CN109191440A - Glass blister detection and method of counting - Google Patents

Glass blister detection and method of counting Download PDF

Info

Publication number
CN109191440A
CN109191440A CN201810975382.8A CN201810975382A CN109191440A CN 109191440 A CN109191440 A CN 109191440A CN 201810975382 A CN201810975382 A CN 201810975382A CN 109191440 A CN109191440 A CN 109191440A
Authority
CN
China
Prior art keywords
glass
input picture
blister
density
glass blister
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
Application number
CN201810975382.8A
Other languages
Chinese (zh)
Inventor
黄良军
张亚妮
张晓宁
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Institute of Technology
Original Assignee
Shanghai Institute of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shanghai Institute of Technology filed Critical Shanghai Institute of Technology
Priority to CN201810975382.8A priority Critical patent/CN109191440A/en
Publication of CN109191440A publication Critical patent/CN109191440A/en
Pending legal-status Critical Current

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
    • G06T7/0008—Industrial image inspection checking presence/absence
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06M—COUNTING MECHANISMS; COUNTING OF OBJECTS NOT OTHERWISE PROVIDED FOR
    • G06M11/00—Counting of objects distributed at random, e.g. on a surface
    • G06M11/02—Counting of objects distributed at random, e.g. on a surface using an electron beam scanning a surface line by line, e.g. of blood cells on a substrate
    • 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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10004—Still image; Photographic image

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)

Abstract

The present invention provides a kind of detection of glass blister and method of counting, are trained by using convolutional neural networks, and using sample set to the convolutional neural networks, obtain network weight parameter θ；According to the network weight parameter θ, so that the convolutional neural networks obtain any input picture X in sample setiGlass blister density map F (Xi, θ) and input picture XiActual glass bubble density figure FiBetween Euclidean distance it is minimum；According to the input picture XiActual glass bubble density figure FiBetween the smallest glass blister density map F (X of Euclidean distancei, θ), obtain the input picture XiThe covering position of middle glass blister and the number of glass blister are able to carry out the glass blister position detection accurately carried out and count.

Description

Glass blister detection and method of counting
Technical field
The present invention relates to a kind of detection of glass blister and method of counting.
Background technique
Existing glass blister detection algorithm is bigger to image Segmentation Technology dependence, it is fixed etc. all to input picture size More limitations, many existing methods can not well detected bubble present in glass, this glass safe for automobile etc. The performance of glass influences very big.Therefore, accurate glass blister detection algorithm has glass defect abnormality detection highly important Meaning.
Summary of the invention
The purpose of the present invention is to provide a kind of detection of glass blister and method of counting.
The present invention provides a kind of detection of glass blister and method of counting, comprising:
The convolutional neural networks are trained using convolutional neural networks, and using sample set, obtain network weight Parameter θ；
According to the network weight parameter θ, so that the convolutional neural networks obtain any input picture X in sample seti Glass blister density map F (Xi, θ) and input picture XiActual glass bubble density figure FiBetween Euclidean distance it is minimum；
According to the input picture XiActual glass bubble density figure FiBetween the smallest glass blister of Euclidean distance it is close Degree figure F (Xi, θ), obtain the input picture XiThe covering position of middle glass blister and the number of glass blister.
Further, in the above-mentioned methods, according to the input picture XiActual glass bubble density figure FiBetween Europe Formula is apart from the smallest glass blister density map F (Xi, θ), obtain the input picture XiThe covering position of middle glass blister and glass The number of bubble, comprising:
With the input picture XiActual glass bubble density figure FiBetween the smallest glass blister density map of Euclidean distance F(Xi, θ) in, high bright part is the covering position of glass blister；
With the input picture XiActual glass bubble density figure FiBetween the smallest glass blister density map of Euclidean distance F(Xi, θ) in, all values summation is this input picture XiGlass blister number.
Further, in the above-mentioned methods, the convolutional neural networks include 5 convolutional layers, 3 full articulamentums and 2 Pond layer, 2 pond layers are respectively after the 1st convolutional layer and the 2nd convolutional layer.
Further, in the above-mentioned methods, according to the network weight parameter θ, so that the convolutional neural networks obtain Any input picture X in sample setiGlass blister density map F (Xi, θ) and input picture XiActual glass bubble density Scheme FiBetween Euclidean distance it is minimum, comprising:
The input picture XiInto after the convolutional neural networks, by convolution, Chi Hua, full attended operation obtain with The input picture XiActual glass bubble density figure FiBetween the smallest glass blister density map F (X of Euclidean distancei,θ)。
Further, in the above-mentioned methods, the loss function of the convolutional neural networks is input picture XiGlass gas Steep density map F (Xi；θ) with input picture XiActual glass bubble density figure FiBetween Euclidean distance L (θ):
Wherein, θ represents the convolutional neural networks parameter to be optimized, and N represents the number of training set picture, XiRepresent input figure Picture, FiRepresenting input images XiCorresponding glass density figure, F (Xi；θ) represent network-evaluated glass blister density map, the volume After product neural network sets an initial value, go out to input the loss of picture according to standard density graphic calculation: L (θ), then each time The parameter θ that whole network is updated in Optimized Iterative, until penalty values converge to the value for being less than preset threshold.
Further, in the above-mentioned methods, the parameter θ of whole network is updated in Optimized Iterative each time, comprising:
Using stochastic gradient descent method, the network weight ginseng of the convolutional neural networks is updated in Optimized Iterative each time Number θ.
Further, in the above-mentioned methods, the input picture XiCorresponding actual glass bubble density figure FiIt indicates are as follows:
In formula: N is input picture XiThe number of middle glass blister,Indicate input picture XiIn each pixel position, xi It is i-th of glass blister in input picture XiIn position, δ () be unit impulse function, * is convolution operation,For mark Quasi- difference is σiGaussian kernel.
Compared with prior art, by the present invention in that with convolutional neural networks, and using sample set to the convolutional Neural Network is trained, and obtains network weight parameter θ；According to the network weight parameter θ, so that the convolutional neural networks obtain Any input picture X in sample setiGlass blister density map F (Xi, θ) and input picture XiActual glass bubble density Scheme FiBetween Euclidean distance it is minimum；According to the input picture XiActual glass bubble density figure FiBetween Euclidean distance most Small glass blister density map F (Xi, θ), obtain the input picture XiThe covering position of middle glass blister and of glass blister Number is able to carry out the glass blister position detection accurately carried out and counts.
