CN110473179A - A kind of film surface defects detection method, system and equipment based on deep learning - Google Patents
A kind of film surface defects detection method, system and equipment based on deep learning Download PDFInfo
- Publication number
- CN110473179A CN110473179A CN201910694129.XA CN201910694129A CN110473179A CN 110473179 A CN110473179 A CN 110473179A CN 201910694129 A CN201910694129 A CN 201910694129A CN 110473179 A CN110473179 A CN 110473179A
- Authority
- CN
- China
- Prior art keywords
- image
- network model
- module
- sample
- defect
- 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 202
- 238000001514 detection method Methods 0.000 title claims abstract description 131
- 238000013135 deep learning Methods 0.000 title claims abstract description 38
- 230000004927 fusion Effects 0.000 claims abstract description 38
- 230000003287 optical effect Effects 0.000 claims abstract description 27
- 238000007781 pre-processing Methods 0.000 claims abstract description 27
- 238000013136 deep learning model Methods 0.000 claims abstract description 20
- 230000008676 import Effects 0.000 claims abstract description 12
- 238000004422 calculation algorithm Methods 0.000 claims description 52
- 230000011218 segmentation Effects 0.000 claims description 41
- 238000000034 method Methods 0.000 claims description 31
- 238000012360 testing method Methods 0.000 claims description 25
- 238000004590 computer program Methods 0.000 claims description 20
- 238000000576 coating method Methods 0.000 claims description 11
- 238000002156 mixing Methods 0.000 claims description 11
- 238000005520 cutting process Methods 0.000 claims description 10
- 238000007689 inspection Methods 0.000 claims description 7
- 238000000926 separation method Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 5
- 230000009182 swimming Effects 0.000 claims description 5
- 238000012549 training Methods 0.000 claims description 5
- 239000000155 melt Substances 0.000 claims description 2
- 230000004069 differentiation Effects 0.000 claims 1
- 239000000463 material Substances 0.000 claims 1
- 239000010408 film Substances 0.000 abstract description 33
- 239000010409 thin film Substances 0.000 abstract description 14
- 230000009286 beneficial effect Effects 0.000 abstract 1
- 238000004364 calculation method Methods 0.000 description 8
- 230000036548 skin texture Effects 0.000 description 8
- 230000000694 effects Effects 0.000 description 5
- 230000000295 complement effect Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 235000013399 edible fruits Nutrition 0.000 description 3
- 238000007635 classification algorithm Methods 0.000 description 2
- 239000011248 coating agent Substances 0.000 description 2
- 230000002708 enhancing effect Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 230000001965 increasing effect Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000003860 storage Methods 0.000 description 2
- 206010027146 Melanoderma Diseases 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000011031 large-scale manufacturing process Methods 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000000465 moulding Methods 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- 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
Abstract
The present invention provides a kind of film surface defects detection method, system and equipment based on deep learning.A kind of film surface defects detection method based on deep learning includes S1: carrying out Image Acquisition to sample front；S2: described image acquisition is carried out to the sample reverse side；S3: image preprocessing is carried out；S4: profile appearances defects detection is carried out；S5: judge contour detecting result；S6: being merged using multichannel, obtains Multichannel fusion image；S7: it imports in trained deep learning model in advance.A kind of film surface defects detection system based on deep learning includes image capture module, image pre-processing module, image co-registration module and image detection module；A kind of film surface defects detection device based on deep learning includes memory and processor.The beneficial effects of the present invention are: can effectively be identified under the premise of low in cost to optical thin film defect, cost of labor is reduced, improves recognition efficiency, and is disposed simply, repeatability is higher.
Description
Technical field
The present invention relates to optical thin film detection field, in particular to a kind of film surface defects detection based on deep learning
Method, system and equipment.
Background technique
Optical thin film is widely used in the liquid crystal display of mobile phone, computer, TV, glasses plated film, solar panel etc.
Field.Optical thin film is made of thin layered medium, propagates light beam by medium, mainly includes reflectance coating, anti-reflection film, optical filtering
Film, optical protection layer, polarizing coating etc. have surface quality high requirement.And in line flow procedure, due to shake
The reasons such as dynamic, air flowing, can generate a plurality of types of defects in Optical Coatings Surface, such as scratch, impression, dog-ear, pit, black
Spot, stain, hickie, fold, bubble and foreign matter etc. cause the using effect of product to decline.
In traditional optical thin film production field, due to the optical characteristics of optical thin film, the imaging and identification of various defects
It is more difficult, cause to be difficult to realize automatic monitoring, optical thin film can only largely be detected by manually estimating, but
On the one hand this detection method is easy to be influenced by subjective judgement, cause testing result unstable, poor reliability, another aspect labor
Fatigue resistance is big, at high cost, it is difficult to meet the needs of large-scale production.And it can be realized can only then detecting less for automatic detection
The defect for measuring type cannot detect all defects simultaneously, and deployment is complicated, and repeatability is poor.
Therefore market needs one kind and can realize automation optical thin film detection, deployment under the premise of higher accuracy rate
Simply, repeatability is high, has method, system and the equipment for adapting to various types of defects.
Summary of the invention
In order to solve the above-mentioned technical problem, a kind of film surface defects detection based on deep learning is disclosed in the present invention
Method, system, equipment, the technical scheme is that be implemented:
A kind of film surface defects detection method based on deep learning, comprising: S1: image is carried out to sample front and is adopted
Collection, obtains multiple sample direct pictures；S2: described image acquisition is carried out to the sample reverse side, obtains multiple sample reverse side figures
Picture；S3: image preprocessing is carried out to multiple described sample direct pictures and multiple described sample verso images, obtains the sample
Pretreatment image；S4: profile appearances defects detection is carried out to the sample pretreatment image, obtains sample contour detecting result；
S5: judging the sample contour detecting as a result, otherwise exporting profile mistake if profile appearances enter correctly into next step and terminating
Method；S6: the sample pretreatment image is merged using multichannel, obtains Multichannel fusion image；S7: will be described more
Channel blending image imports in trained deep learning model in advance, obtains sample and determines result.
