CN110147570A - It is a kind of that method for distinguishing is known based on the electronic component of texture and shape feature - Google Patents
It is a kind of that method for distinguishing is known based on the electronic component of texture and shape feature Download PDFInfo
- Publication number
- CN110147570A CN110147570A CN201910288579.9A CN201910288579A CN110147570A CN 110147570 A CN110147570 A CN 110147570A CN 201910288579 A CN201910288579 A CN 201910288579A CN 110147570 A CN110147570 A CN 110147570A
- Authority
- CN
- China
- Prior art keywords
- image
- electronic component
- texture
- feature
- shape
- 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
- 238000000034 method Methods 0.000 title claims abstract description 28
- 238000005452 bending Methods 0.000 claims abstract description 15
- 230000009466 transformation Effects 0.000 claims abstract description 11
- 238000012549 training Methods 0.000 claims abstract description 5
- 239000013598 vector Substances 0.000 claims description 18
- 238000007781 pre-processing Methods 0.000 claims description 9
- 150000001875 compounds Chemical class 0.000 claims description 8
- 238000013519 translation Methods 0.000 claims description 8
- 238000000605 extraction Methods 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 7
- 238000012360 testing method Methods 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 5
- 238000013528 artificial neural network Methods 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 4
- 239000000203 mixture Substances 0.000 claims description 4
- 230000006978 adaptation Effects 0.000 claims description 3
- 238000012804 iterative process Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 238000000354 decomposition reaction Methods 0.000 claims description 2
- 239000011800 void material Substances 0.000 claims 1
- 238000013527 convolutional neural network Methods 0.000 abstract description 3
- 238000009776 industrial production Methods 0.000 abstract 1
- 230000001360 synchronised effect Effects 0.000 abstract 1
- 239000003990 capacitor Substances 0.000 description 6
- 230000006870 function Effects 0.000 description 3
- 230000004927 fusion Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 210000000232 gallbladder Anatomy 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- XUIMIQQOPSSXEZ-UHFFFAOYSA-N Silicon Chemical group [Si] XUIMIQQOPSSXEZ-UHFFFAOYSA-N 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 239000004568 cement Substances 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000010363 phase shift Effects 0.000 description 1
- 238000002203 pretreatment Methods 0.000 description 1
- 229910052710 silicon Inorganic materials 0.000 description 1
- 239000010703 silicon Substances 0.000 description 1
- 230000031068 symbiosis, encompassing mutualism through parasitism Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of based on the electronic component of texture and shape feature knowledge method for distinguishing, comprising the following steps: by translating to target in image, target is normalized pretreatment stage for rotation and scaling;Gabor filter and gray level co-occurrence matrixes model the textural characteristics of electronic component;The shape of electronic component is captured by one group of Curvelet transformation coefficient and not bending moment;Training neuro-fuzzy classifier is to identify target electronic components.The present invention is based on the texture of target image and shape features, in conjunction with convolutional neural networks, identification is synchronized to multiple target, both the accuracy and efficiency that electronic component identifies in industrial production had been improved, the robustness of recognition methods is improved, again so as to be expanded to the identification application range of other rigidity targets.
Description
Technical field
Method for distinguishing is known based on the electronic component of texture and shape feature the present invention relates to a kind of.Use Gabor filter
(GF) and gray level co-occurrence matrixes (GLCM) carry out feature extraction modeling to the texture of electronic component, convert (CT) using one group of curve
Coefficient and not bending moment (IM) capture the shape of electronic component.It is used as classifier to identify by Fuzzy Neural Controller (NFC)
Electronic component.The present invention can will have different texture, and the electronic component of shape, size and Orientation is effectively known under complex environment
Not.
Background technique
Electronic component plays the role of very important in terms of industrial development, it is many kinds of, and towards micromation, chip
The direction evolution of change.The production of electronic component, scientific research, application and in terms of, Classification and Identification is a Xiang Feichang
Important element task, therefore design a kind of electronic component automatic identifying method that can be handled in real time there is highly important reality
Meaning.Current image classification method is broadly divided into two major classes, and the first kind is to utilize convolutional neural networks (convolution-al
Neural networks, CNN) study characteristics of image progress image classification figure, the second class are based on image space domain or frequency automatically
Classify to image in domain.First kind method is by years of researches and accumulation, although ideal in terms of nicety of grading,
Its data volume is excessive, and computation complexity is higher, is unable to satisfy the efficiency requirements handled in time;Second class method precision is inadequate,
It cannot achieve while classifying to a variety of components.
Summary of the invention
Based on above-mentioned deficiency, method for distinguishing is known based on the electronic component of texture and shape feature the invention proposes a kind of.
The texture of electronic component is modeled using Gabor filter (GF) and gray level co-occurrence matrixes (GLCM), uses one group of curve
Transformation (CT) and not bending moment (IM) obtain the shape of electronic component.Since these features are usually by the side of electronic component in image
Influenced to size, thus pretreatment (PP) stage before application feature extraction by translate to target in image, rotation with
Scaling, target is normalized.By using neuro-fuzzy classifier (NFC) to identify electronic component.The present invention can
There to be different texture, the electronic component of shape, size and Orientation efficiently identifies.
Texture and shape are the important features for Classification and Identification, and method of the invention, which uses, is based on textural characteristics and shape
Shape feature modeling it is compound.Use the line of compound Gabor filter (GF) and gray level co-occurrence matrixes (GLCM) capture electronic component
Reason, while (CT) and the not shape of bending moment (IM) capture electronic component are converted using curve.The step of execution, is summarized as follows: S1,
Pretreatment (PP), S2, Gabor filter (GF), S3, gray level co-occurrence matrixes (GLCM), S4, the feature (FT) based on texture, S5,
Curvelet converts (CT), S6, not bending moment (IM), S7, the feature (FS) based on shape, S8, Fusion Features (FC), S9, obscures
Neural classifier (NFC).
The technical solution of the present invention is as follows: it is a kind of based on the electronic component of texture and shape feature knowledge method for distinguishing, including with
Lower step:
Image preprocessing (PP): by color image gray processing, binaryzation, by the translation to target in image, rotation and
Scaling, target is normalized;
Gabor filter (GF): in conjunction with the thought of convolutional network, it is applied to the feature extraction of electronic component, Gabor
Filter and gray level image convolution generate complex signal, the characteristic information of electronic component are tentatively obtained, by the real part (rgg) of signal
It is separated with imaginary part (igg), the gray level co-occurrence matrixes of image is calculated by imaginary part, further characteristic information is handled;
Gray level co-occurrence matrixes (GLCM): electronic component is rigid structure, and texture is simple and symmetrical, and feature passes through gray scale symbiosis
Matrix Formula calculates, and obtains 3 feature representation F1、F2、F3, F1 is the energy of texture, and F2 is texture contrast, and F3 is texture phase
Guan Xing, the primary expression texture information of rigid body;
The feature (FT) of texture is extracted: being used gray level co-occurrence matrixes, is calculated in multiple angle values according to the image of filtering
The textural characteristics of portion's signal overlay electronic element reconnect characteristic value F1、F2、F3Obtain the texture feature vector of multiple elements, table
Up to the electronic component textural characteristics as rigid body;
Second Generation Curvelet Transform (CT): carrying out Curvelet transformation to electronic component image using Wrap algorithm, point
Orientation and the position of electronic component image different scale difference frequency domain are solved and then effectively obtain, 20 kinds of electronic components are in simple rule
Geometry, take four Curvelet coefficients, retain coefficient that those energy are positive and rank binaryzation electronic component image,
Composition is in the feature vector element of the electronic component of regular geometric shapes together;
Not bending moment (IM): image shape feature in translation, scaling, rotation does not change, by shape information with constant
Square (IM) expression processing, the square m of the pixel P (x, y) at position (x, y) are defined as the product of pixel value and its coordinate distance i.e.
M=xyP (x, y), graphical rule are p × q, and the square of whole image is the summation of the square of its all pixels: mpq=∑x∑yxpyqI(x,y);
The feature (FS) of shape is extracted: being obtained not bending moment, then is carried out quickly by Wrap algorithm to pretreated image
Curvelet transformation, and calculate three curve ripples coefficient CC1, CC2, CC3 calculates the first not bending moment M for each coefficient1;Its
Give expression to the shape of regular geometric rigid body.
Compound characteristics model (FC): textural characteristics (FT) are used for for capturing texture content, shape feature (FS) to its shape
Shape is modeled;
Classification and Identification: neuro-fuzzy classifier is combined using the automatic adaptation process of Fuzzy Classifier and neural network
Advantage identifies 20 kinds of electronics members such as resistance, capacitor, diode using the neuro-fuzzy classifier of conjugate gradient algorithms (NFC)
Part.
Further, described image pretreatment (PP) converts the image into gray scale and bi-level fashion, to guarantee target rotation not
Become, the angle map of the main shafts of target electronic components relative to its orientation, to guarantee target translation invariant, by background reduce until
Electronic component is suitble to bounding rectangles just, and to guarantee that target scale is constant, image is readjusted as standard size.
Further, using Gabor filter (GF) and gray level image convolution, complex signal J is generated, the preliminary of feature is obtained
Information isolates the imaginary part of J, calculates the gray level co-occurrence matrixes (GLCM) of image, and calculate and be based on gray level co-occurrence matrixes
(GLCM) feature.
Further, using the textural characteristics of gray level co-occurrence matrixes (GLCM) statistics electronic component, texture is by intensity profile
Occur repeatedly on spatial position and is formed, thus the meeting between two pixels for being separated by certain distance in electronic component image space
There are certain gray-scale relation, electronic component itself belongs to the rigid structure of simple symmetric, complete using three feature representation parameters
The whole texture information that must reflect target in image, F1 have reacted image grayscale and have been evenly distributed degree and texture fineness degree, F2 reaction
The clarity of image and the rill depth of texture, the gray level that F3 is used to measure image be expert at or column direction on similar journey
Degree.
Further, the feature based on texture (FT) is extracted, and by d=1, the gray level co-occurrence matrixes of N=256 calculate four
Angle corresponds to the four direction of electronic component, to filtered image calculate for four angle value θ=0 °, 45 °, 90 °,
135 ° } internal signal J, calculate characteristic value F1, F2, F3 by gray level co-occurrence matrixes, the final textural characteristics of each image are
Connect 12 vectors that 3 features are formed in 4 angles, just the expressed intact textural characteristics of electronic component.
Further, Curvelet transformation is carried out to image using Wrap algorithm, decomposition obtains each son of image different scale
The gray level image of the orientation of the different frequency domains of band and position, M × N is converted by the fast discrete Curvelet of Wrap algorithm, is produced
Raw one group of curve wave system number C, is scale a, direction b, space coordinate (p, q) respectively, 0≤m≤M, 0≤n≤N are curves here
Waveform, generates four Curvelet coefficients at this time, and four grades from 1 to 4 only retain those energy in all possible coefficient
The coefficient being positive is measured, is sorted from large to small, the binary image one of the first two coefficient (CC1 and CC2) and 1 coefficient of rank (CC3)
Act the element for forming shape eigenvectors.
Further, in compound characteristics model (FC), 12 texture vectors are connected into 3 shape vectors to generate 15 element groups
Resultant vector FC.
Further, using neuro-fuzzy classifier, kinds of electronic components, the feature of test image and certain training class are identified
Coincidence factor highest, target image is identified as certain kinds, while adjusting weight in an iterative process, enables a system to by following
Mode learning data set feature output and input observation between execute Nonlinear Mapping, then by the derivative model be applied to not
The test sample known identifies their type.
The present invention has following technical effect that
During texture feature extraction, the texture feature information that convolution is extracted is handled by gray level co-occurrence matrixes, is used
Texture energy, texture contrast and texture correlation, accurate and effective must express this kind of simple structure volume textures of electronic component
Information;Calculate internal signal when, choose 0,45 °, 90 °, 135 ° of angles, just overlay electronic element up and down, left and right and two
The cornerwise direction of item, processing electronic component are efficiently and accurate;By the Fusion Features of texture and shape, for small electrically realized
Not missing inspection, multiple electronic components not false retrieval, there is the migration of good robustness and the identification to other rigidity targets.
Detailed description of the invention
Fig. 1 is a kind of electronic component recognition methods flow chart based on texture, shape feature.
Fig. 2 is a kind of electronic component recognition methods experimental result based on texture, shape feature.
Specific embodiment
In order to deepen the understanding of the present invention, with reference to the accompanying drawing to of the present invention a kind of special based on texture and shape
The electronic component of sign is known method for distinguishing and is described in further detail.
The present invention is applied to electronic component and identifies field, as shown in Figure 1, being algorithm flow chart of the invention.
Railway freight-car latch deflection fault detection method proposed by the present invention includes filtering pre-treatment step, Gabor
Step, gray level co-occurrence matrixes processing step, the characterization step for extracting texture, Curvelet shift step, not bending moment step, extraction
Feature, texture and the shape feature fusion steps of shape train neuro-fuzzy classifier and identification step.Specific implementation includes such as
Under:
(a) (PP) is pre-processed to image, the size of electronic component and side in standardized images before feature extraction
To.Original image is usually the color image of random angles, direction and size, converts the image into gray scale and two-value shape first
Formula, to guarantee target invariable rotary, angle map of the main shaft of target electronic components relative to its orientation, to guarantee target translation
It is constant, background is reduced until electronic component is suitble to bounding rectangles just, to guarantee that target scale is constant, image is readjusted
For standard size.
(b) using Gabor filter (GF) and gray level image convolution, complex signal J is generated, the imaginary part of J is isolated,
The gray level co-occurrence matrixes (GLCM) of image are calculated, and calculate the feature for being based on gray level co-occurrence matrixes (GLCM).
Gabor filter is the product of Gaussian function and sine wave.The known 2D Gauss in the x and y direction with σ extension
Curve is expressed as follows:
Sine curve is defined as follows, wherein u representation space frequency, and θ indicates direction,Indicate phase shift, formula is as follows:
Gabor function is as follows:
Gray level image I (x, y) and Gabor filter h convolution generate complex signal J.
The real part (rgg) and imaginary part (igg) of signal are separated.
Rgg (x, y)=Re { J (x, y) }
Igg (x, y)=Im { J (x, y) }
(c) using the textural characteristics of gray level co-occurrence matrixes (GLCM) statistics electronic component, texture is by intensity profile in sky
Between occur repeatedly and formed on position, thus be separated by between two pixels of certain distance in image space and can have certain ash
Degree relationship calculates the mathematics in image between certain distance and the two o'clock gray scale of certain orientation by gray level co-occurrence matrixes function and closes
It is the texture information of target in Lai Fanying image.F1 is the energy of texture, has reacted image grayscale and has been evenly distributed degree and texture
Fineness degree;F2 is texture contrast, has reacted the clarity of image and the rill depth of texture;F3 is texture correlation, is used to
Measurement image gray level be expert at or column direction on similarity degree, reacted local gray level correlation.S (i, j) indicates (i, j)
First element size of normalization GLCM is calculated, N is image gray levels, and μ is gray value mean value, the standard deviation of σ.
(d) feature based on texture (FT) is extracted, and passes through d=1, the gray level co-occurrence matrixes of N=256, to filtered figure
As calculating for four angle value θ={ 0 °, 45 °, 90 °, 135 ° } internal signal J.Logical feature forms every in 4 angles
12 vector FTθ=(F1,F2,F3)θ。
FT={ FT0, FT45, FT90, FT135}
(e) Wrap algorithm is a kind of implementation of fast discrete Curvelet transformation, using Wrap algorithm to image into
The orientation for obtaining the different frequency domains of each subband of image different scale and position are decomposed in row Curvelet transformation.
The gray level image of M × N is converted by the fast discrete Curvelet of Wrap algorithm, generates one group of curve wave system number C,
It is scale a, direction b, spatial position p and q respectively.Here 0≤m≤M, 0≤n≤N,It is curve waveform.
The quantity of ratio is calculated according to following relationship:
A=ceil [log2{min(M,N)}-3]
Experiments have shown that effect is best when a=4, four Curvelet coefficients, four grades from 1 to 4 are generated at this time.All
In possible coefficient, only retains the coefficient that those energy are positive, sort from large to small, the first two coefficient (CC1 and CC2) and rank
The binary image of 1 coefficient (CC3) is formed together the element of shape eigenvectors.
(f) image shape feature in translation, scaling, rotation does not change, and bending moment (IM) is not an expression image
The important method of middle target shape feature invariant.
For digital picture scale be p × q, the square m of the pixel P (x, y) at position (x, y) be defined as pixel value and its
The product of coordinate distance, that is, m=xyP (x, y).
The square of whole image is the summation of the square of its all pixels.
mpq=ΣxΣyxpyqI(x,y)
(g) feature based on shape (FS) carries out quick Curvelet transformation to pretreated image by Wrap algorithm,
And calculate three curve ripple coefficients CC1, CC2, CC3.For each coefficient, the first not bending moment M is calculated1:
F4=M1(CC1), F5=M1(CC2), F6=M1(CC3)
For each image, shape feature is made of 3 element vectors comprising three square items:
FS={ F4, F5, F6}
(h) 12 textural characteristics of shape and 3 shape feature compositions compound characteristics (FC) are connected.
Textural characteristics (FT) are for capturing texture content, and shape feature (FS) is for modeling its shape, by 12
Texture vector connects 3 shape vectors to generate 15 element combinations vector F C.Element 1-3 is with 0 ° of capture texture content of angle, member
Plain 4-6 is caught with 45 ° of capture texture contents of angle, element 7-9 with 90 ° of capture texture contents of angle, element 9-12 with 135 ° of angle
Texture content is obtained, element 13-15 captures shape content, compound characteristics FC={ FT, FS }.
(i) neuro-fuzzy classifier combines the advantages of automatic adaptation process using Fuzzy Classifier and neural network,
Using the neuro-fuzzy classifier of conjugate gradient algorithms (NFC), all kinds of kinds of electronic components are identified.
Each electron-like element I by one group of n member image TI=t1, t2 ... .., tn composition, it is characterized in that in training
The set of the characteristic value obtained during stage.The feature of test image and certain training class coincidence factor highest, target image are identified
For certain kinds, while weight is adjusted in an iterative process, enable a system to learning data set feature in the following manner and inputting
Nonlinear Mapping is executed between output observation, the derivative model is then applied to the kind that unknown test sample identifies them
Class.Fuzzy Classifier is in such a case, it is possible to which the probability for providing sample belongs to several classes rather than exclusiveness classification.
The present invention can will have different texture, shape, and the electronic component of size and Orientation discrimination under complex environment is
92.32%, as shown in Figure 2.
All experiments are realized on PC computer.The parameter of computer are as follows: central processing unitCoreTM i5-
3210, memory 4GB, processing speed are 4 frames/second;Environment configurations: the python3.6+OpenCV+ under Windows10 operating system
TensorFlow。
Table 1:20 kind electronic component recognition accuracy
Electronic component title | Accuracy rate |
Chip-R | 95.634% |
Cement resistor | 93.482% |
Conventional, electric-resistance | 89.974% |
Gallbladder capacitor | 83.563% |
MPB capacitor | 94.683% |
Electrolytic capacitor | 88.394% |
Gallbladder patch capacitor | 98.382% |
PEI capacitor | 91.234% |
Chip inductor | 92.342% |
Diode | 93.414% |
Triode | 88.327% |
Touch key switch | 87.295% |
Microswitch | 91.457% |
Temperature switch | 86.375% |
Relay | 91.341% |
Thermal Cutoffs | 94.295% |
It is silicon-controlled | 96.234% |
SAW filter | 98.654% |
Bridge rectifier diode | 95.238% |
Air core inductor | 89.765% |
The series of detailed descriptions listed above only for feasible embodiment of the invention specifically
Protection scope bright, that they are not intended to limit the invention, it is all without departing from equivalent implementations made by technical spirit of the present invention
Or change should all be included in the protection scope of the present invention.
Claims (8)
1. a kind of know method for distinguishing based on the electronic component of texture and shape feature, which comprises the following steps:
Image preprocessing (PP): by color image gray processing, binaryzation, by the translation to target in image, rotation and scaling,
Target is normalized;
Gabor filter (GF): in conjunction with the thought of convolutional network, it is applied to the feature extraction of electronic component, Gabor filtering
Device and gray level image convolution generate complex signal, the characteristic information of electronic component are tentatively obtained, by the real part (rgg) and void of signal
(igg) is separated in portion, and the gray level co-occurrence matrixes of image are calculated by imaginary part, are further handled characteristic information;
Gray level co-occurrence matrixes (GLCM): electronic component is rigid structure, and texture is simple and symmetrical, and feature passes through gray level co-occurrence matrixes
Formula calculates, and obtains 3 feature representation F1、F2、F3, F1 is the energy of texture, and F2 is texture contrast, and F3 is texture correlation,
The primary expression texture information of rigid body;
The feature (FT) of texture is extracted: being used gray level co-occurrence matrixes, is believed according to the inside that the image of filtering calculates multiple angle values
The textural characteristics of number overlay electronic element reconnect characteristic value F1、F2、F3The texture feature vector of multiple elements is obtained, expression is made
For the electronic component textural characteristics of rigid body;
Second Generation Curvelet Transform (CT): using Wrap algorithm to electronic component image carry out Curvelet transformation, decompose into
And orientation and the position of electronic component image different scale difference frequency domain are effectively obtained, 20 kinds of electronic components are several in simple rule
What shape, takes four Curvelet coefficients, retains coefficient and rank binaryzation electronic component image that those energy are positive, together
Composition is in the feature vector element of the electronic component of regular geometric shapes;
Not bending moment (IM): image shape feature in translation, scaling, rotation does not change, by shape information not bending moment
(IM) expression is handled, and the square m of the pixel P (x, y) at position (x, y) is defined as the product i.e. m of pixel value and its coordinate distance
=xyP (x, y), graphical rule are p × q, and the square of whole image is the summation of the square of its all pixels: mpq=∑x∑yxpyqI(x,y);
The feature (FS) of shape is extracted: being obtained not bending moment, then is carried out quickly by Wrap algorithm to pretreated image
Curvelet transformation, and calculate three curve ripples coefficient CC1, CC2, CC3 calculates the first not bending moment M for each coefficient1;Its
Give expression to the shape of regular geometric rigid body.
Compound characteristics model (FC): textural characteristics (FT) for capturing texture content, shape feature (FS) be used for its shape into
Row modeling;
Classification and Identification: neuro-fuzzy classifier is combined using the excellent of the automatic adaptation process of Fuzzy Classifier and neural network
Point identifies electronic component using the neuro-fuzzy classifier of conjugate gradient algorithms (NFC).
2. according to claim 1 a kind of based on the electronic component of texture and shape feature knowledge method for distinguishing, feature exists
In described image pretreatment (PP) converts the image into gray scale and bi-level fashion, to guarantee target invariable rotary, target electronic member
Angle map of the main shaft of part relative to its orientation reduces background until electronic component is lucky to guarantee target translation invariant
It is suitble to bounding rectangles, to guarantee that target scale is constant, image is readjusted as standard size.
3. according to claim 1 a kind of based on the electronic component of texture and shape feature knowledge method for distinguishing, feature exists
In using Gabor filter (GF) and gray level image convolution, generation complex signal J obtains the preliminary information of feature, isolates J
Imaginary part, calculate the gray level co-occurrence matrixes (GLCM) of image, and calculate the feature for being based on gray level co-occurrence matrixes (GLCM).
4. according to claim 1 a kind of based on the electronic component of texture and shape feature knowledge method for distinguishing, feature exists
In, using gray level co-occurrence matrixes (GLCM) statistics electronic component textural characteristics, texture be by intensity profile on spatial position
Occur repeatedly and formed, thus there can be certain ash between two pixels for being separated by certain distance in electronic component image space
Degree relationship, electronic component itself belong to the rigid structure of simple symmetric, must reflect image using three feature representation parameters are complete
The texture information of middle target, F1 have reacted image grayscale and have been evenly distributed degree and texture fineness degree, and F2 has reacted the clear of image
Degree and texture the rill depth, F3 be used to measure image gray level be expert at or column direction on similarity degree.
5. according to claim 1 a kind of based on the electronic component of texture and shape feature knowledge method for distinguishing, feature exists
In the feature (FT) based on texture is extracted, and by d=1, the gray level co-occurrence matrixes of N=256 calculate four angles and correspond to electronics
The four direction of element believes four angle value θ={ 0 °, 45 °, 90 °, 135 ° } inside to what filtered image calculated
Number J calculates characteristic value F1, F2, F3 by gray level co-occurrence matrixes, and the final textural characteristics of each image are 3 feature shapes of connection
At 12 vectors in 4 angles, just expressed intact textural characteristics of electronic component.
6. according to claim 1 a kind of based on the electronic component of texture and shape feature knowledge method for distinguishing, feature exists
In using Wrap algorithm to image progress Curvelet transformation, decomposition obtains the different frequency domains of each subband of image different scale
Orientation and position, the gray level image of M × N converted by the fast discrete Curvelet of Wrap algorithm, generate one group of curve ripple
Coefficient C is scale a, direction b, space coordinate (p, q) respectively, and 0≤m≤M, 0≤n≤N are curve waveforms, generate at this time here
Four Curvelet coefficients, four grades from 1 to 4 only retain the coefficient that those energy are positive in all possible coefficient,
It sorts from large to small, the binary image of the first two coefficient (CC1 and CC2) and 1 coefficient of rank (CC3) is formed together shape feature
The element of vector.
7. according to claim 1 a kind of based on the electronic component of texture and shape feature knowledge method for distinguishing, feature exists
In in compound characteristics model (FC), by 12 texture vectors 3 shape vectors of connection to generate 15 element combinations vector F C.
8. according to claim 1 a kind of based on the electronic component of texture and shape feature knowledge method for distinguishing, feature exists
In, using neuro-fuzzy classifier, identify kinds of electronic components, the feature of test image and certain training class coincidence factor highest,
Target image is identified as certain kinds, while adjusting weight in an iterative process, enables a system to learn number in the following manner
According to collection feature output and input observation between execute Nonlinear Mapping, then by the derivative model be applied to unknown test specimens
Product identify their type.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910288579.9A CN110147570A (en) | 2019-04-11 | 2019-04-11 | It is a kind of that method for distinguishing is known based on the electronic component of texture and shape feature |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910288579.9A CN110147570A (en) | 2019-04-11 | 2019-04-11 | It is a kind of that method for distinguishing is known based on the electronic component of texture and shape feature |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110147570A true CN110147570A (en) | 2019-08-20 |
Family
ID=67589536
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910288579.9A Pending CN110147570A (en) | 2019-04-11 | 2019-04-11 | It is a kind of that method for distinguishing is known based on the electronic component of texture and shape feature |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110147570A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110490207A (en) * | 2019-08-27 | 2019-11-22 | 河北科技大学 | Bar section character picture recognition methods based on bianry image gray level co-occurrence matrixes |
CN110826514A (en) * | 2019-11-13 | 2020-02-21 | 国网青海省电力公司海东供电公司 | Construction site violation intelligent identification method based on deep learning |
CN111010545A (en) * | 2019-12-20 | 2020-04-14 | 深圳市中天安驰有限责任公司 | Vehicle driving decision method, system, terminal and storage medium |
CN116664529A (en) * | 2023-06-05 | 2023-08-29 | 青岛信驰电子科技有限公司 | Electronic element flat cable calibration method based on feature recognition |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105938563A (en) * | 2016-04-14 | 2016-09-14 | 北京工业大学 | Weld surface defect identification method based on image texture |
CN106446804A (en) * | 2016-09-08 | 2017-02-22 | 西安电子科技大学 | ELM-based multi-granularity iris recognition method |
-
2019
- 2019-04-11 CN CN201910288579.9A patent/CN110147570A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105938563A (en) * | 2016-04-14 | 2016-09-14 | 北京工业大学 | Weld surface defect identification method based on image texture |
CN106446804A (en) * | 2016-09-08 | 2017-02-22 | 西安电子科技大学 | ELM-based multi-granularity iris recognition method |
Non-Patent Citations (2)
Title |
---|
JYOTISMITACHAKI 等: "lant leaf recognition using texture and shape features with neural classifiers", 《PATTERN RECOGNITION LETTERS》 * |
恩德等: "基于集成神经网络的植物叶片识别方法", 《浙江农业学报》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110490207A (en) * | 2019-08-27 | 2019-11-22 | 河北科技大学 | Bar section character picture recognition methods based on bianry image gray level co-occurrence matrixes |
CN110490207B (en) * | 2019-08-27 | 2023-07-18 | 河北科技大学 | Bar end face character image recognition method based on binary image gray level co-occurrence matrix |
CN110826514A (en) * | 2019-11-13 | 2020-02-21 | 国网青海省电力公司海东供电公司 | Construction site violation intelligent identification method based on deep learning |
CN111010545A (en) * | 2019-12-20 | 2020-04-14 | 深圳市中天安驰有限责任公司 | Vehicle driving decision method, system, terminal and storage medium |
CN116664529A (en) * | 2023-06-05 | 2023-08-29 | 青岛信驰电子科技有限公司 | Electronic element flat cable calibration method based on feature recognition |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110147570A (en) | It is a kind of that method for distinguishing is known based on the electronic component of texture and shape feature | |
CN110443293B (en) | Zero sample image classification method for generating confrontation network text reconstruction based on double discrimination | |
CN110569905B (en) | Fine-grained image classification method based on generation of confrontation network and attention network | |
CN108734138A (en) | A kind of melanoma skin disease image classification method based on integrated study | |
CN106610969A (en) | Multimodal information-based video content auditing system and method | |
Abidin et al. | Copy-move image forgery detection using deep learning methods: a review | |
CN108520215B (en) | Single-sample face recognition method based on multi-scale joint feature encoder | |
CN108073940B (en) | Method for detecting 3D target example object in unstructured environment | |
CN106327534A (en) | Tire inner wall texture identification method based on locating block | |
CN113506239B (en) | Strip steel surface defect detection method based on cross-stage local network | |
CN111709313A (en) | Pedestrian re-identification method based on local and channel combination characteristics | |
Yang et al. | Dynamic fractal texture analysis for PolSAR land cover classification | |
CN108564130B (en) | Infrared target identification method based on monogenic features and multi-kernel learning | |
Khan et al. | Texture representation through overlapped multi-oriented tri-scale local binary pattern | |
CN110457996A (en) | Moving Objects in Video Sequences based on VGG-11 convolutional neural networks distorts evidence collecting method | |
CN112597798A (en) | Method for identifying authenticity of commodity by using neural network | |
Del Espiritu et al. | Neural network based partial fingerprint recognition as support for forensics | |
CN110443306A (en) | The classification method of grape wine stopper | |
CN114723953A (en) | Deep neural network for image source detection | |
Wiling | Locust Genetic Image Processing Classification Model-Based Brain Tumor Classification in MRI Images for Early Diagnosis | |
Jain | Industrial objects recognition in intelligent manufacturing for computer vision | |
Jiang et al. | MFL data feature extraction based on KPCA-BOMW model | |
Jain et al. | Unmanned machine vision system for automated recognition of mechanical parts | |
Asha et al. | Automatic detection of defects on periodically patterned textures | |
Sossa et al. | Modified dendrite morphological neural network applied to 3D object recognition on RGB-D data |
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: 20190820 |