CN108154498A - A kind of rift defect detecting system and its implementation - Google Patents
A kind of rift defect detecting system and its implementation Download PDFInfo
- Publication number
- CN108154498A CN108154498A CN201711276677.8A CN201711276677A CN108154498A CN 108154498 A CN108154498 A CN 108154498A CN 201711276677 A CN201711276677 A CN 201711276677A CN 108154498 A CN108154498 A CN 108154498A
- Authority
- CN
- China
- Prior art keywords
- image
- subgraph
- value
- detected
- gray
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
- G06T7/0006—Industrial image inspection using a design-rule based approach
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30132—Masonry; Concrete
Abstract
The invention discloses a kind of rift defect detecting system and its implementation, system includes image deflects judgment module and Crack Detection module, implementation method include:Image to be detected is acquired by rift defect detecting system, judges to whether there is defect in image to be detected of acquisition by threshold value comparison method;Whether the defects of being classified to the characteristic results of defect using the method for support vector machines, determine image to be detected is crack.The present invention greatly reduces cost of labor and improves the precision of detection compared to traditional artificial detection method;In addition, the defects of present invention is compared to existing use ultrasonic Detection Method and fiber laser arrays method detection method, reduces operation difficulty, it is more convenient.The present invention is used as a kind of rift defect detecting system and its implementation, can be widely applied to defects detection field.
Description
Technical field
The present invention relates to defects detection field, especially a kind of rift defect detecting system and its implementation.
Background technology
Crack is the most common type defect in reinforced concrete structure, and the presence in crack can seriously affect the impervious of structure
Performance causes moisture and harmful substance to be penetrated into, and induces steel bar corrosion or accelerates the natural aging of concrete, so as to seriously damage work
The bearing capacity of journey structure, has an impact safety.Even if not yet directly affecting safe to use, applicability and resistance to is also resulted in
The decline of long property.Wherein, fracture width detection can not be ignored in Crack Detection, often assessed as component, delivery receiving acceptance,
Accident identifies the important evidence remedied with repair.And nuclear island factory building (includes reactor building, fuel plant, electrical building and core
Auxiliary plant) be entire nuclear power plant nucleus, crack not only influences the bearing capacity of nuclear island factory building structure in itself, but also and core
Safety is closely related, and therefore, the stringent fracture width for controlling nuclear island factory building concrete wall is of great significance.At present, it reacts
The fracture width detection of the containment concrete of heap workshop is typically to be completed in the containment integrality test of 10 years by a definite date, phase
Adjacent check time interval twice is longer, and can not cover entire nuclear island factory building region, and the control and tracking for crack are more not
Profit.To sum up, there is an urgent need for change the method and technology of existing nuclear power plant concrete fracture width detection.
At present, the method for testing concrete defect is mainly the following:(1) it is real by clearance gauge or defect width versus card
Existing defect width measurement;(2) defect width measurement is realized by defect microscope, i.e., using micro- with certain amplification factor
Mirror directly observes defect width;(3) it is artificial to read defect width by display defect image;(4) by supercritical ultrasonics technology to defect
Depth and width be detected;(5) optical fiber method, by being embedded in structure or being attached to structure table by fibre optical sensor in advance
On face, crack width and position are measured using the variation of light intensity.First three methods are required to manually be detected, labor intensity
It is larger, and there are larger errors in reading.Supercritical ultrasonics technology needs coupling agent to measure, inconvenient in practical operation, and in light
Fibre optical sensor is laid in method for fiber structure can make measurement more complicated.
Invention content
In order to solve the above technical problems, first purpose of the present invention is:There is provided that a kind of cost of labor is low, accuracy is high
And rift defect detecting system easy to operate.
Second object of the present invention is:There is provided that a kind of cost of labor is low, accuracy is high and crack easy to operate lacks
Fall into the implementation method of detecting system.
First technical solution being taken of the present invention be:
A kind of rift defect detecting system, including:
Image deflects judgment module, for being judged by threshold value comparison method in image to be detected of acquisition with the presence or absence of scarce
It falls into;
Crack Detection module, for being directed to the image of existing defects, using the method for support vector machines to the feature of defect
Whether the defects of as a result classifying, determining image to be detected is crack.
Further, described image defect dipoles module includes:
Preprocessing module pre-processes for the image to acquisition;
Subgraph gray scale difference characteristic value module for pretreated image to be divided into multiple subgraphs, and is extracted every
The gray scale difference characteristic value of a subgraph;
Color histogram feature module, for carrying out color histogram feature extraction to pretreated image;
Learning training module, for the color histogram feature according to extraction and gray scale difference characteristic value, training is averaged
Gray scale difference, standard deviation and average color histogram;
Threshold value comparison module, for the handling result according to learning training module and the threshold value T of settingB, calculate to be detected
The parameter difference of image and standard picture, if parameter difference is more than given threshold TB, then image to be detected existing defects;Conversely, it then treats
Defect is not present in detection image, wherein, parameter difference EBijCalculation formula be:
BijTRepresent that the average gray of standard picture is poor,Represent that the average gray of image to be detected is poor, T represents setting
Threshold value.
Further, the subgraph gray scale difference characteristic value module includes:
Image split cells, for pretreated image to be divided into equal-sized multiple subgraphs;
Subgraph split cells, for each subgraph to be divided into the identical multiple figures of number of pixels in the width direction
Picture;
Translation unit, for each image for obtaining subgraph split cells decile respectively along width direction and height
Direction translates N-1 times, wherein, N is the image number that subgraph split cells decile obtains;
Effective subgraph number computing unit, for according to translation as a result, calculating effective subgraph number, institute
The calculation formula for stating effective subgraph number is:
Wherein, NshIt represents along the subgraph number in the short transverse of pretreated image, NswIt represents along after pretreatment
Image width direction on subgraph number, S represent shape be size S × S square subgraph the length of side, H is pre-
The height of treated image, W are the width of pretreated image.
Further, the subgraph gray scale difference characteristic value module further includes:
Gray value sum calculation unit for the subgraph number according to calculating, calculates the gray value summation of subgraph, institute
The calculation formula for stating gray value summation is:
K=0,1,2,3 ..., N × N-1,
I=0,1,2,3 ..., Nsh- 1,
J=0,1,2,3 ..., Nsw- 1,
Wherein, AijkK-th of gray value summation of subgraph (i, j) is represented, (i, j) represents pretreated image neutron
The position of image, f (x, y) represent pretreated image;
Gray scale difference characteristic value computing unit for the gray value summation according to calculating, calculates the gray scale difference of each subgraph
Characteristic value, the calculation formula of the gray scale difference characteristic value are:
Bij=max (Aijk)-min(Aijk),
Wherein, BijRepresent the gray scale difference characteristic value of subgraph (i, j), max (Aijk) represent AijkMaximum value, min (Aijk)
Represent AijkMinimum value.
Further, the color histogram feature module includes:
Interval division unit, for 0 to 255 image pixel intensities section in single channel to be divided equally into multiple areas
Between;
Image pixel intensities computing unit, the number of pixels in each section obtained after being divided for computation interval division unit,
Wherein, number of pixels IkExpression formula be:
S is the image pixel intensities that each section includes, and k represents k-th of image pixel intensities section.
Further, the learning training module includes:
First training subelement, for the gray scale difference characteristic value according to subgraph, calculate subgraph average gray difference and
Standard variance, the average gray are poorWith standard variance σBCalculation formula be:
Second training subelement, for the image pixel intensities number according to each pixel range, calculates tri- channels of RGB respectively
Mean value, obtains average color histogram, and the mean value computation formula of each Color Channel is:
Wherein, nckRepresent the number of pixels in k-th of section of c-th of channel, Z represents the number of pixel range.
Further, the Crack Detection module includes:
Profiler construction unit, it is empty for be converted into higher-dimension there will be the image of defect using non-linear transformation method
Between image, then solve higher dimensional space image in optimum linearity classification plane, structure obtain profiler;
Crack Detection unit, for the profiler according to structure, whether the defects of determining image to be detected is crack.
Second technical solution being taken of the present invention be:
A kind of implementation method of rift defect detecting system, includes the following steps:
Image to be detected is acquired by rift defect detecting system, image to be detected of acquisition is judged by threshold value comparison method
In whether there is defect;
Whether the defects of being classified to the characteristic results of defect using the method for support vector machines, determine image to be detected
For crack.
Further, it is described that image to be detected is acquired by rift defect detecting system, judge to acquire by threshold value comparison method
Image to be detected in whether there is defect the step for, include the following steps:
The image of acquisition is pre-processed;
Pretreated image is divided into multiple subgraphs, and extracts the gray scale difference characteristic value of each subgraph;
Color histogram feature extraction is carried out to pretreated image;
According to the color histogram feature of extraction and gray scale difference characteristic value, training obtains that average gray is poor, standard deviation is peaceful
Equal color histogram;
According to the handling result of learning training module and the threshold value T of settingB, calculate image to be detected and the ginseng of standard picture
Number is poor, if parameter difference is more than given threshold TB, then image to be detected existing defects;Conversely, then defect is not present in image to be detected,
Wherein, parameter difference EBijCalculation formula be:
BijTRepresent that the average gray of standard picture is poor,Represent that the average gray of image to be detected is poor, T represents setting
Threshold value.
Further, the gray scale difference that pretreated image is divided into multiple subgraphs, and extracts each subgraph
The step for characteristic value, includes the following steps:
Pretreated image is divided into equal-sized multiple subgraphs;
Each subgraph is divided into the identical multiple images of number of pixels in the width direction;
Each image that subgraph split cells decile obtains is translated into N-1 respectively along width direction and short transverse
It is secondary, wherein, N is the image number that subgraph split cells decile obtains;
According to translation as a result, calculating effective subgraph number, the calculating of the effective subgraph number is public
Formula is:
Wherein, NshIt represents along the subgraph number in the short transverse of pretreated image, NswIt represents along after pretreatment
Image width direction on subgraph number, S represent shape be size S × S square subgraph the length of side, H is pre-
The height of treated image, W are the width of pretreated image;
According to the subgraph number of calculating, the gray value summation of subgraph, the calculation formula of the gray value summation are calculated
For:
K=0,1,2,3 ..., N × N-1,
I=0,1,2,3 ..., Nsh- 1,
J=0,1,2,3 ..., Nsw- 1,
Wherein, AijkK-th of gray value summation of subgraph (i, j) is represented, (i, j) represents pretreated image neutron
The position of image, f (x, y) represent pretreated image;
According to the gray value summation of calculating, the gray scale difference characteristic value of each subgraph is calculated, the gray scale difference characteristic value
Calculation formula is:
Bij=max (Aijk)-min(Aijk),
Wherein, BijRepresent the gray scale difference characteristic value of subgraph (i, j), max (Aijk) represent AijkMaximum value, min (Aijk)
Represent AijkMinimum value.
The advantageous effect of system of the present invention is:The system of the present invention passes through threshold first with image deflects judgment module
It is worth comparison method to judge to whether there is defect in image to be detected of acquisition, image to be detected is then determined by Crack Detection module
The defects of whether be crack, compared to traditional artificial detection method, system of the invention greatly reduces cost of labor and carries
The high precision of detection;In addition, the system of the present invention is compared to existing lacking using ultrasonic Detection Method and fiber laser arrays method
Detecting system is fallen into, reduces operation difficulty, it is more convenient.
The advantageous effect of implementation method of the present invention is:The method of the present invention is acquired to be checked using rift defect detecting system
Altimetric image is judged in image to be detected of acquisition with the presence or absence of defect by threshold value comparison method, then using support vector machines
Whether the defects of method classifies to the characteristic results of defect, determines image to be detected is crack, artificial compared to traditional
Detection method, method of the invention greatly reduce cost of labor and improve the precision of detection;In addition, the method for the present invention
Compared to existing ultrasonic Detection Method and fiber laser arrays method, operation difficulty is reduced, it is more convenient.
Description of the drawings
Fig. 1 is a kind of overall structure block diagram of rift defect detecting system of the present invention.
Specific embodiment
A kind of rift defect detecting system of the present invention, including:
Image deflects judgment module, for being judged by threshold value comparison method in image to be detected of acquisition with the presence or absence of scarce
It falls into;
Crack Detection module, for being directed to the image of existing defects, using the method for support vector machines to the feature of defect
Whether the defects of as a result classifying, determining image to be detected is crack.
Preferred embodiment is further used as, described image defect dipoles module includes:
Preprocessing module pre-processes for the image to acquisition;
Subgraph gray scale difference characteristic value module for pretreated image to be divided into multiple subgraphs, and is extracted every
The gray scale difference characteristic value of a subgraph;
Color histogram feature module, for carrying out color histogram feature extraction to pretreated image;
Learning training module, for the color histogram feature according to extraction and gray scale difference characteristic value, training is averaged
Gray scale difference, standard deviation and average color histogram;
Threshold value comparison module, for the handling result according to learning training module and the threshold value T of settingB, calculate to be detected
The parameter difference of image and standard picture, if parameter difference is more than given threshold TB, then image to be detected existing defects;Conversely, it then treats
Defect is not present in detection image, wherein, parameter difference EBijCalculation formula be:
BijTRepresent that the average gray of standard picture is poor,Represent that the average gray of image to be detected is poor, T represents setting
Threshold value.
Wherein, the pretreatment includes the pretreatment modes such as Morphological scale-space and filtering operation.
Preferred embodiment is further used as, the subgraph gray scale difference characteristic value module includes:
Image split cells, for pretreated image to be divided into equal-sized multiple subgraphs;
Subgraph split cells, for each subgraph to be divided into the identical multiple figures of number of pixels in the width direction
Picture;
Translation unit, for each image for obtaining subgraph split cells decile respectively along width direction and height
Direction translates N-1 times, wherein, N is the image number that subgraph split cells decile obtains;
Effective subgraph number computing unit, for according to translation as a result, calculating effective subgraph number, institute
The calculation formula for stating effective subgraph number is:
Wherein, NshIt represents along the subgraph number in the short transverse of pretreated image, NswIt represents along after pretreatment
Image width direction on subgraph number, S represent shape be size S × S square subgraph the length of side, H is pre-
The height of treated image, W are the width of pretreated image.
Preferred embodiment is further used as, the subgraph gray scale difference characteristic value module further includes:
Gray value sum calculation unit for the subgraph number according to calculating, calculates the gray value summation of subgraph, institute
The calculation formula for stating gray value summation is:
K=0,1,2,3 ..., N × N-1,
I=0,1,2,3 ..., Nsh- 1,
J=0,1,2,3 ..., Nsw- 1,
Wherein, AijkK-th of gray value summation of subgraph (i, j) is represented, (i, j) represents pretreated image neutron
The position of image, f (x, y) represent pretreated image;
Gray scale difference characteristic value computing unit for the gray value summation according to calculating, calculates the gray scale difference of each subgraph
Characteristic value, the calculation formula of the gray scale difference characteristic value are:
Bij=max (Aijk)-min(Aijk),
Wherein, BijRepresent the gray scale difference characteristic value of subgraph (i, j), max (Aijk) represent AijkMaximum value, min (Aijk)
Represent AijkMinimum value.
Preferred embodiment is further used as, the color histogram feature module includes:
Interval division unit, for 0 to 255 image pixel intensities section in single channel to be divided equally into multiple areas
Between;
Image pixel intensities computing unit, the number of pixels in each section obtained after being divided for computation interval division unit,
Wherein, number of pixels IkExpression formula be:
S is the image pixel intensities that each section includes, and k represents k-th of image pixel intensities section.
Wherein, color histogram feature is pixel value in three channels in the RGB models according to the coloured image of image
It is distributed to describe, rgb space is most commonly used to description coloured image.
Preferred embodiment is further used as, the learning training module includes:
First training subelement, for the gray scale difference characteristic value according to subgraph, calculate subgraph average gray difference and
Standard variance, the average gray are poorWith standard variance σBCalculation formula be:
Second training subelement, for the image pixel intensities number according to each pixel range, calculates tri- channels of RGB respectively
Mean value, obtains average color histogram, and the mean value computation formula of each Color Channel is:
Wherein, nckRepresent the number of pixels in k-th of section of c-th of channel, Z represents the number of pixel range.
Preferred embodiment is further used as, the Crack Detection module includes:
Profiler construction unit, it is empty for be converted into higher-dimension there will be the image of defect using non-linear transformation method
Between image, then solve higher dimensional space image in optimum linearity classification plane, structure obtain profiler;
Crack Detection unit, for the profiler according to structure, whether the defects of determining image to be detected is crack.
In solving practical problems, most of sample sets are all linearly inseparables in luv space, are usually used
The sample of luv space is mapped in high-dimensional feature space by the method for Nonlinear Mapping, enables sample in this feature space
Middle linear separability.Linear learning device needs to select a suitable nonlinear characteristic collection when learning a nonlinear relationship,
And it needs to be write data into as new expression formula.Assuming that there are sample (xi,yi), i=1,2 ..., l, x ∈ Rd, y ∈ { -1,1 }, l
For sample number, d is input dimension, then linear discriminant function expression formula is:
Wherein, ω is weight vector, φ:X → F is the mapping from the input space to some feature space, and b is the ginseng of setting
Number.The foundation of one Nonlinear Learning device needs to map the data into feature space using a Nonlinear Mapping, then at this
Feature space is classified using linear learning device.
The significant characteristic of one of linear learning device is can be expressed as dual form, decision rule can use test point and
The inner products of training points represents:
Wherein αiFor the Lagrange multiplier of each sample,<·>It is inner product.
In the case that linear inseparable, optimization problem can be described as:
s.t.yi[<ω·xi>+b]≥1-ξi,
ξi>=0, i=1,2 ..., l,
Wherein, ξ is slack variable, and C is appropriate punishment parameter.
Therefore, the input space is mapped to the optimization problem after some feature space and can be described as:
s.t.yi[<ω·φi(x)>+b]≥1-ξi,
ξi>=0, i=1,2 ..., l,
Decision function can be expressed as:
Wherein,<φi(x),φ(x)>For kernel function.
By extracting rift defect feature, characteristic is normalized, classification results can be made more accurate, so
After start to carry out classification based training to experimental data, choose the feature vector of training set, build svm classifier model, by setting not
With parameter different graders is selected to be trained, finally sort merge is got up to can be formed by profiler, with
The defects of detecting surface is crack.
A kind of implementation method of rift defect detecting system of the present invention, includes the following steps:
Image to be detected is acquired by rift defect detecting system, image to be detected of acquisition is judged by threshold value comparison method
In whether there is defect;
Whether the defects of being classified to the characteristic results of defect using the method for support vector machines, determine image to be detected
For crack.
Preferred embodiment is further used as, it is described that image to be detected is acquired by rift defect detecting system, pass through
The step for threshold value comparison method judges to whether there is defect in image to be detected of acquisition, includes the following steps:
The image of acquisition is pre-processed;
Pretreated image is divided into multiple subgraphs, and extracts the gray scale difference characteristic value of each subgraph;
Color histogram feature extraction is carried out to pretreated image;
According to the color histogram feature of extraction and gray scale difference characteristic value, training obtains that average gray is poor, standard deviation is peaceful
Equal color histogram;
According to the handling result of learning training module and the threshold value T of settingB, calculate image to be detected and the ginseng of standard picture
Number is poor, if parameter difference is more than given threshold TB, then image to be detected existing defects;Conversely, then defect is not present in image to be detected,
Wherein, parameter difference EBijCalculation formula be:
BijTRepresent that the average gray of standard picture is poor,Represent that the average gray of image to be detected is poor, T represents setting
Threshold value.
Preferred embodiment is further used as, it is described that pretreated image is divided into multiple subgraphs, and extract
The step for gray scale difference characteristic value of each subgraph, include the following steps:
Pretreated image is divided into equal-sized multiple subgraphs;
Each subgraph is divided into the identical multiple images of number of pixels in the width direction;
Each image that subgraph split cells decile obtains is translated into N-1 respectively along width direction and short transverse
It is secondary, wherein, N is the image number that subgraph split cells decile obtains;
According to translation as a result, calculating effective subgraph number, the calculating of the effective subgraph number is public
Formula is:
Wherein, NshIt represents along the subgraph number in the short transverse of pretreated image, NswIt represents along after pretreatment
Image width direction on subgraph number, S represent shape be size S × S square subgraph the length of side, H is pre-
The height of treated image, W are the width of pretreated image;
According to the subgraph number of calculating, the gray value summation of subgraph, the calculation formula of the gray value summation are calculated
For:
K=0,1,2,3 ..., N × N-1,
I=0,1,2,3 ..., Nsh- 1,
J=0,1,2,3 ..., Nsw- 1,
Wherein, AijkK-th of gray value summation of subgraph (i, j) is represented, (i, j) represents pretreated image neutron
The position of image, f (x, y) represent pretreated image;
According to the gray value summation of calculating, the gray scale difference characteristic value of each subgraph is calculated, the gray scale difference characteristic value
Calculation formula is:
Bij=max (Aijk)-min(Aijk),
Wherein, BijRepresent the gray scale difference characteristic value of subgraph (i, j), max (Aijk) represent AijkMaximum value, min (Aijk)
Represent AijkMinimum value.
The present invention is further explained and illustrated with specific embodiment with reference to the accompanying drawings of the specification.For of the invention real
The step number in example is applied, is set only for the purposes of illustrating explanation, the sequence between step does not do any restriction, implements
The execution sequence of each step in example can be adaptively adjusted according to the understanding of those skilled in the art.
Embodiment one
For existing concrete rift defect detection method high labor cost, accuracy of detection it is low, operation complexity is high etc.
Problem, the present invention propose a kind of rift defect detecting system and its implementation.The system of the present invention is lacked first by image
It falls into judgment module to judge to whether there is defect in image to be detected of acquisition by threshold value comparison method, then passes through Crack Detection mould
Block determines whether the defects of image to be detected is crack, and compared to traditional artificial detection method, system of the invention drops significantly
Low cost of labor and the precision for improving detection;In addition, the system of the present invention uses ultrasonic Detection Method compared to existing
With detecting system the defects of fiber laser arrays method, operation difficulty is reduced, it is more convenient.
With reference to Fig. 1, a kind of rift defect detecting system of the present invention includes image deflects judgment module and Crack Detection module.
Image deflects judgment module includes preprocessing module, color histogram feature module, subgraph gray scale difference characteristic value module, study
Training module and threshold value comparison module;Color histogram feature module includes interval division unit and image pixel intensities computing unit;
Subgraph gray scale difference characteristic value module includes image split cells, subgraph split cells, translation unit, effective subgraph number
Computing unit, gray value sum calculation unit and gray scale difference characteristic value computing unit;Learning training module includes the first training
Unit and the second training subelement.Crack Detection module includes profiler construction unit and Crack Detection unit.
Wherein, preprocessing module is connect respectively with interval division unit and image split cells, standard picture split cells
Be sequentially connected subgraph split cells, translation unit, effective subgraph number computing unit, gray value sum calculation unit and
Gray scale difference characteristic value computing unit, gray scale difference characteristic value computing unit are connect with the first training subelement, and image pixel intensities calculate single
Member is connect with the second training subelement, and the first training subelement and the second training subelement are connect with threshold value comparison module, threshold
Value comparison module is connect with profiler construction unit, and profiler construction unit is connect with Crack Detection unit.
A kind of specific steps flow of the implementation method of rift defect detecting system of the present invention is as follows:
S1, image to be detected is acquired by rift defect detecting system, the to be detected of acquisition is judged by threshold value comparison method
It whether there is defect in image;
Wherein, the step S1 specifically includes following steps:
S11, the image of acquisition is pre-processed, the pretreatment includes filtering operation;
S12, color histogram feature extraction is carried out to pretreated image;
S13, pretreated image is divided into multiple subgraphs, and extracts the gray scale difference characteristic value of each subgraph;
S14, according to the color histogram feature of extraction and gray scale difference characteristic value, training obtains that average gray is poor, standard deviation
With average color histogram;
S15, according to the handling result of learning training module and the threshold value T of settingB, calculate image to be detected and standard picture
Parameter difference, if parameter difference is more than given threshold TB, then image to be detected existing defects;Conversely, then there is no lack for image to be detected
It falls into, wherein, parameter difference EBijCalculation formula be:
BijTRepresent that the average gray of standard picture is poor,Represent that the average gray of image to be detected is poor, T represents setting
Threshold value plays the role of mark.
Wherein, the step S12 specifically includes following steps:
S121,0 to 255 image pixel intensities section in single channel is divided equally into multiple sections;
The number of pixels in each section that S122, computation interval division unit obtain after dividing, wherein, number of pixels Ik's
Expression formula is:
S is the image pixel intensities that each section includes, and k represents k-th of image pixel intensities section.
The step S13 specifically includes following steps:
S131, pretreated image is divided into equal-sized multiple subgraphs;
S132, each subgraph is divided into the identical multiple images of number of pixels in the width direction;
S133, each image that subgraph split cells decile obtains is translated respectively along width direction and short transverse
N-1 times, wherein, N is the image number that subgraph split cells decile obtains;
S134, effective subgraph number computing unit, for according to translation as a result, calculating effective subgraph
Number, the calculation formula of the effective subgraph number are:
Wherein, NshIt represents along the subgraph number in the short transverse of pretreated image, NswIt represents along after pretreatment
Image width direction on subgraph number, S represent shape be size S × S square subgraph the length of side, H is pre-
The height of treated image, W are the width of pretreated image;
S135, the subgraph number according to calculating calculate the gray value summation of subgraph, the calculating of the gray value summation
Formula is:
K=0,1,2,3 ..., N × N-1,
I=0,1,2,3 ..., Nsh- 1,
J=0,1,2,3 ..., Nsw- 1,
Wherein, AijkK-th of gray value summation of subgraph (i, j) is represented, (i, j) represents pretreated image neutron
The position of image, f (x, y) represent pretreated image;
S136, the gray value summation according to calculating calculate the gray scale difference characteristic value of each subgraph, the gray scale difference feature
The calculation formula of value is:
Bij=max (Aijk)-min(Aijk),
Wherein, BijRepresent the gray scale difference characteristic value of subgraph (i, j), max (Aijk) represent AijkMaximum value, min (Aijk)
Represent AijkMinimum value.
The step S14 specifically includes following steps:
S141, the gray scale difference characteristic value according to subgraph calculate the average gray difference and standard variance of subgraph, described flat
Equal gray scale differenceWith standard variance σBCalculation formula be:
S142, the image pixel intensities number according to each pixel range calculate the mean value of tri- channels of RGB, are averaged respectively
Color histogram, the mean value computation formula of each Color Channel are:
Wherein, nckRepresent the number of pixels in k-th of section of c-th of channel, Z represents the number of pixel range.
S2, according to training study as a result, to image to be detected carry out rift defect detection.
Wherein, the step S2 specifically includes following steps:
S21, defect classification is carried out to the result of training study using binary-tree support vector machine, wherein, defect classification packet
Include the polishing cleaning of metope, wall cracking, impregnating with silane spraying, BONNA pipe surfaces crackle, surface deformation, hole, bolt rust
Erosion and bolt deformation;
Step S21 is specially:Using support vector machine method counterincision seam defect construction feature describer, i.e., split by extraction
Seam defect feature, is normalized characteristic, and classification results can be made more accurate, then start to experimental data into
Row classification based training chooses the feature vector of training set, builds svm classifier model, difference is selected by setting different parameters
Grader be trained, finally sort merge is got up to can be formed by profiler.
S22, according to defect classify result and setting threshold value, image to be detected and standard picture are compared behaviour
Make, determine image to be detected defect classification.
Step S22 is specially:Whether it is to split according to the defects of building obtained profiler, determine image to be detected
Seam.
In conclusion a kind of rift defect detecting system of the present invention and its implementation have the following advantages:
1), rift defect detecting system of the invention does not need to human intervention, automatically controls each module in itself by system
Orderly operation compared to traditional artificial detection method, greatly reduces cost of labor and improves the precision of detection.
2) the defects of, system of the invention is compared to existing use ultrasonic Detection Method and fiber laser arrays method detecting system,
Operation difficulty is reduced, it is more convenient.
3), detecting system can identify polishing, cleaning, impregnating with silane spraying effect, crack, list of bolts the defects of the present invention
The defects of face corrosion and deformation, common all defect situation is covered, is not omitted, detection is more comprehensive.
4), the present invention is due to being integrated with the modules such as preprocessing module, image deflects judgment module and Crack Detection module, energy
Enough in real time, the problem of being detected incessantly to region to be checked, can find pipeline in time, real-time is high.
It is that the preferable of the present invention is implemented to be illustrated, but the present invention is not limited to the embodiment above, it is ripe
Various equivalent variations or replacement can also be made under the premise of without prejudice to spirit of the invention by knowing those skilled in the art, this
Equivalent deformation or replacement are all contained in the application claim limited range a bit.
Claims (10)
1. a kind of rift defect detecting system, it is characterised in that:Including:
Image deflects judgment module, for judging to whether there is defect in image to be detected of acquisition by threshold value comparison method;
Crack Detection module, for being directed to the image of existing defects, using the method for support vector machines to the characteristic results of defect
Whether the defects of classifying, determining image to be detected is crack.
2. a kind of rift defect detecting system according to claim 1, it is characterised in that:Described image defect dipoles module
Including:
Preprocessing module pre-processes for the image to acquisition;
Subgraph gray scale difference characteristic value module for pretreated image to be divided into multiple subgraphs, and is extracted per height
The gray scale difference characteristic value of image;
Color histogram feature module, for carrying out color histogram feature extraction to pretreated image;
Learning training module, for the color histogram feature according to extraction and gray scale difference characteristic value, training obtains average gray
Difference, standard deviation and average color histogram;
Threshold value comparison module, for the handling result according to learning training module and the threshold value T of settingB, calculate image to be detected with
The parameter difference of standard picture, if parameter difference is more than given threshold TB, then image to be detected existing defects;Conversely, then mapping to be checked
As defect is not present, wherein, parameter difference EBijCalculation formula be:
BijTRepresent that the average gray of standard picture is poor,Represent that the average gray of image to be detected is poor, T represents the threshold value of setting.
3. a kind of rift defect detecting system according to claim 2, it is characterised in that:The subgraph gray scale difference feature
Value module includes:
Image split cells, for pretreated image to be divided into equal-sized multiple subgraphs;
Subgraph split cells, for each subgraph to be divided into the identical multiple images of number of pixels in the width direction;
Translation unit, for each image for obtaining subgraph split cells decile respectively along width direction and short transverse
Translation N-1 times, wherein, N is the image number that subgraph split cells decile obtains;
Effective subgraph number computing unit, for according to translation as a result, calculate effective subgraph number, it is described to have
The calculation formula of the subgraph number of effect is:
Wherein, NshIt represents along the subgraph number in the short transverse of pretreated image, NswIt represents along pretreated figure
Subgraph number in the width direction of picture, S represent the length of side for the square subgraph that shape is size S × S, and H is pretreatment
The height of image afterwards, W are the width of pretreated image.
4. a kind of rift defect detecting system according to claim 3, it is characterised in that:The subgraph gray scale difference feature
Value module further includes:
Gray value sum calculation unit for the subgraph number according to calculating, calculates the gray value summation of subgraph, the ash
The calculation formula of angle value summation is:
K=0,1,2,3 ..., N × N-1,
I=0,1,2,3 ..., Nsh- 1,
J=0,1,2,3 ..., Nsw- 1,
Wherein, AijkK-th of gray value summation of subgraph (i, j) is represented, (i, j) represents pretreated image neutron image
Position, f (x, y) represents pretreated image;
Gray scale difference characteristic value computing unit for the gray value summation according to calculating, calculates the gray scale difference feature of each subgraph
Value, the calculation formula of the gray scale difference characteristic value are:
Bij=max (Aijk)-min(Aijk),
Wherein, BijRepresent the gray scale difference characteristic value of subgraph (i, j), max (Aijk) represent AijkMaximum value, min (Aijk) represent
AijkMinimum value.
5. a kind of rift defect detecting system according to claim 4, it is characterised in that:The color histogram feature mould
Block includes:
Interval division unit, for 0 to 255 image pixel intensities section in single channel to be divided equally into multiple sections;
Image pixel intensities computing unit, the number of pixels in each section obtained after being divided for computation interval division unit, wherein,
Number of pixels IkExpression formula be:
S is the image pixel intensities that each section includes, and k represents k-th of image pixel intensities section.
6. a kind of rift defect detecting system according to claim 5, it is characterised in that:The learning training module packet
It includes:
First training subelement for the gray scale difference characteristic value according to subgraph, calculates the average gray difference and standard of subgraph
Variance, the average gray are poorWith standard variance σBCalculation formula be:
Second training subelement, for the image pixel intensities number according to each pixel range, calculates the equal of tri- channels of RGB respectively
Value, obtains average color histogram, the mean value computation formula of each Color Channel is:
Wherein, nckRepresent the number of pixels in k-th of section of c-th of channel, Z represents the number of pixel range.
7. a kind of rift defect detecting system according to claim 1, it is characterised in that:The Crack Detection module packet
It includes:
Profiler construction unit, for there will be the images of defect to be converted into higher dimensional space figure using non-linear transformation method
Picture, then solves the optimum linearity classification plane in higher dimensional space image, and structure obtains profiler;
Crack Detection unit, for the profiler according to structure, whether the defects of determining image to be detected is crack.
8. a kind of implementation method of rift defect detecting system, which is characterized in that include the following steps:
Image to be detected is acquired by rift defect detecting system, is in image to be detected by threshold value comparison method judgement acquisition
No existing defects;
Whether the defects of being classified to the characteristic results of defect using the method for support vector machines, determine image to be detected is to split
Seam.
9. a kind of implementation method of rift defect detecting system according to claim 8, it is characterised in that:It is described by splitting
Seam defect detecting system acquires image to be detected, judges to whether there is defect in image to be detected of acquisition by threshold value comparison method
The step for, include the following steps:
The image of acquisition is pre-processed;
Pretreated image is divided into multiple subgraphs, and extracts the gray scale difference characteristic value of each subgraph;
Color histogram feature extraction is carried out to pretreated image;
According to the color histogram feature of extraction and gray scale difference characteristic value, training obtains poor average gray, standard deviation and average coloured silk
Color Histogram;
According to the handling result of learning training module and the threshold value T of settingB, image to be detected and the parameter difference of standard picture are calculated,
If parameter difference is more than given threshold TB, then image to be detected existing defects;Conversely, then defect is not present in image to be detected, wherein,
Parameter difference EBijCalculation formula be:
BijTRepresent that the average gray of standard picture is poor,Represent that the average gray of image to be detected is poor, T represents the threshold value of setting.
10. a kind of implementation method of rift defect detecting system according to claim 9, it is characterised in that:It is described to incite somebody to action in advance
Treated, and image is divided into multiple subgraphs, and the step for extract the gray scale difference characteristic value of each subgraph, including following
Step:
Pretreated image is divided into equal-sized multiple subgraphs;
Each subgraph is divided into the identical multiple images of number of pixels in the width direction;
Each image that subgraph split cells decile obtains is translated N-1 times respectively along width direction and short transverse,
In, N is the image number that subgraph split cells decile obtains;
According to translation as a result, calculating effective subgraph number, the calculation formula of the effective subgraph number is:
Wherein, NshIt represents along the subgraph number in the short transverse of pretreated image, NswIt represents along pretreated figure
Subgraph number in the width direction of picture, S represent the length of side for the square subgraph that shape is size S × S, and H is pretreatment
The height of image afterwards, W are the width of pretreated image;
According to the subgraph number of calculating, the gray value summation of subgraph is calculated, the calculation formula of the gray value summation is:
K=0,1,2,3 ..., N × N-1,
I=0,1,2,3 ..., Nsh- 1,
J=0,1,2,3 ..., Nsw- 1,
Wherein, AijkK-th of gray value summation of subgraph (i, j) is represented, (i, j) represents pretreated image neutron image
Position, f (x, y) represents pretreated image;
According to the gray value summation of calculating, the gray scale difference characteristic value of each subgraph, the calculating of the gray scale difference characteristic value are calculated
Formula is:
Bij=max (Aijk)-min(Aijk),
Wherein, BijRepresent the gray scale difference characteristic value of subgraph (i, j), max (Aijk) represent AijkMaximum value, min (Aijk) represent
AijkMinimum value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711276677.8A CN108154498A (en) | 2017-12-06 | 2017-12-06 | A kind of rift defect detecting system and its implementation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711276677.8A CN108154498A (en) | 2017-12-06 | 2017-12-06 | A kind of rift defect detecting system and its implementation |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108154498A true CN108154498A (en) | 2018-06-12 |
Family
ID=62466487
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711276677.8A Pending CN108154498A (en) | 2017-12-06 | 2017-12-06 | A kind of rift defect detecting system and its implementation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108154498A (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109187755A (en) * | 2018-09-12 | 2019-01-11 | 西安交通大学 | A kind of online supersonic damage-free detection method based on 3D printing |
CN109829883A (en) * | 2018-12-19 | 2019-05-31 | 歌尔股份有限公司 | Product quality detection method and device |
CN110221186A (en) * | 2019-06-13 | 2019-09-10 | 国网山东省电力公司青州市供电公司 | A kind of darkroom alternating-current voltage resistance test device and test method |
CN110399911A (en) * | 2019-07-12 | 2019-11-01 | 国网上海市电力公司 | A kind of transformer mechanical failure diagnostic method based on sound-filed simulation cloud atlas |
CN110619343A (en) * | 2018-06-20 | 2019-12-27 | 亚摩信息技术股份有限公司 | Automatic defect classification method based on machine learning |
CN112184622A (en) * | 2020-08-31 | 2021-01-05 | 广东三和管桩股份有限公司 | Pipe pile appearance quality inspection method and system, computer equipment and storage medium |
CN115861987A (en) * | 2023-02-27 | 2023-03-28 | 江苏天南电力股份有限公司 | Intelligent electric power fitting defect identification method for on-line monitoring of power transmission line |
CN116990310A (en) * | 2023-09-27 | 2023-11-03 | 中国水利水电第九工程局有限公司 | Wall concrete crack monitoring and early warning system based on data analysis |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101915764A (en) * | 2010-08-10 | 2010-12-15 | 武汉武大卓越科技有限责任公司 | Road surface crack detection method based on dynamic programming |
CN103258206A (en) * | 2012-11-28 | 2013-08-21 | 河海大学常州校区 | Silicon solar cell surface defect detection and identification method |
CN104458755A (en) * | 2014-11-26 | 2015-03-25 | 吴晓军 | Multi-type material surface defect detection method based on machine vision |
CN105320962A (en) * | 2015-10-21 | 2016-02-10 | 东南大学 | Pavement damage type identification method based on classifier ensemble |
-
2017
- 2017-12-06 CN CN201711276677.8A patent/CN108154498A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101915764A (en) * | 2010-08-10 | 2010-12-15 | 武汉武大卓越科技有限责任公司 | Road surface crack detection method based on dynamic programming |
CN103258206A (en) * | 2012-11-28 | 2013-08-21 | 河海大学常州校区 | Silicon solar cell surface defect detection and identification method |
CN104458755A (en) * | 2014-11-26 | 2015-03-25 | 吴晓军 | Multi-type material surface defect detection method based on machine vision |
CN105320962A (en) * | 2015-10-21 | 2016-02-10 | 东南大学 | Pavement damage type identification method based on classifier ensemble |
Non-Patent Citations (1)
Title |
---|
熊慧江: "基于机器视觉的表面缺陷检测方法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110619343A (en) * | 2018-06-20 | 2019-12-27 | 亚摩信息技术股份有限公司 | Automatic defect classification method based on machine learning |
CN110619343B (en) * | 2018-06-20 | 2023-06-06 | 亚摩信息技术(上海)有限公司 | Automatic defect classification method based on machine learning |
CN109187755A (en) * | 2018-09-12 | 2019-01-11 | 西安交通大学 | A kind of online supersonic damage-free detection method based on 3D printing |
CN109187755B (en) * | 2018-09-12 | 2020-05-22 | 西安交通大学 | Online ultrasonic nondestructive testing method based on 3D printing |
CN109829883A (en) * | 2018-12-19 | 2019-05-31 | 歌尔股份有限公司 | Product quality detection method and device |
CN109829883B (en) * | 2018-12-19 | 2020-11-17 | 歌尔光学科技有限公司 | Product quality detection method and device |
CN110221186A (en) * | 2019-06-13 | 2019-09-10 | 国网山东省电力公司青州市供电公司 | A kind of darkroom alternating-current voltage resistance test device and test method |
CN110399911A (en) * | 2019-07-12 | 2019-11-01 | 国网上海市电力公司 | A kind of transformer mechanical failure diagnostic method based on sound-filed simulation cloud atlas |
CN112184622A (en) * | 2020-08-31 | 2021-01-05 | 广东三和管桩股份有限公司 | Pipe pile appearance quality inspection method and system, computer equipment and storage medium |
CN115861987A (en) * | 2023-02-27 | 2023-03-28 | 江苏天南电力股份有限公司 | Intelligent electric power fitting defect identification method for on-line monitoring of power transmission line |
CN116990310A (en) * | 2023-09-27 | 2023-11-03 | 中国水利水电第九工程局有限公司 | Wall concrete crack monitoring and early warning system based on data analysis |
CN116990310B (en) * | 2023-09-27 | 2023-12-08 | 中国水利水电第九工程局有限公司 | Wall concrete crack monitoring and early warning system based on data analysis |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108154498A (en) | A kind of rift defect detecting system and its implementation | |
Wang et al. | Autonomous damage segmentation and measurement of glazed tiles in historic buildings via deep learning | |
CN104483330B (en) | The dynamic assessment method of concrete surface crack real-time monitoring system and cracking risk | |
CN108711148B (en) | Tire defect intelligent detection method based on deep learning | |
CN108921165B (en) | Water level identification method based on water gauge image | |
CN104964886A (en) | Welded member fatigue stress and strain real-time non-contact type monitoring method | |
CN109754368A (en) | A kind of crack joining method in bridge quality testing | |
CN104330023A (en) | Acquisition system and identification method of concrete surface initial crack information | |
CN114998852A (en) | Intelligent detection method for road pavement diseases based on deep learning | |
CN110334750A (en) | Iron tower of power transmission line bolt corrosion degree image classification recognition methods | |
CN111160270B (en) | Bridge monitoring method based on intelligent video recognition | |
Guldur et al. | Condition assessment of bridges using terrestrial laser scanners | |
CN115526515B (en) | Safety monitoring system of gate for water conservancy and hydropower | |
CN112613454A (en) | Electric power infrastructure construction site violation identification method and system | |
CN115993096A (en) | High-rise building deformation measuring method | |
CN112528979B (en) | Transformer substation inspection robot obstacle distinguishing method and system | |
CN115327065A (en) | Hydraulic and hydroelectric engineering sample detection method and system | |
CN202548896U (en) | System for measuring steel packaging parameters | |
CN116486289A (en) | Gas pipeline high-consequence area identification method driven by multi-source data and knowledge | |
Bruno et al. | Decay detection in historic buildings through image-based deep learning | |
CN115995056A (en) | Automatic bridge disease identification method based on deep learning | |
CN103063674B (en) | Detection method for copper grade of copper block, and detection system thereof | |
CN117371337B (en) | Water conservancy model construction method and system based on digital twin | |
CN116664846B (en) | Method and system for realizing 3D printing bridge deck construction quality monitoring based on semantic segmentation | |
CN112801975A (en) | Binocular vision-based track ballast inspection system and working method thereof |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180612 |
|
RJ01 | Rejection of invention patent application after publication |