CN109187549A - A kind of detection method for backing layer edging defect detection station after rearview mirror - Google Patents
A kind of detection method for backing layer edging defect detection station after rearview mirror Download PDFInfo
- Publication number
- CN109187549A CN109187549A CN201810830935.0A CN201810830935A CN109187549A CN 109187549 A CN109187549 A CN 109187549A CN 201810830935 A CN201810830935 A CN 201810830935A CN 109187549 A CN109187549 A CN 109187549A
- Authority
- CN
- China
- Prior art keywords
- rearview mirror
- detection
- detected
- backing layer
- image
- 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
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
Abstract
The invention belongs to the technical fields of intellectualized detection, disclose a kind of detection method for backing layer edging defect detection station after rearview mirror, the following steps are included: obtaining the image of the rearview mirror to be detected using black and white camera Step 1: rearview mirror to be detected is placed into detection station using sucker suction mechanism；Step 2: carrying out denoising and normalized to described image, the image input neural network detection model after denoising and normalized is subjected to defects detection, testing result is shown by display；Step 3: if it is detected that the rearview mirror to be detected be defective products, rejected using the sucker suction mechanism；Step 4 repeats step 1 to three, carries out the defects detection of next rearview mirror to be detected.The full-automation to backing layer edging defect detection after rearview mirror is realized using method of the invention, while guaranteeing to examine accuracy rate, is improved checkability, is saved production cost.
Description
Technical field
The invention belongs to the technical fields of intellectualized detection, and in particular to one kind is examined for backing layer edging defect after rearview mirror
Survey the detection method of station.
Background technique
Automobile rearview mirror detection is the important environment of automobile rearview mirror production, is seemed especially for the safety of automobile rearview mirror
Important, the production of rearview mirror at present has formd collection on a large scale, realizes mechanical automation, but about in rearview mirror production
Flaw and substandard products detection are but because without reliable detection technique and always based on artificial detection.Due to the flaw of rearview mirror
Situations such as many kinds of, and flaw is smaller, and artificial detection will cause missing inspection, erroneous detection more.And it is non-due to artificial detection environment
The reasons such as closure, rearview mirror easily cause secondary damage in the detection, and detection process is also easy by factors such as dust, fingerprints
Interference, so that detection accuracy substantially reduces.
Summary of the invention
The present invention provides a kind of detection methods for backing layer edging defect detection station after rearview mirror, solve existing
The limitation of artificial detection method is very big, and time-consuming, and faint flaw is difficult to detect, and asks vulnerable to dust in air and electrostatic influence etc.
Topic.
The present invention can be achieved through the following technical solutions:
A kind of detection method for backing layer edging defect detection station after rearview mirror, comprising the following steps:
Step 1: rearview mirror to be detected is placed into detection station using sucker suction mechanism, obtained using black and white camera
The image of the rearview mirror to be detected；
Step 2: carrying out denoising and normalized to described image, the image after denoising and normalized is inputted
Neural network detection model carries out defects detection, shows testing result by display；
Step 3: if it is detected that the rearview mirror to be detected be defective products, rejected using the sucker suction mechanism；
Step 4 repeats step 1 to three, carries out the defects detection of next rearview mirror to be detected.
Further, the neural network detection model uses Alexnet network structure, including four convolutional layers, and four most
Great Chiization layer, two full articulamentums and a softmax classification layer, by being instructed with preset rearview mirror image pattern collection
Practice, completes classification learning.
Further, multiple rearview mirror gray level images are collected according to the difference of defect shape, using labelImg tool to institute
It states the defects of gray level image to be labeled, establishes preset rearview mirror image pattern collection.
Further, the rearview mirror gray level image is provided with three thousand sheets, and the shape includes arc-shaped, multilateral shape, ladder
Shape and triangular shape.
