CN110378433A - The classifying identification method of bridge cable surface defect based on PSO-SVM - Google Patents

The classifying identification method of bridge cable surface defect based on PSO-SVM Download PDF

Info

Publication number
CN110378433A
CN110378433A CN201910672440.4A CN201910672440A CN110378433A CN 110378433 A CN110378433 A CN 110378433A CN 201910672440 A CN201910672440 A CN 201910672440A CN 110378433 A CN110378433 A CN 110378433A
Authority
CN
China
Prior art keywords
defect
svm
pso
measured
drag
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
Application number
CN201910672440.4A
Other languages
Chinese (zh)
Inventor
李新科
高潮
郭永彩
邵延华
贺付亮
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Chongqing University
Original Assignee
Chongqing University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Chongqing University filed Critical Chongqing University
Priority to CN201910672440.4A priority Critical patent/CN110378433A/en
Publication of CN110378433A publication Critical patent/CN110378433A/en
Pending legal-status Critical Current

Links

Classifications

    • 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
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)

Abstract

The invention discloses the classifying identification methods of the bridge cable surface defect based on PSO-SVM, include the following steps: S1, obtain drag-line surface image information to be measured；S2, surface defects characteristic information to be measured is extracted from drag-line surface image information to be measured；S3, by surface defects characteristic information input PSO-SVM classifier to be measured, obtain the surface defect Classification and Identification information of drag-line to be measured.The peculiar advantage that the present invention is shown in solution small sample, non-linear, high dimensional pattern identify in view of supporting vector, SVM algorithm is applied to drag-line surface defects detection, also optimize SVM model parameter using particle swarm optimization algorithm, further improves Classification and Identification rate.

Description

The classifying identification method of bridge cable surface defect based on PSO-SVM
Technical field
The present invention relates to bridge machinery fields, and in particular to the classification of the bridge cable surface defect based on PSO-SVM is known Other method.
Background technique
With the high speed development of bridge transport development, the cable-stayed bridge and suspension bridge of large span and super-long span bridges are extensive Using.Drag-line is the main stressed member of this kind of bridge, and the reliability and durability of drag-line will be directly related to the safety of bridge And service life.Since the polyethylene (PE) or high density polyethylene (HDPE) (HDPE) protective layer of drag-line appearance are exposed to nature for a long time In environment and alternating load is born, easily generation corrosion failure, drag-line surface is caused longitudinal cracking, Transverse Cracks, surface occur The defects of corroding with scar pit hole hole.These defects can seriously affect the service performance of drag-line, need to detect it and identified. Mainly use artificial detection method both at home and abroad at present, but artificial process is time-consuming and laborious, it is inefficient, it is easy to produce safety accident.From Movement machine visible detection method has also been developed and deployed, and has critically important research significance and application prospect.
Surface defect classification is that link important in drag-line automatic Surface Defect Detection System can by the classification of defect Influence of all kinds of defects to drag-line service performance is analyzed, and improves the method for producing of drag-line surface protection layer material.It lacks Falling into most common method in assorting process has the methods of artificial neural network identification and support vector machines, but artificial neural network is calculated Method is the traditional statistics based on progressive theory, only has ideal application effect in the larger Shi Caineng of the quantity of learning sample. In limited sample, the good artificial neural network of training effect is possible to show very poor generalization ability.
Support vector machines (SVM) has stronger Generalization Ability when solving small sample decision problem, while can solve again Higher-dimension problem and local extremum problem in neural network algorithm, structure is also very simple, thus in practical engineering applications With stronger advantage, a kind of effective tool is provided for the practical application of Statistical Learning Theory.It is being solved in view of supporting vector Certainly small sample, the peculiar advantage shown in the identification of non-linear, high dimensional pattern, the invention proposes be based on PSO-SVM (population The support vector machines of algorithm optimization) bridge cable surface defect classifying identification method, by SVM algorithm be applied to drag-line surface Defects detection.In order to further increase Classification and Identification rate, using particle swarm optimization algorithm (PSO) Lai Youhua SVM model parameter.
Summary of the invention
In view of the above shortcomings of the prior art, the present invention provides points of the bridge cable surface defect based on PSO-SVM Class recognition methods will in view of the peculiar advantage that supporting vector is shown in solution small sample, non-linear, high dimensional pattern identify SVM algorithm is applied to drag-line surface defects detection, also optimizes SVM model parameter using particle swarm optimization algorithm, further mentions High Classification and Identification rate.
In order to solve the above-mentioned technical problem, present invention employs the following technical solutions:
The classifying identification method of bridge cable surface defect based on PSO-SVM, includes the following steps:
S1, drag-line surface image information to be measured is obtained；
S2, surface defects characteristic information to be measured is extracted from drag-line surface image information to be measured；
S3, by surface defects characteristic information input PSO-SVM classifier to be measured, obtain the surface defect classification of drag-line to be measured Identification information.
Preferably, surface defects characteristic information to be measured includes any one in shape feature, gray feature and textural characteristics Item is multinomial, and surface defect Classification and Identification information includes any one or more in hole, transversal crack, longitudinal crack and scar hole ?.
Preferably, shape feature includes the area of defect area, further includes the defect major diameter L of defect area1It is short with defect Diameter L2The ratio between Rb, the extracting method of shape feature includes:
Defect area is identified from drag-line surface image information to be measured using boundary scan；
Pixel number N in statistical shortcomings regionaTo obtain the area of defect area, G (x, y) indicates coordinate is the gray value of the pixel of (x, y), and A indicates defect area；
It obtains the maximum distance on defect area boundary between any two points and is denoted as defect major diameter L1, obtain defect major diameter L1 The shortest distance on defect area boundary is denoted as defect minor axis L in vertical direction2；
Based on Rb=L1/L2Obtain defect major diameter L1With defect minor axis L2The ratio between Rb。
Preferably, gray feature include based on borderline region scanning after defect area calculate average gray, variance and The extracting method of gradient, gray feature includes:
It is based onCalculate average grayIn formula, the single order histogram of imageL= 0,1 ..., L-1, N indicate total pixel, and n (l) indicates that gray level is the sum of all pixels of l, and L is tonal gradation；
It is based onCalculate variances sigma2；
It is based onCalculate gradient Ske.
Preferably, textural characteristics include based on borderline region scanning after defect area calculate angular second moment, contrast, Related, entropy and unfavourable balance are away from the extracting method of textural characteristics includes:
Based on formulaCalculate angular second moment ASM, in formula, p (i, j, δ, θ)={ (x, y) | f (x, y)=i, f (x+Dx, y+Dy)=j；X, y=0,1,2 ..., N-1 }, i and j are respectively the gray scale of any two pixel Value, (Dx, Dy) are the distance of described two pixels, and i, j=0,1,2 ..., L'-1, (x, y) is image coordinate, and L' is gray scale The number of grade, δ are the number of pixels of adjacent spaces, and θ indicates direction, and p (i, j, δ, θ) is that i and j two restore the probability occurred；
Based on formulaCalculate contrast C ON；
Based on formulaCalculate correlation COR, wherein u1、u2、s1And s2It is Calculating process parameter；
Based on formulaCalculate entropy ENT；
Based on formulaUnfavourable balance is calculated away from IDM.
Preferably, during training PSO-SVM classifier, the punishment system based on particle swarm algorithm Support Vector Machines Optimized Number c and kernel functional parameter g.
In conclusion the invention discloses the classifying identification methods of the bridge cable surface defect based on PSO-SVM, including Following steps: S1, drag-line surface image information to be measured is obtained；S2, surface to be measured is extracted from drag-line surface image information to be measured Defect characteristic information；S3, by surface defects characteristic information input PSO-SVM classifier to be measured, the surface for obtaining drag-line to be measured lacks Fall into Classification and Identification information.The spy that the present invention is shown in solution small sample, non-linear, high dimensional pattern identify in view of supporting vector It is advantageous, SVM algorithm is applied to drag-line surface defects detection, also optimizes SVM model parameter using particle swarm optimization algorithm, Further improve Classification and Identification rate.
Detailed description of the invention
In order to keep the purposes, technical schemes and advantages of invention clearer, the present invention is made into one below in conjunction with attached drawing The detailed description of step, in which:
Fig. 1 is the flow chart of the classifying identification method of the bridge cable surface defect disclosed by the invention based on PSO-SVM；
Fig. 2 is the classification recognition result schematic diagram of PSO-SVM.
Specific embodiment
The present invention is described in further detail with reference to the accompanying drawing.
As shown in Figure 1, the invention discloses the classifying identification method of the bridge cable surface defect based on PSO-SVM, packet Include following steps:
S1, drag-line surface image information to be measured is obtained；
S2, surface defects characteristic information to be measured is extracted from drag-line surface image information to be measured；
S3, by surface defects characteristic information input PSO-SVM classifier to be measured, obtain the surface defect classification of drag-line to be measured Identification information.
Compared with prior art, the present invention in view of supporting vector solve small sample, table in the identification of non-linear, high dimensional pattern SVM algorithm is applied to drag-line surface defects detection by the peculiar advantage revealed, can be in the case where sample is less to drag-line Surface defect is accurately classified, and the present invention uses particle swarm optimization algorithm also to optimize SVM model parameter, is further increased Classification and Identification rate.
When it is implemented, surface defects characteristic information to be measured includes appointing in shape feature, gray feature and textural characteristics It anticipates one or more, surface defect Classification and Identification information includes any one in hole, transversal crack, longitudinal crack and scar hole Or it is multinomial.
Drag-line surface image information to be measured is needed through image preprocessing, and is extracted its defect characteristic and could be passed through classifier Carry out effective Classification and Identification.Image is characterized in that the attribute that can be used as indicating in image, these characteristic parameters need not only to Difference between the 4 class sample defect such as longitudinal cracking, Transverse Cracks, surface erosion, scar pit hole hole is enough described, also reply surface lacks It falls into change caused by size and illumination of image etc. and keep relative stability.The 4 class sample defects proposed according to the present invention, can pass through Shape feature, gray feature and textural characteristics of surface defect etc. are extracted to be described.
When it is implemented, shape feature includes the area of defect area, it further include the defect major diameter L of defect area1With lack Fall into minor axis L2The ratio between Rb, the extracting method of shape feature includes:
Defect area is identified from drag-line surface image information to be measured using boundary scan；
Pixel number N in statistical shortcomings regionaTo obtain the area of defect area, G (x, y) indicates coordinate is the gray value of the pixel of (x, y), and A indicates defect area；
It obtains the maximum distance on defect area boundary between any two points and is denoted as defect major diameter L1, obtain defect major diameter L1 The shortest distance on defect area boundary is denoted as defect minor axis L in vertical direction2；
Based on Rb=L1/L2Obtain defect major diameter L1With defect minor axis L2The ratio between Rb。
Parameter RbThe shape feature for reflecting defect to a certain extent is a kind of easy measurement made to defect shape. Hole and the parameter R in scar holebUsually it is closer to 1, and the parameter R of transversal crack, longitudinal crackbIt is often much larger than or is much smaller than 1, Therefore parameter R is usedbThe classification of defect is carried out, easily and effectively, classification difficulty can be reduced.
When it is implemented, gray feature includes the average gray calculated based on the defect area after borderline region scanning, side Difference and gradient, the extracting method of gray feature include:
It is based onCalculate average grayIn formula, the single order histogram of imageL= 0,1 ..., L-1, N indicate total pixel, and n (l) indicates that gray level is the sum of all pixels of l, and L is tonal gradation；
It is based onCalculate variances sigma2；
It is based onCalculate gradient Ske.
Piece image f (x, y) regards a sample of a bivariate stochastic process as, can be retouched with joint probability distribution It states.By the range value of each pixel of the image measured, it is estimated that the probability distribution of image, to form the histogram of image Feature.Average gray, the main overall gray level for reflecting defect image is horizontal, and the overall gray level between different types of defect is horizontal Be it is discrepant, therefore, the classification and identification of different type defect can be carried out with average gray.The variance of defect image, Reflect the dispersion degree of defect intensity profile.The gradient of defect image, the gray scale that can reflect out between different defects are straight The asymmetric degree of square figure.Due to different types of defect, its intensity profile is different, it is possible to special to characterize different defects Sign.
When it is implemented, textural characteristics include the angular second moment, right calculated based on the defect area after borderline region scanning Than degree, correlation, entropy and unfavourable balance away from the extracting method of textural characteristics includes:
Based on formulaAngular second moment ASM is calculated, in formula, p (i, j, δ, θ)=(x, y) | F (x, y)=i, f (x+Dx, y+Dy)=j；X, y=0,1,2 ..., N-1 }, i and j are respectively the gray scale of any two pixel Value, (Dx, Dy) are the distance of described two pixels, and i, j=0,1,2 ..., L'-1, (x, y) is image coordinate, and L' is gray scale The number of grade, δ are the number of pixels of adjacent spaces, and θ indicates direction, and p (i, j, δ, θ) is that i and j two restore the probability occurred；
Based on formulaCalculate contrast C ON；
Based on formulaCalculate correlation COR, wherein u1、u2、s1And s2It is Calculating process parameter；
Based on formulaCalculate entropy ENT；
Based on formulaUnfavourable balance is calculated away from IDM.
Textural characteristics describe the surface nature of scenery corresponding to image or image-region, are a kind of global characteristics, are not Feature based on pixel, but a kind of feature based on region property.Gray level co-occurrence matrixes are extractable to adapt to human vision spy The textural characteristics of point, many scholars obtain extremely successful and wide as the characteristic quantity of Classification and Identification in image procossing General application.The space coordinate conversion of (x, y) can be " the gray scale of (i, j) by 2 simultaneous probability of pixel grayscale It is right " description, they formed matrix be known as gray level co-occurrence matrixes.
When it is implemented, during training PSO-SVM classifier, based on punishing for particle swarm algorithm Support Vector Machines Optimized Penalty factor c and kernel functional parameter g.
The present invention is based on algorithm of support vector machine to carry out Classification and Identification to drag-line surface defect.Know to improve svm classifier Not rate optimizes the penalty coefficient c and kernel functional parameter g of SVM model, i.e. PSO-SVM algorithm using particle swarm optimization algorithm.
Support vector machines is to solve two classification tool of one kind of Machine Learning Problems in data mining by optimal method, It is successfully applied in classification problem, more more excellent than conventional method on for small sample, non-linear and higher-dimension identification problem Gesture.The basic thought of SVM method for classifying modes is the optimal classification surface for finding two class samples, so that the class interval of two class samples It is maximum.SVM method be based on linear separability in the case of optimal classification surface and propose.If the general type of linear discriminant function Are as follows:
G (x)=wx+b；
Wherein w is adjustable weight vector, and b is the constant term of biasing.If g (x) is linear function, for the super of classification Plane equation is g (x)=0.To find optimal separating hyper plane, classification prediction is carried out to target, that is, it is (false to find optimal w and b Fixed its is w0And b0), then optimal Optimal Separating Hyperplane is w0X+b=0.This has also meant that the decision curved surface for classification.
Optimal classification surface is found, is equivalent to askMinimum value.For the sample of linearly inseparable, relaxation is introduced Variable ξiWith penalty coefficient c, the problem of optimization, can be indicated are as follows:
s.t.yi(w·xi+b)≥1-ξi, i=1 ..., N (15)
Quadratic programming problem is converted by the optimal separating hyper plane that above formula indicates, and is former problem with this problem, is solved Its dual problem:
For Nonlinear separability situation, need the straight line curve (curved surface) in lower dimensional space being mapped as in higher dimensional space Or plane.According to Mercer theorem, kernel function is introduced to it at this time:
K(xi,xj)=φ (xi)·φ(xj)
The effect of kernel function is, while x is mapped to high dimension linear space, also calculate two data in higher-dimension The inner product in space, makes calculation amount revert to xi·xjMagnitude.Kernel function include Polynomial kernel function, Radial basis kernel function and Sigmoid kernel function etc..Radial basis kernel function is selected under normal circumstances:
K(xi,xj)=exp (- γ | xi-xj|2)
Here γ is the parameter of Radial basis kernel function, is indicated with g, generallys use cross validation method in limited data Parameter optimization is carried out on collection.
Finally kernel function is updated in above formula, finds out optimal solution α accordingly*,b*, final discriminant function may be expressed as:
Particle swarm optimization algorithm main thought is from the research to birds group behavior.Since its process is simple, is easy to Realize, parameter is few, is not necessarily to the features such as gradient information, and good effect is shown in all kinds of complicated optimum problems, by It is widely used in data classification, pattern-recognition etc..
SVM model has highly important influence to the selection of penalty coefficient c and kernel functional parameter g to classification accuracy, It needs to optimize the two parameters.Rational choice is carried out to it based on PSO algorithm, the classification of SVM classifier can be improved Accuracy rate.PSO-SVM design of algorithm process is as follows:
(i) needs for combining SVM model, initialize population and setup parameter.Determine population scale and search space, this In only search for optimal penalty coefficient c and kernel functional parameter g, and in specified range limit population position and speed. Inertia weight parameter ω=1, Studying factors c are set herein1=c2=2.
(ii) fitness function is determined.Supporting vector is established using the corresponding penalty coefficient of particle individual and kernel functional parameter The learning model of machine, and bring training sample into, calculate the corresponding fitness value of particle individual.
(iii) extreme value updates and speed updates.By the fitness value of particle individual respectively with individual extreme value and global extremum It is compared.If the fitness value of individual is less than individual extreme value or global extremum, the fitness value of individual is replaced into individual Extreme value or global extremum, and newly-generated extreme value is carried out to the update of speed.
(iv) new extreme value and evolution are searched for.Each particle re-starts search according to new speed and position.It will search The fitness value arrived carries out the operation of step (iii).
(v) iteration is completed.If meeting termination condition (error is good enough or reaches maximum cycle) to exit.It will be optimal Parameter c, g are assigned to support vector machine classifier, are tested using test sample.
The accuracy rate of method disclosed by the invention is verified by following experiment:
A large amount of surface defect sample is arrived by the way that drag-line surface defect detection apparatus is available, during the experiment, from (hole is laterally split the defect of higher 4 quasi-representatives being affected with Surface Quality of the selection frequency of occurrences in these defect sample Line, longitudinal crack, scar hole) each 30 carry out identification experiment as research object.Sample defect size uniformly takes 256*256.
In Classification and Identification experimentation, we choose the surfaces such as longitudinal cracking, Transverse Cracks, surface erosion, scar pit hole hole Defect sample each 30, sample defect is 256*256 by pretreatment and unified size of choosing.Each sample standard deviation by defect face Product, Ratio of long radius to short radius, the average gray of defect image, variance, gradient, energy, contrast, correlation, entropy, inverse difference moment etc. totally 10 Eigenvalue cluster at.
The category label 1,2,3,4 of model output respectively represents longitudinal crack, transversal crack, surface erosion and scar pit hole Hole extracts 10 from every class sample and is used as training sample, and the part output of training sample is as shown in table 1.Remaining 20 works For test sample, they are respectively fed to PSO-SVM and carries out classification prediction.
The category label of 1 training sample of table
10 characteristic parameters are respectively fed to PSO-SVM and carry out classification prediction, (1- is vertical as shown in Figure 2 for the Visual Graph of identification To cracking, 2- Transverse Cracks, 3- surface erosion, 4- scar pit hole hole).Rightlabels indicates correct label, Predictlabels is prediction label.
Figure it is seen that the classification predicted value and target labels of transversal crack and longitudinal crack value are completely the same, and have Two surface erosion defects are accidentally divided into scar pit hole hole, while having a scar pit hole hole defect to be accidentally divided into surface erosion.Table 2 arranges PSO-SVM classifier is gone out to the classification results of 4 class defects.
Classification results of the table 2 based on PSO-SVM classifier
The present invention carries out parameter selection, the c=1 of selection, g=0.05 to SVM model by PSO algorithm.It achieves preferably Effect.PSO algorithm and artificial experience and genetic algorithm (GA) are compared the selection result of SVM model parameter by table 3, adopt With the Classification and Identification rate highest of PSO algorithm, there is better effect to the selection of parameter with genetic algorithm relative to artificial choose. Therefore the progress parameter selection of PSO algorithm is introduced to be necessary.
The different optimizing parametric techniques of table 3 and Classification and Identification rate
Longitudinal cracking, Transverse Cracks, surface erosion, scar pit hole hole that the present invention is mainly primarily present drag-line surface etc. 4 Class defect carries out Study on Classification and Recognition.After image preprocessing, defect characteristic extraction is carried out to defect image.The present invention mentions 10 characteristic parameters in total such as shape feature, gray feature and textural characteristics of defect are taken.Then pass through support vector machines pair Defect image is classified, and in order to improve Classification and Identification rate, SVM model parameter is chosen using particle swarm optimization algorithm.Pass through Classification and Identification experiment to drag-line surface defect image, the Classification and Identification accuracy rate of the method for the present invention have reached 96.25%, as a result Show PSO-SVM discrimination with higher
Finally, it is stated that the above examples are only used to illustrate the technical scheme of the present invention and are not limiting, although passing through ginseng According to the preferred embodiment of the present invention, invention has been described, it should be appreciated by those of ordinary skill in the art that can To make various changes to it in the form and details, without departing from the present invention defined by the appended claims Spirit and scope.

Claims (6)

1. the classifying identification method of the bridge cable surface defect based on PSO-SVM, which comprises the steps of:
S1, drag-line surface image information to be measured is obtained；
S2, surface defects characteristic information to be measured is extracted from drag-line surface image information to be measured；
S3, by surface defects characteristic information input PSO-SVM classifier to be measured, obtain the surface defect Classification and Identification of drag-line to be measured Information.
2. the classifying identification method of the bridge cable surface defect based on PSO-SVM, feature exist as described in claim 1 In surface defects characteristic information to be measured includes any one or more in shape feature, gray feature and textural characteristics, surface Defect Classification and Identification information includes any one or more in hole, transversal crack, longitudinal crack and scar hole.
3. the classifying identification method of the bridge cable surface defect based on PSO-SVM, feature exist as claimed in claim 2 In shape feature includes the area of defect area, further includes the defect major diameter L of defect area1With defect minor axis L2The ratio between Rb, shape The extracting method of shape feature includes:
Defect area is identified from drag-line surface image information to be measured using boundary scan；
Pixel number N in statistical shortcomings regionaTo obtain the area of defect area, G (x, y) indicates coordinate is the gray value of the pixel of (x, y), and A indicates defect area；
It obtains the maximum distance on defect area boundary between any two points and is denoted as defect major diameter L1, obtain defect major diameter L1Vertically The shortest distance on defect area boundary is denoted as defect minor axis L on direction2；
Based on Rb=L1/L2Obtain defect major diameter L1With defect minor axis L2The ratio between Rb。
4. the classifying identification method of the bridge cable surface defect based on PSO-SVM, feature exist as claimed in claim 2 In gray feature includes average gray, variance and the gradient calculated based on the defect area after borderline region scanning, and gray scale is special The extracting method of sign includes:
It is based onCalculate average grayIn formula, the single order histogram of imageN indicates total pixel, and n (l) indicates that gray level is the sum of all pixels of l, and L is tonal gradation；
It is based onCalculate variances sigma2；
It is based onCalculate gradient Ske.
5. the classifying identification method of the bridge cable surface defect based on PSO-SVM, feature exist as claimed in claim 2 In, textural characteristics include the angular second moment calculated based on the defect area after borderline region scanning, contrast, correlation, entropy and inverse The extracting method of gap, textural characteristics includes:
Based on formulaAngular second moment ASM is calculated, in formula, p (i, j, δ, θ)=(x, y) | f (x, Y)=i, f (x+Dx, y+Dy)=j；X, y=0,1,2 ..., N-1 }, i and j are respectively the gray value of any two pixel, (Dx, Dy) is the distance of described two pixels, and i, j=0,1,2 ..., L'-1, (x, y) is image coordinate, and L' is gray level Number, δ are the number of pixels of adjacent spaces, and θ indicates direction, and p (i, j, δ, θ) is that i and j two restore the probability occurred；
Based on formulaCalculate contrast C ON；
Based on formulaCalculate correlation COR, wherein u1、u2、s1And s2It is Calculating process parameter；
Based on formulaCalculate entropy ENT；
Based on formulaUnfavourable balance is calculated away from IDM.
6. such as the classifying identification method of the bridge cable surface defect described in any one of claim 1 to 5 based on PSO-SVM, It is characterized in that, during training PSO-SVM classifier, the penalty coefficient c based on particle swarm algorithm Support Vector Machines Optimized With kernel functional parameter g.
CN201910672440.4A 2019-07-24 2019-07-24 The classifying identification method of bridge cable surface defect based on PSO-SVM Pending CN110378433A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910672440.4A CN110378433A (en) 2019-07-24 2019-07-24 The classifying identification method of bridge cable surface defect based on PSO-SVM

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910672440.4A CN110378433A (en) 2019-07-24 2019-07-24 The classifying identification method of bridge cable surface defect based on PSO-SVM

Publications (1)

Publication Number Publication Date
CN110378433A true CN110378433A (en) 2019-10-25

Family

ID=68255660

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910672440.4A Pending CN110378433A (en) 2019-07-24 2019-07-24 The classifying identification method of bridge cable surface defect based on PSO-SVM

Country Status (1)

Country Link
CN (1) CN110378433A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111487250A (en) * 2020-04-02 2020-08-04 苏州奥创智能科技有限公司 Intelligent visual detection method and system applied to injection molding defective product detection
CN111582510A (en) * 2020-05-13 2020-08-25 中国民用航空飞行学院 Intelligent identification method and system based on support vector machine and civil aircraft engine
CN112884748A (en) * 2021-03-02 2021-06-01 江苏海洋大学 Non-woven fabric surface small defect detection method based on multi-core support vector machine

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108961233A (en) * 2018-06-28 2018-12-07 华侨大学 A kind of composite polycrystal-diamond surface defect classifying identification method
CN109308697A (en) * 2018-09-18 2019-02-05 安徽工业大学 A kind of leaf disease recognition method based on machine learning algorithm

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108961233A (en) * 2018-06-28 2018-12-07 华侨大学 A kind of composite polycrystal-diamond surface defect classifying identification method
CN109308697A (en) * 2018-09-18 2019-02-05 安徽工业大学 A kind of leaf disease recognition method based on machine learning algorithm

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
李新科: "桥梁拉索表面缺陷图像检测关键技术的研究", 《中国博士学位论文全文数据库 信息科技辑》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111487250A (en) * 2020-04-02 2020-08-04 苏州奥创智能科技有限公司 Intelligent visual detection method and system applied to injection molding defective product detection
CN111582510A (en) * 2020-05-13 2020-08-25 中国民用航空飞行学院 Intelligent identification method and system based on support vector machine and civil aircraft engine
CN111582510B (en) * 2020-05-13 2021-08-31 中国民用航空飞行学院 Intelligent identification method and system based on support vector machine and civil aircraft engine
CN112884748A (en) * 2021-03-02 2021-06-01 江苏海洋大学 Non-woven fabric surface small defect detection method based on multi-core support vector machine

Similar Documents

Publication Publication Date Title
Hüthwohl et al. Multi-classifier for reinforced concrete bridge defects
Kou et al. Development of a YOLO-V3-based model for detecting defects on steel strip surface
CN113409314B (en) Unmanned aerial vehicle visual detection and evaluation method and system for corrosion of high-altitude steel structure
CN108765412B (en) Strip steel surface defect classification method
CN109784203B (en) Method for inspecting contraband in weak supervision X-ray image based on layered propagation and activation
Li et al. Sewer pipe defect detection via deep learning with local and global feature fusion
CN110555467B (en) Industrial data classification method based on model migration
CN110378433A (en) The classifying identification method of bridge cable surface defect based on PSO-SVM
CN108956614A (en) A kind of pit rope dynamic method for detection fault detection and device based on machine vision
CN113920107A (en) Insulator damage detection method based on improved yolov5 algorithm
Li et al. Particle swarm optimization-based SVM for classification of cable surface defects of the cable-stayed bridges
CN110147760A (en) A kind of efficient electrical energy power quality disturbance image characteristics extraction and identification new method
CN110176143A (en) A kind of highway traffic congestion detection method based on deep learning algorithm
TW202127371A (en) Image-based defect detection method and computer readable medium thereof
Wang et al. Knowledge transfer for structural damage detection through re-weighted adversarial domain adaptation
CN114565314B (en) Digital twinning-based hot rolled steel coil end face quality control system and method
Sundarrajan et al. Explainable efficient and optimized feature fusion network for surface defect detection
Alfaz et al. Bridge crack detection using dense convolutional network (densenet)
CN115830403B (en) Automatic defect classification system and method based on deep learning
CN110991485B (en) Performance evaluation method and system of target detection algorithm
Minhas et al. Defect Detection using Deep Learning from Minimal Annotations.
CN109034172B (en) Product appearance defect detection method based on fuzzy relaxation constraint multi-core learning
Platzer et al. On the suitability of different features for anomaly detection in wire ropes
CN110390355A (en) The new defect identification method of pipeline based on evolution maximum fuzzy minimum neural network
Li et al. Research on surface defect detection in submersible pump impellers based on the novel MXNet deep learning model

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20191025