CN110097544A - A kind of display panel open defect detection method - Google Patents
A kind of display panel open defect detection method Download PDFInfo
- Publication number
- CN110097544A CN110097544A CN201910341237.9A CN201910341237A CN110097544A CN 110097544 A CN110097544 A CN 110097544A CN 201910341237 A CN201910341237 A CN 201910341237A CN 110097544 A CN110097544 A CN 110097544A
- Authority
- CN
- China
- Prior art keywords
- display panel
- image
- semantic segmentation
- deep learning
- detection method
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- 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
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- 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
Abstract
The invention belongs to display panel detection technique fields, disclose a kind of display panel open defect detection method, and display panel image enhancement processing is cut to multiple subgraphs；Several subgraphs are labeled with polygon frame using image, semantic segmentation annotation tool, the defects of sub-image region marks；Image pair is formed by the subgraph after subgraph and mark without mark；The part of selection image pair is remaining to be used as test set as training set；Modified deeplab v3+ deep learning semantic segmentation model is trained with training set data, and the model is verified using test set；Modified deeplab v3+ deep learning semantic segmentation model is using SENet sorter network as backbone network；Display panel image to be measured is predicted using trained modified deeplab v3+ deep learning semantic segmentation model, exports shape, the area of target defect.
Description
Technical field
The invention belongs to the automation defect detecting technique fields of display panel, more particularly, to a kind of display panel
Open defect detection method.
Background technique
With the universal of electronic product and quickly update, there is great yield to need on LCD screen, OLED screen curtain
It asks.In screen production process, because of situations such as raw material, production technology, accident, various defects, example are likely to occur on screen
Such as fragmentation, bubble, scuffing, unfilled corner, slight crack, faulty goods will affect its performance or reduce user experience, be not allow flow into
Market, it is therefore desirable to which the display screen of production is detected.
Detection to display screen includes appearance detection and backlight detection, and the existing method of open defect detection includes: that (1) passes
System Automatic Optic Inspection (automatic optics inspection, AOI) method: will first extract ROI region, then with tradition side
The detection of method algorithm, this method are affected by factors such as background, illumination, and to different defects using at different algorithms
Reason, the processing time is long when detecting number of drawbacks.(2) classification and detection method of deep learning: the advantages of these methods is
Mark is simple, and classification only needs that whole image is marked, and detection only needs to mark target defect with rectangle frame, but also because
For this reason, there is any discrepancy between the target area detected and actual target area, can not be with area come control defect
Output.(3) the semantic segmentation method of deep learning: the target detected is pixel scale, can be well from testing result
The shape of defect, the characteristics such as area are judged, its shortcoming is that mark needs very big workload.
Summary of the invention
Aiming at the above defects or improvement requirements of the prior art, the present invention provides a kind of detections of display panel open defect
Method effectively promotes open defect inspection its object is to which the method for deep learning semantic segmentation is used for open defect detection
The accuracy of survey.
To achieve the above object, according to one aspect of the present invention, a kind of display panel open defect detection side is provided
Method includes the following steps:
(1) display panel image is carried out as detection image after enhancing processing, and each detection image is cut to more
A subgraph；
(2) several subgraphs are labeled with polygon frame using image, semantic segmentation annotation tool, in sub-image
Defect area mark；
(3) image pair is formed by the subgraph after subgraph and mark without mark；Make the part for choosing image pair
It is remaining to be used as test set for training set；
(4) deep learning semantic segmentation model is trained with training set data, and the model is verified using test set, until
Reach preset accuracy；
(5) display panel image to be measured is predicted using trained deep learning semantic segmentation model, exports mesh
Mark shape, the area of defect.
Preferably, above-mentioned display panel open defect detection method, deep learning semantic segmentation model use deeplab
V3+ deep learning semantic segmentation model, the deeplab v3+ deep learning semantic segmentation model use SENet sorter network
As backbone network.
Preferably, above-mentioned display panel open defect detection method, the sample among training set or test set are input to
Deeplab v3+ deep learning semantic segmentation model, is first passed through compiler processes, is mentioned using the SENet sorter network in model
The feature for taking original image increases the receptive field of characteristic pattern using the void space pyramid pond layer in model.
Preferably, above-mentioned display panel open defect detection method, SENet sorter network extract the method packet of original image feature
It includes: the new characteristic signal with multiple channels being generated by convolution transform to original image feature, is learnt by paying attention to power module
The weight in each channel out, so that effective characteristic spectrum weight is big, invalid or small effect characteristic spectrum weight is small.
Preferably, above-mentioned display panel open defect detection method learns each channel weight out by paying attention to power module
During, first the spatiality of feature is reduce, only reserve channel dimension, then generate a weight for each feature channel,
For explicitly modeling the correlation between channel, different interchannel weights on Weight to original channel, will be finally constituted
Calibration.
Preferably, above-mentioned display panel open defect detection method, the sample among training set or test set, which is input to, to be changed
After carrying out encoder into type deeplab v3+ deep learning semantic segmentation model, the characteristic information that obtains in conjunction with encoder into
Row up-sampling operation, exports forecast image.
Preferably, above-mentioned display panel open defect detection method, forecast image are semantic segmentation as a result, forecast image
Pixel value correspond to default defect type.
Preferably, above-mentioned display panel open defect detection method represents background with pixel value 0,1 represents air blister defect, 2
Represent fragmentation defect, 3 represent crack defect, 4 represent unfilled corner defect, 5 represent scratch defects.
Preferably, above-mentioned display panel open defect detection method, modified deeplab is characterized using mIOU accuracy
The accuracy of v3+ deep learning semantic segmentation model；MIOU accuracy is the two set of the actual value and predicted value of test set
Intersection and the ratio between union.
Preferably, above-mentioned display panel open defect detection method, subgraph are unified for the triple channel of 512*512 size
RGB image.
In general, through the invention it is contemplated above technical scheme is compared with the prior art, can obtain down and show
Beneficial effect:
Display panel open defect detection method provided by the invention will be classified using deep learning semantic segmentation model
Network is detected as the backbone network in deep learning semantic segmentation for display panel open defect；On the one hand, adaptability
By force, relatively or the even special circumstances of uneven illumination especially for background and defect, it needs to increase when with conventional process
Add additional expense to cope with these special circumstances, and only needs to be instructed with these special samples using method provided by the invention
It is also relatively accurate when practicing model, allow the feature of model learning special sample, then giving a forecast, it is increased without the additional processing time.
On the other hand, relative to the classification of deep learning and detection method, target defect can directly be exported using method of the invention
Shape, area；Another aspect, relatively existing semantic segmentation method are passed through using sorter network SENet as backbone network
Network goes learning characteristic weight, so that effective characteristic spectrum weight is big, invalid or small effect characteristic spectrum weight is small, so that
Accuracy in detection has obtained very big promotion.
Detailed description of the invention
Fig. 1 is the deep learning semantic segmentation model schematic used in the embodiment of the present invention；
Fig. 2 is the SENet sorter network schematic diagram in the deep learning semantic segmentation model used in the embodiment of the present invention.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.As long as in addition, technical characteristic involved in the various embodiments of the present invention described below
Not constituting a conflict with each other can be combined with each other.
Referring to Fig.1, the display panel open defect detection method that embodiment provides, includes the following steps:
(1) display panel image is acquired；
(2) it pre-processes: display panel image is carried out as detection image after enhancing processing, and by each panel detection figure
As being cut to the subgraph that multiple sizes are K*K pixel；The pixel K*K of its neutron image is according to the processing frequency of video card and aobvious
Deposit determination；
(3) it marks: annotation tool being divided using labelme image, semantic, with the defects of polygon frame sub-image area
Domain marks；
(4) training set and test set are established: several subgraphs are labeled, after subgraph and mark without mark
Subgraph composition original image and mark image pair；The part of selection image pair is remaining to be used as test set as training set；?
In one embodiment, the 90% of image pair is chosen as training set, remaining 10% is used as test set；
(5) deep learning semantic segmentation model is trained with training set data, and the model is verified using test set, until
The average friendship of test set is simultaneously no longer promoted than mIOU (Mean Intersection over Union) accuracy；
Wherein, mIOU accuracy is the standard characterization of semantic segmentation, is actual value and predicted value the two intersection of sets collection
The ratio between with union；
(6) display panel image to be measured is predicted using trained deep learning semantic segmentation model, it will be to be measured
Display panel image cropping is input to the model at subgraph, exports shape, the area of target defect.
In a preferred embodiment, deep learning semantic segmentation model is using one kind in deeplab v3+ depth
Improved model has been made on idiom justice parted pattern；Deeplab v3+ deep learning semantic segmentation model is Google at 2018 3
The model that the moon increases income, in the Open Source Code of the deeplab v3+ deep learning semantic segmentation model of Google, DCNN is used
Mobilenet or Xception；The model is improved in the present invention, using SENet (Squeeze-and-
Excitation Networks) attention mechanism module is as DCCN.
Referring to Fig.1, modified deeplab v3+ deep learning semantic segmentation model, training set employed in embodiment
Or the sample among test set is input in the model by coder processes；In the model, encoder includes depth convolution
Neural network (DCCN) and void space pyramid pond (ASPP)；Depth convolutional neural networks are used to extract the spy of original image
Sign, feature here mainly includes the color, shape, textural characteristics of defect；ASPP is used to increase the receptive field of characteristic pattern, impression
Open country is the size that convolution kernel is seen on the image.
In embodiment, using SENet attention mechanism module as DCCN, pass through transformation to picture feature is originally inputted,
Such as convolution transform, generate new characteristic signal U；Characteristic signal U has C channel, is learnt by paying attention to power module each out
The weight in channel, to generate the attention of channel region；Specifically, first the spatiality of feature is reduce, only reserve channel is tieed up
Degree, then a weight is generated for each feature channel, for explicitly modeling the correlation between channel, finally by Weight
Onto original channel, different interchannel recalibrations are constituted；By e-learning feature weight, so that effective characteristic spectrum power
Great, invalid or small effect characteristic spectrum weight is small, so that training pattern reaches better result.
The spreading rate of empty convolution is respectively 6,12,18 in embodiment, with the semantic letter in larger range of perception original image
Breath, accurately to divide.
It is handled by modified deeplab v3+ deep learning semantic segmentation solution to model code device (decoder) to export
Forecast image；Decoder treatment process is up-sampled mainly in conjunction with the shallower characteristic information in the part compiler encoder
Operation.
Forecast image is semantic segmentation as a result, under an application scenarios of embodiment, exports image and input picture
Size is 1:1；The pixel of output image has multiclass, using pixel value 0 represent background, 1 represent air blister defect, 2 represent it is broken
Piece defect, 3 represent crack defect, 4 represent unfilled corner defect, 5 represent scratch defects.
In a preferred embodiment, the input sample system of modified deeplab v3+ deep learning semantic segmentation model
One be 512*512 size, triple channel RGB image, it is too big also for sample-size is overcome to facilitate model treatment, by GPU video memory
Limitation, the less problem of the sample data handled simultaneously when training, can handle multiple samples when training simultaneously.
In embodiment, the mIOU that the deeplab v3+ for using SENet as backbone network verifies segmentation on verifying collection is accurate
Degree is 99.5%.And the mIOU that the deeplab v3+ for using xception as backbone network verifies segmentation on verifying collection is accurate
Degree is 97%, and the mIOU accuracy that the deeplab v3+ for using mobilenet as backbone network verifies segmentation on verifying collection is
95%.
This method that embodiment provides, improves deeplab v3+, and DCCN is made using SENet sorter network
For the attention mechanism module of backbone network, learn the weight in each channel out by paying attention to power module, to generate channel
The attention in domain；Referring to Fig. 2, core concept is to go learning characteristic weight according to loss by network, makes effective feature
Value weight is big, invalid or small effect characteristic value weight is small, and training pattern reaches better result in this way.
Using this modified deeplab v3+ deep learning semantic segmentation model, using sorter network SENET as depth
Learn the backbone network in semantic segmentation, is detected for display panel open defect, it can be input by image, semantic segmentation
Each pixel on image is classified, and can be background from which pixel is pixel scale distinguish on output image, which picture
Element belongs to various types of defects；Adaptable feature strong, accuracy in detection is high.
As it will be easily appreciated by one skilled in the art that the foregoing is merely illustrative of the preferred embodiments of the present invention, not to
The limitation present invention, any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should all include
Within protection scope of the present invention.
Claims (10)
1. a kind of display panel open defect detection method, which comprises the steps of:
(1) it is used as detection image after display panel image being carried out enhancing processing, and each detection image is cut to multiple sons
Image；
(2) several subgraphs are labeled with polygon frame using image, semantic segmentation annotation tool, lacking in sub-image
Sunken region marks；
(3) image pair is formed by the subgraph after subgraph and mark without mark；The part of image pair is chosen as instruction
Practice collection, it is remaining to be used as test set；
(4) deep learning semantic segmentation model is trained with training set data, the model is verified using test set, until reaching
Preset accuracy；
(5) display panel image to be measured is predicted using trained deep learning semantic segmentation model, output target lacks
Sunken shape, area.
2. display panel open defect detection method as described in claim 1, which is characterized in that the deep learning is semantic
Parted pattern uses deeplab v3+ deep learning semantic segmentation model, the deeplab v3+ deep learning semantic segmentation mould
Type is using SENet sorter network as backbone network.
3. display panel open defect detection method as claimed in claim 2, which is characterized in that among training set or test set
Sample be input to deep learning semantic segmentation model, first pass through compiler processes, utilize the SENet sorter network to extract former
The feature of figure increases the receptive field of characteristic pattern using the void space pyramid pond layer in model.
4. display panel open defect detection method as claimed in claim 2 or claim 3, which is characterized in that SENet sorter network mentions
The method for taking original image feature includes: that the new characteristic signal with multiple channels is generated by convolution transform to original image feature, is led to
The weight for paying attention to power module to learn each channel out is crossed, so that the spy that effective characteristic spectrum weight is big, invalid or small effect
It is small to levy map weight.
5. display panel open defect detection method as claimed in claim 4, which is characterized in that learned by paying attention to power module
During practising out each channel weight, first the spatiality of feature is reduce, only reserve channel dimension, then is each feature channel
A weight is generated, for explicitly modeling the correlation between channel, finally by Weight to original channel, is constituted
Different interchannel recalibrations.
6. display panel open defect detection method as claimed in claim 2 or claim 3, which is characterized in that training set or test set
Among sample be input to after the deep learning semantic segmentation model is compiled, carried out in conjunction with the characteristic information that compiling obtains
Up-sampling operation, exports forecast image.
7. display panel open defect detection method as claimed in claim 2 or claim 3, which is characterized in that forecast image is semanteme
Segmentation as a result, the pixel value of forecast image corresponds to default defect type.
8. display panel open defect detection method as claimed in claim 7, which is characterized in that with pixel value 0 represent background,
1 represent air blister defect, 2 represent fragmentation defect, 3 represent crack defect, 4 represent unfilled corner defect, 5 represent scratch defects.
9. display panel open defect detection method as claimed in claim 1 or 2, which is characterized in that use mIOU accuracy
To characterize the accuracy of the deep learning semantic segmentation model；For test set actual value and predicted value the two intersection of sets collection
The ratio between with union.
10. display panel open defect detection method as described in claim 1, which is characterized in that subgraph is unified for 512*
The triple channel RGB image of 512 sizes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910341237.9A CN110097544A (en) | 2019-04-25 | 2019-04-25 | A kind of display panel open defect detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910341237.9A CN110097544A (en) | 2019-04-25 | 2019-04-25 | A kind of display panel open defect detection method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110097544A true CN110097544A (en) | 2019-08-06 |
Family
ID=67445995
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910341237.9A Pending CN110097544A (en) | 2019-04-25 | 2019-04-25 | A kind of display panel open defect detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110097544A (en) |
Cited By (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110648310A (en) * | 2019-08-30 | 2020-01-03 | 上海理工大学 | Weak supervision casting defect identification method based on attention mechanism |
| CN110689056A (en) * | 2019-09-10 | 2020-01-14 | Oppo广东移动通信有限公司 | Classification method and device, equipment and storage medium |
| CN110910343A (en) * | 2019-09-29 | 2020-03-24 | 北京建筑大学 | Method and device for detecting pavement cracks and computer equipment |
| CN110991617A (en) * | 2019-12-02 | 2020-04-10 | 华东师范大学 | Construction method of kaleidoscope convolution network |
| CN111080641A (en) * | 2019-12-30 | 2020-04-28 | 上海商汤智能科技有限公司 | Crack detection method and device, computer equipment and storage medium |
| CN111080633A (en) * | 2019-12-20 | 2020-04-28 | Oppo（重庆）智能科技有限公司 | Screen defect detection method and device, terminal equipment and storage medium |
| CN111145145A (en) * | 2019-12-10 | 2020-05-12 | 太原科技大学 | Image surface defect detection method based on MobileNet |
| CN111179214A (en) * | 2019-11-29 | 2020-05-19 | 苏州优纳医疗器械有限公司 | Pathological section tissue area identification system based on image semantic segmentation |
| CN111462124A (en) * | 2020-03-31 | 2020-07-28 | 武汉卓目科技有限公司 | Remote sensing satellite cloud detection method based on Deep L abV3+ |
| CN111815628A (en) * | 2020-08-24 | 2020-10-23 | 武汉精测电子集团股份有限公司 | Display panel defect detection method, device, equipment and readable storage medium |
| CN111862092A (en) * | 2020-08-05 | 2020-10-30 | 复旦大学 | Express delivery outer package defect detection method and device based on deep learning |
| CN111968093A (en) * | 2020-08-19 | 2020-11-20 | 创新奇智(上海)科技有限公司 | Magnetic shoe surface defect detection method and device, electronic equipment and storage medium |
| CN112465835A (en) * | 2020-11-26 | 2021-03-09 | 深圳市对庄科技有限公司 | Method for jadeite image segmentation and model training method |
| CN112489019A (en) * | 2020-12-01 | 2021-03-12 | 合肥工业大学 | Method for rapidly identifying chopped fibers in GFRC image based on deep learning |
| CN112580649A (en) * | 2020-12-15 | 2021-03-30 | 重庆邮电大学 | Semantic segmentation method based on regional context relation module |
| CN112712503A (en) * | 2020-12-30 | 2021-04-27 | 厦门福信光电集成有限公司 | Display panel appearance detection method based on deep learning |
| CN112734690A (en) * | 2020-12-17 | 2021-04-30 | 郑州金惠计算机系统工程有限公司 | Surface defect detection method and device and computer readable storage medium |
| WO2021082919A1 (en) * | 2019-10-28 | 2021-05-06 | 上海悦易网络信息技术有限公司 | Defect detecting method and equipment for screen region of electronic equipment |
| CN112950584A (en) * | 2021-03-01 | 2021-06-11 | 哈尔滨工程大学 | Coating surface defect identification method based on deep learning |
| CN113160204A (en) * | 2021-04-30 | 2021-07-23 | 聚时科技(上海)有限公司 | Semantic segmentation network training method for generating defect area based on target detection information |
| WO2021147386A1 (en) * | 2020-01-21 | 2021-07-29 | 上海万物新生环保科技集团有限公司 | Screen scratch and crack detection method and device |
| WO2021238030A1 (en) * | 2020-05-26 | 2021-12-02 | 浙江大学 | Water level monitoring method for performing scale recognition on the basis of partitioning by clustering |
| CN114092874A (en) * | 2021-10-29 | 2022-02-25 | 北京百度网讯科技有限公司 | Training method of target detection model, target detection method and related equipment thereof |
| CN114663418A (en) * | 2022-04-06 | 2022-06-24 | 京东安联财产保险有限公司 | Image processing method and device, storage medium and electronic equipment |
| 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 |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109509172A (en) * | 2018-09-25 | 2019-03-22 | 无锡动视宫原科技有限公司 | A kind of liquid crystal display flaw detection method and system based on deep learning |
-
2019
- 2019-04-25 CN CN201910341237.9A patent/CN110097544A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109509172A (en) * | 2018-09-25 | 2019-03-22 | 无锡动视宫原科技有限公司 | A kind of liquid crystal display flaw detection method and system based on deep learning |
Non-Patent Citations (3)
| Title |
|---|
| 焦李成 等: "《雷达图像解译技术》", 31 December 2017, 国防工业出版社 * |
| 黄刚 等: "基于深度学习的道路标线自动提取与分类方法研究", 《中国激光》 * |
| 黄跃珍 等: "基于改进型MobileNet 网络的车型识别方法", 《电子技术与软件工程》 * |
Cited By (32)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110648310B (en) * | 2019-08-30 | 2023-08-04 | 上海理工大学 | Weak supervision casting defect identification method based on attention mechanism |
| CN110648310A (en) * | 2019-08-30 | 2020-01-03 | 上海理工大学 | Weak supervision casting defect identification method based on attention mechanism |
| CN110689056A (en) * | 2019-09-10 | 2020-01-14 | Oppo广东移动通信有限公司 | Classification method and device, equipment and storage medium |
| CN110910343A (en) * | 2019-09-29 | 2020-03-24 | 北京建筑大学 | Method and device for detecting pavement cracks and computer equipment |
| WO2021082919A1 (en) * | 2019-10-28 | 2021-05-06 | 上海悦易网络信息技术有限公司 | Defect detecting method and equipment for screen region of electronic equipment |
| CN111179214A (en) * | 2019-11-29 | 2020-05-19 | 苏州优纳医疗器械有限公司 | Pathological section tissue area identification system based on image semantic segmentation |
| CN110991617A (en) * | 2019-12-02 | 2020-04-10 | 华东师范大学 | Construction method of kaleidoscope convolution network |
| CN110991617B (en) * | 2019-12-02 | 2020-12-01 | 华东师范大学 | Construction method of kaleidoscope convolution network |
| CN111145145A (en) * | 2019-12-10 | 2020-05-12 | 太原科技大学 | Image surface defect detection method based on MobileNet |
| CN111080633A (en) * | 2019-12-20 | 2020-04-28 | Oppo（重庆）智能科技有限公司 | Screen defect detection method and device, terminal equipment and storage medium |
| CN111080641A (en) * | 2019-12-30 | 2020-04-28 | 上海商汤智能科技有限公司 | Crack detection method and device, computer equipment and storage medium |
| WO2021147386A1 (en) * | 2020-01-21 | 2021-07-29 | 上海万物新生环保科技集团有限公司 | Screen scratch and crack detection method and device |
| CN111462124A (en) * | 2020-03-31 | 2020-07-28 | 武汉卓目科技有限公司 | Remote sensing satellite cloud detection method based on Deep L abV3+ |
| WO2021238030A1 (en) * | 2020-05-26 | 2021-12-02 | 浙江大学 | Water level monitoring method for performing scale recognition on the basis of partitioning by clustering |
| CN111862092A (en) * | 2020-08-05 | 2020-10-30 | 复旦大学 | Express delivery outer package defect detection method and device based on deep learning |
| CN111968093A (en) * | 2020-08-19 | 2020-11-20 | 创新奇智(上海)科技有限公司 | Magnetic shoe surface defect detection method and device, electronic equipment and storage medium |
| CN111815628B (en) * | 2020-08-24 | 2021-02-19 | 武汉精测电子集团股份有限公司 | Display panel defect detection method, device, equipment and readable storage medium |
| CN111815628A (en) * | 2020-08-24 | 2020-10-23 | 武汉精测电子集团股份有限公司 | Display panel defect detection method, device, equipment and readable storage medium |
| CN112465835A (en) * | 2020-11-26 | 2021-03-09 | 深圳市对庄科技有限公司 | Method for jadeite image segmentation and model training method |
| CN112465835B (en) * | 2020-11-26 | 2022-07-08 | 深圳市对庄科技有限公司 | Method for jadeite image segmentation and model training method |
| CN112489019A (en) * | 2020-12-01 | 2021-03-12 | 合肥工业大学 | Method for rapidly identifying chopped fibers in GFRC image based on deep learning |
| CN112580649A (en) * | 2020-12-15 | 2021-03-30 | 重庆邮电大学 | Semantic segmentation method based on regional context relation module |
| CN112734690A (en) * | 2020-12-17 | 2021-04-30 | 郑州金惠计算机系统工程有限公司 | Surface defect detection method and device and computer readable storage medium |
| CN112712503B (en) * | 2020-12-30 | 2023-05-02 | 厦门福信光电集成有限公司 | Display panel appearance detection method based on deep learning |
| CN112712503A (en) * | 2020-12-30 | 2021-04-27 | 厦门福信光电集成有限公司 | Display panel appearance detection method based on deep learning |
| CN112950584A (en) * | 2021-03-01 | 2021-06-11 | 哈尔滨工程大学 | Coating surface defect identification method based on deep learning |
| CN112950584B (en) * | 2021-03-01 | 2022-07-29 | 哈尔滨工程大学 | Coating surface defect identification method based on deep learning |
| CN113160204A (en) * | 2021-04-30 | 2021-07-23 | 聚时科技(上海)有限公司 | Semantic segmentation network training method for generating defect area based on target detection information |
| CN114092874A (en) * | 2021-10-29 | 2022-02-25 | 北京百度网讯科技有限公司 | Training method of target detection model, target detection method and related equipment thereof |
| CN114092874B (en) * | 2021-10-29 | 2023-07-25 | 北京百度网讯科技有限公司 | Training method of target detection model, target detection method and related equipment thereof |
| CN114663418A (en) * | 2022-04-06 | 2022-06-24 | 京东安联财产保险有限公司 | Image processing method and device, storage medium and electronic equipment |
| 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 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110097544A (en) | A kind of display panel open defect detection method | |
| US10445557B2 (en) | Learning pixel visual context from object characteristics to generate rich semantic images | |
| JP7458519B2 (en) | Image analysis method, image analysis device, program, method for producing trained deep learning algorithm, and trained deep learning algorithm | |
| US11074692B2 (en) | Image analysis method, apparatus, program, and learned deep learning algorithm | |
| CN110349126B (en) | Convolutional neural network-based marked steel plate surface defect detection method | |
| US11436718B2 (en) | Image analysis method, image analysis apparatus, program, learned deep layer learning algorithm manufacturing method and learned deep layer learning algorithm | |
| US10474874B2 (en) | Applying pixelwise descriptors to a target image that are generated by segmenting objects in other images | |
| US10438096B2 (en) | Identifying and excluding blurred areas of images of stained tissue to improve cancer scoring | |
| CN105046277B (en) | Robust mechanism study method of the feature significance in image quality evaluation | |
| CN103518224B (en) | Method for analysing microbial growth | |
| CN109509172A (en) | A kind of liquid crystal display flaw detection method and system based on deep learning | |
| CN103076288B (en) | A kind of flesh of fish self-grading device based on computer vision and method | |
| CN111257341B (en) | Underwater building crack detection method based on multi-scale features and stacked full convolution network | |
| CN108564589A (en) | A kind of plant leaf blade dividing method based on the full convolutional neural networks of improvement | |
| CN110910343A (en) | Method and device for detecting pavement cracks and computer equipment | |
| CN110443262A (en) | Fish body freshness fast non-destructive detection method and its device based on computer vision | |
| CN110910372B (en) | Deep convolutional neural network-based uniform light plate defect detection method | |
| CN105043998B (en) | One kind differentiates the haploid method of corn | |
| CN103439348A (en) | Remote controller key defect detection method based on difference image method | |
| CN116309191B (en) | Intelligent gas inspection display method based on image enhancement | |
| CN106845545A (en) | Image-recognizing method and device in digital electric measuring instrument automatic calibrator | |
| CN109685030A (en) | A kind of mug rim of a cup defects detection classification method based on convolutional neural networks | |
| CN110110773A (en) | A kind of confidence calculations method of image, semantic segmentation object | |
| CN103854009B (en) | A kind of information processing method and electronic equipment | |
| CN115423802A (en) | Automatic classification and segmentation method for squamous epithelial tumor cell picture based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190806 |
|
| RJ01 | Rejection of invention patent application after publication |