CN109146873A - A kind of display screen defect intelligent detecting method and device based on study - Google Patents
A kind of display screen defect intelligent detecting method and device based on study Download PDFInfo
- Publication number
- CN109146873A CN109146873A CN201811024270.0A CN201811024270A CN109146873A CN 109146873 A CN109146873 A CN 109146873A CN 201811024270 A CN201811024270 A CN 201811024270A CN 109146873 A CN109146873 A CN 109146873A
- Authority
- CN
- China
- Prior art keywords
- result
- network parameter
- defect
- feedback
- default network
- 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
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
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30121—CRT, LCD or plasma display
Landscapes
- Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
This application provides a kind of display screen defect intelligent detecting method and device based on study, wherein the described method includes: obtaining image to be detected of screen；According to default network parameter, the suspected defects information of described image to be detected is determined；According to the suspected defects information, defect characteristic numerical value is extracted；According to the defect characteristic numerical value, final detection result is determined；Summarizing all final detection results is results set；The final detection result in the results set is randomly selected as sampling observation sample, and detects the sampling observation sample again, determines that the testing result result different from the final detection result is result to be feedback；Feed back described to be feedback as a result, being learnt according to the result to be feedback and generating next default network parameter；According to next default network parameter, the default network parameter is updated, and according to updated default network parameter, detects next image to be detected.
Description
Technical field
This application involves screen detection technique field more particularly to a kind of display screen defect intelligent measurement sides based on study
Method and device.
Background technique
One of the emerging strategic industry that " novel display industry " is supported as the Chinese government being helped in the positive of government in recent years
It holds down and achieves significant progress.Currently, China's liquid crystal display industry be undergoing can often expand, shipment increase and technology mention
The stage of liter.Display screen defects detection is the important link that display screen manufacturing enterprise guarantees shipment quality.Currently, screen detection method
Mainly there are artificial detection and AOI equipment to detect two kinds.Wherein, AOI equipment detection method gradually substitute traditional artificial detection at
For mainstream detection method, but it is also faced with that defect benchmark is unstable, defect type is complicated, the detection parameters tune of detection device
Reduction is miscellaneous, equipment debugging takes a long time and the problems such as to high-incidence defect response lag.
The original visible detection method used in AOI detection device is mainly made of four parts, respectively image acquisition part
Divide, defect extracts part, signature analysis part and data summarization part.Wherein, defect extracts part and mainly uses Threshold segmentation
Image processing method is needed for different type defect and scene using different extracting methods, poor robustness；Signature analysis portion
Divide the method using manually given threshold, according to the principle for being determined as defect when defect characteristic value is greater than threshold value, still
Parameter adjustment is complicated, and is affected by artificial experience, responds slower.
Summary of the invention
This application provides a kind of display screen defect intelligent detecting method and device based on study, to solve existing screen
The problem of defect inspection method is cumbersome, low-response.
The application first aspect provides a kind of display screen defect intelligent detecting method based on study, the method packet
It includes:
Obtain image to be detected of screen；
According to default network parameter, the suspected defects information of described image to be detected is determined, the suspected defects information is extremely
It less include: suspected defects region and corresponding suspected defects type；
According to the suspected defects information, defect characteristic numerical value is extracted；
According to the defect characteristic numerical value, determine that final detection result, the final detection result include at least: suspicious to lack
It falls into information and determines result and the suspected defects information；
Summarizing all final detection results is results set；
The final detection result in the results set is randomly selected as sampling observation sample, and detects the sampling observation again
Sample determines that the testing result result different from the final detection result is result to be feedback；
Feed back described to be feedback as a result, being learnt according to the result to be feedback and generating next default network parameter；
According to next default network parameter, the default network parameter is updated, and according to updated default network
Parameter detects next image to be detected.
With reference to first aspect, the first in first aspect can be described according to the defect characteristic numerical value in realization mode,
The specific steps for determining final detection result include:
Compare the defect characteristic numerical value and default characteristics defect numerical value；
It determines and is greater than the suspected defects area corresponding to the defect characteristic numerical value of the default characteristics defect numerical value
Domain is machine examination defect；
Determine be less than or equal to the default characteristics defect numerical value the defect characteristic numerical value corresponding to it is described can
Doubting defect is to arrange to doubt defect；
Determine that above-mentioned judging result is final detection result.
With reference to first aspect, described to summarize all final detections in second of achievable mode of first aspect
As a result include: for the specific steps of results set
It marks and numbers the final detection result；
Summarizing all numbers is results set, and the results set includes: that the number is tied with the final detection
The corresponding relationship of fruit and the corresponding final detection result of the number.
With reference to first aspect, the third in first aspect can be described to randomly select the results set in realization mode
In the final detection result be inspect by random samples sample specific steps include:
Randomly select the number；
According to the number and the corresponding relationship of the final detection result, obtain the corresponding final detection result and
The corresponding suspected defects region of the final detection result；
Determine the suspected defects region for sampling observation sample.
With reference to first aspect, in the 4th kind of achievable mode of first aspect, the feedback is described to be feedback as a result, root
Learn according to the result to be feedback and the specific steps for generating next default network parameter include:
According to pre-set dimension, the defects of sampling observation pattern detection result described in the result to be feedback image is extracted, is made
For defect sample image；
Mark the defects of defect sample image region；
According to annotation results, mark file is generated；
Divide network using training Framework and deep learning, learn the mark file and generates next defect extraction
Network parameter.
With reference to first aspect, described according to next default network in the 5th kind of achievable mode of first aspect
Parameter, the specific steps for updating the default network parameter include:
It the defects of extracts and rejects the default network parameter and extract network parameter section；
It updates next defect and extracts network parameter to the default network parameter, form next default network parameter.
With reference to first aspect, in the 6th kind of achievable mode of first aspect, the feedback is described to be feedback as a result, root
Learn according to the result to be feedback and the specific steps for generating next default network parameter include:
It obtains and summarizes the defects of the corresponding testing result of sampling observation sample described in the result to be feedback character numerical value
With the defects of final detection result character numerical value, obtain summarizing data；
Using training Framework and deep learning network structure, summarizes data described in study, generate next defect analysis
Network parameter.
With reference to first aspect, described according to next default network in the 7th kind of achievable mode of first aspect
Parameter, the specific steps for updating the default network parameter include:
The defects of extract and reject the default network parameter analysis network parameter section；
Next defect analysis network parameter is updated to the default network parameter, forms next default network parameter.
The application second aspect provides a kind of display screen defect intelligent detection device based on study, described device packet
It includes:
Image acquisition unit, for obtaining image to be detected of screen；
Suspected defects information determination unit, for according to network parameter is preset, determining that the suspicious of described image to be detected lacks
Information is fallen into, the suspected defects information includes at least: suspected defects region and corresponding suspected defects type；
Characteristics extraction unit, for extracting defect characteristic numerical value according to the suspected defects information；
Final detection result determination unit, for according to the defect characteristic numerical value, determining final detection result, it is described most
Whole testing result includes at least: suspected defects information determines result；
Collection unit is results set for summarizing all final detection results；
Result determination unit to be feedback；The final detection result in the results set is randomly selected as sampling observation sample
This, and the sampling observation sample is detected again, determine that the testing result result different from the final detection result is knot to be feedback
Fruit；
Result feedback unit, for feeding back described to be feedback as a result, learning and generating next according to the result to be feedback
Default network parameter；
Network parameter updating unit, for updating the default network parameter according to next default network parameter, and
According to updated default network parameter, next image to be detected is detected.
By the above technology it is found that this application provides a kind of display screen defect intelligent detecting method and dress based on study
It sets, wherein the described method includes: obtaining image to be detected of screen；
According to default network parameter, the suspected defects information of described image to be detected is determined, the suspected defects information is extremely
It less include: suspected defects region and corresponding suspected defects type；According to the suspected defects information, defect characteristic number is extracted
Value；According to the defect characteristic numerical value, determine that final detection result, the final detection result include at least: suspected defects letter
Breath determines result and the suspected defects information；Summarizing all final detection results is results set；It randomly selects described
The final detection result in results set is sampling observation sample, and detects the sampling observation sample again, determine testing result with
The different result of the final detection result is result to be feedback；It feeds back described to be feedback as a result, according to the result to be feedback
Learn and generates next default network parameter；According to next default network parameter, the default network parameter, and root are updated
According to updated default network parameter, next image to be detected is detected.It, can be according to warp when detection device carries out primary detection
It tests or history detection records, one group of default network parameter is preset, to carry out first to the defects of image to be detected
Secondary standard analysis, primarily determines those suspected defects, obtains suspected defects information.It extracts in suspected defects information about those suspected defects
Defect characteristic numerical value, and according to defect characteristic numerical value carry out second of benchmark analysis, further judge those suspected defects whether be
The defect that detection device is authenticated, and using judging result as final detection result.In whole final detection results carry out with
Machine extracts, and finds corresponding sampling observation sample, is detected again to sampling observation sample using artificial or other methods, to sentence
Surely whether sampling observation sample is really defect, and is which kind of defect type etc., and is judged and testing result obtained by detection device
Similarity, as result to be feedback.Result to be feedback is fed back into defect extraction and defect analysis part respectively, and according to anti-
It presents result and carries out deep learning, generate new, more suitably default network parameter and be used for as next default network parameter
Screen defects detection next time.And the learning process of next default network parameter of above-mentioned elaboration, in the detection of every secondary screen
Can all occur in the process, the testing result that can be detected to every secondary screen is all fed back and learnt, and then continues to optimize default
Network parameter.The application can be directly using resulting next default network parameter be learnt, and detection device can be according to next pre-
If network parameter directly carries out defect extraction to image to be detected, instead of original various image segmentation processing methods, it is suitable for multiple
The detection of miscellaneous defect type and different scenes, improves the robustness of detection method；Moreover, being detected by constantly feedback learning
As a result, being continued to optimize in turn to default network parameter, detection device can be enabled to reach efficient, intelligent detection effect, had
Effect solves the problems, such as parameter adjustment complexity and low-response.
Detailed description of the invention
In order to illustrate more clearly of the technical solution of the application, letter will be made to attached drawing needed in the embodiment below
Singly introduce, it should be apparent that, for those of ordinary skills, without any creative labor,
It is also possible to obtain other drawings based on these drawings.
Fig. 1 is a kind of flow chart of the display screen defect intelligent detecting method based on study provided by the embodiments of the present application；
Fig. 2 is the schematic diagram of defects detection result before and after a kind of study provided by the embodiments of the present application, wherein left figure be to
Detection image, right figure are detection result image after study；
Fig. 3 is a kind of method flow diagram of determining final detection result provided by the embodiments of the present application；
Fig. 4 is a kind of method flow diagram for generating results set provided by the embodiments of the present application；
Fig. 5 is a kind of method flow diagram of determining sampling observation sample provided by the embodiments of the present application；
Fig. 6 is a kind of method flow diagram for generating next default network parameter provided by the embodiments of the present application；
Fig. 7 is that network structure is used in a kind of defect extraction part deep learning detection provided by the embodiments of the present application；
Fig. 8 is a kind of method flow diagram for updating default network parameter provided by the embodiments of the present application；
Fig. 9 is another method flow diagram for generating next default network parameter provided by the embodiments of the present application；
Figure 10 is that network structure is used in defect analysis partial depth learning data provided by the embodiments of the present application analysis；
Figure 11 is another method flow diagram for updating default network parameter provided by the embodiments of the present application；
Figure 12 is that a kind of structure of the display screen defect intelligent detection device based on study provided by the embodiments of the present application is shown
It is intended to.
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
Whole 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.
Referring to Fig. 1, being a kind of display screen defect intelligent detecting method based on study provided by the embodiments of the present application
Flow chart, which comprises
Step 100, the image to be detected for obtaining screen；
Step 200, basis preset network parameter, determine the suspected defects information of described image to be detected, described suspicious scarce
Sunken information includes at least: suspected defects region and corresponding suspected defects type；
Step 300, according to the suspected defects information, extract defect characteristic numerical value；
Step 400, according to the defect characteristic numerical value, determine that final detection result, the final detection result are at least wrapped
Include: suspected defects information determines result and the suspected defects information；
Step 500, to summarize all final detection results be results set；
Step 600 is randomly selected the final detection result in the results set as sampling observation sample, and is detected again
The sampling observation sample determines that the testing result result different from the final detection result is result to be feedback；
Step 700, feedback are described to be feedback as a result, being learnt according to the result to be feedback and generating next default network ginseng
Number；
Step 800, according to next default network parameter, update the default network parameter, and according to updated
Default network parameter, detects next image to be detected.
In use, detection device obtains image to be detected of screen first to display screen shot detected.Specifically
Ground can be used industrial camera and acquire display image in real time, and image to be detected can be whole display images, be also possible to
Representative part display image.
Image to be detected that detection device will acquire is sent to defect and extracts part progress defect primary detection, when detection fills
When setting progress primary detection, one group of default network parameter can be preset according to artificial experience or history detection record, from
And the analysis of first time benchmark is carried out to image to be detected, it primarily determines those suspected defects, obtains suspected defects information, including at least can
Defect area and corresponding suspected defects type are doubted, such as according to the gray value of image, divides suspected defects region, then determines
Suspected defects type is bubble, scratch, is mingled with.
Image to be detected after detection device analyzes first time benchmark is sent to defect analysis part, by defect analysis
It is detected again for suspected defects region part.And according to suspected defects region and suspected defects type, doubtful lack is obtained
Sunken defect characteristic numerical value, and second of benchmark analysis is carried out according to defect characteristic numerical value, further whether judge those suspected defects
For real defect defined in detection device, corresponding final detection result is obtained, for example, if suspected defects are defect, then most
Whole testing result is OK；If suspected defects are not defect, final detection result NG.
Whole final detection results are summarized for results set, it is seen then that second of basis point is included at least in results set
It analyses and analyzes first time benchmark suspected defects information corresponding to made judging result and judging result.
The method of randomly selecting can exclude artificial subjective consciousness interference, avoid the contingency and particularity that extract result, because
This, is randomly selected in whole final detection results, obtains sampling observation sample.Artificial or its other party is utilized to sampling observation sample
Method re-starts detection, therefore, it is determined that whether sampling observation sample is actual defects, and is which kind of defect type etc., obtains sampling observation knot
Fruit, and will test the result result different from final detection result and be determined as result to be feedback, it is seen then that result to be feedback is at least wrapped
Include the erroneous judgement image and its data such as corresponding defect characteristic value numerical value of second benchmark analysis detection error.
Examined is fed back to defect extraction and defect analysis part respectively, and depth is carried out according to result to be feedback
It practises, new, more suitably default network parameter is generated, as next default network parameter, for screen defect next time
Detection.And the learning process of next default network parameter of above-mentioned elaboration, can all occur during the detection of every secondary screen, it can
It is all fed back and is learnt with the testing result detected to every secondary screen, and then continue to optimize default network parameter.
The application can be directly using resulting next default network parameter be learnt, as shown in Fig. 2, detection device can root
Defect extraction directly is carried out to image to be detected on the left of Fig. 2 according to next default network parameter, instead of original various image segmentations
Processing method obtains the image for indicating defect area on the right side of Fig. 2, is suitable for the inspection of complicated defect type and different scenes
It surveys, improves the robustness of detection method；Moreover, by constantly feedback learning testing result, so to default network parameter into
Row is continued to optimize, and detection device can be enabled to reach efficient, intelligent detection effect, effectively solves parameter adjustment complexity and low-response
The problem of.
Referring to Fig. 3, being a kind of method flow diagram of determining final detection result provided by the embodiments of the present application.In this Shen
Please be described according to the defect characteristic numerical value in embodiment, determine that the specific steps of final detection result include:
Step 401, the comparison defect characteristic numerical value and default characteristics defect numerical value；
Step 402, determine be greater than the default characteristics defect numerical value the defect characteristic numerical value corresponding to it is described can
Doubting defect area is machine examination defect；
Step 403 determines corresponding to the defect characteristic numerical value for being less than or equal to the default characteristics defect numerical value
The suspected defects be arrange doubt defect；
Step 404 determines that above-mentioned judging result is final detection result.
By the defect characteristic numerical value obtained according to suspected defects obtained by first time benchmaring and pre-set default spy
Sign defect numerical value be compared, the default characteristics defect numerical value can according to artificial experience such as defect numerical value of the same race
Relative recording setting.And will be greater than suspected defects region corresponding to default characteristics defect numerical value and be defined as machine examination defect, it is described
The defect that machine examination defect is finally assert for detection device used by the application, is not exclusively equal to actual defects；It will be less than
Or be defined as arranging doubtful defect equal to suspected defects region corresponding to default characteristics defect numerical value, it is to pass through that the row, which doubts defect,
It crosses second of benchmaring and determines the defect that defect obtained by first time benchmaring is not assert by detection device.It needs second
Secondary standard detection gained, the either judging result of approval or opposition first time benchmaring result, are defined as most final inspection
Survey result.
Referring to Fig. 4, a kind of method flow diagram for generating results set provided by the embodiments of the present application.Implement in the application
In example, all final detection results that summarizes are that the specific steps of results set include:
Step 501 marks and numbers the final detection result；
Step 502, summarize all it is described number be results set, the results set include: it is described number with it is described most
The corresponding relationship of whole testing result and the corresponding final detection result of the number.
Final detection result includes at least suspected defects information and determines result and suspected defects information, it is seen then that comprising a large amount of
Data drag the slow speed of service if will certainly consume biggish computing resource directly using final detection result as object.Therefore,
The present embodiment provides one kind by marking and numbering final detection result, and one kind will be established between final detection result and number
Corresponding relationship to represent final detection result with number, and then will summarize object and replace with number by final detection result, this
Sample can greatly reduce data entrained when operation, can effectively improve the speed of service.
All numbers are aggregated to form results set, it is known that, number, number and final detection are included at least in results set
As a result corresponding relationship, final detection result between.As it can be seen that the number and most final inspection can be passed through when extracting a number
The corresponding relationship for surveying result, finds corresponding final detection result.
Referring to Fig. 5, for a kind of method flow diagram of determining sampling observation sample provided by the embodiments of the present application.In the application reality
It applies in example, the final detection result randomly selected in the results set is that the specific steps of sampling observation sample include:
Step 601 randomly selects the number；
Step 602, according to the number and the corresponding relationship of the final detection result, obtain the corresponding most final inspection
Survey result and the corresponding suspected defects region of the final detection result；
Step 603 determines the suspected defects region for sampling observation sample.
In the results set being aggregated to form by number, arbitrary number is randomly selected, extracting quantity can be according to practical feelings
Condition is formulated, and extraction quantity is more, and subsequent learning process quality is higher.According to the corresponding pass between number and final detection result
System, available final detection result corresponding with number, it might even be possible to obtain corresponding suspected defects region and suspicious
Sub- image to be detected where defect area, described sub- image to be detected are in whole image to be detected with the suspected defects area
Centered on domain, pre-determined distance is part image to be detected that radius is formed, by suspected defects region or defect area institute
Image to be detected as sampling observation sample, it is basic to provide detection for follow-up work.
Referring to Fig. 6, being a kind of method flow diagram for generating next default network parameter provided by the embodiments of the present application.?
In the embodiment of the present application, the feedback is described to be feedback as a result, being learnt according to the result to be feedback and generating next default net
The specific steps of network parameter include:
Step 711, according to pre-set dimension, extract the defects of sampling observation pattern detection result described in the result to be feedback
Image, as defect sample image；
The defects of step 712, mark defect sample image region；
Step 713, according to annotation results, generate mark file；
Step 714 divides network using training Framework and deep learning, learns the mark file and generates next
Defect extracts network parameter.
Sampling observation sample can obtain the testing result of real defect according to artificial or other detection methods, firstly, according to
Pre-set dimension comes out the defects of testing result image zooming-out, and pre-set dimension can according to actual needs or experience
Value setting, such as 128 × 128 or 256 × 256, but it is not limited only to both.
Then, using the defect image extracted as defect sample image, and to the defects of defect sample image region
It is labeled, at this point it is possible to be labeled by the way of manually marking using ImageLabel (picture annotation tool) etc.,
And generate the mark file of the formats such as json, txt, xml.
Finally, dividing network using the training Framework of deep learning, deep learning, mark file is iterated
It practises, and generates model parameter.Specifically, referring to Fig. 7, study part uses FCN (Fully Convolutional
Networks, full convolutional network structure), network by 6 layers of convolutional layer Convol1_1, Convol1_2, Convol2_1,
Convol2_2,Convol3_1,Convol3_2；4 layers of pond layer Pool1, Pool2, Pool3, FusePool 1；2 layers of full connection
Layer FC6, FC7；2 layers of warp lamination Deconvol1, Deconvol2 and other auxiliary layers form, provided by the embodiment of the present application
Network amounts to 26 layer network structures, but the limitation of this network number of layers is not limited only in practical operation.Instruction used by deep learning
Practicing frame is the tool software that can be disposed, and is realized using encapsulating in Caffe frame foundation.
Deep learning is carried out to the defect of detection device error in judgement in sampling observation sample, the defect after finally being trained mentions
Take the network parameter of part.
Referring to Fig. 8, for a kind of method flow diagram for updating default network parameter provided by the embodiments of the present application.In this Shen
Please be in embodiment, described according to next default network parameter, the specific steps for updating the default network parameter include:
Step 811 the defects of is extracted and rejects the default network parameter and extracts network parameter section；
Step 812 updates next defect extraction network parameter to the default network parameter, forms next default net
Network parameter.
Default network parameter includes that defect extracts the network parameter of part and the network parameter of defect analysis part, passes through depth
Degree study can targetedly obtain the network of relation parameter that defect extracts part, when updating, need the default net of original first
Defect extracts network parameter section and extracts and reject in network parameter, and the defect obtained after deep learning is then extracted network parameter
It is padded to former default network parameter, forms new default network parameter, i.e., next default network parameter.
Referring to Fig. 9, for another method flow diagram for generating next default network parameter provided by the embodiments of the present application.
In the embodiment of the present application, the feedback is described to be feedback as a result, being learnt according to the result to be feedback and being generated next default
The specific steps of network parameter include:
Step 721 obtains and summarizes and inspects the defects of corresponding testing result of sample described in the result to be feedback by random samples
The defects of character numerical value and final detection result character numerical value obtain summarizing data；
Step 722, using training Framework and deep learning network structure, summarize data described in study, generate next
Defect analysis network parameter.
It will test the defect characteristic numerical value for inspecting sample by random samples and the artificial or detected pumping of other methods that device obtains
The defect characteristic numerical value of sample sheet summarizes, and carries out deep learning and training to data are summarized.
Specifically, referring to Figure 10, using CNN (Convolutional Neural Networks, convolutional network structure),
Network is by 4 layers of convolutional layer Convol1_1, Convol1_2, Convol2_1, Convol2_2；2 layers of pond layer Pool 1, Pool2；
2 layers of full articulamentum FC6, FC7 and other auxiliary layers form, and network provided by the embodiment of the present application amounts to 14 layer network structures,
But the limitation of network number of layers is not limited only in practical operation.Training frame used by deep learning is that the tool that can dispose is soft
Part, and realized using being encapsulated in Caffe frame foundation.
Figure 11 is please referred to, for another method flow diagram for updating default network parameter provided by the embodiments of the present application.?
It is described according to next default network parameter in the embodiment of the present application, update the specific steps packet of the default network parameter
It includes:
Analysis network parameter section the defects of is extracted and reject the default network parameter to step 821；
Step 822 updates next defect analysis network parameter to the default network parameter, forms next default net
Network parameter.
The network of relation parameter that defect analysis part can be targetedly obtained by deep learning is needed when updating
Defect analysis network parameter section in the default network parameter of original is extracted and rejected first, is then lacked what is obtained after deep learning
It falls into analysis network parameter and is padded to former default network parameter, form new default network parameter, i.e., next default network parameter.
Figure 12 is please referred to, is a kind of display screen defect intelligent detection device based on study provided by the embodiments of the present application
Structural schematic diagram.In the embodiment of the present application, described device includes:
Image acquisition unit 1, for obtaining image to be detected of screen；
Suspected defects information determination unit 2, for according to network parameter is preset, determining that the suspicious of described image to be detected lacks
Information is fallen into, the suspected defects information includes at least: suspected defects region and corresponding suspected defects type；
Characteristics extraction unit 3, for extracting defect characteristic numerical value according to the suspected defects information；
Final detection result determination unit 4, it is described for determining final detection result according to the defect characteristic numerical value
Final detection result includes at least: suspected defects information determines result and the suspected defects information；
Collection unit 5 is results set for summarizing all final detection results；
Result determination unit 6 to be feedback；The final detection result in the results set is randomly selected as sampling observation sample
This, and the sampling observation sample is detected again, determine that the testing result result different from the final detection result is knot to be feedback
Fruit；
Result feedback unit 7, for feeding back described to be feedback as a result, learning and generating next according to the result to be feedback
Default network parameter；
Network parameter updating unit 8, for updating the default network parameter according to next default network parameter,
And according to updated default network parameter, next image to be detected is detected.
In a kind of another embodiment of display screen defect intelligent detection device based on study provided by the present application, the dress
It sets further include: comparison unit, for comparing the defect characteristic numerical value and default characteristics defect numerical value；Machine examination defect determines single
Member is greater than the suspected defects region corresponding to the defect characteristic numerical value of the default characteristics defect numerical value for determining
For machine examination defect；It arranges and doubts defect determination unit, for determining that being less than or equal to the described of the default characteristics defect numerical value lacks
Falling into the suspected defects corresponding to character numerical value is to arrange to doubt defect；Final detection result determination unit above-mentioned is sentenced for determining
Disconnected result is final detection result.
In a kind of another embodiment of display screen defect intelligent detection device based on study provided by the present application, the dress
It sets further include: numbered cell, for marking and numbering the final detection result；Collection unit, for summarizing all volumes
It number is results set, the results set includes: the number and the corresponding relationship of the final detection result and the volume
Number corresponding final detection result.
In a kind of another embodiment of display screen defect intelligent detection device based on study provided by the present application, the dress
It sets further include: extracting unit, for randomly selecting the number；Suspected defects area acquisition unit, for according to the number
With the corresponding relationship of the final detection result, obtains the corresponding final detection result and the final detection result is corresponding
The suspected defects region；Sample determination unit is inspected by random samples, for determining the suspected defects region for sampling observation sample.
In a kind of another embodiment of display screen defect intelligent detection device based on study provided by the present application, the dress
It sets further include: defect sample image extraction unit, for extracting and inspecting sample by random samples described in the result to be feedback according to pre-set dimension
The defects of this testing result image, as defect sample image；Unit is marked, for marking in the defect sample image
Defect area；File generating unit, for generating mark file according to annotation results；Learning training unit, for utilizing training
Framework and deep learning divide network, learn the mark file and generate next defect extraction network parameter.
In a kind of another embodiment of display screen defect intelligent detection device based on study provided by the present application, the dress
It sets further include: parameter extraction unit extracts network parameter section for extracting and rejecting the defects of described default network parameter；
Parameter updating unit extracts network parameter to the default network parameter for updating next defect, is formed next default
Network parameter.
In a kind of another embodiment of display screen defect intelligent detection device based on study provided by the present application, the dress
It sets further include: data summarization unit, for obtaining and summarizing the corresponding detection knot of sampling observation sample described in the result to be feedback
The defects of the defects of fruit character numerical value and final detection result character numerical value, obtain summarizing data；Learning training unit is used
In using Framework and deep learning network structure is trained, summarize data described in study, generates next defect analysis network ginseng
Number.
In a kind of another embodiment of display screen defect intelligent detection device based on study provided by the present application, the dress
It sets further include: parameter extraction unit, for extracting and rejecting the defects of described default network parameter analysis network parameter section；
Parameter updating unit is formed next default for updating next defect analysis network parameter to the default network parameter
Network parameter.
It is worth noting that, in the specific implementation, the present invention also provides a kind of computer storage mediums, wherein the computer
Storage medium can be stored with program, which may include the service providing method or use of user identity provided by the invention when executing
Step some or all of in each embodiment of family register method.The storage medium can be magnetic disk, CD, read-only storage note
Recall body (English: read-only memory, abbreviation: ROM) or random access memory (English: random access
Memory, referred to as: RAM) etc..
It is required that those skilled in the art can be understood that the technology in the embodiment of the present invention can add by software
The mode of general hardware platform realize.Based on this understanding, the technical solution in the embodiment of the present invention substantially or
Say that the part that contributes to existing technology can be embodied in the form of software products, which can deposit
Storage is in storage medium, such as ROM/RAM, magnetic disk, CD, including some instructions are used so that computer equipment (can be with
It is personal computer, server or the network equipment etc.) execute certain part institutes of each embodiment of the present invention or embodiment
The method stated.
Those skilled in the art after considering the specification and implementing the invention disclosed here, will readily occur to of the invention its
Its embodiment.This application is intended to cover any variations, uses, or adaptations of the invention, these modifications, purposes or
Person's adaptive change follows general principle of the invention and including the undocumented common knowledge in the art of the present invention
Or conventional techniques.The description and examples are only to be considered as illustrative, and true scope and spirit of the invention are by following
Claim is pointed out.
It should be understood that the application is not limited to the precise structure that has been described above and shown in the drawings, and
And various modifications and changes may be made without departing from the scope thereof.Scope of the present application is only limited by the accompanying claims.
Claims (9)
1. a kind of display screen defect intelligent detecting method based on study, which is characterized in that the described method includes:
Obtain image to be detected of screen；
According to default network parameter, determine that the suspected defects information of described image to be detected, the suspected defects information are at least wrapped
It includes: suspected defects region and corresponding suspected defects type；
According to the suspected defects information, defect characteristic numerical value is extracted；
According to the defect characteristic numerical value, determine that final detection result, the final detection result include at least: suspected defects letter
Breath determines result and the suspected defects information；
Summarizing all final detection results is results set；
Final detection result described in the results set is randomly selected as sampling observation sample, and detects the sampling observation sample again,
Determine that the testing result result different from the final detection result is result to be feedback；
Feed back described to be feedback as a result, being learnt according to the result to be feedback and generating next default network parameter；
According to next default network parameter, the default network parameter is updated, and according to updated default network parameter,
Detect next image to be detected.
2. detection method according to claim 1, which is characterized in that it is described according to the defect characteristic numerical value, it determines most
The specific steps of whole testing result include:
Compare the defect characteristic numerical value and default characteristics defect numerical value；
Determine that being greater than the suspected defects region corresponding to the defect characteristic numerical value of the default characteristics defect numerical value is
Machine examination defect；
It determines described suspicious scarce less than or equal to corresponding to the defect characteristic numerical value of the default characteristics defect numerical value
It falls into arrange and doubting defect；
Determine that above-mentioned judging result is final detection result.
3. detection method according to claim 1, which is characterized in that all final detection results that summarizes are knot
The specific steps of fruit set include:
It marks and numbers the final detection result；
Summarizing all numbers is results set, and the results set includes: the number and the final detection result
Corresponding relationship and the corresponding final detection result of the number.
4. detection method according to claim 3, which is characterized in that it is described randomly select it is described in the results set
Final detection result is that the specific steps of sampling observation sample include:
Randomly select the number；
According to the number and the corresponding relationship of the final detection result, the corresponding final detection result and described is obtained
The corresponding suspected defects region of final detection result；
Determine the suspected defects region for sampling observation sample.
5. detection method according to claim 1, which is characterized in that the feedback is described to be feedback as a result, according to described
The specific steps that result to be feedback learns and generate next default network parameter include:
According to pre-set dimension, the defects of sampling observation pattern detection result described in the result to be feedback image is extracted, as scarce
Fall into sample image；
Mark the defects of defect sample image region；
According to annotation results, mark file is generated；
Divide network using training Framework and deep learning, learn the mark file and generates next defect extraction network
Parameter.
6. detection method according to claim 5, which is characterized in that it is described according to next default network parameter, more
Newly the specific steps of the default network parameter include:
It the defects of extracts and rejects the default network parameter and extract network parameter section；
It updates next defect and extracts network parameter to the default network parameter, form next default network parameter.
7. detection method according to claim 1, which is characterized in that the feedback is described to be feedback as a result, according to described
The specific steps that result to be feedback learns and generate next default network parameter include:
Obtain and summarize described in the result to be feedback the defects of the corresponding testing result of sampling observation sample character numerical value and most
The defects of whole testing result character numerical value, obtains summarizing data；
Using training Framework and deep learning network structure, summarizes data described in study, generate next defect analysis network
Parameter.
8. detection method according to claim 7, which is characterized in that it is described according to next default network parameter, more
Newly the specific steps of the default network parameter include:
The defects of extract and reject the default network parameter analysis network parameter section；
Next defect analysis network parameter is updated to the default network parameter, forms next default network parameter.
9. a kind of display screen defect intelligent detection device based on study, which is characterized in that described device includes:
Image acquisition unit, for obtaining image to be detected of screen；
Suspected defects information determination unit, for according to network parameter is preset, determining the suspected defects letter of described image to be detected
Breath, the suspected defects information include at least: suspected defects region and corresponding suspected defects type；
Characteristics extraction unit, for extracting defect characteristic numerical value according to the suspected defects information；
Final detection result determination unit, for determining final detection result, the most final inspection according to the defect characteristic numerical value
Survey result to include at least: suspected defects information determines result and the suspected defects information；
Collection unit is results set for summarizing all final detection results；
Result determination unit to be feedback；The final detection result in the results set is randomly selected as sampling observation sample, and
Again the sampling observation sample is detected, determines that the testing result result different from the final detection result is result to be feedback；
Result feedback unit, for feeding back described to be feedback as a result, being learnt according to the result to be feedback and being generated next default
Network parameter；
Network parameter updating unit is used for according to next default network parameter, the update default network parameter, and according to
Updated default network parameter, detects next image to be detected.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811024270.0A CN109146873B (en) | 2018-09-04 | 2018-09-04 | Learning-based intelligent detection method and device for defects of display screen |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811024270.0A CN109146873B (en) | 2018-09-04 | 2018-09-04 | Learning-based intelligent detection method and device for defects of display screen |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109146873A true CN109146873A (en) | 2019-01-04 |
| CN109146873B CN109146873B (en) | 2020-12-29 |
Family
ID=64826633
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811024270.0A Active CN109146873B (en) | 2018-09-04 | 2018-09-04 | Learning-based intelligent detection method and device for defects of display screen |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109146873B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111337512A (en) * | 2020-05-22 | 2020-06-26 | 深圳新视智科技术有限公司 | Defect extraction method for AOI defect detection |
| CN111563869A (en) * | 2019-01-25 | 2020-08-21 | 宁波舜宇光电信息有限公司 | Stain testing method for quality inspection of camera module |
| WO2020211823A1 (en) * | 2019-04-18 | 2020-10-22 | 杭州海康威视数字技术股份有限公司 | Model training method and device, and defect detection method and device |
| CN111986195A (en) * | 2020-09-07 | 2020-11-24 | 北京凌云光技术集团有限责任公司 | Appearance defect detection method and system |
| CN112508426A (en) * | 2020-12-15 | 2021-03-16 | 深圳市华星光电半导体显示技术有限公司 | Chip selective inspection method for lighting machine and chip selective inspection lighting machine |
| TWI745767B (en) * | 2019-10-18 | 2021-11-11 | 汎思數據股份有限公司 | Optical inspection secondary image classification method |
| CN114723682A (en) * | 2022-03-22 | 2022-07-08 | 厦门微亚智能科技有限公司 | Product defect management method, system and computer readable storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6408429B1 (en) * | 1997-01-17 | 2002-06-18 | Cognex Corporation | Machine vision system for identifying and assessing features of an article |
| CN106875381A (en) * | 2017-01-17 | 2017-06-20 | 同济大学 | A kind of phone housing defect inspection method based on deep learning |
| CN108230317A (en) * | 2018-01-09 | 2018-06-29 | 北京百度网讯科技有限公司 | Steel plate defect detection sorting technique, device, equipment and computer-readable medium |
| CN108242054A (en) * | 2018-01-09 | 2018-07-03 | 北京百度网讯科技有限公司 | A kind of steel plate defect detection method, device, equipment and server |
| CN108428247A (en) * | 2018-02-27 | 2018-08-21 | 广州视源电子科技股份有限公司 | The detection method and system in bump direction |
-
2018
- 2018-09-04 CN CN201811024270.0A patent/CN109146873B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6408429B1 (en) * | 1997-01-17 | 2002-06-18 | Cognex Corporation | Machine vision system for identifying and assessing features of an article |
| CN106875381A (en) * | 2017-01-17 | 2017-06-20 | 同济大学 | A kind of phone housing defect inspection method based on deep learning |
| CN108230317A (en) * | 2018-01-09 | 2018-06-29 | 北京百度网讯科技有限公司 | Steel plate defect detection sorting technique, device, equipment and computer-readable medium |
| CN108242054A (en) * | 2018-01-09 | 2018-07-03 | 北京百度网讯科技有限公司 | A kind of steel plate defect detection method, device, equipment and server |
| CN108428247A (en) * | 2018-02-27 | 2018-08-21 | 广州视源电子科技股份有限公司 | The detection method and system in bump direction |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111563869A (en) * | 2019-01-25 | 2020-08-21 | 宁波舜宇光电信息有限公司 | Stain testing method for quality inspection of camera module |
| CN111563869B (en) * | 2019-01-25 | 2023-07-21 | 宁波舜宇光电信息有限公司 | Stain test method for quality inspection of camera module |
| WO2020211823A1 (en) * | 2019-04-18 | 2020-10-22 | 杭州海康威视数字技术股份有限公司 | Model training method and device, and defect detection method and device |
| TWI745767B (en) * | 2019-10-18 | 2021-11-11 | 汎思數據股份有限公司 | Optical inspection secondary image classification method |
| CN111337512A (en) * | 2020-05-22 | 2020-06-26 | 深圳新视智科技术有限公司 | Defect extraction method for AOI defect detection |
| CN111337512B (en) * | 2020-05-22 | 2020-09-08 | 深圳新视智科技术有限公司 | Defect extraction method for AOI defect detection |
| CN111986195A (en) * | 2020-09-07 | 2020-11-24 | 北京凌云光技术集团有限责任公司 | Appearance defect detection method and system |
| CN111986195B (en) * | 2020-09-07 | 2024-02-20 | 凌云光技术股份有限公司 | Appearance defect detection method and system |
| CN112508426A (en) * | 2020-12-15 | 2021-03-16 | 深圳市华星光电半导体显示技术有限公司 | Chip selective inspection method for lighting machine and chip selective inspection lighting machine |
| CN112508426B (en) * | 2020-12-15 | 2023-12-01 | 深圳市华星光电半导体显示技术有限公司 | Lighting machine and fragment sampling inspection method thereof |
| CN114723682A (en) * | 2022-03-22 | 2022-07-08 | 厦门微亚智能科技有限公司 | Product defect management method, system and computer readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109146873B (en) | 2020-12-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109146873A (en) | A kind of display screen defect intelligent detecting method and device based on study | |
| CN104616664B (en) | A kind of audio identification methods detected based on sonograph conspicuousness | |
| CN109285139A (en) | A kind of x-ray imaging weld inspection method based on deep learning | |
| CN108846835A (en) | The image change detection method of convolutional network is separated based on depth | |
| CN108229580A (en) | Sugared net ranking of features device in a kind of eyeground figure based on attention mechanism and Fusion Features | |
| CN110378232B (en) | Improved test room examinee position rapid detection method of SSD dual-network | |
| CN106709936A (en) | Single target tracking method based on convolution neural network | |
| CN109598287A (en) | The apparent flaws detection method that confrontation network sample generates is generated based on depth convolution | |
| CN114577812A (en) | Method and system for training efficient quality inspection model based on ultra-small sample | |
| CN104200478B (en) | Low-resolution touch screen image defect detection method based on sparse representation | |
| Gao et al. | A Novel Deep Convolutional Neural Network Based on ResNet‐18 and Transfer Learning for Detection of Wood Knot Defects | |
| CN109191444A (en) | Video area based on depth residual error network removes altering detecting method and device | |
| CN113222913B (en) | Circuit board defect detection positioning method, device and storage medium | |
| CN109410184A (en) | Live streaming pornographic image detection method based on dense confrontation network semi-supervised learning | |
| CN112949408A (en) | Real-time identification method and system for target fish passing through fish channel | |
| CN111127454A (en) | Method and system for generating industrial defect sample based on deep learning | |
| CN109472790A (en) | A kind of machine components defect inspection method and system | |
| CN109685756A (en) | Image feature automatic identifier, system and method | |
| CN110956656A (en) | Spindle positioning method based on depth target detection | |
| KR101782364B1 (en) | Vision inspection method based on learning data | |
| CN110321867A (en) | Shelter target detection method based on part constraint network | |
| CN106528665A (en) | Method and system for searching testing document of AOI equipment | |
| CN108765391A (en) | A kind of plate glass foreign matter image analysis methods based on deep learning | |
| US11727673B1 (en) | Visual analysis method for cable element identification | |
| CN110033443A (en) | A kind of feature extraction network and its defects of display panel 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 | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 100094 Beijing city Haidian District Cui Hunan loop 13 Hospital No. 7 Building 7 room 701 Applicant after: Lingyunguang Technology Co.,Ltd. Address before: 100094 Beijing city Haidian District Cui Hunan loop 13 Hospital No. 7 Building 7 room 701 Applicant before: LUSTER LIGHTTECH GROUP Co.,Ltd. |
|
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 100094 701, 7 floor, 7 building, 13 Cui Hunan Ring Road, Haidian District, Beijing. Applicant after: Lingyunguang Technology Co.,Ltd. Address before: 100094 701, 7 floor, 7 building, 13 Cui Hunan Ring Road, Haidian District, Beijing. Applicant before: Beijing lingyunguang Technology Group Co.,Ltd. Address after: 100094 701, 7 floor, 7 building, 13 Cui Hunan Ring Road, Haidian District, Beijing. Applicant after: Beijing lingyunguang Technology Group Co.,Ltd. Address before: 100094 701, 7 floor, 7 building, 13 Cui Hunan Ring Road, Haidian District, Beijing. Applicant before: LUSTER LIGHTTECH GROUP Co.,Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |