CN110223269A - A kind of FPC defect inspection method and device - Google Patents
A kind of FPC defect inspection method and device Download PDFInfo
- Publication number
- CN110223269A CN110223269A CN201910334223.4A CN201910334223A CN110223269A CN 110223269 A CN110223269 A CN 110223269A CN 201910334223 A CN201910334223 A CN 201910334223A CN 110223269 A CN110223269 A CN 110223269A
- Authority
- CN
- China
- Prior art keywords
- defect
- image
- classification
- defects
- disaggregated model
- 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
Classifications
-
- 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
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- 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
- G01N2021/8854—Grading and classifying of flaws
-
- 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
- G06T2207/30141—Printed circuit board [PCB]
Abstract
A kind of FPC defect inspection method and device, comprising: image is split according to the structure of product, finds out the defects of image block；The defects of image block image is cut down, and zooms to predefined size；According to defect rank evaluation of classification strategy, using there are the convolutional neural networks of supervision to be trained the defects of defect image, and generates disaggregated model and classify to defect, defect is detected according to the classification of defect.The inconsistency that the embodiment of the present application makes complicated background become simply to eliminate optical environment influences, and reduces the influence that background classifies to defect；According to defect rank evaluation of classification strategy, using there is the convolutional neural networks of supervision to be trained the defects of defect image, disaggregated model is generated to classify to defect, defect is detected using detection method according to defect classification, reduce sample usage quantity, the accuracy for improving detection and classification, different Flaw discriminations is come out, to promote the recall rate of FPC product.
Description
Technical field
This application involves this application involves flexible circuit board more particularly to a kind of FPC defect inspection methods and device.
Background technique
Flexible printed circuit board (The Flexible Printed Circuit board, abbreviation FPC) is with flexible
Printed circuit made of insulating substrate.The manufacturing industry of FPC at home has been popularized at present, but in industrial manufacturing process often
It will appear defective defective products.
The accuracy of defect classification will directly affect the recall rate of FPC product.Existing defect sorting technique includes using
BP neural network is classified for the first time, to can be carried out again with defect classification SVM (Support Vector Machine, support to
Amount machine) secondary classification.Since existing sorting technique use is machine learning and BP neural network algorithm, BP neural network algorithm
The network number of plies is less in practical applications, feature that can not be more advanced in abstract image, therefore is difficult the defect to background complexity
Classified and is difficult to be suitable for the more classification task of defect number.And existing defect sorting technique is to the demand of sample
Greatly, the data of thousands of defect sample is needed to be trained.
Summary of the invention
The application provides a kind of FPC defect inspection method and device.
According to a first aspect of the present application, the application provides a kind of FPC defect inspection method,
Cutting image step: being split image according to the structure of product, detects, looks for the image block after segmentation
The defects of image block out；
Cutting defect step: the defects of image block image is cut down, and zooms to predefined size；
It generates disaggregated model step: defect rank evaluation of classification strategy being arranged according to defect Stringency, according to defect etc.
Grade classification Evaluation Strategy, using there is the convolutional neural networks of supervision to be trained the defects of defect image, and generates classification
Model；
Defect classifying step: classified using disaggregated model to defect in defect image；
Defect detection procedure: defect is detected according to the classification of defect.
According to a second aspect of the present application, the application provides a kind of FPC defect detecting device, comprising:
Cutting image module examines the image block after segmentation for being split according to the structure of product to image
It surveys, finds out the defects of image block；
Defect module is cut, for cutting down the defects of image block image, and zooms to predefined size；
Disaggregated model module is generated, for defect rank evaluation of classification strategy to be arranged according to defect Stringency, according to scarce
Grade separation Evaluation Strategy is fallen into, using there are the convolutional neural networks of supervision to be trained the defects of defect image, and is generated
Disaggregated model；
Defect categorization module, for being classified using disaggregated model to defect in defect image；
Defects detection module, for being detected according to the classification of defect to defect.
According to the third aspect of the application, the application provides a kind of FPC defect detecting device, comprising:
Memory, for storing computer program；
Processor, for by executing the step of computer program is to realize above-mentioned FPC defects detection.
Due to using above technical scheme, the beneficial effect for having the application is:
Since the embodiment of the present application includes being split product image according to structural region, the defects of image block is schemed
As cutting down, predefined size is zoomed to, the inconsistency for making complicated background become simply to eliminate optical environment influences, and subtracts
The influence that small background classifies to defect；Defect rank evaluation of classification strategy is set according to defect Stringency, using there is supervision
Convolutional neural networks the defects of defect image is trained, and generate disaggregated model and defect in defect image divided
Class detects defect using different detection methods according to different defect classifications, reduces sample usage quantity, improves
The accuracy of detection and classification, different Flaw discriminations is come out, to promote the recall rate of FPC product.
Detailed description of the invention
Fig. 1 is the flow chart of the FPC defect inspection method of the application in one embodiment；
Fig. 2 is the flow chart of the FPC defect inspection method of the application in another embodiment；
Fig. 3 is that the FPC defect inspection method of the application exports the flow chart of disaggregated model in one embodiment；
Fig. 4 is the schematic diagram of FPC product ACF disconnection defect；
Fig. 5 is the schematic diagram of FPC product ACF foreign matter defect；
Fig. 6 is the schematic diagram that FPC product ACF scratches defect；
Fig. 7 is the program module schematic diagram of the FPC defect detecting device of the application in one embodiment；
Fig. 8 is the program module schematic diagram of the FPC defect detecting device of the application in another embodiment.
Specific embodiment
Below by specific embodiment combination attached drawing, invention is further described in detail.The application can be with a variety of
Different forms is realized, however it is not limited to embodiment described in the present embodiment.The purpose of following specific embodiments is provided
It is easy for becoming apparent from present disclosure thorough explanation, wherein the words of the indicating positions such as upper and lower, left and right is only needle
To shown structure in respective figure for position.
However, those skilled in the art may be aware that one or more detail description can be by
Omit, or can also adopt with other methods, component or material.In some instances, some embodiments are not described
Or it is not described later in detail.
It is herein component institute serialization number itself, such as " first ", " second " etc., is only used for distinguishing described object,
Without any sequence or art-recognized meanings.
In addition, technical characteristic described herein, technical solution can also be in one or more embodiments arbitrarily to close
Suitable mode combines.For those skilled in the art, it should be readily appreciated that method related with embodiment provided herein
Step or operation order can also change.Therefore, any sequence in drawings and examples is merely illustrative purposes, not secretly
Show requirement in a certain order, is required unless expressly stated according to a certain sequence.
Embodiment one:
As shown in Figure 1, the FPC defect inspection method of the application, a kind of embodiment, comprising the following steps:
Step 101: cutting image step is split image according to the structure of product, to the image block after segmentation into
Row detection, finds out the defects of image block.
Step 102: cutting defect step cuts down the defects of image block image, and zoom to predefined size.
Step 103: generating disaggregated model step, defect rank evaluation of classification strategy, root are arranged according to defect Stringency
According to defect rank evaluation of classification strategy, using there is the convolutional neural networks of supervision to be trained the defects of defect image, and
Generate disaggregated model.Wherein, defect rank evaluation of classification strategy can predefine.
Further, defect rank evaluation of classification strategy, can specifically include:
Various types of defects are successively sorted from high to low according to Stringency；
Allow to rank low defect and be categorized into and rank high defect, but ranks high defect cannot to be categorized into ranking low
Defect.
By taking golden finger (Anisotropic Conductive Film, golden finger) defect as an example, ACF foreign matter, ACF broken string
It is on optical imagery and its similar with ACF removing defect, it is accurately identified although deep learning can reach it, it can not
Guarantee that 100% classification is correct.Therefore, the application designs a relatively reasonable evaluation of classification strategy:
Firstly, ACF foreign matter, ACF broken string and the examination criteria of ACF removing defect are different, and defect severity is not yet
Together, it can be arranged according to examination criteria Stringency is descending are as follows: ACF broken string > ACF removing > ACF foreign matter, if Fig. 4 is ACF
Broken string, Fig. 5 are ACF foreign matter, and Fig. 6 is ACF scuffing.
Defect rank sequence: defects detection is successively sorted from high to low according to Stringency.
It provides the mistake allowed: allowing will test not stringent defect and be categorized into the stringent defect of detection, such as can incite somebody to action
The classification of ACF foreign matter becomes ACF and breaks, but ACF cannot break and be categorized into ACF foreign matter.
Further, step 103 can specifically include:
Step 1031: training sample database is built, defect sample is added, data is trained, the data after making training
Classification performance evaluation index reaches preset first threshold；
Step 1032: input test data are tested, if the classification performance evaluation index of test data reaches preset
Second threshold then exports disaggregated model.
Further, step 103 can also include:
Step 1033: if the classification performance evaluation index of test data is not up to preset second threshold, finding out misclassification
Sample is arranged to participate in training every time, is re-added to training sample database.
Step 104: defect classifying step classifies to defect in defect image using disaggregated model.
Step 105: defect detection procedure detects defect according to the classification of defect.Wherein according to the classification of defect
The specific method detected to defect is identical as existing classification and Detection method.
Further, step 105 can specifically include:
Step 1051: identifying defect according to the examination criteria of different defects, detection threshold value is set；
Step 1052: the product judgement that defect is not up to detection threshold value requirement is become into non-defective unit.
Defect is identified according to the examination criteria of different defects, and suitable detection threshold value is set, to not being defect or defect
The product judgement of not up to standard becomes non-defective unit, promotes the recall rate of FPC product.
As shown in Fig. 2, the FPC defect inspection method of the application, a kind of specific embodiment, comprising the following steps:
Step 201: shooting product image.Image is shot using CCD area array cameras, different defects is used different
Brightness is shot.
Step 202: image is cut by image block according to the structural region of product.Such as by ACF, GND of image
The regions such as (Ground, ground wire), pad, silverskin, guarantor's glue are respectively cut.
Step 203: the image block cut down being detected, defective image block is found out.
Step 204: defective image block being handled, and is expanded along defect boundary, defect part is then cut out.
In the present embodiment, 5 pixels can be expanded along defect boundary.
Step 205: defective image is uniformly zoomed into predetermined size.In the present embodiment, it specifically may be scalable to
The size of 80 pixel *, 80 pixel.
Step 206: rejecting abnormalities data.Defective data is pre-processed, such as image denoising, deblurring, with rejecting abnormalities
Data.
Step 207: use trained CNN convolutional neural networks disaggregated model, according to structural region to defect type into
Row classification, can specifically classify according to defect Stringency.
Step 208: output category result.
Step 209: defect is filtered.I.e. according to the examination criteria in different structure region, suitable detection threshold is set
Value will not be that defect or defect do not meet the product judgement of detection threshold value as non-defective unit.
Step 210: output test result.Testing result includes non-defective unit or two kinds of defective products.
As shown in figure 3, the method for the generation disaggregated model of the application, can specifically include following steps:
Step 301: input defect rank sequencing table.Specifically can according to the structural region of product, as product ACF, GND,
Pad, protects the regions such as glue at silverskin, and the various defects of each structural region are ranked up from high to low by defect Stringency.
Step 302: the error criteria that regulation defect classification allows.Allow will test not stringent defect and is categorized into detection sternly
The defect of lattice, such as ACF foreign matter can be classified and be broken as ACF, but ACF cannot be broken and be categorized into ACF foreign matter.
Step 303: building training sample database.
Step 304: judging ROC (receiver operator characteristic, Receiver operating curve)
Whether reach preset first threshold, if going to step 305, otherwise goes to step 307.
Classification performance evaluation index selects ROC, and (receiver operator characteristic, subject work special
Levy curve), the value of ROC, which can according to need, to be configured, and in the present embodiment, the first threshold of ROC is set as 0.99.
Step 305: input test is data.
Step 306: judging whether test data ROC reaches preset second threshold.If going to step 311, otherwise turn to walk
Rapid 308.In the present embodiment, the second threshold of ROC is set as 0.98.
Step 307: being re-added to defect sample, go to step 303.
Step 308: finding out the sample of classification error.
Step 309: setting the sample of classification error to participate in training every time.
Step 310: being re-added to defect sample, go to step 303.
Step 311: export disaggregated model.
Embodiment two:
As shown in Figure 7, Figure 8, the FPC defect detecting device of the application, a kind of embodiment, including cutting image mould
Block, generates disaggregated model module, defect categorization module and defects detection module at cutting defect module.
Cutting image module examines the image block after segmentation for being split according to the structure of product to image
It surveys, finds out the defects of image block；
Defect module is cut, for cutting down the defects of image block image, and zooms to predefined size；
Disaggregated model module is generated, for defect rank evaluation of classification strategy to be arranged according to defect Stringency, according to scarce
Grade separation Evaluation Strategy is fallen into, using there are the convolutional neural networks of supervision to be trained the defects of defect image, and is generated
Disaggregated model；
Defect categorization module, for being classified using disaggregated model to defect in defect image；
Defects detection module, for being detected according to the classification of defect to defect.
Further, generating disaggregated model module may include training unit and detection unit.
Training unit adds defect sample, is trained to data, make the number after training for building training sample database
According to classification performance evaluation index reach preset first threshold；
Detection unit is tested for input test data, if the classification performance evaluation index of test data reaches pre-
If second threshold, then export disaggregated model.
Further, generating disaggregated model module can also include reset cell.
Reset cell, for finding out when the classification performance evaluation index of test data is not up to preset second threshold
The sample of misclassification is arranged to participate in training every time, is re-added to training sample database.
Further, defects detection module can also include setting unit and filter element.
For identifying defect according to the examination criteria of different defects, and detection threshold value is arranged in setting unit；
Filter element, the product judgement for defect to be not up to detection threshold value requirement become non-defective unit.
Embodiment three:
A kind of FPC defect detecting device of the application, a kind of embodiment, comprising:
Memory, for storing computer program；
Processor, for by executing the computer program to realize the defect detecting device side FPC in embodiment one
The step of method.
It will be understood by those skilled in the art that all or part of the steps of various methods can pass through in above embodiment
Program instructs related hardware to complete, which can be stored in a computer readable storage medium, storage medium can wrap
It includes: read-only memory, random access memory, disk or CD etc..The above content is combination specific embodiments to the application institute
The further description of work is, and it cannot be said that the specific implementation of the application is only limited to these instructions.Skill affiliated for the application
For the those of ordinary skill in art field, without departing from the concept of this application, can also make it is several it is simple deduction or
Replacement.
Claims (10)
1. a kind of FPC defect inspection method characterized by comprising
Cutting image step: being split image according to the structure of product, detects to the image block after segmentation, finds out figure
As the defects of block；
Cutting defect step: the defects of image block image is cut down, and zooms to predefined size；
It generates disaggregated model step: defect rank evaluation of classification strategy being arranged according to defect Stringency, according to defect rank point
Class Evaluation Strategy using there is the convolutional neural networks of supervision to be trained the defects of defect image, and generates disaggregated model；
Defect classifying step: classified using disaggregated model to defect in defect image；
Defect detection procedure: defect is detected according to the classification of defect.
2. the method as described in claim 1, which is characterized in that the defect rank evaluation of classification strategy specifically includes:
Various types of defects are successively sorted from high to low according to Stringency；
Allow to rank low defect and be categorized into and rank high defect, but ranks high defect to be categorized into that ranking is low to be lacked
It falls into.
3. method according to claim 2, which is characterized in that the generation disaggregated model step, comprising:
Training sample database is built, defect sample is added, data is trained, the classification performance evaluation of the data after making training refers to
Mark reaches preset first threshold；
Input test data are tested, if the classification performance evaluation index of test data reaches preset second threshold, are led
Disaggregated model out.
4. method as claimed in claim 3, which is characterized in that the generation disaggregated model step, further includes:
If the classification performance evaluation index of test data is not up to preset second threshold, the sample of misclassification is found out, is arranged to every
Secondary participation training, is re-added to training sample database.
5. method according to any one of claims 1 to 4, which is characterized in that the defect detection procedure further include:
Defect is identified according to the examination criteria of different defects, and detection threshold value is set；
The product judgement that defect is not up to detection threshold value requirement is become into non-defective unit.
6. a kind of FPC defect detecting device characterized by comprising
Cutting image module detects the image block after segmentation, looks for for being split according to the structure of product to image
The defects of image block out；
Defect module is cut, for cutting down the defects of image block image, and zooms to predefined size；
Disaggregated model module is generated, for defect rank evaluation of classification strategy to be arranged according to defect Stringency, according to defect etc.
Grade classification Evaluation Strategy, using there is the convolutional neural networks of supervision to be trained the defects of defect image, and generates classification
Model；
Defect categorization module, for being classified using disaggregated model to defect in defect image；
Defects detection module, for being detected according to the classification of defect to defect.
7. device as claimed in claim 6, which is characterized in that the generation disaggregated model module, comprising:
Training unit adds defect sample, is trained to data, make the data after training for building training sample database
Classification performance evaluation index reaches preset first threshold；
Detection unit is tested for input test data, if the classification performance evaluation index of test data reaches preset
Second threshold then exports disaggregated model.
8. device as claimed in claim 7, which is characterized in that the generation disaggregated model module, further includes:
Reset cell, for finding out misclassification when the classification performance evaluation index of test data is not up to preset second threshold
Sample, be arranged to every time participate in training, be re-added to training sample database.
9. the device as described in any one of claim 6 to 8, which is characterized in that the defects detection module further include:
For identifying defect according to the examination criteria of different defects, and detection threshold value is arranged in setting unit；
Filter element, the product judgement for defect to be not up to detection threshold value requirement become non-defective unit.
10. a kind of FPC defect detecting device characterized by comprising
Memory, for storing computer program；
Processor, for by executing the computer program to realize FPC defect according to any one of claims 1 to 5
The step of detection.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910334223.4A CN110223269A (en) | 2019-04-24 | 2019-04-24 | A kind of FPC defect inspection method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910334223.4A CN110223269A (en) | 2019-04-24 | 2019-04-24 | A kind of FPC defect inspection method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110223269A true CN110223269A (en) | 2019-09-10 |
Family
ID=67819992
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910334223.4A Pending CN110223269A (en) | 2019-04-24 | 2019-04-24 | A kind of FPC defect inspection method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110223269A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111060520A (en) * | 2019-12-30 | 2020-04-24 | 歌尔股份有限公司 | Product defect detection method, device and system |
| CN111127571A (en) * | 2019-12-03 | 2020-05-08 | 歌尔股份有限公司 | Small sample defect classification method and device |
| CN111161246A (en) * | 2019-12-30 | 2020-05-15 | 歌尔股份有限公司 | Product defect detection method, device and system |
| CN111179253A (en) * | 2019-12-30 | 2020-05-19 | 歌尔股份有限公司 | Product defect detection method, device and system |
| CN112184665A (en) * | 2020-09-27 | 2021-01-05 | 苏州嘉展科技有限公司 | Artificial intelligence defect detecting system applied to paper-plastic industry |
| CN112934741A (en) * | 2021-01-27 | 2021-06-11 | 扬州虹扬科技发展有限公司 | Diode screening method and system |
| TWI775140B (en) * | 2019-09-19 | 2022-08-21 | 日商斯庫林集團股份有限公司 | Learning apparatus, inspection apparatus, learning method and inspection method |
| TWI810016B (en) * | 2022-08-11 | 2023-07-21 | 易華電子股份有限公司 | System for inspecting soft circuit board |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140185919A1 (en) * | 2013-01-02 | 2014-07-03 | Kla-Tencor Corporation | Detecting Defects on a Wafer |
| CN109215009A (en) * | 2017-06-29 | 2019-01-15 | 上海金艺检测技术有限公司 | Continuous casting billet surface image defect inspection method based on depth convolutional neural networks |
| CN109239102A (en) * | 2018-08-21 | 2019-01-18 | 南京理工大学 | A kind of flexible circuit board open defect detection method based on CNN |
-
2019
- 2019-04-24 CN CN201910334223.4A patent/CN110223269A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140185919A1 (en) * | 2013-01-02 | 2014-07-03 | Kla-Tencor Corporation | Detecting Defects on a Wafer |
| CN109215009A (en) * | 2017-06-29 | 2019-01-15 | 上海金艺检测技术有限公司 | Continuous casting billet surface image defect inspection method based on depth convolutional neural networks |
| CN109239102A (en) * | 2018-08-21 | 2019-01-18 | 南京理工大学 | A kind of flexible circuit board open defect detection method based on CNN |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI775140B (en) * | 2019-09-19 | 2022-08-21 | 日商斯庫林集團股份有限公司 | Learning apparatus, inspection apparatus, learning method and inspection method |
| CN111127571A (en) * | 2019-12-03 | 2020-05-08 | 歌尔股份有限公司 | Small sample defect classification method and device |
| US11836967B2 (en) | 2019-12-03 | 2023-12-05 | Goertek, Inc. | Method and device for small sample defect classification and computing equipment |
| CN111127571B (en) * | 2019-12-03 | 2023-12-29 | 歌尔股份有限公司 | Small sample defect classification method and device |
| CN111060520A (en) * | 2019-12-30 | 2020-04-24 | 歌尔股份有限公司 | Product defect detection method, device and system |
| CN111161246A (en) * | 2019-12-30 | 2020-05-15 | 歌尔股份有限公司 | Product defect detection method, device and system |
| CN111179253A (en) * | 2019-12-30 | 2020-05-19 | 歌尔股份有限公司 | Product defect detection method, device and system |
| CN111179253B (en) * | 2019-12-30 | 2023-11-24 | 歌尔股份有限公司 | Product defect detection method, device and system |
| CN112184665A (en) * | 2020-09-27 | 2021-01-05 | 苏州嘉展科技有限公司 | Artificial intelligence defect detecting system applied to paper-plastic industry |
| CN112934741A (en) * | 2021-01-27 | 2021-06-11 | 扬州虹扬科技发展有限公司 | Diode screening method and system |
| TWI810016B (en) * | 2022-08-11 | 2023-07-21 | 易華電子股份有限公司 | System for inspecting soft circuit board |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110223269A (en) | A kind of FPC defect inspection method and device | |
| JP4552749B2 (en) | Inspection standard setting device and method, and process inspection device | |
| CN109959661A (en) | Automatic optical detection method, equipment and its deep learning system | |
| CN109919908B (en) | Method and device for detecting defects of light-emitting diode chip | |
| CN108932523B (en) | Image classification and classified data creation system and method, and storage medium | |
| JP5718781B2 (en) | Image classification apparatus and image classification method | |
| US11423531B2 (en) | Image-recognition apparatus, image-recognition method, and non-transitory computer-readable storage medium thereof | |
| CN111242899B (en) | Image-based flaw detection method and computer-readable storage medium | |
| JP2006098155A (en) | Method and device for inspection | |
| US11521015B2 (en) | Image-recognition apparatus, image-recognition method, and non-transitory computer-readable storage medium thereof | |
| KR20210038303A (en) | System and method of classifying manufactured products | |
| WO2022130762A1 (en) | Defect inspection device, defect inspection method, and manufacturing method | |
| JP2017054331A (en) | Image sorting method, method for configuring sorter, and image sorting device | |
| CN115546211B (en) | Welding spot defect classification method, terminal and computer storage medium | |
| KR20210105335A (en) | Classification of defects in semiconductor specimens | |
| US11455490B2 (en) | Image-recognition apparatus, image-recognition method, and non-transitory computer-readable storage medium thereof | |
| CN111191670A (en) | Neural network-based classification device and classification method | |
| CN113781483A (en) | Industrial product appearance defect detection method and device | |
| JP6871807B2 (en) | Classifier construction method, classifier and classifier construction device | |
| CN116934732A (en) | Photovoltaic module detection method and device and electronic equipment | |
| TWI761806B (en) | Image recognition apparatus, image recognition method, and computer program product thereof | |
| CN204857670U (en) | Detection and treatment facility of bad granule among chip resistor | |
| CN114387230A (en) | PCB defect detection method based on re-verification detection | |
| TWI727595B (en) | Neural-network-based classification device and classification method | |
| TW202018600A (en) | Neural-network-based classification device and classification 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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190910 |
|
| RJ01 | Rejection of invention patent application after publication |