Further, the sucker suction mechanism is arranged in the transport mechanism using the double-deck conveyer belt, the double-deck transmission
One of band is for transmitting rearview mirror to be detected, and another for transmitting defective products rearview mirror, the sucker suction mechanism utilization
Sucker suction defective products rearview mirror is placed into another conveyer belt for transmitting defective products rearview mirror.
Further, the double-deck conveyer belt of the transport mechanism is using being arranged in parallel, the sucker suction mechanism include sucker,
Two sliding rails being arranged in a mutually vertical manner, the sucker is for adsorbing rearview mirror, moving direction of the sucker along one sliding rail
It is parallel with the direction of transfer of conveyer belt.
Further, the rearview mirror to be detected in the step 1 has puted up SN bar code
The present invention is beneficial to be had the technical effect that
The present invention completes the quick detection of rearview mirror using simplified AlexNet network, is shown and is detected by display screen
As a result, completing detection transport and the defective products of rearview mirror using the transport mechanism of sucker suction Agency layer conveyer belt
It rejects, realizes the full-automation to backing layer edging defect detection after rearview mirror, whole process does not need manually to participate in, and is guaranteeing to examine
While testing accuracy rate, checkability is improved, saves production cost, and avoid the secondary pollution to rearview mirror, guarantees to produce
The quality of product improves the reliability of product.
Detailed description of the invention
Fig. 1 is overview flow chart of the invention；
Fig. 2 is the structural schematic diagram of neural network of the invention；
Fig. 3 is that the present invention utilizes labelImg tool to mark schematic diagram to the defects of rearview mirror gray level image；
Fig. 4 is the contrast schematic diagram before and after denoising of the invention, before mark A indicates denoising, after mark B indicates denoising；
Fig. 5 is the schematic diagram of the associated mechanisms of testing station of the invention；
Fig. 6 is the comparison for carrying out accuracy rate test to neural network detection model of the invention using training set and test set
Figure, wherein straight line indicates that training set, dotted line indicate test set；
Wherein, 1- sucker suction mechanism, 2- edging defects detection station, 3- transport mechanism, 4- rearview mirror to be detected.
Specific embodiment
With reference to the accompanying drawing and the preferred embodiment specific embodiment that the present invention will be described in detail.
Referring to attached drawing 1, the present invention provides a kind of detection method for backing layer edging defect detection station after rearview mirror,
The following steps are included:
Step 1: rearview mirror 4 to be detected is placed into detection station 2 using sucker suction mechanism 1, obtained using black and white camera
Take the image of the rearview mirror to be detected；
The rearview mirror 4 to be detected is transported using the transport mechanism 3 using the double-deck conveyer belt, and a conveyer belt is for passing
Rearview mirror 4 to be detected is sent, another conveyer belt is arranged for transmitting defective products rearview mirror, sucker suction mechanism 1 in transport mechanism 3
On, including sucker and two sliding rails being arranged in a mutually vertical manner, the sucker is for adsorbing rearview mirror, after the sucker suction is to be detected
For visor along one sliding rail, the setting direction of the sliding rail and the direction of transfer of conveyer belt are parallel, are moved to detection station 2, pass through
After light source light filling, is taken pictures using the black and white camera that 2 top of detection station is arranged in the rear backing layer of rearview mirror, obtain it
The image of backing layer afterwards.Retrospect to the entire production procedure of rearview mirror for convenience, can first put up SN bar code, then put on it
Onto detection station 2, when detection, SN bar code scanning is first carried out, in this way, the information of detection whole process can be somebody's turn to do
SN bar code saves, convenient for retrospect.
Step 2: carrying out denoising and normalized to above-mentioned image, the image after denoising and normalized is inputted
Neural network detection model carries out defects detection, shows testing result by display；
The neural network detection model uses Alexnet network structure, since image of the invention is single channel, that is, gray scale
Image, the low-level feature such as color, edge in feature is more, and high-level characteristic such as texture is less, is easy to learn, so only needing
Less convolutional layer can extract with distinguish property feature, therefore the AlexNet network of script and on the basis of carry out letter
Change operation, to accelerate training and detection speed, which includes four convolutional layers, four maximum pond layers, two full articulamentums
Classification learning is completed, as shown in Figure 2 by being trained with preset rearview mirror image pattern collection with an output layer.