Preferably, it is that light source is located above the sample that described image, which acquires used method, and camera receives the light
Source reflected light carries out described image acquisition or the light source is located at below the sample, and the camera receives one in transmitted light
Kind；It is by multiple described sample direct pictures and multiple described sample verso images schools that described image, which pre-processes used method,
Quasi- alignment, then cuts multiple described sample direct pictures and multiple described sample verso images；The profile appearances
Method used in defects detection is Contour tracing algorithm, based on image subset algorithm and based on one of distance of swimming algorithm or more
Kind；The deep learning model include the sorter network model of complex background texture, defect recognition segmentation network model and lack
Fall into recognition detection network model；It is described differentiate result include the sample whether have defect, the position of defect, defect it is big
The grade of small, defect classification and defect.
Preferably, the S7 includes: S7-1: Multichannel fusion image is imported to the sorter network model of complex background texture
In, obtain background texture classification results；S7-2: the background texture classification results are imported to the segmentation network of the defect recognition
In model, defect recognition classification results are obtained；S7-3: the defect recognition classification results are imported into the defect recognition and detect net
In network model, defect rank classification results are obtained；S7-4: by the background texture classification results, defect recognition classification knot
Fruit and the defect rank classification results, which merge, generates the sample judgement as a result, and sample is determined result output.
It preferably, further include S0: the training deep learning model；The S0 includes: S0-1: just to multiple samples
Face carries out described image acquisition, obtains multiple multiple described sample direct pictures；S0-2: institute is carried out to multiple sample reverse side
Image Acquisition is stated, multiple multiple described sample verso images are obtained；S0-3: to multiple multiple described sample direct pictures with it is multiple
Multiple described sample verso images carry out image preprocessing, obtain multiple sample pretreatment images；S0-4: will be multiple described
Sample pretreatment image is merged using the multichannel, obtains multiple Multichannel fusion images；S0-5: by multiple institutes
It states Multichannel fusion image to import in complex background Texture classification network model generating algorithm, obtains the complex background texture point
Class network model；S0-6: multiple Multichannel fusion images and the complex background Texture classification network model are imported and are lacked
It falls into the segmentation network model generating algorithm of identification, obtains the segmentation network model of the defect recognition；S0-7: will be multiple described
The segmentation network model of Multichannel fusion image and the complex background Texture classification network model, the defect recognition, which imports, to be lacked
It falls into recognition detection network model generating algorithm, obtains the defect recognition detection network model；S0-8: it is described multiple to merge output
The sorter network model of miscellaneous background texture, the segmentation network model of the defect recognition and the defect recognition detect network mould
Type.
A kind of film surface defects detection system based on deep learning uses a kind of film based on deep learning
Detection method of surface flaw, including image capture module, image pre-processing module, image co-registration module and image detection module；
Described image acquisition module includes conveyer belt, light source, camera, camera frame, for acquiring direct picture and verso images；The figure
As preprocessing module includes image calibration module and image cutting module, the connection described image acquisition of described image preprocessing module
Module receives the direct picture and the verso images and generates pretreatment image；Described image Fusion Module is based on more
The computer program of channel fusion, described image Fusion Module connect described image preprocessing module, receive the pretreatment figure
Picture simultaneously generates blending image；Described image detection module includes the sorter network model of profile detection module, complex background texture
Module, the segmentation network model module of defect recognition and defect recognition detect network model module, and described image detection module connects
Described image Fusion Module is connect, the blending image is received and generates image detection result.
Preferably, described image calibration module is one of computer program based on image calibration algorithm or a variety of；
Described image cutting module is the computer program based on foreground detection algorithm；The profile detection module is based on profile appearances
The computer program of recognizer.
It preferably, further include deep learning model generation module；The deep learning model generation module includes complicated back
The sorter network model generation module of scape texture, the segmentation network model generation module of defect recognition and defect recognition detect network
Model generation module.
Preferably, the sorter network model generation module of the complex background texture is based on complex background Texture classification net
The computer program of network model generation algorithm；The segmentation network model generation module of the defect recognition is based on defect recognition
Divide the computer program of network model generating algorithm；The defect recognition detection network model generation module is to be known based on defect
Not Jian Ce network model generating algorithm computer program.
Preferably, the profile detection module receives the pretreatment image, generates image outline testing result；It is described multiple
The sorter network model module of miscellaneous background texture receives described image contour detecting as a result, generating image background skin texture detection knot
Fruit；The segmentation network model module of the defect recognition receives described image background texture testing result, generates image deflects and knows
Other testing result；Defect recognition detection network model module receives described image defect recognition testing result, described in generation
Image deflects grade separation testing result, and described image background texture testing result, the detection of described image defect recognition are tied
Fruit and described image defect rank classification and Detection result merge output.
A kind of Optical Coatings Surface defect detection equipment of deep learning, including memory, processor；In the memory
It is stored at least one instruction, at least a Duan Chengxu, code set or instruction set, at least one instruction, an at least Duan Cheng
Sequence, code set or instruction set are executed to run a kind of Optical Coatings Surface defects detection of deep learning by the processor
System is to realize the Optical Coatings Surface defect inspection method of deep learning a kind of.
Human cost is high, accuracy rate is low, deployment is complicated needed for implementation technical solution of the present invention can solve in the prior art,
Repeatability is poor, only can be suitably used for the technical issues of single type defect；Implement technical solution of the present invention, it can be achieved that higher
Under the premise of accuracy rate, automation optical thin film detection is realized, deployment is simple, and repeatability is high, can adapt to various types of
The technical effect of defect.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
A kind of embodiment of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is a kind of a kind of flow chart of embodiment of film surface defects detection method based on deep learning；
Fig. 2 is a kind of a kind of system construction drawing of embodiment of film surface defects detection system based on deep learning.
In above-mentioned attached drawing, each figure number label is respectively indicated:
1- image capture module；
2- image pre-processing module
21- image calibration module；22- image cutting module；
3- image co-registration module；
4- image detection module；
41- profile detection module；The sorter network model module of 42- complex background texture；The segmentation net of 43- defect recognition
Network model module；44- defect recognition detects network model module；
5- deep learning model generation module；
The sorter network model generation module of 51- complex background texture；The segmentation network model of 52- defect recognition generates mould
Block；53- defect recognition detects network model generation module.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
In a kind of specific embodiment, as shown in Figure 1, a kind of film surface defects detection side based on deep learning
Method, comprising: S1: Image Acquisition is carried out to sample front, obtains multiple sample direct pictures；S2: image is carried out to sample reverse side
Acquisition, obtains multiple sample verso images；S3: image is carried out with multiple sample verso images to multiple sample direct pictures and is located in advance
Reason, obtains sample pretreatment image；S4: carrying out profile appearances defects detection to sample pretreatment image, obtains the inspection of sample profile
Survey result；S5: judging sample contour detecting as a result, otherwise exporting profile mistake simultaneously if profile appearances enter correctly into next step
Ending method；S6: sample pretreatment image is merged using multichannel, obtains Multichannel fusion image；S7: by multichannel
Blending image imports in trained deep learning model in advance, obtains sample and determines result.
In this kind of specific embodiment, due to the optical characteristics of film surface, there is the ability for changing beam direction, from
And cause to need to carry out Image Acquisition respectively in the tow sides of S1, S2 to film at different picture in tow sides,
To guarantee the integrality for film surface Image Acquisition；S3 pre-processes the image of tow sides, on the one hand can be with
Information unrelated in image is eliminated, useful real information is restored, on the other hand the detectability of enhancing for information about is simultaneously maximum
Simplify data to limit, to enhance the reliability of following model identification；Contour detecting belongs to Preliminary detection, relatively simple reliable
And required calculation amount is less, and subsequent step belongs to complicated further detection, and required calculation power is more, for passing through
The obvious shortcoming that simple profile measurement can be distinguished, S4, S5 can be higher to avoid its required calculation power after entering
Step, to calculate power consumed by efficiently reducing；It, can be by channel for traditional image recognition technology based on deep learning
Parameter setting is 3, to input the image information in tri- channels R, G, B, but this method produced on by assembly line it is optically thin
On this application scenarios of film, the Limited information that on the one hand includes due to the image of single angle and there are more noises, separately
On the one hand there is a large amount of redundancies to lead to inefficiency between three channels, and if setting each channel and individually entering
Multiple images, are combined into the multichannel image for being used for network inputs, then may be used by the image of different illumination condition and different angle
To utilize the abundant information in multiple complementary images simultaneously, and the influence of noise can be reduced, improve the robustness of system；It will pass through
The Multichannel fusion image of pretreatment and fusion treatment imports trained deep learning model in advance, by trained in advance
The identification of deep learning model, obtains recognition result；It through the above steps, can be under the premise of cheap cost, to optics
Film defects are effectively identified, cost of labor is reduced, and improve recognition efficiency, and have deployment simply, repeatability
Higher feature.
In a kind of preferred embodiment, as shown in Figure 1, method used in Image Acquisition is that light source is located on sample
Side, camera receives light source reflected light progress Image Acquisition or light source is located at below sample, and camera receives one of transmitted light；
Method used in image preprocessing is by multiple sample direct pictures and multiple sample verso images alignements, then to more
Sample direct picture is opened to be cut with multiple sample verso images；Method used in profile appearances defects detection chases after for profile
Track algorithm, based on image subset algorithm and based on one of distance of swimming algorithm or a variety of；Deep learning model includes complex background
The sorter network model of texture, the segmentation network model of defect recognition and defect recognition detect network model；Differentiate that result includes
Whether sample has the grade of defect, the position of defect, the size of defect, the classification of defect and defect.
In this kind of preferred embodiment, a kind of method used in Image Acquisition is light source and camera all the one of sample
Side, light source irradiate sample, camera acquisition be light source reflected light, another method is light source and camera on the both sides of sample,
Light source irradiates sample, and camera acquisition is that the transmitted light that light source penetrates sample helps to collect sample more in this way
The different characteristic of a angle helps to improve the reliability of subsequent defective identification, during Image Acquisition, camera and light source
Angle can be adjusted according to actual needs；After obtaining sample direct picture and sample verso images, in order to utilize
Its both complementary image information, it is necessary to be carried out alignement, the image calibration pair based on characteristic point generally can be used
Together, while for the ease of profile appearances defects detection, image cutting is carried out after the calibration, sample direct picture and sample is anti-
Face image segmentation is subregion one by one；Profile appearances defects detection is a kind of preliminary detection method, should be as far as possible
The calculation power needed for it is reduced, Contour tracing algorithm does not need higher calculation based on image subset algorithm and based on distance of swimming algorithm
Power can complete the effect of contour detecting, specifically select which kind of algorithm that can be carried out according to practical software and hardware resources and running environment
Corresponding selection；The sorter network model of complex background texture for being identified to the texture type of film, defect recognition
For identifying to the type of defect, defect recognition detects network model and is used to carry out the grade of model segmentation network model
Classification, to obtain accurate defect characteristics.
In a kind of preferred embodiment, as shown in Figure 1, S7 includes: S7-1: Multichannel fusion image is imported complicated back
In the sorter network model of scape texture, background texture classification results are obtained；S7-2: background texture classification results are imported into defect and are known
In other segmentation network model, defect recognition classification results are obtained；S7-3: defect recognition classification results are imported into defect recognition inspection
It surveys in network model, obtains defect rank classification results；S7-4: by background texture classification results, defect recognition classification results and
Defect rank classification results, which merge, generates sample judgement as a result, and sample is determined result output.
In this kind of preferred embodiment, the background texture very different of different types of film, the background of a kind of film
Texture is likely to similar with the imaging of certain defect texture in another kind of film, is easy to produce erroneous detection, therefore should be first to film
Background texture is classified, and background texture classification results are the classification results one for including smooth grain, rough grain, random grain
Kind, according to background texture classification results, the segmentation network model of defect recognition, which just can be learnt, should be applicable in any defect inspection
It surveys, so that corresponding detection path be selected to be detected, similarly, defect recognition detection network model can be obtained according to the above results
Know the division that any defect rank criteria for classifying should be selected to be carried out in grade to film defects.
In a kind of preferred embodiment, as shown in Figure 1, further including S0: training deep learning model；S0 includes: S0-1:
Image Acquisition is carried out to multiple sample fronts, obtains multiple multiple sample direct pictures；S0-2: figure is carried out to multiple sample reverse side
As acquisition, multiple multiple sample verso images are obtained；S0-3: to multiple multiple sample direct pictures and multiple multiple sample reverse side
Image carries out image preprocessing, obtains multiple sample pretreatment images；S0-4: multiple sample pretreatment images are used into multichannel
It is merged, obtains multiple Multichannel fusion images；S0-5: multiple Multichannel fusion images are imported into complex background Texture classification
In network model generating algorithm, complex background Texture classification network model is obtained；S0-6: by multiple Multichannel fusion images and again
Miscellaneous background texture sorter network model imports in the segmentation network model generating algorithm of defect recognition, obtains the segmentation of defect recognition
Network model；S0-7: by the segmentation net of multiple Multichannel fusion images and complex background Texture classification network model, defect recognition
Network model imports in defect recognition detection network model generating algorithm, obtains defect recognition detection network model；S0-8: merge defeated
The sorter network model of complex background texture, the segmentation network model of defect recognition and defect recognition detect network model out.
In this kind of preferred embodiment, as shown in Figure 1, since the basis of deep learning is statistics, therefore in S0-1 and
As much as possible more images should be acquired in S0-2, help to improve the quality of deep learning model；S0-3 is to positive and negative two
The image in face is pre-processed, and information unrelated in image on the one hand can be eliminated, and restores useful real information, on the other hand
Enhance detectability for information about and simplify data to the maximum extent, to enhance the reliability of following model identification, simultaneously
It is the type that deep learning algorithm can identify by the data type conversion of image；The same S6 of the effect of S0-4, can be improved system
Robustness；Complex background Texture classification network model generating algorithm construct bottom network model, can be include VGG,
One in the neural network classification algorithm of the classics of MobileNet, GoogLeNet, Inception, Residual Network
Kind；The segmentation network model generating algorithm of defect recognition be for based on VGG, MobileNet, ResNet, U-Net, DeepLab,
One of computer program of RefineNet, PSPNet or a variety of deformations, convolutional layer are divided into five layers, each convolutional layer
It is extracted feature abundant, with the increase of receptive field, the feature that convolutional layer extracts becomes coarseness by fine granularity, to obtain
Then characteristics of image on different grain size again merges the convolutional layer feature on different grain size, higher accurate to reach
Degree, while blending image loss function is set during training, to improve its accuracy；Due to different grades of defect pair
It is different in the influence of optical thin film, it needs for different grades of defect film using different measures, therefore needs to be arranged corresponding
Defect recognition detection network model generating algorithm to generate corresponding defect rank disaggregated model, defect recognition detects network mould
Type generating algorithm can be that it can be the classical neural network classification calculation for including VGG, GoogLeNet, Residual Network
One of method, according to the subregion of defect recognition network model output as input, to obtain accurately defect characteristics.
In a kind of specific embodiment, as shown in Fig. 2, a kind of film surface defects detection system based on deep learning
System, uses a kind of film surface defects detection method based on deep learning, including image capture module 1, image preprocessing mould
Block 2, image co-registration module 3 and image detection module 4；Image capture module 1 includes conveyer belt, light source, camera, camera frame, is used
In acquisition direct picture and verso images；Image pre-processing module 2 includes image calibration module 21 and image cutting module 22, figure
As preprocessing module 2 connects image capture module 1, reception direct picture and verso images and generates pretreatment image；Image melts
Molding block 3 is the computer program based on Multichannel fusion, and image co-registration module 3 connects image pre-processing module 2, receives pre- place
Reason image simultaneously generates blending image；Image detection module 4 includes the sorter network mould of profile detection module 41, complex background texture
Pattern block 42, the segmentation network model module 43 of defect recognition and defect recognition detect network model module 44, image detection mould
Block 4 connects image Fusion Module 3, receives blending image and generates image detection result.
In this kind of specific embodiment, when detecting, film is placed in acquisition sample image, and sample image transmitting is given
Image pre-processing module 2, image pre-processing module 2 carry out pretreatment generation pretreatment image after receiving sample image, and will be pre-
Image transmitting is handled to image co-registration module 3, image co-registration module 3 generates Multichannel fusion image after receiving pretreatment image,
And by Multichannel fusion image transmitting to image detection module 4, to generate image detection result；Camera is generally high definition phase
Machine, can recognize more tiny defect, and light source can choose strip light and ring light as needed；2 one side of image pre-processing module
Face can eliminate information unrelated in image, restore useful real information, on the other hand the detectability of enhancing for information about
And simplify data to the maximum extent, to enhance the reliability of following model identification, image calibration module 21 is for obtaining
After sample direct picture and sample verso images, the leading processing carried out using both complementary image informations, Zhi Houzai
It is cut by the image after 22 pairs of image cutting module calibrations to complete preprocessing process；For traditional based on deep learning
Image recognition technology, channel parameters can be set as 3, to input the image information in tri- channels R, G, B, but this method exists
By on optical thin film this application scenarios for being produced on assembly line, on the one hand due to Limited information that the image of single angle includes
And there are more noises, and on the other hand there is a large amount of redundancies to lead to inefficiency between three channels, and if
The image that each channel individually enters different illumination condition and different angle is set, multiple images are combined into and are used for network
The multichannel image of input then can reduce the influence of noise simultaneously using the abundant information in multiple complementary images, mention
The robustness of high system, to more effectively utilize the steric information of product image, image co-registration module 3 plays raising and improves
The robustness of system and the steric information for effectively increasing product image；It will melt by pretreatment and the multichannel of fusion treatment
It closes image importing image detection module 4 and obtains recognition result by the identification of image detection module 4；Profile detection module 41 is
Preliminary detection can directly detect the product of faulty goods either profile existing defects itself, point of complex background texture
Class network model module 42 is for classifying to the texture of film, and the segmentation network model module 43 of defect recognition is for identification
Size, position and the type information of defect, defect recognition detection network model module 44 is for classifying to the grade of defect；
Optical thin film defect can effectively be identified under the premise of cheap cost by the cooperation between modules,
Reduce cost of labor, improve recognition efficiency, and there is deployment simply, the higher feature of repeatability.
In a kind of preferred embodiment, as shown in Fig. 2, image calibration module 21 is the calculating based on image calibration algorithm
One of machine program is a variety of；Image cutting module 22, will be unrelated in shooting image as one of pretreated important step
Background parts detection, and according to actual needs cut off image edge, retain target area, image cutting module 22 be base
In the computer program of foreground detection algorithm；Profile detection module 41 is the computer program based on profile appearances recognizer.
