CN109671058A - A kind of defect inspection method and system of big image in different resolution - Google Patents
A kind of defect inspection method and system of big image in different resolution Download PDFInfo
- Publication number
- CN109671058A CN109671058A CN201811482944.1A CN201811482944A CN109671058A CN 109671058 A CN109671058 A CN 109671058A CN 201811482944 A CN201811482944 A CN 201811482944A CN 109671058 A CN109671058 A CN 109671058A
- Authority
- CN
- China
- Prior art keywords
- target detection
- image
- defect
- different resolution
- detection frame
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30121—CRT, LCD or plasma display
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30168—Image quality inspection
Abstract
The invention discloses the defect inspection methods and system of a kind of big image in different resolution, method includes the following steps: S1: carrying out horizontal and vertical sliding on defect image using a certain size sliding window, obtain several sectional images;S2: defects detection is carried out to each sectional image by the target detection model obtained based on the training of small image in different resolution, the results set of a description target detection frame is obtained, the testing result of each target detection frame includes the probabilistic classification value for the defect type for being included in the target detection frame;S3: the successively Euclidean distance between two target detection frames adjacent in calculated result set merges two target detection frames when Euclidean distance is less than window fusion threshold value, takes the maximum target detection frame of probabilistic classification value as defects detection result;The present invention realizes the defects detection of big image in different resolution by way of sliding cutting and window fusion, not will lead to tiny flaw loss, avoided inspection or missing inspection.
Description
Technical field
The invention belongs to automate defect detecting technique field, more particularly, to a kind of big point based on deep learning
Resolution image defect detection method and system.
Background technique
During LCD panel production, often because of the scuffing on backlight (Back Light Unit, BLU), ash
The defects of dirt, spot, leads to the defect for occurring similar in final LCD panel, this directly affect the final quality of LCD panel and
Therefore the output of Yield Grade is as a result, detect the defect for being possible to occur in BLU backlight at the initial stage of LCD processing procedure
It is vital.
Currently, defects detection in BLU backlight be mainly realized by following three kinds of methods, first is that by Quality Inspector into
Capable artificial detection one by one;Second is that being detected by traditional image processing algorithm;Third is that the deep learning algorithm based on mainstream
It has been detected that, deep learning is since 2012 just come onto stage in the match of ImageNet image recognition, and deep learning algorithm is not
New breakthrough is constantly only obtained in field of image recognition, is even more to yield unusually brilliant results in object detection field;But three of the above method
It is respectively present following defect:
1, there are very strong subjective factors for human eye detection, and human eye carries out long-term detection can also visual fatigue;
In addition, the defect in BLU backlight is often fainter, factors above is integrated, and can be directly contributed a large amount of in BLU backlight
The missing inspection of defect and excessively inspection;And detection efficiency is low, brings a large amount of human cost and time cost;
Although 2, the automatic detection of BLU backlight defect may be implemented in traditional image processing algorithm to a certain extent,
It is that traditional image processing algorithm generalization ability is not strong, the parameter for needing to adjust is excessive, and detection process needs human intervention, in needle
When to diversified BLU backlight defects detection, this method not only can prevent entire defect inspection process from being fully automated, and
And because uncontrollable factor is numerous, it will also result in and largely cross inspection and missing inspection;
3, the defects detection of the mainstream algorithm of target detection in deep learning field works well, still, these mainstreams
The test object of algorithm of target detection is all the image of small resolution ratio, can not be directly used in as similar BLU defect image
Big image in different resolution;Also, since the defects of BLU defect image is very faint, wherein 50% defect area is essentially 10
A 10 pixels of pixel * or so, therefore also can not directly calculate the target detection that big resolution ratio defect image directly passes through mainstream
The image of down-sampled to the small resolution ratio of pyramid down-sampling algorithm in method, because directly down-sampled will make genetic defects figure
A large amount of defects as in are lost, and will also be will cause a large amount of LCD defect and be crossed inspection and missing inspection.
Summary of the invention
For at least one defect or Improvement requirement of the prior art, the present invention provides a kind of lacking for big image in different resolution
Detection method and system are fallen into, its object is to solve directly detect big image in different resolution existing for existing detection method, be easy
There is the problem of leak detection inspection.
To achieve the above object, according to one aspect of the present invention, a kind of defects detection of big image in different resolution is provided
Method, comprising the following steps:
S1: carrying out horizontal and vertical sliding using a certain size sliding window on defect image to be measured, lacks to be measured
Sunken image is cut into several sectional images, and the region of the sliding window covering is one point when laterally or longitudinally sliding each time
Area's image;It does not overlap each other, and laterally or longitudinally slides several times generated between adjacent laterally or longitudinally sliding twice
The set of sectional image need to cover whole defect images to be measured;
S2: defect is carried out to each sectional image by the target detection model obtained based on the training of small image in different resolution
Detection obtains the results set of a description target detection frame, and the testing result of each target detection frame includes the mesh
The probabilistic classification value for the defect type for being included in mark detection block;
S3: the Euclidean distance between two target detection frames adjacent in the results set is successively calculated, when the Europe
Formula distance merges two target detection frames when being less than preset window fusion threshold value, takes the maximum target of probabilistic classification value
Detection block is as defects detection result.
Preferably, drawbacks described above detection method, the testing result of target detection frame further include the point in the detection block upper left corner
Coordinate, width and height.
Preferably, drawbacks described above detection method, in step S3 further include:
According to the point coordinate in the target detection frame upper left corner, width, highly calculate the seat of each target detection frame central point
Mark, and each target detection frame in the results set according to the coordinate pair of the central point is ranked up.
Preferably, drawbacks described above detection method, before step S1 further include:
The defect sample image of big resolution ratio is cut into several small image in different resolution, in the small image in different resolution
Defect type is marked, and generates training set;And according to the training set to the target detection model based on deep learning
It is trained.
Preferably, drawbacks described above detection method, the size of sliding window be 312 pixels × 312 pixels, 416 pixels ×
Any one of 416 pixels, 512 pixels × 512 pixels, 608 pixels × 608 pixels.
Other side according to the invention additionally provides a kind of defect detecting system of big image in different resolution, including place
Device, memory are managed, and stores the computer program that can be run in the memory and in the processor, the calculating
Machine program realizes step described in any of the above embodiments when being executed by processor.
Preferably, drawbacks described above detection system, processor include sliding block, detection module and Fusion Module;
The sliding block is used to carry out on defect image to be measured using a certain size sliding window horizontal and vertical
Sliding, is cut into multiple sectional images for defect image to be measured, and laterally or longitudinally slide corresponding sectional image several times
Set need to cover entire defect image;
The detection module is used for the target detection model by obtaining based on the training of small image in different resolution to each point
Area's image carries out defects detection, obtains the results set of a description target detection frame, the inspection of each target detection frame
Survey the probabilistic classification value that result includes the defect type for being included in the target detection frame;
The Fusion Module is European between two target detection frames adjacent in the results set for successively calculating
Distance merges two target detection frames when the Euclidean distance is less than preset window fusion threshold value, takes probability point
Class is worth maximum target detection frame as defects detection result.
Preferably, drawbacks described above detection system, the testing result of target detection frame further include the point in the detection block upper left corner
Coordinate, width and height.
Preferably, drawbacks described above detection system further includes sorting module;
The sorting module is used for the point coordinate according to the target detection frame upper left corner, width, highly calculates each target
The coordinate of detection block central point, and each target detection frame in the results set according to the coordinate pair of the central point is arranged
Sequence.
Preferably, drawbacks described above detection system further includes segmentation module and training module;
The segmentation module is used to the defect sample image of big resolution ratio being cut into several small image in different resolution, to described
The defects of small image in different resolution type is marked, and generates training set;
The training module is used to be trained the target detection model based on deep learning according to the training set.
In general, through the invention it is contemplated above technical scheme is compared with the prior art, can obtain down and show
Beneficial effect:
(1) defect inspection method and system of a kind of big image in different resolution provided by the invention uses sliding window first
By the defect image cutting of big resolution ratio be small image in different resolution, then by small resolution image be sent into target detection network model into
Row defects detection is realized the defects detection of big image in different resolution by way of this sliding cutting, not will lead to genetic defects
Tiny flaw in image is lost, and avoids defect and crosses inspection or missing inspection;
(2) defect inspection method and system of a kind of big image in different resolution provided by the invention, passes through target detection window
The method of fusion carries out secondary treatment to multiple target detection frames of the same defect, and takes the maximum target inspection of probabilistic classification value
Frame is surveyed as defects detection as a result, the case where can effectively avoiding the occurrence of leak detection error detection inspection;
(3) defect inspection method and system of a kind of big image in different resolution provided by the invention, not only can be to defect map
As the defects of detected and positioned, while can classify to the defect that detected, by traditional target detection and
Classify two procedure mergings among a frame, improves the performance of AOI detection system, simplify defects detection process, subtract
The human input of few artificial detection;
(4) defect inspection method and system of a kind of big image in different resolution provided by the invention, without to current AOI knot
Structure is modified, and not will increase any hardware cost, has the characteristics that easy to accomplish, at low cost, practicability is high.
Detailed description of the invention
Fig. 1 is the flow chart of the defect inspection method of big image in different resolution provided in an embodiment of the present invention;
Fig. 2 is the visualization interface schematic diagram of target detection frame provided in an embodiment of the present invention;
Fig. 3 is the logic diagram of SOPC chip provided in an 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.
A kind of defect inspection method of big image in different resolution provided by the present embodiment, by the target in deep learning field
Detection algorithm is introduced into LCD object detection field, is primarily adapted for use in the defects detection to the big image in different resolution of BLU, solves depth
Mainstream algorithm of target detection in learning areas is not suitable for big image in different resolution detection, there are problems that leak detection inspection situation.
Fig. 1 is the flow chart of the defect inspection method of big image in different resolution provided in this embodiment, as shown in Figure 1, the party
Method the following steps are included:
S1: a large amount of big resolution ratio BLU defect image is collected, which is cut into small resolution
Rate image, and the defect inside each small image in different resolution is marked according to different defect types, make training set;Make
The target detection model based on deep learning is trained with the training set;The present embodiment is using the convolution in deep learning
Neural network is as target detection model, it is only necessary to which the design for carrying out primary network model can be to all kinds of in training set
The defect characteristic of defect extracts, without designing different feature extracting methods according to different defect types;
S2: cunning from left to right, from top to bottom is carried out on BLU defect image to be measured using a certain size sliding window
It is dynamic, it is not overlapped preferably between adjacent laterally or longitudinally sliding twice, laterally or longitudinally sliding need to cover whole several times
BLU defect image to be measured;The sliding order of sliding window is not especially limited, can be longitudinal sliding using first sliding laterally again
Dynamic, first longitudinal sliding motion slides laterally again, laterally longitudinal alternately sliding etc. modes carry out;Every sliding is primary, just covers sliding window
The target detection network that sectional image under lid is sent into after training carries out the detection of defect;
The size of sliding window is not specifically limited, and is in general selected according to the resolution ratio of BLU defect image to be measured
It selects, the resolution ratio of BLU defect image to be measured is bigger, then can select biggish sliding window;The size of common sliding window
For 312 pixels × 312 pixels, 416 pixels × 416 pixels, 512 pixels × 512 pixels, 608 pixels × 608 pixels etc.;This reality
Apply example use first sliding window by the BLU defect image cutting of big resolution ratio be small image in different resolution, then by small resolutions figure
Defects detection is carried out as being sent into target detection network model, lacking for big image in different resolution can be realized by way of this sliding
Detection is fallen into, and not will lead to the loss of the tiny flaw in genetic defects image, defect is avoided and crosses inspection or missing inspection.
S3: being slided covered sectional image each time to sliding window by target detection model and carry out defects detection,
Obtain the results set (bbox of a description target detection frame1, bbox2, bbox3..., bboxn), the inspection that these bbox are represented
After survey result is visualized, for the target detection frame being shown on BLU defect image as shown in Figure 2;Each bbox tool
The testing result that body includes is bbox (x, y, w, h, prob), wherein variable x, y are the coordinate of target detection frame upper left angle point,
W, h are respectively the width and height of target detection frame, and prob is by the probabilistic classification for the defect type for including in target detection frame
Value;
S4: each target detection frame is calculated according to the point coordinate (x, y), width w, height h in the target detection frame upper left corner
The coordinate (center_x, center_y) of central point, and according to the coordinate of central point (center_x, center_y) to result set
Each target detection frame in conjunction is ranked up;Since the same defect may will detect that multiple target detection frames, shown in Fig. 2
Three target detection frames be the same defect, two minor sorts are carried out to each target detection frame using quick sorting algorithm, it is first
It is first ranked up according to x coordinate, is then ranked up according to y-coordinate;After sequence, the corresponding multiple mesh of same defect
The center point coordinate (center_x, center_y) for marking detection block will be adjacent;
S5: the successively Euclidean distance between two target detection frames adjacent in calculated result set, by the Euclidean distance
It merges threshold value iWinFuseT with preset window to be compared, if the Euclidean distance of two target detection frames melts less than window
Close threshold value iWinFuseT, show the two target detection frames it is corresponding be the same defect, then to the two target detection frames into
Row fusion;
Wherein, the value range of n is [1, N-1], and N is the number of target detection frame contained in results set;
S6: the maximum target detection frame of probabilistic classification value prob is examined as defect in the target detection frame that two merge takes
It surveys as a result, the defects detection result that output is finally merged.
The present embodiment carries out two to multiple target detection frames of the same defect by the method that target detection window merges
Secondary processing, and take the maximum target detection frame of probabilistic classification value as defects detection as a result, leak detection can be avoided the occurrence of effectively
The case where error detection is examined;And realize the classification to defect type simultaneously during defect location and detection, it will be traditional
Two procedure mergings of target detection and classification improve the performance of AOI detection system among a frame, simplify defect inspection
Flow gauge reduces the human input of artificial detection.
The present embodiment additionally provides a kind of defect detecting system of big image in different resolution, including SOPC chip, memory, should
The computer program that can be run on SOPC chip is stored in memory, which is performed the realization above method
The step of;Fig. 3 is the logic diagram of SOPC chip provided in an embodiment of the present invention, as shown in figure 3, on SOPC chip example have it is more
A functional module, including sliding block, detection module and Fusion Module;
Wherein, it is loaded with sliding window algorithm in sliding block, for the sliding window using a certain size in BLU to be measured
Horizontal and vertical sliding is carried out on defect image, obtains several sectional images;Between adjacent laterally or longitudinally sliding twice most
It does not overlap well, the sectional image laterally or longitudinally slided several times need to cover whole BLU defect images to be measured;Sliding window
The sliding order of mouth is not especially limited, and can be used and first be slid laterally longitudinal sliding motion again, first longitudinal sliding motion slides laterally again, is horizontal
It is carried out to longitudinal alternately sliding etc. modes;Every sliding is primary, and the image under sliding window covering is just sent into the target after training
Detect the detection that network carries out defect;The size of sliding window is not specifically limited, in general according to BLU defect image to be measured
Resolution ratio selected, the resolution ratio of BLU defect image to be measured is bigger, then can select biggish sliding window;Commonly
The size of sliding window be 312 pixels × 312 pixels, 416 pixels × 416 pixels, 512 pixels × 512 pixels, 608 pixels ×
608 pixels etc.;
Detection module is used to export sliding block by the target detection model obtained based on the training of small image in different resolution
Each sectional image carry out defects detection, obtain one description target detection frame results set, each target detection
The testing result of frame includes the interior defect for being included of point coordinate, width, height and the target detection frame in the detection block upper left corner
The probabilistic classification value of type;
Target detection window blending algorithm is loaded in Fusion Module, for two adjacent in successively calculated result set
Euclidean distance between target detection frame, when the Euclidean distance of two target detection frames be less than preset window fusion threshold value when pair
The two target detection frames are merged, and take the maximum target detection frame of probabilistic classification value as defects detection result.
One as the present embodiment is preferred, which further includes sorting module;Quick row is loaded in sorting module
Sequence algorithm, for according to the target detection frame upper left corner point coordinate, width, highly calculate each target detection frame central point
Coordinate, and be ranked up according to each target detection frame in the coordinate pair results set of central point.
One as the present embodiment is preferred, which further includes segmentation module and training module;
Segmentation module is used to the defect sample image of big resolution ratio being cut into several small image in different resolution, to each small point
The defects of resolution image type is marked, and generates training set;
Training module is used to be trained the target detection model based on deep learning according to above-mentioned training set;This reality
Applying example preferably uses the convolutional neural networks in deep learning as target detection model, it is only necessary to carry out primary network model
Design can extract the defect characteristic of all kinds of defects in training set, without according to different defect type designs
Different feature extracting methods.
In above-mentioned technical proposal, SOPC chip can be replaced central processor unit (Central Processing
Unit, CPU), general processor, digital signal processor (Digital Signal Processing, DSP), dedicated integrated electricity
Road (Application Specific Integrated Circuit, ASIC), field-programmable gate array (Field-
Programmable Gate Array, FPGA) or other programmable logic device, transistor logic etc..
Compared to existing defect inspection method, a kind of defect inspection method of big image in different resolution provided by the invention and
The defect image cutting of big resolution ratio is first small image in different resolution using sliding window, then by small resolution image by system
It is sent into target detection network model and carries out defects detection, the defect of big image in different resolution is realized by way of this sliding cutting
Detection, the tiny flaw that not will lead in genetic defects image are lost;By method that target detection window merges to same
Multiple target detection frames of defect carry out secondary treatment, and take the maximum target detection frame of probabilistic classification value as defects detection knot
Fruit can effectively avoid the occurrence of the case where leak detection error detection is examined.
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 defect inspection method of big image in different resolution, which comprises the following steps:
S1: carrying out horizontal and vertical sliding using sliding window on defect image to be measured, defect image to be measured is cut into more
A sectional image, and the set for laterally or longitudinally sliding covered sectional image several times need to cover entire defect image;
S2: defect is carried out each described sectional image by the target detection model obtained based on the training of small image in different resolution
Detection obtains the results set of a description target detection frame, and the testing result of each target detection frame includes the mesh
The probabilistic classification value for the defect type for being included in mark detection block;
S3: successively calculating the Euclidean distance between two target detection frames adjacent in the results set, when it is described it is European away from
Two target detection frames are merged from when being less than preset window fusion threshold value, maximum target detection is worth with probabilistic classification
Frame is as defects detection result.
2. defect inspection method as described in claim 1, which is characterized in that the testing result of the target detection frame further includes
Point coordinate, width and the height in the detection block upper left corner.
3. defect inspection method as claimed in claim 2, which is characterized in that in step S3 further include:
According to the point coordinate in the target detection frame upper left corner, width, highly the coordinate of each target detection frame central point is calculated, and
It is ranked up according to each target detection frame in results set described in the coordinate pair of the central point.
4. defect inspection method as claimed in claim 1 or 3, which is characterized in that before step S1 further include:
The sample defect image of big resolution ratio is cut into several small image in different resolution, to the defects of described small image in different resolution
Type is marked, and generates training set;And the target detection model based on deep learning is carried out according to the training set
Training.
5. defect inspection method as described in claim 1, which is characterized in that the size of the sliding window be 312 pixels ×
Any one of 312 pixels, 416 pixels × 416 pixels, 512 pixels × 512 pixels, 608 pixels × 608 pixels.
6. a kind of defect detecting system of big image in different resolution, which is characterized in that including processor, memory, and be stored in
It is real when the computer program is executed by processor in the memory and the computer program that can be run in the processor
The step of any one of existing claim 1-5 the method.
7. defect detecting system as claimed in claim 6, which is characterized in that the processor includes sliding block, detection mould
Block and Fusion Module;
The sliding block is used to carry out horizontal and vertical sliding on defect image to be measured using sliding window, by defect to be measured
Image is cut into multiple sectional images, and the set for laterally or longitudinally sliding corresponding sectional image several times need to cover entire lack
Fall into image;
The detection module is used for by training obtained target detection model described in each based on small image in different resolution
Sectional image carries out defects detection, obtains the results set of a description target detection frame, each target detection frame
Testing result includes the probabilistic classification value for the defect type for being included in the target detection frame;
The Fusion Module is used to successively calculate the Euclidean distance between two target detection frames adjacent in the results set,
Two target detection frames are merged when the Euclidean distance is less than preset window fusion threshold value, take probabilistic classification value most
Big target detection frame is as defects detection result.
8. defect detecting system as claimed in claim 7, which is characterized in that the testing result of the target detection frame further includes
Point coordinate, width and the height in the detection block upper left corner.
9. defect detecting system as claimed in claim 8, which is characterized in that further include sorting module;
The sorting module is used for the point coordinate according to the target detection frame upper left corner, width, highly calculates each target detection
The coordinate of frame central point, and each target detection frame in the results set according to the coordinate pair of the central point is ranked up.
10. the defect detecting system as described in claim 7 or 9, which is characterized in that further include segmentation module and training module;
The segmentation module is used to the defect sample image of big resolution ratio being cut into several small image in different resolution, to described small point
The defects of resolution image type is marked, and generates training set;
The training module is used to be trained the target detection model based on deep learning according to the training set.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811482944.1A CN109671058B (en) | 2018-12-05 | 2018-12-05 | Defect detection method and system for large-resolution image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811482944.1A CN109671058B (en) | 2018-12-05 | 2018-12-05 | Defect detection method and system for large-resolution image |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109671058A true CN109671058A (en) | 2019-04-23 |
CN109671058B CN109671058B (en) | 2021-04-20 |
Family
ID=66144265
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811482944.1A Active CN109671058B (en) | 2018-12-05 | 2018-12-05 | Defect detection method and system for large-resolution image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109671058B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110298819A (en) * | 2019-05-23 | 2019-10-01 | 网宿科技股份有限公司 | A kind of quality detecting method and edge calculations equipment of target object |
CN111178445A (en) * | 2019-12-31 | 2020-05-19 | 上海商汤智能科技有限公司 | Image processing method and device |
CN112084886A (en) * | 2020-08-18 | 2020-12-15 | 眸芯科技(上海)有限公司 | Method and device for improving detection performance of neural network target detection |
CN112884803A (en) * | 2020-08-18 | 2021-06-01 | 眸芯科技(上海)有限公司 | Real-time intelligent monitoring target detection method and device based on DSP |
CN113008895A (en) * | 2021-01-29 | 2021-06-22 | 广州信邦智能装备股份有限公司 | Block fitting defect detection method based on three-dimensional data |
CN113269055A (en) * | 2021-05-06 | 2021-08-17 | 中国矿业大学 | Method for calculating loss function of fallen leaf detection prediction frame |
CN114718587A (en) * | 2021-01-02 | 2022-07-08 | 杨立鹏 | Bucket tooth of weak surrounding rock tunnel excavator |
WO2022147966A1 (en) * | 2021-01-08 | 2022-07-14 | 长鑫存储技术有限公司 | Wafer testing method and apparatus, and device and storage medium |
EP4050560A4 (en) * | 2021-01-08 | 2022-08-31 | Changxin Memory Technologies, Inc. | Wafer testing method and apparatus, and device and storage medium |
CN115984268A (en) * | 2023-03-20 | 2023-04-18 | 杭州百子尖科技股份有限公司 | Target detection method and device based on machine vision, electronic equipment and medium |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101996405A (en) * | 2010-08-30 | 2011-03-30 | 中国科学院计算技术研究所 | Method and device for rapidly detecting and classifying defects of glass image |
US20120027288A1 (en) * | 2010-07-29 | 2012-02-02 | Chang Yuan | Methods, Systems and Apparatus for Defect Detection |
WO2016047377A1 (en) * | 2014-09-22 | 2016-03-31 | 富士フイルム株式会社 | Image-recording device, and image defect detection device and method |
CN106251361A (en) * | 2016-08-30 | 2016-12-21 | 兰州交通大学 | A kind of rail surface defects image self-adapting division method |
CN106780455A (en) * | 2016-12-07 | 2017-05-31 | 五邑大学 | A kind of product surface detection method based on the local neighborhood window for sliding |
CN108038853A (en) * | 2017-12-18 | 2018-05-15 | 浙江工业大学 | A kind of ceramic tile surface defect identification method based on convolutional neural networks and Active Learning |
CN108846311A (en) * | 2018-04-28 | 2018-11-20 | 北京羽医甘蓝信息技术有限公司 | The method and device of the facial pieces of skin shape defect of detection based on deep learning |
-
2018
- 2018-12-05 CN CN201811482944.1A patent/CN109671058B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120027288A1 (en) * | 2010-07-29 | 2012-02-02 | Chang Yuan | Methods, Systems and Apparatus for Defect Detection |
CN101996405A (en) * | 2010-08-30 | 2011-03-30 | 中国科学院计算技术研究所 | Method and device for rapidly detecting and classifying defects of glass image |
WO2016047377A1 (en) * | 2014-09-22 | 2016-03-31 | 富士フイルム株式会社 | Image-recording device, and image defect detection device and method |
CN106251361A (en) * | 2016-08-30 | 2016-12-21 | 兰州交通大学 | A kind of rail surface defects image self-adapting division method |
CN106780455A (en) * | 2016-12-07 | 2017-05-31 | 五邑大学 | A kind of product surface detection method based on the local neighborhood window for sliding |
CN108038853A (en) * | 2017-12-18 | 2018-05-15 | 浙江工业大学 | A kind of ceramic tile surface defect identification method based on convolutional neural networks and Active Learning |
CN108846311A (en) * | 2018-04-28 | 2018-11-20 | 北京羽医甘蓝信息技术有限公司 | The method and device of the facial pieces of skin shape defect of detection based on deep learning |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110298819A (en) * | 2019-05-23 | 2019-10-01 | 网宿科技股份有限公司 | A kind of quality detecting method and edge calculations equipment of target object |
CN111178445A (en) * | 2019-12-31 | 2020-05-19 | 上海商汤智能科技有限公司 | Image processing method and device |
CN112084886B (en) * | 2020-08-18 | 2022-03-15 | 眸芯科技(上海)有限公司 | Method and device for improving detection performance of neural network target detection |
CN112884803A (en) * | 2020-08-18 | 2021-06-01 | 眸芯科技(上海)有限公司 | Real-time intelligent monitoring target detection method and device based on DSP |
CN112084886A (en) * | 2020-08-18 | 2020-12-15 | 眸芯科技(上海)有限公司 | Method and device for improving detection performance of neural network target detection |
CN112884803B (en) * | 2020-08-18 | 2023-11-24 | 眸芯科技(上海)有限公司 | Real-time intelligent monitoring target detection method and device based on DSP |
CN114718587A (en) * | 2021-01-02 | 2022-07-08 | 杨立鹏 | Bucket tooth of weak surrounding rock tunnel excavator |
WO2022147966A1 (en) * | 2021-01-08 | 2022-07-14 | 长鑫存储技术有限公司 | Wafer testing method and apparatus, and device and storage medium |
EP4050560A4 (en) * | 2021-01-08 | 2022-08-31 | Changxin Memory Technologies, Inc. | Wafer testing method and apparatus, and device and storage medium |
US11928808B2 (en) | 2021-01-08 | 2024-03-12 | Changxin Memory Technologies, Inc. | Wafer detection method, device, apparatus, and storage medium |
CN113008895A (en) * | 2021-01-29 | 2021-06-22 | 广州信邦智能装备股份有限公司 | Block fitting defect detection method based on three-dimensional data |
CN113269055A (en) * | 2021-05-06 | 2021-08-17 | 中国矿业大学 | Method for calculating loss function of fallen leaf detection prediction frame |
CN113269055B (en) * | 2021-05-06 | 2024-02-13 | 中国矿业大学 | Calculation method for loss function of fallen leaf detection prediction frame |
CN115984268A (en) * | 2023-03-20 | 2023-04-18 | 杭州百子尖科技股份有限公司 | Target detection method and device based on machine vision, electronic equipment and medium |
Also Published As
Publication number | Publication date |
---|---|
CN109671058B (en) | 2021-04-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109671058A (en) | A kind of defect inspection method and system of big image in different resolution | |
US11915407B2 (en) | Automated system and method for clarity measurements and clarity grading | |
CN111310558B (en) | Intelligent pavement disease extraction method based on deep learning and image processing method | |
CN109636772A (en) | The defect inspection method on the irregular shape intermetallic composite coating surface based on deep learning | |
CN109711288A (en) | Remote sensing ship detecting method based on feature pyramid and distance restraint FCN | |
CN112419250A (en) | Pavement crack digital image extraction, crack repair and crack parameter calculation method | |
CN109886282A (en) | Method for checking object, device, computer readable storage medium and computer equipment | |
CN106846344A (en) | A kind of image segmentation optimal identification method based on the complete degree in edge | |
CN109598224A (en) | Recommend white blood cell detection method in the Sections of Bone Marrow of convolutional neural networks based on region | |
CN109087274A (en) | Electronic device defect inspection method and device based on multidimensional fusion and semantic segmentation | |
CN110047073A (en) | A kind of X-ray weld image fault grading method and system | |
CN110853005A (en) | Immunohistochemical membrane staining section diagnosis method and device | |
CN102974551A (en) | Machine vision-based method for detecting and sorting polycrystalline silicon solar energy | |
CN110458301A (en) | A kind of damage identification method of vehicle part, device, computer equipment and storage medium | |
CN109886960A (en) | The method of glass edge defects detection based on machine vision | |
CN109697719A (en) | A kind of image quality measure method, apparatus and computer readable storage medium | |
CN108186051A (en) | A kind of image processing method and processing system of the automatic measurement fetus Double Tops electrical path length from ultrasonoscopy | |
CN113505865B (en) | Sheet surface defect image recognition processing method based on convolutional neural network | |
CN103852034B (en) | A kind of method for measuring perendicular | |
CN112215217B (en) | Digital image recognition method and device for simulating doctor to read film | |
CN113240626A (en) | Neural network-based method for detecting and classifying concave-convex flaws of glass cover plate | |
CN114926407A (en) | Steel surface defect detection system based on deep learning | |
CN108764328A (en) | The recognition methods of Terahertz image dangerous material, device, equipment and readable storage medium storing program for executing | |
CN113240623B (en) | Pavement disease detection method and device | |
CN106056608A (en) | Image dot-line defect detection method and device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |