CN109859207A - A kind of defect inspection method of high density flexible substrate - Google Patents
A kind of defect inspection method of high density flexible substrate Download PDFInfo
- Publication number
- CN109859207A CN109859207A CN201910166760.2A CN201910166760A CN109859207A CN 109859207 A CN109859207 A CN 109859207A CN 201910166760 A CN201910166760 A CN 201910166760A CN 109859207 A CN109859207 A CN 109859207A
- Authority
- CN
- China
- Prior art keywords
- defect
- network
- convolutional neural
- inspection method
- candidate frame
- 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
- 230000007547 defect Effects 0.000 title claims abstract description 55
- 239000000758 substrate Substances 0.000 title claims abstract description 14
- 238000000034 method Methods 0.000 title claims description 32
- 238000007689 inspection Methods 0.000 title claims description 15
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 51
- 238000001514 detection method Methods 0.000 claims abstract description 24
- 238000012549 training Methods 0.000 claims abstract description 18
- 238000003062 neural network model Methods 0.000 claims abstract description 9
- 230000002950 deficient Effects 0.000 claims description 8
- 239000011159 matrix material Substances 0.000 claims description 7
- 230000007704 transition Effects 0.000 claims description 7
- 230000007246 mechanism Effects 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000011478 gradient descent method Methods 0.000 claims description 2
- 238000013519 translation Methods 0.000 claims description 2
- 238000003475 lamination Methods 0.000 claims 1
- 238000010586 diagram Methods 0.000 description 5
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000011179 visual inspection Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 229910052760 oxygen Inorganic materials 0.000 description 1
- 239000001301 oxygen Substances 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000003908 quality control method Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
本发明公开了一种高密度柔性基板的缺陷检测方法，包括采集带有外观缺陷的FICS图像，将图像预处理后统一为标准尺寸，标记每张图像中的缺陷位置和类别，作为faster R‑CNN卷积神经网络模型的训练样本；将标记号的训练样本作为faster R‑CNN卷积神经网络模型的输入，并输出FICS缺陷的位置与类型信息，得到训练好的基于faster R‑CNN卷积神经网络模型；然后将待检测的FICS图像输入训练好的基于faster R‑CNN卷积神经网络模型，输出是否有缺陷，如果有缺陷，则输出缺陷位置及类型。本发明实现高密度柔性基板外观缺陷的快速定位与类型判断，解决了传统缺陷检测方法速度慢，正确率低的问题。
The invention discloses a defect detection method for a high-density flexible substrate. The training sample of the CNN convolutional neural network model; the training sample of the label number is used as the input of the faster R-CNN convolutional neural network model, and the position and type information of the FICS defect is output, and the trained convolution based on the faster R-CNN is obtained. Neural network model; then input the FICS image to be detected into the trained convolutional neural network model based on faster R-CNN, and output whether there is a defect, and if there is a defect, output the defect location and type. The invention realizes rapid positioning and type judgment of appearance defects of high-density flexible substrates, and solves the problems of slow speed and low accuracy of traditional defect detection methods.
Description
Technical field
The present invention relates to machine vision surface defects detection technical fields, and in particular to a kind of high density flexible substrate lacks
Fall into detection method.
Background technique
High density flexible substrate (Flexible Integrated Circuit Substrate, abbreviation FICS) is a kind of
It can be used as the high density flexible print wiring board of IC package substrate.In the production process of high density FICS, due to technical process
The precision problem of control inevitably generates apparent defect.By high-precision vision-based detection, to the various appearances defect of FICS
It is quickly positioned and type identification, is the key that realize quality control in high density FICS manufacturing process.
The method that flexible base board manufacturer mainly uses Manual Visual Inspection carries out the inspection of open defect to high density FICS
It surveys.The detection efficiency of Manual Visual Inspection is relatively low, causes the significant wastage of human resources, and this method false detection rate is higher, difficult
To guarantee the quality of detection.Some scholars propose the open defect detection method based on characteristics of image, but these methods detect
Speed is slower, it is difficult to meet the practical application request of FICS detection；Meanwhile these methods can only be carried out for certain special defect
Detection can not identify simultaneously a plurality of types of open defects on FICS.
Summary of the invention
In order to overcome shortcoming and deficiency of the existing technology, the present invention provides a kind of defect inspection of high density flexible substrate
Survey method.
The present invention adopts the following technical scheme:
A kind of defect inspection method of high density flexible substrate, comprising:
Deep neural network model training step, specifically:
Acquisition has the FICS image of open defect, and standard size will be unified for after image preprocessing, marks every image
The defects of position and classification, the training sample as faster R-CNN convolutional neural networks model；
Using the training sample of label number as the input of faster R-CNN convolutional neural networks model, and exports FICS and lack
Sunken position and type information obtains trained based on faster R-CNN convolutional neural networks model；
Defect detection procedure, specifically:
FICS image to be detected input is trained based on faster R-CNN convolutional neural networks model, and output is
It is no defective, if defective, export defective locations and type.
The faster R-CNN convolutional neural networks model includes
Using pretreated FICS image to be detected as input, the shared convolutional neural networks of characteristic pattern are exported
After carrying out sample point coordinate to output characteristic pattern, by bilinear interpolation, the value of sample point coordinate is calculated,
And export the parallel space converting network that characteristic pattern corresponds to target point value；
Input feature vector figure corresponds to target point value and obtains the candidate frame generation network of candidate frame；
Candidate frame is inputted, the area-of-interest pond sorter network of the location information and classification for defect is exported.
The shared convolutional neural networks are made of 5 convolutional layers and 2 pond layers.
The parallel space converting network includes
The positioning network being made of three full articulamentums, for obtaining transition matrix θ by characteristic pattern；
By transition matrix θ, position of each sampled point on input feature vector figure on output characteristic pattern is obtained, it is then right
The pixel positioner that characteristic pattern is converted；
By bilinear interpolation, value is obtained to sample point coordinate and is calculated, exports the sampling of target point value on characteristic pattern
Device.
Pixel positioner includes the mapping mode of scaling, translation and rotation.
The candidate frame generates the mechanism that network uses sliding window combination anchor, determines the target of sliding window.
Area-of-interest pond sorter network includes an area-of-interest layer and two full articulamentums.
The loss function of area-of-interest pond sorter network are as follows:
Lhard=Lclass(p, u)+δ Lregre(t, v)
Lclass(p, u) represents the log loss of classification, Lregre(t, v) represents the recurrence loss of prediction block coordinate；P is represented
The prediction block to it includes object true classification predicted value, 4 set of coordinates that t and v have respectively represented prediction block close
The vector that the vector sum actual position coordinate come combines；
4 coordinates are respectively (x, y, w, h), and x represents the opposite abscissa of candidate frame, and y represents the relatively vertical of candidate frame and sits
Mark, w represent the width of candidate frame, and h represents the height of candidate frame, and δ returns loss for balanced sort loss and coordinate；
In the training process, preceding 128 candidate frames of the largest loss are considered as difficult sample, using under small batch stochastic gradient
Drop method carries out backpropagation, updates network parameter.
Uniform sizes of the present invention are specially the standard size of 224 × 224 pixels.
Beneficial effects of the present invention:
(1) a kind of high density flexible exterior substrate defective vision detection side based on improved faster R-CNN of the present invention
Method, it can be applied to the quick detection of open defect in flexible base board manufacturing process, replace Manual Visual Inspection, avoid artificial mesh
The waste of human resources brought by examining, and greatly improve working efficiency；
(2) detection method can be quickly detected from different types of defect on high density flexible substrate automatically, and will
The position of defect and type mark come out, and the defect solved in the case of the insoluble random site of traditional images method is quick
Positioning and test problems；
(3) without carrying out image characteristic analysis, after the neural network model that picture to be detected input has been trained, Ke Yizhi
Position and the classification information for connecing output defect, to solve traditional image processing method speed based on signature analysis slowly simultaneously
And the problem of multiple types defect can not be detected simultaneously.
Detailed description of the invention
Fig. 1 is work flow diagram of the invention；
Fig. 2 is that the present invention is based on the structure charts of the deep neural network model of faster R-CNN；
Fig. 3 is the schematic diagram of the shared convolutional neural networks in Fig. 2；
Fig. 4 is the schematic diagram of parallel space converting network in Fig. 2；
Fig. 5 is the schematic diagram that candidate frame generates network in Fig. 2；
Fig. 6 is the schematic diagram of area-of-interest pond sorter network in Fig. 2；
Fig. 7 (a)-Fig. 7 (f) is the detection effect figure of FICS in the present invention, is short circuit, open circuit, scratch, pin hole, oxygen respectively
Change and holes defect.
Specific embodiment
Below with reference to examples and drawings, the present invention is described in further detail, but embodiments of the present invention are not
It is limited to this.
Embodiment
As shown in Figure 1, a kind of defect inspection method of high density flexible substrate, including deep neural network model training and
Two steps of defects detection.
The deep neural network model training step, deep neural network model uses fasterR- in the present embodiment
CNN, specific as follows:
S1: the FICS image for largely having variety classes open defect is collected；
S2: the size for the image being collected into is pre-processed, and is unified for 224 × 224 standard size；
S3: marking the defects of every image position and classification, using the image marked as next step faster R-
The training sample of CNN convolutional neural networks；
S4: using the training sample image that has marked as the input of depth convolutional neural networks model, and FICS is lacked
Sunken position and type information is exported as model, is carried out to the neural network detection model based on improved faster R-CNN
Training, obtains the deep neural network model for FICS picture appearance defect location and type identification.
Defect detection procedure, specific as follows:
All FICS images to be detected are unified for standard size 224 × 224；
FICS image is inputted into trained convolutional neural networks model in order, exports whether defective, defect
Position and type information, and by information preservation into database；
The image inputted to next detects, until detection terminates.
As shown in Fig. 2, based on the depth convolutional neural networks model of improved faster R-CNN in the present embodiment.It by
4 parts form, and first part is shared convolutional neural networks, and second part is parallel space converting network, and Part III is
Candidate frame generates network, and Part IV is area-of-interest and pond network.
As shown in figure 3, first part is shared convolutional neural networks.This subnetwork is by 5 convolutional layers and 2 ponds
Layer composition, input are the images that size is 224*224 after pre-processing, and output is characteristic pattern.Pond layer uses the side in maximum pond
Method calculates to reduce a large amount of parameter, and prevents over-fitting.
As shown in figure 4, second part is parallel space converting network.Each spatial alternation network contains 3 portions
Point, it is respectively: 1) positions network, it is made of 3 full articulamentums.Its input is that width is w, and a height of h, port number is the spy of c
Sign figure/, output is a transition matrix θ；2) pixel positioner, it obtains each on output characteristic pattern by transition matrix θ
Position of the point on input feature vector figure, when carrying out becoming privileged processing to transition matrix θ, so that it may produce different types of feature
Figure transformation, the parallel space converting network contain scaling, translate, rotate 3 kinds of mapping modes；3) sampler passes through previous step
Output characteristic pattern is obtained after the sample point coordinate on input feature vector figure, by bilinear interpolation, to the value of sample point coordinate
It is calculated, it is determined that the value of target point is corresponded on output characteristic pattern.
As shown in Figures 5 and 6, it is one image of input that the candidate frame, which generates the effect of network, then output a batch point
The higher candidate frame of number.In the network, the generation of candidate frame uses the mechanism of sliding window combination anchor, to determine each cunning
Whether with the presence of target inside region corresponding to dynamic window.Since the length and width of target are inconsistent, it is therefore desirable to a variety of scales
Window is covered.The mechanism of anchor is on the basis of a benchmark window size, according to length and width and multiple proportional next life
At different candidate frames.The invention generates candidate frame using (8,16,32) three kinds of multiples and (0.5,1,2) three kinds of ratios,
The anchor of available 9 kinds of different scales.
Part IV is the area-of-interest pond sorter network in conjunction with difficult pattern detection.Area-of-interest pondization classification net
Network is made of an area-of-interest layer and two full articulamentums.Its input is a series of candidate frame, by a upper part
Candidate frame generate network generated, output is position and classification information of the neural network for defect, and loss function is as follows:
Lhard=Lclass(p, u)+δ Lregre(t, v)
Lclass(p, u) represents the log loss of classification, Lregre(t, v) represents the recurrence loss of prediction block coordinate；P is represented
The prediction block to it includes object true classification predicted value, 4 set of coordinates that t and v have respectively represented prediction block close
The vector that the vector sum actual position coordinate come combines.4 coordinates are respectively (x, y, w, h), and x represents the opposite of candidate frame
Abscissa, y represent the opposite ordinate of candidate frame, and w represents the width of candidate frame, and h represents the height of candidate frame.δ is for balancing
Classification Loss and coordinate return loss.In the training process, preceding 128 candidate frames of the largest loss are considered as difficult sample, utilized
Small batch stochastic gradient descent method carries out backpropagation, updates network parameter.
The training step of depth convolutional neural networks model of the present embodiment based on improved faster R-CNN are as follows:
1) convolutional neural networks of four parts are initialized respectively；
2) to shared convolutional neural networks, parallel space converting network, candidate frame generates network and is trained, and obtains a system
The candidate frame of column；
3) candidate frame generated using step 2, to shared convolutional neural networks, parallel space converting network and region of interest
Domain pond sorter network is trained, and is picked out the candidate frame of preceding 128 the largest loss every time, is utilized small batch stochastic gradient
Descent algorithm carries out network parameter update；
4) fixed shared convolutional neural networks, parallel space converting network generate network to candidate frame and are individually trained,
Obtain a series of candidate frame；
5) candidate frame generated using step 4, to shared convolutional neural networks, parallel space converting network and region of interest
Domain pond sorter network is trained, and 128 candidate frames of the largest loss is picked out every time, using under small batch stochastic gradient
It drops algorithm and carries out network parameter update；
Open defect detection system of the present invention and method are used for the short circuit, open circuit of route in FICS image, broken
The detection of the defects of damage, holes defect, testing result such as Fig. 7 (a), Fig. 7 (b), Fig. 7 (c), Fig. 7 (d), Fig. 7 (e) and Fig. 7 (f)
It is shown.The method of the present invention is defeated by FICS image to be detected by training the depth convolutional network model based on deep learning
Enter the model trained, can quickly export position and its type information of the defect on FICS image.
The above embodiment is a preferred embodiment of the present invention, but embodiments of the present invention are not by the embodiment
Limitation, other any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present invention,
It should be equivalent substitute mode, be included within the scope of the present invention.
Claims (9)
1. a kind of defect inspection method of high density flexible substrate characterized by comprising
Deep neural network model training step, specifically:
Acquisition has the FICS image of open defect, and standard size will be unified for after image preprocessing, marks in every image
Defective locations and classification, the training sample as faster R-CNN convolutional neural networks model；
Using the training sample of label number as the input of faster R-CNN convolutional neural networks model, and export FICS defect
Position and type information obtain trained based on faster R-CNN convolutional neural networks model；
Defect detection procedure, specifically:
FICS image to be detected input is trained based on faster R-CNN convolutional neural networks model, and whether output has
Defect exports defective locations and type if defective.
2. defect inspection method according to claim 1, which is characterized in that the faster R-CNN convolutional neural networks
Model includes
Using pretreated FICS image to be detected as input, the shared convolutional neural networks of characteristic pattern are exported
After carrying out sample point coordinate to output characteristic pattern, by bilinear interpolation, the value of sample point coordinate is calculated, and defeated
Characteristic pattern corresponds to the parallel space converting network of target point value out；
Input feature vector figure corresponds to target point value and obtains the candidate frame generation network of candidate frame；
Candidate frame is inputted, the area-of-interest pond sorter network of the location information and classification for defect is exported.
3. defect inspection method according to claim 2, which is characterized in that the shared convolutional neural networks are rolled up by 5
Lamination and 2 pond layers are constituted.
4. defect inspection method according to claim 2, which is characterized in that the parallel space converting network includes
The positioning network being made of three full articulamentums, for obtaining transition matrix θ by characteristic pattern；
By transition matrix θ, position of each sampled point on input feature vector figure on output characteristic pattern is obtained, then to feature
The pixel positioner that figure is converted；
By bilinear interpolation, value is obtained to sample point coordinate and is calculated, exports the sampler of target point value on characteristic pattern.
5. defect inspection method according to claim 4, which is characterized in that pixel positioner includes scaling, translation and rotation
The mapping mode turned.
6. defect inspection method according to claim 2, which is characterized in that the candidate frame generates network and uses sliding window
Mouth combines the mechanism of anchor, determines the target of sliding window.
7. defect inspection method according to claim 2, which is characterized in that the area-of-interest pond sorter network packet
Include an area-of-interest layer and two full articulamentums.
8. defect inspection method according to claim 7, which is characterized in that area-of-interest pond sorter network
Loss function are as follows:
Lhard=Lclass(p, u)+δ Lregre(t, v)
Lclass(p, u) represents the log loss of classification, Lregre(t, v) represents the recurrence loss of prediction block coordinate；It is pre- that p represents this
Survey frame to it includes object true classification predicted value, what 4 combinatorial coordinates that t and v have respectively represented prediction block got up
The vector that vector sum actual position coordinate combines；
4 coordinates are respectively (x, y, w, h), and x represents the opposite abscissa of candidate frame, and y represents the opposite ordinate of candidate frame, w
The width of candidate frame is represented, h represents the height of candidate frame, and δ returns loss for balanced sort loss and coordinate；
In the training process, preceding 128 candidate frames of the largest loss are considered as difficult sample, utilize small batch stochastic gradient descent method
Backpropagation is carried out, network parameter is updated.
9. defect inspection method according to claim 1, which is characterized in that uniform sizes are specially 224 × 224 pixels
Standard size.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910166760.2A CN109859207B (en) | 2019-03-06 | 2019-03-06 | A defect detection method for high-density flexible substrates |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910166760.2A CN109859207B (en) | 2019-03-06 | 2019-03-06 | A defect detection method for high-density flexible substrates |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109859207A true CN109859207A (en) | 2019-06-07 |
| CN109859207B CN109859207B (en) | 2023-06-23 |
Family
ID=66899962
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910166760.2A Active CN109859207B (en) | 2019-03-06 | 2019-03-06 | A defect detection method for high-density flexible substrates |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109859207B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110349145A (en) * | 2019-07-09 | 2019-10-18 | 京东方科技集团股份有限公司 | Defect inspection method, device, electronic equipment and storage medium |
| CN110400296A (en) * | 2019-07-19 | 2019-11-01 | 重庆邮电大学 | Binocular scanning and deep learning fusion recognition method and system for continuous casting slab surface defects |
| CN110560376A (en) * | 2019-07-19 | 2019-12-13 | 华瑞新智科技（北京）有限公司 | Product surface defect detection method and device |
| CN111080615A (en) * | 2019-12-12 | 2020-04-28 | 创新奇智(重庆)科技有限公司 | PCB defect detection system and detection method based on convolutional neural network |
| CN111415325A (en) * | 2019-11-11 | 2020-07-14 | 杭州电子科技大学 | A method for defect detection of copper foil substrate based on convolutional neural network |
| CN111429431A (en) * | 2020-03-24 | 2020-07-17 | 深圳市振邦智能科技股份有限公司 | Element positioning and identifying method based on convolutional neural network |
| CN111462043A (en) * | 2020-03-05 | 2020-07-28 | 维库（厦门）信息技术有限公司 | Defect detection method, device, equipment and medium based on Internet network |
| CN111563179A (en) * | 2020-03-24 | 2020-08-21 | 维库（厦门）信息技术有限公司 | Method and system for constructing defect image rapid classification model |
| CN112184667A (en) * | 2020-09-28 | 2021-01-05 | 京东方科技集团股份有限公司 | Defect detection, repair method, device and storage medium |
| CN113362277A (en) * | 2021-04-26 | 2021-09-07 | 辛米尔视觉科技(上海)有限公司 | Workpiece surface defect detection and segmentation method based on deep learning |
| CN114862845A (en) * | 2022-07-04 | 2022-08-05 | 深圳市瑞桔电子有限公司 | Defect detection method, device and equipment for mobile phone touch screen and storage medium |
| CN117670876A (en) * | 2024-01-31 | 2024-03-08 | 成都数之联科技股份有限公司 | Panel defect severity level judging method, system, equipment and storage medium |
| WO2024169750A1 (en) * | 2023-02-17 | 2024-08-22 | 杭州长川科技股份有限公司 | Defect detection model training method and apparatus, and electronic device |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106952250A (en) * | 2017-02-28 | 2017-07-14 | 北京科技大学 | A metal strip surface defect detection method and device based on Faster R-CNN network |
| CN107274451A (en) * | 2017-05-17 | 2017-10-20 | 北京工业大学 | Isolator detecting method and device based on shared convolutional neural networks |
| US20180268257A1 (en) * | 2017-03-20 | 2018-09-20 | Rolls-Royce Plc | Surface defect detection |
| CN109142371A (en) * | 2018-07-31 | 2019-01-04 | 华南理工大学 | High density flexible exterior substrate defect detecting system and method based on deep learning |
| CN109190668A (en) * | 2018-08-01 | 2019-01-11 | 福州大学 | The detection of multiclass certificate and classification method based on Faster-RCNN |
| CN109360204A (en) * | 2018-11-28 | 2019-02-19 | 燕山大学 | An internal defect detection method for multilayer metal lattice structure materials based on Faster R-CNN |
-
2019
- 2019-03-06 CN CN201910166760.2A patent/CN109859207B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106952250A (en) * | 2017-02-28 | 2017-07-14 | 北京科技大学 | A metal strip surface defect detection method and device based on Faster R-CNN network |
| US20180268257A1 (en) * | 2017-03-20 | 2018-09-20 | Rolls-Royce Plc | Surface defect detection |
| CN107274451A (en) * | 2017-05-17 | 2017-10-20 | 北京工业大学 | Isolator detecting method and device based on shared convolutional neural networks |
| CN109142371A (en) * | 2018-07-31 | 2019-01-04 | 华南理工大学 | High density flexible exterior substrate defect detecting system and method based on deep learning |
| CN109190668A (en) * | 2018-08-01 | 2019-01-11 | 福州大学 | The detection of multiclass certificate and classification method based on Faster-RCNN |
| CN109360204A (en) * | 2018-11-28 | 2019-02-19 | 燕山大学 | An internal defect detection method for multilayer metal lattice structure materials based on Faster R-CNN |
Non-Patent Citations (3)
| Title |
|---|
| 何俊杰等: "基于区域神经网络的TFT-LCD电路缺陷检测方法", 《计算机与现代化》 * |
| 常海涛等: "Faster R-CNN在工业CT图像缺陷检测中的应用", 《中国图象图形学报》 * |
| 邡鑫等: "基于卷积神经网络的晶圆缺陷检测与分类算法", 《计算机工程》 * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110349145A (en) * | 2019-07-09 | 2019-10-18 | 京东方科技集团股份有限公司 | Defect inspection method, device, electronic equipment and storage medium |
| CN110400296A (en) * | 2019-07-19 | 2019-11-01 | 重庆邮电大学 | Binocular scanning and deep learning fusion recognition method and system for continuous casting slab surface defects |
| CN110560376A (en) * | 2019-07-19 | 2019-12-13 | 华瑞新智科技（北京）有限公司 | Product surface defect detection method and device |
| CN111415325B (en) * | 2019-11-11 | 2023-04-25 | 杭州电子科技大学 | A Copper Foil Substrate Defect Detection Method Based on Convolutional Neural Network |
| CN111415325A (en) * | 2019-11-11 | 2020-07-14 | 杭州电子科技大学 | A method for defect detection of copper foil substrate based on convolutional neural network |
| CN111080615A (en) * | 2019-12-12 | 2020-04-28 | 创新奇智(重庆)科技有限公司 | PCB defect detection system and detection method based on convolutional neural network |
| CN111080615B (en) * | 2019-12-12 | 2023-06-16 | 创新奇智(重庆)科技有限公司 | PCB defect detection system and detection method based on convolutional neural network |
| CN111462043B (en) * | 2020-03-05 | 2023-10-24 | 维库（厦门）信息技术有限公司 | Defect detection method, device, equipment and medium based on Internet |
| CN111462043A (en) * | 2020-03-05 | 2020-07-28 | 维库（厦门）信息技术有限公司 | Defect detection method, device, equipment and medium based on Internet network |
| CN111563179A (en) * | 2020-03-24 | 2020-08-21 | 维库（厦门）信息技术有限公司 | Method and system for constructing defect image rapid classification model |
| CN111429431B (en) * | 2020-03-24 | 2023-09-19 | 深圳市振邦智能科技股份有限公司 | Element positioning and identifying method based on convolutional neural network |
| CN111429431A (en) * | 2020-03-24 | 2020-07-17 | 深圳市振邦智能科技股份有限公司 | Element positioning and identifying method based on convolutional neural network |
| CN112184667A (en) * | 2020-09-28 | 2021-01-05 | 京东方科技集团股份有限公司 | Defect detection, repair method, device and storage medium |
| CN113362277A (en) * | 2021-04-26 | 2021-09-07 | 辛米尔视觉科技(上海)有限公司 | Workpiece surface defect detection and segmentation method based on deep learning |
| CN114862845A (en) * | 2022-07-04 | 2022-08-05 | 深圳市瑞桔电子有限公司 | Defect detection method, device and equipment for mobile phone touch screen and storage medium |
| CN114862845B (en) * | 2022-07-04 | 2022-09-06 | 深圳市瑞桔电子有限公司 | Defect detection method, device and equipment for mobile phone touch screen and storage medium |
| WO2024169750A1 (en) * | 2023-02-17 | 2024-08-22 | 杭州长川科技股份有限公司 | Defect detection model training method and apparatus, and electronic device |
| CN117670876A (en) * | 2024-01-31 | 2024-03-08 | 成都数之联科技股份有限公司 | Panel defect severity level judging method, system, equipment and storage medium |
| CN117670876B (en) * | 2024-01-31 | 2024-05-03 | 成都数之联科技股份有限公司 | Panel defect severity level judging method, system, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109859207B (en) | 2023-06-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109859207A (en) | A kind of defect inspection method of high density flexible substrate | |
| Yuan et al. | Vision-based defect detection for mobile phone cover glass using deep neural networks | |
| CN104458755B (en) | Multi-type material surface defect detection method based on machine vision | |
| CN109142371A (en) | High density flexible exterior substrate defect detecting system and method based on deep learning | |
| CN109509187A (en) | A kind of efficient check algorithm for the nibs in big resolution ratio cloth image | |
| CN109711474A (en) | An Algorithm for Detection of Surface Defects of Aluminum Materials Based on Deep Learning | |
| CN109584227A (en) | A kind of quality of welding spot detection method and its realization system based on deep learning algorithm of target detection | |
| CN110473173A (en) | A kind of defect inspection method based on deep learning semantic segmentation | |
| CN110927171A (en) | A method for detecting defects on the chamfered surface of bearing rollers based on machine vision | |
| CN111667455A (en) | AI detection method for various defects of brush | |
| CN109829891A (en) | A kind of magnetic tile surface defect detection method based on intensive generation confrontation neural network | |
| KR20210150970A (en) | Detecting defects in semiconductor specimens using weak labeling | |
| CN113128335B (en) | Method, system and application for detecting, classifying and finding micro-living ancient fossil image | |
| CN109543688A (en) | A kind of novel meter reading detection and knowledge method for distinguishing based on multilayer convolutional neural networks | |
| CN111047655A (en) | High-definition camera cloth defect detection method based on convolutional neural network | |
| CN109948522A (en) | A method for interpreting the maturity of X-ray hand bone based on deep neural network | |
| CN112232263B (en) | Tomato identification method based on deep learning | |
| CN108764134A (en) | A kind of automatic positioning of polymorphic type instrument and recognition methods suitable for crusing robot | |
| CN109187579A (en) | Fabric defect detection method and device, computer equipment and computer-readable medium | |
| CN109978014A (en) | A kind of flexible base board defect inspection method merging intensive connection structure | |
| CN108985337A (en) | A kind of product surface scratch detection method based on picture depth study | |
| CN115082444B (en) | Copper pipe weld defect detection method and system based on image processing | |
| CN109543585A (en) | Underwater optics object detection and recognition method based on convolutional neural networks | |
| CN115205255A (en) | Stone automatic grading method and system based on deep learning | |
| CN118014972A (en) | Drilling core image identification and cataloging method and system |
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 |