According to the difference of defect shape, including arc-shaped, multilateral shape, trapezoidal shape and triangular shape, multiple rearview mirrors are collected
Gray level image, about three thousand sheets, is labeled the defects of these gray level images using labelImg tool, as shown in figure 3,
Establish preset rearview mirror image pattern collection.
The activation primitive of the network still uses ReLu function, is divided according to preset rearview mirror image pattern collection defect
Class is set as five classifications using the output of the output layer of softmax classifier, respectively corresponding four kinds of defects and a back
Scape largely reduces the risk of over-fitting, decreases training in this way, the parameter of whole network is about 900,000
And detection time, actually detected speed are about 12 frames, accuracy rate is up to 98%.
It is utilized using the tall and handsome GPU up to model GTX1080,8G video memory in conjunction with the mode of propagated forward and backpropagation
Sample set is trained AlexNet network as described above, in order to avoid excessive numerical value causes memory to overflow, training
Iteration is set as 3000 times, and batch_size is set as 64, and the size and step-length of convolution kernel still use AlexNet first four
The numerical value of convolutional layer.
The training of the network is divided into two parts, i.e. propagated forward and backpropagation.In the forward propagation process, rearview mirror figure
As obtaining required characteristic pattern, then converting one-dimensional characteristic square for characteristic pattern after network convolution operation and pondization operation
One-dimensional vector is finally combined by the full articulamentum of network again and is identified by battle array；For reversal phase: the reality output of network
There are errors with ideal output, the partial derivative of each biasing and weight are sought error function, to keep error most fast along reduction of speed
Direction adjust each weight and biasing.But before starting training, all parameter, that is, weights and biasing will use difference
Random number initialized.
To the defects of training set mark and background, so-called background is set as removing the part of defect mark in image, leads to
The propagated forward process for crossing neural network obtains the eigenmatrix corresponding to original image, since defect and background have in original image
There is different features, the feature vector in eigenmatrix generated must have a certain difference, at this time according to the reality of network
Border output is the difference of the classification that softmax is generated according to feature vector and ideal output i.e. between the defect and image background of mark
Value to adjust the weight of network and bigoted, makes error reach minimum using error back propagation.To sum up, by before signal to biography
The training process with error back propagation is broadcast, network is made to reach convergence.
In training, 500 composition training sets are chosen from 3000 sample sets, for detecting the nerve after the completion of training
The Detection accuracy of network detection model, since the bad type that the product of different batches occurs might have difference, again
1000 image composition test sets are additionally acquired, the detection of the neural network detection model after the completion of detection training is equally used for
Accuracy rate obtains the corresponding accuracy rate of different the number of iterations by detection, as shown in fig. 6, taking second place when the number of iterations reaches 2000
Afterwards, network convergence is slack-off, and after iteration 3000 times, network convergence is gradually stable, illustrates that network does not need to be further continued for optimizing.
Since training sample of the invention is less, in order to enhance the accuracy in detection of network, need to carry out data to sample
Enhancing generates some new datas by some transformation from existing sample set, to expand trained sample size, mainly adopts
With following several: (1) flip horizontal image；(2) some images are converted out from original image random translation；(3) increase to image
Some random rotation angles.It the use of mask probability is 0.4 in two full articulamentums of model in order to avoid over-fitting
Dropout, i.e., the random drop nerve unit from neural network, allows partial nerve member to be not involved in propagated forward, is also not involved in anti-
To propagation, in this way, forcing e-learning to more healthy and stronger feature, the risk of network over-fitting is reduced.
The salt-pepper noise as present in image filters out noise frequency range therein using Fast Fourier Transform (FFT), before denoising
Then referring to attached drawing 4, the image of 244*244*1 is normalized into image afterwards, i.e. the length and width of image are respectively 244,
Single pass gray level image, the image input neural network detection model and carry out defects detection, by its volume of first convolutional layer
Product core size is that 3*3 completes a convolution operation and obtains the child node matrix that size is 112*112*64, then most by the second layer
Great Chiization layer extracts the greatest measure in convolution mask, obtains new node matrix equation, and so to the last a convolutional layer obtains
It is the characteristics of image of higher-dimension by the low-dimensional characteristics of image convolution of script to the node matrix equation of 28*28*256.When network training is completed
Afterwards, network parameter reaches excellent, and different defects and background will generate biggish difference in high dimensional feature, i.e., in 28*28*
The different location of 256 node matrix equation has different data values, at this time after two full articulamentums, obtains 1*1*2048
Node matrix equation, then classify using Softmax classifier, to detect different defect types, and by defect
It marks out and, on a display screen, which is arranged in the top of transport mechanism, so that track walker checks at any time for display.
Step 3: if it is detected that above-mentioned rearview mirror to be detected 4 be defective products, rejected using sucker suction mechanism 1, i.e.,
Using the sucker suction defective products, along another sliding rail, the setting of the sliding rail is put to, movement vertical with the direction of transfer of conveyer belt
To another conveyer belt, it is transported to reinspection station, member to be tested, which rechecks, to be verified, as shown in Figure 5.
Step 4 repeats step 1 to three, carries out the defects detection of next rearview mirror 4 to be detected.
The present invention completes the quick detection of rearview mirror using simplified AlexNet network, is shown and is detected by display screen
As a result, completing detection transport and the defective products of rearview mirror using the transport mechanism of sucker suction Agency layer conveyer belt
It rejects, realizes the full-automation to backing layer edging defect detection after rearview mirror, whole process does not need manually to participate in, and is guaranteeing to examine
While testing accuracy rate, checkability is improved, saves production cost, and avoid the secondary pollution to rearview mirror, guarantees to produce
The quality of product improves the reliability of product.
Although specific embodiments of the present invention have been described above, it will be appreciated by those of skill in the art that these
Be merely illustrative of, under the premise of without departing substantially from of the invention and essence, these embodiments can be made numerous variations or
Modification, therefore, protection scope of the present invention is defined by the appended claims.
Claims (7)
1. a kind of detection method for backing layer edging defect detection station after rearview mirror, it is characterised in that the following steps are included:
Step 1: rearview mirror to be detected is placed into detection station using sucker suction mechanism, obtained using black and white camera described in
The image of rearview mirror to be detected；
Step 2: carrying out denoising and normalized to described image, the image after denoising and normalized is inputted into nerve
Network detection model carries out defects detection, shows testing result by display；
Step 3: if it is detected that the rearview mirror to be detected be defective products, rejected using the sucker suction mechanism；
Step 4 repeats step 1 to three, carries out the defects detection of next rearview mirror to be detected.
2. the detection method according to claim 1 for backing layer edging defect detection station after rearview mirror, feature exist
In: the neural network detection model uses Alexnet network structure, including four convolutional layers, four maximum pond layers, two
Full articulamentum and a softmax classification layer, by being trained with preset rearview mirror image pattern collection, completion taxology
It practises.
3. the detection method according to claim 2 for backing layer edging defect detection station after rearview mirror, feature exist
In: multiple rearview mirror gray level images are collected according to the difference of defect shape, using labelImg tool in the gray level image
Defect be labeled, establish preset rearview mirror image pattern collection.
4. the detection method according to claim 3 for backing layer edging defect detection station after rearview mirror, feature exist
In: the rearview mirror gray level image is provided with three thousand sheets, and the shape includes arc-shaped, multilateral shape, trapezoidal shape and triangular shape.
5. the detection method according to claim 1 for backing layer edging defect detection station after rearview mirror, feature exist
In: the sucker suction mechanism is arranged in the transport mechanism using the double-deck conveyer belt, and one of the bilayer conveyer belt is used for
Rearview mirror to be detected is transmitted, another is used to transmit defective products rearview mirror, and the sucker suction mechanism is bad using sucker suction
Product rearview mirror is placed into another conveyer belt for transmitting defective products rearview mirror.
6. the detection method according to claim 5 for backing layer edging defect detection station after rearview mirror, feature exist
In: using being arranged in parallel, the sucker suction mechanism includes sucker, is arranged in a mutually vertical manner the double-deck conveyer belt of the transport mechanism
Two sliding rails, the sucker is for adsorbing rearview mirror, and the sucker is along the moving direction of one sliding rail and the biography of conveyer belt
Send direction parallel.
7. the detection method according to claim 1 for backing layer edging defect detection station after rearview mirror, feature exist
In: the rearview mirror to be detected in the step 1 has puted up SN bar code.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810830935.0A CN109187549A (en) | 2018-07-26 | 2018-07-26 | A kind of detection method for backing layer edging defect detection station after rearview mirror |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810830935.0A CN109187549A (en) | 2018-07-26 | 2018-07-26 | A kind of detection method for backing layer edging defect detection station after rearview mirror |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109187549A true CN109187549A (en) | 2019-01-11 |
Family
ID=64937454
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810830935.0A Pending CN109187549A (en) | 2018-07-26 | 2018-07-26 | A kind of detection method for backing layer edging defect detection station after rearview mirror |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109187549A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110220909A (en) * | 2019-04-28 | 2019-09-10 | 浙江大学 | A kind of Shield-bored tunnels Defect inspection method based on deep learning |
| CN116499920A (en) * | 2023-06-30 | 2023-07-28 | 青岛冠宝林活性炭有限公司 | Online monitoring method for adsorption state of tail gas activated carbon |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN203259481U (en) * | 2013-04-24 | 2013-10-30 | 合肥京东方光电科技有限公司 | Glass substrate detecting device |
| CN105424718A (en) * | 2015-11-02 | 2016-03-23 | 东华大学 | Car mirror flaw online automatic detection device and method based on double stations |
| CN105938105A (en) * | 2016-06-21 | 2016-09-14 | 深圳市振华兴科技有限公司 | Substrate detection equipment |
| CN206223679U (en) * | 2015-11-09 | 2017-06-06 | 艾斯迈科技股份有限公司 | Photomask detection device |
| CN107328787A (en) * | 2017-07-05 | 2017-11-07 | 北京科技大学 | A kind of metal plate and belt surface defects detection system based on depth convolutional neural networks |
| CN107862692A (en) * | 2017-11-30 | 2018-03-30 | 中山大学 | A kind of ribbon mark of break defect inspection method based on convolutional neural networks |
| CN108154508A (en) * | 2018-01-09 | 2018-06-12 | 北京百度网讯科技有限公司 | Method, apparatus, storage medium and the terminal device of product defects detection positioning |
-
2018
- 2018-07-26 CN CN201810830935.0A patent/CN109187549A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN203259481U (en) * | 2013-04-24 | 2013-10-30 | 合肥京东方光电科技有限公司 | Glass substrate detecting device |
| CN105424718A (en) * | 2015-11-02 | 2016-03-23 | 东华大学 | Car mirror flaw online automatic detection device and method based on double stations |
| CN206223679U (en) * | 2015-11-09 | 2017-06-06 | 艾斯迈科技股份有限公司 | Photomask detection device |
| CN105938105A (en) * | 2016-06-21 | 2016-09-14 | 深圳市振华兴科技有限公司 | Substrate detection equipment |
| CN107328787A (en) * | 2017-07-05 | 2017-11-07 | 北京科技大学 | A kind of metal plate and belt surface defects detection system based on depth convolutional neural networks |
| CN107862692A (en) * | 2017-11-30 | 2018-03-30 | 中山大学 | A kind of ribbon mark of break defect inspection method based on convolutional neural networks |
| CN108154508A (en) * | 2018-01-09 | 2018-06-12 | 北京百度网讯科技有限公司 | Method, apparatus, storage medium and the terminal device of product defects detection positioning |
Non-Patent Citations (2)
| Title |
|---|
| 赵俊冉等: "基于机器视觉的玻璃磨边缺陷检测", 《烟台大学学报( 自然科学与工程版)》 * |
| 颜伟鑫: "深度学习及其在工件缺陷自动检测中的应用研究", 《中国优秀硕士学位论文全文数据库工程科技I辑》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110220909A (en) * | 2019-04-28 | 2019-09-10 | 浙江大学 | A kind of Shield-bored tunnels Defect inspection method based on deep learning |
| CN116499920A (en) * | 2023-06-30 | 2023-07-28 | 青岛冠宝林活性炭有限公司 | Online monitoring method for adsorption state of tail gas activated carbon |
| CN116499920B (en) * | 2023-06-30 | 2023-09-12 | 青岛冠宝林活性炭有限公司 | Online monitoring method for adsorption state of tail gas activated carbon |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109239102B (en) | CNN-based flexible circuit board appearance defect detection method | |
| CN106875373B (en) | Mobile phone screen MURA defect detection method based on convolutional neural network pruning algorithm | |
| CN107330453B (en) | Pornographic image identification method based on step-by-step identification and fusion key part detection | |
| US11132786B2 (en) | Board defect filtering method based on defect list and circuit layout image and device thereof and computer-readable recording medium | |
| CN110473173A (en) | A kind of defect inspection method based on deep learning semantic segmentation | |
| CN105654121A (en) | Complex jacquard fabric defect detection method based on deep learning | |
| CN109509172A (en) | A kind of liquid crystal display flaw detection method and system based on deep learning | |
| CN107966454A (en) | A kind of end plug defect detecting device and detection method based on FPGA | |
| CN111932511B (en) | Electronic component quality detection method and system based on deep learning | |
| CN107944504B (en) | Board recognition and machine learning method and device for board recognition and electronic equipment | |
| US20200402221A1 (en) | Inspection system, image discrimination system, discrimination system, discriminator generation system, and learning data generation device | |
| CN110490842B (en) | Strip steel surface defect detection method based on deep learning | |
| CN112053317A (en) | Workpiece surface defect detection method based on cascade neural network | |
| CN102854194A (en) | Object surface defect detection method and apparatus based on linear array CCD | |
| CN103295021A (en) | Method and system for detecting and recognizing feature of vehicle in static image | |
| CN104200478A (en) | Low-resolution touch screen image defect detection method based on sparse representation | |
| CN108672316A (en) | A kind of micro parts quality detecting system based on convolutional neural networks | |
| CN109187549A (en) | A kind of detection method for backing layer edging defect detection station after rearview mirror | |
| US20220405586A1 (en) | Model generation apparatus, estimation apparatus, model generation method, and computer-readable storage medium storing a model generation program | |
| TW201930908A (en) | Board defect filtering method and device thereof and computer-readabel recording medium | |
| CN114742818A (en) | Visual defect detection system based on industrial area array camera and detection method thereof | |
| CN112215907A (en) | Automatic extraction method for weld defects | |
| CN109685030A (en) | A kind of mug rim of a cup defects detection classification method based on convolutional neural networks | |
| CN115861170A (en) | Surface defect detection method based on improved YOLO V4 algorithm | |
| CN112907562A (en) | MobileNet-based SMT defect classification algorithm |
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: 20190111 |
|
| RJ01 | Rejection of invention patent application after publication |