CN107024532A - A kind of leakage field defect of pipeline position extracting method based on forms feature - Google Patents
A kind of leakage field defect of pipeline position extracting method based on forms feature Download PDFInfo
- Publication number
- CN107024532A CN107024532A CN201710236188.3A CN201710236188A CN107024532A CN 107024532 A CN107024532 A CN 107024532A CN 201710236188 A CN201710236188 A CN 201710236188A CN 107024532 A CN107024532 A CN 107024532A
- Authority
- CN
- China
- Prior art keywords
- defect
- acp
- forms
- sample
- abnormal area
- 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
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N27/00—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means
- G01N27/72—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating magnetic variables
- G01N27/82—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating magnetic variables for investigating the presence of flaws
- G01N27/83—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating magnetic variables for investigating the presence of flaws by investigating stray magnetic fields
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
- G06F2218/10—Feature extraction by analysing the shape of a waveform, e.g. extracting parameters relating to peaks
Abstract
The invention discloses a kind of leakage field defect of pipeline position extracting method based on forms feature, it includes：Step 1, extract after sample feature extraction to be carried out to the sample that is extracted based on forms, using history magnetic flux leakage data learning characteristic parameter and set up identification model；The feature includes significant characteristics, contrast metric, center point feature and fingerprint characteristic；Step 2, abnormality detection is carried out to any one section of testing data first, to determine corresponding abnormal area；Secondly in two stages identified abnormal area is carried out defect position extracting to obtain corresponding recognition result；Two described stages referred to the cognitive phase of individual defect and the identification to multiple defects and segmentation stage；Step 3, based on obtained recognition result carry out recruitment evaluation.This method considers the difference degree that measurement forms are sent out with ambient background by the flexible of inner and outer ring, enhances Model Identification effect and avoids excessive human intervention, adds the portability of algorithm.
Description
Technical field
The present invention relates to fault diagnosis and the technical field of artificial intelligence, particularly relate to a kind of based on forms feature
Leakage field defect of pipeline position extracting method.
Background technology
Magnetic conductivity steel pipe is widely used in the conveying of oil and natural gas.With the increasing of in-service pipeline run time
Plus, the potential risk of leakage of pipeline is also in lifting.Therefore periodically carrying out security evaluation to pipeline becomes to be even more important.It is used as prevention
One of key technology of pipe leakage, Non-Destructive Testing is widely used.At present, pipeline detection lossless detection method includes：Leakage
Magnetic testi, EDDY CURRENT and ultrasound detection.Wherein, Magnetic Flux Leakage Inspecting technology is widely used in almost 90% in-service pipeline.
This unquestionable leadership comes from：1) Magnetic Flux Leakage Inspecting technology is low to detection environmental requirement, is not influenceed by transmission medium,
Inside and outside defect can be detected simultaneously；2) initial quantization of defect can be realized.
Generally, for a successful defect position extracting method, there should be good defect and non-defective first
Nicety of grading, secondly it should be ensured that the defect recognized has the border that well defines.If the identification of defect is asked
Topic, such as the missing inspection of large-scale major defect, a unexpected incident of leakage will occur.This will bring huge economic damage
Become estranged environmental pollution.Meanwhile, if the border of defect can not be well defined, perhaps defect area can be more than actual defects
Region turned into inclusion region, or turned into deficient inclusion region less than actual defects region.This defect not defined well
Many problems will be brought to follow-up security evaluation.Such as, for the three-D profile inversion problem of defect, if defect area
Belonged to and included, the free degree will certainly increase, and ultimately resulted in the increase of inverting iteration time.Meanwhile, if defect area belongs to
Owe to include, the final inverting size of defect will be affected.
Learnt by consulting literatures, the identification problem of defect is usually to pass through from the angle of model to some direct features
Extract, finally reach the purpose of identification.But the Partial Feature of these features only inclusion region, it is difficult to ensure that good classification
Precision.Meanwhile, how few literature research extract the exact boundary of defect, such as (many defects are gathered in one to many defects
Rise region) segmentation problem.In fact, few independent defects in real corrosive environment, most defects are all poly-
Gather together.Therefore, the segmentation of many defects, which seems, is even more important.In addition, most documents only consider identification problem, recognize
Region is assumed, it is known that but how to go to determine that identification region is seldom described.
The content of the invention
In view of the defect that prior art is present, the invention aims to provide a kind of leakage field pipeline based on forms feature
Defect position extracting method,
It proposes four kinds of features containing parameter based on forms：Significant characteristics, contrast metric, center point feature
And fingerprint characteristic, and naive Bayesian network is applied in parameter learning ensure optimal characteristic ginseng value and avoid people
Work threshold value is intervened, in addition, the present invention proposes a kind of defect area extracting method in 2 stage simultaneously.
To achieve these goals, technical scheme：
A kind of leakage field defect of pipeline position extracting method based on forms feature, it is characterised in that comprise the following steps：
Step 1, set up identification model：Corresponding feature is carried out after extraction sample to the sample extracted based on forms to carry
Take, learn corresponding characteristic parameter using history magnetic flux leakage data and set up corresponding identification model；The sample includes defect sample
And non-defective sample, described feature includes significant characteristics, contrast metric, center point feature and fingerprint characteristic；
Step 2, defective locations extracted region is carried out to any one section of magnetic flux leakage data to be measured：First to any one section of number to be measured
According to abnormality detection is carried out, to determine corresponding abnormal area；Secondly defect is carried out to identified abnormal area in two stages
Extract to obtain corresponding recognition result in position；Two described stages refer to the cognitive phase of individual defect and lacked to multiple
Sunken identification and segmentation stage；
Step 3, based on obtained recognition result carry out recruitment evaluation, the effect assessed include Classification and Identification effect and side
Define effect in boundary.
It is preferred that, the step 1 includes：
Step 11, progress sample extraction process are the extraction process for carrying out defect sample and non-defective sample, the defect
The extraction process of sample is included carrying out defect area non-defective sample by handmarking, the extraction process of the non-defective sample
Sample and normal region non-defective sample is sampled；
Step 12, the corresponding feature extraction of sample progress based on forms to being extracted, wherein described significant characteristics
Corresponding formula is
Wherein, w represents to measure forms, xiRepresent i-th of element, I (xi) energy of the measurement forms is represented, it is expressed asxmaxThe forms element maximum of the measurement forms is represented,The average of all elements in the measurement forms is represented,Represented with Θ, and | | Θ | | implication be defined as follows：If Θ represents condition, | | Θ | | it is boolean
Value, even condition Θ is met, then | | Θ | | it is worth for 1, otherwise value is 0；If Θ representing matrixs, | | Θ | | element in representing matrix
Number, θsRepresent parameter to be learned；
The contrast metric includes outer shroud contrast OC and inner ring contrast IC, wherein outer shroud contrast OC (w, θOC) be
Refer to the difference degree of the rectangular area between measurement forms and its corresponding outer shroud, the size of the rectangular area is joined by sliding
Number θOCObtain, i.e.,
Wherein, θOC={ θOCL, θOCC, θOCLAnd θOCCThe axial contrast level parameter of outer shroud and the circumferential contrast of outer shroud are represented respectively
Parameter, both spans are 0~1, then corresponding outer shroud contrast is obtained by calculating card side's distance, i.e.,：
OC(w,θOC)=χ2(MSE(w),MSE(O(w,θOC))) (23)；
Wherein inner ring contrast IC (w, θIC) refer to the difference that measures the rectangular area between forms and its corresponding inner ring
Degree, In (w, θIC) measurement forms its corresponding inner rectangular ring is expressed as, pass through control parameter θICObtain, i.e.,：
Wherein, θIC={ θICL,θICC,It is to measure the rectangular area between forms and inner ring, θICLWith θICCPoint
Not Biao Shi the axial contrast level parameter of inner ring and the circumferential contrast level parameter of inner ring, both spans are 0~1, then it is corresponding in
Ring contrast is：
The center point feature CP is：
Wherein：
θCP={ θCPL,θCPC, θCPLWith θCPCThe axial contrast level parameter of inner ring and the circumferential contrast level parameter of inner ring are represented respectively,
Both spans are 0~1, In (w, θCP) internal slide straight-flanked ring is represented,Represent between forms and straight-flanked ring
Region；
The fingerprint characteristic includes axial fingerprint characteristic and circumferential fingerprint characteristic, wherein, axial finger print information and circumferentially refer to
Line information is respectively expressed as α=(X1,X2,...,Xj) and β=(Y1,Y2,...Yi), wherein：
Then the corresponding formula of axial fingerprint characteristic and circumferential fingerprint characteristic is respectively：
FL(w)=fp (dt (x) * α), FC(w)=fp (dt (x) * β) (30)
Dt represents wavelet filtering function, represents that dt (x) * α or dt (x) * β, fp () represent peak function, w is considered to be
One line number M, columns N matrix, xijIt is the element in the matrix；
Step 13, learn corresponding characteristic parameter using history magnetic flux leakage data and set up corresponding identification model；
θ is learnt by naive Bayesian network firstOC、θIC、θCP, due to θOC、θIC、θCPLearning method is identical, in order to retouch
Facility is stated, only to θOCIt is described, that is, defines θ=θOC, G (w, θ)=OC (w, θ), sets target forms areNon-targeted window
Body isFor any one parameter θ, positive sample possibility p (G (w, θ) | obj) and negative sample possibility p (G (w, θ) is set up
| bg) after obtained by maximizing posterior probability：I.e.
Wherein,pθ(bg)=1-pθ(obj), pθAnd p (obj)θ(bg) respectively
The ratio shared by the ratio and non-defective sample of defect sample, as prior probability in sample set are represented, c represents positive negative sample ratio
Example, it belongs to obj and bg；
Next learning parameter θs, optimal θ is determined by maximizing defective locations precisionsValue：I.e.
Wherein o represents target defect region, finally, and identification model is set up with Intelligence Classifier, and characteristic parameter is mapped
To discrete space.
The step 2 includes：
Step 21, abnormality detection is carried out to any one section of testing data, to determine corresponding abnormal area；Specifically, will
Testing data is divided into size to be s1×s2Grid, pass through θsPixilated grid is determined, most identified each pixilated grid at last
Merge and obtain abnormal area, wherein s2It is set to 1, s1In θsObtained in parameter learning；
Step 22, defect position extracting secondly is carried out to obtain corresponding knowledge to identified abnormal area in two stages
Other result；Two described stages referred to the cognitive phase of individual defect and the identification to multiple defects and segmentation stage；
Wherein, the cognitive phase to individual defect includes step 221, identified abnormal area is carried out first
Direct Recognition, recognizes that the abnormal area whether there is the defect area with unimodal defect characteristic, is to proceed identification simultaneously
Abnormal area is further merged according to ASME criterions, the abnormal area merging criterion is the axle if between two abnormal areas
It is less than set threshold value to distance, then merges two abnormal areas, the abnormal area after merging is as an entirety continuation point
Analyse, if threshold value set by the axial distance between two exceptions, the two abnormal areas are as independent abnormal area；
Step 222, the created identification model that passes through recognize abnormal area and judged, if recognition result belongs to scarce
Fall into, then carry out follow-up overlap test, otherwise carry out step 24, the overlap test principle states are as follows；If two identifications
Model output area D1And D2It can be merged, then：
Wherein, ξ rule of thumb values；
Step 223, to identified abnormal area carry out approximate center point detection be ACP detect, with find out presence it is many
The approximate center point ACP of each independent defect of the abnormal area of defect area；Specifically, entering first to the abnormal area
Row mesh generation so that the average amplitude of each grid is AA, and the average amplitude matrix of whole region is G, that is, is caused：AAi,j
∈ G, and if there is ACP in current grid, then need to meet following condition：
Wherein, RGAnd CGDifference representing matrix G line number and columns；
Step 224, detected central point is identified, if abnormal area has ACP, proved in abnormal area
It may contain defective, then carry out step 225, if ACP is not present in abnormal area, the region is identified as non-defective region；
Step 225, split and recognized to defective abnormal area may be contained in step 224 in region, i.e., for
One possesses NAIndividual ACP multiple target region is abnormal area, first initialization sampling minimum windowFrame size is
m1×m2And new sampling forms are updated toWherein, L1Represent the minimum axial both sides of sampling forms per side
Extension points, L2Represent that extension of the minimum circumferential both sides of sampling forms per side is counted；And axially expanded with circumferential
Exhibition, in extension, if there is new ACP in the forms newly extended, samples when front direction and stops and obtain when front direction
Border, when both direction all has border, the sampled point of current ACP points terminates, NACentered on put quantity；
Step 226, bimodal defect recognition is carried out to the abnormal area in step 225, according to close ACP can combination principle,
Update ACP and the ACP after renewal is sampled；Two close ACP are ACPiAnd ACPjFollowing two conditions should be met：
Wherein, C (ACPi,ACPj) represent ACPiAnd ACPjCircumferential distance, A (ACPx,ACPy) represent ACPxAnd ACPyAxle
To distance；ρ is experience value；
Step 227, pass through the abnormal area corresponding to created identification model identification step 226 and judged, if
Recognition result belongs to defect, then carries out follow-up overlap test, otherwise carries out step 224, the overlap test principle states are such as
Under；If two identification model institute output area D1And D2It can be merged, then：
It is preferred that, the step 3 carries out recruitment evaluation according to the result finally recognized, and it is carried out according to following interpretational criterias
Recruitment evaluation：
Classification and Identification effect assessment criterion 1：
Classification and Identification effect assessment criterion 2：
Wherein, TP is the number that defect is identified as defect；FN is the number that defect is identified as non-defective；TN is non-lacks
Fall into the number for being identified as non-defective；FP is the number that non-defective is identified as defect；
Classification and Identification effect assessment criterion 3：
α2Take 0.3；
Boundary definition accuracy rate interpretational criteria is as follows：
Wherein, w represents output identification form region, and o represents to demarcate defect area, represents w ∩ o or w ∪ o, | | |
| represent data points.
Compared with prior art, beneficial effects of the present invention：
(1) general feature extracting method is different from, the present invention proposes a kind of feature extracting method based on forms, should
Method considers the difference degree that measurement forms are sent out with ambient background by the flexible of inner and outer ring.Enhance Model Identification effect.
(2) carry out the optimizing of parameter with naive Bayesian network, it is to avoid excessive human intervention, add algorithm
It is portable.
(3) consider the need for being Practical Project, it is proposed that a kind of defective locations method for extracting region based on 2 stages,
It is for the real corrosive environment of actual description so that method, which has more, to have significant practical applications.
(4) the inventive method has been applied to Practical Project and has detected that Detection results are good.
Brief description of the drawings
Fig. 1 is conspicuousness parameter θ of the present inventionsIllustration；
Fig. 2 is contrast metric of the present invention, center point feature and fingerprint characteristic parameter learning figure；
Fig. 3 is that outer region divides figure in measurement window body of the present invention, and (a) is that outer region divides figure, and (b) is inner ring
Zoning plan；
Fig. 4 is abnormality detection flow chart of the present invention；
Fig. 5 is that defect area of the present invention extracts flow chart；
Fig. 6 is many defect area ACP samplings flow charts of the present invention；
Fig. 7 is the Classification and Identification result of different classifications device of the present invention；
Fig. 8 is certain northern city in-service pipeline road route map of China of the present invention；
Fig. 9 is the recognition effect schematic diagram that in-service pipeline of the present invention detects this pipeline fragment.
Embodiment
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with attached in the embodiment of the present invention
Figure, technical scheme is clearly and completely described, it is clear that described embodiment is that a part of the invention is real
Apply example, rather than whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art are not making creation
Property work under the premise of the every other embodiment that is obtained, belong to the scope of protection of the invention.
In view of the deficienciess of the prior art, the present invention is different from general direct feature, such as signal peak-valley difference feature
Deng first proposed a kind of based on forms feature i.e. significant characteristics, contrast metric, center point feature and fingerprint characteristic
Leakage field defect of pipeline position extracting method, these four features focus more on difference of the whole forms with ambient background in itself；Its
It is secondary, in order to ensure optimal characteristic ginseng value and avoid artificial threshold value intervention, joined accordingly using naive Bayesian network
Mathematics is practised, and proposes a kind of two stage defect area extracting method in defect area extraction process again, i.e., in the first rank
Duan Zhong, if measurement window body surface reveals independent defect characteristic and possesses the border well defined, it will be identified as defect
Region；The segmentation problem of many defects is then primarily upon in second stage, contains complete defect but by mistake knowledge simultaneously for those
Other forms extract exact boundary；Finally, machine learning method is used for Model Identification.
Based on above-mentioned principle, then the present invention specifically includes following steps：
Step 1, set up identification model：Corresponding feature is carried out after extraction sample to the sample extracted based on forms to carry
Take, learn corresponding characteristic parameter using history magnetic flux leakage data and set up corresponding identification model；The sample includes defect sample
And non-defective sample, described feature includes significant characteristics, contrast metric, center point feature and fingerprint characteristic；
Wherein, the step 1 includes
Step 11, progress sample extraction process are the extraction process for carrying out defect sample and non-defective sample, the defect
The extraction process of sample is by handmarking, in order to increase the diversity of non-defective sample, the extraction process of the non-defective sample
Including being sampled to defect area non-defective sample and normal region non-defective sample being sampled；Wherein defect area is non-
Defect sample sampling process is：We obtain several non-defective regions around the defect that each is marked, and this example is 8
It is individual, the size of one of defect is set as m × n, corresponding centre coordinate X (x, y), i.e.,Simultaneously
In order to reduce the registration of defect and non-defective region, while ensureing that non-defective region includes segmental defect area information, sampling
Non-defective region centre coordinate X ' (x ', y ') should be met：And non-defective regional center is selected at random
Take, size is consistent with corresponding defect size；In order to increase the classification boundaries of defect and non-defective sample, normal region is added non-
Defect is sampled, and wherein non-defective sampling process in normal region is：In order to increase the classification boundaries of defect and non-defective sample, this mistake
Journey is normally carrying out non-defective specimen sample without defect area；If o represents defect, measurement forms w is met：W ∩ o=
φ.And all samples sources are in Tianjin test site True Data.
Step 12, the corresponding feature extraction of sample progress based on forms to being extracted：
One real defect possesses the outward appearance for being different from its ambient background, and based on this uniqueness, we have proposed aobvious
Work property feature, the formula corresponding to described significant characteristics is
Wherein, w represents to measure forms, xiRepresent i-th of element, I (xi) energy of the measurement forms is represented, it is expressed asxmaxThe forms element maximum of the measurement forms is represented,The average of all elements in the measurement forms is represented,Represented with Θ, and | | Θ | | implication be defined as follows：If Θ represents condition, | | Θ | | it is boolean
Value, even condition Θ is met, then | | Θ | | it is worth for 1, otherwise value is 0；If Θ representing matrixs, | | Θ | | element in representing matrix
Number, θsRepresent parameter to be learned；This feature is designed to whether measurement forms wrap exception.
According to the feature of magnetic leakage signal, it is contemplated that defect and the relation of its ambient background, outer shroud contrast metric and inner ring
Contrast metric is suggested, and the contrast metric includes outer shroud contrast OC and inner ring contrast IC, wherein due to pipeline
Corrosive environment is complicated, and too many background information loses contrast meaning, therefore the background in each background direction of measurement forms is big
It is small to be equal to measurement forms size itself, outer shroud contrast OC (w, θOC) refer to measure the rectangle between forms and its corresponding outer shroud
The difference degree in region, the size of the rectangular area is by sliding parameter θOCObtain, i.e.,
Wherein, θOC={ θOCL, θOCC, θOCLAnd θOCCThe axial contrast level parameter of outer shroud and the circumferential contrast of outer shroud are represented respectively
Parameter, both spans are 0~1, then corresponding outer shroud contrast is obtained by calculating card side's distance, i.e.,：
OC(w,θOC)=χ2(MSE(w),MSE(O(w,θOC))) (43)；
OC can measure whether forms belong to deficient comprising (measurement forms include segmental defect).OC values are higher, and forms are included
The possibility of complete defect is bigger.
Under many circumstances, perhaps forms occurred the situation for including (form region is more than actual defects region), this
When OC cannot weigh forms well, therefore we have proposed inner ring contrast IC (w, θIC) measure forms in itself to weigh
The difference degree of rectangular area between its corresponding inner ring, In (w, θIC) it is expressed as measurement forms its corresponding inner rectangular
Ring, passes through control parameter θICObtain, i.e.,：
Wherein, θIC={ θICL,θICC,It is to measure the rectangular area between forms and inner ring, θICLWith θICCPoint
Not Biao Shi the axial contrast level parameter of inner ring and the circumferential contrast level parameter of inner ring, both spans are 0~1, then it is corresponding in
Ring contrast is：
IC can weigh whether measurement forms are crossed comprising defect, that is to say, that IC can weigh measurement window body interior background
The size in region, the smaller explanation interior background region of IC value is bigger, and otherwise explanation measurement forms are comprising complete defect and very
The possibility of few background is bigger.In actually detected, by the flexible of inner and outer ring, OC and IC can preferably determine defect together
Border.
Possess the defect for defining border very well as one, the measured value of central area can be higher than the measurement of fringe region
Value.In general, a defect is in the region of than one, the region defect in measurement window body edge at measurement window body center more
It is possible to be received.Based on this measurement characteristicses, the difference of forms center and peripheral is weighed we have proposed center point feature
Degree.Similar with inner ring feature, center point feature is determined again by inner ring is slided.Center point feature CP is defined as：
Wherein：
θCP={ θCPL,θCPC, θCPLWith θCPCThe axial contrast level parameter of inner ring and the circumferential contrast level parameter of inner ring are represented respectively,
Both spans are 0~1.
According to Magnetic Flux Leakage Inspecting principle, the measurement on defect axial direction is up to a peak value, in circumferential direction at most
One peak value.According to this feature of defect and magnetic leakage signal, the peak value of measurement forms is weighed invention defines fingerprint characteristic
Number.The fingerprint characteristic includes axial fingerprint characteristic and circumferential fingerprint characteristic, wherein, axial finger print information and circumferential fingerprint letter
Breath is respectively expressed as α=(X1,X2,...,Xj) and β=(Y1,Y2,...Yi), wherein：
Then the corresponding formula of axial fingerprint characteristic and circumferential fingerprint characteristic is respectively：
FL(w)=fp (dt (x) * α), FC(w)=fp (dt (x) * β) (50)
Dt represents wavelet filtering function, represents that dt (x) * α or dt (x) * β, fp () represent peak function, w is considered to be
One line number M, columns N matrix, xijIt is the element in the matrix；
First step 13, learn corresponding characteristic parameter using history magnetic flux leakage data and set up corresponding identification model, history building
Magnetic flux leakage data derives from true pipeline data；
θ is learnt by the naive Bayesian network for parameter optimization firstOC、θIC、θCP, due to θOC、θIC、θCPStudy side
Method is identical, in order to describe facility, only to θOCIt is described, that is, defines θ=θOC, G (w, θ)=OC (w, θ), sets target forms
ForNon-targeted forms areFor any one parameter θ, positive sample possibility p (G (w, θ) | obj) and negative sample are set up
Obtained after this possibility p (G (w, θ) | bg) by maximizing posterior probability：
Wherein,pθ(bg)=1-pθ(obj), pθAnd p (obj)θ(bg) respectively
The ratio shared by the ratio and non-defective sample of defect sample, as prior probability in sample set are represented, c represents positive negative sample ratio
Example, it belongs to obj and bg；
Next learning parameter θs, optimal s values are determined by maximizing defective locations precision, that is to say, that we seek
Look for optimal θsValue causes measurement forms w at utmost coverage goal defect o：
Finally, identification model is set up with Intelligence Classifier, characteristic parameter is mapped to discrete space.
Instantiation, as shown in figure 1, being split to calculate each ginseng by depicted area around each measurement forms
The possibility value of several lower features.Significant characteristics parameter learning is as shown in Fig. 2 s ∈ { 2,4,8,16,32 } in figure, we make
With θsThe defect changed under different mesh generations is capped the change of precision.With θsChange, defective locations covering precision is all
It is first to rise to decline afterwards, maximum is between 0.1~0.2.Therefore optimal θsValue is 0.15, due in θsIn 0.1~
Precision highest in the case of s=4 when between 0.2.Therefore optimal mesh generation size is 4 × 1.Characteristic parameter θOC, θIC, θCPGinseng
Mathematics is practised as shown in Fig. 2 each characteristic parameter includes two parametric components of axial and circumferential.According to positive negative sample each
Characteristic probability distribution under individual parameter value value, we find the positive and negative sample probability deviation point of maximum as optimal value of the parameter.
It is subsequently based on the carry out feature extraction of forms.As shown in figure 3, the interior exterior domain of measurement forms is divided into 8 regions.It is each
Individual contrast metric is broken down into axial and circumferential contrast.Axial contrast reflects measurement forms and respective shaft to background (3
With 7 regions) difference.Equally, circumferential contrast reflects the difference of the circumferential background (1 and 5 region) of correspondence of measurement forms.Press
According to Optimal Learning parameter extraction conspicuousness, contrast, central point and fingerprint characteristic.Identification model is finally set up, and according to extraction
The feature samples arrived, train identification model.Due to the present invention, develop is universal identification algorithm, therefore, the present invention SVM, RF
Model is respectively trained with KNN.
Step 2, defective locations extracted region is carried out to any one section of magnetic flux leakage data to be measured：First to any one section of number to be measured
According to abnormality detection is carried out, to determine corresponding abnormal area；Secondly defect is carried out to identified abnormal area in two stages
Extract to obtain corresponding recognition result in position；Two described stages refer to the cognitive phase of individual defect and lacked to multiple
Sunken identification and segmentation stage；Further, the step 2 includes：
Step 21, abnormality detection is carried out to any one section of testing data, to determine corresponding abnormal area；Specifically, will
Magnetic leakage measurement amount to be measured is divided into size to be s1×s2Grid, pass through θsPixilated grid is determined, if following condition is set up：
Then grid is defined as pixilated grid, and most identified each pixilated grid merges and obtains abnormal area at last,
Wherein in order to reduce the influence between sensor passage, s2It is set to 1, s1In θsObtained in parameter learning；Specifically such as Fig. 4 institutes
Show, magnetic leakage measurement to be measured is divided into the grid that size is 4 × 1.Then, θ is passed throughsPixilated grid is determined, most grid enters at last
Row merging obtains abnormal area.
Step 22, defect position extracting secondly is carried out to obtain to identified abnormal area in two stages as shown in Figure 5
Obtain corresponding recognition result；Two described stages：1) cognitive phase (IDI) of separate defect is separate defect to individual defect
Cognitive phase, this stage mainly recognizes independent individual defect；2) the segmentation stage (MDS) of many defects, to multiple defects
Identification and the segmentation stage, the stage is designed to recognize many defects and is individual defect by many defect Segmentations.When region is identified
During for individual defect, overlap test will be performed, and be ultimately formed with the good defect for defining border.
Wherein, the cognitive phase to individual defect, which includes step 221, the step, includes two parts, i.e., unimodal
The identification of defect and the identification of bimodal defect are to carry out Direct Recognition to identified abnormal area first, it is intended to recognize that those are only
Vertical has defect characteristic and there is the good region for defining border to be to recognize that the abnormal area whether there is with unimodal defect
The defect area of feature, is then to continue to recognize bimodal defect, double due to the presence of plateau region between two peaks of bimodal defect
Peak defect may be divided into two independent parts in the axial direction.Therefore we further merge institute according to ASME criterions
The abnormal area of determination, the abnormal area merging criterion is if the axial distance between two abnormal areas is less than set
The wall thickness (3t, t are wall thickness) of threshold value, such as three times, then merge two abnormal areas, and the abnormal area after merging is whole as one
Body continues the analysis of next step, if threshold value set by axial distance between two exceptions, the two abnormal areas as
Independent abnormal area；
Step 222, separate defect differentiate：Pass through created identification model identification abnormal area and judged, if knowing
Other result belongs to defect, then carries out follow-up overlap test, otherwise carries out step 224, and the overlap test principle states are as follows；
If two identification model institute output area D1And D2It can be merged, then：
ξ values are set to 0.5 by us according to PASCAL criterions.
Step 223, to identified abnormal area carry out approximate center point detection be ACP detect, with find out presence it is many
The approximate center point ACP of each independent defect of the abnormal area of defect area；In the segmentation stage of many defects, it is intended to
Find the approximate center point (ACP) of each independent defect of many defect areas.Each approximate center point may correspond to one
Individual defect, the identification of central point will provide foundation for follow-up sampling.Mesh generation during with abnormality detection is similar, specifically
, mesh generation is carried out to the abnormal area first so that the average amplitude of each grid is AA, whole region is averaged
Amplitude matrix is G, i.e.,：AAi,j∈ G, and if there is ACP in current grid, then need to meet following condition：
Wherein, RGAnd CGDifference representing matrix G line number and columns；If that is, current grid AA values exceed threshold value
And exist it is upper, under, left, right grid, and AA values are local maximums, then there is ACP values in the grid.
Step 224, ACP identifications are carried out to detected central point, if abnormal area has ACP, prove exceptions area
It may contain defective in domain, then carry out step 225, if ACP is not present in abnormal area, the region is identified as non-defective area
Domain；
Step 225, many defect areas split unimodal defect recognition, and multi_region problem is actually to each ACP
Sampled the problem of recognizing again：The sampling process of each many defect area is as shown in fig. 6, specific to area in step 224
Defective abnormal area may be contained in domain to be split and recognized, i.e., possess N for oneAIndividual ACP multiple target region is
Abnormal area, first initialization sampling minimum windowAnd new sampling forms are updated toL1Represent
Extension of the axial both sides of minimum sampling forms per side is counted, L2Represent extension point of the minimum circumferential both sides of sampling forms per side
Number；And be axially extended with circumferential, in extension, if there is new ACP in the forms newly extended, work as front direction
Sampling stops and obtained when the border of front direction, and when both direction all has border, the sampled point of current ACP points terminates；
ACP is updated, the sampling process then carried out to the ACP after renewal.
Step 226, many defect areas split bimodal defect recognition, and ACP identification is it is possible that a bimodal defect quilt
Separate situation about sampling respectively.Therefore, the present invention is directed to bimodal identification problem, and existing ACP is carried out by certain criterion
Update so that bimodal defect can be detected：Specifically, carrying out bimodal defect recognition to the abnormal area in step 225, press
Take a picture nearly ACP can combination principle, update ACP simultaneously the ACP after renewal is sampled as shown in Figure 6；Two close ACP ACPi
And ACPjFollowing two conditions should be met to meet：
Wherein, C (ACPi,ACPj) represent ACPiAnd ACPjCircumferential distance；ρ of the present invention value is 3.Then after updating
ACP samplings it is consistent with described in Fig. 6.
Step 227, pass through the abnormal area corresponding to created identification model identification step 226 and judged, if
Recognition result belongs to defect, then carries out follow-up overlap test, otherwise carries out step 224, the overlap test principle states are such as
Under；If two identification model institute output area D1And D2It can be merged, then：
Step 3, based on obtained recognition result carry out recruitment evaluation, the effect assessed include Classification and Identification effect and side
Define effect in boundary.
Further, the step 3 carries out recruitment evaluation according to the result finally recognized, and it enters according to following interpretational criterias
Row recruitment evaluation：
Classification and Identification effect assessment criterion 1：
Classification and Identification effect assessment criterion 2：
Wherein, TP is the number that defect is identified as defect；FN is the number that defect is identified as non-defective；TN is non-lacks
Fall into the number for being identified as non-defective；FP is the number that non-defective is identified as defect；
In practical application, in order to preferably weigh Precision and Recall, addition Classification and Identification effect assessment criterion 3：
α2Take 0.3；
Boundary definition accuracy rate interpretational criteria is as follows：
Wherein, w represents output identification form region, and o represents to demarcate defect area, represents w ∩ o or w ∪ o, | | |
| represent data points.
Embodiment of the present invention is to be estimated test respectively with experimental field pipeline and in-service pipeline.
Step 3.1：Experimental field pipeline result.
Testing ground pipeline of the present invention is described as follows：Duct length is 800m, pipe diameter 219mm, 9.5mm duct wall
Thickness, pipe material is the carbon steel material being widely adopted, and pipeline internal medium is water, and flow velocity 0.5m/s, operating pressure is 3MPa.Institute
Stating pipeline includes 1280 separate defects, defect more than 65.Wherein separate defect includes 800 unimodal defects and 480
Bimodal defect.Defect is artificial defect in experiment, and the length and width value of defect is 10mm to 60mm, and depth value 1 is arrived
9mm.Three different classifications devices SVM, RF and KNN Classification and Identification effect are as shown in Figure 7；As seen from the figure, RF and KNN possess closely
The classification accuracy of patibhaga-nimitta etc..Although SVM possesses highest Precision, but Recall is less than other two graders；So
And, the evaluation index of three graders is above 80%, meets practical engineering application demand.Table 1 lists three grader effects
Boundary definition ability under fruit.Equally, classification indicators can also be higher than 80%.
In order to further illustrate the recognition effect of the present invention, the recognition capability to both sides' defect makes statistics respectively, and
The segmentation ability of many defects is assessed, the recognition capability of bimodal defect is as shown in table 2, the defect Segmentation effect such as table of many defect areas
Shown in 3.
Step 3.2：In-service field pipe detection result.The inventive method is applied to the defects detection of actual pipeline at present
In.In-service pipeline in embodiment is from a northern city of China.Whole length of pipeline is 88.99m, running route
As shown in Figure 8.Mark 1 represents initial station, and mark 5 represents terminal.Running route is by two valve chambers (mark 2 and 4), and identifying 3 is
Heating station.Pumped (conveying) medium is 0.The crude oil of 5% moisture content, condensation point is 26 DEG C.The present invention considers to lack in actual corrosive environment more
Fall into and be in the great majority, because RF many defect Segmentation effects are preferable, therefore the present invention carries out in-service pipeline with RF graders
Detection.The recognition result of in-service 10 pipelines is as shown in table 4.Although global index is less than experimental level, it can still meet
Actual requirement of engineering.The identification of a wherein segment pipe asks result as shown in Figure 9.
Table 1
| SVM | RF | KNN | |
| BA | 86% | 92% | 85% |
Table 2
| SVM | RF | KNN | |
| Recall | 83% | 88% | 95% |
Table 3
| Method | Number of defects | Estimated result | Recall | BA |
| SVM | 169 | 140 | 83% | 78% |
| RF | 169 | 155 | 92% | 86% |
| KNN | 169 | 148 | 88% | 80% |
Table 4
| Precision | Recall | F | BA |
| 78% | 82% | 79% | 84% |
The foregoing is only a preferred embodiment of the present invention, but protection scope of the present invention be not limited thereto,
Any one skilled in the art the invention discloses technical scope in, technique according to the invention scheme and its
Inventive concept is subject to equivalent substitution or change, should all be included within the scope of the present invention.
Claims (3)
1. a kind of leakage field defect of pipeline position extracting method based on forms feature, it is characterised in that comprise the following steps：
Step 1, set up identification model：Extract the corresponding feature extraction of sample progress based on forms to being extracted, profit after sample
Learn corresponding characteristic parameter with history magnetic flux leakage data and set up corresponding identification model；The sample includes defect sample and non-scarce
Sample is fallen into, described feature includes significant characteristics, contrast metric, center point feature and fingerprint characteristic；
Step 2, defective locations extracted region is carried out to any one section of magnetic flux leakage data to be measured：Any one section of testing data is entered first
Row abnormality detection, to determine corresponding abnormal area；Secondly defective locations are carried out to identified abnormal area in two stages
Extract to obtain corresponding recognition result；Two described stages refer to the cognitive phase of individual defect and to multiple defects
Identification and segmentation stage；
Step 3, based on obtained recognition result carry out recruitment evaluation, the effect assessed include Classification and Identification effect and border circle
Determine effect.
2. according to the method described in claim 1, it is characterised in that：
The step 1 includes：
Step 11, progress sample extraction process are the extraction process for carrying out defect sample and non-defective sample, the defect sample
Extraction process by handmarking, the extraction process of the non-defective sample includes sampling to defect area non-defective sample
And normal region non-defective sample is sampled；
Step 12, the corresponding feature extraction of sample progress based on forms to being extracted, wherein described significant characteristics institute is right
The formula answered is
Wherein, w represents to measure forms, xiRepresent i-th of element, I (xi) energy of the measurement forms is represented, it is expressed asxmaxThe forms element maximum of the measurement forms is represented,The average of all elements in the measurement forms is represented,Represented with Θ, and | | Θ | | implication be defined as follows：If Θ represents condition, | | Θ | | it is Boolean,
Even condition Θ is met, then | | Θ | | it is worth for 1, otherwise value is 0；If Θ representing matrixs, | | Θ | | element in representing matrix
Number, θsRepresent parameter to be learned；
The contrast metric includes outer shroud contrast OC and inner ring contrast IC, wherein outer shroud contrast OC (w, θOC) refer to survey
The difference degree of the rectangular area between forms and its corresponding outer shroud is measured, the size of the rectangular area is by sliding parameter θOC
Obtain, i.e.,
Wherein, θOC={ θOCL, θOCC, θOCLAnd θOCCThe axial contrast level parameter of outer shroud and the circumferential contrast ginseng of outer shroud are represented respectively
Number, both spans are 0~1, then corresponding outer shroud contrast is obtained by calculating card side's distance, i.e.,：
OC(w,θOC)=χ2(MSE(w),MSE(O(w,θOC))) (3)；
Wherein inner ring contrast IC (w, θIC) refer to the difference degree that measures the rectangular area between forms and its corresponding inner ring,
In(w,θIC) measurement forms its corresponding inner rectangular ring is expressed as, pass through control parameter θICObtain, i.e.,：
Wherein, θIC={ θICL,θICC,It is to measure the rectangular area between forms and inner ring, θICLWith θICCDifference table
Show the axial contrast level parameter of inner ring and the circumferential contrast level parameter of inner ring, both spans are 0~1, then corresponding inner ring pair
It is than degree：
The center point feature CP is：
Wherein：
θCP={ θCPL,θCPC, θCPLWith θCPCThe axial contrast level parameter of inner ring and the circumferential contrast level parameter of inner ring are represented respectively, both
Span is 0~1, In (w, θCP) internal slide straight-flanked ring is represented,Represent the area between forms and straight-flanked ring
Domain；
The fingerprint characteristic includes axial fingerprint characteristic and circumferential fingerprint characteristic, wherein, axial finger print information and circumferential fingerprint letter
Breath is respectively expressed as α=(X1,X2,...,Xj) and β=(Y1,Y2,...Yi), wherein：
Then the corresponding formula of axial fingerprint characteristic and circumferential fingerprint characteristic is respectively：
FL(w)=fp (dt (x) * α), FC(w)=fp (dt (x) * β) (10)
Dt represents wavelet filtering function, represents that dt (x) * α or dt (x) * β, fp () represent peak function, w is considered to be
One line number M, columns N matrix, xijIt is the element in the matrix；
Step 13, learn corresponding characteristic parameter using history magnetic flux leakage data and set up corresponding identification model；
θ is learnt by naive Bayesian network firstOC、θIC、θCP, due to θOC、θIC、θCPLearning method is identical, in order to describe just
Profit, only to θOCIt is described, that is, defines θ=θOC, G (w, θ)=OC (w, θ), sets target forms areNon-targeted forms areFor any one parameter θ, positive sample possibility p (G (w, θ) | obj) and negative sample possibility p (G (w, θ) | bg) are set up
Obtained afterwards by maximizing posterior probability：I.e.
Wherein,pθ(bg)=1-pθ(obj), pθAnd p (obj)θ(bg) represent respectively
The ratio of defect sample and the ratio shared by non-defective sample, as prior probability in sample set；
Next learning parameter θs, optimal θ is determined by maximizing defective locations precisionsValue：I.e.
Wherein o represents target defect region, finally, and identification model is set up with Intelligence Classifier, by characteristic parameter be mapped to from
Dissipate space.
3. method according to claim 2, it is characterised in that：
The step 2 includes：
Step 21, abnormality detection is carried out to any one section of testing data, to determine corresponding abnormal area；Specifically, will be to be measured
Data are divided into size to be s1×s2Grid, pass through θsPixilated grid is determined, most identified each pixilated grid is carried out at last
Merging obtains abnormal area, wherein s2It is set to 1, s1In θsObtained in parameter learning；
Step 22, defect position extracting secondly is carried out to identified abnormal area in two stages tied with obtaining corresponding identification
Really；Two described stages referred to the cognitive phase of individual defect and the identification to multiple defects and segmentation stage；
Wherein, the cognitive phase to individual defect includes step 221, identified abnormal area is carried out directly first
Identification, recognizes that the abnormal area whether there is the defect area with unimodal defect characteristic, is to proceed identification and basis
ASME criterions further merge abnormal area, the abnormal area merging criterion be if between two abnormal areas axially away from
From less than set threshold value, then two abnormal areas are merged, the abnormal area after merging continues to analyze as an entirety, such as
Really threshold value set by the axial distance between two exceptions, then the two abnormal areas are as independent abnormal area；
Step 222, the created identification model that passes through recognize abnormal area and judged, if recognition result belongs to defect,
Follow-up overlap test is carried out, step 24 is otherwise carried out, the overlap test principle states are as follows；If two identification model institutes
Output area D1And D2It can be merged, then：
Wherein, ξ rule of thumb values；
Step 223, to identified abnormal area carry out approximate center point detection be ACP detect, there are many defects to find out
The approximate center point ACP of each independent defect of the abnormal area in region；Specifically, carrying out net to the abnormal area first
Lattice are divided so that the average amplitude of each grid is AA, and the average amplitude matrix of whole region is G, that is, is caused：AAi,j∈ G,
And if there is ACP in current grid, then need to meet following condition：
Wherein, RGAnd CGDifference representing matrix G line number and columns；
Step 224, detected central point is identified, if abnormal area has ACP, proving may in abnormal area
Containing defective, then step 225 is carried out, if ACP is not present in abnormal area, the region is identified as non-defective region；
Step 225, split and recognized to defective abnormal area may be contained in step 224 in region, i.e., for one
Possess NAIndividual ACP multiple target region is abnormal area, first initialization sampling minimum windowFrame size is m1×m2
And new sampling forms are updated toWherein, L1Represent extension of the minimum axial both sides of sampling forms per side
Points, L2Represent that extension of the minimum circumferential both sides of sampling forms per side is counted；And be axially extended with circumferential, expanding
Zhan Shi, if there is new ACP in the forms newly extended, samples when front direction and stops and obtain when the border of front direction, when
When both direction all has border, the sampled point of current ACP points terminates, NACentered on put quantity；
Step 226, bimodal defect recognition is carried out to the abnormal area in step 225, according to close ACP can combination principle, update
ACP simultaneously samples to the ACP after renewal；Two close ACP are ACPiAnd ACPjFollowing two conditions should be met：
Wherein, C (ACPi,ACPj) represent ACPiAnd ACPjCircumferential distance, A (ACPx,ACPy) represent ACPxAnd ACPyAxially away from
From；ρ is experience value；
Step 227, pass through the abnormal area corresponding to created identification model identification step 226 and judged, if identification
As a result belong to defect, then carry out follow-up overlap test, otherwise carry out step 224, the overlap test principle states are as follows；Such as
Really two identification model institute output area D1And D2It can be merged, then：
It is preferred that the step 3 recruitment evaluation is carried out according to the result that finally recognizes, it carries out effect according to following interpretational criterias
Assess：
Classification and Identification effect assessment criterion 1：
Classification and Identification effect assessment criterion 2：
Wherein, TP is the number that defect is identified as defect；FN is the number that defect is identified as non-defective；TN is non-defective quilt
It is identified as the number of non-defective；FP is the number that non-defective is identified as defect；
Classification and Identification effect assessment criterion 3：
α2Take 0.3；
Boundary definition accuracy rate interpretational criteria is as follows：
Wherein, w represents output identification form region, and o represents to demarcate defect area, represents w ∩ o or w ∪ o, | | | | table
Registration strong point number.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710236188.3A CN107024532B (en) | 2017-04-12 | 2017-04-12 | Method for extracting defect positions of magnetic leakage pipeline based on window features |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710236188.3A CN107024532B (en) | 2017-04-12 | 2017-04-12 | Method for extracting defect positions of magnetic leakage pipeline based on window features |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107024532A true CN107024532A (en) | 2017-08-08 |
| CN107024532B CN107024532B (en) | 2019-12-27 |
Family
ID=59526708
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710236188.3A Expired - Fee Related CN107024532B (en) | 2017-04-12 | 2017-04-12 | Method for extracting defect positions of magnetic leakage pipeline based on window features |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107024532B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110082424A (en) * | 2019-05-10 | 2019-08-02 | 东北大学 | A kind of multiple dimensioned abnormal area recommender system of rapid pipeline magnetic flux leakage data and method |
| CN110516589A (en) * | 2019-08-26 | 2019-11-29 | 东北大学 | A kind of boundary precise recognition method of pipe leakage data |
| WO2020261554A1 (en) * | 2019-06-28 | 2020-12-30 | Ｊｆｅスチール株式会社 | Round bar surface defect inspection method and surface defect inspection device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100002929A1 (en) * | 2004-05-13 | 2010-01-07 | The Charles Stark Draper Laboratory, Inc. | Image-based methods for measuring global nuclear patterns as epigenetic markers of cell differentiation |
| CN102122351A (en) * | 2011-03-01 | 2011-07-13 | 哈尔滨工程大学 | Intelligent identification method for pipeline defect on basis of RBF (Radical Basis Function) neural network |
| CN104034794A (en) * | 2014-06-12 | 2014-09-10 | 东北大学 | Extreme learning machine-based pipeline magnetic flux leakage defect detection method |
| CN104514987A (en) * | 2014-12-19 | 2015-04-15 | 清华大学 | Three-dimensional pipeline flux leakage imaging defect quantizing method |
| CN106018545A (en) * | 2016-06-29 | 2016-10-12 | 东北大学 | Pipeline defect magnetic flux leakage inversion method based on Adaboost-RBF synergy |
-
2017
- 2017-04-12 CN CN201710236188.3A patent/CN107024532B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100002929A1 (en) * | 2004-05-13 | 2010-01-07 | The Charles Stark Draper Laboratory, Inc. | Image-based methods for measuring global nuclear patterns as epigenetic markers of cell differentiation |
| CN102122351A (en) * | 2011-03-01 | 2011-07-13 | 哈尔滨工程大学 | Intelligent identification method for pipeline defect on basis of RBF (Radical Basis Function) neural network |
| CN104034794A (en) * | 2014-06-12 | 2014-09-10 | 东北大学 | Extreme learning machine-based pipeline magnetic flux leakage defect detection method |
| CN104514987A (en) * | 2014-12-19 | 2015-04-15 | 清华大学 | Three-dimensional pipeline flux leakage imaging defect quantizing method |
| CN106018545A (en) * | 2016-06-29 | 2016-10-12 | 东北大学 | Pipeline defect magnetic flux leakage inversion method based on Adaboost-RBF synergy |
Non-Patent Citations (4)
| Title |
|---|
| MARYAM RAVAN1,REZA K. AMINEH1, SLAWOMIR KOZIEL2, NATALIA K. NIKO: "《Sizing of Multiple Cracks Using Magnetic Flux Leakage Measurements》", 《IET SCIENCE, MEASUREMENT & TECHNOLOGY》 * |
| WIJERATHNA, BUDDHI; VIDAL-CALLEJA, TERESA; KODAGODA, SARATH: "《Multiple defect interpretation based on Gaussian processes for MFL technology》", 《SPIE 8694,NONDESTRUCTIVE CHARACTERIZATION FOR COMPOSITE MATERIALS, AEROSPACE ENGINEERING, CIVIL INFRASTRUCTURE, AND HOMELAND SECURITY》 * |
| 刘金海: "《基于漏磁内检测的缺陷识别方法》", 《仪器仪表学报》 * |
| 赵重阳: "管道漏磁检测数据特征提取及特征分析方法研究", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110082424A (en) * | 2019-05-10 | 2019-08-02 | 东北大学 | A kind of multiple dimensioned abnormal area recommender system of rapid pipeline magnetic flux leakage data and method |
| WO2020228044A1 (en) * | 2019-05-10 | 2020-11-19 | 东北大学 | Multiscale abnormal region rapid recommendation system and method for pipeline magnetic flux leakage data |
| WO2020261554A1 (en) * | 2019-06-28 | 2020-12-30 | Ｊｆｅスチール株式会社 | Round bar surface defect inspection method and surface defect inspection device |
| JPWO2020261554A1 (en) * | 2019-06-28 | 2021-11-04 | Ｊｆｅスチール株式会社 | Surface defect inspection method and surface defect inspection device for round bars |
| JP7107439B2 (en) | 2019-06-28 | 2022-07-27 | Ｊｆｅスチール株式会社 | Round bar surface flaw inspection method and surface flaw inspection device |
| CN110516589A (en) * | 2019-08-26 | 2019-11-29 | 东北大学 | A kind of boundary precise recognition method of pipe leakage data |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107024532B (en) | 2019-12-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Hou et al. | Deep features based on a DCNN model for classifying imbalanced weld flaw types | |
| Malof et al. | Automatic detection of solar photovoltaic arrays in high resolution aerial imagery | |
| Chen et al. | A self organizing map optimization based image recognition and processing model for bridge crack inspection | |
| Xiao et al. | Surface defect detection using image pyramid | |
| Rai et al. | A novel pipeline leak detection approach independent of prior failure information | |
| Song et al. | Gas leak detection in galvanised steel pipe with internal flow noise using convolutional neural network | |
| CN102063720B (en) | Treelets-based method for detecting remote sensing image changes | |
| Chen et al. | Bottom-up image detection of water channel slope damages based on superpixel segmentation and support vector machine | |
| CN102867195B (en) | Method for detecting and identifying a plurality of types of objects in remote sensing image | |
| CN107024532A (en) | A kind of leakage field defect of pipeline position extracting method based on forms feature | |
| CN108447055A (en) | SAR image change detection based on SPL and CCN | |
| CN114581764B (en) | Underground structure crack disease discriminating method based on deep learning algorithm | |
| Ye et al. | Diagnosis of sewer pipe defects on image recognition of multi-features and support vector machine in a southern Chinese city | |
| CN113409314A (en) | Unmanned aerial vehicle visual detection and evaluation method and system for corrosion of high-altitude steel structure | |
| Harmouche et al. | Long-term monitoring for leaks in water distribution networks using association rules mining | |
| Guan et al. | Automatic fault diagnosis algorithm for hot water pipes based on infrared thermal images | |
| Dang et al. | Deep learning-based masonry crack segmentation and real-life crack length measurement | |
| CN107516082A (en) | Based on the SAR image change region detection method from step study | |
| CN111753877B (en) | Product quality detection method based on deep neural network migration learning | |
| Li et al. | A two-stage crack detection method for concrete bridges using Convolutional Neural Networks | |
| CN105374047A (en) | Improved bilateral filtering and clustered SAR based image change detection method | |
| Jiang et al. | Anomaly detection of industrial multi-sensor signals based on enhanced spatiotemporal features | |
| CN103971362A (en) | Synthetic aperture radar (SAR) imagine change detection based on histogram and elite genetic clustering algorithm | |
| CN110057588B (en) | Bearing early fault detection and diagnosis method and system based on fusion of singular value and graph theory characteristics | |
| Fakhri et al. | Road crack detection using gaussian/prewitt filter |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20191227 |