Detailed description of the invention
Fig. 1 is the convolutional network structure that uses of one embodiment of the invention, and in figure, Conv represents convolution operation, and FC represents complete Articulamentum, the number under Conv module represent the size of convolution kernel and the port number of convolutional layer, and the number under FC module represents mind Through first number；
Fig. 2 is the input picture X of one embodiment of the inventioni；
Fig. 3 is the actual glass bubble density figure F of one embodiment of the inventioni。
Specific embodiment
In order to make the foregoing objectives, features and advantages of the present invention clearer and more comprehensible, with reference to the accompanying drawing and specific real Applying mode, the present invention is described in further detail.
As shown in Figures 1 to 3, the present invention provides a kind of detection of glass blister and method of counting, comprising:
Step S1 instructs the convolutional neural networks using convolutional neural networks shown in Fig. 1, and using sample set Practice, obtains network weight parameter θ；
Here, Fig. 1 is the convolutional network structure that uses of the present invention, in figure, Conv represents convolution operation, and FC represents full connection Layer, the number under Conv module represents the size of convolution kernel and the port number of convolutional layer, and the number under FC module represents neuron Number；
Step S2, according to the network weight parameter θ, so that the convolutional neural networks obtain arbitrarily inputting in sample set Image XiGlass blister density map F (Xi, θ) and input picture XiActual glass bubble density figure FiBetween Euclidean distance It is minimum；
Step S3, according to the input picture XiActual glass bubble density figure FiBetween the smallest glass of Euclidean distance Glass bubble density figure F (Xi, θ), obtain the input picture XiThe covering position of middle glass blister and the number of glass blister.
Here, carrying out accurate glass gas the present invention provides single image captured by a kind of foundation industry camera Method bubble detection and counted.
Glass blister detection of the invention in one embodiment of method of counting, step S3, according to the input picture Xi Actual glass bubble density figure FiBetween the smallest glass blister density map F (X of Euclidean distancei, θ), obtain the input picture XiThe covering position of middle glass blister and the number of glass blister, comprising:
With the input picture XiActual glass bubble density figure FiBetween the smallest glass blister density map of Euclidean distance F(Xi, θ) in, high bright part is the covering position of glass blister；
With the input picture XiActual glass bubble density figure FiBetween the smallest glass blister density map of Euclidean distance F(Xi, θ) in, all values summation is this input picture XiGlass blister number.
Here, network inputs image XiAfterwards, it can operate, obtain and the input figure by multiple convolution and pondization twice As XiActual glass bubble density figure FiBetween the smallest glass blister density map F (X of Euclidean distancei, θ), wherein with it is described Input picture XiActual glass bubble density figure FiBetween the smallest glass blister density map F (X of Euclidean distancei, θ) in, it highlights Position, that is, bubble position, the summation of density map all values are glass blister sum.The present invention is successfully by convolutional neural networks application In glass defect detection task.
With one embodiment of method of counting, the convolutional neural networks include 5 convolution for glass blister detection of the invention Layer, 3 full articulamentums, 2 pond layers, 2 pond layers are respectively after the 1st convolutional layer and the 2nd convolutional layer.
In glass blister detection of the invention and one embodiment of method of counting, step S2, according to the network weight parameter θ, so that the convolutional neural networks obtain any input picture X in sample setiGlass blister density map F (Xi, θ) and defeated with this Enter image XiActual glass bubble density figure FiBetween Euclidean distance it is minimum, comprising:
The input picture XiInto after the convolutional neural networks, by convolution, Chi Hua, full attended operation obtain with The input picture XiActual glass bubble density figure FiBetween the smallest glass blister density map F (X of Euclidean distancei,θ)。
With one embodiment of method of counting, the loss function of the convolutional neural networks is for glass blister detection of the invention Input picture XiGlass blister density map F (Xi；θ) with input picture XiActual glass bubble density figure FiBetween Europe Formula distance L (θ):
Wherein, θ represents the convolutional neural networks parameter to be optimized, and N represents the number of training set picture, XiRepresent input figure Picture, FiRepresenting input images XiCorresponding glass density figure, F (Xi；θ) represent network-evaluated glass blister density map, the volume After product neural network sets an initial value, go out to input the loss of picture according to standard density graphic calculation: L (θ), then each time The parameter θ that whole network is updated in Optimized Iterative, until penalty values converge to the value for being less than preset threshold.
Glass blister detection of the invention updates entire net with one embodiment of method of counting in Optimized Iterative each time The parameter θ of network, comprising:
Using stochastic gradient descent method, the network weight ginseng of the convolutional neural networks is updated in Optimized Iterative each time Number θ.
In glass blister detection of the invention and one embodiment of method of counting, the input picture XiCorresponding actual glass Bubble density figure FiIt indicates are as follows:
In formula: N is input picture XiThe number of middle glass blister,Indicate input picture XiIn each pixel position, xi It is i-th of glass blister in input picture XiIn position, δ () be unit impulse function, * is convolution operation,For mark Quasi- difference is σiGaussian kernel.
In addition, mean absolute error can be used in the present inventionAnd Averaged Square Error of Multivariate Carry out the quality of evaluation algorithms performance.Wherein, N represents the sum of cycle tests picture, ziTrue bubble in representing input images Number,Estimation bubble number in representing input images.
Specifically, needing to solve in one embodiment of the invention to give a glass blister image, it is each then to detect the image Total bubble number is simultaneously estimated in the position of the glass blister in a region, includes the following steps:
Known input picture can be expressed as the matrix of m x n: x ∈ Rm×n, then glass gas corresponding to input picture x Bubble density map can indicate are as follows:In formula: N is the glass blister total number in image, Indicate the position of each pixel in image, xiFor the position of i-th of bubble in the picture, δ () is unit impulse function, and * is Convolution operation,It is σ for standard deviationiGaussian kernel.By the detection of the single image glass blister of convolutional neural networks with based on The target for figuring method is to learn mapping function F:x → F (x) ≈ M by input picture x to the glass blister in the image (x), in formula, F (x) is estimation glass density figure.In order to learn F, need to solve the problems, such as follows:
In formula, F (Xi；It is θ) estimation glass density figure, θ is parameter to be learned.In general, F is a complex nonlinear Mapping.
In the present invention, using convolutional neural networks as shown in Figure 1 come learning of nonlinear functions F.Its structure such as Fig. 1 institute Show.Characteristic pattern is sent into full articulamentum and revert to glass blister density map by input picture behind convolution pond.Network in Fig. 1 The port number and convolution kernel size of the digital representation of the lower section convolutional layer, such as the size of first convolutional layer convolution kernel is 7x7, Port number is 32.
The loss function of the convolutional neural networks is estimation density map F (Xi；θ) and actual density figure FiBetween it is European away from From L (θ),In this equation, θ represents the entire counting and network parameter to be optimized, N Represent the number of training set picture, XiRepresent input picture, FiRepresent the corresponding glass density figure of input picture.F(Xi；θ) represent Network-evaluated glass density figure.First to after one initial value of network settings, go out to input picture according to standard density graphic calculation Loss: L (θ) then updates the parameter θ of whole network using stochastic gradient descent method in Optimized Iterative each time, until damage Mistake value converges to a lesser value.
Fig. 2 is the glass blister image for inputting network, and Fig. 3 is network-evaluated glass blister density map.Wherein, scheme High bright part is the position overlay area that glass blister is likely to occur in 3, and the summation of Fig. 3 all values is that the input picture is corresponding Glass blister number.
Each embodiment in this specification is described in a progressive manner, the highlights of each of the examples are with other The difference of embodiment, the same or similar parts in each embodiment may refer to each other.
Professional further appreciates that, unit described in conjunction with the examples disclosed in the embodiments of the present disclosure And algorithm steps, can be realized with electronic hardware, computer software, or a combination of the two, in order to clearly demonstrate hardware and The interchangeability of software generally describes each exemplary composition and step according to function in the above description.These Function is implemented in hardware or software actually, the specific application and design constraint depending on technical solution.Profession Technical staff can use different methods to achieve the described function each specific application, but this realization is not answered Think beyond the scope of this invention.
Obviously, those skilled in the art can carry out various modification and variations without departing from spirit of the invention to invention And range.If in this way, these modifications and changes of the present invention belong to the claims in the present invention and its equivalent technologies range it Interior, then the invention is also intended to include including these modification and variations.

Claims (7)

1. a kind of glass blister detection and method of counting characterized by comprising
The convolutional neural networks are trained using convolutional neural networks, and using sample set, obtain network weight parameter θ；
According to the network weight parameter θ, so that the convolutional neural networks obtain any input picture X in sample setiGlass Bubble density figure F (Xi, θ) and input picture XiActual glass bubble density figure FiBetween Euclidean distance it is minimum；
According to the input picture XiActual glass bubble density figure FiBetween the smallest glass blister density map of Euclidean distance F(Xi, θ), obtain the input picture XiThe covering position of middle glass blister and the number of glass blister.
2. glass blister as described in claim 1 detection and method of counting, which is characterized in that according to the input picture Xi Actual glass bubble density figure FiBetween the smallest glass blister density map F (X of Euclidean distancei, θ), obtain the input picture XiThe covering position of middle glass blister and the number of glass blister, comprising:
With the input picture XiActual glass bubble density figure FiBetween the smallest glass blister density map F of Euclidean distance (Xi, θ) in, high bright part is the covering position of glass blister；
With the input picture XiActual glass bubble density figure FiBetween the smallest glass blister density map F of Euclidean distance (Xi, θ) in, all values summation is this input picture XiGlass blister number.
3. glass blister as described in claim 1 detection and method of counting, which is characterized in that the convolutional neural networks include 5 convolutional layers, 3 full articulamentums and 2 pond layers, 2 pond layers are respectively after the 1st convolutional layer and the 2nd convolutional layer.
4. glass blister detection as claimed in claim 3 and method of counting, which is characterized in that according to the network weight parameter θ, so that the convolutional neural networks obtain any input picture X in sample setiGlass blister density map F (Xi, θ) and defeated with this Enter image XiActual glass bubble density figure FiBetween Euclidean distance it is minimum, comprising:
The input picture XiInto after the convolutional neural networks, by convolution, Chi Hua, full attended operation obtain with it is described Input picture XiActual glass bubble density figure FiBetween the smallest glass blister density map F (X of Euclidean distancei,θ)。
5. glass blister detection as claimed in claim 4 and method of counting, which is characterized in that the damage of the convolutional neural networks Losing function is input picture XiGlass blister density map F (Xi；θ) with input picture XiActual glass bubble density figure Fi Between Euclidean distance L (θ):
Wherein, θ represents the convolutional neural networks parameter to be optimized, and N represents the number of training set picture, XiRepresenting input images, Fi Representing input images XiCorresponding glass density figure, F (Xi；θ) represent network-evaluated glass blister density map, the convolution mind After one initial value of network settings, go out to input the loss of picture according to standard density graphic calculation: L (θ), then in each suboptimization The parameter θ that whole network is updated in iteration, until penalty values converge to the value for being less than preset threshold.
6. glass blister detection as claimed in claim 5 and method of counting, which is characterized in that in Optimized Iterative each time more The parameter θ of new whole network, comprising:
Using stochastic gradient descent method, the network weight parameter θ of the convolutional neural networks is updated in Optimized Iterative each time.
7. glass blister detection as claimed in claim 5 and method of counting, which is characterized in that the input picture XiIt is corresponding Actual glass bubble density figure FiIt indicates are as follows:
In formula: N is input picture XiThe number of middle glass blister,Indicate input picture XiIn each pixel position, xiIt is i-th A glass blister is in input picture XiIn position, δ () be unit impulse function, * is convolution operation,It is for standard deviation σiGaussian kernel.
CN201810975382.8A 2018-08-24 2018-08-24 Glass blister detection and method of counting Pending CN109191440A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810975382.8A CN109191440A (en) 2018-08-24 2018-08-24 Glass blister detection and method of counting

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810975382.8A CN109191440A (en) 2018-08-24 2018-08-24 Glass blister detection and method of counting

Publications (1)

Publication Number Publication Date
CN109191440A true CN109191440A (en) 2019-01-11

Family

ID=64919837

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810975382.8A Pending CN109191440A (en) 2018-08-24 2018-08-24 Glass blister detection and method of counting

Country Status (1)

Country Link
CN (1) CN109191440A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109840909A (en) * 2019-01-18 2019-06-04 西安科技大学 A kind of crucible bubble counting device and method of counting
CN115610739A (en) * 2022-09-26 2023-01-17 江阴瑞兴塑料玻璃制品有限公司 Bubble state detection platform for glass film pasting machine

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1181136A (en) * 1995-12-22 1998-05-06 圣戈班电影及控制公司 Method for checking glass container
US5982920A (en) * 1997-01-08 1999-11-09 Lockheed Martin Energy Research Corp. Oak Ridge National Laboratory Automated defect spatial signature analysis for semiconductor manufacturing process
CN101750422A (en) * 2010-01-07 2010-06-23 秦皇岛凯维科技有限公司 On-line automatic detection device for glass defect
CN202916216U (en) * 2012-10-25 2013-05-01 淄博中材庞贝捷金晶玻纤有限公司 Glass bubble amount detection device
CN104730087A (en) * 2014-12-12 2015-06-24 南通路博石英材料有限公司 Device for observing air bubbles in transparent layer of quartz glass crucible
CN105717137A (en) * 2016-01-27 2016-06-29 中国建筑材料科学研究总院 Silica-glass micro-defect detecting method
CN106650913A (en) * 2016-12-31 2017-05-10 中国科学技术大学 Deep convolution neural network-based traffic flow density estimation method
CN107657226A (en) * 2017-09-22 2018-02-02 电子科技大学 A kind of Population size estimation method based on deep learning

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1181136A (en) * 1995-12-22 1998-05-06 圣戈班电影及控制公司 Method for checking glass container
US5982920A (en) * 1997-01-08 1999-11-09 Lockheed Martin Energy Research Corp. Oak Ridge National Laboratory Automated defect spatial signature analysis for semiconductor manufacturing process
CN101750422A (en) * 2010-01-07 2010-06-23 秦皇岛凯维科技有限公司 On-line automatic detection device for glass defect
CN202916216U (en) * 2012-10-25 2013-05-01 淄博中材庞贝捷金晶玻纤有限公司 Glass bubble amount detection device
CN104730087A (en) * 2014-12-12 2015-06-24 南通路博石英材料有限公司 Device for observing air bubbles in transparent layer of quartz glass crucible
CN105717137A (en) * 2016-01-27 2016-06-29 中国建筑材料科学研究总院 Silica-glass micro-defect detecting method
CN106650913A (en) * 2016-12-31 2017-05-10 中国科学技术大学 Deep convolution neural network-based traffic flow density estimation method
CN107657226A (en) * 2017-09-22 2018-02-02 电子科技大学 A kind of Population size estimation method based on deep learning

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
周云龙等: ""基于图像法的气液两相稀疏泡状流气泡参数分析"", 《核科学与工程》 *
陈艳燕等: ""机器视觉钢化玻璃绝缘子气泡的缺陷检测研究"", 《自动化仪表》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109840909A (en) * 2019-01-18 2019-06-04 西安科技大学 A kind of crucible bubble counting device and method of counting
CN109840909B (en) * 2019-01-18 2021-05-25 西安科技大学 Crucible bubble counting device and counting method
CN115610739A (en) * 2022-09-26 2023-01-17 江阴瑞兴塑料玻璃制品有限公司 Bubble state detection platform for glass film pasting machine
CN115610739B (en) * 2022-09-26 2023-05-16 江阴瑞兴塑料玻璃制品有限公司 Glass film pasting machinery bubble state detection platform

Similar Documents

Publication Publication Date Title
CN105528589B (en) Single image crowd's counting algorithm based on multiple row convolutional neural networks
WO2020239015A1 (en) Image recognition method and apparatus, image classification method and apparatus, electronic device, and storage medium
CN106875373B (en) Mobile phone screen MURA defect detection method based on convolutional neural network pruning algorithm
CN105657402B (en) A kind of depth map restoration methods
CN108960135B (en) Dense ship target accurate detection method based on high-resolution remote sensing image
CN104202547B (en) Method, projection interactive approach and its system of target object are extracted in projected picture
US11610289B2 (en) Image processing method and apparatus, storage medium, and terminal
CN107844753A (en) Pedestrian in video image recognition methods, device, storage medium and processor again
CN110766058B (en) Battlefield target detection method based on optimized RPN (resilient packet network)
CN106934408A (en) Identity card picture sorting technique based on convolutional neural networks
CN111275660B (en) Flat panel display defect detection method and device
CN109918523B (en) Circuit board component detection method based on YOLO9000 algorithm
CN111242026B (en) Remote sensing image target detection method based on spatial hierarchy perception module and metric learning
CN107680053A (en) A kind of fuzzy core Optimized Iterative initial value method of estimation based on deep learning classification
CN107240126B (en) Array image calibration method
CN107909053B (en) Face detection method based on hierarchical learning cascade convolution neural network
CN109191440A (en) Glass blister detection and method of counting
CN113592866A (en) Semiconductor lead frame exposure defect detection method
CN107343196A (en) One kind mixing distortion non-reference picture quality appraisement method
CN113888542B (en) Product defect detection method and device
CN110659637A (en) Electric energy meter number and label automatic identification method combining deep neural network and SIFT features
CN114021704B (en) AI neural network model training method and related device
CN110544249A (en) Convolutional neural network quality identification method for arbitrary-angle case assembly visual inspection
CN108089753B (en) Positioning method for predicting fingertip position by using fast-RCNN
US11508187B2 (en) Pupil detection 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
RJ01 Rejection of invention patent application after publication

Application publication date: 20190111

RJ01 Rejection of invention patent application after publication