CN111652883B - Glass surface defect detection method based on deep learning - Google Patents
Glass surface defect detection method based on deep learning Download PDFInfo
- Publication number
- CN111652883B CN111652883B CN202010673882.3A CN202010673882A CN111652883B CN 111652883 B CN111652883 B CN 111652883B CN 202010673882 A CN202010673882 A CN 202010673882A CN 111652883 B CN111652883 B CN 111652883B
- Authority
- CN
- China
- Prior art keywords
- defect
- training
- network
- model
- 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.)
- Active
Links
- 230000007547 defect Effects 0.000 title claims abstract description 143
- 238000001514 detection method Methods 0.000 title claims abstract description 42
- 238000013135 deep learning Methods 0.000 title claims abstract description 16
- 239000011521 glass Substances 0.000 title claims abstract description 14
- 238000012549 training Methods 0.000 claims abstract description 49
- 238000000034 method Methods 0.000 claims abstract description 22
- 238000002372 labelling Methods 0.000 claims description 14
- 238000012360 testing method Methods 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 7
- 238000013145 classification model Methods 0.000 claims description 6
- 238000003384 imaging method Methods 0.000 claims description 6
- 238000012795 verification Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 3
- 230000001105 regulatory effect Effects 0.000 claims description 2
- 239000000428 dust Substances 0.000 abstract description 16
- 230000000694 effects Effects 0.000 abstract description 8
- 238000012805 post-processing Methods 0.000 abstract description 3
- 239000000047 product Substances 0.000 description 19
- 230000001276 controlling effect Effects 0.000 description 4
- 238000004519 manufacturing process Methods 0.000 description 4
- 239000013589 supplement Substances 0.000 description 3
- 238000013136 deep learning model Methods 0.000 description 2
- 238000003672 processing method Methods 0.000 description 2
- 241000519995 Stachys sylvatica Species 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 230000009194 climbing Effects 0.000 description 1
- 238000007621 cluster analysis Methods 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 239000012780 transparent material Substances 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- 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
- 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
- G06T2207/30168—Image quality inspection
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Quality & Reliability (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a glass surface defect detection method based on deep learning, which comprises the steps of training a network model by utilizing a residual network ResNet18 and a residual network ResNet101, detecting and judging defects of a sample image through the trained network model after the training of the network model is finished, and simultaneously managing and controlling defect specifications. Compared with the original method for classifying dirt, dust and defects by using the traditional algorithm, the classification accuracy of the model is improved from 70% to more than 90%; compared with the method for controlling the defect specification by using the score value of the deep learning classification result, the method for controlling the defect specification based on the post-processing of the deep learning detection result by using the traditional image algorithm has the advantages that the parameters such as the length, the width, the length-width ratio, the contrast ratio and the like are utilized to control the parameters of the defect, and a better control effect is achieved.
Description
Technical Field
The invention relates to the technical field of visual image processing, in particular to a glass surface defect detection method based on deep learning.
Background
The full-name Residual Network of ResNet. The best paper for CVPR was obtained by Kaiming He, deep Residual Learning for Image Recognition. The depth residual error network provided by the user reaches 152 layers of depth on the premise of ensuring network precision, and then further reaches 1000 layers of depth, the feature level becomes higher along with the deepening of the network, and the expression capability of the network is also greatly improved.
The existing case for detecting industrial defects by deep learning mainly comprises the following steps of collecting sample pictures and marking defects, setting network weight parameters for model training, and detecting OK and NG of the pictures by using test pictures.
Thus, the prior art has the following drawbacks:
1. the existing deep learning detection model is mostly used for judging the OK/NG of the product, and in the reality, not only the OK/NG of the product needs to be judged, but also the NG type needs to be classified, and defect specification management and control are carried out based on the classification result. For glass products, dust and dirt belong to dirt defects, the glass products can be repaired by a cleaning means, scratches and bubbles belong to breakage defects and are irreparable, and the two products must be distinguished. Because the imaging characteristics of dust, white spots, linear dirt and scratches are very similar, the existing traditional algorithm and application cases of deep learning have poor distinguishing effect on the two defects.
2. The existing deep learning detection model lacks a quantitative analysis means, and the output detection result is unfavorable for parameter control of defect specifications. The existing application cases can only output the heat map score of the product NG, and the probability of the product NG is higher when the score is higher, but the heat map score does not have practical physical significance, so that the defect specification is not convenient to control by a manufacturing staff. Often, a process person wants to define and manage the specifications of defects by intuitive parameters such as length, width, aspect ratio, contrast, number of defects in a certain area, etc.
Disclosure of Invention
The invention aims to solve the technical problems that: the difficulty in detecting the defects on the surface of the glass is that dirt and dust are difficult to avoid and are easy to confuse with real defects; in addition, the defect specification control scale is difficult to control, the control is too tight, a large amount of overkill can cause the product detection yield to be too low, and if the parameter control is too loose, too many omission tests can be caused, so that the shipment quality is affected; in order to solve the above problems, a glass surface defect detection method based on deep learning is provided.
The technical scheme adopted for solving the technical problems is as follows: the deep learning-based glass surface defect detection method comprises the steps of training a network model by utilizing a residual network ResNet18 and a residual network ResNet101, detecting and judging defects of a sample image through the trained network model after the network model is trained, and managing and controlling defect specifications.
Further, the training process of the network model comprises the following steps,
1) Collecting various defect samples according to defect detection requirements, and carrying out image acquisition on the samples by using an imaging system;
2) Dividing a region to be detected from an image, setting a gray value of a background region to 0, and adding the gray value into a training sample image set for defect labeling;
3) Labeling the sample image;
4) Model training, including training of a defect detection network and training of a defect discrimination network;
5) Testing sample pictures by using the network model generated by training, and counting the detected data and the classified data of the model;
6) And collecting the undetected pictures and the pictures with wrong classification, adding the pictures into a training sample to label defects, training the model again, testing the model, and iterating and optimizing the model once.
In step 3), the method of integrally marking the body projection is adopted, and specifically, the body and the projection are connected and marked into a whole by using a marking brush.
In step 4), the defect detection network is trained first, and then the defect discrimination network is trained; the defect detection network uses a residual network ResNet18 to segment all defects from the image, and the defect discrimination network uses a residual network ResNet101 to classify the defects.
Further, the invention detects the defects of the image with the gray value of 0 in the background area, which is input by using the trained defect detection model; if the defect is detected, transmitting the defect area image to a defect judging network for further defect classification; if no suspected defect is detected, the product is judged to be an OK product.
Further, the defect discriminating network of the present invention performs defect classification discrimination on the incoming suspicious defect area image, and each suspicious defect obtains a classification result.
Further, the defect specification management and control method of the invention is to calculate defect parameters which can be defined mathematically by using an image processing algorithm; if the defect parameter is within the control specification, judging that the defect parameter is an OK product; if outside the regulatory specification, then a NG is determined.
The beneficial effects of the invention are as follows:
1. compared with the original method for classifying dirt, dust and defects by using the traditional algorithm, the classification accuracy of the model is improved from 70% to more than 90%.
2. Compared with the method for controlling the defect specification by using the score value of the deep learning classification result, the method for controlling the defect specification based on the post-processing of the deep learning detection result by using the traditional image algorithm has the advantages that the parameters such as the length, the width, the length-width ratio, the contrast ratio and the like are utilized to control the parameters of the defect, and a better control effect is achieved. These parameters have the actual physical meaning, which better conforms to the human knowledge of the defect specification.
Drawings
FIG. 1 is a flow chart of the network training of the present invention;
fig. 2 is a flow chart of the invention for detection using a trained network.
Detailed Description
The invention will now be described in further detail with reference to the drawings and a preferred embodiment. The drawings are simplified schematic representations which merely illustrate the basic structure of the invention and therefore show only the structures which are relevant to the invention.
1-2, firstly, performing image segmentation and extraction of a detection area by using a traditional image processing algorithm, then performing defect detection on an image in the detection area by using a deep learning model based on ResNet18, classifying detected defects by using ResNet101, distinguishing dust, dirt and real defects, and finally performing defect specification management and control by combining the traditional image processing algorithm, thereby achieving the effects of defect classification detection and individual specification management and control of each type of defects.
The model has the following steps in the training process:
1. and collecting samples, namely collecting various defect samples according to defect detection requirements, and collecting images of the samples by using an imaging system. The defective samples in the glass inspection include dust, dirt, bubbles, scratches, and edge chipping. In the sample collection stage, a sample library needs to be established, balance of various samples in the training samples is kept, and the phenomenon of overfitting caused by unbalance of the samples is prevented.
2. And extracting a detection area, dividing an area to be detected from an image by using a traditional image processing method, setting the gray value of a background area to 0 for eliminating interference of the background area, and finally adding the image with the gray value of the background area set to 0 into a training sample image set for defect labeling.
3. And (3) defect labeling, namely creating 7 types of labels including dust, dirt, broken filaments, bubbles, scratches, broken edges and OK by using labeling software, and labeling the sample image. In the labeling process, a method of integrally labeling by body projection and imaging characteristics of glass transparent materials are adopted, so that when light irradiates on the surface of glass from a certain inclination angle, objects with protruding characteristics such as dust, broken filaments and the like are encountered, projection phenomenon exists in imaging, and other defects do not have the characteristics. Therefore, in the sample labeling stage, the dust body and the projection are connected and labeled into a whole by the labeling brush in the embodiment. Practical tests show that the model trained by the body projection integrated marking mode has higher classifying capability on dust, bubbles, hairline, scratches and dirt than the model trained by the body projection separation marking mode, and the classifying accuracy is 15% higher when few initial training samples are used, so that the classifying accuracy required by customers can be achieved more quickly.
4. Model training, wherein the model training comprises a defect detection network and a defect judging network, and the defect detection network is trained first and then the defect judging network is trained. The ResNet18 is a 2-class model, which is responsible for dividing all defects from images, and the improvement of the model effect mainly depends on the supplement of training samples, in particular the supplement training of missed detection samples. The defect judging network is a multi-classification model, and in the product proofing stage, sufficient defect samples cannot be collected for model training, and the model trained based on the defect judging network is easy to be subjected to fitting phenomenon, and particularly has the advantages that the test effect of the trained classification model on a training set is good, the test effect of the trained classification model on the testing set is poor, and the generalization capability of the model is weak. In order to solve the problem of over-fitting, in this embodiment, an Early Stopping method (Early Stopping) is adopted in the training process of the classification model, the accuracy of the verification data is calculated at the end of each training period (Epoch), the best verification accuracy up to date is recorded, and when the best verification accuracy is not reached in 10 continuous training periods, iteration is stopped. The classification effect of the trained model in the proofing stage and the small-batch trial production stage is relatively good. After the product enters the production capacity climbing and formal mass production stage, the means for preventing the overfitting phenomenon from occurring becomes management of a training sample library along with continuous supplement training of training samples, and the quantity of the training samples with various defects is maintained to be balanced.
5. And model testing, namely testing sample pictures by using a model network generated by training, and counting the detected data and the classified data of the model.
6. Model optimization, namely collecting undetected pictures and pictures with wrong classification, adding the pictures into a training sample to label defects, training the model again, testing the model, and optimizing the model for one time iteration.
The product detection flow based on the defect detection network and the defect discrimination network as shown in fig. 2 comprises the following steps:
(1) And extracting a detection area, dividing an area to be detected from an image by using a traditional image processing method, setting the gray value of a background area to 0 for eliminating interference of the background area, and finally transmitting the image with the gray value of the background area set to 0 into a model for detection.
(2) And the defect detection network is used for detecting the defects of the images with the gray value of 0 in the background area, if the defects are detected, the images of the defect areas are transmitted to the defect judgment network for carrying out the next defect classification, and if the suspected defects are not detected, the product is judged to be an OK product.
(3) The defect discriminating network discriminates the defect type of the incoming suspicious defect area image, each suspicious defect can obtain a classification result, and in the embodiment, 6 types of defects including dust, dirt, broken filaments, bubbles, scratches and broken edges are verified, so that the classification result is the score of the defect belonging to each type of defect, and the defect type with the highest score is selected as the classification result.
(4) The defect specification management and control is needed for each type of defect distinguished by the defect distinguishing network, the defect specification management and control mode is to calculate defect parameters which can be defined mathematically and are concerned by customers by utilizing a traditional image processing algorithm, such as region length (minimum circumscribed rectangle length1 of a region), region width (minimum circumscribed rectangle width length2 of a region), region contrast (difference between a region average Gray value Gray1 and a background region average Gray value Gray2 in 5 pixels outside the region) and the like, if the defect parameters are within the management and control specification, a product is judged to be an OK product, and if the defect parameters are outside the management and control specification, a product is judged to be a dirty NG (dust, dirty, broken filaments are judged to be dirty NG, scratches, bubbles and broken edges are judged to be broken NG).
Accordingly, the present invention uses a ResNet 18-based model for defect detection and a ResNet 101-based model for defect classification. The ResNet18 model has excellent capability of extracting defects in a glass-like flat background area, and can extract defects such as dust, dirt, broken filaments, bubbles, scratches, broken edges and the like. The ResNet 101-based model classification capability is excellent, and the trained classification model can accurately classify dust, dirt, broken filaments, bubbles, scratches, broken edges and the like under the condition that training samples such as dust, dirt, broken edges are sufficient and the samples are balanced.
In addition, the invention carries out post-processing on the detection result of the deep learning model by utilizing the traditional image processing algorithm, carries out blob analysis and Cluster analysis, calculates the Length, width, aspect ratio, contrast and Cluster result Cluster of the defects of each defect, and uses the parameters to independently control various defects by a process staff, thereby meeting the requirement of customer classification for controlling defect specifications.
The foregoing description is merely illustrative of specific embodiments of the invention, and the invention is not limited to the details shown, since modifications and variations of the foregoing embodiments may be made by those skilled in the art without departing from the spirit and scope of the invention.
Claims (2)
1. A glass surface defect detection method based on deep learning is characterized by comprising the following steps: training a network model by utilizing a residual network ResNet18 and a residual network ResNet101, and performing defect detection and discrimination on a sample image through the trained network model and defect specification control at the same time after the network model is trained;
the training process of the network model comprises the following steps,
1) Collecting various defect samples according to defect detection requirements, and carrying out image acquisition on the samples by using an imaging system; in the sample collection stage, a sample library is established, and the balance of various samples in the training samples is kept;
2) Dividing a region to be detected from an image, setting a gray value of a background region to 0, and adding the gray value into a training sample image set for defect labeling;
3) Labeling the sample image; in the labeling process, a method of integrally labeling the body projection is adopted, specifically, the body and the projection are connected and labeled into a whole by using a labeling brush;
4) Model training, including training of a defect detection network and training of a defect discrimination network; training a defect detection network and then training a defect discrimination network; the defect detection network utilizes a residual error network ResNet18 to divide all defects from the image, and the defect judgment network utilizes a residual error network ResNet101 to classify the defects; the defect discrimination network discriminates the defect type of the incoming suspicious defect area image, and each suspicious defect can obtain a classification result; an early stopping method is adopted in the training process of the classification model, the accuracy of verification data is calculated at the end of each training period, the best verification accuracy up to the present is recorded, and when the best verification accuracy is not reached in 10 continuous training periods, iteration is stopped;
5) Testing sample pictures by using the network model generated by training, and counting the detected data and the classified data of the model;
6) Model optimization, namely collecting undetected pictures and pictures with wrong classification, adding the pictures into a training sample to label defects, training the model again, testing the model, and carrying out iterative optimization on the model for the first time;
the defect specification management and control mode is that an image processing algorithm is utilized to calculate defect parameters which can be defined mathematically; if the defect parameter is within the control specification, judging that the defect parameter is an OK product; if outside the regulatory specification, then a NG is determined.
2. The method for detecting glass surface defects based on deep learning according to claim 1, wherein: performing defect detection on an image with the gray value of 0 in the input background area by using the trained defect detection model; if the defect is detected, transmitting the defect area image to a defect judging network for further defect classification; if no suspected defect is detected, the product is judged to be an OK product.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010673882.3A CN111652883B (en) | 2020-07-14 | 2020-07-14 | Glass surface defect detection method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010673882.3A CN111652883B (en) | 2020-07-14 | 2020-07-14 | Glass surface defect detection method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111652883A CN111652883A (en) | 2020-09-11 |
| CN111652883B true CN111652883B (en) | 2024-02-13 |
Family
ID=72348818
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010673882.3A Active CN111652883B (en) | 2020-07-14 | 2020-07-14 | Glass surface defect detection method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111652883B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113298765B (en) * | 2021-05-19 | 2023-11-07 | 广东工业大学 | Glass panel surface defect detection method based on image processing and deep learning |
| CN113658108A (en) * | 2021-07-22 | 2021-11-16 | 西南财经大学 | Glass defect detection method based on deep learning |
| CN114577816A (en) * | 2022-01-18 | 2022-06-03 | 广州超音速自动化科技股份有限公司 | Hydrogen fuel bipolar plate detection method |
| CN114441547A (en) * | 2022-04-11 | 2022-05-06 | 深圳市睿阳精视科技有限公司 | Intelligent household appliance cover plate defect detection method |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AU2018100321A4 (en) * | 2018-03-15 | 2018-04-26 | Chen, Jinghan Mr | Person ReID method based on metric learning with hard mining |
| CN109754392A (en) * | 2018-12-20 | 2019-05-14 | 上海集成电路研发中心有限公司 | A kind of device and method that defect image automatically analyzes |
| CN110175988A (en) * | 2019-04-25 | 2019-08-27 | 南京邮电大学 | Cloth defect inspection method based on deep learning |
| CN110400315A (en) * | 2019-08-01 | 2019-11-01 | 北京迈格威科技有限公司 | A kind of defect inspection method, apparatus and system |
| CN110458019A (en) * | 2019-07-10 | 2019-11-15 | 哈尔滨工程大学 | The waterborne target detection method for excluding inverted image interference under rare cognition sample conditions |
| CN110530875A (en) * | 2019-08-29 | 2019-12-03 | 珠海博达创意科技有限公司 | A kind of FPCB open defect automatic detection algorithm based on deep learning |
| CN111179251A (en) * | 2019-12-30 | 2020-05-19 | 上海交通大学 | Defect detection system and method based on twin neural network and by utilizing template comparison |
-
2020
- 2020-07-14 CN CN202010673882.3A patent/CN111652883B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AU2018100321A4 (en) * | 2018-03-15 | 2018-04-26 | Chen, Jinghan Mr | Person ReID method based on metric learning with hard mining |
| CN109754392A (en) * | 2018-12-20 | 2019-05-14 | 上海集成电路研发中心有限公司 | A kind of device and method that defect image automatically analyzes |
| CN110175988A (en) * | 2019-04-25 | 2019-08-27 | 南京邮电大学 | Cloth defect inspection method based on deep learning |
| CN110458019A (en) * | 2019-07-10 | 2019-11-15 | 哈尔滨工程大学 | The waterborne target detection method for excluding inverted image interference under rare cognition sample conditions |
| CN110400315A (en) * | 2019-08-01 | 2019-11-01 | 北京迈格威科技有限公司 | A kind of defect inspection method, apparatus and system |
| CN110530875A (en) * | 2019-08-29 | 2019-12-03 | 珠海博达创意科技有限公司 | A kind of FPCB open defect automatic detection algorithm based on deep learning |
| CN111179251A (en) * | 2019-12-30 | 2020-05-19 | 上海交通大学 | Defect detection system and method based on twin neural network and by utilizing template comparison |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111652883A (en) | 2020-09-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111652883B (en) | Glass surface defect detection method based on deep learning | |
| CN109829914B (en) | Method and device for detecting product defects | |
| Zhou et al. | Automated visual inspection of glass bottle bottom with saliency detection and template matching | |
| TWI525317B (en) | Method of Optical Defect Detection through Image analysis and Data Mining Integrated | |
| CN109508664B (en) | Dish identification pricing method based on deep learning | |
| CN111311542A (en) | Product quality detection method and device | |
| CN102637258B (en) | Method for creating online surface quality detection system defect library | |
| CN109946303A (en) | Check device and method | |
| CN103593670A (en) | Copper sheet and strip surface defect detection method based on-line sequential extreme learning machine | |
| CN104537651B (en) | Proportion detecting method and system for cracks in road surface image | |
| CN103852034B (en) | A kind of method for measuring perendicular | |
| CN108711148A (en) | A kind of wheel tyre defect intelligent detecting method based on deep learning | |
| US20210214765A1 (en) | Methods and systems for automated counting and classifying microorganisms | |
| CN108564577A (en) | Solar cell segment grid defect inspection method based on convolutional neural networks | |
| CN110473201A (en) | A kind of automatic testing method and device of disc surface defect | |
| CN109886932A (en) | Gear ring of wheel speed sensor detection method of surface flaw based on SVM | |
| CN111160451A (en) | Flexible material detection method and storage medium thereof | |
| CN109886931A (en) | Gear ring of wheel speed sensor detection method of surface flaw based on BP neural network | |
| WO2024002187A1 (en) | Defect detection method, defect detection device, and storage medium | |
| CN111986195A (en) | Appearance defect detection method and system | |
| CN107576660B (en) | A kind of double yellow duck egg Automatic Visual Inspection method based on apart from contour | |
| CN110660049A (en) | Tire defect detection method based on deep learning | |
| CN115147363A (en) | Image defect detection and classification method and system based on deep learning algorithm | |
| CN116559111A (en) | Sorghum variety identification method based on hyperspectral imaging technology | |
| CN110866931A (en) | Image segmentation model training method and classification-based enhanced image segmentation 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |