CN108061735A - The recognition methods of component surface defect and device - Google Patents
The recognition methods of component surface defect and device Download PDFInfo
- Publication number
- CN108061735A CN108061735A CN201711246947.0A CN201711246947A CN108061735A CN 108061735 A CN108061735 A CN 108061735A CN 201711246947 A CN201711246947 A CN 201711246947A CN 108061735 A CN108061735 A CN 108061735A
- Authority
- CN
- China
- Prior art keywords
- parts
- defect
- surface defect
- detected
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
- G01N2021/8854—Grading and classifying of flaws
Abstract
A kind of recognition methods of component surface defect and device, wherein, this method includes：Multiple sample images are obtained, wherein, multiple sample images include：The sample image of the sample image of multiple zero defect parts and multiple defective parts；According to multiple sample images, establish to identify the convolutional neural networks of surface defect；Obtain part diagram picture to be detected；According to part diagram picture to be detected and the convolutional neural networks for identifying surface defect, determine that parts to be detected whether there is surface defect.Since the program first with the foundation of multiple sample images by identifying the convolutional neural networks of surface defect, recycle above-mentioned convolutional neural networks identification component surface defect, so as to solve the technical issues of can not identifying component surface defect present in existing method exactly, reach the technique effect that component surface defect can be quickly and accurately identified in the poor environment of illumination.
Description
Technical field
The present invention relates to mechanical part detection technique field, more particularly to a kind of recognition methods of component surface defect
And device.
Background technology
It in production and construction, generally requires to be detected large batch of component of machine, with rapidly from numerous zero
It is rejected in part and does not meet production requirement, parts off quality.For example, the quick parts rejected there are surface defect.
At present, in order to rapidly identify component surface defect, existing method is acquisition part diagram picture mostly, passes through inspection
It surveys the method for gray threshold and the images of parts is subjected to binary conversion treatment, then by the image and template after above-mentioned binary conversion treatment
It is compared, to determine the surface quality situation of parts.But since the method for above-mentioned detection gray threshold compares in itself
Relatively simple, resolution ratio is relatively low, causes when it is implemented, identifying that the accuracy of component surface defect is not high.It is also, existing
Method is during implementation generally for the more demanding of light environment.In the case where light environment is weaker, existing method
Defect can not often be efficiently identified out.In summary, existing method is when it is implemented, zero can not be identified exactly by often existing
The technical issues of parts surface defect.
In view of the above-mentioned problems, currently no effective solution has been proposed.
The content of the invention
The application embodiment provides recognition methods and the device of a kind of component surface defect, to solve to have in method
Existing the technical issues of can not identifying component surface defect exactly.
The embodiment of the present application provides a kind of recognition methods of component surface defect, including：
Multiple sample images are obtained, wherein, the multiple sample image includes：The sample image of multiple zero defect parts
With the sample image of multiple defective parts；
According to the multiple sample image, establish to identify the convolutional neural networks of surface defect；
Obtain part diagram picture to be detected；
According to the part diagram picture to be detected and the convolutional neural networks for being used to identify surface defect, determine to treat
The parts of detection whether there is surface defect.
In one embodiment, the parts include at least one of：Bearing, screw, gear, pneumatic element,
Sealing element, fastener.
In one embodiment, according to the multiple sample image, establish to identify the convolutional Neural of surface defect
Network, including：
According to the sample image of the multiple zero defect parts and the sample image of multiple defective parts, is established
One training set and the first test set；
It is trained using first training set and first test set, to establish parts defect recognition network,
And using the parts defect recognition network as the convolutional neural networks for being used to identify surface defect.
In one embodiment, the sample image of the multiple defective parts includes a variety of parts defect types
Image.
In one embodiment, a variety of parts defect type images include：Crack-type image, cut type
Image, hole types of image, mottling types of image.
In one embodiment, it is described according to the multiple sample image, it establishes to identify the convolution of surface defect
Neutral net, including：
According to the sample image of the multiple zero defect parts and the sample image of multiple defective parts, is established
One training set and the first test set；
It is trained using first training set and first test set, to establish the parts defect recognition net
Network；
According to a variety of parts defect type images, the second training set and the second test set are established；
It is trained using second training set and second test set, to establish defect type sorter network；
The parts defect recognition network and the defect type sorter network are used to identify that surface lacks as described
Sunken convolutional neural networks.
In one embodiment, according to the part diagram picture to be detected and the volume for being used to identify surface defect
Product neutral net determines that parts to be detected whether there is surface defect, including：
The part diagram picture to be detected is identified using the parts defect recognition network, it is described to determine
Parts to be detected whether there is surface defect；
In the case where the definite parts to be detected are there are surface defect, the defect type sorter network is utilized
Determine the type of the component surface defect to be detected.
In one embodiment, after multiple sample images are obtained, the method further includes：
Image procossing is carried out to the multiple sample image, multiple sample images that obtain that treated, wherein, described image
Processing includes at least one of：Rotation processing, mirror image processing and gray scale adjusting processing；
Correspondingly, according to the multiple sample image, establish to identify the convolutional neural networks of surface defect, including：
According to treated multiple sample images, establish described for identifying the convolutional neural networks of surface defect.
In one embodiment, after part diagram picture to be detected is obtained, the method further includes：
Gaussian filtering is carried out to the part diagram picture to be detected, obtains the part diagram picture to be detected after denoising；
Correspondingly, according to the part diagram picture to be detected and the convolutional Neural net for being used to identify surface defect
Network determines that parts to be detected whether there is surface defect, including：
According to the part diagram picture to be detected after the denoising and the convolutional Neural net for being used to identify surface defect
Network determines that parts to be detected whether there is surface defect.
The embodiment of the present application additionally provides a kind of identification device of component surface defect, including：
First acquisition module, for obtaining multiple sample images, wherein, the multiple sample image includes：It is multiple intact
Fall into the sample image of parts and the sample image of multiple defective parts；
Module is established, for according to the multiple sample image, establishing to identify the convolutional neural networks of surface defect；
Second acquisition module, for obtaining part diagram picture to be detected；
Determining module, for according to the part diagram picture to be detected and the convolution god for being used to identify surface defect
Through network, determine that parts to be detected whether there is surface defect.
In the embodiment of the present application, establish to identify the convolutional Neural net of surface defect by using multiple sample images
Network, recycling is above-mentioned for identifying that part diagram picture to be detected is identified in the convolutional neural networks of surface defect, with true
Fixed component surface defect to be detected, can not identify that component surface lacks exactly so as to solve present in existing method
The technical issues of falling into, has reached the technology that component surface defect can be quickly and accurately identified in the poor environment of illumination
Effect.
Description of the drawings
Below by a manner of clearly understandable, preferred embodiment is described with reference to the drawings, the present invention is compiled based on figure
Code is in the one-way data transmission method of physical isolation terminal room and above-mentioned characteristic, technical characteristic, advantage and its realization side of system
Formula is further described.
Fig. 1 is the process chart of the recognition methods of the component surface defect provided according to the application embodiment；
Fig. 2 is the composition structure chart of the identification device of the component surface defect provided according to the application embodiment；
The electronic equipment group of the recognition methods of the component surface defect shown in Fig. 3 provided based on the application embodiment
Into structure chart；
Fig. 4 be in a Sample Scenario using component surface defect provided by the embodiments of the present application recognition methods and
Device carries out certain batch of bearing to be detected the flow diagram of Surface Defect Recognition.
Specific embodiment
It is in order to make those skilled in the art better understand the technical solutions in the application, real below in conjunction with the application
The attached drawing in example is applied, the technical solution in the embodiment of the present application is clearly and completely described, it is clear that described implementation
Example is merely a part but not all of the embodiments of the present application.Based on the embodiment in the application, this field is common
Technical staff's all other embodiments obtained without making creative work should all belong to the application protection
Scope.
It is often that basis is obtained by detecting gray threshold method in view of the recognition methods of existing component surface defect
The surface of the binary image identification parts obtained whether there is defect.Since the image resolution ratio of above-mentioned binaryzation itself is opposite
It is not high, and the above method is very high to the illumination requirement in implementation environment.Cause existing method when it is implemented, often can not be accurate
Really identify component surface defect, especially in the case where light environment is relatively poor, identification error is relatively large.Pin
To generating the basic reason of above-mentioned technical problem, the application considers convolutional neural networks in terms of image procossing and image identification
Advantage, proposition can need not identify surface defect by detecting the binary image of gray threshold method acquisition, but can
First to establish to identify the convolutional neural networks of surface defect using multiple sample images, convolutional neural networks identification is recycled
Component surface defect can not identify that the technology of component surface defect is asked exactly so as to solve present in existing method
Topic, has reached the technique effect that component surface defect can be quickly and accurately identified in the poor environment of illumination.
Based on above-mentioned thinking thinking, the application embodiment provides a kind of recognition methods of component surface defect.Tool
Body please refers to Fig.1 the process flow of the recognition methods of the shown component surface defect provided according to the application embodiment
Figure.The recognition methods for the component surface defect that the application embodiment provides, when it is implemented, may comprise steps of.
S11：Multiple sample images are obtained, wherein, the multiple sample image includes：The sample of multiple zero defect parts
The sample image of image and multiple defective parts.
In one embodiment, when it is implemented, multiple sample parts can be shot by CCD industrial cameras,
To obtain above-mentioned multiple sample images.Wherein, above-mentioned multiple sample parts can specifically include multiple flawless parts
With multiple defective parts.Wherein, above-mentioned CCD (Charge Coupled Device, photosensitive coupling component) is digital phase
It is used to record the semiconductor device of light variation in machine.
In one embodiment, the sum of above-mentioned multiple sample images can specifically be more than or equal to 500.Wherein, it is more
The sum of the sample image of a zero defect parts can be more than or equal to 250, the sample image of multiple defective parts
Sum can be more than or equal to 250.In this way, the sample compared with horn of plenty can be obtained, can subsequently to establish out accuracy of identification
The relatively higher convolutional neural networks for being used to identify surface defect.
In the present embodiment, the size of above-mentioned sample image can be specifically required as 512*512.Certainly, it is embodied
When, the sample image of other sizes can also be obtained, then corresponding image procossing is carried out to the sample image of acquisition, to be accorded with
The size for closing requirement is 512*512 images.
In one embodiment, above-mentioned parts can specifically refer to component of machine.Wherein, above-mentioned parts are specific
It can include but is not limited at least one of：Bearing, screw, gear, pneumatic element, sealing element, fastener etc..Certainly, need
It is noted that above-mentioned cited how middle parts are intended merely to that the application embodiment is better described.When it is implemented,
It can also select as the case may be and construction requirement to the other kinds of parts in addition to above-mentioned cited parts
Carry out the identification of corresponding surface defect.In addition, above-mentioned parts specifically can also be machinery involved in certain projects
Parts.Specifically, for example, it may be parts involved in industry internet integration test bed project.
S12：According to the multiple sample image, establish to identify the convolutional neural networks of surface defect.
In one embodiment, in order to establish the relatively high convolutional neural networks of identification surface defect precision, specifically
During implementation, it can be performed according to following steps：
S12-1：According to the sample image of the multiple zero defect parts and the sample image of multiple defective parts,
Establish the first training set and the first test set；
S12-2：It is trained using first training set and first test set, to establish parts defect recognition
Network, and using the parts defect recognition network as the convolutional neural networks for being used to identify surface defect.
In the present embodiment, above-mentioned convolutional neural networks can be specifically a kind of feedforward neural network, wherein, the nerve
Artificial neuron in network can respond the unit of surrounding.Usual convolutional neural networks can specifically include convolutional layer and pond
Layer etc..In addition, convolutional neural networks have relatively high precision in image procossing, image identification etc..
In the present embodiment, it is necessary to illustrate, above-mentioned convolutional neural networks are the bases in multilayer neural network
A kind of specially designed deep learning method for image classification and identification to grow up on plinth.In two kinds of volumes of this patent
In product neutral net, when it is implemented, input layer can be regarded to the neuron of two-dimensional matrix arrangement.With conventional neutral net
Similar, the neuron needs of input layer are connected with the neuron of hidden layer.But it is not by each input neuron here
It is connected with each hidden neuron, but only creates connection in the regional area of an image.For example, can using size as
Exemplified by the image of 28x28, if the region of a 5x5 of the neuron and input layer of first hidden layer connects, this 5x5
Region can be known as local sensing domain.Wherein, 25 neurons in the local sensing domain and first hidden layer is same
A neuron connection, each connection are upper again there are one weight parameter, therefore above-mentioned local sensing domain shares 5x5 weight.If
By local sensing domain along from left to right, order from top to bottom is slided, and will must correspond to neuron different in hidden layer.It is logical
24x24 neuron in available first hidden layer of the above method is crossed, these neurons are all using same 5x5 power
Weight.Wherein, the output of k-th of neuron can be specifically expressed as in j-th of hidden layer：
Wherein, here σ be neuron sigmoid excitation functions；B is the shared deviation of perception domain connection；ωl,mIt is
A 5x5 shares weight matrix；ax,yThe x in input layer is represented, the input value at y.In order to do the identification of the image of bearing, one complete
Whole convolutional layer needs to include several to tens different Feature Mappings.It is max-pooling to follow closely after convolutional layer
Pond layer, effect are the output of simplified convolutional layer.Each neuron in the layer of pond can be simply by preceding layer 2x2's
Maximum output in input domain.Last layer in network is a full articulamentum, i.e., each neuron of this layer with most
Each neuron connection of Max-pooling layers of the latter.It, must along with an output layer by the multiple collocation of said structure
To a complete convolutional neural networks.
S13：Obtain part diagram picture to be detected.
In the present embodiment, when it is implemented, parts to be detected can be shot by CCD industrial cameras, with
Obtain above-mentioned part diagram picture to be detected.
It in the present embodiment, can be to acquired when it is implemented, after above-mentioned part diagram picture to be detected is obtained
Part diagram picture to be detected carry out image procossing, obtain satisfactory size be 512*512 part diagram to be detected
Picture.
S14：According to the part diagram picture to be detected and described for identifying the convolutional neural networks of surface defect, really
Fixed parts to be detected whether there is surface defect.
In the present embodiment, when it is implemented, can utilize established for identifying the convolutional Neural of surface defect
Network is analyzed above-mentioned part diagram picture to be detected, is handled, to determine above-mentioned part diagram to be detected as corresponding to
Parts whether there is surface defect.
In the present embodiment, it is necessary to explanation, due to be in above-mentioned implementation process using convolutional neural networks for
Part diagram picture to be detected specifically analyze, handled, therefore different from traditional detection method, will not be subject to illumination ring
The influence in border in time in the case where light environment is poor, still has higher accuracy.
In the embodiment of the present application, compared to the prior art, convolutional neural networks are taken full advantage of in image procossing and figure
As the advantage in terms of identification, establish to identify the convolutional neural networks of surface defect, then profit by using multiple sample images
It is to be detected to determine with above-mentioned for identifying that part diagram picture to be detected is identified in the convolutional neural networks of surface defect
Component surface defect, so as to solve the technology that can not identify component surface defect present in existing method exactly
Problem has reached the technique effect that component surface defect can be quickly and accurately identified in the poor environment of illumination.
In one embodiment, after multiple sample images are obtained, in order to obtain huger sample data,
To establish more accurately for identifying the convolutional neural networks of surface defect, when it is implemented, obtaining multiple sample drawings
As after, the above method can also specifically include herein below：Image procossing is carried out to the multiple sample image, after obtaining processing
Multiple sample images, wherein, described image processing include at least one of：Rotation processing, mirror image processing and gray scale are adjusted
Processing.
Correspondingly, it is above-mentioned according to the multiple sample image, it establishes to identify the convolutional neural networks of surface defect, have
Body can include：According to treated multiple sample images, establish described for identifying the convolutional Neural net of surface defect
Network.
In the present embodiment, it is necessary to which explanation, more sample images can be obtained by above-mentioned image procossing, with
Increase sample size.Specifically, for example, 500 new samples can be arrived to 500 sample images progress rotation processing
Image.Again original 500 sample images are carried out with mirror image processing, but in addition can be 500 sample images.Then to round face
500 sample images carry out gray scale adjusting processing, then 500 additional sample images can be obtained.In this way, to above-mentioned 250
After sample image carries out rotation processing, mirror image processing and gray scale adjusting processing respectively, 2000 sample drawings may finally be obtained
Picture so as to effectively expand original sample space, is preferably used for relatively can subsequently to establish out recognition effect
Identify the convolutional neural networks of surface defect.
In one embodiment, in order to lower part diagram picture to be detected, part diagram to be detected is being obtained
As after, the method is when it is implemented, can also include herein below：Gauss filter is carried out to the part diagram picture to be detected
Ripple obtains the part diagram picture to be detected after denoising；Correspondingly, it according to the part diagram picture to be detected and described is used for
It identifies the convolutional neural networks of surface defect, determines that parts to be detected with the presence or absence of surface defect, can specifically include：Root
According to the part diagram picture to be detected after the denoising and the convolutional neural networks for being used to identify surface defect, determine to be checked
The parts of survey whether there is surface defect.
In this way, the Gaussian noise in part diagram picture to be detected can be removed more efficiently, to be detected zero is improved
The resolution ratio and accuracy of identification of image of component, subsequently can more accurately to identify that parts to be detected whether there is
Surface defect.
In one embodiment, it is that can identify zero exactly on the basis of component surface defect is accurately identified
The concrete type of parts surface defect, when specific implementation, can obtain the image of a variety of parts defect types as above-mentioned multiple
The sample image of defective parts, then lacked for specific a variety of parts in the sample image of above-mentioned defective parts
Sunken types of image is trained, learns, and the convolutional neural networks that can identify defect type have been established out.
In one embodiment, the sample image of above-mentioned multiple defective parts can specifically include a variety of parts
Defect type image.Wherein, above-mentioned a variety of parts defect type images specifically can include act set forth below in two kinds or
It is a variety of：Crack-type image, cut types of image, hole types of image, mottling types of image etc..Certainly, it is necessary to illustrate,
Above-mentioned cited a variety of parts defect images are intended merely to that the application embodiment is better described.When it is implemented,
Other zero in addition to above-mentioned cited parts defect type image can be introduced with construction requirement as the case may be
Part defect type image.In this regard, the application is not construed as limiting.
In one embodiment, it is not only able to identify component surface defect in order to establish, moreover it is possible to exactly
Determine the convolutional neural networks of the specific defect type of surface defect, described according to the multiple sample image, foundation is used for
Identify the convolutional neural networks of surface defect, when specific implementation can include herein below：
S1：According to the sample image of the multiple zero defect parts and the sample image of multiple defective parts, build
Vertical first training set and the first test set；
S2：It is trained using first training set and first test set, is known with establishing the parts defect
Other network；
S3：According to a variety of parts defect type images, the second training set and the second test set are established；
S4：It is trained using second training set and second test set, to establish defect type sorter network；
And using the parts defect recognition network and the defect type sorter network as the volume for being used to identify surface defect
Product neutral net.
In one embodiment, when it is implemented, multiple above-mentioned sample images specifically can include multiple defective zero
The sample image of the sample image of component and multiple zero defect parts.Wherein, the sample drawing of multiple above-mentioned defective parts
The number of picture is identical with the number of the sample image of zero defect parts or difference is less than threshold value.
In the present embodiment, when it is implemented, multiple above-mentioned sample images are 500 sample images, wherein, including
The sample image of 250 defective parts and the sample image of 250 zero defect parts.Further, it is above-mentioned defective
Parts sample image is specific can to include a variety of parts defect type images again, such as can include 50 crack-type figures
Picture, 50 cut types of image, 50 hole types of image, 50 mottling types of image, 50 other defect types of image.Again
Multiple above-mentioned sample images are carried out respectively after including the image procossing that rotation processing, mirror image processing and gray scale adjust processing, it can
To obtain the sample image of 1000 zero defect parts, 200 crack-type images, 200 cut types of image, 200
Open hole types of image, 200 mottling types of image, 200 other defect types of image, i.e. 1000 defective parts
Sample image.
In the present embodiment, in order to establish to obtain relatively good first training set of effect and the first test set, specifically
During implementation, 800 flawless sample images can be extracted from the sample image of 1000 zero defect parts, from 1000
800 defective sample images are extracted in the sample image of defective parts, by above-mentioned 800 flawless sample images
It is combined to obtain the first training set with 800 defective sample images.By remaining 200 defective sample images
It is combined to obtain the first test set with 200 flawless sample images.
In the present embodiment, in order to establish to obtain relatively good second training set of effect and the second test set, specifically
It, can be from 200 crack-type images, 200 cut types of image, 200 hole types of image, 200 mottlings during implementation
Types of image, 200 other defect types of image extract 160 crack-type images, 160 cut types of image, 160 respectively
It opens hole types of image, 160 mottling types of image, 160 other defect types of image to be combined, to establish the second training
Collection.By remaining 40 crack-type images, 40 cut types of image, 40 hole types of image, 40 mottling type maps
Picture, 40 other defect types of image are combined, and obtain the second test set.
In one embodiment, in order to identifying the specific defect of component surface defect to be detected exactly
Type, it is above-mentioned according to the part diagram picture to be detected and described for identifying the convolutional neural networks of surface defect, it determines
Parts to be detected whether there is surface defect, and when specific implementation can include herein below：
S1：The part diagram picture to be detected is identified using the parts defect recognition network, to determine
The parts to be detected whether there is surface defect；
S2：In the case where the definite parts to be detected are there are surface defect, classified using the defect type
Network determines the type of the component surface defect to be detected.
In one embodiment, in order to efficiently being identified from large quantities of parts and determine component surface
Defect, when it is implemented, can by using GPU to first with the parts defect recognition network to described to be detected zero
Image of component is identified, to determine that the parts to be detected whether there is surface defect；Again definite described to be detected
Parts there are surface defect in the case of, determine the parts table to be detected using the defect type sorter network
The type of planar defect.
In the present embodiment, above-mentioned GPU (Graphics Processing Unit, image processor) also known as shows core
The heart, vision processor, display chip or accelerator card are that one kind is specifically applied in brain, work station, the mobile equipment in part (such as
Tablet computer, smart mobile phone etc.) on carry out image operation work microprocessor.
In one embodiment, in order to further improve the rate of identification component surface defect, when it is implemented, can
Two panels GPU, that is, GPU1 and GPU2 are integrated in a server, when it is implemented, described zero can be utilized by GPU1
The part diagram picture to be detected is identified in part defect recognition network, to determine whether the parts to be detected are deposited
In surface defect；By GPU2 in the case where the definite parts to be detected are there are surface defect, the defect is utilized
Classification of type network determines the type of the component surface defect to be detected.Share out the work and help one another so as to reach, multithreading
Operation further improves the technique effect for the treatment of effeciency.
In one embodiment, above-mentioned GPU can be specifically Teslap40GPU, and the GPU processing speeds of the model are opposite
Higher, precision is relatively preferable.Certainly, when it is implemented, other models can also be selected with construction requirement as the case may be
GPU.
In one embodiment, the parts specifically can include but is not limited to the parts of act set forth below：Axis
It holds, screw, gear, pneumatic element, sealing element, fastener etc..Certainly, it is necessary to illustrate, when it is implemented, can also basis
Concrete condition and construction requirement select other parts in addition to above-mentioned cited parts type to carry out surface defect
Identification.
It can be seen from the above description that the recognition methods for the component surface defect that the application embodiment provides,
Due to taking full advantage of advantage of the convolutional neural networks in terms of image procossing and image identification, by using multiple sample images
Foundation recycles the above-mentioned convolutional neural networks pair for being used to identify surface defect for identifying the convolutional neural networks of surface defect
Part diagram picture to be detected is identified, to determine component surface defect to be detected, so as to solve in existing method
Existing the technical issues of can not identifying component surface defect exactly, having reached can be fast in the poor environment of illumination
Speed, the technique effect for identifying component surface defect exactly；It is lacked further through parts are established respectively using multiple sample images
Identification network and defect type sorter network are fallen into as the convolutional neural networks for being used to identify surface defect, in this way, can be with
Determine that parts with the presence or absence of surface defect, recycle defect type sorter network to determine using parts defect recognition network
Go out the concrete type of surface defect, reach the technique effect of type the defects of can accurately identifying component surface defect.
Based on same inventive concept, a kind of identification dress of component surface defect is additionally provided in embodiment of the present invention
It puts, as described in following embodiment.Due to the principle that the identification device of component surface defect solves the problems, such as and parts table
The recognition methods of planar defect is similar, therefore the specific implementation of the identification device of component surface defect may refer to component surface
The specific implementation of the recognition methods of defect, overlaps will not be repeated.Used below, term " unit " or " module " can
To realize the combination of the software of predetermined function and/or hardware.Although the described device of following embodiment is preferably come with software
It realizes, but the realization of the combination of hardware or software and hardware is also what may and be contemplated.Referring to Fig. 2, it is the application
A kind of composition structure chart of the identification device of the component surface defect of embodiment, the device can include：First obtains mould
Block 21 establishes module 22, the second acquisition module 23, determining module 24, and the structure is specifically described below.
First acquisition module 21 specifically can be used for obtaining multiple sample images, wherein, the multiple sample image bag
It includes：The sample image of the sample image of multiple zero defect parts and multiple defective parts.
Module 22 is established, specifically can be used for according to the multiple sample image, establishing the volume for identifying surface defect
Product neutral net.
Second acquisition module 23 specifically can be used for obtaining part diagram picture to be detected.
Determining module 24 specifically can be used for according to the part diagram picture to be detected and described for identifying that surface lacks
Sunken convolutional neural networks determine that parts to be detected whether there is surface defect.
In one embodiment, the parts can specifically include at least one of：Bearing, screw, gear, gas
Dynamic element, sealing element, fastener etc..Certainly, it is necessary to which explanation, above-mentioned cited a variety of parts types are intended merely to more
Illustrate the application embodiment well, when it is implemented, can also be introduced with construction requirement as the case may be except above-mentioned listed
Other kinds of parts carry out the identification of corresponding surface defect beyond the parts of act.
In one embodiment, in order to according to the multiple sample image, establishing to identify surface defect
Convolutional neural networks, above-mentioned module 22 of establishing can specifically include following structural unit：
First establishes unit, specifically can be used for according to the sample images of the multiple zero defect parts and it is multiple have it is scarce
The sample image of parts is fallen into, establishes the first training set and the first test set；
Second establishes unit, specifically can be used for being trained using first training set and first test set,
To establish parts defect recognition network, and it is used to identify surface defect using the parts defect recognition network as described
Convolutional neural networks.
In one embodiment, the sample image of the multiple defective parts can specifically include a variety of parts
Defect type image.
In one embodiment, a variety of parts defect type images can specifically include：Crack-type image,
Cut types of image, hole types of image, mottling types of image etc..Certainly, it is necessary to explanation, above-mentioned cited parts
Defect type image is intended merely to that the application embodiment is better described, when it is implemented, can also as the case may be and
Construction requirement introduces the other kinds of defect type image in addition to types of image except above-mentioned cited the defects of.
In one embodiment, a variety of parts defect classes are included in the sample image of the multiple defective parts
In the case of type image, in order to which according to the multiple sample image, it is more accurate to establish, and can identify specific surface defect
The convolutional neural networks for being used to identify surface defect of type, above-mentioned module 22 of establishing can specifically include following structural unit：
First establishes unit, specifically can be used for according to the sample images of the multiple zero defect parts and it is multiple have it is scarce
The sample image of parts is fallen into, establishes the first training set and the first test set；
Second establishes unit, specifically can be used for being trained using first training set and first test set,
To establish the parts defect recognition network；
3rd establishes unit, specifically can be used for, according to a variety of parts defect type images, establishing the second training
Collection and the second test set；
4th establishes unit, specifically can be used for being trained using second training set and second test set,
To establish defect type sorter network；And using the parts defect recognition network and the defect type sorter network as institute
State to identify the convolutional neural networks of surface defect.
In one embodiment, in order to according to the part diagram picture to be detected and described for identifying surface
The convolutional neural networks of defect determine parts to be detected with the presence or absence of surface defect, and above-mentioned determining module 24 specifically can be with
Including following structural unit：
Recognition unit specifically can be used for using the parts defect recognition network to the part diagram to be detected
As being identified, to determine that the parts to be detected whether there is surface defect；
Determination unit specifically can be used in the case where the definite parts to be detected are there are surface defect, profit
The type of the component surface defect to be detected is determined with the defect type sorter network.
In one embodiment, above device can also specifically include image processing module, specifically can be used for obtaining
After taking multiple sample images, image procossing is carried out to the multiple sample image, multiple sample images that obtain that treated,
In, described image processing includes at least one of：Rotation processing, mirror image processing and gray scale adjusting processing.Certainly, it is necessary to illustrate
, above-mentioned cited a variety of image procossing modes are intended merely to that the application embodiment is better described, when it is implemented,
Can also the other kinds of image in addition to above-mentioned cited image procossing mode be introduced with construction requirement as the case may be
Processing.In this regard, the application is not construed as limiting.
Correspondingly, above-mentioned establish can be used for when module 22 is embodied according to treated multiple sample images,
Establish the convolutional neural networks for being used to identify surface defect.
In one embodiment, above device can also specifically include denoising module, specifically can be used for treating in acquisition
After the part diagram picture of detection, gaussian filtering is carried out to the part diagram picture to be detected, is obtained to be detected after denoising
Part diagram picture.
Correspondingly, above-mentioned determining module 24 is when it is implemented, can be according to the part diagram to be detected after the denoising
Picture and the convolutional neural networks for being used to identify surface defect, determine that parts to be detected whether there is surface defect.
Each embodiment in this specification is described by the way of progressive, identical similar portion between each embodiment
Point just to refer each other, and the highlights of each of the examples are difference from other examples.It is real especially for system
For applying example, since it is substantially similar to embodiment of the method, so description is fairly simple, related part is referring to embodiment of the method
Part explanation.
It should be noted that system, device, module or unit that the above embodiment illustrates, it specifically can be by computer
Chip or entity are realized or realized by having the function of certain product.For convenience of description, in the present specification, retouch
It is divided into various units when stating apparatus above with function to describe respectively.It certainly, when implementing the application can be the function of each unit
It realizes in the same or multiple software and or hardware.
In addition, in the present specification, adjective can be only used for an element or dynamic such as first and second
Make to distinguish with another element or action, without requiring or implying any actual this relation or order.Permit in environment
Perhaps in the case of, one in only element, component or step is should not be interpreted as limited to reference to element or component or step (s)
It is a, and can be one or more of element, component or step etc..
It can be seen from the above description that the identification device for the component surface defect that the application embodiment provides,
Due to taking full advantage of advantage of the convolutional neural networks in terms of image procossing and image identification, by establishing module using multiple
Sample image establishes to identify the convolutional neural networks of surface defect, then utilize by determining module and above-mentioned be used to identify surface
Part diagram picture to be detected is identified in the convolutional neural networks of defect, to determine component surface defect to be detected,
So as to solving the technical issues of can not identifying component surface defect present in existing method exactly, having reached can be
The technique effect of component surface defect is quickly and accurately identified in the poor environment of illumination；It is again sharp especially by module is established
Parts defect recognition network and defect type sorter network are established respectively by the use of multiple sample images to be used to identify table as described
The convolutional neural networks of planar defect, in this way, can determine that parts whether there is surface using parts defect recognition network
Defect recycles defect type sorter network to determine the concrete type of surface defect, and can accurately be identified by reaching by zero
The defects of part surface defect type technique effect.
The application can specifically refer to shown in Fig. 3 real based on the application embodiment further provides a kind of electronic equipment
The electronic equipment composition structure chart of the recognition methods for the component surface defect that the mode of applying provides, the electronic equipment specifically can be with
Including input equipment 31, processor 32, memory 33.Wherein, the input equipment 31 specifically can be used for inputting multiple samples
Image, wherein, the multiple sample image includes：The sample image of multiple zero defect parts and multiple defective parts
Sample image and part diagram picture to be detected.The processor 32 specifically can be used for according to the multiple sample image,
It establishes to identify the convolutional neural networks of surface defect；And according to the part diagram picture to be detected and described for identifying
The convolutional neural networks of surface defect determine that parts to be detected whether there is surface defect.The memory 33 specifically may be used
For storing the convolutional neural networks for being used to identify surface defect established by processor 32.
In the present embodiment, the input equipment can be specifically that information exchange is carried out between user and computer system
One of main device.The input equipment can include keyboard, mouse, camera, scanner, light pen, writing input board, language
Sound input unit etc.；Input equipment is used to initial data be input to the programs for handling these numbers in computer.The input
Equipment, which can also obtain, receives the data that other modules, unit, equipment transmit.The processor can be by any appropriate
Mode is realized.For example, processor can take such as microprocessor or processor and storage that can be performed by (micro-) processor
Computer readable program code (such as software or firmware) computer-readable medium, logic gate, switch, application-specific integrated circuit
(Application Specific Integrated Circuit, ASIC), programmable logic controller (PLC) and embedded microcontroller
Form etc..The storage implement body can be used to protect stored memory device in modern information technologies.The storage
Device includes many levels, in digital display circuit, as long as can preserve binary data can be memory；In integrated circuit
In, a circuit with store function without physical form is also memory, such as RAM, FIFO；In systems, have
The storage device of physical form is also memory, such as memory bar, TF card.
In the present embodiment, the function and effect of electronic equipment specific implementation, can compare with other embodiment
It explains, details are not described herein.
A kind of computer storage media is additionally provided in this theory application embodiment, the computer storage media is stored with
Computer program instructions are performed realization in the computer program instructions：Multiple sample images are obtained, wherein, it is described more
A sample image includes：The sample image of the sample image of multiple zero defect parts and multiple defective parts；According to institute
Multiple sample images are stated, establish to identify the convolutional neural networks of surface defect；Obtain part diagram picture to be detected；According to
The part diagram picture to be detected and the convolutional neural networks for being used to identify surface defect, determine parts to be detected
With the presence or absence of surface defect.
In the present embodiment, above-mentioned storage medium includes but not limited to random access memory (Random Access
Memory, RAM), read-only memory (Read-Only Memory, ROM), caching (Cache), hard disk (Hard Disk
Drive, HDD) or storage card (Memory Card).The memory can be used for storing computer program instructions.Network leads to
Believe that unit can be according to standard setting as defined in communication protocol, for carrying out the interface of network connection communication.
In the present embodiment, the function and effect of the program instruction specific implementation of computer storage media storage, can
To compare explanation with other embodiment, details are not described herein.
It is embodied at one in Sample Scenario, recognition methods and dress using the component surface defect of the application offer
Put the surface defect of certain batch of bearing to be detected is identified it is definite.Specific implementation process can refer to herein below and perform.
Referred to during specific implementation it is shown in Fig. 4 in a Sample Scenario apply parts provided by the embodiments of the present application
The recognition methods of surface defect and device carry out certain batch of bearing to be detected the flow diagram of Surface Defect Recognition.
S1：Implement to prepare, including：The training data and its parameter of neutral net are obtained, and establishes bearing surface identification net
Network (i.e. parts defect recognition network) and defect sorter network (i.e. defect type sorter network).When it is implemented, it can wrap
Include herein below：
S1-1：It is the acquisition of image (i.e. sample image) first, CCD industrial camera photographic subjects samples specifically may be employed
The image of product, and the size of image is uniformly converted to the size of common 512x512.
In the present embodiment, by above-mentioned image acquisition, 250 images of zero defect bearing can specifically be obtained (i.e.
The sample image of zero defect parts), 250 images (images of i.e. defective parts) of defective bearing, wherein, defect
Bearing image specifically includes that crackle 50 opens (i.e. crack-type image), cut 50 opens (i.e. cut types of image), hole 50 is opened
(i.e. hole types of image), mottling 50 open (i.e. mottling types of image) and other species defects 50 open (i.e. other defect type map
Picture).
S1-2：The image data collection (i.e. enlarged sample space) of acquisition by using the method for image procossing, can be expanded.
When it is implemented, can be rotated respectively, the adjusting of mirror image and gray scale, the expansion of zero defect bearing image is made to be 1000, is defective
Bearing image expands as 1000.
S1-3：Design two convolutional neural networks (establishing to identify the convolutional neural networks of surface defect)：Bearing
Surface identifies network (i.e. parts defect recognition network) and defect sorter network (i.e. defect type sorter network) to carry out axis
Hold the identification of surface defect and determining for specific defect type.
S1-3-1：When it is implemented, zero defect bearing 800 can be used to open opens image as instruction with defective bearing 800
Practice collection (i.e. the first training set), other 400, as test set (i.e. the first test set), bearing surface are obtained by training study
Identify network；
S1-3-2：Contain that crackle 160 is opened, cut 160 is opened, hole 160 is opened, mottling 160 is opened and other species using bearing
Defect 160 opens image as training set (i.e. the second training set), other to take 200 as test sets (i.e. the second test set), logical
Cross the defects of training study obtains including species defect type sorter network；
S2：Identify component surface defect to be detected, including：Identify that network (i.e. know by parts defect by bearing surface
Other network) determine bearing to be detected with the presence or absence of surface defect and by defect sorter network (i.e. defect type sorter network)
Determine the specific defect type of surface defect.
S2-1：It is the acquisition of image (images of parts to be detected) first, CCD industrial cameras specifically may be employed
Photographic subjects (parts i.e. to be detected) image, and the size of the image is uniformly converted to the size of common 512x512.
S2-2：To improve recognition accuracy of the follow-up neutral net to input picture, when it is implemented, Gauss may be employed
Filtering method is filtered image, to remove Gaussian noise, obtains filtered image.
S2-3：In order to meet in industrial production scene, the requirement to real-time, when specific implementation, can be by 2 pieces
Teslap40GPU, which accelerates to calculate to block, to be integrated in the server, correspondingly, can calculate card GPU1 by acceleration uses bearing table
Filtered image is identified in face identification network, and whether it is defective bearing as a result, to determine to treat that this network can provide
The bearing of detection whether there is surface defect；In the case where the result of output is defective bearing, further pass through accelerometer
Calculate card GPU2 using off-line training it is good the defects of sorter network, classify to image, this network can provide belong to which kind of or
The probability of multiclass, and then the specific defect type of surface defect can be accurately determined out.
By above-mentioned Sample Scenario, demonstrate the component surface defect of the application embodiment offer recognition methods and
Device is established to identify the convolutional neural networks of surface defect by using multiple sample images, and recycling is above-mentioned to be used to know
Part diagram picture to be detected is identified in the convolutional neural networks of other surface defect, to determine component surface to be detected
Defect solves the technical issues of can not identifying component surface defect present in existing method exactly, reaches really really
To the technique effect that component surface defect can be quickly and accurately identified in the poor environment of illumination.
Although mentioning different specific embodiments in teachings herein, the application is not limited to be capable
Industry standard or the described situation of embodiment etc., some professional standards or the implementation described using self-defined mode or embodiment
On the basis of embodiment amended slightly can also realize above-described embodiment it is identical, it is equivalent or it is close or deformation after it is anticipated that
Implementation result.Using the embodiment of these modifications or deformed data acquisition, processing, output, judgment mode etc., still may be used
To belong within the scope of the optional embodiment of the application.
It is also known in the art that in addition to realizing controller in a manner of pure computer readable program code, it is complete
Entirely can by by method and step carry out programming in logic come controller with logic gate, switch, application-specific integrated circuit, may be programmed
The form of logic controller and embedded microcontroller etc. realizes identical function.Therefore this controller is considered one kind
Hardware component, and the structure that can also be considered as to the device for being used to implement various functions that its inside includes in hardware component.Or
The device for being used to implement various functions even, can be considered as either the software module of implementation method can be hardware again by person
Structure in component.
The application can be described in the general context of computer executable instructions, such as program
Module.Usually, program module includes routines performing specific tasks or implementing specific abstract data types, program, object, group
Part, data structure, class etc..The application can also be put into practice in a distributed computing environment, in these distributed computing environment,
By performing task by communication network and connected remote processing devices.In a distributed computing environment, program module can
To be located in the local and remote computer storage media including storage device.
Each embodiment in this specification is described by the way of progressive, the same or similar portion between each embodiment
Point just to refer each other, and the highlights of each of the examples are difference from other examples.The application can be used for crowd
In mostly general or special purpose computing system environments or configuration.Such as：Personal computer, server computer, handheld device or
Portable device, laptop device, multicomputer system, the system based on microprocessor, set top box, programmable electronics are set
Standby, network PC, minicomputer, mainframe computer, distributed computing environment including any of the above system or equipment etc..
Although depicting the application by embodiment, it will be appreciated by the skilled addressee that the application there are many deformation and
Variation is without departing from spirit herein, it is desirable to which appended embodiment includes these deformations and changes without departing from the application.
Claims (10)
1. a kind of recognition methods of component surface defect, which is characterized in that including：
Multiple sample images are obtained, wherein, the multiple sample image includes：The sample image of multiple zero defect parts and more
The sample image of a defective parts；
According to the multiple sample image, establish to identify the convolutional neural networks of surface defect；
Obtain part diagram picture to be detected；
According to the part diagram picture to be detected and the convolutional neural networks for being used to identify surface defect, determine to be detected
Parts whether there is surface defect.
2. according to the method described in claim 1, it is characterized in that, the parts include at least one of：Bearing, spiral shell
Nail, gear, pneumatic element, sealing element, fastener.
3. according to the method described in claim 1, it is characterized in that, according to the multiple sample image, establish to identify table
The convolutional neural networks of planar defect, including：
According to the sample image of the multiple zero defect parts and the sample image of multiple defective parts, the first instruction is established
Practice collection and the first test set；
It is trained using first training set and first test set, to establish parts defect recognition network, and will
The parts defect recognition network is as the convolutional neural networks for being used to identify surface defect.
4. according to the method described in claim 1, it is characterized in that, the sample image of the multiple defective parts is including more
Kind parts defect type image.
5. according to the method described in claim 4, it is characterized in that, a variety of parts defect type images include：Crackle
Types of image, cut types of image, hole types of image, mottling types of image.
6. according to the method described in claim 5, it is characterized in that, described according to the multiple sample image, establish to know
The convolutional neural networks of other surface defect, including：
According to the sample image of the multiple zero defect parts and the sample image of multiple defective parts, the first instruction is established
Practice collection and the first test set；
It is trained using first training set and first test set, to establish the parts defect recognition network；
According to a variety of parts defect type images, the second training set and the second test set are established；
It is trained using second training set and second test set, to establish defect type sorter network；
The parts defect recognition network and the defect type sorter network are used to identify surface defect as described
Convolutional neural networks.
7. it according to the method described in claim 6, it is characterized in that, according to the part diagram picture to be detected and described is used for
It identifies the convolutional neural networks of surface defect, determines that parts to be detected whether there is surface defect, including：
The part diagram picture to be detected is identified using the parts defect recognition network, it is described to be checked to determine
The parts of survey whether there is surface defect；
In the case where the definite parts to be detected are there are surface defect, determined using the defect type sorter network
The type of the component surface defect to be detected.
8. according to the method described in claim 1, it is characterized in that, after multiple sample images are obtained, the method further includes：
Image procossing is carried out to the multiple sample image, multiple sample images that obtain that treated, wherein, described image processing
Including at least one of：Rotation processing, mirror image processing and gray scale adjusting processing；
Correspondingly, according to the multiple sample image, establish to identify the convolutional neural networks of surface defect, including：
According to treated multiple sample images, establish described for identifying the convolutional neural networks of surface defect.
9. according to the method described in claim 1, it is characterized in that, after part diagram picture to be detected is obtained, the method
It further includes：
Gaussian filtering is carried out to the part diagram picture to be detected, obtains the part diagram picture to be detected after denoising；
Correspondingly, according to the part diagram picture to be detected and described for identifying the convolutional neural networks of surface defect, really
Fixed parts to be detected whether there is surface defect, including：
According to the part diagram picture to be detected after the denoising and described for identifying the convolutional neural networks of surface defect, really
Fixed parts to be detected whether there is surface defect.
10. a kind of identification device of component surface defect, which is characterized in that including：
First acquisition module, for obtaining multiple sample images, wherein, the multiple sample image includes：Multiple zero defects zero
The sample image of the sample image of component and multiple defective parts；
Module is established, for according to the multiple sample image, establishing to identify the convolutional neural networks of surface defect；
Second acquisition module, for obtaining part diagram picture to be detected；
Determining module, for according to the part diagram picture to be detected and the convolutional Neural net for being used to identify surface defect
Network determines that parts to be detected whether there is surface defect.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711246947.0A CN108061735A (en) | 2017-12-01 | 2017-12-01 | The recognition methods of component surface defect and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711246947.0A CN108061735A (en) | 2017-12-01 | 2017-12-01 | The recognition methods of component surface defect and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108061735A true CN108061735A (en) | 2018-05-22 |
Family
ID=62135227
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711246947.0A Pending CN108061735A (en) | 2017-12-01 | 2017-12-01 | The recognition methods of component surface defect and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108061735A (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108776808A (en) * | 2018-05-25 | 2018-11-09 | 北京百度网讯科技有限公司 | A kind of method and apparatus for detecting ladle corrosion defect |
| CN108960413A (en) * | 2018-07-11 | 2018-12-07 | 天津工业大学 | A kind of depth convolutional neural networks method applied to screw surface defects detection |
| CN109239075A (en) * | 2018-08-27 | 2019-01-18 | 北京百度网讯科技有限公司 | Battery detection method and device |
| CN109444146A (en) * | 2018-09-17 | 2019-03-08 | 鲁班嫡系机器人（深圳）有限公司 | A kind of defect inspection method, device and the equipment of industrial processes product |
| CN109598721A (en) * | 2018-12-10 | 2019-04-09 | 广州市易鸿智能装备有限公司 | Defect inspection method, device, detection device and the storage medium of battery pole piece |
| CN109767431A (en) * | 2018-12-29 | 2019-05-17 | 北京领邦智能装备股份公司 | Accessory appearance defect inspection method, device, equipment and readable storage medium storing program for executing |
| CN109829903A (en) * | 2019-01-28 | 2019-05-31 | 合肥工业大学 | A kind of chip surface defect inspection method based on convolution denoising self-encoding encoder |
| CN109919908A (en) * | 2019-01-23 | 2019-06-21 | 华灿光电（浙江）有限公司 | The method and apparatus of light-emitting diode chip for backlight unit defects detection |
| CN109949271A (en) * | 2019-02-14 | 2019-06-28 | 腾讯科技（深圳）有限公司 | A kind of detection method based on medical image, the method and device of model training |
| CN110148106A (en) * | 2019-01-18 | 2019-08-20 | 华晨宝马汽车有限公司 | A kind of system and method using deep learning model inspection body surface defect |
| CN110930350A (en) * | 2018-09-19 | 2020-03-27 | 鸿富锦精密电子(天津)有限公司 | Machine learning method and automatic optical detection equipment applying same |
| WO2021046726A1 (en) * | 2019-09-10 | 2021-03-18 | 西门子能源国际公司 | Method and device for detecting mechanical equipment parts |
| CN112555285A (en) * | 2020-12-01 | 2021-03-26 | 徐州东宏机械制造有限公司 | Bearing wear warning bearing sleeve |
| WO2021147386A1 (en) * | 2020-01-21 | 2021-07-29 | 上海万物新生环保科技集团有限公司 | Screen scratch and crack detection method and device |
| US20210339758A1 (en) * | 2018-08-30 | 2021-11-04 | Camso Inc. | Systems and methods for monitoring tracked vehicles |
| CN115797340A (en) * | 2023-02-03 | 2023-03-14 | 西南石油大学 | Industrial surface defect detection method based on multi-instance learning |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4467214A (en) * | 1980-06-13 | 1984-08-21 | Kabushiki Kaisha Toyota Chuo Kenkyusho | Method of and apparatus for detecting external defects of a circular sealing member |
| CN102854191A (en) * | 2012-07-18 | 2013-01-02 | 湖南大学 | Real-time visual detection and identification method for high speed rail surface defect |
| CN105938563A (en) * | 2016-04-14 | 2016-09-14 | 北京工业大学 | Weld surface defect identification method based on image texture |
| CN106556781A (en) * | 2016-11-10 | 2017-04-05 | 华乘电气科技（上海）股份有限公司 | Shelf depreciation defect image diagnostic method and system based on deep learning |
| CN106568783A (en) * | 2016-11-08 | 2017-04-19 | 广东工业大学 | Hardware part defect detecting system and method |
| CN106952250A (en) * | 2017-02-28 | 2017-07-14 | 北京科技大学 | A kind of metal plate and belt detection method of surface flaw and device based on Faster R CNN networks |
| CN107292298A (en) * | 2017-08-09 | 2017-10-24 | 北方民族大学 | Ox face recognition method based on convolutional neural networks and sorter model |
| CN107328787A (en) * | 2017-07-05 | 2017-11-07 | 北京科技大学 | A kind of metal plate and belt surface defects detection system based on depth convolutional neural networks |
| CN107369155A (en) * | 2017-07-24 | 2017-11-21 | 广东工业大学 | A kind of cloth surface defect detection method and its system based on machine vision |
-
2017
- 2017-12-01 CN CN201711246947.0A patent/CN108061735A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4467214A (en) * | 1980-06-13 | 1984-08-21 | Kabushiki Kaisha Toyota Chuo Kenkyusho | Method of and apparatus for detecting external defects of a circular sealing member |
| CN102854191A (en) * | 2012-07-18 | 2013-01-02 | 湖南大学 | Real-time visual detection and identification method for high speed rail surface defect |
| CN105938563A (en) * | 2016-04-14 | 2016-09-14 | 北京工业大学 | Weld surface defect identification method based on image texture |
| CN106568783A (en) * | 2016-11-08 | 2017-04-19 | 广东工业大学 | Hardware part defect detecting system and method |
| CN106556781A (en) * | 2016-11-10 | 2017-04-05 | 华乘电气科技（上海）股份有限公司 | Shelf depreciation defect image diagnostic method and system based on deep learning |
| CN106952250A (en) * | 2017-02-28 | 2017-07-14 | 北京科技大学 | A kind of metal plate and belt detection method of surface flaw and device based on Faster R CNN networks |
| CN107328787A (en) * | 2017-07-05 | 2017-11-07 | 北京科技大学 | A kind of metal plate and belt surface defects detection system based on depth convolutional neural networks |
| CN107369155A (en) * | 2017-07-24 | 2017-11-21 | 广东工业大学 | A kind of cloth surface defect detection method and its system based on machine vision |
| CN107292298A (en) * | 2017-08-09 | 2017-10-24 | 北方民族大学 | Ox face recognition method based on convolutional neural networks and sorter model |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11276158B2 (en) | 2018-05-25 | 2022-03-15 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Method and apparatus for inspecting corrosion defect of ladle |
| CN108776808A (en) * | 2018-05-25 | 2018-11-09 | 北京百度网讯科技有限公司 | A kind of method and apparatus for detecting ladle corrosion defect |
| CN108960413A (en) * | 2018-07-11 | 2018-12-07 | 天津工业大学 | A kind of depth convolutional neural networks method applied to screw surface defects detection |
| CN109239075A (en) * | 2018-08-27 | 2019-01-18 | 北京百度网讯科技有限公司 | Battery detection method and device |
| US20210339758A1 (en) * | 2018-08-30 | 2021-11-04 | Camso Inc. | Systems and methods for monitoring tracked vehicles |
| CN109444146A (en) * | 2018-09-17 | 2019-03-08 | 鲁班嫡系机器人（深圳）有限公司 | A kind of defect inspection method, device and the equipment of industrial processes product |
| CN110930350A (en) * | 2018-09-19 | 2020-03-27 | 鸿富锦精密电子(天津)有限公司 | Machine learning method and automatic optical detection equipment applying same |
| CN109598721B (en) * | 2018-12-10 | 2021-08-31 | 广州市易鸿智能装备有限公司 | Defect detection method and device for battery pole piece, detection equipment and storage medium |
| CN109598721A (en) * | 2018-12-10 | 2019-04-09 | 广州市易鸿智能装备有限公司 | Defect inspection method, device, detection device and the storage medium of battery pole piece |
| CN109767431A (en) * | 2018-12-29 | 2019-05-17 | 北京领邦智能装备股份公司 | Accessory appearance defect inspection method, device, equipment and readable storage medium storing program for executing |
| CN110148106A (en) * | 2019-01-18 | 2019-08-20 | 华晨宝马汽车有限公司 | A kind of system and method using deep learning model inspection body surface defect |
| CN110148106B (en) * | 2019-01-18 | 2023-06-27 | 华晨宝马汽车有限公司 | System and method for detecting object surface defects by using deep learning model |
| CN109919908A (en) * | 2019-01-23 | 2019-06-21 | 华灿光电（浙江）有限公司 | The method and apparatus of light-emitting diode chip for backlight unit defects detection |
| CN109829903A (en) * | 2019-01-28 | 2019-05-31 | 合肥工业大学 | A kind of chip surface defect inspection method based on convolution denoising self-encoding encoder |
| CN109949271A (en) * | 2019-02-14 | 2019-06-28 | 腾讯科技（深圳）有限公司 | A kind of detection method based on medical image, the method and device of model training |
| CN109949271B (en) * | 2019-02-14 | 2021-03-16 | 腾讯科技（深圳）有限公司 | Detection method based on medical image, model training method and device |
| WO2021046726A1 (en) * | 2019-09-10 | 2021-03-18 | 西门子能源国际公司 | Method and device for detecting mechanical equipment parts |
| WO2021147386A1 (en) * | 2020-01-21 | 2021-07-29 | 上海万物新生环保科技集团有限公司 | Screen scratch and crack detection method and device |
| CN112555285A (en) * | 2020-12-01 | 2021-03-26 | 徐州东宏机械制造有限公司 | Bearing wear warning bearing sleeve |
| CN115797340A (en) * | 2023-02-03 | 2023-03-14 | 西南石油大学 | Industrial surface defect detection method based on multi-instance learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108061735A (en) | The recognition methods of component surface defect and device | |
| US20200210702A1 (en) | Apparatus and method for image processing to calculate likelihood of image of target object detected from input image | |
| CN109508688B (en) | Skeleton-based behavior detection method, terminal equipment and computer storage medium | |
| Zhao et al. | A visual long-short-term memory based integrated CNN model for fabric defect image classification | |
| Deshpande et al. | One-shot recognition of manufacturing defects in steel surfaces | |
| Zhu et al. | Modified densenet for automatic fabric defect detection with edge computing for minimizing latency | |
| CN107808161B (en) | Underwater target identification method based on optical vision | |
| CN109558864A (en) | Face critical point detection method, apparatus and storage medium | |
| Yu et al. | Detecting gear surface defects using background-weakening method and convolutional neural network | |
| CN111445459A (en) | Image defect detection method and system based on depth twin network | |
| CN112700435B (en) | Wall defect detection method based on deep learning | |
| CN109816634B (en) | Detection method, model training method, device and equipment | |
| US20210390282A1 (en) | Training data increment method, electronic apparatus and computer-readable medium | |
| CN109584206A (en) | The synthetic method of the training sample of neural network in piece surface Defect Detection | |
| WO2024002187A1 (en) | Defect detection method, defect detection device, and storage medium | |
| Rai et al. | Extraction of Facial Features for Detection of Human Emotions under Noisy Condition | |
| Maiti et al. | Classification of materials in cylindrical workpieces using image processing and machine learning techniques | |
| WO2021036185A1 (en) | Certificate identification method and device | |
| KR102178238B1 (en) | Apparatus and method of defect classification using rotating kernel based on machine-learning | |
| CN114170589A (en) | Rock lithology identification method based on NAS, terminal equipment and storage medium | |
| Schak et al. | Gesture MNIST: a new free-hand gesture dataset | |
| WO2012092132A2 (en) | Determining the uniqueness of a model for machine vision | |
| Stryczek | Finite point sets in recognizing location and orientation of machine parts of complex shapes | |
| CN111402196A (en) | Bearing roller image generation method based on countermeasure generation network | |
| Rane et al. | A Deep Learning-Based Regression Scheme for Angle Estimation in Image Dataset |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180522 |
|
| RJ01 | Rejection of invention patent application after publication |