CN109035226A - Mura defects detection method based on LSTM model - Google Patents
Mura defects detection method based on LSTM model Download PDFInfo
- Publication number
- CN109035226A CN109035226A CN201810763443.4A CN201810763443A CN109035226A CN 109035226 A CN109035226 A CN 109035226A CN 201810763443 A CN201810763443 A CN 201810763443A CN 109035226 A CN109035226 A CN 109035226A
- Authority
- CN
- China
- Prior art keywords
- lstm model
- sample
- region
- mura defects
- panel
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
- G01N2021/8887—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges based on image processing techniques
-
- 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]
Abstract
The invention discloses a kind of Mura defects detection methods based on LSTM model, comprising steps of panel 1) is divided into m region, acquire the qualified product sample and defective sample in m region；2) the qualified product sample and defective sample in each region form a sample sequence, and each sequence establishes a LSTM model；3) sample sequence in m region is subjected to LSTM model training respectively；4) picture in m region of measured panel is input to corresponding LSTM model, the classification results of m LSTM model output qualification or defect；5) classification results of m LSTM model are comprehensive, the classification results for judging whether there is any region are defect.The present invention establishes defects detection model by deep neural network, effectively evade traditional algorithm and carried out background segment, the limitation of parameter extraction, while having reduced cost of labor, can quickly export Mura defects, defect is accurately detected, the accuracy and efficiency of Mura defects detection is improved.
Description
Technical field
The present invention relates to the automation defects detection field of panel, especially deep learning and image procossing, in particular to
A kind of Mura defects detection method based on LSTM model.
Background technique
In AOI (Automatic Optic Inspection automatic optics inspection) defects detection, Mura defects detection
It is most important, it directly affects final LCD panel defect drop etc. and determines result.Mura defects have in irregular shape, size not
Uniformly, position be not fixed, the partly or wholly low characteristic of brightness irregularities, contrast, while by human eye perception and master
The limitation of sight factor is difficult that rapidly and accurately Mura is rapidly and accurately detected and evaluated, from figure 1 it appears that
Mura defects are very low relative to the contrast of background, and edge blurry is unclear, are visually not easy to recognize.Simultaneously with the update of processing procedure
It regenerates, various unpredictable Mura defects can also be supervened.Traditional algorithm forms reliable and stable detection not yet at present
System needs to seek a kind of more stable, higher Mura defects detection method of precision.
As shown in Fig. 2, existing Mura defects testing process is generally divided into image acquisition, brightness correction, image enhancement and lacks
Fall into 4 processes of detection.With the big picture of LCD, lightening, high-resolution development trend, the probability that Mura defects occur is significantly
Increase, while local environment can become increasingly complex, the fuzzy side of duplicate grain background, the uneven and defect of overall brightness
Edge, lower contrast make brightness correction and picture enhancing algorithm traditional in image procossing be difficult directly to examine Mura defects
Out.Therefore, it is necessary to which developing a set of new method deacclimatizes existing industry requirement.
Summary of the invention
In view of the deficiencies of the prior art, the object of the present invention is to provide a kind of Mura defects detection sides based on LSTM model
Method establishes defects detection model by deep neural network, has effectively evaded traditional algorithm and has carried out background segment, parameter extraction
Limitation, while cost of labor is reduced, Mura defects can be quickly exported, defect is accurately detected, Mura is improved and lacks
Fall into the accuracy and efficiency of detection.
To achieve the above object, a kind of Mura defects detection method based on LSTM model designed by the present invention, it is special
Different place is, includes the following steps:
1) panel is divided into m region, acquires the qualified product sample and defective sample in m region；
2) the qualified product sample and defective sample in each region form a sample sequence, and each sequence establishes one
LSTM model；
3) sample sequence in m region is subjected to LSTM model training respectively, LSTM model is extracted by deep learning and closed
Lattice product sample characteristics and defective sample characteristics；
4) picture in m region of measured panel is input to corresponding LSTM model, m LSTM model output it is qualified or
The classification results of person's defect；
5) classification results of m LSTM model are comprehensive, it is then that the classification results for judging whether there is any region, which are defect,
Judge that measured panel positions for Mura defects and by defective locations, otherwise judges measured panel for qualified product.
Preferably, the ratio of the qualified product sample and defective sample is (0.8~1.2): (0.8~1.2).
Preferably, m=a*a, a are natural number.Most preferably, a=10.
Preferably, the measured panel is with a batch of n panel product.
Preferably, the panel of the qualified product sample and defective sample and measured panel are that the panel of same model produces
Product.
The invention proposes the Mura defects detection methods of complete set, form sample sequence according to the panel production time,
Memory training is carried out by LSTM network model, after network model is trained sample sequence, allows rapid screening defect area
Domain and normal region, to carry out defects detection.In this way, the Mura defects of panel can be carried out with accurate detailed inspection
It surveys, can successfully evade the limitation of traditional algorithm, while greatly reducing human cost.
The present invention has the advantages that
1) panel zone is divided into multiple subregions, improves the sequence information of each subregion, to be more advantageous to
LSTM detection；
2) it is based on LSTM neural network, a large amount of sample is learnt, the interference of complex background is reduced, improves Mura
The accuracy of detection；
3) detection method is suitble to on-line checking, replaces personnel's detection process, to reduce cost of labor and time cost.
Detailed description of the invention
Fig. 1 is LCD master drawing.
Fig. 2 is the flow chart of existing Mura detection.
Fig. 3 is LSTM modular concept figure.
Fig. 4 is that the present invention is based on the flow charts of the Mura defects detection method of LSTM model.
Fig. 5 is sample sequence structural schematic diagram.
Specific embodiment
Below in conjunction with the drawings and specific embodiments, the present invention is described in further detail.
The present invention detects Mura defects using LSTM model, and LSTM model is a kind of RNN of particular form
(Recurrent neural network, Recognition with Recurrent Neural Network), and RNN is a series of nerve nets for being capable of handling sequence data
The general name of network.
Panel product is produced according to batch at present, and the product of all same batches can regard a sequence as
Sample, the difference between normal sample can be smaller, and in a sequence per adjacent sample between difference can be smaller, once occur
Obvious difference, then maximum probability is to produce Mura defects.
LSTM model avoids long-term Dependence Problem by design deliberately.Remember that long-term information is LSTM in practice
Default behavior.The key of LSTM model is exactly location mode, and horizontal line is on the diagram just through operation.Location mode, which is similar to, to be passed
Send band.It is directly run on entire chain, only some a small amount of linear reciprocals.Information is spread above to be remained unchanged and can hold very much
Easily.LSTM model has through the well-designed structure for being referred to as " door " energy for removing or increasing information to location mode
Power.Door is a kind of method for allowing information selecting type to pass through, and is multiplied comprising a Sigmoid neural net layer and a pointwise
Method operation.Numerical value between Sigmoid layers of output 0 to 1 describes that how many amount of each part can pass through.0 representative mustn't be any
Amount passes through, and 1 just refers to that permission any amount passes through.LSTM model is gathered around there are three door, to protect and control unit state, structural principle
As shown in Figure 3.
The LSTM model that the present invention establishes is found mutual between normal sample by the panel sample of one sequence of study
Relationship, and control the information for needing to retain or delete by forgeing door, by the corresponding output normal sample of out gate and
Mura defects sample.
As shown in figure 4, the specific steps of the Mura defects detection method proposed by the present invention based on LSTM model include:
1) panel is divided into m=a*a region, acquires the qualified product sample and defective sample in m region.This implementation
M=100, a=10 are taken in example.Each panel sample is divided into 10 × 10 totally 100 regions；And it is labeled as 1~100, such as
5-a in Fig. 5 is denoted as n with a batch of number of panels.
2) the qualified product sample and defective sample in each region form a sample sequence, as shown in 5-b in Fig. 5,
Each sequence establishes a LSTM model.
3) sample sequence in m region is subjected to LSTM model training respectively, LSTM model is extracted by deep learning and closed
Lattice product sample characteristics and defective sample characteristics.Training sample generally takes 500~2000 under normal conditions, qualified product sample and
The ratio of defective sample is 1:1 or so.
100 sequence samples carry out LSTM model training respectively, and each sample controls input, forgetting and defeated by door
Out.Door is actually one layer of full articulamentum, and input is feature vector x, and output is the real vector between one 0 to 1.Formula
Are as follows:
G (x)=σ (Wx+b)
In formula, W: weight matrix is controlled whether to retain or be forgotten by setting 0 or 1；B: the bias term of door, σ are indicated
Sigmoid function is used as the threshold function table of neural network, by variable mappings to 0, between 1,
LSTM is according to the information of the sample sequence learnt, and by the control of input gate, each unit can be according to upper one
To calculate location mode currently entered, it is more that input gate determines that the input of current time network has for secondary output and this input
It is saved in location mode less；By forgeing the control of door, determine last moment location mode how many when remaining into current
It carves；How many is output to the current output value of LSTM to output door control unit state.It is very bright due to having with a batch of sample
Aobvious sequence information, LSTM can learn previous useful information, how many information is eventually controlled by out gate can be at
For the output valve of current LSTM；The feature that LSTM is extracted according to oneself is established the characterization rules of qualified product and defective, is then led to
It crosses three doors to go to control the reserved and output quantity of the feature of each sample extraction, to establish more preferably model, realizes that Mura is lacked
Fall into detection.
4) picture in m region of measured panel is input to corresponding LSTM model, m LSTM model output it is qualified or
The classification results of person's defect.In above-mentioned LSTM learning process, 100 LSTM networks can be generated, when testing, equally will
Panel zone is divided into 10 × 10 (100 regions), each region is detected using corresponding model.
5) classification results of m LSTM model are comprehensive, it is then that the classification results for judging whether there is any region, which are defect,
Judge that measured panel positions for Mura defects and by defective locations, otherwise judges measured panel for qualified product.At 100 of output
As a result in, as long as having a result is Mura defects, which is judged as Mura defects, while defective locations can be quasi-
Determine position.
The above is only the preferred embodiment of the present invention, it is noted that those skilled in the art are come
It says, without departing from the principle of the present invention, can be devised by several improvement, these improvement also should be regarded as guarantor of the invention
Protect range.
The content that this specification is not described in detail belongs to the prior art well known to professional and technical personnel in the field.
Claims (6)
1. a kind of Mura defects detection method based on LSTM model, characterized by the following steps:
1) panel is divided into m region, acquires the qualified product sample and defective sample in m region；
2) the qualified product sample and defective sample in each region form a sample sequence, and each sequence establishes a LSTM
Model；
3) sample sequence in m region is subjected to LSTM model training respectively, LSTM model extracts qualified product by deep learning
Sample characteristics and defective sample characteristics；
4) picture in m region of measured panel is input to corresponding LSTM model, m LSTM model output is qualified or scarce
Sunken classification results；
5) classification results of m LSTM model are comprehensive, it is to judge that the classification results for judging whether there is any region, which are defect,
Measured panel is Mura defects and positions defective locations, otherwise judges measured panel for qualified product.
2. the Mura defects detection method according to claim 1 based on LSTM model, it is characterised in that: the qualified product
The ratio of sample and defective sample is (0.8~1.2): (0.8~1.2).
3. the Mura defects detection method according to claim 1 based on LSTM model, it is characterised in that: m=a*a, a are
Natural number.
4. the Mura defects detection method according to claim 1 based on LSTM model, it is characterised in that: the tested surface
The panel product that plate is same batch n.
5. the Mura defects detection method according to claim 1 based on LSTM model, it is characterised in that: the qualified product
The panel of sample and defective sample and measured panel are the panel product of same model.
6. the Mura defects detection method according to claim 1 based on LSTM model, it is characterised in that: a=10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810763443.4A CN109035226B (en) | 2018-07-12 | 2018-07-12 | Mura defect detection method based on LSTM model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810763443.4A CN109035226B (en) | 2018-07-12 | 2018-07-12 | Mura defect detection method based on LSTM model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109035226A true CN109035226A (en) | 2018-12-18 |
| CN109035226B CN109035226B (en) | 2021-11-23 |
Family
ID=64641301
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810763443.4A Active CN109035226B (en) | 2018-07-12 | 2018-07-12 | Mura defect detection method based on LSTM model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109035226B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109829914A (en) * | 2019-02-26 | 2019-05-31 | 视睿（杭州）信息科技有限公司 | The method and apparatus of testing product defect |
| CN109856157A (en) * | 2019-01-24 | 2019-06-07 | 武汉精立电子技术有限公司 | It is a kind of based on the LCD panel defect inspection method adaptively focused |
| CN110174409A (en) * | 2019-06-14 | 2019-08-27 | 北京科技大学 | A kind of cut deal periodicity defect control method based on real-time detection result |
| CN110232406A (en) * | 2019-05-28 | 2019-09-13 | 厦门大学 | A kind of liquid crystal display panel CF image identification method based on statistical learning |
| CN110322429A (en) * | 2019-05-09 | 2019-10-11 | 中南大学 | A kind of cellular composite material defect classification method based on deep learning |
| CN111191739A (en) * | 2020-01-09 | 2020-05-22 | 电子科技大学 | Wall surface defect detection method based on attention mechanism |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104318256A (en) * | 2014-11-17 | 2015-01-28 | 上海埃蒙特自动化系统有限公司 | Method for achieving automatic detection and classification of glass defects on basis of computer software |
| CN107423756A (en) * | 2017-07-05 | 2017-12-01 | 武汉科恩斯医疗科技有限公司 | Nuclear magnetic resonance image sequence sorting technique based on depth convolutional neural networks combination shot and long term memory models |
| US20170372475A1 (en) * | 2016-06-23 | 2017-12-28 | Siemens Healthcare Gmbh | Method and System for Vascular Disease Detection Using Recurrent Neural Networks |
| CN108197648A (en) * | 2017-12-28 | 2018-06-22 | 华中科技大学 | A kind of Fault Diagnosis Method of Hydro-generating Unit and system based on LSTM deep learning models |
| CN108255656A (en) * | 2018-02-28 | 2018-07-06 | 湖州师范学院 | A kind of fault detection method applied to batch process |
-
2018
- 2018-07-12 CN CN201810763443.4A patent/CN109035226B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104318256A (en) * | 2014-11-17 | 2015-01-28 | 上海埃蒙特自动化系统有限公司 | Method for achieving automatic detection and classification of glass defects on basis of computer software |
| US20170372475A1 (en) * | 2016-06-23 | 2017-12-28 | Siemens Healthcare Gmbh | Method and System for Vascular Disease Detection Using Recurrent Neural Networks |
| CN107423756A (en) * | 2017-07-05 | 2017-12-01 | 武汉科恩斯医疗科技有限公司 | Nuclear magnetic resonance image sequence sorting technique based on depth convolutional neural networks combination shot and long term memory models |
| CN108197648A (en) * | 2017-12-28 | 2018-06-22 | 华中科技大学 | A kind of Fault Diagnosis Method of Hydro-generating Unit and system based on LSTM deep learning models |
| CN108255656A (en) * | 2018-02-28 | 2018-07-06 | 湖州师范学院 | A kind of fault detection method applied to batch process |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109856157A (en) * | 2019-01-24 | 2019-06-07 | 武汉精立电子技术有限公司 | It is a kind of based on the LCD panel defect inspection method adaptively focused |
| CN109829914A (en) * | 2019-02-26 | 2019-05-31 | 视睿（杭州）信息科技有限公司 | The method and apparatus of testing product defect |
| CN110322429A (en) * | 2019-05-09 | 2019-10-11 | 中南大学 | A kind of cellular composite material defect classification method based on deep learning |
| CN110232406A (en) * | 2019-05-28 | 2019-09-13 | 厦门大学 | A kind of liquid crystal display panel CF image identification method based on statistical learning |
| CN110232406B (en) * | 2019-05-28 | 2021-07-06 | 厦门大学 | Liquid crystal panel CF image identification method based on statistical learning |
| CN110174409A (en) * | 2019-06-14 | 2019-08-27 | 北京科技大学 | A kind of cut deal periodicity defect control method based on real-time detection result |
| CN111191739A (en) * | 2020-01-09 | 2020-05-22 | 电子科技大学 | Wall surface defect detection method based on attention mechanism |
| CN111191739B (en) * | 2020-01-09 | 2022-09-27 | 电子科技大学 | Wall surface defect detection method based on attention mechanism |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109035226B (en) | 2021-11-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109035226A (en) | Mura defects detection method based on LSTM model | |
| CN110658202B (en) | Industrial component appearance defect detection method based on deep learning | |
| CN107563999A (en) | A kind of chip defect recognition methods based on convolutional neural networks | |
| CN110717481B (en) | Method for realizing face detection by using cascaded convolutional neural network | |
| CN105118044B (en) | A kind of wheel shape cast article defect automatic testing method | |
| CN104458755B (en) | Multi-type material surface defect detection method based on machine vision | |
| CN112183667B (en) | Insulator fault detection method in cooperation with deep learning | |
| CN110689531A (en) | Automatic power transmission line machine inspection image defect identification method based on yolo | |
| CN109308447A (en) | The method of equipment operating parameter and operating status is automatically extracted in remote monitoriong of electric power | |
| CN110274908A (en) | Flaw detection apparatus, defect detecting method and computer readable recording medium | |
| CN108171707A (en) | A kind of Mura defects level evaluation method and device based on deep learning | |
| CN106875373A (en) | Mobile phone screen MURA defect inspection methods based on convolutional neural networks pruning algorithms | |
| CN109376792A (en) | Photovoltaic cell open defect classification method based on multichannel residual error neural network | |
| CN107016405A (en) | A kind of insect image classification method based on classification prediction convolutional neural networks | |
| CN107133943A (en) | A kind of visible detection method of stockbridge damper defects detection | |
| CN106845556A (en) | A kind of fabric defect detection method based on convolutional neural networks | |
| CN109087277B (en) | Method for measuring PM2.5 of fine air particles | |
| CN106250911B (en) | A kind of picture classification method based on convolutional neural networks | |
| CN106650823A (en) | Probability extreme learning machine integration-based foam nickel surface defect classification method | |
| CN109215015A (en) | A kind of online visible detection method of silk cocoon based on convolutional neural networks | |
| CN112258490A (en) | Low-emissivity coating intelligent damage detection method based on optical and infrared image fusion | |
| CN108960413A (en) | A kind of depth convolutional neural networks method applied to screw surface defects detection | |
| CN108764243A (en) | A kind of image processing method and device | |
| CN115797694A (en) | Display panel microdefect classification method based on multi-scale twin neural network | |
| CN110567967B (en) | Display panel detection method, system, terminal device and computer readable 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |