CN107292885A - A kind of product defects classifying identification method and device based on autocoder - Google Patents
A kind of product defects classifying identification method and device based on autocoder Download PDFInfo
- Publication number
- CN107292885A CN107292885A CN201710671719.1A CN201710671719A CN107292885A CN 107292885 A CN107292885 A CN 107292885A CN 201710671719 A CN201710671719 A CN 201710671719A CN 107292885 A CN107292885 A CN 107292885A
- Authority
- CN
- China
- Prior art keywords
- defect
- image
- feature
- mode image
- sparse
- 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
- 230000007547 defect Effects 0.000 title claims abstract description 327
- 238000000034 method Methods 0.000 title claims abstract description 37
- 238000000605 extraction Methods 0.000 claims abstract description 30
- 230000009467 reduction Effects 0.000 claims description 90
- 230000005540 biological transmission Effects 0.000 claims description 28
- 108010001267 Protein Subunits Proteins 0.000 claims description 15
- 239000000284 extract Substances 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 abstract description 5
- 230000011218 segmentation Effects 0.000 abstract description 3
- 238000012549 training Methods 0.000 abstract description 3
- 230000000694 effects Effects 0.000 abstract description 2
- 238000011946 reduction process Methods 0.000 abstract 1
- 238000012545 processing Methods 0.000 description 16
- 230000008569 process Effects 0.000 description 10
- 210000004205 output neuron Anatomy 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 239000000463 material Substances 0.000 description 3
- 210000002569 neuron Anatomy 0.000 description 3
- 239000004576 sand Substances 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000005520 cutting process Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 210000002364 input neuron Anatomy 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 239000002994 raw material Substances 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000006641 stabilisation Effects 0.000 description 1
- 238000011105 stabilization Methods 0.000 description 1
- 238000012360 testing method Methods 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
- G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
Abstract
The invention discloses a kind of product defects classifying identification method based on autocoder, including:Obtain defect image;The defect characteristic of the defect image is extracted by stacking at least four sparse self-encoding encoders;The corresponding defect classification of defect characteristic is recognized by grader.It can be seen that, in this programme, depth network is formd by the stacking for stacking at least four sparse self-encoding encoders, noise reduction process is carried out to sparse self-encoding encoder simultaneously, so that the defect characteristic extracted is more abstract, the defect characteristic for improving the discrimination to product defects and extracting has good robustness.Meanwhile, sparse self-encoding encoder is using the successively unsupervised training of greed, it is not necessary to substantial amounts of sample, meanwhile, the step such as it can be pre-processed with alternate image, defect Segmentation, extraction defect characteristic vector reduces the fussy degree of image procossing, reduces amount of calculation;The invention also discloses a kind of product defects Classification and Identification device based on autocoder, above-mentioned technique effect can be equally realized.
Description
Technical field
The present invention relates to field of machine vision, more particularly to a kind of product defects Classification and Identification side based on autocoder
Method and device.
Background technology
Machine vision is also known as computer vision, is that research imitates human eye and brain respectively using camera and computer, with
Just replace people to detect and judge with machine, complete the science of the task such as target identification and industrial detection.In recent years, with nerve
Network and the theoretical constantly improve of deep learning and its extensive use in field of machine vision, have promoted computer vision
It is fast-developing.
Generally, the quality of defect mainly has the requirement in terms of size, outward appearance;In process of manufacture, by
In the change of raw material physical parameter, the factor such as technological parameter is unreasonable and processing equipment performance is bad, blemish surface occur damage,
The surface defect such as sand holes, scratch, scarce material, deformation, pit.These surface defects can not only destroy the outward appearance of defect, and can shadow
Its performance is rung to lead to not use.
Traditional defect inspection method needs to carry out numerous and diverse pretreatment process to the image of input, for example, image conversion,
Image restoration and reconstruction, compression of images, morphological images processing, image segmentation, image recognition etc., different images are generally required
Different Preprocessing Algorithms are, it is necessary to which repetition test just can determine which kind of processing scheme taken.And in feature extraction, need first
Accurately to determine threshold value, the higher feature of discrimination of segmentation.In addition, effectively choosing the higher feature of discrimination and it being entered
Row description is often relatively difficult, it is necessary to very professional knowledge and preferable priori.
With the development of artificial neural network theories and perfect, many algorithms have been applied to the field of image procossing.But,
Because traditional neural network structure hidden layer number is few, and the substantial amounts of sample data of needs is trained to it, it is difficult to portray
Complex relationship between network inputs and output, causes that the feature of defect image can not be carried out the description of high abstraction, therefore,
Need to be combined with traditional image processing algorithm, cause complex disposal process and computationally intensive.
Therefore, computationally intensive, the problem of processing procedure is cumbersome during identification product defects feature how is solved, obtains abstract
The defect characteristic of defect image simultaneously improves discrimination to product defects, is the problem of those skilled in the art need solution.
The content of the invention
It is an object of the invention to provide a kind of product defects classifying identification method and device based on autocoder, with
Computationally intensive, the problem of processing procedure is cumbersome during identification product defects is solved, the defect characteristic of abstract defect image is obtained
And raising is to the discrimination of product defects.
To achieve the above object, the embodiments of the invention provide following technical scheme:
A kind of product defects classifying identification method based on autocoder, including:
Obtain defect image;
The defect characteristic of the defect image is extracted by stacking at least four sparse self-encoding encoders;
The defect characteristic is sent to grader, the corresponding defect class of the defect characteristic is recognized by the grader
Not.
It is preferred that, it is described to extract the defect image defect characteristic by stacking at least four sparse self-encoding encoders, including:
The first mode image, second mode image and the 3rd mode image of the defect image are obtained, the defect image is colour
Image, the first mode image is the R passages of defect image, and second mode image is the G passages of defect image, the 3rd pattern
Image is the channel B of defect image;The first mode image, described the are extracted respectively by stacking six sparse self-encoding encoders
The defect characteristic of two modes image and the 3rd mode image.
It is preferred that, it is described to extract the first mode image, described second respectively by stacking six sparse self-encoding encoders
The defect characteristic of mode image and the 3rd mode image, including:Carried respectively by stacking six sparse noise reduction self-encoding encoders
Take the defect characteristic of the first mode image, the second mode image and the 3rd mode image.
It is preferred that, it is described by stack six sparse noise reduction self-encoding encoders extract the first mode image respectively, it is described
The defect characteristic of second mode image and the 3rd mode image, including:
The first mode image, the second mode image and institute are obtained respectively using the first sparse noise reduction self-encoding encoder
The pixel value of the 3rd mode image is stated, so that the first sparse noise reduction self-encoding encoder obtains the first spy to pixel value coding
Levy;
The fisrt feature is sent to the second sparse noise reduction self-encoding encoder, so that the second sparse noise reduction self-encoding encoder
Second feature is obtained to fisrt feature coding;
The second feature is sent to the 3rd sparse noise reduction self-encoding encoder, so that the 3rd sparse noise reduction self-encoding encoder
Third feature is obtained to second feature coding;
The third feature is sent to the 4th sparse noise reduction self-encoding encoder, so that the 4th sparse noise reduction self-encoding encoder
Fourth feature is obtained to third feature coding;
The fourth feature is sent to the 5th sparse noise reduction self-encoding encoder, so that the 5th sparse noise reduction self-encoding encoder
Fifth feature is obtained to fourth feature coding;
The fifth feature is sent to the 6th sparse noise reduction self-encoding encoder, so that the 6th sparse noise reduction self-encoding encoder
Sixth feature is obtained to fifth feature coding;It regard the sixth feature as the defect characteristic, the defect characteristic bag
Include first mode image deflects feature, second mode image deflects feature and the 3rd mode image defect characteristic.
It is preferred that, it is described to send the defect characteristic to grader, defect characteristic correspondence is recognized by the grader
Defect classification include:The defect characteristic is sent to the input layer of grader;The defect is calculated by the input layer
Feature corresponds to the probability of each defect type;The maximum probability in the probability is selected, the maximum probability is corresponding scarce
Type is fallen into as the defect type of the defect characteristic.
A kind of product defects Classification and Identification device based on autocoder, including:
Defect image acquisition module, for obtaining defect image;
Defect characteristic extraction module, for extracting the defect image acquisition by stacking at least four sparse self-encoding encoders
The defect characteristic for the defect image that module is obtained;
Defect characteristic sending module, the defect characteristic for defect characteristic extraction module to be obtained is sent to classification
Device, the corresponding defect classification of defect characteristic is recognized by the grader.
It is preferred that, the defect characteristic extraction module includes:
Acquiring unit, first mode image, second mode image and the 3rd ideograph for obtaining the defect image
Picture, the defect image is coloured image, and the first mode image is the R passages of defect image, and second mode image is scarce
The G passages of image are fallen into, the 3rd mode image is the channel B of defect image;
First extraction unit, for extracting the institute that the acquiring unit is obtained respectively by stacking six sparse self-encoding encoders
State the defect characteristic of first mode image, the second mode image and the 3rd mode image.
It is preferred that, first extraction unit includes:Second extraction unit, for self-editing by stacking six sparse noise reductions
Code device extracts the defect characteristic of the first mode image, the second mode image and the 3rd mode image respectively.
It is preferred that, second extraction unit includes:
Subelement is obtained, for obtaining the first mode image respectively using the first sparse noise reduction self-encoding encoder, described
The pixel value of second mode image and the 3rd mode image, so that the first sparse noise reduction self-encoding encoder is to the pixel
Value coding obtains fisrt feature;
First transmission sub-unit, the fisrt feature for acquisition subelement to be obtained is sent to the second sparse noise reduction certainly
Encoder, so that the second sparse noise reduction self-encoding encoder obtains second feature to fisrt feature coding;
Second transmission sub-unit, the second feature for first transmission sub-unit to be obtained is sent to the 3rd dilute
Noise reduction self-encoding encoder is dredged, so that the 3rd sparse noise reduction self-encoding encoder obtains third feature to second feature coding;
3rd transmission sub-unit, the third feature for second transmission sub-unit to be obtained is sent to the 4th sparse drop
Make an uproar self-encoding encoder, so that the 4th sparse noise reduction self-encoding encoder obtains fourth feature to third feature coding;
4th transmission sub-unit, the fourth feature for the 3rd transmission sub-unit to be obtained is sent to the 5th sparse drop
Make an uproar self-encoding encoder, so that the 5th sparse noise reduction self-encoding encoder obtains fifth feature to fourth feature coding;
5th transmission sub-unit, the fifth feature for the 4th transmission sub-unit to be obtained is sent to the 6th sparse drop
Make an uproar self-encoding encoder, so that the 6th sparse noise reduction self-encoding encoder obtains sixth feature to fifth feature coding;Will be described
Sixth feature is as the defect characteristic, and the defect characteristic includes first mode image deflects feature, second mode image and lacked
Fall into feature and the 3rd mode image defect characteristic.
It is preferred that, the defect characteristic sending module includes:
Transmitting element, for the defect characteristic to be sent to grader input layer;
Computing unit, for calculating the defect characteristic pair that the transmitting element is sent by the grader input layer
Answer the probability of each defect type;
Selecting unit, the maximum probability in the probability for selecting the computing unit calculating is most general by described in
The corresponding defect type of rate as the defect characteristic defect type.
By above scheme, a kind of product defects classification based on autocoder provided in an embodiment of the present invention is known
Other method, including:Obtain defect image;The defect spy that the defect image is extracted by stacking at least four sparse self-encoding encoders
Levy;The defect characteristic is sent to grader, the corresponding defect classification of the defect characteristic is recognized by the grader.
It can be seen that, in this programme, by inputting defect image, defect is extracted using at least four sparse self-encoding encoders are stacked
Defect characteristic in image, the defect characteristic is inputted to grader again and carries out defect Classification and Identification;Wherein, by being stacked to
The stacking of few four sparse self-encoding encoders adds network depth so that the defect characteristic extracted is more abstract, obtains more
Plus it is adapted to the feature of the defect image of division, improve the discrimination to defect image.Sparse self-encoding encoder uses greed successively
Unsupervised training, it is not necessary to substantial amounts of sample, meanwhile, it can be pre-processed with alternate image, defect Segmentation, extract defect characteristic to
The intermediate steps such as amount, reduce the fussy degree of image procossing, reduce amount of calculation;The invention also discloses one kind based on automatic
The product defects Classification and Identification device of encoder, can equally realize above-mentioned technique effect.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
There is the accompanying drawing used required in technology description to be briefly described, it should be apparent that, drawings in the following description are only this
Some embodiments of invention, for those of ordinary skill in the art, on the premise of not paying creative work, can be with
Other accompanying drawings are obtained according to these accompanying drawings.
Fig. 1 is that a kind of product defects classifying identification method flow based on autocoder disclosed in the embodiment of the present invention is shown
It is intended to;
Fig. 2 is a kind of autocoder structural representation disclosed in the embodiment of the present invention;
Fig. 3 is a kind of sparse self-encoding encoder and grader structural representation disclosed in the embodiment of the present invention;
Fig. 4 is a kind of structural representation for stacking sparse self-encoding encoder and grader disclosed in the embodiment of the present invention;
Fig. 5 is a kind of schematic flow sheet for the defect characteristic for extracting RGB color image disclosed in the embodiment of the present invention;
Fig. 6 is a kind of structure of the product defects Classification and Identification device based on autocoder disclosed in the embodiment of the present invention
Schematic diagram.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Site preparation is described, it is clear that described embodiment is only a part of embodiment of the invention, rather than whole embodiments.It is based on
Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under the premise of creative work is not made
Embodiment, belongs to the scope of protection of the invention.
The embodiment of the invention discloses a kind of product defects classifying identification method and device based on autocoder, to solve
Computationally intensive, the problem of processing procedure is cumbersome when certainly recognizing product defects, obtains abstract defect characteristic and product is lacked with improving
Sunken discrimination.
Referring to Fig. 1, a kind of product defects classifying identification method based on autocoder provided in an embodiment of the present invention, bag
Include:
S101, acquisition defect image;
Specifically, the defect image in the present embodiment is to be shot by industrial camera equipment or other image picking-up apparatus
Defect image comprising product defects, it can be the image of RGB color pattern, naturally it is also possible to be the figure of other color modes
Picture.In the present embodiment by taking the image of RGB color color mode as an example, defect image is by red in RGB color pattern
(R), green (G), the change of blue (B) three Color Channels and their superpositions each other obtain constituting each of defect image
Color is planted, R, G, B are the color for representing three passages of red, green, blue.
S102, the defect characteristic by stacking at least four sparse self-encoding encoders extraction defect images;
Obtain, please join specifically, the sparse self-encoding encoder in the present embodiment is subject to openness limitation to self-encoding encoder
See Fig. 2, self-encoding encoder typically has three layers, including input layer, hidden layer and output layer, wherein, hidden layer is also referred to as intermediate layer;
When autocoder learns to input sample, its training objective is to reconstruct to come from objective expression by input signal.One
The structure of individual autocoder can be divided into coding and decoding two parts, and cataloged procedure is that input layer to hidden layer passes through weights structure
The mapping relations of input layer and hidden layer are built, and decoding process is to reconstruct input layer by the weights relation of hidden layer and output layer
Data;When inputting piece image, by the pixel value of R (G or B) passage of hidden layer entire image, according to from left to right from
The order of top to bottm, deploys successively, and each pixel value is inputted to a neuron of input layer, and how many pixel input layer is just
How many neuron, it is assumed that be originally inputted as x, the weights of input layer to hidden layer are (W, b), mapping function is Sigmoid,
Hidden layer output is characterized as y, wherein, w is weight matrix, and b is bias vector, therefore self-encoding encoder completes input vector x to hidden
The Mapping and Converting between the output y of layer is hidden, its typical expression formula is:Y=fΦ(x)=u (wx+b);Wherein, the y that hidden layer is obtained
The characteristics of image of the defect image of input layer input is represented, now, y is referred to as code;By hidden layer and output layer, then lead to
Code reconstruct input datas are crossed, are obtained The data of as sparse autocoder reconstruct.
For example, one secondary 128*128 of input defect image, has 16384 pixels, so the section of input layer and output layer
Point quantity is 16384.It is 8000 to take hidden layer number of nodes.Now, 16384 pixel values of input layer input are through self-encoding encoder
After coded portion coding, 16384 pixel values of input layer are encoded to 8000 data of hidden layer, now, hidden layer
8000 data are the characteristics of image of the defect image for the 128*128 for representing input.If hidden layer number of nodes is very big, very
During to than more than input layer quantity, at this moment need to add openness limitation, openness limitation requires most of hidden layer
Node is output as 0, here it is sparse self-encoding encoder, the most number state of output for being embodied in sigmoid functions is 0, therefore
Sparse self-encoding encoder has the ability of more excellent learning data.
S103, the defect characteristic sent to grader, recognize that the defect characteristic is corresponding by the grader
Defect classification.
Specifically, the softmax (grader) in the present embodiment only has two layers, i.e. input layer and output layer, grader
Input layer refers to Fig. 3, the input neuron of softmax input layers to stack the characteristics of image that sparse self-encoding encoder is extracted
Number determines that the neuron number of softmax output layers is sparse certainly by stacking by stacking the characteristics of image that sparse self-encoding encoder is extracted
The classification of the defect characteristic included in the characteristics of image that encoder is extracted is determined.The defect image extracted according to sparse self-encoding encoder
Characteristics of image classification, set softmax corresponding product defect output unit number;Softmax each exports list
Member represents to stack respectively the probability of the classification of the defect characteristic corresponding to the characteristics of image of sparse self-encoding encoder extraction, will
The most probable value of Softmax output units output is used as the defect classification for input defect image.
For example, identification sand holes, scratch are needed in a class product defects, lacks material, deformation, pit, the class defect of greasy dirt six, so
Setting softmax has 6 output units, a kind of defect classification of each output unit correspondence;After defect image input, by heap
After the network operations for folding sparse self-encoding encoder, softmax6 output unit exports P (0 | x) (probability for representing sand holes) and is respectively
0.1st, output P (1 | x) (probability for representing scratch) be 0.5, output P (2 | x) (representing the probability for lacking material) be 0.08, export P (3 |
X) probability of deformation (represent) be that 0.6, output P (4 | x) (probability for representing pit) is 0.04, exports P (5 | x) and (represent greasy dirt
Probability) be 0.1, now, it is defeated that we, which just think to stack the corresponding defect classification of characteristics of image that sparse self-encoding encoder extracts,
Go out P (3 | x) corresponding deformation defect.
It should be noted that in the present embodiment, stacking can be formed by least four sparse self-encoding encoders stackings self-editing
Code device, refers to Fig. 4, self-encoding encoder is had depth by stacking, and forms depth self-encoding encoder.Stack sparse self-encoding encoder
Number is more, and its network depth is deeper, and the characteristics of image finally given also can be more abstract, by last layer of sparse self-encoding encoder
The characteristics of image that hidden layer is obtained is exported to grader, realizes the defect recognition to defect image.Further, it is contemplated that there is ring
The influence of border factor, carries out noise processed to the defect image of input layer, the sparse self-encoding encoder of noise reduction is obtained, to obtain more may be used
Lean on, stable defect image feature, meanwhile, during autocoder identification defect image, using GPU parallel computations,
Calculating process is accelerated, and the more image of number of pixels can be handled.
Further, constituted in the present embodiment by stacking at least four sparse self-encoding encoders in depth network, the present embodiment
Exemplified by stacking six sparse self-encoding encoders and be constituted depth network, it is of course also possible to stack other positive integers for being more than or equal to four
Individual sparse self-encoding encoder constitutes depth network, for example, stack 9,12, even more many, herein and is not construed as limiting.
It can be seen that, a kind of product defects classifying identification method based on autocoder provided in an embodiment of the present invention passes through
Defect image is obtained, using the defect characteristic stacked at least four sparse self-encoding encoders extraction defect images, by the defect
Feature inputs to grader and carries out defect Classification and Identification again;Wherein, the stacking structure of at least four sparse self-encoding encoders of stacking is passed through
Into depth self-encoding encoder, the abstract characteristics of the image extracted by depth self-encoding encoder improve the identification to defect image
Rate.Meanwhile, the abstract characteristics of defect image are extracted by depth self-encoding encoder, are not required to combine image conversion, cutting defect image
Etc. a series of image processing algorithm, amount of calculation is substantially reduced, and inputs defect image by the input layer of sparse self-encoding encoder,
The hidden layer of sparse self-encoding encoder device directly obtains the defect characteristic of the defect image of input layer, and extraction process is simple, meanwhile, it is right
The defect image of input layer carries out noise processed, relatively reliable to obtain, the defect image feature of stabilization, in calculating process,
Using GPU parallel computations, calculating process is accelerated.
It is in the present embodiment, described to extract described by stacking at least four sparse self-encoding encoders based on above-described embodiment
Defect image defect characteristic, including:
Obtain the first mode image, second mode image and the 3rd mode image of the defect image, the defect map
As being coloured image, the first mode image is the R passages of defect image, and second mode image is the G passages of defect image,
3rd mode image is the channel B of defect image;
By stack six sparse self-encoding encoders extract respectively the first mode image, the second mode image and
The defect characteristic of 3rd mode image.
Specifically, defect image can be the coloured image of rgb color pattern in the present embodiment, it is of course also possible to be it
The image of its pattern, herein and is not construed as limiting, and in the present embodiment, by taking the coloured image of rgb color pattern as an example, refers to Fig. 5,
By obtained after the rasterized processing of RGB defect images R passages, G passages, channel B three kinds of images, wherein, the defect map of R passages
Picture color is only red, the like;Using the defect image of R passages as first mode image, the like, by three passages
Defect image input to stacking sparse self-encoding encoder, pass through hidden layer and output layer the reconstruct input of each sparse self-encoding encoder
Defect image data, and constantly update the weights and error of each sparse self-encoding encoder, it is defeated when error is in allowed band
Characteristic value to the grader classification for going out three passages is exported.
Further, the sparse self-encoding encoder of stacking constituted by stacking six sparse self-encoding encoders extracts R passages, G respectively
The characteristics of image of passage, channel B, i.e., input R passages, G passages, the pixel value of the defect image of channel B respectively, dilute by stacking
The hidden layer for dredging self-encoding encoder extracts R passages, G passages, the defect characteristic of the defect image of three passages of channel B.
It should be noted that constituting depth network by stacking six sparse self-encoding encoders in the present embodiment, certainly, also may be used
To stack more sparse self-encoding encoders to increase network depth, so that more abstract characteristics of image is extracted, for example, stacking 9
It is individual, 12, it is even more many, herein and be not construed as limiting.
It can be seen that, in the present embodiment, processing is color defect image, is led to respectively in the R passages, G passages, B of defect image
Characteristics of image is extracted on the different passages in three, road, the characteristics of image of different passages is obtained, is different from general gray level image, it is colored
Image comprising containing much information, to the defect characteristic description of product defects more it is accurate in detail, wherein, the defect map of different passage
As sharing a depth self-encoding encoder network, network size is smaller, therefore significantly reduces amount of calculation.By extracting three passages
Characteristics of image, the description of the defect image feature to being included in characteristics of image is more careful, accurate, and classifying quality is more preferably, right
The discrimination of product defects is also higher.
It is in the present embodiment, described to be extracted respectively by stacking six sparse self-encoding encoders based on above-mentioned any embodiment
The defect characteristic of the first mode image, the second mode image and the 3rd mode image, including:
The first mode image, the second mode image are extracted respectively using six sparse noise reduction self-encoding encoders are stacked
And the defect characteristic of the 3rd mode image.
Specifically, in the present embodiment, the sparse self-encoding encoder of noise reduction is on the basis of sparse self-encoding encoder, to input data
Carry out plus make an uproar processing so that sparse self-encoding encoder study removes this noise;Generally add the method for processing of making an uproar:One kind is that addition is high
This noise;Another is, with bi-distribution random process data, input data to be set to 0 with certain probability;It is, of course, also possible to there is it
His method, herein and is not construed as limiting.
It can be seen that, the embodiment of the present invention improves sparse self-encoding encoder pair by processing that sparse self-encoding encoder is carried out plus made an uproar
Input the generalization ability of defect image so that the characteristics of image that sparse self-encoding encoder is extracted has good robustness, that is, carries
The characteristics of image of the defect image taken is more stablized, reliably.
It is in the present embodiment, described to be distinguished by stacking six sparse noise reduction self-encoding encoders based on above-mentioned any embodiment
The defect characteristic of the first mode image, the second mode image and the 3rd mode image is extracted, including:
The first mode image, the second mode image and institute are obtained respectively using the first sparse noise reduction self-encoding encoder
The pixel value of the 3rd mode image is stated, so that pixel value coding is obtained the first spy by the first sparse noise reduction self-encoding encoder
Levy;
The fisrt feature is sent to the second sparse noise reduction self-encoding encoder, so that the second sparse noise reduction self-encoding encoder
Second feature is obtained to fisrt feature coding;
The second feature is sent to the 3rd sparse noise reduction self-encoding encoder, so that the 3rd sparse noise reduction self-encoding encoder
Third feature is obtained to second feature coding;
The third feature is sent to the 4th sparse noise reduction self-encoding encoder, so that the 4th sparse noise reduction self-encoding encoder
Fourth feature is obtained to third feature coding;
The fourth feature is sent to the 5th sparse noise reduction self-encoding encoder, so that the 5th sparse noise reduction self-encoding encoder
Fifth feature is obtained to fourth feature coding;
The fifth feature is sent to the 6th sparse noise reduction self-encoding encoder, so that the 6th sparse noise reduction self-encoding encoder
Sixth feature is obtained to fifth feature coding;It regard the sixth feature as the defect characteristic, the defect characteristic bag
Include first mode image deflects feature, second mode image deflects feature and the 3rd mode image defect characteristic.
The sparse self-encoding encoder of noise reduction formed is stacked by the sparse self-encoding encoder of multiple noise reductions, and to be properly termed as stack noise reduction sparse
Self-encoding encoder.The deeper level characteristics of input defect image can be obtained by the sparse self-encoding encoder of stack noise reduction.By preceding layer
The output of the hidden layer of the sparse self-encoding encoder of noise reduction is successively carried out as the input of the sparse self-encoding encoder of later layer noise reduction.For
To the sparse self-encoding encoder of noise reduction, first with by adding the defect image that is originally inputted for processing of making an uproar to train first noise reduction sparse self-editing
Code device, the fisrt feature for obtaining inputting defect image represents that it is sparse certainly that the fisrt feature then is denoted as into next noise reduction
The input of encoder, obtains second feature and represents, the like, n-th of rank character representation is sparse self-editing as n-th of noise reduction
The input of code device, obtains n-th of character representation, wherein, n is the positive integer more than or equal to 4, sparse after noise processed by adding
Self-encoding encoder is used as the sparse self-encoding encoder of noise reduction for recognizing product defects.The sparse self-encoding encoder of stack noise reduction is used in the present embodiment
Network contains 6 layers of hidden layer, it is of course also possible to there is any positive integer hidden layer more than 6, herein and is not construed as limiting.
It should be noted that in the present embodiment, self-encoding encoder net is added by stacking six sparse noise reduction self-encoding encoders
Network depth, it is of course also possible to which it is deep to form deeper self-encoding encoder network to stack the sparse noise reduction self-encoding encoder of more
Degree, herein and is not construed as limiting.
It can be seen that, in the present embodiment, the sparse noise reduction own coding of stack is obtained by stacking six sparse noise reduction self-encoding encoders
Device, increases network depth so that the characteristics of image extracted is more abstract by stacking sparse noise reduction self-encoding encoder, identification production
The accuracy of product defect is higher;And improve sparse self-encoding encoder to input defect image generalization ability so that it is sparse from
The characteristics of image that encoder is extracted has good robustness, that is, the characteristics of image of the defect image extracted more stablizes, can
Lean on.
It is described to send the defect characteristic to grader in the present embodiment based on above-described embodiment, pass through the classification
The corresponding defect classification of device identification defect characteristic includes:
The defect characteristic is sent to the input layer of grader;
The probability of each defect type of the defect characteristic correspondence is calculated by the input layer;
The maximum probability in the probability is selected, the corresponding defect type of the maximum probability is regard as the defect characteristic
Defect type.
Specifically, the grader in the present embodiment has been discussed in detail above, it will not be repeated here.Use softmax points
Class device, when the defect classification increase of defect, it is only necessary to which the output neuron number of corresponding increase softmax graders just can be with
Represent corresponding classification.For example, product defects classification is initially 6, due to the influence of environmental factor, the defect classification of defect increases
7 are added as, now, need to be in softmax graders output neuron only on the basis of 6, then to increasing softmax graders more
Plus an output neuron.
It can be seen that, in the present embodiment, using softmax graders, when defect classification increase, it is only necessary to corresponding increase output
Neuron number can just represent corresponding classification.Whole product defects identifying system need not be adjusted, processing procedure is simplified.
The product defects Classification and Identification device provided in an embodiment of the present invention based on autocoder is introduced below,
Product defects Classification and Identification device described below based on autocoder and the above-described production based on autocoder
Product defect classifying identification method can be with cross-referenced.
Referring to Fig. 6, a kind of product defects Classification and Identification device based on autocoder provided in an embodiment of the present invention, bag
Include:
Defect image acquisition module 100, for obtaining defect image;
Defect characteristic extraction module 200, for extracting the defect image by stacking at least four sparse self-encoding encoders
The defect characteristic for the defect image that acquisition module is obtained;
Defect characteristic sending module 300, the defect characteristic for defect characteristic extraction module to be obtained send to point
Class device, the corresponding defect classification of defect characteristic is recognized by the grader.
Based on above-described embodiment, in the present embodiment, the defect characteristic extraction module 200 includes:
Acquiring unit, first mode image, second mode image and the 3rd ideograph for obtaining the defect image
Picture, the defect image is coloured image, and the first mode image is the R passages of defect image, and second mode image is scarce
The G passages of image are fallen into, the 3rd mode image is the channel B of defect image;
First extraction unit, for extracting the institute that the acquiring unit is obtained respectively by stacking six sparse self-encoding encoders
State the defect characteristic of first mode image, the second mode image and the 3rd mode image.
Based on above-described embodiment, in the present embodiment, first extraction unit includes:
Second extraction unit, for extracting the first mode figure respectively by stacking six sparse noise reduction self-encoding encoders
The defect characteristic of picture, the second mode image and the 3rd mode image.
Based on above-described embodiment, in the present embodiment, second extraction unit includes:
Subelement is obtained, for obtaining the first mode image respectively using the first sparse noise reduction self-encoding encoder, described
The pixel value of second mode image and the 3rd mode image, so that the first sparse noise reduction self-encoding encoder is by the pixel
Value coding obtains fisrt feature;
First transmission sub-unit, sends the fisrt feature to the second sparse noise reduction for obtain acquisition subelement
Self-encoding encoder, so that the second sparse noise reduction self-encoding encoder obtains second feature to fisrt feature coding;
Second transmission sub-unit, sends the second feature to the 3rd for obtain first transmission sub-unit
Sparse noise reduction self-encoding encoder, so that the 3rd sparse noise reduction self-encoding encoder obtains third feature to second feature coding;
3rd transmission sub-unit, the third feature for second transmission sub-unit to be obtained is sent to the 4th sparse drop
Make an uproar self-encoding encoder, so that the 4th sparse noise reduction self-encoding encoder obtains fourth feature to third feature coding;
4th transmission sub-unit, the fourth feature for the 3rd transmission sub-unit to be obtained is sent to the 5th sparse drop
Make an uproar self-encoding encoder, so that the 5th sparse noise reduction self-encoding encoder obtains fifth feature to fourth feature coding;
5th transmission sub-unit, the fifth feature for the 4th transmission sub-unit to be obtained is sent to the 6th sparse drop
Make an uproar self-encoding encoder, so that the 6th sparse noise reduction self-encoding encoder obtains sixth feature to fifth feature coding;Will be described
Sixth feature is as the defect characteristic, and the defect characteristic includes first mode image deflects feature, second mode image and lacked
Fall into feature and the 3rd mode image defect characteristic.
The defect characteristic sending module 300 includes:
Transmitting element, for the defect characteristic to be sent to grader input layer;
Computing unit, for calculating the defect characteristic pair that the transmitting element is sent by the grader input layer
Answer the probability of each defect type;
Selecting unit, the maximum probability in the probability for selecting the computing unit calculating is most general by described in
The corresponding defect type of rate as the defect characteristic defect type.
It can be seen that, in this programme, defect characteristic extraction module 200 extracts scarce using at least four sparse self-encoding encoders are stacked
The defect characteristic is inputted to grader and carries out defect point by the defect characteristic fallen into image, defect characteristic sending module 300 again
Class is recognized;Wherein, depth self-encoding encoder is constituted by the stacking of at least four sparse self-encoding encoders, carried by depth self-encoding encoder
The abstract characteristics of image got, improves the discrimination to defect image.Meanwhile, defect characteristic extraction module 200, which is extracted, to be lacked
The defect characteristic of image is fallen into, is not required to combine a series of image processing algorithms such as image conversion, cutting defect image, amount of calculation shows
Write and reduce, and defect image is inputted by the input layer of sparse self-encoding encoder, the hidden layer of sparse self-encoding encoder device is directly obtained
The defect characteristic of the defect image of input layer, extraction process is simple, meanwhile, noise processed is carried out to the defect image of input layer,
Obtain relatively reliable, stable defect image feature in calculating process, using GPU parallel computations, accelerates calculating process.
The embodiment of each in this specification is described by the way of progressive, and what each embodiment was stressed is and other
Between the difference of embodiment, each embodiment identical similar portion mutually referring to.
The foregoing description of the disclosed embodiments, enables professional and technical personnel in the field to realize or using the present invention.
A variety of modifications to these embodiments will be apparent for those skilled in the art, as defined herein
General Principle can be realized in other embodiments without departing from the spirit or scope of the present invention.Therefore, it is of the invention
The embodiments shown herein is not intended to be limited to, and is to fit to and principles disclosed herein and features of novelty phase one
The most wide scope caused.
Claims (10)
1. a kind of product defects classifying identification method based on autocoder, it is characterised in that including:
Obtain defect image;
The defect characteristic of the defect image is extracted by stacking at least four sparse self-encoding encoders;
The defect characteristic is sent to grader, the corresponding defect classification of the defect characteristic is recognized by the grader.
2. the product defects classifying identification method according to claim 1 based on autocoder, it is characterised in that described
The defect characteristic of the defect image is extracted by stacking at least four sparse self-encoding encoders, including:
The first mode image, second mode image and the 3rd mode image of the defect image are obtained, the defect image is
Coloured image, the first mode image is the R passages of defect image, and second mode image is the G passages of defect image, the 3rd
Mode image is the channel B of defect image;
The first mode image, the second mode image and described are extracted respectively by stacking six sparse self-encoding encoders
The defect characteristic of 3rd mode image.
3. the product defects classifying identification method according to claim 2 based on autocoder, it is characterised in that described
The first mode image, the second mode image and the 3rd mould are extracted respectively by stacking six sparse self-encoding encoders
The defect characteristic of formula image, including:
The first mode image, the second mode image and institute are extracted respectively by stacking six sparse noise reduction self-encoding encoders
State the defect characteristic of the 3rd mode image.
4. the product defects classifying identification method according to claim 3 based on autocoder, it is characterised in that described
The first mode image, the second mode image and described the are extracted respectively by stacking six sparse noise reduction self-encoding encoders
The defect characteristic of three mode images, including:
The first mode image, the second mode image and described are obtained respectively using the first sparse noise reduction self-encoding encoder
The pixel value of three mode images, so that the first sparse noise reduction self-encoding encoder obtains fisrt feature to pixel value coding;
The fisrt feature is sent to the second sparse noise reduction self-encoding encoder, so that the second sparse noise reduction self-encoding encoder is to institute
State fisrt feature coding and obtain second feature;
The second feature is sent to the 3rd sparse noise reduction self-encoding encoder, so that the 3rd sparse noise reduction self-encoding encoder is to institute
State second feature coding and obtain third feature;
The third feature is sent to the 4th sparse noise reduction self-encoding encoder, so that the 4th sparse noise reduction self-encoding encoder is to institute
State third feature coding and obtain fourth feature;
The fourth feature is sent to the 5th sparse noise reduction self-encoding encoder, so that the 5th sparse noise reduction self-encoding encoder is to institute
State fourth feature coding and obtain fifth feature;
The fifth feature is sent to the 6th sparse noise reduction self-encoding encoder, so that the 6th sparse noise reduction self-encoding encoder is to institute
State fifth feature coding and obtain sixth feature;Using the sixth feature as the defect characteristic, the defect characteristic includes the
One mode image defect characteristic, second mode image deflects feature and the 3rd mode image defect characteristic.
5. the product defects classifying identification method based on autocoder according to claim 1-4 any one, it is special
Levy and be, it is described to send the defect characteristic to grader, the corresponding defect class of defect characteristic is recognized by the grader
Do not include:
The defect characteristic is sent to the input layer of grader;
The probability of each defect type of the defect characteristic correspondence is calculated by the input layer;
The maximum probability in the probability is selected, the corresponding defect type of the maximum probability is regard as the scarce of the defect characteristic
Fall into type.
6. a kind of product defects Classification and Identification device based on autocoder, it is characterised in that including:
Defect image acquisition module, for obtaining defect image;
Defect characteristic extraction module, for extracting the defect image acquisition module by stacking at least four sparse self-encoding encoders
The defect characteristic of the obtained defect image;
Defect characteristic sending module, the defect characteristic for defect characteristic extraction module to be obtained is sent to grader, is led to
Cross the corresponding defect classification of the grader identification defect characteristic.
7. the product defects Classification and Identification device according to claim 6 based on autocoder, it is characterised in that described
Defect characteristic extraction module includes:
Acquiring unit, first mode image, second mode image and the 3rd mode image for obtaining the defect image, institute
Defect image is stated for coloured image, the first mode image is the R passages of defect image, second mode image is defect image
G passages, the 3rd mode image be defect image channel B;
First extraction unit, for described the by stacking that six sparse self-encoding encoders extract that the acquiring unit obtains respectively
The defect characteristic of one mode image, the second mode image and the 3rd mode image.
8. the product defects Classification and Identification device according to claim 7 based on autocoder, it is characterised in that described
First extraction unit includes:
Second extraction unit, for extracting the first mode image, institute respectively by stacking six sparse noise reduction self-encoding encoders
State the defect characteristic of second mode image and the 3rd mode image.
9. the product defects Classification and Identification device according to claim 8 based on autocoder, it is characterised in that described
Second extraction unit includes:
Subelement is obtained, for obtaining the first mode image, described second respectively using the first sparse noise reduction self-encoding encoder
The pixel value of mode image and the 3rd mode image, so that the first sparse noise reduction self-encoding encoder compiles the pixel value
Code obtains fisrt feature;
First transmission sub-unit, the fisrt feature for acquisition subelement to be obtained is sent to the second sparse noise reduction own coding
Device, so that the second sparse noise reduction self-encoding encoder obtains second feature to fisrt feature coding;
Second transmission sub-unit, the second feature for first transmission sub-unit to be obtained is sent to the 3rd sparse drop
Make an uproar self-encoding encoder, so that the 3rd sparse noise reduction self-encoding encoder obtains third feature to second feature coding;
3rd transmission sub-unit, the third feature for second transmission sub-unit to be obtained is sent to the 4th sparse drop
Make an uproar self-encoding encoder, so that the 4th sparse noise reduction self-encoding encoder obtains fourth feature to third feature coding;
4th transmission sub-unit, the fourth feature for the 3rd transmission sub-unit to be obtained is sent to the 5th sparse drop
Make an uproar self-encoding encoder, so that the 5th sparse noise reduction self-encoding encoder obtains fifth feature to fourth feature coding;
5th transmission sub-unit, the fifth feature for the 4th transmission sub-unit to be obtained is sent to the 6th sparse drop
Make an uproar self-encoding encoder, so that the 6th sparse noise reduction self-encoding encoder obtains sixth feature to fifth feature coding;Will be described
Sixth feature is as the defect characteristic, and the defect characteristic includes first mode image deflects feature, second mode image and lacked
Fall into feature and the 3rd mode image defect characteristic.
10. the product defects Classification and Identification device based on autocoder according to claim 6-9 any one, it is special
Levy and be, the defect characteristic sending module includes:
Transmitting element, for the defect characteristic to be sent to grader input layer;
Computing unit, the defect characteristic correspondence for calculating the transmitting element transmission by the grader input layer is every
A kind of probability of defect type;
Selecting unit, the maximum probability in the probability for selecting the computing unit calculating, by the maximum probability pair
The defect type answered as the defect characteristic defect type.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710671719.1A CN107292885A (en) | 2017-08-08 | 2017-08-08 | A kind of product defects classifying identification method and device based on autocoder |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710671719.1A CN107292885A (en) | 2017-08-08 | 2017-08-08 | A kind of product defects classifying identification method and device based on autocoder |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107292885A true CN107292885A (en) | 2017-10-24 |
Family
ID=60104529
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710671719.1A Pending CN107292885A (en) | 2017-08-08 | 2017-08-08 | A kind of product defects classifying identification method and device based on autocoder |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107292885A (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108596892A (en) * | 2018-04-23 | 2018-09-28 | 西安交通大学 | A kind of identification of Weld Defects based on improvement LeNet-5 models |
CN109259733A (en) * | 2018-10-25 | 2019-01-25 | 深圳和而泰智能控制股份有限公司 | Apnea detection method, apparatus and detection device in a kind of sleep |
CN110096632A (en) * | 2019-04-16 | 2019-08-06 | 华东师范大学 | A kind of brush list person's detection method based on sparse self-encoding encoder |
CN110188774A (en) * | 2019-05-27 | 2019-08-30 | 昆明理工大学 | A kind of current vortex scan image classifying identification method based on deep learning |
CN110619618A (en) * | 2018-06-04 | 2019-12-27 | 杭州海康威视数字技术股份有限公司 | Surface defect detection method and device and electronic equipment |
CN111028210A (en) * | 2019-11-25 | 2020-04-17 | 北京航天控制仪器研究所 | Deep neural network glass tube end surface defect detection method based on sparse automatic encoder |
CN111595790A (en) * | 2020-05-30 | 2020-08-28 | 南京林业大学 | Hyperspectral image-based green plum acidity prediction method |
CN112861625A (en) * | 2021-01-05 | 2021-05-28 | 深圳技术大学 | Method for determining stacking denoising autoencoder model |
CN112884712A (en) * | 2021-01-22 | 2021-06-01 | 深圳精智达技术股份有限公司 | Method and related device for classifying defects of display panel |
EP3770851A4 (en) * | 2019-05-23 | 2021-07-21 | Wangsu Science & Technology Co., Ltd. | Quality inspection method for target object, and edge computing device |
CN114637270A (en) * | 2022-05-17 | 2022-06-17 | 成都秦川物联网科技股份有限公司 | Intelligent manufacturing industry Internet of things based on distributed control and control method |
CN116228769A (en) * | 2023-05-09 | 2023-06-06 | 和峻(广州)胶管有限公司 | Device and method suitable for flaw detection of steel wire braided tube |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104166859A (en) * | 2014-08-13 | 2014-11-26 | 西安电子科技大学 | Polarization SAR image classification based on SSAE and FSALS-SVM |
CN104484658A (en) * | 2014-12-30 | 2015-04-01 | 中科创达软件股份有限公司 | Face gender recognition method and device based on multi-channel convolution neural network |
CN104978580A (en) * | 2015-06-15 | 2015-10-14 | 国网山东省电力公司电力科学研究院 | Insulator identification method for unmanned aerial vehicle polling electric transmission line |
CN105654121A (en) * | 2016-04-11 | 2016-06-08 | 李云栋 | Complex jacquard fabric defect detection method based on deep learning |
CN106446930A (en) * | 2016-06-28 | 2017-02-22 | 沈阳工业大学 | Deep convolutional neural network-based robot working scene identification method |
CN106803062A (en) * | 2016-12-20 | 2017-06-06 | 陕西师范大学 | The recognition methods of stack noise reduction own coding neutral net images of gestures |
-
2017
- 2017-08-08 CN CN201710671719.1A patent/CN107292885A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104166859A (en) * | 2014-08-13 | 2014-11-26 | 西安电子科技大学 | Polarization SAR image classification based on SSAE and FSALS-SVM |
CN104484658A (en) * | 2014-12-30 | 2015-04-01 | 中科创达软件股份有限公司 | Face gender recognition method and device based on multi-channel convolution neural network |
CN104978580A (en) * | 2015-06-15 | 2015-10-14 | 国网山东省电力公司电力科学研究院 | Insulator identification method for unmanned aerial vehicle polling electric transmission line |
CN105654121A (en) * | 2016-04-11 | 2016-06-08 | 李云栋 | Complex jacquard fabric defect detection method based on deep learning |
CN106446930A (en) * | 2016-06-28 | 2017-02-22 | 沈阳工业大学 | Deep convolutional neural network-based robot working scene identification method |
CN106803062A (en) * | 2016-12-20 | 2017-06-06 | 陕西师范大学 | The recognition methods of stack noise reduction own coding neutral net images of gestures |
Non-Patent Citations (3)
Title |
---|
ZUNLIN FAN ET AL.: "Low-level structure feature extraction for image processing via stacked sparse denoising autoencoder", 《NEUROCOMPUTING》 * |
戴晓爱 等: "基于堆栈式稀疏自编码器的高光谱影像分类", 《电子科技大学学报》 * |
赵飞翔 等: "基于栈式降噪稀疏自动编码器的雷达目标识别方法", 《雷达学报》 * |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108596892A (en) * | 2018-04-23 | 2018-09-28 | 西安交通大学 | A kind of identification of Weld Defects based on improvement LeNet-5 models |
CN108596892B (en) * | 2018-04-23 | 2020-08-18 | 西安交通大学 | Weld defect identification method based on improved LeNet-5 model |
CN110619618B (en) * | 2018-06-04 | 2023-04-07 | 杭州海康威视数字技术股份有限公司 | Surface defect detection method and device and electronic equipment |
CN110619618A (en) * | 2018-06-04 | 2019-12-27 | 杭州海康威视数字技术股份有限公司 | Surface defect detection method and device and electronic equipment |
CN109259733A (en) * | 2018-10-25 | 2019-01-25 | 深圳和而泰智能控制股份有限公司 | Apnea detection method, apparatus and detection device in a kind of sleep |
CN110096632A (en) * | 2019-04-16 | 2019-08-06 | 华东师范大学 | A kind of brush list person's detection method based on sparse self-encoding encoder |
EP3770851A4 (en) * | 2019-05-23 | 2021-07-21 | Wangsu Science & Technology Co., Ltd. | Quality inspection method for target object, and edge computing device |
CN110188774B (en) * | 2019-05-27 | 2022-12-02 | 昆明理工大学 | Eddy current scanning image classification and identification method based on deep learning |
CN110188774A (en) * | 2019-05-27 | 2019-08-30 | 昆明理工大学 | A kind of current vortex scan image classifying identification method based on deep learning |
CN111028210A (en) * | 2019-11-25 | 2020-04-17 | 北京航天控制仪器研究所 | Deep neural network glass tube end surface defect detection method based on sparse automatic encoder |
CN111595790A (en) * | 2020-05-30 | 2020-08-28 | 南京林业大学 | Hyperspectral image-based green plum acidity prediction method |
CN112861625A (en) * | 2021-01-05 | 2021-05-28 | 深圳技术大学 | Method for determining stacking denoising autoencoder model |
CN112861625B (en) * | 2021-01-05 | 2023-07-04 | 深圳技术大学 | Determination method for stacked denoising self-encoder model |
CN112884712A (en) * | 2021-01-22 | 2021-06-01 | 深圳精智达技术股份有限公司 | Method and related device for classifying defects of display panel |
CN114637270A (en) * | 2022-05-17 | 2022-06-17 | 成都秦川物联网科技股份有限公司 | Intelligent manufacturing industry Internet of things based on distributed control and control method |
US11681283B1 (en) | 2022-05-17 | 2023-06-20 | Chengdu Qinchuan Iot Technology Co., Ltd. | Intelligent manufacturing industrial Internet of Things based on distributed control, control methods and media thereof |
CN116228769A (en) * | 2023-05-09 | 2023-06-06 | 和峻(广州)胶管有限公司 | Device and method suitable for flaw detection of steel wire braided tube |
CN116228769B (en) * | 2023-05-09 | 2023-10-17 | 和峻(广州)胶管有限公司 | Device and method suitable for flaw detection of steel wire braided tube |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107292885A (en) | A kind of product defects classifying identification method and device based on autocoder | |
CN105046277B (en) | Robust mechanism study method of the feature significance in image quality evaluation | |
CN107862668A (en) | A kind of cultural relic images restored method based on GNN | |
CN112837295A (en) | Rubber glove defect detection method based on generation of countermeasure network | |
CN105657402B (en) | A kind of depth map restoration methods | |
CN107392925A (en) | Remote sensing image terrain classification method based on super-pixel coding and convolutional neural networks | |
CN110109060A (en) | A kind of radar emitter signal method for separating and system based on deep learning network | |
CN107392314A (en) | A kind of deep layer convolutional neural networks method that connection is abandoned based on certainty | |
CN111709290B (en) | Crowd counting method based on coding and decoding-jump connection scale pyramid network | |
CN108171701A (en) | Conspicuousness detection method based on U networks and confrontation study | |
CN111862143B (en) | Automatic monitoring method for river dike collapse | |
CN108010034A (en) | Commodity image dividing method and device | |
CN108921942B (en) | Method and device for 2D (two-dimensional) conversion of image into 3D (three-dimensional) | |
CN108108751A (en) | A kind of scene recognition method based on convolution multiple features and depth random forest | |
CN110210608A (en) | The enhancement method of low-illumination image merged based on attention mechanism and multi-level features | |
CN104182771B (en) | Based on the graphical analysis method of the time series data with packet loss automatic coding | |
CN113627376B (en) | Facial expression recognition method based on multi-scale dense connection depth separable network | |
CN107944459A (en) | A kind of RGB D object identification methods | |
CN107423747A (en) | A kind of conspicuousness object detection method based on depth convolutional network | |
CN104463242B (en) | The multiple features action identification method that feature based is converted and dictionary learns | |
CN106780546A (en) | The personal identification method of the motion blur encoded point based on convolutional neural networks | |
CN110210577A (en) | A kind of deep learning and recognition methods for intensive flock of birds | |
CN107480723A (en) | Texture Recognition based on partial binary threshold learning network | |
CN115471423A (en) | Point cloud denoising method based on generation countermeasure network and self-attention mechanism | |
CN111553873A (en) | Automatic brain neuron detection method based on multi-scale convolutional neural network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171024 |