from a defect library by a traditional method to obtain product defect data, and generating corresponding labeling data.
Step 2: and constructing a double-input network model, wherein the first half part of the network is a twin network based on a residual error network, and the second half part of the network is a feature fusion and extraction network.
And step 3: and putting the training data into the network, inputting the network into good products and corresponding generated defect samples, and outputting the network into a label to train the model.
And 4, step 4: the learned network model has the capability of comparing the difference between the two samples, so the network model is called a comparison network, the output of the network is the data difference degree of the two input samples, and whether the product has defects, the sizes of the positions of the defects and the like are obtained through a certain threshold value and a certain traditional algorithm.
By adopting the technical scheme, the defect detection can be realized by comparing the difference degree between two products output by the network, so that the time cost of artificial marking is saved for the production detection of the template product, the early preparation work of model training is reduced, the fast switching detection of the product in the working production is realized, and the production efficiency of the product is improved.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (4)
1. A template defect detection method is characterized by comprising the following steps:
step 1, collecting template good product data: acquiring product defect data by fusing small defects and good products acquired from a defect library, and generating corresponding marking data;
step 2, constructing a double-input network model: the first half of the network is a twin network based on a residual error network, and the second half is a feature fusion and extraction network;
step 3, putting training data into the network: inputting good products and corresponding defect sample generation and outputting as a label by a network, and training a model;
step 4, obtaining a comparison network through training and learning, wherein the comparison network has the capability of comparing the difference between two samples, and the output of the network is the data difference degree of the two input samples;
and 5, outputting the difference degree between the two products through a comparison network to further realize defect detection.
2. The template defect method of claim 1, wherein the defect detection comprises the presence and location, size, etc. of the defect.
3. The template defect detection method of claim 2, wherein the comparison network outputs a thermodynamic diagram, wherein a larger value in the thermodynamic diagram indicates a larger difference between the two samples.
4. The template defect detection method according to claim 3, wherein whether a product has defects, corresponding defect sizes and the like are calculated by setting a threshold and the like.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010483881.2A CN111709920A (en) | 2020-06-01 | 2020-06-01 | Template defect detection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010483881.2A CN111709920A (en) | 2020-06-01 | 2020-06-01 | Template defect detection method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111709920A true CN111709920A (en) | 2020-09-25 |
Family
ID=72538808
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010483881.2A Pending CN111709920A (en) | 2020-06-01 | 2020-06-01 | Template defect detection method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111709920A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112907543A (en) * | 2021-02-24 | 2021-06-04 | 胡志雄 | Product appearance defect detection method based on random defect model |
CN113160200A (en) * | 2021-04-30 | 2021-07-23 | 聚时科技(上海)有限公司 | Industrial image defect detection method and system based on multitask twin network |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109727246A (en) * | 2019-01-26 | 2019-05-07 | 福州大学 | Comparative learning image quality evaluation method based on twin network |
CN110222792A (en) * | 2019-06-20 | 2019-09-10 | 杭州电子科技大学 | A kind of label defects detection algorithm based on twin network |
CN110288013A (en) * | 2019-06-20 | 2019-09-27 | 杭州电子科技大学 | A kind of defective labels recognition methods based on block segmentation and the multiple twin convolutional neural networks of input |
CN110533654A (en) * | 2019-08-30 | 2019-12-03 | 北京明略软件系统有限公司 | The method for detecting abnormality and device of components |
CN110852373A (en) * | 2019-11-08 | 2020-02-28 | 深圳市深视创新科技有限公司 | Defect-free sample deep learning network training method based on vision |
US20200073850A1 (en) * | 2018-09-05 | 2020-03-05 | General Electric Company | Methods and systems for generating device-specific machine learning model |
CN111179251A (en) * | 2019-12-30 | 2020-05-19 | 上海交通大学 | Defect detection system and method based on twin neural network and by utilizing template comparison |
-
2020
- 2020-06-01 CN CN202010483881.2A patent/CN111709920A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20200073850A1 (en) * | 2018-09-05 | 2020-03-05 | General Electric Company | Methods and systems for generating device-specific machine learning model |
CN109727246A (en) * | 2019-01-26 | 2019-05-07 | 福州大学 | Comparative learning image quality evaluation method based on twin network |
CN110222792A (en) * | 2019-06-20 | 2019-09-10 | 杭州电子科技大学 | A kind of label defects detection algorithm based on twin network |
CN110288013A (en) * | 2019-06-20 | 2019-09-27 | 杭州电子科技大学 | A kind of defective labels recognition methods based on block segmentation and the multiple twin convolutional neural networks of input |
CN110533654A (en) * | 2019-08-30 | 2019-12-03 | 北京明略软件系统有限公司 | The method for detecting abnormality and device of components |
CN110852373A (en) * | 2019-11-08 | 2020-02-28 | 深圳市深视创新科技有限公司 | Defect-free sample deep learning network training method based on vision |
CN111179251A (en) * | 2019-12-30 | 2020-05-19 | 上海交通大学 | Defect detection system and method based on twin neural network and by utilizing template comparison |
Non-Patent Citations (1)
Title |
---|
吴运雄: "基于孪生网络的缺陷检测方法", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112907543A (en) * | 2021-02-24 | 2021-06-04 | 胡志雄 | Product appearance defect detection method based on random defect model |
CN112907543B (en) * | 2021-02-24 | 2024-03-26 | 胡志雄 | Product appearance defect detection method based on random defect model |
CN113160200A (en) * | 2021-04-30 | 2021-07-23 | 聚时科技(上海)有限公司 | Industrial image defect detection method and system based on multitask twin network |
CN113160200B (en) * | 2021-04-30 | 2024-04-12 | 聚时科技(上海)有限公司 | Industrial image defect detection method and system based on multi-task twin network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP6573226B2 (en) | DATA GENERATION DEVICE, DATA GENERATION METHOD, AND DATA GENERATION PROGRAM | |
CN110992317B (en) | PCB defect detection method based on semantic segmentation | |
CN111709920A (en) | Template defect detection method | |
CN107977686B (en) | Industrial X-ray image classification method | |
EP2478423A1 (en) | Supervised fault learning using rule-generated samples for machine condition monitoring | |
CN104268940B (en) | MEMS structure reconstruction and detection method based on CT scanned images | |
CN110660049A (en) | Tire defect detection method based on deep learning | |
CN113962951B (en) | Training method and device for detecting segmentation model, and target detection method and device | |
CN110852373A (en) | Defect-free sample deep learning network training method based on vision | |
CN101853393B (en) | Automatic production and automatic learning method of machine vision system detection algorithm | |
CN117056688A (en) | New material production data management system and method based on data analysis | |
CN116843107A (en) | Building information intelligent management system based on BIM technology | |
CN111913619A (en) | Fire-fighting data reading and interaction method based on BIM lightweight | |
CN111737067A (en) | Hard disk fault prediction model interpretation method and device | |
CN108895839B (en) | sintering machine air leakage fault online diagnosis method | |
CN115937555A (en) | Industrial defect detection algorithm based on standardized flow model | |
CN112598142B (en) | Wind turbine maintenance working quality inspection auxiliary method and system | |
CN114155914A (en) | Detection and correction system based on metagenome splicing error | |
CN110796187A (en) | Method and device for classifying defects | |
CN112255383A (en) | Water quality monitoring analysis system based on big data | |
KR101904643B1 (en) | Article generation method using the decision tree | |
CN112070712B (en) | Printing defect detection method based on self-encoder network | |
CN117349462B (en) | Remote sensing intelligent interpretation sample data set generation method | |
CN114037679A (en) | Product image defect detection method and device based on unsupervised feature combination | |
CN117572996B (en) | Digital twin display method, device and equipment for intelligent identification and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200925 |