CN108596892A - A kind of identification of Weld Defects based on improvement LeNet-5 models - Google Patents
A kind of identification of Weld Defects based on improvement LeNet-5 models Download PDFInfo
- Publication number
- CN108596892A CN108596892A CN201810368229.9A CN201810368229A CN108596892A CN 108596892 A CN108596892 A CN 108596892A CN 201810368229 A CN201810368229 A CN 201810368229A CN 108596892 A CN108596892 A CN 108596892A
- Authority
- CN
- China
- Prior art keywords
- lenet
- models
- channel
- image
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 230000007547 defect Effects 0.000 title claims abstract description 56
- 230000006872 improvement Effects 0.000 title claims abstract description 17
- 238000013528 artificial neural network Methods 0.000 claims abstract description 31
- 238000005516 engineering process Methods 0.000 claims abstract description 9
- 238000009472 formulation Methods 0.000 claims abstract description 6
- 239000000203 mixture Substances 0.000 claims abstract description 6
- 230000009466 transformation Effects 0.000 claims description 14
- 238000000605 extraction Methods 0.000 claims description 13
- 238000000034 method Methods 0.000 claims description 13
- 238000001514 detection method Methods 0.000 claims description 5
- 238000003062 neural network model Methods 0.000 claims description 4
- 230000035515 penetration Effects 0.000 claims description 4
- 208000037656 Respiratory Sounds Diseases 0.000 claims description 3
- 230000001131 transforming effect Effects 0.000 claims description 2
- 239000010410 layer Substances 0.000 description 18
- 238000013527 convolutional neural network Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000007796 conventional method Methods 0.000 description 4
- 230000004927 fusion Effects 0.000 description 4
- 230000008859 change Effects 0.000 description 2
- 238000012512 characterization method Methods 0.000 description 2
- 230000007704 transition Effects 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 244000287680 Garcinia dulcis Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 239000011229 interlayer Substances 0.000 description 1
- 238000002156 mixing Methods 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 230000000474 nursing effect Effects 0.000 description 1
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
- 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
-
- G06T5/92—
-
- 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
- G06T2207/30152—Solder
Abstract
The invention discloses a kind of based on the identification of Weld Defects for improving 5 models of LeNet, first against weld seam gray level image, improvement is made to the input in the 5 model tradition convolution kernel channels LeNet, gray level image is converted into coloured image by virtual color display technology, and using obtained coloured image as the input of neural network；Then improvement is made to 5 model convolution kernels of LeNet, adds the convolution kernel channel with Gabor filter；In neural network layer 6, the feature that multiple channels obtain is merged, obtains characteristic set T；Finally, SoftMax graders are used in neural network layer 7 (output layer), obtains the defect type of weld seam and its belongs to probability of all categories, for comment piece personnel to judge that egative film type and the formulation of live reworking plan provide reference.The present invention improves neural network characteristics extractability, to improve the accuracy of defect recognition；The probability that recognition result belongs to certain classification with defect provides, to comment piece personnel to provide more from the reference information divided.
Description
Technical field
The invention belongs to Welding Line Flaw Detection technical fields, and in particular to a kind of based on the weld seam for improving LeNet-5 models
Defect identification method.
Background technology
In weld defect automatic identification field, conventional method will inevitably undergo the process of artificial selected characteristic, should
Process time and effort consuming, and whether the selection of feature rationally has prodigious subjectivity, has larger impact for recognition correct rate.
Weld image is gray level image, and gray level image is directly inputted in neural network, and there are primitive character characterizations not
Enough abundant problems.
The convolution kernel that existing convolutional neural networks often rely on single type carries out convolution process, is easy to cause feature extraction
Insufficient problem, to influence the accuracy of defect recognition.
Meanwhile at present due to relevant policies specification, defect type must be by commenting piece personnel to judge, the knot of defect inspection
Fruit is only as the reference for commenting piece personnel.Automatic identification algorithm is mostly only to provide a kind of final recognition result at present, lacks result
The quantificational description of probability comments the value that piece personnel can refer to limited, and is unfavorable for formulating reworking plan.
Invention content
In view of the above-mentioned deficiencies in the prior art, the technical problem to be solved by the present invention is that providing a kind of based on improvement
The identification of Weld Defects of LeNet-5 models avoids the process of the artificial selected characteristic of conventional method；And pass through pseudo-colours
Converter technique has expanded the information content of neural network input, by adding the convolutional channel of Gabor filter, makes nerve after improvement
Network had not only had traditional convolution kernel channel, but also had Gabor filter channel, neural network characteristics extractability was improved, to carry
The accuracy of high defect recognition；The probability that recognition result belongs to certain classification with defect provides, for comment piece personnel provide more from
The reference information divided.
The present invention uses following technical scheme：
A kind of identification of Weld Defects based on improvement LeNet-5 models is right first against weld seam gray level image
Improvement is made in the input in LeNet-5 model tradition convolution kernels channel, and weld seam gray level image is converted by virtual color display technology
For coloured image, and using obtained coloured image as the input of neural network tradition convolutional channel；Then to LeNet-5 models
Convolution kernel makes improvement, adds the convolutional channel with Gabor filter；In neural network layer 6, two channels are obtained
Feature merged to obtain characteristic set T；Finally, SoftMax graders are used in neural network output layer, obtains weld seam
Defect type and its belong to probability of all categories, for comment piece personnel to judge that egative film type and live reworking plan formulation provide
With reference to.
Specifically, by gray level image by virtual color display technology, RGB color transformation is carried out to original-gray image, and
Obtained coloured image is specific as follows as the input of neural network：
S101, it obtains comprising crackle, does not merge, is lack of penetration, elongated or circular flaw type weld seam detection gray scale image；
S102, weld seam detection gray level image is subjected to RGB color space transformation, then is input to LeNet-5 model conventional rolls
Product core channel.
Further, gray level image pseudocolor transformation, the three-component change of red, green, blue are realized using pixel itself converter technique
Exchange the letters number is as follows：
Wherein, f (x, y) is the gray value at gray-scale map picture point (x, y).
Specifically, on the basis of LeNet-5 models, the Gabor filter convolutional channel with different directions is built, i.e.,
Convolution is carried out to input picture using Gabor filter in the channel first layer, the characteristics of for weld defect edge blurry, is adopted
Feature extraction is carried out to imaginary part with Gabor filter.
Further, specific as follows to imaginary part progress feature extraction using Gabor filter：
Wherein, x '=xcos θ+sin θ；Y '=- xsin θ+cos θ；λ, θ, ψ, σ, γ are respectively wavelength, direction, and phase is inclined
It moves, the standard deviation and length-width ratio of Gaussian function.
Further, the direction of Gabor filter convolutional channel is 0 degree, 45 degree, 90 degree and 135 degree.
Specifically, structure has 7 layers of structure and includes the binary channels god in traditional convolution kernel channel and Gabor filter channel
Through network model, the feature that each channel obtains is merged in neural network layer 6, obtains twin-channel characteristic set
T。
Specifically, using SoftMax multi classifiers, the probability of the corresponding output result Yi of i-th kind of defect of image is obtained
Information P (Y=Yi), the result type based on Probability Forms is obtained, to comment piece personnel to judge egative film type, live reworking plan system
It is fixed that reference is provided.
Further, probabilistic information P (Y=Yi) mathematic(al) representation it is as follows：
Wherein, YiThe corresponding output of i-th kind of defect of image is indicated as a result, i=1,2,3,4,5,6, K be classification number.
Compared with prior art, the present invention at least has the advantages that：
The present invention is a kind of based on the identification of Weld Defects for improving LeNet-5 models, first against weld seam gray level image,
Improvement is made to the input in LeNet-5 model tradition convolution kernels channel, gray level image is converted to by virtual color display technology
Coloured image, and using obtained coloured image as the input of neural network；Then LeNet-5 model convolution kernels are made and changes
Into convolutional channel of the addition with Gabor filter；In neural network layer 6, the feature that multiple channels obtain is melted
It closes；Finally the feature after fusion is input in SoftMax multi classifiers, obtain the defect type of weld seam and its belongs to all kinds of
Other probability, for comment piece personnel to judge that egative film type and the formulation of live reworking plan provide reference, weld seam egative film is defeated
Enter to before neural network, it is pre-processed, i.e., is the cromogram of RGB triple channels by single pass greyscale image transitions
Picture has expanded the information content of input picture；It is added to the convolutional channel with Gabor cores, enhances neural network characteristics extraction
Ability, avoid the process of artificial selected characteristic, for comment piece personnel judge defect type more sufficient information is provided, be alternative
The formulation of reworking plan provides reference.
Further, coloured image has more abundant information than gray level image, will especially in terms of details characterization
Input of the coloured image as neural network enhances ability to express of the image to weld seam internal flaw.
It further, can be to avoid artificial in conventional method when weld defect being identified using convolutional neural networks
Extraction characteristic procedure；Since original weld image is gray level image, directly it is entered into neural network, there are original welderings
The insufficient problem of seam defect information representation has higher requirement to the ability in feature extraction of neural network.
Further, in order to obtain feature of the image along different directions, and also to which set direction is avoided excessively to be led
The problem of feature redundancy of cause, choice direction are that the filter in 0 degree, 45 degree, 90 degree and 135 degree direction carries out feature extraction.
Further, it by traditional convolution kernel and different from Gabor filter convolution nuclear structure, can extract accordingly not
Same feature enhances the ability in feature extraction of model using binary channels neural network model；The feature that each channel is obtained
It is merged, obtains twin-channel characteristic set T so that the characteristics of image extracted is more fully.
Further, using SoftMax multi classifiers, defect type and its probability are obtained, to comment piece personnel to provide
More from point reference information.
Below by drawings and examples, technical scheme of the present invention will be described in further detail.
Description of the drawings
Fig. 1 is block diagram of the present invention；
Fig. 2 is neural network model figure of the present invention；
Fig. 3 is original weld image recognition correct rate figure under different iterations on LeNet-5 models；
Fig. 4 is image recognition correct rate figure under different iterations on LeNet-5 models after virtual color display；
Fig. 5 is image recognition correct rate figure under different iterations on LeNet-5 models after improvement after virtual color display.
Specific implementation mode
The present invention provides a kind of based on the identification of Weld Defects for improving LeNet-5 models, for weld seam gray-scale map
Picture makes improvement to the input in LeNet-5 model tradition convolution kernels channel, i.e., by gray level image by virtual color display technology,
Coloured image is converted to, and using obtained coloured image as the input of neural network；LeNet-5 model convolution kernels are made and are changed
Into, convolutional channel of the addition with Gabor filter, and in neural network layer 6, the feature that multiple channels are obtained carries out
Fusion；SoftMax multi classifiers, the defect for obtaining weld seam is used to belong to probability of all categories in output layer.
Referring to Fig. 1, the present invention is a kind of based on the identification of Weld Defects for improving LeNet-5 models, including following step
Suddenly：
S1, it is directed to weld seam gray level image, improvement is made to the input in traditional convolution kernel channel of LeNet-5 models, i.e., will
Gray level image carries out RGB color transformation by virtual color display technology, to original-gray image, and obtained coloured image is made
For the input of neural network tradition convolutional channel；
S101, it obtains and is examined comprising defect type (such as crackle does not merge, is lack of penetration, elongated defect, circular flaw) weld seam
Survey gray scale image；
S102, it proposes weld seam gray level image carrying out RGB color space transformation so that image has than original-gray image
More information, then it is input to LeNet-5 model tradition convolution kernels channel.
Wherein pixel itself converter technique is used to realize gray level image pseudocolor transformation, the three-component transforming function transformation function of red, green, blue
As follows：
Wherein, f (x, y) is the gray value at gray-scale map picture point (x, y).
S2, improvement is made to LeNet-5 model convolution kernels, add the convolutional channel with Gabor filter, make nerve net
The existing traditional convolution kernel channel of network, and there is Gabor filter convolutional channel, to enhance the ability in feature extraction of model；
It is proposed on the basis of LeNet-5 models, input layer with output interlayer structure size be 5 × 5 and with 4 kinds not
The convolutional channel of equidirectional (0 degree, 45 degree, 90 degree, 135 degree of directions) Gabor filter, and determine each layer Gabor filter ginseng
Number, each layer filter construction are as shown in Figure 2；The characteristics of for weld defect edge blurry, is carried out using Gabor filter imaginary part
Feature extraction, shown in mathematic(al) representation such as formula (2)：
Wherein, x '=xcos θ+sin θ；Y '=- xsin θ+cos θ；λ, θ, ψ, σ, γ are respectively wavelength, direction, and phase is inclined
It moves, the standard deviation and length-width ratio of Gaussian function.
S3, on the basis of LeNet-5 models, build the binary channels neural network model with 7 layers of structure, i.e. model includes
Traditional convolution kernel channel and Gabor cores channel, realize feature extraction；Wherein, multiple channels are obtained in neural network layer 6
Feature merged, obtain two channels characteristic set T；
S301, structure binary channels convolutional neural networks model (had not only had traditional convolution kernel channel, but also with Gabor filtering
The model of the convolutional channel of device) it is 6 layers first；
S302, the feature for extracting each channel carry out Fusion Features at the 6th layer, obtain characteristic set T.
S4, it is to obtain weld defect classification results and its probabilistic information, using SoftMax multi classifiers, in network the 7th
Layer is classified, and be can get the defect type of weld seam and its is belonged to probability of all categories, to comment piece personnel to judge egative film type, showing
Field reworking plan, which is formulated, provides reference.
To obtain each defect classification results Yi(YiIndicate image theiThe corresponding output of kind of defect is as a result, wherein i=1, and 2,3,
4,5,6) probabilistic information P (Y=Yi), using SoftMax multi classifiers, obtain output YiProbability, probabilistic information P (Y=
Yi) mathematic(al) representation it is as follows：
Wherein, K is classification number.
The result type based on Probability Forms is obtained, to comment piece personnel to judge that egative film type, live reworking plan formulation carry
For reference.
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.The present invention being described and shown in usually here in attached drawing is real
Applying the component of example can be arranged and be designed by a variety of different configurations.Therefore, the present invention to providing in the accompanying drawings below
The detailed description of embodiment be not intended to limit the range of claimed invention, but be merely representative of the selected of the present invention
Embodiment.Based on the embodiments of the present invention, those of ordinary skill in the art are obtained without creative efforts
The every other embodiment obtained, shall fall within the protection scope of the present invention.
First, it is coloured image by greyscale image transitions by the pixel Self-variation method based on rgb space, and will
Input of the coloured image arrived as neural network.
Later, on the basis of LeNet-5 models, addition contains four direction (0 degree, 45 degree, 90 degree, 135 degree)
The convolutional channel of Gabor filter carries out image using Gabor filter in Gabor filter convolutional channel first layer
Convolution, and merge the feature that multiple channels are extracted in layer 6, obtain improved convolutional neural networks.
Feature after fusion is input in SoftMax multi classifiers, the defect for obtaining weld seam belongs to of all categories general
Rate.
It is illustrated below with specific example, it is 32x32 weld images to choose 626 small greatly, wherein being split comprising 153
Print image, 87 non-blending images, 180 lack of penetration images, 64 elongated defect maps, 94 circular flaw figures.It is wherein all kinds of
80% is not accounted for by training set, test set accounts for 20% pro rate.Finally, training set totally 488, test set are schemed for 138 totally
Picture.
Training set picture is input in LeNet-5, the accuracy that observing and nursing identifies on test set, in different iteration
Recognition correct rate shown in Fig. 2 can be obtained under number.
As seen from Figure 2, directly original weld image is input in neural network, in 180 times to 200 times iteration mistakes
The accuracy of Cheng Zhong, defect recognition maintain 71% or so, and the correct recognition rata of defect classification is relatively low.
For weld seam original image, [0,1] section is normalized to according to formula (1) using pixel transformation into itself's method
R, the value of G, B all, and the value of R, G, B all is multiplied by obtain behind 255 and rounding and is used as coloured image RGB triple channel values, i.e., it will weldering
Seam original-gray image is changed into the coloured image with RGB triple channels, then obtained coloured image training set collection is input to
In LeNet-5 models, under different iterations, the recognition correct rate on test set is observed, can be obtained as shown in Figure 3
As a result.
By the result of Fig. 3 it is found that when original image is after pseudocolor transformation, then it is input in LeNet-5 models and carries out
Identification, accuracy reaches 80% or more after 200 iteration, and under 160 times to 200 times iterations, and accuracy maintains
80% or so.It can be seen that after being input to again in LeNet-5 models after original weld image diameter pseudocolor transformation, weld seam lacks
Sunken recognition correct rate, which has, to be obviously improved, but the accuracy compared with conventional method also has a certain distance.
Improvement is made to LeNet-5 models, convolutional channel of the addition with Gabor filter between input and output, and
Coloured image training set after virtual color display is input in improved LeNet-5 models, it is corresponding with most probable value
Defect type is output type, and observation test set obtains recognition correct rate as shown in Figure 4 in different iteration.
As shown in Figure 4, defect recognition accuracy further rises, and reaches 90% or more.
It is the recognition result for randomly selecting 3 images in table 1
Obtain the output valve Y of neural networki, and the probability that a certain image belongs to i-th kind of defect has been obtained according to formula (3)
P (Y=Yi), to comment piece personnel to provide more detailed reference information.
The problem of needing artificial selected characteristic the invention avoids traditional shortcoming identification；Convolutional neural networks input figure is expanded
The information content of picture has been extracted compared with the more sufficient features of LeNet-5.It is demonstrated experimentally that the present invention defect recognition rate compared with
LeNet-5 models, which have, to be obviously improved.
The above content is merely illustrative of the invention's technical idea, and protection scope of the present invention cannot be limited with this, every to press
According to technological thought proposed by the present invention, any change done on the basis of technical solution each falls within claims of the present invention
Protection domain within.
Claims (9)
1. a kind of based on the identification of Weld Defects for improving LeNet-5 models, which is characterized in that first against weld seam gray-scale map
Picture makes improvement to the input in LeNet-5 model tradition convolution kernels channel, weld seam gray level image is passed through virtual color display technology
Coloured image is converted to, and using obtained coloured image as the input of neural network tradition convolutional channel；Then to LeNet-5
Model convolution kernel makes improvement, adds the convolutional channel with Gabor filter；In neural network layer 6, by two channels
Obtained feature is merged to obtain characteristic set T；Finally, SoftMax graders are used in neural network output layer, is welded
The defect type of seam and its belong to probability of all categories, for comment piece personnel to judge that egative film type and live reworking plan are formulated
Reference is provided.
2. according to claim 1 a kind of based on the identification of Weld Defects for improving LeNet-5 models, feature exists
In by gray level image by virtual color display technology, RGB color transformation, and the colour that will be obtained are carried out to original-gray image
Input of the image as neural network tradition convolutional channel, it is specific as follows：
S101, it obtains comprising crackle, does not merge, is lack of penetration, elongated or circular flaw type weld seam detection gray scale image；
S102, weld seam detection gray level image is subjected to RGB color space transformation, then is input to LeNet-5 model tradition convolution kernels
Channel.
3. according to claim 2 a kind of based on the identification of Weld Defects for improving LeNet-5 models, feature exists
In using pixel itself converter technique realization gray level image pseudocolor transformation, the three-component transforming function transformation function of red, green, blue is as follows：
Wherein, f (x, y) is the gray value at gray-scale map picture point (x, y).
4. according to claim 1 a kind of based on the identification of Weld Defects for improving LeNet-5 models, feature exists
In on the basis of LeNet-5 models, building the convolutional channel of the Gabor filter with different directions, i.e., in the channel the
One layer carries out convolution using Gabor filter to input picture, the characteristics of for weld defect edge blurry, is filtered using Gabor
Wave device carries out feature extraction to imaginary part.
5. according to claim 4 a kind of based on the identification of Weld Defects for improving LeNet-5 models, feature exists
In specific as follows to imaginary part progress feature extraction using Gabor filter：
Wherein, x '=xcos θ+sin θ；Y '=- xsin θ+cos θ；λ, θ, ψ, σ, γ are respectively wavelength, direction, phase offset, height
The standard deviation and length-width ratio of this function.
6. according to claim 5 a kind of based on the identification of Weld Defects for improving LeNet-5 models, feature exists
In the direction of Gabor filter convolutional channel is 0 degree, 45 degree, 90 degree and 135 degree.
7. according to claim 1 a kind of based on the identification of Weld Defects for improving LeNet-5 models, feature exists
In, the binary channels neural network model that there are 7 layers of structure and include traditional convolution kernel channel and Gabor filter channel is built,
The feature that each channel obtains is merged in neural network layer 6, obtains twin-channel characteristic set T.
8. according to claim 1 a kind of based on the identification of Weld Defects for improving LeNet-5 models, feature exists
In using SoftMax multi classifiers, the corresponding output result Y of acquisition i-th kind of defect of imageiProbabilistic information P (Y=Yi),
The result type based on Probability Forms is obtained, to comment piece personnel to judge that egative film type, live reworking plan formulation provide reference.
9. according to claim 8 a kind of based on the identification of Weld Defects for improving LeNet-5 models, feature exists
In probabilistic information P (Y=Yi) mathematic(al) representation it is as follows：
Wherein, YiThe corresponding output of i-th kind of defect of image is indicated as a result, i=1,2,3,4,5,6, K be classification number.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810368229.9A CN108596892B (en) | 2018-04-23 | 2018-04-23 | Weld defect identification method based on improved LeNet-5 model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810368229.9A CN108596892B (en) | 2018-04-23 | 2018-04-23 | Weld defect identification method based on improved LeNet-5 model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108596892A true CN108596892A (en) | 2018-09-28 |
| CN108596892B CN108596892B (en) | 2020-08-18 |
Family
ID=63614087
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810368229.9A Active CN108596892B (en) | 2018-04-23 | 2018-04-23 | Weld defect identification method based on improved LeNet-5 model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108596892B (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109741295A (en) * | 2018-11-28 | 2019-05-10 | 歌尔股份有限公司 | Product quality detection method and device |
| CN109902584A (en) * | 2019-01-28 | 2019-06-18 | 深圳大学 | A kind of recognition methods, device, equipment and the storage medium of mask defect |
| CN109916906A (en) * | 2018-10-11 | 2019-06-21 | 乐人株式会社 | Defect detecting device and method |
| CN110210497A (en) * | 2019-05-27 | 2019-09-06 | 华南理工大学 | A kind of real-time characteristics of weld seam detection method of robust |
| CN110458828A (en) * | 2019-08-12 | 2019-11-15 | 广东工业大学 | A kind of laser welding defect identification method and device based on multi-modal fusion network |
| CN110572626A (en) * | 2019-09-03 | 2019-12-13 | 云南白药集团健康产品有限公司 | Image processing method and device |
| CN111178398A (en) * | 2019-12-13 | 2020-05-19 | 天翼电子商务有限公司 | Method, system, storage medium and device for detecting tampering of image information of identity card |
| CN111193916A (en) * | 2018-12-29 | 2020-05-22 | 中科寒武纪科技股份有限公司 | Operation method, device and related product |
| CN111612730A (en) * | 2019-02-23 | 2020-09-01 | 妙微(杭州)科技有限公司 | Welding seam detection and positioning method |
| CN112288678A (en) * | 2019-09-11 | 2021-01-29 | 中北大学 | Pseudo-color display method and system for weld image |
| CN113240653A (en) * | 2021-05-19 | 2021-08-10 | 中国联合网络通信集团有限公司 | Rice quality detection method, device, server and system |
| CN113537176A (en) * | 2021-09-16 | 2021-10-22 | 武汉未来幻影科技有限公司 | Method, device and equipment for determining fatigue state of driver |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120294535A1 (en) * | 2010-02-05 | 2012-11-22 | S1 Corporation | Face detection method and apparatus |
| CN107292885A (en) * | 2017-08-08 | 2017-10-24 | 广东工业大学 | A kind of product defects classifying identification method and device based on autocoder |
-
2018
- 2018-04-23 CN CN201810368229.9A patent/CN108596892B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120294535A1 (en) * | 2010-02-05 | 2012-11-22 | S1 Corporation | Face detection method and apparatus |
| CN107292885A (en) * | 2017-08-08 | 2017-10-24 | 广东工业大学 | A kind of product defects classifying identification method and device based on autocoder |
Non-Patent Citations (5)
| Title |
|---|
| 吴杰: "基于卷积神经网络的行为识别研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 张鹏: "基于改进卷积神经网络的交通标志牌识别算法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 曹茂永等: "基于RGB三基色和像素自身变换的灰度图像伪彩色编码", 《工程图学学报》 * |
| 牛新亚: "基于深度学习的人脸表情识别研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 邹焱飚等: "基于深度分层特征的激光视觉焊缝检测与跟踪系统研究", 《中国激光》 * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109916906A (en) * | 2018-10-11 | 2019-06-21 | 乐人株式会社 | Defect detecting device and method |
| CN109916906B (en) * | 2018-10-11 | 2022-10-14 | 乐人株式会社 | Defect detection device and method |
| CN109741295A (en) * | 2018-11-28 | 2019-05-10 | 歌尔股份有限公司 | Product quality detection method and device |
| CN109741295B (en) * | 2018-11-28 | 2021-04-20 | 歌尔股份有限公司 | Product quality detection method and device |
| CN111193916A (en) * | 2018-12-29 | 2020-05-22 | 中科寒武纪科技股份有限公司 | Operation method, device and related product |
| CN111193916B (en) * | 2018-12-29 | 2022-03-29 | 中科寒武纪科技股份有限公司 | Operation method |
| CN109902584B (en) * | 2019-01-28 | 2022-02-22 | 深圳大学 | Mask defect identification method, device, equipment and storage medium |
| CN109902584A (en) * | 2019-01-28 | 2019-06-18 | 深圳大学 | A kind of recognition methods, device, equipment and the storage medium of mask defect |
| CN111612730B (en) * | 2019-02-23 | 2022-04-22 | 妙微(杭州)科技有限公司 | Welding seam detection and positioning method |
| CN111612730A (en) * | 2019-02-23 | 2020-09-01 | 妙微(杭州)科技有限公司 | Welding seam detection and positioning method |
| CN110210497A (en) * | 2019-05-27 | 2019-09-06 | 华南理工大学 | A kind of real-time characteristics of weld seam detection method of robust |
| CN110458828A (en) * | 2019-08-12 | 2019-11-15 | 广东工业大学 | A kind of laser welding defect identification method and device based on multi-modal fusion network |
| CN110572626A (en) * | 2019-09-03 | 2019-12-13 | 云南白药集团健康产品有限公司 | Image processing method and device |
| CN112288678A (en) * | 2019-09-11 | 2021-01-29 | 中北大学 | Pseudo-color display method and system for weld image |
| CN112288678B (en) * | 2019-09-11 | 2022-05-27 | 中北大学 | Pseudo-color display method and system for welding seam image |
| CN111178398A (en) * | 2019-12-13 | 2020-05-19 | 天翼电子商务有限公司 | Method, system, storage medium and device for detecting tampering of image information of identity card |
| CN111178398B (en) * | 2019-12-13 | 2023-08-22 | 天翼商业保理有限公司 | Method, system, storage medium and device for detecting tampering of identity card image information |
| CN113240653A (en) * | 2021-05-19 | 2021-08-10 | 中国联合网络通信集团有限公司 | Rice quality detection method, device, server and system |
| CN113537176A (en) * | 2021-09-16 | 2021-10-22 | 武汉未来幻影科技有限公司 | Method, device and equipment for determining fatigue state of driver |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108596892B (en) | 2020-08-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108596892A (en) | A kind of identification of Weld Defects based on improvement LeNet-5 models | |
| CN111612763B (en) | Mobile phone screen defect detection method, device and system, computer equipment and medium | |
| CN106204779B (en) | Check class attendance method based on plurality of human faces data collection strategy and deep learning | |
| CN104834933B (en) | A kind of detection method and device in saliency region | |
| CN106650786A (en) | Image recognition method based on multi-column convolutional neural network fuzzy evaluation | |
| CN107657257A (en) | A kind of semantic image dividing method based on multichannel convolutive neutral net | |
| CN111985499B (en) | High-precision bridge apparent disease identification method based on computer vision | |
| CN106934404A (en) | A kind of image flame identifying system based on CNN convolutional neural networks | |
| CN112132196B (en) | Cigarette case defect identification method combining deep learning and image processing | |
| JP2021005266A (en) | Image discrimination model construction method, image discrimination model, and image discrimination method | |
| US11790499B2 (en) | Certificate image extraction method and terminal device | |
| CN111105389B (en) | Road surface crack detection method integrating Gabor filter and convolutional neural network | |
| CN109920012A (en) | Image colorant system and method based on convolutional neural networks | |
| CN104463881B (en) | A kind of multi-spectrum remote sensing image change detecting method based on spectral reflectivity neighborhood disparity map and neighborhood probability fusion | |
| CN108062554B (en) | Method and device for identifying color of vehicle annual inspection label | |
| CN111008647B (en) | Sample extraction and image classification method based on void convolution and residual linkage | |
| CN109978854A (en) | A kind of screen content image quality measure method based on edge and structure feature | |
| CN106339984A (en) | Distributed image super-resolution method based on K-means driven convolutional neural network | |
| CN113160023A (en) | Land utilization checking system | |
| CN111259891B (en) | Method, device, equipment and medium for identifying identity card in natural scene | |
| CN113590906A (en) | Land utilization change proof checking system | |
| CN111080574A (en) | Fabric defect detection method based on information entropy and visual attention mechanism | |
| CN103544716B (en) | A kind of pixel to image carries out method and the device of color classification | |
| CN106645180A (en) | Method for checking defects of substrate glass, field terminal and server | |
| CN110909772B (en) | High-precision real-time multi-scale dial pointer detection method and system |
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 |