CN108956614A - A kind of pit rope dynamic method for detection fault detection and device based on machine vision - Google Patents
A kind of pit rope dynamic method for detection fault detection and device based on machine vision Download PDFInfo
- Publication number
- CN108956614A CN108956614A CN201810432434.7A CN201810432434A CN108956614A CN 108956614 A CN108956614 A CN 108956614A CN 201810432434 A CN201810432434 A CN 201810432434A CN 108956614 A CN108956614 A CN 108956614A
- Authority
- CN
- China
- Prior art keywords
- wirerope
- image
- damage
- neural network
- matrix
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
- G01N2021/8854—Grading and classifying of flaws
Abstract
The invention belongs to technical field of image processing, propose a kind of pit rope dynamic method for detection fault detection and device based on machine vision.Method includes the following steps: recording the wirerope in dynamic motion by video camera sequence photography obtains vision signal；Vision signal is handled, moving target is extracted and obtains the oscillation trajectory of each pixel on wirerope edge feature；Obtain the flexibility matrix of wirerope to be measured；Step S4: the damage position and degree of injury of wirerope to be measured are obtained using the BP neural network after sample training；The damage position unit judged by BP neural network is extracted from whole image as ROI region and saves into new damage of steel cable initial pictures；Damage of steel cable initial pictures are handled and extracted with feature, the surface defect based on textural characteristics is made using linear classifier and differentiates.The present invention solves the problems, such as image detection fault, easy to operate, at low cost.
Description
Technical field
The invention belongs to technical field of image processing, and in particular to a kind of pit rope dynamic spy based on machine vision
Triage surveys method and apparatus.
Background technique
With the diversification of coal mining method, the winch of various models also plays in coal mine sloping lane transport important
Role.Personnel, equipment disengaging mine are required by winch jacking system, and wirerope needs to undertake heavy burden when work,
And subsurface environment is special, and wirerope is easy to appear the peace such as steel rope abrasion, crackle, burn into fracture of wire in the long-term use
Full hidden danger, more change steel rope is easy to appear wire cable rupture not in time, and great accident will occur.If can go out in wirerope
Timely more change steel rope, can effectively avoid serious accident from occurring before existing rope-broken accident.
Wirerope generally uses no more than 2 years, in spite of damaging, must all replace.The lifting of coal mine winch
The value of wirerope used in system is this to only rely on the side that service life decides whether replacement generally four or five million or so
On the one hand there is very big waste in formula, another reverse side cannot guarantee that safety, sometimes if using if incorrect, wirerope
It just will appear the abnormal conditions of abrasion, fracture, distortion, deformation using 1 year or so.
Wire rope flaw detection detection includes Manual Visual Inspection and electromagnetic detection two major classes.Manual Visual Inspection, which refers to, is equipped with special work
Whether personnel are periodically had damage with naked-eye observation wirerope.This method detection time is long, large labor intensity, and testing staff is easily tired
Labor, inefficiency and the larger specialized capability and working attitude for relying on staff of this method, subjectivity lead to by force omission factor height.
Then electromagnetic detection, which refers to magnetize on wirerope to be measured, detects leakage field.The device is complicated for such method, with high costs, and signal is easy
Distortion is generated by external interference, magnetization unevenness can also bring very big error to testing result.
The technological deficiency for how overcoming the wire rope flaw detection detection system in traditional technology, be those skilled in the art urgently
Technical problem to be solved.
Summary of the invention
The present invention overcomes the shortcomings of the prior art, technical problem to be solved are as follows: it is accurate to provide a kind of detection, behaviour
Make simply pit rope dynamic method for detection fault detection and device based on machine vision.
In order to solve the above-mentioned technical problem, a kind of the technical solution adopted by the present invention are as follows: mining steel based on machine vision
Cord dynamic method for detection fault detection, comprising the following steps:
Step S1: setting up video camera in front of wirerope, and actual range representated by calibrating camera unit pixel obtains
Distance calibration parameter, moves wirerope at the uniform velocity, records the wirerope in dynamic motion by video camera sequence photography and is regarded
Frequency signal；
Step S2: handling vision signal, extracts moving target and obtains each pixel on wirerope edge feature
Oscillation trajectory；
Step S3: according to the oscillation trajectory of pixel each on wirerope edge feature, wirerope is obtained along rope length direction point
The displacement space sequence matrix of cloth makees discrete Fourier transform to each column of the matrix, obtains the frequency of each point on corresponding wirerope
Function is rung, intrinsic frequency, and the amplitude of the response frequency according to the frequency response function of each point are identified according to the peak value of frequency response function
The Mode Shape of wirerope is obtained, the flexibility matrix of wirerope to be measured is finally obtained；
Step S4: setting BP neural network sets the input parameter of BP neural network to the flexibility matrix of wirerope, defeated
Output is set as structural damage unit and damaged degree quantized value, every wirerope is divided into multiple units, and give each list
Multiple Stiffness values are arranged in member, obtain the flexibility matrix of a variety of damaged wirerope, are trained as training sample；According to
The flexibility matrix for surveying wirerope obtains the damage position and damage journey of wirerope to be measured using the BP neural network after sample training
Degree；
Step S5: the damage position unit judged by BP neural network is mentioned from whole image as ROI region
It takes out and saves into new damage of steel cable initial pictures；
Step S6: according to Retinex principle, high-pass filtering is first carried out to damage of steel cable initial pictures, then be based on
The greyscale transformation of imadjust；Then the smoothness R and entropy e of the image after extraction process make base using linear classifier
Differentiate in the surface defect of textural characteristics.
In the step S2, the step of processing vision signal are as follows:
Step S201: background model the foundation of background model: is established using the method that multiple image is averaged；
Step S202: the detection of region of variation: two field pictures continuous in video sequence are subjected to difference processing, determine back
The region of scape and the region of motion change；
Step S203: moving object detection: to current frame image processing, only image in region of variation is done with background image
Difference detects the object of movement；
Step S204: edge grey scale change curve object edge identification: is fitted by using 5 order polynomials to extract sub- picture
The marginal point of plain precision；To a series of images carry out processing obtain wirerope edge feature on each pixel in certain time period
Oscillation trajectory.
In the step S4, when the training sample of sample training BP neural network is set, every wirerope is divided into 8
Unit, according to the difference of fracture of wire situation, it is assumed that its rigidity declines 25%, 40% respectively, obtains the damaged condition of 16 kinds of degree, meter
The intrinsic frequency and Mode Shape for calculating 16 kinds of damaged wirerope vibrations, take preceding two order frequency and the vibration shape, obtain 16 class values, to its into
Row GSL transformation, as the training sample of neural network.
In the step S4, when sample training BP neural network is arranged, selection Sigmoid function is excitation function, is selected
Error signal back-propagation algorithm is training algorithm, the renewal process formula of weight are as follows:
In formula, the gradient of first item Representative errors average value, Section 2 represents instantaneous item, and Section 3 represents random noise
, t indicates the number of iterations,Indicate the L layers of weight of the t times iteration,Indicate delta error,It indicates
The output of L layers of j-th of neuron of k-th of training sample, η indicate that Studying factors, μ indicate momentary constant,It indicates
Random noise item.
In the step S6, the calculation formula of smoothness R are as follows: R=1-1/ (1+ σ2)；The calculation formula of entropy e are as follows:In formula, σ indicates the standard deviation of image, calculation formula are as follows:
L indicates gray level, ziIndicating each pixel gray value in image-region, m indicates average gray level,p
(zi) indicate that each pixel gray value is z in image-regioniProbability.
In the step S6, high-pass filtering is carried out to damage of steel cable initial pictures and based on the greyscale transformation of imadjust
Afterwards, image is divided further into smaller part, extracts the smoothness value and entropy of each part, passes through linear classifier pair
Each section carries out classification judgement.
In the step S6, linear classifier includes smoothness classification thresholds T1 and entropy classification thresholds T2, when putting down for image
Slippery R > T1 and when entropy e > T2, determines that surface is defective, as the smoothness R < T1 of image and entropy e < T2, determines surface
It is intact.
The present invention also provides a kind of pit rope dynamic carrying out flaw detection device based on machine vision, including image obtain
Take equipment, controller, memory and display；Described image obtains the vision signal that equipment is used to obtain wirerope；The control
Device processed is for executing following procedure:
Moving target is extracted from vision signal, obtains the oscillation trajectory of each pixel on wirerope edge feature；
Displacement space sequence matrix of the wirerope along rope length directional spreding is obtained according to oscillation trajectory, to each column of the matrix
Make discrete Fourier transform, obtains the frequency response function of each point on corresponding wirerope, identified inherently according to the peak value of frequency response function
Frequency, and the Mode Shape of wirerope is obtained according to the amplitude of the response frequency of the frequency response function of each point, finally obtain wirerope
Flexibility matrix；
BP neural network is set, sets the input parameter of BP neural network to the flexibility matrix of wirerope, output quantity is set
It is set to structural damage unit and damaged degree quantized value, every wirerope is divided into multiple units, and be arranged to each unit
Multiple Stiffness values obtain the flexibility matrix of a variety of damaged wirerope, are trained as training sample；According to steel wire to be measured
The flexibility matrix of rope obtains the damage position and degree of injury of wirerope to be measured using the BP neural network after sample training；
The damage position unit judged by BP neural network is extracted simultaneously from whole image as ROI region
Save into new damage of steel cable initial pictures；
According to Retinex principle, high-pass filtering is first carried out to damage of steel cable initial pictures, then is carried out based on imadjust
Greyscale transformation；Then the smoothness R and entropy e of the image after extraction process are made using linear classifier based on textural characteristics
Surface defect differentiate；
The memory is used to store the wirerope picture of damage；The display equipment is used to show the wirerope figure of damage
Piece simultaneously marks damage position and degree of injury.
Compared with the prior art, the invention has the following beneficial effects: the present invention can be during normal production in real time
Video photography is carried out to the state of wirerope, and background segment is carried out to wirerope, be capable of providing most complete characteristic and
Moving target is extracted, position is accurate, and speed is fast, and can solve single-frame images to seek background color when verge searching target bright
Degree grey scale change bring target seeks failure.Using flexibility matrix non-destructive tests in such a way that image detection combines, benefit
Influence of the wirerope stress variation to flexibility caused by damage solves merely using image detecting technique because of wirerope surface
The spot that the greasy dirt of attachment and other industrial impurity greasy dirts are formed is fallen in the rope strand image range of extraction, caused by image examine
Dendrometry misses problem.Complicated equipment is not needed simultaneously, it is easy to operate, it is at low cost.
Detailed description of the invention
Fig. 1 is that a kind of process of pit rope dynamic method for detection fault detection based on machine vision of the invention is illustrated
Figure；
Fig. 2 is Marx's Weir mechanical model of wirerope；
Fig. 3 is the algorithm flow chart of BP neural network；
Fig. 4 is the flow diagram of linear classifier of the invention；
Fig. 5 is a kind of structural representation of pit rope dynamic carrying out flaw detection device based on machine vision of the invention
Figure.
Specific embodiment
It in order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below will be in the embodiment of the present invention
Technical solution be clearly and completely described, it is clear that described embodiment is a part of the embodiments of the present invention, without
It is whole embodiments；Based on the embodiments of the present invention, those of ordinary skill in the art are not before making creative work
Every other embodiment obtained is put, shall fall within the protection scope of the present invention.
As shown in Figure 1, the present invention provides a kind of pit rope dynamic method for detection fault detection based on machine vision, packet
Include following steps:
Step S1: setting up industrial camera in front of wirerope, demarcates reality representated by industrial camera unit pixel
Distance obtains distance calibration parameter, moves wirerope at the uniform velocity, records the steel wire in dynamic motion by video camera sequence photography
Rope obtains vision signal.
Step S2: handling vision signal, extracts moving target and obtains each pixel on wirerope edge feature
Oscillation trajectory.
Wherein, the step of vision signal being handled are as follows:
Step S201: background model the foundation of background model: is established using the method that multiple image is averaged.Using more
The method that frame image is averaged establishes background model, and formulae express is as follows:
Wherein, N is the image sequence frame number rebuild, BKFor the background image of reconstruction, fkFor kth frame image, in background image
The value of each pixel is the cumulative mean of pixel N frame image grayscale.
Step S202: the detection of region of variation: two field pictures continuous in video sequence are subjected to difference processing, determine back
The region of scape and the region of motion change.
Wherein, frame differential method is capable of detecting when the region of variation that adjacent two interframe occurs, this practical region includes fortune
The region P that animal body is covered in former frame, region i.e. moving object that moving object is covered in the current frame Q itself.If
fk(i, j) and fk+1(i, j) is continuous two field pictures in video sequence, wherein i, and j indicates pixel coordinate value, by this two frame
Image carries out difference processing, and detected rule is as follows:
Here T is the threshold value of detection, since region of variation needs to be further processed with background image to be partitioned into
Moving object, therefore here need not be accurate to the selection of T value, adaptation range is very wide, takes 15 here.Bk+1After (i, j) indicates difference
It is determined as the region of background, Mk+1It is determined as the region of motion change after (i, j) expression difference.
Step S203: moving object detection: to current frame image processing, only image in region of variation is done with background image
Difference detects the object of movement.
Wherein, it distinguishes in image after region of variation and non-changing region, for current frame image, only to region of variation
Middle image and background image do difference to detect the object of movement.
After current frame image and background image difference, chosen according to specific application and picture quality different
Threshold value carries out Threshold segmentation, and difference image is become bianry image.After Threshold segmentation, since the jamming target of noise will appear
Some small holes and burr are removed some holes and burr using the method corroded in morphological image to bianry image and expanded
It removes.
Step S204: edge grey scale change curve object edge identification: is fitted by using 5 order polynomials to extract sub- picture
The marginal point of plain precision；To a series of images carry out processing obtain wirerope edge feature on each pixel in certain time period
Oscillation trajectory.
Different moments, corresponding every frame image was represented by m row n column matrix, adjacent two field pictures during object vibration
The difference of middle edge feature spatial position is to reflect the vibration processes of actual object, to the difference of edge feature spatial position into
Rower is fixed and converts, and can find out the oscillation trajectory of corresponding moment object.Carrying out processing to a series of images can be obtained structure
In the oscillation trajectory and vibration performance of certain time period, wherein the identification and its positioning of edge feature are the passes that this method is realized
Key.
Marginal position can be determined by the mathematical feature of the gray value of edge.The grey scale change of the image border of wirerope
It is impulse type edge, using the boundary point set of the adjacent two width gray level image of classical operators such as sobel operator extraction, only extracts here
The boundary point set of wirerope image right.Two one-dimension arrays containing n whole pixel edge positions are obtained, K is denoted as1(n)、K2
(n), it is expressed as follows:
K1(n)=[k1(1),k1(2),…,k1(n)]； (3)
K2(n)=[k2(1),k2(2),…,k2(n)]； (4)
But the pixel pixel of above-mentioned gray level image is integer, it is clear that the precision of displacement only has whole Pixel-level.Below
The marginal point of sub-pixel precision will be extracted by polynomial fitting method.
Edge grey scale change curve is fitted using 5 order polynomials:
I (z)=(c0+c1z+c2z2+…+c5z5)； (5)
In formula (5), I indicates that gray value at z point, z indicate sub-pixel edge position, c0,…,c5Indicate that fitting is multinomial
Formula coefficient.
Enable K1(i)=j, j ∈ [1, m] indicate the edge group number K of 1 image of state1(n) element in, every frame image
It is expressed as m row n column matrix, then (i, j) is the position at edge.Extract 6 pixels of the marginal point (i, j) near column direction
Point (including the edge) and its corresponding gray value.Polynomial function is carried it into solve parameter c to be determined0,…,c5,
Such as following formula:
The parameter c that above formula is acquired0,…,c5Take back I (z)=(c0+c1z+c2z2+…+c5z5) its second dervative is sought, enable it
Second dervative is the zero location point Z that can solve sub-pixel edge, and Z ∈ (j-1, j+1).
Fitting by column is carried out to the edge gray scale of state 1 and 2 image of state and is solved, available a series of sub-pix
Marginal position is denoted as Z respectively1(n) and Z2(n), then Z1(n) and Z2(n) difference is obtained Displacement, is denoted as y
(n).It has corresponded to the vibration displacement that wirerope is in 2 moment of state, using this method to each image of wirerope vibration processes
It is handled frame by frame, the dynamic displacement that each point on wirerope is in each vibration moment can be acquired.
Each frame image that wirerope vibrates as above is handled using fitting of a polynomial gray-scale intensity method, acquires its feature
The oscillation trajectory of each pixel on edge, is denoted as y (t), then the displacement space sequence along the distribution of steel wire rope length direction can be denoted as
y1,y2,…,yn, each element therein is the function of time, and then y (t) can be expressed as the form of one-dimension array, it may be assumed that
Y (t)=[y1(t),y2(t),…,yi(t),…,yn(t)], i=1,2 ... n； (7)
Wherein yi(t) displacement function about time t of some pixel on the edge feature of object vibrational image is indicated.
Step S3: according to the oscillation trajectory of pixel each on wirerope edge feature, wirerope is obtained along rope length direction point
The displacement space sequence matrix of cloth makees discrete Fourier transform to each column of the matrix, obtains the frequency of each point on corresponding wirerope
Function is rung, intrinsic frequency, and the amplitude of the response frequency according to the frequency response function of each point are identified according to the peak value of frequency response function
The Mode Shape of wirerope is obtained, the flexibility matrix of wirerope to be measured is finally obtained.
Wherein, wirerope is denoted as y (t, i) along the displacement space sequence matrix of rope length directional spreding, i=1 in formula, 2 ... n
Indicate pixel position, the frequency response function of each point on corresponding wirerope can be obtained by making discrete Fourier transform to each column of the matrix
H, it may be assumed that
H (ω, i)=FFT (y (t, i)) ω ∈ [0,1/ △ t]； (8)
S width image is wherein acquired in time t, then respectively t at the time of recording corresponding when each image1,t2,…,
ts, time interval is denoted as △ t, then has: △ t=t2-t1=t3-t2=...=ts-ts-1；Data in frequency response function H (ω, i) are equal
For plural number, it can be indicated, can also be indicated by real and imaginary parts by amplitude and phase, be the vector of complex plane, expression
Formula are as follows:
H (ω)=HR(ω)i+HI(ω)j； (9)
Amplitude-versus-frequency curve is the relation curve between frequency response function amplitude and frequency, and extreme point occurs in structure
Natural frequency ω0Place, i.e., | H (ω0) | for displacement peak response amplitude, therefore by frequency response function structure natural frequency of vibration meeting
There is this characteristic of peak value, can identify intrinsic frequency.
Mode Shape can obtain by the amplitude combinations of the response frequency of the response spectra of each point, and with response spectra amplitude maximum
Point the response spectral amplitude ratio of other each points is normalized, i.e., when frequency be the i-th rank intrinsic frequency when, frequency response function width
Value can approximate representation are as follows:
In formula, Φri、Mi、ζiIt is all the constant determined by the i-th rank mode, φ1i……φNiIndicate each freedom degree vibration of i rank
Type matrix.Thus frequency response function is directly proportional to the i-th rank mode of structure, ignores the influence of remaining mode, then each freedom degree frequency response
The ratio between the peak value of function amplitude-versus-frequency curve at certain rank modal frequency approximate can be used as the rank Mode Shape.
Mine hoisting steel cable is really a linear glutinous, elastomer, rather than rigid body.We can use Marx's Weir
Model (C.Maxwell model) simulates the characteristic of wirerope.The model is by spring (elastic model (H)) and damper
(viscosity model (N)) is composed in series, and the structural schematic diagram is as shown in Fig. 2, model symbol indicates are as follows: M=H-N.One structure
Rigidity and flexibility matrix can be acquired by its modal parameter, i.e. eigenfrequncies and vibration models, if known system mass matrix M, Gu
There is frequencies omegai(i=1~n), vibration shape matrix φ, then stiffness matrix K and flexibility matrix distinguish F are as follows:
Wherein, Λ is frequency matrix,N is number of degrees of freedom,.
Vibration shape matrix Φ meets regular conditions: ΦTM Φ=I ΦTK Φ=Λ.
It can be seen that square being inversely proportional for stiffness matrix and intrinsic frequency from (11) formula, it is therefore, accurate in order to obtain one
Stiffness matrix estimated value, it is necessary to measure all modal parameters, or at least high-order.But can see from (12) formula,
Flexibility matrix and intrinsic frequency square are inversely proportional, it means that and with the increase of intrinsic frequency, flexibility matrix is restrained quickly,
Therefore, it can be obtained by the fine estimation of flexibility matrix from the modal parameter of some low orders.
When crack or local damage occurs in steel cord structure, its stiffness matrix can reduce, since rigidity and flexibility are deposited
Relationship FK=I (I is unit matrix), then the reduction of stiffness matrix inherently causes the increase of flexibility matrix, if by flexibility
As the sensitive parameter of characterization structural failure, explicit physical meaning, and implement also more convenient.Based on flexibility matrix
The characteristics of, we diagnose degree of impairment the sensitive parameter by the change of flexibility matrix as characterization structural failure.
After structural failure, (FA+△F)(KA+ △ K)=I；Above formula expansion arranges :-FE△ K=(FE-FA)KA；Enable E=-
FE△ K, then E=(FE-FA)KA.Wherein,△ F, △ K are respectively the variation of flexibility matrix and stiffness matrix
Amount, I are unit matrix, FA, FEFlexibility matrix respectively before structural failure and after damaged, KAFor the rigidity square before structural failure
Battle array, ΦiFor the i-th first order mode for meeting regular conditions, ωiFor the i-th order frequency, n is number of degrees of freedom,.
If degree of disturbing direction freedom degree is only considered, for preceding order frequency and the vibration shape, by E=-FEThe available matrix E of △ Kl.If
δeiFor matrix ElCorrespondence row element dijThe maximum value of absolute value, it may be assumed that δei=max | dij|, i, j are degree of disturbing direction number of degrees of freedom,.
Therefore, by being analyzed above it is found that the flexibility matrix of a structure can be by its modal parameter, i.e. intrinsic frequency and vibration
Type acquires.It is calculated by the above-mentioned processing to picture, after obtaining the eigenfrequncies and vibration models of wirerope, available wirerope
Flexibility matrix.
Step S4: setting BP neural network sets the input parameter of BP neural network to the flexibility matrix of wirerope, defeated
Output is set as structural damage unit and damaged degree quantized value, every wirerope is divided into multiple units, and give each list
Multiple Stiffness values are arranged in member, obtain the flexibility matrix of a variety of damaged wirerope, are trained as training sample；According to
The flexibility matrix for surveying wirerope obtains the damage position and damage journey of wirerope to be measured using the BP neural network after sample training
Degree.
Wherein, BP neural network is made of Three Tiered Network Architecture, and first layer is input layer, and the second layer is hidden layer, finally
One layer be output layer, when the output of signal does not reach the preset aim of learning, then will by backpropagation again into
Row training, when carrying out back transfer, error signal is successively returned along the neuron path most started, and when every layer of return is continuous
The weight size for modifying the connection between neuron, is finally completed backpropagation.BP neural network algorithm flow chart such as Fig. 3 institute
Show, in Fig. 3, p representative sample, t represents the number of iterations.Capitalizing P is number of samples, and small letter p is to indicate p-th of sample.It sets first
T=1 calculates first time iteration.Initial value 1 is assigned to p, if the total number of samples of small letter p < capitalizes P, small letter p is constantly followed from adding 1
Ring, until small letter p then calculates reversion error not less than capitalization P..Such as be not up to standard, then the number of iterations t from plus one, calculate the
Second iteration does not stop circulation and is iterated calculating, until reaching standard, then calculates stopping.Based on forward-propagating and reversed biography
The continuous iteration of process is broadcast, trained final purpose is to make error signal control in defined zone of reasonableness.This implementation
In example, when sample training BP neural network is arranged, selection Sigmoid function is excitation function, selects error signal backpropagation
Algorithm is training algorithm.Assuming that the transfer function of L-1 layers of j-th of neuron of p-th of training sample and output are respectivelyWithThen:
In formula (13), E is global error；For weight；For delta error.
In formula (14), m indicates m-th of neuron, and logarithm tangent function is one of Sigmoid function functional form,
Its citation form isWeightRenewal process formula are as follows:
In formula (15), t indicates the number of iterations,Indicate delta error,Indicate the L of p-th of training sample
The output of j-th of neuron of layer,Indicate that the L layers of weight of the t times iteration, μ indicate that momentary constant takes 0.8, η to indicate
Studying factors take 0.1,Indicate random noise item.
In addition, the acquisition process of training sample is as follows: every wirerope being divided into 8 units, according to fracture of wire situation
Difference, rigidity decline 25%, 40% respectively, thus obtain the damaged condition of 16 kinds of degree, calculate 16 kinds of damaged wirerope
The frequency and the vibration shape of vibration take preceding two order frequency and the vibration shape, obtain 16 class values, it is carried out GSL transformation (i.e. orthogonal intersection space method,
It is capable of the Localization Phenomenon of pointwise correction data point, is uniformly distributed it in data space), as the input of neural network
Parameter.Studying factors η is determined as 0.1.
BP neural network finally exports that the results are shown in Table 1.
Table 1BP neural network finally exports
Step S5: the damage position unit judged by BP neural network is mentioned from whole image as ROI region
It takes out and saves into new damage of steel cable initial pictures.
Step S6: according to Retinex principle, high-pass filtering is first carried out to damage of steel cable initial pictures, then be based on
The greyscale transformation of imadjust；Then the smoothness R and entropy e of the image after extraction process make base using linear classifier
Differentiate in the surface defect of textural characteristics.
Wherein, defect texture properties and metastable feature can be reflected to pretreated image zooming-out, as knowledge
The foundation of other defect.The selection of the defect characteristic of the present embodiment depends on the signature analysis and experiment analysis results of defect.
The criterion of selection feature is characteristic value difference of the same race most it will be evident that ensuring that feature has biggish mutual independence, final choice
Smoothness and entropy are analyzed.
The wherein calculation formula of smoothness are as follows:
R=1-1/ (1+ σ2)； (16)
In formula (16), σ indicates the standard deviation of image, calculation formula are as follows:L indicates ash
Spend grade, ziIndicating each pixel gray value in image-region, m indicates average gray level,p(zi) indicate
Each pixel gray value is z in image-regioniProbability.The roughening of image texture, smoothness can be described using smoothness R
Low texture texture higher than smoothness in gray level has smaller changeability, i.e. smoothness is lower, and image is more flat.
The calculation formula of entropy are as follows:
In formula, p (zi) it is that each pixel gray value is z in image-regioniProbability；L is gray level.It can be retouched using entropy e
The randomness for stating image texture, in wirerope image, surface has the entropy e of the wirerope of defect than the entropy of intact wirerope
It beats, representative has the randomness of defect wirerope texture bigger.
In addition, have certain correlation in view of between each feature of wirerope image, in order to reduce the complexity of recognizer, this
Embodiment utilizes linear classifier when determining, form and the decision rule of linear classifier discriminant function are shown below:
G (x)=WTX+w0； (18)
WhereinN is characterized the dimension in space.
If
If g (X)=0, refuses to classify.
Using above-mentioned textural characteristics, the state of wirerope is differentiated.Characteristic threshold value is by experimental analysis and feature
Analysis result obtains.In decision process, in order to increase the accuracy of testing result, first by the image of wirerope further into
Row is divided into smaller part, extracts the smoothness value and entropy of each part, and be compared with classification thresholds.Such as Fig. 4
It is shown, it is the judgement flow diagram of linear classifier, wherein T1, T2The respectively classification thresholds of smoothness and entropy.When image
Smoothness R > T1 and when entropy e > T2, determines that surface is defective, as the smoothness R < T1 of image and entropy e < T2, decision table
Face is intact.
After the completion of judgement, finally needs will test result and damage picture sorts out archive, so that staff checks, again
Inspection.
In addition, as shown in figure 5, the present invention also provides a kind of pit rope dynamic carrying out flaw detection based on machine vision
Device, including image acquisition equipment 1, controller 2, memory 3 and display 4；Wherein, described image obtains equipment 1 for obtaining
Take the vision signal of wirerope 5；The controller 2 is for executing following procedure:
Moving target is extracted from vision signal, obtains the oscillation trajectory of each pixel on wirerope edge feature；
Displacement space sequence matrix of the wirerope along rope length directional spreding is obtained according to oscillation trajectory, to each column of the matrix
Make discrete Fourier transform, obtains the frequency response function of each point on corresponding wirerope, identified inherently according to the peak value of frequency response function
Frequency, and the Mode Shape of wirerope is obtained according to the amplitude of the response frequency of the frequency response function of each point, finally obtain wirerope
Flexibility matrix；
BP neural network is set, sets the input parameter of BP neural network to the flexibility matrix of wirerope, output quantity is set
It is set to structural damage unit and damaged degree quantized value, every wirerope is divided into multiple units, and be arranged to each unit
Multiple Stiffness values obtain the flexibility matrix of a variety of damaged wirerope, are trained as training sample；According to steel wire to be measured
The flexibility matrix of rope obtains the damage position and degree of injury of wirerope to be measured using the BP neural network after sample training；
The damage position unit judged by BP neural network is extracted simultaneously from whole image as ROI region
Save into new damage of steel cable initial pictures；
According to Retinex principle, high-pass filtering is first carried out to damage of steel cable initial pictures, then is carried out based on imadjust
Greyscale transformation；Then the smoothness R and entropy e of the image after extraction process are made using linear classifier based on textural characteristics
Surface defect differentiate；
The memory 3 is used to store the wirerope picture of damage；
The display equipment 4 is used to show the wirerope picture of damage and marks damage position and degree of injury.
The present invention can carry out video photography to the state of wirerope in real time during normal production, and to wirerope
Background segment is carried out, this image processing method is capable of providing most complete characteristic and extracts moving target, and position is accurate,
Speed is fast, and can solve background color brightness/gray scale variation bring target when single-frame images seeks verge searching target and ask
Take failure.Using flexibility matrix non-destructive tests in such a way that image detection combines, utilize wirerope stress caused by damage
Change influence to flexibility, solves simple greasy dirt adhere to using image detecting technique by wirerope surface and other are industrial
Impurity greasy dirt formed spot fall in the rope strand image range of extraction, caused by image detection make mistakes problem.It does not need simultaneously
Complicated equipment, it is easy to operate, it is at low cost.
Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention., rather than its limitations；To the greatest extent
Pipe present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: its according to
So be possible to modify the technical solutions described in the foregoing embodiments, or to some or all of the technical features into
Row equivalent replacement；And these are modified or replaceed, various embodiments of the present invention technology that it does not separate the essence of the corresponding technical solution
The range of scheme.
Claims (8)
1. a kind of pit rope dynamic method for detection fault detection based on machine vision, which comprises the following steps:
Step S1: setting up video camera in front of wirerope, and actual range representated by calibrating camera unit pixel obtains distance
Calibrating parameters move wirerope at the uniform velocity, record the wirerope in dynamic motion by video camera sequence photography and obtain video letter
Number；
Step S2: handling vision signal, extracts moving target and obtains the vibration of each pixel on wirerope edge feature
Track；
Step S3: according to the oscillation trajectory of pixel each on wirerope edge feature, wirerope is obtained along rope length directional spreding
Displacement space sequence matrix makees discrete Fourier transform to each column of the matrix, obtains the frequency response letter of each point on corresponding wirerope
Number, identifies intrinsic frequency according to the peak value of frequency response function, and obtain according to the amplitude of the response frequency of the frequency response function of each point
The Mode Shape of wirerope finally obtains the flexibility matrix of wirerope to be measured；
Step S4: setting BP neural network sets the input parameter of BP neural network to the flexibility matrix of wirerope, output quantity
It is set as structural damage unit and damaged degree quantized value, every wirerope is divided into multiple units, and set to each unit
Multiple Stiffness values are set, the flexibility matrix of a variety of damaged wirerope is obtained, is trained as training sample；According to steel to be measured
The flexibility matrix of cord obtains the damage position and degree of injury of wirerope to be measured using the BP neural network after sample training；
Step S5: the damage position unit judged by BP neural network is extracted from whole image as ROI region
Come and saves into new damage of steel cable initial pictures；
Step S6: according to Retinex principle, high-pass filtering is first carried out to damage of steel cable initial pictures, then be based on
The greyscale transformation of imadjust；Then the smoothness R and entropy e of the image after extraction process make and being based on using linear classifier
The surface defect of textural characteristics differentiates.
2. a kind of pit rope dynamic method for detection fault detection based on machine vision according to claim 1, feature
It is, in the step S2, the step of processing vision signal are as follows:
Step S201: background model the foundation of background model: is established using the method that multiple image is averaged；
Step S202: the detection of region of variation: two field pictures continuous in video sequence are subjected to difference processing, determine background
The region in region and motion change；
Step S203: to current frame image processing, difference only moving object detection: is done to image in region of variation and background image
To detect the object of movement；
Step S204: edge grey scale change curve object edge identification: is fitted by using 5 order polynomials to extract sub-pix essence
The marginal point of degree；To a series of images carry out processing obtain wirerope edge feature on each pixel certain time period vibration
Track.
3. a kind of pit rope dynamic method for detection fault detection based on machine vision according to claim 1, feature
It is, in the step S4, when the training sample of sample training BP neural network is set, every wirerope is divided into 8 lists
Member, according to the difference of fracture of wire situation, it is assumed that its rigidity declines 25%, 40% respectively, obtains the damaged condition of 16 kinds of degree, calculates
The intrinsic frequency and Mode Shape of 16 kinds of damaged wirerope vibrations, take preceding two order frequency and the vibration shape, obtain 16 class values, carry out to it
GSL transformation, as the training sample of neural network.
4. a kind of pit rope dynamic method for detection fault detection based on machine vision according to claim 1, feature
It is, in the step S4, when sample training BP neural network is arranged, selection Sigmoid function is excitation function, selects error
Signal back-propagation algorithm is training algorithm, the renewal process formula of weight are as follows:
In formula, the gradient of first item Representative errors average value, Section 2 represents instantaneous item, and Section 3 represents random noise item, t table
Show the number of iterations,Indicate the L layers of weight of the t times iteration,Indicate delta error,Indicate k-th of instruction
Practicing the output of L layers of j-th of neuron of sample, η indicates that Studying factors, μ indicate momentary constant,Expression is made an uproar at random
Sound item.
5. a kind of pit rope dynamic method for detection fault detection based on machine vision according to claim 1, feature
It is, in the step S6, the calculation formula of smoothness R are as follows: R=1-1/ (1+ σ2)；The calculation formula of entropy e are as follows:In formula, σ indicates the standard deviation of image, calculation formula are as follows:
L indicates gray level, ziIndicating each pixel gray value in image-region, m indicates average gray level,p
(zi) indicate that each pixel gray value is z in image-regioniProbability.
6. a kind of pit rope dynamic method for detection fault detection based on machine vision according to claim 1, feature
It is, in the step S6, to the progress high-pass filtering of damage of steel cable initial pictures and after the greyscale transformation based on imadjust,
Image is divided further into smaller part, the smoothness value and entropy of each part are extracted, by linear classifier to every
A part carries out classification judgement.
7. a kind of pit rope dynamic method for detection fault detection based on machine vision according to claim 1, feature
It is, in the step S6, linear classifier includes smoothness classification thresholds T1 and entropy classification thresholds T2, when the smoothness of image
R > T1 and when entropy e > T2, determines that surface is defective, as the smoothness R < T1 of image and entropy e < T2, determines that surface is intact.
8. a kind of pit rope dynamic carrying out flaw detection device based on machine vision, which is characterized in that set including image acquisition
Standby, controller, memory and display；
Described image obtains the vision signal that equipment is used to obtain wirerope；
The controller is for executing following procedure:
Moving target is extracted from vision signal, obtains the oscillation trajectory of each pixel on wirerope edge feature；
Obtain displacement space sequence matrix of the wirerope along rope length directional spreding according to oscillation trajectory, to each column of the matrix make from
Fourier transformation is dissipated, the frequency response function of each point on corresponding wirerope is obtained, intrinsic frequency is identified according to the peak value of frequency response function,
And the Mode Shape of wirerope is obtained according to the amplitude of the response frequency of the frequency response function of each point, finally obtain the flexibility of wirerope
Matrix；
BP neural network is set, sets the input parameter of BP neural network to the flexibility matrix of wirerope, output quantity is set as
Every wirerope is divided into multiple units by structural damage unit and damaged degree quantized value, and multiple to the setting of each unit
Stiffness value obtains the flexibility matrix of a variety of damaged wirerope, is trained as training sample；According to wirerope to be measured
Flexibility matrix obtains the damage position and degree of injury of wirerope to be measured using the BP neural network after sample training；
The damage position unit judged by BP neural network is extracted and saved from whole image as ROI region
The damage of steel cable initial pictures of Cheng Xin；
According to Retinex principle, high-pass filtering is first carried out to damage of steel cable initial pictures, then carry out the ash based on imadjust
Degree transformation；Then the smoothness R and entropy e of the image after extraction process make the table based on textural characteristics using linear classifier
Planar defect differentiates；
The memory is used to store the wirerope picture of damage；
The display equipment is used to show the wirerope picture of damage and marks damage position and degree of injury.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810432434.7A CN108956614B (en) | 2018-05-08 | 2018-05-08 | Mining steel wire rope dynamic flaw detection method and device based on machine vision |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810432434.7A CN108956614B (en) | 2018-05-08 | 2018-05-08 | Mining steel wire rope dynamic flaw detection method and device based on machine vision |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108956614A true CN108956614A (en) | 2018-12-07 |
| CN108956614B CN108956614B (en) | 2020-12-29 |
Family
ID=64499692
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810432434.7A Active CN108956614B (en) | 2018-05-08 | 2018-05-08 | Mining steel wire rope dynamic flaw detection method and device based on machine vision |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108956614B (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109613117A (en) * | 2018-12-19 | 2019-04-12 | 广州广电计量检测股份有限公司 | Vibrate defectoscope performance parameter preparation method and device |
| CN109859170A (en) * | 2019-01-04 | 2019-06-07 | 中国矿业大学 | A kind of steel wire rope surface damage intelligent monitoring method and system based on LBP feature |
| CN110617873A (en) * | 2019-04-26 | 2019-12-27 | 深圳市豪视智能科技有限公司 | Method for detecting vibration of cable and related product |
| CN111796710A (en) * | 2020-06-02 | 2020-10-20 | 南京信息工程大学 | Method for reproducing image contour characteristics on touch screen |
| CN111862083A (en) * | 2020-07-31 | 2020-10-30 | 中国矿业大学 | Comprehensive monitoring system and method for steel wire rope state based on vision-electromagnetic detection |
| CN111983973A (en) * | 2020-08-11 | 2020-11-24 | 武汉万邦德新科技有限公司 | Cast workpiece edge deburring processing method based on template matching |
| CN112085787A (en) * | 2020-07-20 | 2020-12-15 | 中国矿业大学 | Monocular vision-based hoisting steel wire rope spatial vibration measurement method |
| CN112307932A (en) * | 2020-10-27 | 2021-02-02 | 上海交通大学 | Parameterized full-field visual vibration modal decomposition method |
| WO2021042907A1 (en) * | 2019-04-26 | 2021-03-11 | 深圳市豪视智能科技有限公司 | Rope vibration measurement method and related apparatus |
| WO2021159569A1 (en) * | 2020-02-10 | 2021-08-19 | Dalian University Of Technology | Method for engineering structural fiexibility identification considering the impact of noise |
| CN113567460A (en) * | 2020-04-29 | 2021-10-29 | 中国石油化工股份有限公司 | Automatic detection device and detection method for broken filaments in drafting process based on image recognition |
| CN114445404A (en) * | 2022-04-06 | 2022-05-06 | 湖南大学 | Automatic structural vibration response identification method and system based on sub-pixel edge detection |
| CN114541197A (en) * | 2021-12-31 | 2022-05-27 | 上海天链轨道交通检测技术有限公司 | Dynamic measurement method, device and system for track settlement and deviation |
| CN115147450A (en) * | 2022-09-05 | 2022-10-04 | 中印云端(深圳)科技有限公司 | Moving target detection method and detection device based on motion frame difference image |
| CN112085787B (en) * | 2020-07-20 | 2024-04-23 | 中国矿业大学 | Method for measuring space vibration of hoisting steel wire rope based on monocular vision |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020065637A1 (en) * | 2000-11-30 | 2002-05-30 | Thompson John S. | Method and apparatus for simulating the measurement of a part without using a physical measurement system |
| CN101158650A (en) * | 2007-11-09 | 2008-04-09 | 无锡东望科技有限公司 | On-line detecting method of machine vision system for printed calico flaw |
| CN104280406A (en) * | 2014-09-16 | 2015-01-14 | 中国科学院广州能源研究所 | Machine vision system for detecting surface defects of copper part |
| CN107316295A (en) * | 2017-07-02 | 2017-11-03 | 苏州大学 | A kind of fabric defects detection method based on deep neural network |
| CN107618533A (en) * | 2017-09-29 | 2018-01-23 | 兰州交通大学 | A kind of machine vision detection device and method of the discrete defect of Rail Surface |
| CN107764839A (en) * | 2017-10-12 | 2018-03-06 | 中南大学 | A kind of steel wire rope surface defect online test method and device based on machine vision |
-
2018
- 2018-05-08 CN CN201810432434.7A patent/CN108956614B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020065637A1 (en) * | 2000-11-30 | 2002-05-30 | Thompson John S. | Method and apparatus for simulating the measurement of a part without using a physical measurement system |
| CN101158650A (en) * | 2007-11-09 | 2008-04-09 | 无锡东望科技有限公司 | On-line detecting method of machine vision system for printed calico flaw |
| CN104280406A (en) * | 2014-09-16 | 2015-01-14 | 中国科学院广州能源研究所 | Machine vision system for detecting surface defects of copper part |
| CN107316295A (en) * | 2017-07-02 | 2017-11-03 | 苏州大学 | A kind of fabric defects detection method based on deep neural network |
| CN107618533A (en) * | 2017-09-29 | 2018-01-23 | 兰州交通大学 | A kind of machine vision detection device and method of the discrete defect of Rail Surface |
| CN107764839A (en) * | 2017-10-12 | 2018-03-06 | 中南大学 | A kind of steel wire rope surface defect online test method and device based on machine vision |
Non-Patent Citations (2)
| Title |
|---|
| 石晟 等: "基于机器视觉的矿用钢丝绳无损监测系统研究", 《煤矿机电》 * |
| 谢飞 等: "一种钢丝绳芯输送带早期故障检测方法", 《工矿自动化》 * |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109613117A (en) * | 2018-12-19 | 2019-04-12 | 广州广电计量检测股份有限公司 | Vibrate defectoscope performance parameter preparation method and device |
| CN109613117B (en) * | 2018-12-19 | 2021-11-09 | 广州广电计量检测股份有限公司 | Method and device for obtaining performance parameters of vibration flaw detector |
| CN109859170A (en) * | 2019-01-04 | 2019-06-07 | 中国矿业大学 | A kind of steel wire rope surface damage intelligent monitoring method and system based on LBP feature |
| CN109859170B (en) * | 2019-01-04 | 2023-04-18 | 中国矿业大学 | LBP (local binary pattern) feature-based intelligent monitoring method and system for surface damage of steel wire rope |
| CN110617873A (en) * | 2019-04-26 | 2019-12-27 | 深圳市豪视智能科技有限公司 | Method for detecting vibration of cable and related product |
| WO2021042907A1 (en) * | 2019-04-26 | 2021-03-11 | 深圳市豪视智能科技有限公司 | Rope vibration measurement method and related apparatus |
| WO2021159569A1 (en) * | 2020-02-10 | 2021-08-19 | Dalian University Of Technology | Method for engineering structural fiexibility identification considering the impact of noise |
| CN113567460A (en) * | 2020-04-29 | 2021-10-29 | 中国石油化工股份有限公司 | Automatic detection device and detection method for broken filaments in drafting process based on image recognition |
| CN111796710A (en) * | 2020-06-02 | 2020-10-20 | 南京信息工程大学 | Method for reproducing image contour characteristics on touch screen |
| CN111796710B (en) * | 2020-06-02 | 2023-05-23 | 南京信息工程大学 | Method for reproducing image contour features on touch screen |
| CN112085787A (en) * | 2020-07-20 | 2020-12-15 | 中国矿业大学 | Monocular vision-based hoisting steel wire rope spatial vibration measurement method |
| CN112085787B (en) * | 2020-07-20 | 2024-04-23 | 中国矿业大学 | Method for measuring space vibration of hoisting steel wire rope based on monocular vision |
| CN111862083B (en) * | 2020-07-31 | 2023-09-19 | 中国矿业大学 | Visual-electromagnetic detection-based steel wire rope state comprehensive monitoring system and method |
| CN111862083A (en) * | 2020-07-31 | 2020-10-30 | 中国矿业大学 | Comprehensive monitoring system and method for steel wire rope state based on vision-electromagnetic detection |
| CN111983973A (en) * | 2020-08-11 | 2020-11-24 | 武汉万邦德新科技有限公司 | Cast workpiece edge deburring processing method based on template matching |
| CN111983973B (en) * | 2020-08-11 | 2021-08-06 | 重庆见芒信息技术咨询服务有限公司 | Cast workpiece edge deburring processing method based on template matching |
| CN112307932B (en) * | 2020-10-27 | 2023-02-17 | 上海交通大学 | Parameterized full-field visual vibration modal decomposition method |
| CN112307932A (en) * | 2020-10-27 | 2021-02-02 | 上海交通大学 | Parameterized full-field visual vibration modal decomposition method |
| CN114541197A (en) * | 2021-12-31 | 2022-05-27 | 上海天链轨道交通检测技术有限公司 | Dynamic measurement method, device and system for track settlement and deviation |
| CN114541197B (en) * | 2021-12-31 | 2024-04-05 | 上海天链轨道交通检测技术有限公司 | Dynamic measurement method, device and system for track settlement and offset |
| CN114445404A (en) * | 2022-04-06 | 2022-05-06 | 湖南大学 | Automatic structural vibration response identification method and system based on sub-pixel edge detection |
| CN115147450A (en) * | 2022-09-05 | 2022-10-04 | 中印云端(深圳)科技有限公司 | Moving target detection method and detection device based on motion frame difference image |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108956614B (en) | 2020-12-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108956614A (en) | A kind of pit rope dynamic method for detection fault detection and device based on machine vision | |
| CN109118479B (en) | Capsule network-based insulator defect identification and positioning device and method | |
| CN108984893B (en) | Gradient lifting method-based trend prediction method | |
| CN108242054A (en) | A kind of steel plate defect detection method, device, equipment and server | |
| KR102026449B1 (en) | Simulation Data Preprocessing Technique for Development of Damage Detecting Method for Bridges Based on Convolutional Neural Network | |
| CN110992349A (en) | Underground pipeline abnormity automatic positioning and identification method based on deep learning | |
| CN109472822A (en) | Dimension of object measurement method based on depth image processing | |
| CN104458755A (en) | Multi-type material surface defect detection method based on machine vision | |
| CN109389599A (en) | A kind of defect inspection method and device based on deep learning | |
| CN103994724A (en) | Method for monitoring two-dimensional displacement and strain of structure based on digital image processing technology | |
| CN107292879B (en) | A kind of sheet metal surface method for detecting abnormality based on image analysis | |
| CN109444169A (en) | A kind of bearing defect detection method and system | |
| CN108921812B (en) | Intelligent evaluation method for fatigue state of breaker spring based on image recognition | |
| CN113409314A (en) | Unmanned aerial vehicle visual detection and evaluation method and system for corrosion of high-altitude steel structure | |
| CN116758491B (en) | Printing monitoring image analysis method and system applied to 3D printing | |
| CN111079822A (en) | Method for identifying dislocation fault image of middle rubber and upper and lower plates of axle box rubber pad | |
| CN106651906A (en) | Test method for motion characteristic of high-voltage breaker based on improved Gaussian mixture model | |
| Savino et al. | Automated classification of civil structure defects based on convolutional neural network | |
| Widodo et al. | Confirmation of thermal images and vibration signals for intelligent machine fault diagnostics | |
| CN114627383A (en) | Small sample defect detection method based on metric learning | |
| CN113902710A (en) | Method and system for detecting surface defects of industrial parts based on anomaly detection algorithm | |
| CN115619708A (en) | Method for judging fault based on image recognition of oil quality change of main shaft bearing lubricating oil | |
| Ning et al. | Research on surface defect detection algorithm of strip steel based on improved YOLOV3 | |
| KR102189951B1 (en) | System and method for inspection of ship painting condition using image analysis | |
| CN115070780A (en) | Industrial robot grabbing method and device based on digital twinning and storage medium |
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 |