CN107749058B - Machine vision detection method and system for boiler pipeline surface defects - Google Patents

Machine vision detection method and system for boiler pipeline surface defects Download PDF

Info

Publication number
CN107749058B
CN107749058B CN201710994305.2A CN201710994305A CN107749058B CN 107749058 B CN107749058 B CN 107749058B CN 201710994305 A CN201710994305 A CN 201710994305A CN 107749058 B CN107749058 B CN 107749058B
Authority
CN
China
Prior art keywords
image
boiler
vector
pipeline
lbp
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.)
Active
Application number
CN201710994305.2A
Other languages
Chinese (zh)
Other versions
CN107749058A (en
Inventor
谭建平
李臻
方宇
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.)
Central South University
Original Assignee
Central South 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 Central South University filed Critical Central South University
Priority to CN201710994305.2A priority Critical patent/CN107749058B/en
Publication of CN107749058A publication Critical patent/CN107749058A/en
Application granted granted Critical
Publication of CN107749058B publication Critical patent/CN107749058B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—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
    • G06T5/00—Image enhancement or restoration
    • G06T5/40—Image enhancement or restoration using histogram techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20024—Filtering details
    • G06T2207/20032—Median filtering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30168—Image quality inspection

Landscapes

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

Abstract

The invention discloses a machine vision detection method and a system for boiler pipeline surface defects. And then, carrying out real-time detection on the surface image of the boiler pipeline to be detected, which is acquired by the industrial camera, through a decision function. The classification model is simple and reliable, the defect identification accuracy is high, and compared with the manual detection of the boiler surface defects, the detection efficiency is greatly improved.

Description

Machine vision detection method and system for boiler pipeline surface defects
Technical Field
The invention relates to the technical field of machine vision, in particular to a machine vision detection method and system for boiler pipeline surface defects.
Background
Because the energy in China is mainly fossil fuels such as coal, the fire power generation still occupies the leading position of the electricity supply in China in a period of time in the future. According to statistics, by 2 months in 2015, 1241 households are in thermal power plants operated in China at present, and the total power generation amount is 9.16 hundred million kilowatts and is about 67% of the total power generation amount. Thermal power generation is still the main power generation mode of the power energy industry in China at present, boilers occupy an important position in various equipment of thermal power generation, are called three main machines of a power plant together with a steam turbine and a generator, are energy sources of the whole coal-fired power plant, and the running stability of the three main machines directly influences the safety of the whole coal-fired power plant system.
The boiler is called as a steam generator, is an energy conversion device, has an extremely wide application range, needs to be regularly detected and maintained due to the fact that the operating condition of the boiler is very poor and the boiler is subjected to the scouring and corrosion of steam, furnace water and the like for a long time, so that the defects of scale accumulation, falling of oxide skin on the surface of a pipeline, uneven internal and external deformation and the like often occur, otherwise the normal operation of the device is affected once a safety accident occurs, and in the existing boiler faults, the four-tube (a superheater tube, a reheater tube, a water wall tube and a coal economizer tube) explosion and leakage are the most common and multiple faults, and the peeling of the pipeline is an important reason for the four-tube leakage. At the initial stage of peeling off of the oxide skin of the pipeline, the leakage amount of the pipeline is not large, the fault part is not easy to determine and judge, the leakage degree is generally increased after a few days or longer, destructive leakage or pipe explosion is developed, and the safe and stable operation of an ignition power plant is seriously threatened, so that the detection of four pipes of a boiler in the processes of production, installation and operation is strengthened, and the safe operation of the coal-fired power plant is vital.
At present, thermal power plants all over the country mainly depend on the most original method of manual regular detection, which is also the most widely applied method at present. A worker wears the mask, wears the working clothes to enter the interior of the boiler, and determines whether the boiler pipeline has defects or not and maintains the boiler pipeline by the aid of the flashlight and the naked eyes. Although simple and easy, the method has great limitation in practical use. First, the working environment inside the boiler is extremely harsh, and workers working in such an environment for a long time have a high possibility of occupational diseases, resulting in an increasing number of young people reluctant to do the work. Secondly, the inspection of the boiler pipeline requires extremely rich working experience, and workers who first contact the boiler pipeline cannot well distinguish defects under the irradiation of a flashlight, but the defects often have serious consequences and directly cause shutdown. Therefore, finding a new surface defect detection method to replace the traditional manual detection is one of the problems that many enterprises need to solve urgently.
The machine vision system is a scientific technology for simulating biological vision by researching a computer, the primary objective of the machine vision system is to create or restore a real world model by using an image, then the real world is known, the machine vision is a quite new and rapidly developed research field and becomes one of important research fields of computer science, and in decades, with the improvement of the performance of hardware such as an industrial camera, an image acquisition card, lighting equipment and the like and the continuous improvement of an image processing algorithm, the detection technology based on the machine vision has higher precision, stronger anti-interference capability and better reliability, so that the machine vision detection method based on the image can provide a good alternative scheme and solution for the defect identification of the pipeline surface.
Disclosure of Invention
The technical problems to be solved by the invention are as follows: aiming at the defects of the existing pipeline surface defect detection which is mainly caused by the observation of workers by naked eyes, the invention provides a machine vision detection method and a system for the boiler pipeline surface defect, which have the advantages of high detection speed and high detection precision.
In order to solve the technical problems, the invention adopts the following technical scheme:
a machine vision detection method for boiler pipeline surface defects comprises the following steps:
step 1, acquiring z boiler pipeline surface images as sample images by an industrial camera with the aid of an illumination system, wherein the sample images comprise normal pipeline surface images and pipeline surface defect images;
step 2, respectively preprocessing each sample image and extracting a feature vector of the image;
step 3, respectively taking the feature vectors of z sample images as input vectors of a support vector machine, then establishing an optimal classification hyperplane according to a design criterion of maximum classification interval, and determining a decision function by the optimal classification hyperplane;
step 4, acquiring a surface image of the boiler pipeline to be detected by an industrial camera with the aid of an illumination system;
step 5, preprocessing the surface image of the boiler pipeline to be detected, and extracting a feature vector of the image;
and 6, inputting the characteristic vector of the surface image of the boiler pipe to be detected into a decision function, and judging whether the surface of the boiler pipe to be detected has defects or not.
The method for preprocessing the image in the steps 2 and 5 comprises the following steps: firstly, increasing the contrast of an image by adopting a histogram equalization method; and then, removing the noise interference in the image acquisition process by adopting a self-adaptive median filtering method.
The step 2 and the step 5 for extracting the feature vector of the image comprise the following steps:
s1, extracting a binary vector of the image: segmenting the preprocessed image by adopting a self-adaptive threshold segmentation processing method to generate a binary image; converting the binary image into a vector form, namely generating a corresponding binary vector by the gray value of each pixel point on the binary image;
s2, extracting LBP vector of the image: firstly, processing a preprocessed image by adopting a local binarization method: the method comprises the steps of determining an LBP value of each pixel in an image through a circular LBP operator with the radius of 2 pixels, establishing an LBP statistical histogram by taking the LBP value as a horizontal coordinate and the occurrence frequency of each value as a vertical coordinate, and setting the size of the LBP statistical histogram to be consistent with that of a binary image, even if the number of pixels contained in the horizontal length and the longitudinal length of the LBP statistical histogram is kept to be consistent with that of the binary image; then, the LBP statistical histogram is respectively converted into a vector form, namely, a corresponding LBP vector is generated according to the gray value of each pixel point on the LBP statistical histogram;
s3, feature dimension reduction: and respectively carrying out dimensionality reduction on the binary vector and the LBP vector of the image, and merging the two vectors obtained after dimensionality reduction to serve as the feature vector of the image.
The above steps are specifically described below:
(1) histogram equalization: because there is not any lighting device in the boiler, so the industrial camera must collect the high-quality, high-definition picture under the assistance of the lighting system, and the industrial camera is influenced by the sensitivity of the lighting, photosensitive device in the course of collecting the picture, the degradation of the picture may appear, the invention is through the equalization of the histogram, use the cumulative function to adjust the gray value, and then increase the contrast of the surface picture of the pipeline, after the surface picture of the pipeline is carried on the equalization of the histogram, its effect is as shown in fig. 6;
(2) adaptive median filtering: because the noise in the pipeline surface image is most salt and pepper noise, and the gray value of the noise point is most larger than that of the field pixel, the invention uses a window scanning image with the size of 5 multiplied by 5, and takes the median of the neighborhood as the gray value of the pixel of the center point of the window, and the method not only eliminates the salt and pepper noise to a certain extent, but also protects the edge characteristic of the image;
(3) and (3) adaptive threshold processing: the invention adopts self-adaptive threshold processing as an image segmentation method, uses a window with the size of 3 multiplied by 3 to scan the filtered image, uses the average value of 9 gray values in the window as the threshold value of the region, further completes the threshold processing of the window, and finally realizes the segmentation of the whole image. The binary image after the adaptive thresholding is shown in fig. 7 and 8;
(4) local binarization processing: the sizes and the shapes of different boiler pipelines are almost the same, and the color of the defect part of the pipeline is similar to the color of the normal surface of the pipeline, so that the extraction of the shape characteristic and the color characteristic of the image is not substantially helpful for identifying whether the surface of the pipeline has defects or not. The pipeline surface defect has the characteristics of large defect range, obvious defect boundary and the like, so that the texture features of the pipeline surface can be extracted, the extracted texture features are used as a classification basis, the calculation is simple when the texture features are extracted, and the extracted features have rotation invariance and gray scale invariance, so that the local binarization processing is adopted to extract the texture of the image to obtain the corresponding LBP statistical histogram. The normal LBP statistical histogram of the surface of the pipeline and the LBP statistical histogram of the surface of the pipeline with defects are respectively shown in FIG. 9 and FIG. 10;
(5) and (3) feature dimension reduction treatment: when the binary image and the LBP statistical histogram are converted into a vector form, for example, the sizes of the binary image and the LBP statistical histogram are both 400 × 600, when the binary image and the LBP statistical histogram are converted into a vector, the dimensions of the vectors corresponding to the two types of images are 240000 dimensions, and it can be found that the dimensions of the vectors are very high, if dimension reduction is not performed and direct substitution calculation is performed, the complexity of calculation is increased, and burden is brought to subsequent classification problems, therefore, the invention establishes a new feature subset by performing linear combination on the original feature vectors, and further realizes dimension reduction of the feature vectors, and the specific steps are as follows:
1) firstly, respectively converting binary images corresponding to z sample images and an LBP statistical histogram into a vector form, setting the dimension of the vector as n, and then sequentially arranging the vectors obtained by converting the binary images corresponding to the z sample images from top to bottom to form a matrix with z rows and n columns, and recording the matrix as A[z×n]Sequentially arranging vectors obtained by converting LBP statistical histograms corresponding to z sample images from top to bottom to form a matrix with z rows and n columns, which is marked as B[z×n]；
2) Will matrix A[z×n]And B[z×n]Subtracting the mean of each column of (a) to obtain two new matrices, denoted as a'[z×n]And B'[z×n]；
3) Let matrix RA＝(A′TA′)n×nThe matrix RB＝(B′TB′)n×nSeparately solving the two matrices RAAnd RBCharacteristic value λ ofAiAnd λBiAnd corresponding feature vectors
Figure GDA0002937681070000041
And
Figure GDA0002937681070000042
wherein i is 1,2,3, … n;
4) the characteristic value lambda is measuredAiAnd λBiSequentially arranging the characteristic vectors from large to small, combining the characteristic vectors corresponding to the first k characteristic values into a matrix with the row number n and the column number k, and recording the matrix as UA[n×k]And UB[n×k]Wherein k is<<n；
5) Will matrix A'[z×n]And B'[z×n]Are respectively provided withAnd matrix UA[n×k]And UB[n×k]Multiplying to obtain two matrixes with z rows and k columns, and recording the matrixes as A ″[z×k]And B ″)[z×k]；
6) The matrix A' after dimension reduction[z×k]And B ″)[z×k]Respectively converted into z column vectors, which are respectively marked as
Figure GDA0002937681070000043
And
Figure GDA0002937681070000044
wherein each column vector contains k elements, i ═ 1,2, … z; therefore, the dimension of the vector corresponding to each image is reduced to k dimension.
Step 3 is described in detail below:
the shapes of the defects on the surface of the pipeline are various, no rule can be followed, the defect distribution is scattered, the optimal hyperplane is established only by means of the LBP statistical histogram, and the classification accuracy is low. The method comprises the following specific steps:
3.1, two-valued vector of the same image
Figure GDA0002937681070000051
And LBP vector
Figure GDA0002937681070000052
Merge, record as
Figure GDA0002937681070000053
Wherein the vector
Figure GDA0002937681070000054
Contains 2k elements, i ═ 1,2, …, z;
3.2 at each feature vector
Figure GDA0002937681070000055
To all aboveMarking a class label yiBecause the shot image is only divided into normal image and defect image, the class label corresponding to the feature vector of the normal image on the surface of the pipeline is 1, and the class label corresponding to the feature vector of the image with defect on the surface of the pipeline is-1;
3.3, because the distribution of the defects on the surface of the pipeline has stronger randomness and the sizes and the shapes of the defects are different, in order to ensure that the classification hyperplane can accurately classify the images on the surface of the pipeline, the functional relation can take the following form:
Figure GDA0002937681070000056
the corresponding decision function is
Figure GDA0002937681070000057
Figure GDA0002937681070000058
Is the feature vector of the ith sample image, yiA class label corresponding to the feature vector of the ith sample image; c. CiFor lagrange multipliers, each eigenvector
Figure GDA0002937681070000059
Corresponds to one ci(ii) a b is a deviation term, H is a penalty parameter, and gamma is a kernel function parameter. The classification hyperplane should have maximum classification interval when classifying normal images and defect images, so that the problem of solving the function relation can be converted into the objective function
Figure GDA00029376810700000510
In that
Figure GDA00029376810700000511
The problem of the maximum value under the conditions;
3.4, it can be known from the functional relation of the objective function L (c), taking different penalty parameters H and kernel parameters gamma will change the maximum value of the objective function L (c), and further affect the accuracy of the classification plane to the defect identification, but it still has the following advantagesIn most cases, the values of the parameters H and gamma are determined only by experience, while the invention determines the values of the punishment parameter H and the nuclear parameter gamma by using a genetic algorithm, in the calculation, the defect identification accuracy rate is used as the iteration until the preset iteration times are finished, and then the highest classification accuracy rate in the iteration process and the corresponding punishment parameter H are output*With the kernel function parameter gamma*. The method comprises the following specific steps:
1) determining iteration parameters including the population number N, the iteration times T, the cross probability p, the variation probability q, the variation range of a penalty parameter H and the variation range of a kernel function parameter gamma;
2) randomly selecting T groups of punishment parameters H and kernel function parameters gamma as N groups of original data, then respectively calculating the defect identification accuracy rate corresponding to each group of data as the fitness value of the defect identification accuracy rate, and recording the maximum fitness value and a group of data corresponding to the fitness value; the defect identification accuracy corresponding to each group of data can be calculated through the svmtrain function in the libsvm toolbox, and the method comprises the following steps: firstly, setting parameters in an svmtrain function, including an svm type, a kernel function type and a cross validation number, substituting the N groups of original data pairs into the svmtrain function, and inputting training data and a class label corresponding to the training data to obtain the defect identification accuracy;
3) selecting iterative data from the current N groups of data by using a roulette algorithm, converting the iterative data into a binary form, and then performing intersection and variation to obtain N groups of new data;
4) calculating the defect identification accuracy corresponding to the N groups of new data to be used as the fitness values of the N groups of new data, and recording the maximum fitness value and a group of data corresponding to the fitness value;
5) returning to the step 3) to obtain data of the next iteration by using the roulette algorithm again, and repeating the steps until 180 iterations are completed;
6) outputting the maximum fitness value recorded in the T iterative processes and a group of data corresponding to the maximum fitness value, and recording as a penalty parameter H*With the kernel function parameter gamma*。
7) Will find outPenalty parameter H of*With the kernel function parameter gamma*Substituting the H and the gamma into an objective function L (c), and solving the maximum value of the objective function L (c) and the c corresponding to the maximum value under the constraint condition by using an SMO algorithm (the idea of coordinate rising)iAnd the value of b, denoted as ci *And b*；
Step 7) comprises the following specific steps:
(1) in that
Figure GDA0002937681070000061
Under the condition, all Lagrange multipliers ciGiving an initial value and recording as ci oldThen, any lagrange multiplier c in the range of (0, H) is selectedh oldCorrespond it to
Figure GDA0002937681070000062
As
Figure GDA0002937681070000063
Substitution into
Figure GDA0002937681070000064
In turn, solve for boldBecause according to the KKT condition, when 0<ci old<When the hydrogen content is H, the reaction is carried out,
Figure GDA0002937681070000065
so that there are
Figure GDA0002937681070000066
And obtaining the initial classification hyperplane expression which is recorded as
Figure GDA0002937681070000067
(2) Determining a first variable; traverse the entire lagrange multiplier sample set (i.e., c)i oldI-1, 2, …, z), the lagrange multiplier that violates the KKT condition is selected as the first variable, denoted as cuU e {1,2, …, z } where the KKT condition is:
Figure GDA0002937681070000071
(3) determining a second variable; is chosen such that | Eu-EvThe maximum lagrangian multiplier is used as the second variable, v is formed by {1,2, …, z }, and is marked as cv；
Figure GDA0002937681070000073
(4) Considering the remaining Z-2 Lagrangian multipliers as fixed values, i.e. taking the initial values in step (1), L (c) is a bivariate linear equation, where c isu、cvAs independent variables, simultaneous equations
Figure GDA0002937681070000074
The maximum value of L (c) and the corresponding cu newAnd cv new
(5) According to cu newAnd cv newB can be determinednewValue of (a), (b)newThe value ranges are as follows:
Figure GDA0002937681070000072
(6) c to be obtainedu new、cv newAnd bnewRespectively as cu old、cv oldAnd boldIs substituted into
Figure GDA0002937681070000075
Obtaining a new classification hyperplane function h (x);
(7) and (4) repeating the steps (2) to (6) until all the Lagrangian multipliers meet the KKT condition and are marked as ci *The corresponding offset term is denoted as b*(i.e., b obtained by repeating the step (5) for the last timenew)；
C obtained in step (7)i *And b*Respectively substitute into the functional relational expressions
Figure GDA0002937681070000077
And obtaining a classification hyperplane with the highest classification accuracy under the z images, wherein the relation of the classification hyperplane is as follows:
Figure GDA0002937681070000076
further, in the step 4, the industrial camera is fixed on the lifting platform through the two-degree-of-freedom cradle head, the lifting platform and the two-degree-of-freedom cradle head drive the industrial camera to move up and down, left and right, under the assistance of the illumination system, the industrial camera continuously collects the surface images of the boiler pipeline to be detected according to a set time interval, outputs the images to the upper computer for real-time detection, and triggers an alarm signal once the defects exist on the surface of the boiler pipeline.
The invention also provides a machine vision detection system for the surface defects of the boiler pipeline, which comprises an industrial camera, a two-degree-of-freedom cradle head, a lifting platform, an illumination system and an upper computer; the industrial camera and the illumination system are fixed on the lifting platform through the two-degree-of-freedom holder, and the lifting platform and the two-degree-of-freedom holder drive the industrial camera to move up and down, left and right for realizing continuous acquisition of the surface images of the boiler pipeline; the industrial camera collects the surface image of the boiler pipeline under the assistance of the illumination system and outputs the image to the upper computer for detection; the method of the system realizes the surface defect detection of the boiler pipeline.
And the upper computer is provided with an alarm module, and once the defects on the surface of the boiler pipeline are detected, an alarm signal is triggered.
Has the advantages that:
the invention provides a machine vision detection method and a system for boiler pipeline surface defects, which have the following advantages:
1) the judgment model of the vision system is simple and reliable, the accuracy of defect identification can reach 97%, and the defect that a worker who is in contact with the work for the first time cannot well distinguish the defects under the irradiation of a flashlight is thoroughly overcome;
2) the visual system has strong adaptability to the internal environment of the boiler, realizes full automation of defect identification, greatly improves the detection efficiency compared with the manual detection of the surface defects of the boiler, greatly shortens the shutdown detection time of the boiler, and further reduces the huge loss caused by shutdown;
3) the visual system is small in size, the industrial camera can reach any region inside the boiler by depending on the lifting platform and the two-degree-of-freedom holder, particularly the region with a large danger coefficient, and the region which cannot be obtained by workers due to the problems of internal construction and the like of the boiler, so that the all-round dynamic detection of the whole boiler pipeline is realized.
4) The industrial camera adopted by the vision system has corresponding finished products on the market, and can be purchased without customization. The two-degree-of-freedom holder for building the camera has the advantages of simple structure, easiness in manufacturing and processing, low production cost and the like.
Drawings
FIG. 1 is a schematic view of a machine vision system for detecting surface defects of boiler tubes
FIG. 2 is a diagram of a machine vision system for detecting surface defects in a boiler tube
FIG. 3 arrangement diagram of the illumination system
FIG. 4 flow chart of pipeline surface defect identification algorithm
FIG. 5 Gray scale map of a defective image
FIG. 6 Gray level map of defect image after histogram equalization
FIG. 7 Normal image after adaptive threshold segmentation processing
FIG. 8 Defect image after adaptive threshold segmentation processing
FIG. 9 LBP histogram of Normal image
FIG. 10 LBP histogram of a defective image
FIG. 11 is a graph showing test sample results
Detailed Description
The invention will be further explained with reference to the drawings and examples.
As shown in FIG. 1, the machine vision inspection system for the surface defects of the boiler pipeline disclosed by the invention comprises an image acquisition unit, an image transmission unit, an image processing unit, a fault judgment unit and a fault alarm unit. The specific structure of the device is shown in fig. 2 and 3, and the device comprises an industrial camera 3, a two-degree-of-freedom holder 2, a lifting platform 1, an illumination system and an upper computer; the industrial camera 3 and the illumination system are fixed on the lifting platform 1 through the two-degree-of-freedom holder 2, and the lifting platform 1 and the two-degree-of-freedom holder 2 drive the industrial camera 3 to move up and down, left and right for realizing continuous acquisition of surface images of the boiler pipeline; the industrial camera 3 collects the surface image of the boiler pipeline with the assistance of the illumination system and outputs the image to the upper computer for detection; the system firstly collects a certain number of boiler pipeline surface images, preprocesses, reduces dimensions and combines features on the images, and then solves a classification hyperplane with the highest accuracy and a corresponding decision function by utilizing an iterative algorithm and a coordinate ascending idea;
then, real-time detection of the surface defects of the boiler pipes to be detected in the same size is carried out, and the process is shown in fig. 4 and comprises the following steps:
1. the industrial camera is fixed on the lifting platform through the two-degree-of-freedom cradle head, the lifting platform drives the industrial camera to move up and down, left and right, under the assistance of the illumination system, the industrial camera continuously collects the surface images of the boiler pipeline to be detected according to a set time interval, and outputs the images to the upper computer in a wired transmission mode;
2. the upper computer carries out the steps of preprocessing, dimensionality reduction, feature combination and the like on the collected image to obtain the feature vector of the surface image of the boiler pipeline to be detected
Figure GDA0002937681070000092
3. Will vector
Figure GDA0002937681070000091
Substituting the relational expression
Figure GDA0002937681070000093
In the formula of function
Figure GDA0002937681070000094
Judging whether the surface of the boiler pipeline to be detected has defects or not, if so, judging whether the surface of the boiler pipeline to be detected has defects or not
Figure GDA0002937681070000095
If the value is 1, the surface of the boiler pipeline to be detected is normal, and if the value is not 1, the surface of the boiler pipeline to be detected is normal
Figure GDA0002937681070000096
If the value is-1, detecting that the surface of the boiler pipeline has defects;
4. once the surface of the boiler pipeline is detected to have defects, an alarm signal is triggered, so that workers can timely and accurately repair the surface of the boiler pipeline in the later period.
In order to verify the detection effect of the invention, 38 test images are taken as a test set, and the steps of image preprocessing, dimension reduction, feature merging and the like are repeated to obtain the feature vectors of the test images
Figure GDA0002937681070000101
Then the feature vector is processed
Figure GDA0002937681070000102
Importing into a function relation of the classification hyperplane, and
Figure GDA0002937681070000103
namely, it is
Figure GDA0002937681070000104
Greater than 0, y' is equal to 1,
Figure GDA0002937681070000105
if the value is less than 0, y ' is equal to-1, and the test value y ' corresponding to each test image is obtained 'iAs shown in fig. 11. Test value y'iWith the true value y of the imageiCompared with the prior art, the classification accuracy of the classification hyperplane on the test set is 97%.

Claims (7)

1. A machine vision detection method for boiler pipeline surface defects is characterized by comprising the following steps:
step 1, acquiring z boiler pipeline surface images as sample images by an industrial camera with the aid of an illumination system, wherein the sample images comprise normal pipeline surface images and pipeline surface defect images;
step 2, respectively preprocessing each sample image and extracting a feature vector of the image;
step 3, respectively taking the feature vectors of z sample images as input vectors of a support vector machine, then establishing an optimal classification hyperplane according to a design criterion of maximum classification interval, and determining a decision function by the optimal classification hyperplane;
step 4, acquiring a surface image of the boiler pipeline to be detected by an industrial camera with the aid of an illumination system;
step 5, preprocessing the surface image of the boiler pipeline to be detected, and extracting a feature vector of the image;
step 6, inputting the characteristic vector of the surface image of the boiler pipeline to be detected into a decision function, and judging whether the surface of the boiler pipeline to be detected has defects or not;
the step 2 and the step 5 for extracting the feature vector of the image comprise the following steps:
s1, extracting a binary vector of the image: segmenting the preprocessed image by adopting a self-adaptive threshold segmentation processing method to generate a binary image; converting the binary image into a vector form, namely generating a corresponding binary vector by the gray value of each pixel point on the binary image;
s2, extracting LBP vector of the image: firstly, processing a preprocessed image by adopting a local binarization method: the method comprises the steps of determining an LBP value of each pixel in an image through a circular LBP operator with the radius of 2 pixels, establishing an LBP statistical histogram by taking the LBP value as a horizontal coordinate and the occurrence frequency of each value as a vertical coordinate, and setting the size of the LBP statistical histogram to be consistent with that of a binary image, even if the number of pixels contained in the horizontal length and the longitudinal length of the LBP statistical histogram is kept to be consistent with that of the binary image; then, the LBP statistical histogram is respectively converted into a vector form, namely, a corresponding LBP vector is generated according to the gray value of each pixel point on the LBP statistical histogram;
s3, feature dimension reduction: and respectively carrying out dimensionality reduction on the binary vector and the LBP vector of the image, and merging the two vectors obtained after dimensionality reduction to serve as the feature vector of the image.
2. The method for machine vision inspection of boiler tube surface defects according to claim 1, wherein the step 2 and the step 5 pre-process the images by: firstly, increasing the contrast of an image by adopting a histogram equalization method; and then, removing the noise interference in the image acquisition process by adopting a self-adaptive median filtering method.
3. The method of claim 1, wherein in step 3, the functional relationship of the classification hyperplane is as follows:
Figure FDA0002937681060000011
the decision function is
Figure FDA0002937681060000021
Wherein y is a class label of the surface image of the boiler pipeline to be detected,
Figure FDA0002937681060000022
is the characteristic vector of the surface image of the boiler pipeline to be detected,
Figure FDA0002937681060000023
is the feature vector of the ith sample image, yiA class label corresponding to the feature vector of the ith sample image; c. CiFor lagrange multipliers, each eigenvector
Figure FDA0002937681060000024
Corresponds to one ci，0≤ciH is less than or equal to H, H is a punishment parameter, b is a deviation term, gamma>0, gamma is a kernel function parameter; c. CiB, H and gamma are parameters to be optimized;
converting the problem of solving the function relation into the objective function
Figure FDA0002937681060000025
Under the constraint condition
Figure FDA0002937681060000026
0≤ci≤H，γ>The maximum at 0; and solving through a genetic algorithm and an SMO algorithm, and the method comprises the following steps:
1) determining iteration parameters including the population number N, the iteration times T, the cross probability p, the variation probability q, the variation range of a penalty parameter H and the variation range of a kernel function parameter gamma;
2) randomly selecting N sets of punishment parameters H and kernel function parameters gamma as N sets of original data, then respectively calculating the defect identification accuracy rate corresponding to each set of data as the fitness value of the defect identification accuracy rate, and recording the maximum fitness value and a set of data corresponding to the fitness value;
3) selecting iterative data from the current N groups of data by using a roulette algorithm, converting the iterative data into a binary form, and then performing intersection and variation to obtain N groups of new data;
4) calculating the defect identification accuracy corresponding to the N groups of new data to be used as the fitness values of the N groups of new data, and recording the maximum fitness value and a group of data corresponding to the fitness value;
5) returning to the step 3) to obtain data of the next iteration by using the roulette algorithm again, and repeating the steps until T iterations are completed;
6) outputting the maximum fitness value recorded in the T iterative processes and a group of data corresponding to the maximum fitness value, and recording as a penalty parameter H*With the kernel function parameter gamma*；
7) Penalty parameter H to be obtained*With the kernel function parameter gamma*Substituting the H and the gamma into an objective function L (c), and solving the maximum value of the objective function L (c) and the c corresponding to the maximum value under the constraint condition by using an SMO algorithmiAnd the value of b, denoted as ci *And b*；
8) Will find gamma*、ci *And b*Substituting the function relation into the function relation of the classification hyperplane to obtain the function relation of the classification hyperplane as follows:
Figure FDA0002937681060000031
4. the machine vision inspection method for the surface defects of the boiler tube according to claim 3, wherein in the step 1), a population number N is set to 20, an iteration number T is set to 180, a cross probability p is set to 0.4, a variation probability q is set to 0.01, and a penalty parameter H is set to*Has a variation range of (0.1,100) and a kernel function parameter gamma*The variation range of (c) is (0.01,1000).
5. The machine vision inspection method for the surface defects of the boiler pipes according to claim 3, wherein in the step 4, the industrial camera is fixed on the lifting platform through the two-degree-of-freedom pan-tilt, the lifting platform and the two-degree-of-freedom pan-tilt drive the industrial camera to move up and down, left and right, under the assistance of the illumination system, the industrial camera continuously acquires the surface images of the boiler pipes to be inspected according to a set time interval and outputs the images to an upper computer for real-time inspection, and once the defects exist on the surface of the boiler pipes, an alarm signal is triggered.
6. A machine vision detection system for surface defects of boiler pipelines is characterized by comprising an industrial camera, a two-degree-of-freedom cradle head, a lifting platform, an illumination system and an upper computer; the industrial camera and the illumination system are fixed on the lifting platform through the two-degree-of-freedom holder, and the lifting platform and the two-degree-of-freedom holder drive the industrial camera to move up and down, left and right for realizing continuous acquisition of the surface images of the boiler pipeline; the industrial camera collects the surface image of the boiler pipeline under the assistance of the illumination system and outputs the image to the upper computer for detection; the system adopts the method of any one of claims 1-4 to realize the surface defect detection of the boiler pipeline.
7. The machine vision detection system for the surface defects of the boiler pipes as claimed in claim 6, wherein an alarm module is arranged on the upper computer, and an alarm signal is triggered once the defects on the surface of the boiler pipe are detected.
CN201710994305.2A 2017-10-23 2017-10-23 Machine vision detection method and system for boiler pipeline surface defects Active CN107749058B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710994305.2A CN107749058B (en) 2017-10-23 2017-10-23 Machine vision detection method and system for boiler pipeline surface defects

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710994305.2A CN107749058B (en) 2017-10-23 2017-10-23 Machine vision detection method and system for boiler pipeline surface defects

Publications (2)

Publication Number Publication Date
CN107749058A CN107749058A (en) 2018-03-02
CN107749058B true CN107749058B (en) 2021-05-04

Family

ID=61253136

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710994305.2A Active CN107749058B (en) 2017-10-23 2017-10-23 Machine vision detection method and system for boiler pipeline surface defects

Country Status (1)

Country Link
CN (1) CN107749058B (en)

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108645865A (en) * 2018-05-10 2018-10-12 中国石油天然气集团有限公司 A kind of measurement method of the submerged-arc welding steel pipe weld seam amount of the being partially welded parameter based on CCD
CN109800824B (en) * 2019-02-25 2019-12-20 中国矿业大学(北京) Pipeline defect identification method based on computer vision and machine learning
CN111257422B (en) * 2020-02-28 2023-09-08 北京新联铁集团股份有限公司 Wheel axle defect identification model construction method and defect identification method based on machine vision
CN111768404A (en) * 2020-07-08 2020-10-13 北京滴普科技有限公司 Mask appearance defect detection system, method and device and storage medium
CN112432954A (en) * 2020-12-09 2021-03-02 浙江理工大学 Braided tube flaw detection method
CN113034480B (en) * 2021-04-01 2023-12-19 艾德领客(上海)数字技术有限公司 Blast furnace damage analysis method based on artificial intelligence and image processing
CN113658094A (en) * 2021-07-06 2021-11-16 大唐东北电力试验研究院有限公司 Thermal power plant boiler four-pipe intelligent detection method based on vision
CN115082923B (en) * 2022-08-24 2022-11-08 成都工业学院 Milk packing box production date identification method based on machine vision
CN115755717B (en) * 2022-11-29 2023-08-29 潮州市索力德机电设备有限公司 Kiln equipment operation detecting system based on thing networking
CN115797340A (en) * 2023-02-03 2023-03-14 西南石油大学 Industrial surface defect detection method based on multi-instance learning

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1656371A (en) * 2002-05-21 2005-08-17 杰富意钢铁株式会社 Surface defect judging method
CN101661004A (en) * 2009-07-21 2010-03-03 湖南大学 Visible detection method of welding quality of circuit board based on support vector machine
CN102680478A (en) * 2012-04-25 2012-09-19 华南农业大学 Detection method and device of surface defect of mechanical part based on machine vision
CN104198497A (en) * 2014-09-12 2014-12-10 苏州大学 Surface defect detection method based on visual saliency map and support vector machine
CN106568783A (en) * 2016-11-08 2017-04-19 广东工业大学 Hardware part defect detecting system and method

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8345949B2 (en) * 2010-03-09 2013-01-01 General Electric Company Sequential approach for automatic defect recognition

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1656371A (en) * 2002-05-21 2005-08-17 杰富意钢铁株式会社 Surface defect judging method
CN101661004A (en) * 2009-07-21 2010-03-03 湖南大学 Visible detection method of welding quality of circuit board based on support vector machine
CN102680478A (en) * 2012-04-25 2012-09-19 华南农业大学 Detection method and device of surface defect of mechanical part based on machine vision
CN104198497A (en) * 2014-09-12 2014-12-10 苏州大学 Surface defect detection method based on visual saliency map and support vector machine
CN106568783A (en) * 2016-11-08 2017-04-19 广东工业大学 Hardware part defect detecting system and method

Also Published As

Publication number Publication date
CN107749058A (en) 2018-03-02

Similar Documents

Publication Publication Date Title
CN107749058B (en) Machine vision detection method and system for boiler pipeline surface defects
CN107782733B (en) Image recognition nondestructive detection device and method for metal surface defects
CN104268505A (en) Automatic cloth defect point detection and recognition device and method based on machine vision
CN110689011A (en) Solar cell panel defect detection method of multi-scale combined convolution neural network
CN109544522A (en) A kind of Surface Defects in Steel Plate detection method and system
CN107481237B (en) A kind of infrared array image hot spot detection method based on multiframe temperature characterisitic
CN108108736A (en) A kind of solar energy photovoltaic panel spot identification method
CN109376740A (en) A kind of water gauge reading detection method based on video
CN114241364A (en) Method for quickly calibrating foreign object target of overhead transmission line
CN116452594B (en) Visualized monitoring and early warning method and system for power transmission line state
CN115862073B (en) Substation hazard bird species target detection and identification method based on machine vision
CN115597494B (en) Precision detection method and system for prefabricated part preformed hole based on point cloud
CN111402249B (en) Image evolution analysis method based on deep learning
CN110009601A (en) Large-Scale Equipment irregular contour detection method of surface flaw based on HOG
Wu et al. An end-to-end learning method for industrial defect detection
CN115147377A (en) Training method and device for CycleGAN model for generating defect images of photovoltaic panel
CN116229052A (en) Method for detecting state change of substation equipment based on twin network
Jia et al. A modified centernet for crack detection of sanitary ceramics
CN106989672A (en) A kind of workpiece measuring based on machine vision
CN117269168B (en) New energy automobile precision part surface defect detection device and detection method
Daogang et al. Anomaly identification of critical power plant facilities based on YOLOX-CBAM
CN116485802B (en) Insulator flashover defect detection method, device, equipment and storage medium
Özer et al. An approach based on deep learning methods to detect the condition of solar panels in solar power plants
CN116664941A (en) Visual detection method for surface defects of bearing ring
CN113034465B (en) Power equipment thermal fault monitoring method, device and medium based on infrared image

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant