CN110135477A - A kind of strip surface quality defect classifier and its classification method based on serial/parallel row integrated study frame - Google Patents
A kind of strip surface quality defect classifier and its classification method based on serial/parallel row integrated study frame Download PDFInfo
- Publication number
- CN110135477A CN110135477A CN201910351038.6A CN201910351038A CN110135477A CN 110135477 A CN110135477 A CN 110135477A CN 201910351038 A CN201910351038 A CN 201910351038A CN 110135477 A CN110135477 A CN 110135477A
- Authority
- CN
- China
- Prior art keywords
- defect
- classification
- serial
- classifier
- parallel row
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- 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
- G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
Abstract
The invention proposes a kind of strip surface quality defect classifier and its classification method based on serial/parallel row integrated study frame, specifically include: a kind of classifier structure based on serial/parallel row integrated study frame, merge with the defect that the classifier structure is adapted and stage division and mischief rule table construction method.Compared with traditional technology, the invention has the following advantages that classifier based on serial/parallel row integrated study frame change it is common " multiple classifiers are simply parallel; then final classification result is determined by Nearest Neighbor with Weighted Voting " thinking, traditional parallel sorting device structure of modification at unique serial/parallel row integrated morphology.By sufficiently merging newest machine learning classification identification technology, defect mechanism of production and artificial experience rule, it is aided with the defect being adapted with classifier structure to merge and stage division, the accuracy of strip surface quality defect Classification and Identification is further improved, and improves self-learning capability, real-time performance and the generalization ability of classifier.
Description
Technical field
The invention belongs to strip surface quality defects detections and sorting technique field, are related to computer vision and engineering
Learning method is related specifically to a kind of strip surface quality defect classification method based on serial/parallel row integrated study frame.
Background technique
Quality is the soul of modern enterprise, is the foundation stone that enterprise depends on for existence and development.
In the mainstream industry class of multiple national economy such as steel, papermaking, weaving, printing and dyeing and glass, product surface matter
The superiority and inferiority of amount will directly affect the final quality of product, performance and value, simultaneously as downstream client is to product surface quality
It is required that more and more harsh, good surface quality becomes the important magic weapon that enterprise wins in fierce market competition, thus,
There is extensive product surface qualities to detect demand in these industries.
The present invention relates to one of major products of steel and iron industry --- and the surface quality defect detection classification of strip is asked
Topic.It is traditionally general using artificial range estimation or offline sampling observation or the two in order to accurately identify strip surface quality defect
The mode combined.But with the development of steel and iron industry production level, traditional detection mode obviously has been unable to meet the modern times
The requirement of industrial production high speed, high yield and high-quality is unable to get the surface quality detection result of continuous, stable multiplicity.Machine
The extensive use of device vision technique and flourishing for artificial intelligence theory provide for surface quality detection problem in every profession and trade
Brand-new thinking and solution route.Also it is based on this, has studied the problem gradually using machine vision and artificial intelligence approach
Hot spot as domestic and foreign scholars, academic institution and business research.
Summary of the invention
The classification of strip surface quality defect is that multiple features under the conditions of a complicated production, multi-class pattern-recognition are asked
Topic.The accuracy and stability that improve Classification and Identification are the key scientific problems that classifier design is faced.Existing more classification
Device combined method often " multiple classifiers are simply parallel, and then Nearest Neighbor with Weighted Voting determines classification results ", it is no it is any pretreatment and
The link of centre intervention.Not high for such as classification accuracy present in traditional Combination of Multiple Classifiers method, classifier is learnt by oneself
The problems such as habit ability and generalization ability are weak, real-time performance is bad, the invention proposes one kind to be based on serial/parallel row integrated study frame
Strip surface quality defect classification method.
The technical solution adopted by the present invention includes following technology contents:
A kind of strip surface quality defect classifier based on serial/parallel row integrated study frame, which is characterized in that classifier
It include rule list and Machine learning classifiers for the three layers of serial structure divided from top to bottom, specifically:
Upper layer is rule list,
Middle layer is multiple parallel Machine learning classifiers,
Lower layer is rule list or single Machine learning classifiers.
Rule list is presented as a kind of bivariate table in structure, specific comprising two column, be respectively serial number and to it is regular it is specific in
The description of appearance.The line number of rule list is determined by regular quantity.
In a kind of above-mentioned strip surface quality defect classifier based on serial/parallel row integrated study frame, rule list is
The process mechanism that binding deficient generates, and the experience and knowledge of live expert is used for reference and absorbs, the classification formed after inverted
Decision rule.In selecting when multiple collateral learning classifiers of interbed, focus on their mutual complementarity, rule list
Construction step includes:
S201 finds a value range feature Chong Die with other all categories zero, if having found such feature,
Then this feature is added in the rule list of upper layer, is the defect of upper layer level by the defect classification that this feature is distinguished.
S202 finds a value range and other Chong Die features of particular category (not all classification) zero, if found
Then this feature is added in lower layer's rule list for such feature, is lacking for lower layer level by the defect classification that this feature is distinguished
It falls into.
S203 executes traversal than choosing to all features, until terminating.
In a kind of above-mentioned strip surface quality defect classifier based on serial/parallel row integrated study frame, middle layer dress
With random forest grader, self organizing neural network and the three kinds of classifiers of GentlBoost classifier encoded based on ECOC.This
Invention proposes the defect classifier based on serial/parallel row integrated study frame, is exactly in order to fully integrate existing machine learning point
Class identification technology, defect mechanism of production and artificial experience rule, further increase accuracy and the classifier of Classification and Identification
Self-learning capability, real-time performance and generalization ability.
A kind of classification method of the defect classifier based on serial/parallel row integrated study frame, which is characterized in that including defect
Merging and stage division, the specific steps are as follows:
S1 is isolated as the defect for being included in upper layer rule list the defect of upper layer level;
S2 compares the characteristic value of two kinds of defect sample in training set, if two kinds of defect classification sequence numbers are respectively yiAnd yj, instruction
Practicing the corresponding sample size of concentration is respectively ciAnd cj, characteristic value dimension is n dimension, and defect characteristic value sequence is respectively (xi1,
xi2,…,xik,…,xin) and (xj1,xj2,…,xjk,…,xjn), wherein xikAnd xjkIn respectively two kinds of defect characteristic value sequences
Component, calculate the mean value of two kinds of defect characteristic value respective components, be denoted as:WithK=1 ..., n,
Two mean bias percentages are further calculated, as follows:
Compare the respective components mean bias percentage and preset two threshold value δ1And δ2Size relation.If should
Respective components mean bias percentage is less than threshold value δ1Eigenvalue components number account for all characteristic value dimensions percentage it is big
In threshold value δ2, then corresponding two kinds of defects are merged.
The present invention is by threshold value δ1It is taken as 70%, threshold value δ2It is taken as 90%.
S3, if before not merging in the confusion matrix of two class defect classification results, mutual mistake divides rate to be greater than threshold
Value δ3, then also two class defects are merged.
The present invention is by threshold value δ3It is taken as 20%.
S4, the defect after merging are placed on middle layer level.
S5, other unassorted defects are also placed in middle layer level, i.e., other unassorted defects with merge after
Defect is placed on same level, and with the defect classifier middle layer proposed by the present invention based on serial/parallel row integrated study frame
Multiple parallel machine learning classification methods realize classification.
The invention has the following advantages that the classifier based on serial/parallel row integrated study frame changes commonly " by multiple points
Class device is simply parallel, then determines final classification result by Nearest Neighbor with Weighted Voting " thinking, traditional parallel sorting device structure is changed
Cause unique serial/parallel row integrated morphology.By sufficiently merging newest machine learning classification identification technology, defect mechanism of production
With artificial experience rule, it is aided with the defect being adapted with classifier structure and merges and stage division, further improve strip table
The accuracy of face mass defect Classification and Identification, and improve self-learning capability, real-time performance and the generalization ability of classifier.
Detailed description of the invention
Fig. 1 is the application defect classification method based on serial/parallel row integrated study frame of the invention mentioned to belt steel surface
The flow diagram that mass defect is classified.
A kind of strip surface quality defect classifier based on serial/parallel row integrated study frame that Fig. 2 is mentioned for the present invention
Structure chart.
Fig. 3 a is the five kinds of different types of typical defect pictures applied in the embodiment of the present invention two, and defect type is weldering
Seam, Fig. 3 b are the five kinds of different types of typical defect pictures applied in the embodiment of the present invention two, and defect type is hole;
Fig. 3 c is the five kinds of different types of typical defect pictures applied in the embodiment of the present invention two, and defect type is to draw
Wound;
Fig. 3 d is the five kinds of different types of typical defect pictures applied in the embodiment of the present invention two, and defect type is roller
Print;
Fig. 3 e is the five kinds of different types of typical defect pictures applied in the embodiment of the present invention two, and defect type is oil
Spot.
Specific embodiment
In order to enable those skilled in the art to better understand the solution of the present invention, with reference to the accompanying drawings and detailed description
The present invention is described in further detail.
Embodiment one
Fig. 1 is the application defect classification method based on serial/parallel row integrated study frame of the invention mentioned to belt steel surface
The flow diagram that mass defect is classified.
As shown in Figure 1, detailed process includes 3 key steps, it is as follows respectively:
Step 1, specific, the significant defect of feature difference is separated from all defect classification, with first layer
Regular table sort;
Step 2, then according to process knowledges such as the formation mechenisms of defect, with defect merging method by appearance or feature
It is worth close/similar defect to merge, Machine learning classifiers are only trained and classify to the defect classification after merging;
Step 3, the classification before merging is finely divided by the rule list of third layer and classifier.
Specifically, the construction method of above-mentioned first layer and the rule list of third layer is as follows:
Firstly, the value range feature Chong Die with other all categories zero is found, if having found such feature,
Then this feature is added in the rule list of upper layer, is the defect of upper layer level by the defect classification that this feature is distinguished.
Then, a value range and other Chong Die features of particular category (not all classification) zero are found, if found
Then this feature is added in lower layer's rule list for such feature, is lacking for lower layer level by the defect classification that this feature is distinguished
It falls into.
It should be noted that traversal need to be executed to all features than choosing, until terminating.
A kind of strip surface quality defect classifier based on serial/parallel row integrated study frame that Fig. 2 is mentioned for the present invention
Structure chart.
As shown in Fig. 2, the defect classifier based on serial/parallel row integrated study frame includes rule list and machine learning classification
Two kinds of basic building blocks of device, are generally represented as a kind of serial structure, are divided into three layers from top to bottom, upper layer is rule list, middle layer
It is multiple parallel Machine learning classifiers, lower layer is also rule list or single Machine learning classifiers.In addition to being embodied in structure
Except serial/parallel row integrated study frame, the fusion to rule list is also one of the technical characteristic of the classifier.Rule list is knot
The process mechanism that defect generates is closed, and uses for reference and absorb the experience and knowledge of live expert, the classification formed after inverted is sentenced
Set pattern is then.In selecting when multiple collateral learning classifiers of interbed, focus on their mutual complementarity.
Preferably, the present invention is assembled random forest grader, self organizing neural network in middle layer and is encoded based on ECOC
Three kinds of classification methods such as GentlBoost classifier.
Defect Multiple Classifier Fusion based on serial/parallel row integrated study frame priori rules and expertise, will be common
Parallel sorting device structure of modification at unique serial/parallel row integrated morphology, taken into account strip surface quality defect operational characteristic and
The versatility of classifying identification method is applied to other plate band-type products except strip (in such as paper industry for the classifier
Paper and with the fabric etc. in textile industry) provide necessary prerequisite.When concrete application, in order to meet strip, paper
Demand is detected with the surface quality defect of the different products such as fabric, is write after priori knowledge relevant to Product Process characteristic is concluded
Enter in rule list, versatility is then embodied with the Study strategies and methods via the training of standardization sample library.
In order to give full play to rule list and machine learning classification in the defect classifier based on serial/parallel row integrated study frame
The performance of device, the defect that the present invention also provides a kind of to be adapted with the defect classifier structure merges and stage division.
Specific step is as follows with stage division for defect merging:
S201 is isolated as the defect for being included in upper layer rule list the defect of upper layer level;
S202 compares the characteristic value of two kinds of defect sample in training set, if two kinds of defect classification sequence numbers are respectively yiAnd yj,
Corresponding sample size is respectively c in training setiAnd cj, characteristic value dimension is n dimension, and defect characteristic value sequence is respectively (xi1,
xi2,…,xik,…,xin) and (xj1,xj2,…,xjk,…,xjn), wherein xikAnd xjkIn respectively two kinds of defect characteristic value sequences
Component, calculate the mean value of two kinds of defect characteristic value respective components, be denoted as:WithK=1 ..., n,
Two mean bias percentages are further calculated, as follows:
Compare the respective components mean bias percentage and preset two threshold value δ1And δ2Size relation.If should
Respective components mean bias percentage is less than threshold value δ1Eigenvalue components number account for all characteristic value dimensions percentage it is big
In threshold value δ2, then corresponding two kinds of defects are merged.
Preferably, the present invention is by threshold value δ1It is taken as 70%, threshold value δ2It is taken as 90%.
S203, if mutual mistake point rate is greater than before not merging in the confusion matrix of two class defect classification results
Threshold value δ3, then also two class defects are merged.
Preferably, the present invention is by threshold value δ3It is taken as 20%.
S203, if mutual mistake point rate is greater than before not merging in the confusion matrix of two class defect classification results
Threshold value δ3, then also two class defects are merged.
Preferably, the present invention is by threshold value δ3It is taken as 20%.
S204, the defect after merging are placed on middle layer level.
S205, other unassorted defects are also placed in middle layer level, i.e., other unassorted defects with merge after
Defect be placed on same level, and among the defect classifier proposed by the present invention based on serial/parallel row integrated study frame
The multiple parallel machine learning classification methods of layer realize classification.
Embodiment two
The plan of embodiment two classifies to five kinds of different types of strip surface quality typical defects, this five defective class
Type title is respectively weld seam, hole, scuffing, roll marks and oil mark, and defect image pattern is respectively as shown in Fig. 3 a~Fig. 3 e.
The steel strip surface defect classification method based on serial/parallel row integrated study frame is proposed to above-mentioned five kinds using the present invention
When defect is classified, first the 1st kind of defect --- weld seam is separated from all defect classification, and foundation is real in building
When applying rule list involved by example two, " the defect image width " and " defect image height/width ratio " two of weld defect are found
A feature is Chong Die with the individual features zero of other all defect classifications, it is obvious that the building side of mentioned rule list according to the present invention
Method, the two features should all be added in the rule list of first layer, and the rule list of first layer is as shown in table 1 below:
The rule list of 1 first layer of table
Serial number | Rule description |
1 | Defect image width is equal with strip breadth |
2 | Defect image height/width ratio is less than 1/100 |
The defect merging mentioned further according to the present invention and stage division, should be by " defect image width " and " defect image
Defect of the corresponding defect classification of height/width ratio " as upper layer level, and the two features are directed to weld defect.This
Sample, all weld defects can simply be determined by the rule list of first layer.
Then, it is sent out by merging to the defect not mentioned using the present invention with defect classification results analysis before stage division
It is existing, it scratches and the mutual mistake point rate of two kinds of defects of roll marks is above 20%, for expansion, i.e., scuffing defect is divided into roll marks by mistake and lacks
Sunken percentage is 26.9582%, and roll marks defect is 29.3761% by the percentage that mistake is divided into scuffing defect, according to the present invention
The defect mentioned merges and stage division, is higher than threshold value δ3, so will scratch and two kinds of defects of roll marks merge, rename for
" strip defect ", as the defect of intermediate level, other holes not merged and oil mark are also placed in this level, in this way, in
Between level defect include three kinds, be respectively as follows:
Strip defect
Hole
Oil mark
Mentioned method according to the present invention, using middle layer in the defect classifier based on serial/parallel row integrated study frame
Machine learning classifiers classify to three kinds of defects of above-mentioned intermediate level.Middle layer Machine learning classifiers contain at random
Forest classified device, self organizing neural network and the GentlBoost classifier based on ECOC coding, by three kinds of parallel classifiers
It weights and determines classification results.
So far, the five defective in embodiment two has distinguished four kinds, is determined by upper layer rule list respectively
Weld seam, and the strip defect, hole and the oil mark that are determined by middle layer Machine learning classifiers, it is to be noted, however, that just
In beginning defect classification and strip defect is not included, this is the intersection scratched with two kinds of defects of roll marks, so also needing according to third
Layer rule list and classifier do further classification processing.
In embodiment two third layer rule list be sky, classifier using support vector machines realization, with support vector machines to by
The strip defect sample set that middle layer Machine learning classifiers determine is trained and classifies, and obtains final classification results,
Scuffing and two kinds of defects of roll marks are distinguished i.e. in strip defect.
Comprehensive final classification accuracy, as shown in table 2.
2 defect classification accuracy of table
Defect classification | Weld seam | Hole | It scratches | Roll marks | Oil mark |
Accuracy rate | 100.00% | 98.61% | 96.27% | 98.09% | 99.86% |
In conclusion a kind of strip surface quality defect based on serial/parallel row integrated study frame disclosed by the invention point
Class method is aided with by sufficiently merging newest machine learning classification identification technology, defect mechanism of production and artificial experience rule
The defect adaptable with classifier structure merges and stage division, further improves strip surface quality defect Classification and Identification
Accuracy, and improve self-learning capability, real-time performance and the generalization ability of classifier.
Those of ordinary skill in the art be further appreciated that implement the method for the above embodiments be can
It is completed with instructing relevant hardware by program, the program can store in computer-readable storage medium,
Described storage medium, including ROM/RAM, disk, CD etc..
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.
Claims (4)
1. a kind of strip surface quality defect classifier based on serial/parallel row integrated study frame, which is characterized in that classifier is
The three layers of serial structure divided from top to bottom, comprising rule list and Machine learning classifiers, specifically:
Upper layer is rule list,
Middle layer is multiple parallel Machine learning classifiers,
Lower layer is rule list or single Machine learning classifiers;
Rule list is presented as a kind of bivariate table in structure, specific comprising two column, is serial number and to regular particular content respectively
Description;The line number of rule list is determined by regular quantity.
2. a kind of strip surface quality defect classifier based on serial/parallel row integrated study frame according to claim 1,
It is characterized in that, rule list is the process mechanism that binding deficient generates, and use for reference and absorb the experience and knowledge of live expert, warp
The classification decision rule formed after conversion;In selecting when multiple collateral learning classifiers of interbed, focus on they mutually it
Between complementarity, the construction step of rule list includes:
S201 finds a value range feature Chong Die with other all categories zero, will if having found such feature
This feature is added in the rule list of upper layer, is the defect of upper layer level by the defect classification that this feature is distinguished;
S202 finds a value range and other Chong Die features of particular category (not all classification) zero, if having found this
Then this feature is added in lower layer's rule list for the feature of sample, is the defect of lower layer's level by the defect classification that this feature is distinguished;
S203 executes traversal than choosing to all features, until terminating.
3. a kind of strip surface quality defect classifier based on serial/parallel row integrated study frame according to claim 1,
It is characterized in that, middle layer assembly random forest grader, self organizing neural network and the GentlBoost based on ECOC coding
Three kinds of classifiers of classifier;The present invention proposes the defect classifier based on serial/parallel row integrated study frame, exactly in order to sufficiently melt
Existing machine learning classification identification technology, defect mechanism of production and artificial experience rule are closed, Classification and Identification is further increased
The self-learning capability of accuracy and classifier, real-time performance and generalization ability.
4. a kind of classification method of the defect classifier based on serial/parallel row integrated study frame, which is characterized in that closed including defect
And and stage division, the specific steps are as follows:
S1 is isolated as the defect for being included in upper layer rule list the defect of upper layer level;
S2 compares the characteristic value of two kinds of defect sample in training set, if two kinds of defect classification sequence numbers are respectively yiAnd yj, training set
In corresponding sample size be respectively ciAnd cj, characteristic value dimension is n dimension, and defect characteristic value sequence is respectively (xi1,xi2,…,
xik,…,xin) and (xj1,xj2,…,xjk,…,xjn), wherein xikAnd xjkComponent in respectively two kinds of defect characteristic value sequences,
The mean value for calculating two kinds of defect characteristic value respective components, is denoted as:WithK=1 ..., n are further counted
Two mean bias percentages are calculated, as follows:
Compare the respective components mean bias percentage and preset two threshold value δ1And δ2Size relation;If the correspondence
Component mean bias percentage is less than threshold value δ1Eigenvalue components number account for all characteristic value dimensions percentage be greater than threshold
Value δ2, then corresponding two kinds of defects are merged;
S3, if before not merging in the confusion matrix of two class defect classification results, mutual mistake divides rate to be greater than threshold value δ3,
Then also two class defects are merged;
S4, the defect after merging are placed on middle layer level;
S5, other unassorted defects are also placed in middle layer level, i.e., other unassorted defects with merge after defect
It is placed on same level, and multiple with the defect classifier middle layer proposed by the present invention based on serial/parallel row integrated study frame
Parallel machine learning classification method realizes classification.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910351038.6A CN110135477B (en) | 2019-04-28 | 2019-04-28 | Strip steel surface quality defect classifier based on serial/parallel integrated learning framework and classification method thereof |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910351038.6A CN110135477B (en) | 2019-04-28 | 2019-04-28 | Strip steel surface quality defect classifier based on serial/parallel integrated learning framework and classification method thereof |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110135477A true CN110135477A (en) | 2019-08-16 |
CN110135477B CN110135477B (en) | 2023-03-24 |
Family
ID=67575449
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910351038.6A Active CN110135477B (en) | 2019-04-28 | 2019-04-28 | Strip steel surface quality defect classifier based on serial/parallel integrated learning framework and classification method thereof |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110135477B (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160328837A1 (en) * | 2015-05-08 | 2016-11-10 | Kla-Tencor Corporation | Method and System for Defect Classification |
CN108846831A (en) * | 2018-05-28 | 2018-11-20 | 中冶南方工程技术有限公司 | The steel strip surface defect classification method combined based on statistical nature and characteristics of image |
CN109636772A (en) * | 2018-10-25 | 2019-04-16 | 同济大学 | The defect inspection method on the irregular shape intermetallic composite coating surface based on deep learning |
-
2019
- 2019-04-28 CN CN201910351038.6A patent/CN110135477B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160328837A1 (en) * | 2015-05-08 | 2016-11-10 | Kla-Tencor Corporation | Method and System for Defect Classification |
CN108846831A (en) * | 2018-05-28 | 2018-11-20 | 中冶南方工程技术有限公司 | The steel strip surface defect classification method combined based on statistical nature and characteristics of image |
CN109636772A (en) * | 2018-10-25 | 2019-04-16 | 同济大学 | The defect inspection method on the irregular shape intermetallic composite coating surface based on deep learning |
Also Published As
Publication number | Publication date |
---|---|
CN110135477B (en) | 2023-03-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109584221A (en) | A kind of abnormal image detection method generating confrontation network based on supervised | |
CN109711474A (en) | A kind of aluminium material surface defects detection algorithm based on deep learning | |
CN108520114B (en) | Textile fabric defect detection model and training method and application thereof | |
CN103324946B (en) | A kind of method and system of paper money recognition classification | |
CN108229580A (en) | Sugared net ranking of features device in a kind of eyeground figure based on attention mechanism and Fusion Features | |
CN101216436A (en) | Fabric flaw automatic detection method based on Support Vector data description theory | |
CN109064461A (en) | A kind of detection method of surface flaw of steel rail based on deep learning network | |
CN103267764B (en) | Hot-rolled steel plate surface defect image identification method based on neighborhood information estimation | |
CN109919934A (en) | A kind of liquid crystal display panel defect inspection method based on the study of multi-source domain depth migration | |
CN107169956A (en) | Yarn dyed fabric defect detection method based on convolutional neural networks | |
CN105719291A (en) | Surface defect image classification system with selectable varieties | |
CN103544499A (en) | Method for reducing dimensions of texture features for surface defect detection on basis of machine vision | |
CN108765402A (en) | Non-woven fabrics defects detection and sorting technique | |
CN103413314A (en) | Textile defect detecting algorithm based on texture gradients | |
CN107730500A (en) | A kind of ceramic tile texture detection, system, device and readable storage medium storing program for executing | |
CN111402236B (en) | Hot rolled strip steel surface defect grading method based on image gray value | |
CN109460914B (en) | Semi-supervised error correction learning-based bridge health grade determination method | |
CN110110780A (en) | A kind of picture classification method based on confrontation neural network and magnanimity noise data | |
CN105787046A (en) | Imbalanced data sorting system based on unilateral dynamic downsampling | |
CN106203496A (en) | Hydrographic curve extracting method based on machine learning | |
Zhang et al. | Research on surface defect detection algorithm of strip steel based on improved YOLOV3 | |
CN107423750A (en) | A kind of effective capsicum place of production source tracing method | |
CN103034868A (en) | Classification method aiming at small sample and high dimensional images | |
CN110135477A (en) | A kind of strip surface quality defect classifier and its classification method based on serial/parallel row integrated study frame | |
CN103258218A (en) | Matte detection frame generation method and device and defect detection method and device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |