CN108961233A - A kind of composite polycrystal-diamond surface defect classifying identification method - Google Patents
A kind of composite polycrystal-diamond surface defect classifying identification method Download PDFInfo
- Publication number
- CN108961233A CN108961233A CN201810691154.8A CN201810691154A CN108961233A CN 108961233 A CN108961233 A CN 108961233A CN 201810691154 A CN201810691154 A CN 201810691154A CN 108961233 A CN108961233 A CN 108961233A
- Authority
- CN
- China
- Prior art keywords
- classification
- defect
- composite polycrystal
- identification method
- kernel function
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30132—Masonry; Concrete
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30168—Image quality inspection
Abstract
The invention discloses a kind of composite polycrystal-diamond surface defect classifying identification methods, the problems such as present invention solves artificial detection subjectivity strong, low efficiency, the accuracy of classification results and reliability can not ensure is, it can be achieved that composite polycrystal-diamond face crack, white point, three kinds of defect high efficiency of white edge, high-precision, the Classification and Identification of automation.
Description
Technical field
The present invention relates to a kind of composite polycrystal-diamond surface defect classifying identification methods.
Background technique
Composite polycrystal-diamond (polycrystalline diamond compact, hereinafter referred to as PDC) is by plycrystalline diamond
Diamond and hard alloy are sintered at high temperature under high pressure, because its excellent performance is widely used in oil exploration, machinery
The numerous areas such as processing.In the process of production and processing, side polycrystalline diamond layer (polycrystalline diamond, below
Abbreviation PCD) surface region is inevitably cracked, white point and the defects of white edge, properties of product are impacted.Cause
This, must carry out stringent defects detection and classification before completing product packaging to finished surface.Traditionally, PDC factory
The most of mode using Manual Visual Inspection of family carries out the identification and classification of defect, because defect information protrusion is unobvious, needs long-term
It is observed operation under strong light, human eye is injured larger.And that there is also detection accuracy is low, subjectivity is strong, effect for artificial detection
The problems such as rate is low.
Currently, introducing the defect detecting technique based on machine vision in production practice both at home and abroad, have non-contact, low
The advantages that cost, high degree of automation, can overcome some disadvantages of artificial detection.Jia etc. is made using radial base (RBF) kernel function
It is the kernel function of support vector machines (SVM) classifier the Nonlinear Classification that solves the problems, such as sample, and penalty factor and core is joined
Number is selected, and the rift defect identification of steel plate is realized.Choose six kinds of defects of cold rolling steel surface first 46 such as Choi are several
What feature and 8 gray levels preferably go out RBF kernel function, then as characteristic of division from linear, multinomial and RBF kernel function
Parameter optimization is carried out using grid data service, recognition accuracy reaches 87%-94%.Yellow will letter et al. extracts 6 kinds of gray variance etc.
Beer bottleneck defect characteristic constitutes SVM input vector, kernel function of the RBF as SVM classifier is preferentially chosen, then according to defect
Taxonomic property solves more classification problems using one-to-many classification, and finally six kinds of defects of bottleneck are reached with 91.6% classification standard
True rate.Wang Peng etc. is directed to the feature of image of the defects of bullet surface green statin, crack, carries out in terms of geometry, gray scale and texture three
Defect characteristic extracts, and establishes the bullet open defect disaggregated model based on support vector machines, and recognition accuracy is up to 98.9%.
The defect characteristic of different test objects is different, and the classification method of use is also different.Therefore, herein by right
Common crackle, white point, white edge defect are studied in PDC production process, find the Classification and Identification side for being suitble to tri- kinds of defects of PDC
Method, to realize the Accurate classification of three kinds of defects.
Summary of the invention
It is an object of the invention to provide a kind of composite polycrystal-diamond surface in place of overcome the deficiencies in the prior art
Defect classifying identification method solves the problems in above-mentioned background technique.
The technical solution adopted by the present invention to solve the technical problems is: providing a kind of composite polycrystal-diamond surface
Defect classifying identification method, includes the following steps:
(1) it is directed to composite polycrystal-diamond face crack, three kinds of white point, white edge common deficiency types, establishes defect spy
Levy model;
(2) model established according to step (1) extracts input vector of the defect characteristic amount as classifier, the defect
Characteristic quantity includes the provincial characteristics and shape feature of defect;
(3) according to sample data distribution situation, construct sorter model solve sample data more classification and it is non-linear can
Divide problem, and sorter model parameter is selected, the parameter penalty factor after selection is 1.3195, and nuclear parameter γ is
2.2974。
(4) sorter model obtained through step (3) is trained and is tested to defect sample using three kinds of kernel functions,
The number of supporting vector needed for choosing is minimum, the classification time is most short and the highest polynomial function of Classification and Identification rate is as classifier
The kernel function of model.
In a preferred embodiment of the present invention, the provincial characteristics of the defect includes area, perimeter, length, width, shape
Feature includes solid degree, circularity and length-width ratio.
In a preferred embodiment of the present invention, the classifier of the step (2) includes support vector machines (SVM) classifier.
In a preferred embodiment of the present invention, step (3) the building multi-categorizer solves more classification problems.
In a preferred embodiment of the present invention, the step (3) introduces slack variable and solves the problems, such as linearly inseparable, introduces
Kernel function solves the problems, such as Nonlinear separability.
In a preferred embodiment of the present invention, the step (3) uses grid search and the preferred SVM of cross validation combined method
Model parameter.
In a preferred embodiment of the present invention, three seed nucleus functional forms of the step (4) are (1) linear kernel function: K
(x1,x2)=< x1,x2>;(2) Polynomial kernel function: K (x1,x2)=(γ < x1,x2>+1)d, d is the rank of Polynomial kernel function
Number;(3) RBF kernel function K (x1,x2)=exp (- γ | | x1,x2||2)。
The technical program compared with the background art, it has the following advantages:
The present invention is directed to the characteristics of PDC face crack, three kinds of white point, white edge defect images, establishes three kinds of defect models,
And seven characteristic quantities such as area are extracted according to defect area and shape information, it can preferably reflect the spatial distribution of PDC defect,
With preferable representativeness;Respectively by building multi-categorizer, it is introduced into slack variable and kernel function solves in sample data more points
The problem of class and Nonlinear Classification, and the optimizing of SVM model parameter is carried out using the method that cross validation is combined with grid search;
The optimal Polynomial kernel function of classifying quality has been selected by comparative analysis, system is made to have reached 99% Detection accuracy.Therefore
SVM classifier designed by the present invention is to three kinds of the surface PDC common deficiency Classification and Identification rate with higher and preferable practical
Property.
Detailed description of the invention
Fig. 1 is three kinds of surface defect facet models established by the present invention.
Fig. 2 is the three kinds of defect original images detected needed for the present invention, wherein (a) is crackle, (b) is white point, (c) is white
Side.
Fig. 3 is two-dimensional spatial distribution figure of the sample of the present invention data under seven kinds of characteristic quantities.
Fig. 4 is that the present invention randomly selects crackle and each 20 samples length-width ratio data characteristics amount comparison diagram of white edge.
Fig. 5 is that the present invention uses kernel function that lower dimensional space data are mapped to higher dimensional space schematic diagram.
Fig. 6 is that grid search of the present invention and cross validation carry out parameter preferred result figure.
Specific embodiment
The contents of the present invention are illustrated with reference to the accompanying drawing:
In the process of production and processing, common surface defect mainly has white point, crackle and white edge to composite polycrystal-diamond
Deng, but three kinds of defects hardly occur simultaneously.White point appears at interface cohesion greatly, and white is dotted, not of uniform size;Crackle
Mostly originate in PCD chamfering bottom edge, to interface cohesion prescription to extension, aterrimus elongated shape is different in size;White edge mostly occurs
Below PCD layer chamfered area, white long strip type, width is different.For convenient for being compared analysis, root of the present invention to three kinds of defects
Defect characteristic model is established according to generation process, position, size and the shape feature of three kinds of defects, and three kinds of defects unifications are drawn
In finished product side view, as shown in Figure 1.
The defect model (shown in Fig. 1) established through the invention is it is known that three kinds of defects are both present in PCD layer, and lack
It falls into area size and shape feature gap is larger, digital quantity will be converted by defect information by image procossing in next step, and right
Defect characteristic amount extracts.
It is learnt according to the analysis to defect model, the area size and shape feature of three kinds of defects are inconsistent.Therefore, it is
There is feature vector to defect characteristic preferable descriptive, according to the provincial characteristics of defect, chooses area A, perimeter P, length
Spend tetra- characteristic quantities of L and width W.According to the shape feature of defect, tri- solid degree R, circularity C and length-width ratio E features are chosen
Amount.Wherein, solid degree reflects the departure degree of defect area Yu its boundary rectangle, by calculating target area face to be identified
The long-pending ratio with its minimum circumscribed rectangle area obtains, and specific formula isCircularity is used to measure figure and is biased to circle
The degree of shape, specific formula are
Seven defect characteristic amount interactions, the present invention are incorporated herein support vector machine classifier and carry out machine learning realization
Automatic classification, input vector of seven defect characteristic amounts as classifier.Support vector machines is initially linear for solving two classes
The problem of classification, for given training sample:
{(x1,y1),(x2,y2),…(xl,yl)},
x∈Rn,y∈{+1,-1}
Wherein, l is sample number, and y indicates classification, and n is input dimension.Classifying for task is exactly that a hyperplane is looked for make
These two types of samples are completely separable so that data point corresponding y in hyperplane one side is+1 entirely, data point corresponding y in one side is all-
1, this hyperplane can indicate are as follows:<W, x>+b=0.In formula, W is the normal vector of classification line,<,>indicate interior Product function.
If training pattern set can correctly be separated by hyperplane, and the mode data nearest apart from hyperplane and super flat
The distance between face maximum, as optimal hyperlane.A constrained optimization problem can be finally expressed as:
Wherein, s.t. indicates constraint condition, | | W | | embody confidence risk.
Initial data is by 30 crackles, 300 white points, 120 white edges, and totally 450 sample compositions, every after feature extraction
A sample contains 7 characteristic quantities.In order to intuitively analyze sample data, all defect sample is made in seven kinds of characteristic quantities
Distribution situation in lower two-dimensional space, as shown in Figure 3.Sample data totally three class labels, wherein crackle class label is 1,
White point class label is 2, and white edge class label is 3.
As can be seen that the linear separability of data is different.Wherein, the linear separability of circularity is best, belongs to line
Property problem;When the corresponding sample data of four characteristic quantities of length is not linear separability for area, perimeter, length and width, but inseparable
Data it is less, belong to approximately linear separable problem;Solid degree and width characteristics amount, it is poor to the separability of three kinds of defects, belong to
In fairly linear inseparable problem.In order to realize the Accurate classification of defect, need to solve more classification and the linearly inseparable of sample
Problem.
Classification and Identification is carried out to tri- kinds of defects of PDC, belongs to multi-class problem.SVM algorithm is initially linear for solving two classes
The problem of classification, then, the characteristic quantity of linear separability this kind of for circularity can be by combining multiple two classifiers (i.e. above-mentioned branch
Hold vector machine classifier) carry out structural classification device, common method has one-to-many and two kinds one-to-one.
One-against-rest is the Multiclass SVM method used earliest, classification problems more for K class, it needs to train K two classes
Support vector machine classifier.When constructing i-th of support vector machines sub-classifier, the sample data for belonging to the i-th classification is marked
Be positive class, and the sample data for being not belonging to i classification marks the class that is negative.When test, each sub-classifier is calculated separately to test data
Decision function value, and unknown sample is classified as that class with maximum classification function value.The number of required support vector machines
It is less, but each support vector machines sub-classifier requires all samples and participates in, and the training time is long, and data is easy to cause to incline
Tiltedly.
One-against-one is one classifier of training between every two class, and therefore, classification problems more for K class, one kind is to one
Class method needs to construct K (K-1)/2 classifier.In the SVM sub-classifier of structure classes i and classification j, in sample data set
The sample data that selection belongs to classification i, classification j is positive as training sample data, and by the data markers for belonging to classification i, will
The data markers for belonging to classification j are negative.When classification, the classification that classification samples are treated by the way of " ballot " is judged, most
Afterwards using who gets the most votes's classification as the classification of sample to be sorted.Each sub-classifier need to only train two class samples, so classification
Speed is fast compared with one-against-rest.For the not high classification problem of classification number, nicety of grading is higher, the training time is short.Therefore, this hair
It is bright that one-to-one method is selected to construct multi-categorizer.
For not being linear separability, but the sample data that inseparable data are seldom, the Optimal Separating Hyperplane established is just
There can be certain error in classification.For data extracted in the present invention, can exist between different classes of sample data and hand over
Fork accidentally divides as shown in figure 4, can exist at this time.It is separated to maximize optimal hyperlane by different classes of data, and makes it
Minimum is reached to the average error in classification of entire training set, the present invention solves this problem by introducing non-negative slack variable ξ.
Slack variable ξ is departure degree of the metric data point to ideal position.If slack variable ξi=0, illustrate sample number
According to there is no outliers;If 0 < ξi≤ 1, data point falls in the inside of separated region, in the classification correct side of plane;If ξi
> 1, data point fall in the side of classification plane mistake, and value is bigger, and point just peels off remoter.
The optimization restricted problem of data classification after introducing slack variable are as follows:
Above-mentioned optimization problem is the quadratic programming objective function problem with inequality constraints.Wherein,Embody experience
Risk, C are known as penalty factor, are demonstrated by the degree for paying attention to isolated point, are the balances to empiric risk and confidence risk.C is bigger
Show more to pay attention to loss brought by outlier, but excessively high C will lead to overfitting, cause problem without solution.Therefore, to optimization
When problem solving, needs to find a preferable parameter C and make the Generalization Ability of SVM best.
It is solid degree and two characteristic quantities of width corresponding to sample data belong to it is fairly linear can not divided data.At this moment, originally
Invention is by introducing kernel function K (x1,x2) fairly linear inseparable to solve the problems, such as, the main thought of support vector machines is will be low
Dimension space data are mapped in higher dimensional space by kernel function, realize the linear separability of data, at this time algorithm meter with higher
Calculate complexity.After introducing kernel function, detailed process is first to compare vector (asking inner product or certain distance) in the low-dimensional input space, so
Nonlinear transformation is remake to result afterwards.In this way, a large amount of operation will be in original input space rather than in high-dimensional feature space
It carries out, reduces the complexity of algorithm, mapping process is as shown in Figure 5.
Three common seed nucleus functional forms are as follows: (1) linear kernel function: K (x1,x2)=< x1,x2>;(2) polynomial kernel letter
Number: K (x1,x2)=(γ < x1,x2>+1)d, d is the order of Polynomial kernel function;(3) RBF kernel function K (x1,x2)=exp (-
γ||x1,x2||2).Wherein, nuclear parameter γ > 0.The performance superiority and inferiority of kernel function is directly influenced by parameter γ, and γ is too small to be easy to make
At " overfitting ", γ is excessive, is easy to cause " owing study ".Therefore, it when using kernel function, needs to carry out parameter γ preferred.
It can be seen from the above, the selection of punishment parameter C and nuclear parameter γ are particularly significant in non-linear more classification problems, choosing
The key for selecting optimized parameter is that finding optimal (C, γ) combines, and common optimization method has genetic algorithm, particle group optimizing
Algorithm, grid data service etc..First two algorithm complexity is higher, and is affected by initial value.Grid search is with higher
Learn the advantage that parallel processing was simply easily realized and had to precision, algorithm.Therefore, the present invention is carried out using the method for grid search
Parameter optimization.Overfitting problem is likely to occur in data searching process, and cross validation is for verifying classifier performance
A kind of statistical analysis technique, and the classification accuracy to obtain can evade search process as the performance indicator of classification of assessment device
Fall into over-fitting.Therefore, the present invention carries out parameter optimization using the method that cross validation and grid search combine.
Having main steps that makes C and γ in a certain range, takes different values to carry out cross validation, is searched by grid
Rope repeats k times, estimates it is expected extensive error after k iteration according to the mean square error average value obtained, last selection sort is just
Highest one group of true rate is used as optimized parameter.It is selected roughly first when grid search, selectes the range of one group of parameter, such as
C={ 2-10,2-9,…29,210, γ={ 2-10,2-9,…29,210, according to the result selected roughly select an accuracy rate compared with
High lesser region of search, then finely selected.
Randomly select sample 1/3 is used as training set, constantly looks for the lower instruction of different (C, γ) combination using grid data service
Practice sample classification accuracy rate as a result, as shown in Figure 6.Wherein, the classification accuracy of training set is up to 100%, is corresponded at this time
Parameter C be 1.3195, γ 2.2974, this group of optimal parameter searched is put into SVM and is trained.
SVM model is established using the parameter (C, γ) after one-to-one more classification and optimization, model formation at this time are as follows:
Wherein, kernel function K (x1,x2) it include parameter γ.Further, 150 are randomly selected from 450 sample datas
It is a be used as training sample, remaining 300 be used as test sample, SVM model is trained by training set, then to test set into
Row class prediction.
Test set is tested respectively under three kinds of linear function, polynomial function, radial basis function kernel functions, is used
Optimal parameter, test result are as shown in table 1.
Test set classification results under 1 three kinds of kernel functions of table
Under three kinds of kernel functions, the number of supporting vector needed for SVM model when polynomial function is as kernel function is at least
11, it is 0.0463s that the classification time is most short, classification accuracy reaches 99%.Final choice polynomial function is carried out as kernel function
Defect classification.
Divide quickly, accurately, automatically it can be seen that the present invention can realize three kinds of PDC face crack, white point, white edge defects
Class.
The above is only the preferred embodiment of the present invention, the range implemented of the present invention that therefore, it cannot be limited according to, i.e., according to
Equivalent changes and modifications made by the invention patent range and description, should still be within the scope of the present invention.
Claims (7)
1. a kind of composite polycrystal-diamond surface defect classifying identification method, which comprises the steps of:
(1) it is directed to composite polycrystal-diamond face crack, three kinds of white point, white edge common deficiency types, establishes defect characteristic mould
Type;
(2) model established according to step (1) extracts input vector of the defect characteristic amount as classifier, the defect characteristic
Amount includes the provincial characteristics and shape feature of defect;
(3) according to sample data distribution situation, the more classification and Nonlinear separability for constructing sorter model solution sample data are asked
Topic, and sorter model parameter is selected, the parameter penalty factor after selection is 1.3195, and nuclear parameter γ is 2.2974.
(4) sorter model obtained through step (3) is trained and is tested to defect sample using three kinds of kernel functions, is chosen
The number of required supporting vector is minimum, the classification time is most short and the highest polynomial function of Classification and Identification rate is as sorter model
Kernel function.
2. a kind of composite polycrystal-diamond surface defect classifying identification method according to claim 1, it is characterised in that:
The provincial characteristics of the defect includes area, perimeter, length, width, and shape feature includes solid degree, circularity and length-width ratio.
3. a kind of composite polycrystal-diamond surface defect classifying identification method according to claim 1, it is characterised in that:
The classifier of the step (2) includes support vector machines (SVM) classifier.
4. a kind of composite polycrystal-diamond surface defect classifying identification method according to claim 1, it is characterised in that:
Step (3) the building multi-categorizer solves more classification problems.
5. a kind of composite polycrystal-diamond surface defect classifying identification method according to claim 1, it is characterised in that:
The step (3) introduces slack variable and kernel function solves the problems, such as Nonlinear separability.
6. a kind of composite polycrystal-diamond surface defect classifying identification method according to claim 1, it is characterised in that:
The step (3) uses grid search and the preferred SVM model parameter of cross validation combined method.
7. a kind of composite polycrystal-diamond surface defect classifying identification method according to claim 1, it is characterised in that:
Three seed nucleus functional forms of the step (4) are (1) linear kernel function: K (x1,x2)=< x1,x2>;(2) Polynomial kernel function: K
(x1,x2)=(γ < x1,x2>+1)d, d is the order of Polynomial kernel function;(3) RBF kernel function K (x1,x2)=exp (- γ | | x1,
x2||2)。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810691154.8A CN108961233A (en) | 2018-06-28 | 2018-06-28 | A kind of composite polycrystal-diamond surface defect classifying identification method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810691154.8A CN108961233A (en) | 2018-06-28 | 2018-06-28 | A kind of composite polycrystal-diamond surface defect classifying identification method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108961233A true CN108961233A (en) | 2018-12-07 |
Family
ID=64487899
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810691154.8A Pending CN108961233A (en) | 2018-06-28 | 2018-06-28 | A kind of composite polycrystal-diamond surface defect classifying identification method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108961233A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110378433A (en) * | 2019-07-24 | 2019-10-25 | 重庆大学 | The classifying identification method of bridge cable surface defect based on PSO-SVM |
CN116883394A (en) * | 2023-09-06 | 2023-10-13 | 山东融泽新材料有限公司 | Diamond quality detection method based on image data processing |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20050019865A1 (en) * | 2003-06-27 | 2005-01-27 | Kihm Anthony J. | Cartilage and bone repair and regeneration using postpartum-derived cells |
CN103198322A (en) * | 2013-01-18 | 2013-07-10 | 江南大学 | Magnetic tile surface defect feature extraction and defect classification method based on machine vision |
CN107369136A (en) * | 2017-06-22 | 2017-11-21 | 福建省万龙新材料科技有限公司 | Composite polycrystal-diamond face crack visible detection method |
CN107389701A (en) * | 2017-08-22 | 2017-11-24 | 西北工业大学 | A kind of PCB visual defects automatic checkout system and method based on image |
-
2018
- 2018-06-28 CN CN201810691154.8A patent/CN108961233A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20050019865A1 (en) * | 2003-06-27 | 2005-01-27 | Kihm Anthony J. | Cartilage and bone repair and regeneration using postpartum-derived cells |
CN103198322A (en) * | 2013-01-18 | 2013-07-10 | 江南大学 | Magnetic tile surface defect feature extraction and defect classification method based on machine vision |
CN107369136A (en) * | 2017-06-22 | 2017-11-21 | 福建省万龙新材料科技有限公司 | Composite polycrystal-diamond face crack visible detection method |
CN107389701A (en) * | 2017-08-22 | 2017-11-24 | 西北工业大学 | A kind of PCB visual defects automatic checkout system and method based on image |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110378433A (en) * | 2019-07-24 | 2019-10-25 | 重庆大学 | The classifying identification method of bridge cable surface defect based on PSO-SVM |
CN116883394A (en) * | 2023-09-06 | 2023-10-13 | 山东融泽新材料有限公司 | Diamond quality detection method based on image data processing |
CN116883394B (en) * | 2023-09-06 | 2023-11-17 | 山东融泽新材料有限公司 | Diamond quality detection method based on image data processing |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106371427B (en) | Industrial process Fault Classification based on analytic hierarchy process (AHP) and fuzzy Fusion | |
US9384603B2 (en) | Failure cause classification apparatus | |
JP2022087297A (en) | Apparatuses, methods, and systems for image-based human embryo cell classification | |
CN101915769B (en) | Automatic optical inspection method for printed circuit board comprising resistance element | |
CN109952614A (en) | The categorizing system and method for biomone | |
CN109273096B (en) | Medicine risk grading evaluation method based on machine learning | |
CN109523518B (en) | Tire X-ray defect detection method | |
CN110503245B (en) | Prediction method for large-area delay risk of airport flight | |
CN111612784A (en) | Steel plate surface defect detection method based on classification-first YOLO network | |
US20080168011A1 (en) | Methods and systems for automatic selection of classification and regression trees | |
CN108711148B (en) | Tire defect intelligent detection method based on deep learning | |
CN111540006B (en) | Plant stomata intelligent detection and identification method and system based on deep migration learning | |
CN101751666A (en) | Semi-supervised multi-spectral remote sensing image segmentation method based on spectral clustering | |
CN103632168A (en) | Classifier integration method for machine learning | |
CN112102226A (en) | Data processing method, pattern detection method and wafer defect pattern detection method | |
CN111160401A (en) | Abnormal electricity utilization judging method based on mean shift and XGboost | |
CN104766097A (en) | Aluminum plate surface defect classification method based on BP neural network and support vector machine | |
CN113487533B (en) | Part assembly quality digital detection system and method based on machine learning | |
CN101980251A (en) | Remote sensing classification method for binary tree multi-category support vector machines | |
CN105740914A (en) | Vehicle license plate identification method and system based on neighboring multi-classifier combination | |
CN108961233A (en) | A kind of composite polycrystal-diamond surface defect classifying identification method | |
CN110059713A (en) | Precipitation type identification method based on precipitation particle multi-feature parameters | |
CN108241925A (en) | A kind of discrete manufacture mechanical product quality source tracing method based on outlier detection | |
CN109003266A (en) | A method of based on fuzzy clustering statistical picture quality subjective evaluation result | |
CN112819821A (en) | Cell nucleus image detection method |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20181207 |
|
WD01 | Invention patent application deemed withdrawn after publication |