In this kind of preferred embodiment, the image calibration alignment based on characteristic point generally can be used, it can be according to specific
Image conditions are adjusted calibrating mode, while for the ease of profile appearances defects detection, carrying out image point after the calibration
It cuts, sample direct picture and sample verso images is divided into subregion one by one, partitioning scheme can be according to real image
Situation is taken based on one or more of foreground detection algorithm；The detection that profile detection module 41 is run is a kind of preliminary
Detection method, the calculation power needed for it should be reduced as far as possible, Contour tracing algorithm based on image subset algorithm and is based on the distance of swimming
Algorithm, which does not need higher calculation power, can complete the effect of contour detecting, specifically select which kind of algorithm can be according to practical soft or hard
Part resource and running environment are selected accordingly.
In a kind of preferred embodiment, as shown in Fig. 2, further including deep learning model generation module 5；Deep learning mould
Type generation module 5 includes the segmentation network model life of the sorter network model generation module 51 of complex background texture, defect recognition
Network model generation module 53 is detected at module 52 and defect recognition.
When carrying out model generation, the acquisition sample direct picture as much as possible of image capture module 1 and sample reverse side figure
As it is simultaneously transferred to image pre-processing module 2, image pre-processing module 2 generated after processing pretreatment image and by it
It is transferred to image co-registration module 3, image co-registration module 3 merges pretreatment image and generates sample blending image importing complex background line
The sorter network model generation module 51 of reason, so that the sorter network model of complex background texture is generated, according to complex background line
The classification results of the sorter network model of reason will import the segmentation network model generation module 52 of defect recognition, generates defect and knows
Other segmentation network model, point of the segmentation network model of sorter network model and defect recognition further according to complex background texture
Class detects network model generation module 53 as a result, importing defect recognition, so that defect rank disaggregated model is generated, it finally will be complicated
The sorter network model of background texture, the segmentation network model of defect recognition merge output with defect recognition detection network model and are
Deep learning model imports image detection module 4.
In a kind of preferred embodiment, as shown in Fig. 2, the sorter network model generation module 51 of complex background texture is
Computer program based on complex background Texture classification network model generating algorithm；The segmentation network model of defect recognition generates mould
Block 52 is the computer program of the segmentation network model generating algorithm based on defect recognition；Defect recognition detects network model and generates
Module 53 is that the computer program of network model generating algorithm is detected based on defect recognition.
In this kind of preferred embodiment, complex background Texture classification network model generation module constructs the network mould of bottom
Type, based on algorithm can be the classical nerve net for including VGG, MobileNet, GoogLeNet, Residual Network
One of network sorting algorithm；The algorithm that the segmentation network model generation module 52 of defect recognition is based on be VGG,
One or more of MobileNet, ResNet, U-Net, DeepLab, RefineNet, PSPNet deformation, convolutional layer point
It is five layers, each convolutional layer is extracted feature abundant, and with the increase of receptive field, the feature that convolutional layer extracts is by fine granularity
Become coarseness, to obtain the characteristics of image on different grain size, then the convolutional layer feature on different grain size is melted again
It closes, to reach higher accuracy, while blending image loss function is set during training, to improve its accuracy；
The different grades of defect of influence due to to(for) optical thin film is different, needs for different grades of defect film using different
Measure, therefore need to be arranged corresponding defect recognition detection network model generating algorithm to generate corresponding defect rank classification mould
Type, defect recognition detection network model generating algorithm can be its can be include VGG, Mobilenet, GoogLeNet, Residual
Classical one of the neural network classification algorithm of Network, makees according to the subregion of defect recognition network model output
To input, to obtain accurately defect characteristics.
In a kind of preferred embodiment, as shown in Fig. 2, profile detection module 41 receives pretreatment image, image is generated
Contour detecting result；The sorter network model module 42 of complex background texture receives image outline testing result, generates image back
Scape skin texture detection result；The segmentation network model module 43 of defect recognition receives image background skin texture detection as a result, generating image
Defect recognition testing result；Defect recognition detects network model module 44 and receives image deflects recognition detection as a result, generating image
Defect rank classification and Detection is as a result, and by image background skin texture detection result, image deflects recognition detection result and image deflects
Grade separation testing result merges output.
In this kind of preferred embodiment, profile detection module 41 generates image outline testing result, according to image outline
Testing result judges whether the sample has profile defects, if any obvious shortcoming, is directly determined as faulty goods, does not need to import multiple
Among the sorter network model module 42 of miscellaneous background texture, if cannot be determined as obvious shortcoming, it is conducted into complex background
The sorter network model module 42 of texture is determined by the sorter network model module 42 of complex background texture；Complex background
The image background skin texture detection result that the sorter network model module 42 of texture generates be include smooth grain, rough grain, with
The classification results of machine texture are a kind of；Defect recognition detects network model module 44 and receives image background classification results, and building is different
Image deflects under the background texture of classification identify network, convert the image deflects detection of complex background texture to multiple single
Defects detection under background texture, obtained image deflects recognition detection result include the size position classification of the defect identified
Information, and with Multichannel fusion image and image background skin texture detection result；Defect recognition detection network model module 44 connects
It receives and generates image after image deflects recognition detection result, image background skin texture detection result and the Multichannel fusion image had and lack
Grade separation testing result is fallen into, image deflects grade separation testing result is slightly, in slight, moderate, severe, major defect
One kind, actual defect rank division can be increased or be reduced according to the actual situation, finally by image background skin texture detection
As a result, image deflects recognition detection result and image deflects grade separation testing result merge output as testing result.
In a kind of specific embodiment, a kind of Optical Coatings Surface defect detection equipment of deep learning, including storage
Device, processor；At least one instruction, at least a Duan Chengxu, code set or instruction set are stored in memory, at least one refers to
It enables, an at least Duan Chengxu, code set or instruction set is executed to run a kind of optical thin film table of deep learning by processor
Planar defect detection system is to realize the Optical Coatings Surface defect inspection method of deep learning a kind of.
In the preferred embodiment, memory includes memory and external memory, and physical memory can be used, such as hard disk and interior
It deposits, it is possible to use virtual memory, the cloud storage service such as provided using cloud service provider；Processor includes at least a centre
Device is managed, either the processor of entity, it is possible to use virtual processor, the virtual processor such as provided using cloud service provider；
Memory and processor are used cooperatively to run a kind of open defect detection system merged based on multiple light courcess to realize a kind of base
In the open defect detection method of multiple light courcess fusion, the process that memory and processor are used cooperatively belongs to fields technology people
Known to member, this will not be repeated here.
It should be pointed out that the foregoing is merely illustrative of the preferred embodiments of the present invention, it is not intended to limit the invention, it is all
Within the spirit and principles in the present invention, any modification, equivalent replacement, improvement and so on should be included in guarantor of the invention
Within the scope of shield.
Claims (10)
1. a kind of film surface defects detection method based on deep learning characterized by comprising
S1: Image Acquisition is carried out to sample front, obtains multiple sample direct pictures；
S2: described image acquisition is carried out to the sample reverse side, obtains multiple sample verso images；
S3: image preprocessing is carried out to multiple described sample direct pictures and multiple described sample verso images, obtains the inspection
Material pretreatment image；
S4: profile appearances defects detection is carried out to the sample pretreatment image, obtains sample contour detecting result；
S5: judging the sample contour detecting as a result, otherwise exporting profile mistake simultaneously if profile appearances enter correctly into next step
Ending method；
S6: the sample pretreatment image is merged using multichannel, obtains Multichannel fusion image；
S7: the Multichannel fusion image is imported in trained deep learning model in advance, is obtained sample and is determined result.
2. a kind of film surface defects detection method based on deep learning according to claim 1, which is characterized in that institute
Stating method used in Image Acquisition is that light source is located above the sample, and camera receives the light source reflected light and carries out the figure
As acquisition or the light source are located at one of the sample lower section, described camera reception transmitted light；
It is by multiple described sample direct pictures and multiple described sample verso images that described image, which pre-processes used method,
Then alignement cuts multiple described sample direct pictures and multiple described sample verso images；
Method used in the profile appearances defects detection is Contour tracing algorithm, based on image subset algorithm and based on the distance of swimming
One of algorithm is a variety of；
The deep learning model include the sorter network model of complex background texture, defect recognition segmentation network model and lack
Fall into recognition detection network model；
The differentiation result includes whether the sample has defect, the position of defect, the size of defect, the classification of defect and lack
Sunken grade.
3. a kind of film surface defects detection method based on deep learning according to claim 1, which is characterized in that institute
Stating S7 includes:
S7-1: Multichannel fusion image being imported in the sorter network model of the complex background texture, obtains background texture point
Class result；
S7-2: the background texture classification results are imported in the segmentation network model of the defect recognition, defect recognition is obtained
Classification results；
S7-3: the defect recognition classification results are imported in the defect recognition detection network model, defect rank point is obtained
Class result；
S7-4: the background texture classification results, the defect recognition classification results and the defect rank classification results are closed
And it generates the sample and determines as a result, and sample is determined result output.
4. a kind of film surface defects detection method based on deep learning according to claim 1, which is characterized in that also
Including S0: the training deep learning model；
The S0 includes:
S0-1: carrying out described image acquisition to multiple sample fronts, obtains multiple multiple described sample direct pictures；
S0-2: carrying out described image acquisition to multiple sample reverse side, obtains multiple multiple described sample verso images；
S0-3: carrying out image preprocessing to multiple multiple described sample direct pictures and multiple multiple described sample verso images,
Obtain multiple sample pretreatment images；
S0-4: multiple sample pretreatment images are merged using the multichannel, multiple multichannels is obtained and melts
Close image；
S0-5: multiple Multichannel fusion images are imported in complex background Texture classification network model generating algorithm, are obtained
The complex background Texture classification network model；
S0-6: multiple Multichannel fusion images and the complex background Texture classification network model are imported into defect recognition
Divide in network model generating algorithm, obtains the segmentation network model of the defect recognition；
S0-7: by multiple Multichannel fusion images and the complex background Texture classification network model, the defect recognition
Segmentation network model import in defect recognition detection network model generating algorithm, obtain defect recognition detection network mould
Type；
S0-8: merge the sorter network model, the defect recognition that export the complex background texture segmentation network model and
The defect recognition detects network model.
5. a kind of film surface defects detection system based on deep learning is based on deep using one kind described in Claims 1-4 4
Spend the film surface defects detection method of study, which is characterized in that including image capture module, image pre-processing module, image
Fusion Module and image detection module；
Described image acquisition module includes conveyer belt, light source, camera, camera frame, for acquiring direct picture and verso images；
Described image preprocessing module includes image calibration module and image cutting module, and described image preprocessing module connects institute
Image capture module is stated, the direct picture and the verso images are received and generates pretreatment image；
Described image Fusion Module is the computer program based on Multichannel fusion, and described image Fusion Module connects described image
Preprocessing module receives the pretreatment image and generates blending image；
Described image detection module includes profile detection module, the sorter network model module of complex background texture, defect recognition
Segmentation network model module and defect recognition detect network model module, described image detection module connect described image fusion
Module receives the blending image and generates image detection result.
6. according to right to go 5 described in a kind of film surface defects detection system based on deep learning, which is characterized in that institute
Stating image calibration module is one of computer program based on image calibration algorithm or a variety of；
Described image cutting module is the computer program based on foreground detection algorithm；
The profile detection module is the computer program based on profile appearances recognizer.
7. a kind of film surface defects detection system based on deep learning according to claim 5, which is characterized in that also
Including deep learning model generation module；
The deep learning model generation module includes the sorter network model generation module of complex background texture, defect recognition
Divide network model generation module and defect recognition detects network model generation module.
8. a kind of film surface defects detection system based on deep learning according to claim 7, which is characterized in that institute
The sorter network model generation module for stating complex background texture is based on complex background Texture classification network model generating algorithm
Computer program；
The segmentation network model generation module of the defect recognition is the segmentation network model generating algorithm based on defect recognition
Computer program；
The defect recognition detection network model generation module is the calculating that network model generating algorithm is detected based on defect recognition
Machine program.
9. a kind of film surface defects detection system based on deep learning according to claim 5, which is characterized in that institute
It states profile detection module and receives the pretreatment image, generate image outline testing result；
The sorter network model module of the complex background texture receives described image contour detecting as a result, generating image background line
Manage testing result；
The segmentation network model module of the defect recognition receives described image background texture testing result, generates image deflects and knows
Other testing result；
The defect recognition detection network model module receives described image defect recognition testing result, generates described image defect
Grade separation testing result, and by described image background texture testing result, described image defect recognition testing result and described
Image deflects grade separation testing result merges output.
10. a kind of Optical Coatings Surface defect detection equipment of deep learning, it is characterised in that including memory, processor；
It is stored at least one instruction, at least a Duan Chengxu, code set or instruction set in the memory, described at least one
Instruction, at least a Duan Chengxu, code set or instruction set are executed to run any institute of claim 5 to 9 by the processor
A kind of Optical Coatings Surface defect detecting system for the deep learning stated is to realize a kind of any depth of Claims 1-4
Spend the Optical Coatings Surface defect inspection method of study.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910694129.XA CN110473179B (en) | 2019-07-30 | 2019-07-30 | Method, system and equipment for detecting surface defects of thin film based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910694129.XA CN110473179B (en) | 2019-07-30 | 2019-07-30 | Method, system and equipment for detecting surface defects of thin film based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110473179A true CN110473179A (en) | 2019-11-19 |
| CN110473179B CN110473179B (en) | 2022-03-25 |
Family
ID=68509884
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910694129.XA Active CN110473179B (en) | 2019-07-30 | 2019-07-30 | Method, system and equipment for detecting surface defects of thin film based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110473179B (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111145145A (en) * | 2019-12-10 | 2020-05-12 | 太原科技大学 | Image surface defect detection method based on MobileNet |
| CN111179263A (en) * | 2020-01-06 | 2020-05-19 | 宜通世纪物联网研究院(广州)有限公司 | Industrial image surface defect detection model, method, system and device |
| CN111753707A (en) * | 2020-06-19 | 2020-10-09 | 上海交通大学 | Method and system for detecting imperfect grains of granular crops |
| CN111798408A (en) * | 2020-05-18 | 2020-10-20 | 中国科学院宁波工业技术研究院慈溪生物医学工程研究所 | Endoscope interference image detection and grading system and method |
| CN113092489A (en) * | 2021-05-20 | 2021-07-09 | 鲸朵(上海)智能科技有限公司 | System and method for detecting appearance defects of battery |
| CN113781430A (en) * | 2021-09-09 | 2021-12-10 | 北京云屿科技有限公司 | Glove surface defect detection method and system based on deep learning |
| CN113933296A (en) * | 2021-03-15 | 2022-01-14 | 住华科技股份有限公司 | Method, device and system for detecting optical film |
| CN114115666A (en) * | 2021-11-26 | 2022-03-01 | 长江存储科技有限责任公司 | Semiconductor structure detection method and device |
| CN114998192A (en) * | 2022-04-19 | 2022-09-02 | 深圳格芯集成电路装备有限公司 | Defect detection method, device and equipment based on deep learning and storage medium |
| CN116993704A (en) * | 2023-08-10 | 2023-11-03 | 苏州苏瑞膜纳米科技有限公司 | Reverse osmosis membrane defect detection system, reverse osmosis membrane defect detection method, storage medium and computer |
| CN117114140A (en) * | 2023-09-12 | 2023-11-24 | 深圳大学 | Packaging method, device, equipment and medium of distributed machine learning algorithm library |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102854194A (en) * | 2012-09-11 | 2013-01-02 | 中南大学 | Object surface defect detection method and apparatus based on linear array CCD |
| CN103439338A (en) * | 2013-08-30 | 2013-12-11 | 无锡金视界科技有限公司 | Classification method for film defects |
| CN106872488A (en) * | 2017-04-20 | 2017-06-20 | 广东振华科技股份有限公司 | A kind of double surface defect visible detection methods of rapid large-area transparent substrate and device |
| CN107402216A (en) * | 2017-07-07 | 2017-11-28 | 深圳市维图视技术有限公司 | Coated products detecting system and method |
| CN107462580A (en) * | 2016-06-02 | 2017-12-12 | 住友化学株式会社 | Defect inspecting system, film manufacturing device and defect detecting method |
| CN108108768A (en) * | 2017-12-29 | 2018-06-01 | 清华大学 | Photovoltaic glass defect classification method and device based on convolutional neural networks |
| US20180293721A1 (en) * | 2017-04-07 | 2018-10-11 | Kla-Tencor Corporation | Contour based defect detection |
| KR20190037889A (en) * | 2017-09-29 | 2019-04-08 | 주식회사 엘지씨엔에스 | Method and system for detecting defect in produnt |
| CN109712117A (en) * | 2018-12-11 | 2019-05-03 | 重庆信息通信研究院 | Lightweight TFT-LCD mould group scratch detection method based on convolutional neural networks |
| CN109946303A (en) * | 2017-12-21 | 2019-06-28 | 东友精细化工有限公司 | Check device and method |
-
2019
- 2019-07-30 CN CN201910694129.XA patent/CN110473179B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102854194A (en) * | 2012-09-11 | 2013-01-02 | 中南大学 | Object surface defect detection method and apparatus based on linear array CCD |
| CN103439338A (en) * | 2013-08-30 | 2013-12-11 | 无锡金视界科技有限公司 | Classification method for film defects |
| CN107462580A (en) * | 2016-06-02 | 2017-12-12 | 住友化学株式会社 | Defect inspecting system, film manufacturing device and defect detecting method |
| US20180293721A1 (en) * | 2017-04-07 | 2018-10-11 | Kla-Tencor Corporation | Contour based defect detection |
| CN106872488A (en) * | 2017-04-20 | 2017-06-20 | 广东振华科技股份有限公司 | A kind of double surface defect visible detection methods of rapid large-area transparent substrate and device |
| CN107402216A (en) * | 2017-07-07 | 2017-11-28 | 深圳市维图视技术有限公司 | Coated products detecting system and method |
| KR20190037889A (en) * | 2017-09-29 | 2019-04-08 | 주식회사 엘지씨엔에스 | Method and system for detecting defect in produnt |
| CN109946303A (en) * | 2017-12-21 | 2019-06-28 | 东友精细化工有限公司 | Check device and method |
| CN108108768A (en) * | 2017-12-29 | 2018-06-01 | 清华大学 | Photovoltaic glass defect classification method and device based on convolutional neural networks |
| CN109712117A (en) * | 2018-12-11 | 2019-05-03 | 重庆信息通信研究院 | Lightweight TFT-LCD mould group scratch detection method based on convolutional neural networks |
Non-Patent Citations (2)
| Title |
|---|
| MUHAMMAD FAIZAN SHIRAZI等: "Fast Industrial Inspection of Optical Thin Film Using Optical Coherence Tomography", 《SENSOR》 * |
| 张毅 等: "基于优化背景差分法光学薄膜表面检测系统", 《控制工程》 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111145145B (en) * | 2019-12-10 | 2023-04-07 | 太原科技大学 | Image surface defect detection method based on MobileNet |
| CN111145145A (en) * | 2019-12-10 | 2020-05-12 | 太原科技大学 | Image surface defect detection method based on MobileNet |
| CN111179263A (en) * | 2020-01-06 | 2020-05-19 | 宜通世纪物联网研究院(广州)有限公司 | Industrial image surface defect detection model, method, system and device |
| CN111179263B (en) * | 2020-01-06 | 2023-10-13 | 广东宜通联云智能信息有限公司 | Industrial image surface defect detection model, method, system and device |
| CN111798408A (en) * | 2020-05-18 | 2020-10-20 | 中国科学院宁波工业技术研究院慈溪生物医学工程研究所 | Endoscope interference image detection and grading system and method |
| CN111753707A (en) * | 2020-06-19 | 2020-10-09 | 上海交通大学 | Method and system for detecting imperfect grains of granular crops |
| CN113933296A (en) * | 2021-03-15 | 2022-01-14 | 住华科技股份有限公司 | Method, device and system for detecting optical film |
| CN113092489A (en) * | 2021-05-20 | 2021-07-09 | 鲸朵(上海)智能科技有限公司 | System and method for detecting appearance defects of battery |
| CN113781430B (en) * | 2021-09-09 | 2023-08-25 | 北京云屿科技有限公司 | Glove surface defect detection method and system based on deep learning |
| CN113781430A (en) * | 2021-09-09 | 2021-12-10 | 北京云屿科技有限公司 | Glove surface defect detection method and system based on deep learning |
| CN114115666A (en) * | 2021-11-26 | 2022-03-01 | 长江存储科技有限责任公司 | Semiconductor structure detection method and device |
| CN114115666B (en) * | 2021-11-26 | 2024-04-16 | 长江存储科技有限责任公司 | Semiconductor structure detection method and device |
| CN114998192A (en) * | 2022-04-19 | 2022-09-02 | 深圳格芯集成电路装备有限公司 | Defect detection method, device and equipment based on deep learning and storage medium |
| CN114998192B (en) * | 2022-04-19 | 2023-05-30 | 深圳格芯集成电路装备有限公司 | Defect detection method, device, equipment and storage medium based on deep learning |
| CN116993704A (en) * | 2023-08-10 | 2023-11-03 | 苏州苏瑞膜纳米科技有限公司 | Reverse osmosis membrane defect detection system, reverse osmosis membrane defect detection method, storage medium and computer |
| CN117114140A (en) * | 2023-09-12 | 2023-11-24 | 深圳大学 | Packaging method, device, equipment and medium of distributed machine learning algorithm library |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110473179B (en) | 2022-03-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110473179A (en) | A kind of film surface defects detection method, system and equipment based on deep learning | |
| Ren et al. | State of the art in defect detection based on machine vision | |
| CN111060601B (en) | Weld ultrasonic phased array detection data intelligent analysis method based on deep learning | |
| CN111189837B (en) | Cigarette appearance online detection method and device | |
| CN103286081B (en) | Monocular multi-perspective machine vision-based online automatic sorting device for steel ball surface defect | |
| CN102288613B (en) | Surface defect detecting method for fusing grey and depth information | |
| CN109001230A (en) | Welding point defect detection method based on machine vision | |
| CN110473178A (en) | A kind of open defect detection method and system based on multiple light courcess fusion | |
| CN104977313A (en) | Method and device for detecting and identifying X-ray image defects of welding seam | |
| CN107966454A (en) | A kind of end plug defect detecting device and detection method based on FPGA | |
| CN111951249A (en) | Mobile phone light guide plate defect visual detection method based on multitask learning network | |
| CN103324937A (en) | Method and device for labeling targets | |
| CN111932511B (en) | Electronic component quality detection method and system based on deep learning | |
| CN113421230B (en) | Visual detection method for defects of vehicle-mounted liquid crystal display light guide plate based on target detection network | |
| WO2023168972A1 (en) | Linear array camera-based copper surface defect detection method and apparatus | |
| CN112270659A (en) | Rapid detection method and system for surface defects of pole piece of power battery | |
| CN109283182A (en) | A kind of detection method of battery welding point defect, apparatus and system | |
| CN106872488A (en) | A kind of double surface defect visible detection methods of rapid large-area transparent substrate and device | |
| CN105606628A (en) | Optical lens detecting system and method | |
| CN115861170A (en) | Surface defect detection method based on improved YOLO V4 algorithm | |
| CN114359235A (en) | Wood surface defect detection method based on improved YOLOv5l network | |
| CN108889635A (en) | A kind of online visible detection method of pop can manufacturing defect | |
| CN109685756A (en) | Image feature automatic identifier, system and method | |
| CN109241948A (en) | A kind of NC cutting tool visual identity method and device | |
| CN114881987A (en) | Improved YOLOv 5-based hot-pressing light guide plate defect visual detection 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 | ||
| CP03 | Change of name, title or address | ||
| CP03 | Change of name, title or address |
Address after: Room 709-3, Building 2, No. 21 Intercity Road, High tech Zone, Suzhou City, Jiangsu Province, 215000 Patentee after: Suzhou Shenshi Information Technology Co.,Ltd. Country or region after: China Address before: 200241, room 1027, building B, 555 Dongchuan Road, Shanghai, Minhang District Patentee before: SHANGHAI DEEPSIGHT INFORMATION TECHNOLOGY CO.,LTD. Country or region before: China |