CN110009011A - A kind of solder joint recognition methods based on image procossing - Google Patents
A kind of solder joint recognition methods based on image procossing Download PDFInfo
- Publication number
- CN110009011A CN110009011A CN201910214056.XA CN201910214056A CN110009011A CN 110009011 A CN110009011 A CN 110009011A CN 201910214056 A CN201910214056 A CN 201910214056A CN 110009011 A CN110009011 A CN 110009011A
- Authority
- CN
- China
- Prior art keywords
- solder joint
- image
- svm
- recognition methods
- disaggregated model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 229910000679 solder Inorganic materials 0.000 title claims abstract description 45
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000003466 welding Methods 0.000 claims abstract description 26
- 238000000605 extraction Methods 0.000 claims abstract description 18
- 238000012549 training Methods 0.000 claims abstract description 12
- 210000000349 chromosome Anatomy 0.000 claims description 18
- 238000012545 processing Methods 0.000 claims description 13
- 230000035772 mutation Effects 0.000 claims description 10
- 230000009466 transformation Effects 0.000 claims description 8
- 230000006978 adaptation Effects 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000001914 filtration Methods 0.000 claims description 5
- 239000000284 extract Substances 0.000 claims description 4
- 238000003707 image sharpening Methods 0.000 claims description 4
- 230000003595 spectral effect Effects 0.000 claims description 4
- 238000009825 accumulation Methods 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 2
- 230000006872 improvement Effects 0.000 claims description 2
- 238000004519 manufacturing process Methods 0.000 abstract description 15
- 230000004069 differentiation Effects 0.000 abstract description 2
- 230000000694 effects Effects 0.000 description 7
- 230000002068 genetic effect Effects 0.000 description 6
- 230000009977 dual effect Effects 0.000 description 4
- 238000003708 edge detection Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000005457 optimization Methods 0.000 description 4
- 238000001514 detection method Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 239000003086 colorant Substances 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000033001 locomotion Effects 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 230000001629 suppression Effects 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 239000011248 coating agent Substances 0.000 description 1
- 238000000576 coating method Methods 0.000 description 1
- 238000004040 coloring Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 235000013305 food Nutrition 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 108090000623 proteins and genes Proteins 0.000 description 1
- 238000004080 punching Methods 0.000 description 1
- 238000004904 shortening Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- 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
Abstract
The invention belongs to manufacturing technology field, in particular to a kind of solder joint recognition methods based on image procossing.A kind of solder joint recognition methods based on image procossing, comprising: original image is pre-processed；To the solder joint in pretreated image and circular hole is interfered to carry out position extraction；Butt welding point and interference circular hole carry out texture feature extraction；GA-SVM disaggregated model is constructed, GA-SVM disaggregated model is trained, the GA-SVM disaggregated model after being trained；Based on the GA-SVM disaggregated model after training, butt welding point is identified.Energy intelligent recognition bond pad locations of the present invention and differentiation solder joint and interference information solve the problems, such as that traditional welding robot needs teaching repeatedly, working efficiency lowly and welding quality is not high.
Description
Technical field
The invention belongs to manufacturing technology field, in particular to a kind of solder joint recognition methods based on image procossing.
Background technique
With the continuous improvement of China's macroeconomic level, people start to pursue life while meeting basic problem of food and clothing
The raising of bioplasm amount and life hierarchy, the automobile industry in China are also stepped into the stage of fast development therewith, bring market and
Economic benefit is made that tremendous contribution to the economic development in China.
As the scale of China's automobile market constantly expands, requirement of the consumer to car mass is higher and higher, therefore must
Automation, flexibility and the intelligence degree of automobile production process must be improved.In general, punching press, welding, coating and general assembly be
Essential technique during automobile production, wherein the revamping workload of welding production line is very huge, and enterprise changes in each remodel
Investment substantial contribution is required when making, this is very unfavorable to the operation of enterprise.In order to meet the requirement of automobile product update,
The flexibility degree of automotive welding production line must just be improved.Welding robot possess flexible movements degree is high, operation stability is strong,
The advantages that activity duration is long, flexibility degree is high, can Improving The Quality of Products, shortening production cycle, raising production efficiency and production
Line flexibility ratio plays a crucial role during automotive welding production line flexibility.
Currently, welding robot is applied in welding production line in large quantities, exhausted when carrying out path of welding planning
It is most of that teaching repeatedly is just all had to pass through to keep the motion profile of robot more accurate using the method for manual teaching
And debugging process, it needs to repeat above procedure again when the object of welding changes, a large amount of manpower and time will be expended, greatly
Production efficiency is reduced greatly.Also, in actual production process, size, the solder joint position of all welding workpieces are not ensured that
It sets or weld width strict conformance, teaching robot can not discover these differences out, path still good according to prior teaching
Operation is carried out, the decline of welding quality is necessarily caused.
Summary of the invention
The disadvantage low for welding robot production efficiency, welding quality is not high, the present invention provide a kind of based on image
The solder joint recognition methods of processing.Energy intelligent recognition bond pad locations of the present invention and differentiation solder joint and interference information, solve tradition
The problem that welding robot needs teaching repeatedly, working efficiency is low and welding quality is not high.
The present invention adopts the following technical scheme that realization:
A kind of solder joint recognition methods based on image procossing, comprising:
S1, original image is pre-processed；
S2, in pretreated image solder joint and interference circular hole carry out position extraction；
S3, butt welding point and interference circular hole carry out texture feature extraction；
S4, building GA-SVM disaggregated model, are trained GA-SVM disaggregated model, the GA-SVM classification after being trained
Model；
S5, based on the GA-SVM disaggregated model after training, butt welding point is identified.
Further: image preprocessing includes image gray processing, image denoising, image sharpening and Edge extraction.
Preferably, image denoising is carried out using median filtering.Image sharpening is carried out using laplacian spectral radius method.Using
Canny operator carries out Edge extraction.
Further, in step S2, solder joint and interference circular hole position are carried out using improved random Hough transformation algorithm
It extracts, comprising the following steps:
(1) three pictures being not arranged on the same straight line are randomly selected from all edge point set V in image to be detected
Vegetarian refreshments is used to determine the parameter of candidate circle；
(2) pixel other than set V is traversed, judges the pixel whether on candidate's circle；Record falls in candidate simultaneously
The number of pixel on circle；
(3) after the parameter for determining candidate circle, evidence accumulation is carried out, determines center location and radius size；
(4) according to determining central coordinate of circle position and radius size interception solder joint and circular hole image, complete solder joint and
The position of circular hole is interfered to extract.
Further, in step S3, using textural characteristics of the extraction based on gray level co-occurrence matrixes, comprising:
S31, pretreated image is divided into 16 gray levels, and the distance between capture element is 1；
S32, the gray level co-occurrence matrixes for constructing 0 °, 45 °, 90 °, 135 ° four direction；
S33, gray level co-occurrence matrixes element is constructed into energy, the moment of inertia, four entropy, correlation features using knowledge of statistics
Amount, and calculate its value；
S34, the feature vector of the mean value and standard deviation of each characteristic quantity as textural characteristics, the dimension of feature vector are calculated
It is 8.
Further, step S4 includes:
S41, building svm classifier model, determine the value range and code length of penalty factor and nuclear parameter；
S42, binary coding is carried out to svm classifier model parameter, and N number of chromosome is randomly generated and generates initial population；
S43, each individual in initial population is obtained using the classification accuracy rate of svm classifier model calculating training sample
To the ideal adaptation angle value of Single chromosome, so progress n times, the individual adaptation degree of each chromosome in initial population is obtained
Value；
S44, selection, intersection and mutation operation are executed, forms next-generation population, and the number of iterations of population is added 1；
S45, the judgement for being iterated termination condition, algorithm stops if the number of iterations reaches the specified upper limit, by population
In optimum individual decode optimized parameter as svm classifier model；Otherwise, step S44 is gone to continue to execute；
S46, the best parameter group obtained using decoding, are inputted svm classifier model, obtain GA-SVM disaggregated model；
S47, GA-SVM disaggregated model is trained, the GA-SVM disaggregated model after being trained.
Preferably, the value range of penalty factor be c ∈ [0,100], the value range of nuclear parameter be set as g ∈ [0,
100], the binary coding length of penalty factor and nuclear parameter is 10, and binary coding range is all are as follows: 0000000000-
1111111111。
Preferably, selection operation uses roulette algorithm；Crossover operation is by the way of linear combination；Mutation operation uses
Basic bit mutation mode is realized.
The invention has the following beneficial effects:
1, the automatic identification and positioning for realizing solder joint, enable whole production line autonomous operation, without excessive people
Work intervention, reduces human cost, improves work efficiency.
2, what is proposed carries out the svm classifier model (GA-SVM) of parameter optimization based on genetic algorithm, utilizes genetic algorithm handle
The solution of challenge is modeled to the process of a biological evolution, and has the characteristics that of overall importance, randomness, concurrency, can be fast
Speed finds the global approximate optimal solution of system, effectively improves the efficiency of parameter optimization, classifier is enable quickly to find
Optimal solution, thus more accurate carry out target identification classification.
3, GA-SVM disaggregated model the problems such as solving non-linear, small sample, high dimensional pattern identification when show it is preferable
Performance, and the problems such as falling into local optimum during neural network learning can be overcome.
4, the textural characteristics based on gray level co-occurrence matrixes are extracted, using knowledge of statistics building one on the basis of matrix
A little easily distinguishable characteristic quantities, such as: energy, the moment of inertia, entropy have the texture feature vector based on gray level co-occurrence matrixes more
Low dimension.
5, Hough transform algorithm is improved, improved random Hough transformation algorithm can when carrying out circle detection
To be effectively removed the influence of noise, and it still is able in the case where circle has deformation or even circular portion circular arc is lost
Obtain more satisfactory effect.
Detailed description of the invention
Fig. 1 is Canny operator edge detection effect in one embodiment of the invention；
Fig. 2 is improved random Hough transformation algorithm bond pad locations extraction effect in one embodiment of the invention；
Fig. 3 is the texture feature extraction result based on gray level co-occurrence matrixes in one embodiment of the invention, in which: (a) is
Energy；It (b) is the moment of inertia；It (c) is entropy；It (d) is correlation；
Fig. 4 is GA-SVM disaggregated model design flow diagram in one embodiment of the invention.
Specific embodiment
The present invention is described in further detail below by specific embodiment, but embodiments of the present invention are not
It is limited to this.
A kind of solder joint recognition methods based on image procossing, comprising:
S1, original image is pre-processed, including image gray processing, image denoising, image sharpening, image border mention
It takes.
By image capturing system obtain original color image, most color images all use RGB (represent it is red,
These three green, blue primary colors) color mode.The color of each pixel is determined by tri- color components of RGB in image, every kind of face
The grey scale change range of colouring component is 0 to 255, each pixel of color image has up to more than 1600 million colors variations
Range, so color image committed memory is very big, transmission and processing time are long, are not able to satisfy wanting for real-time in practical application
It asks.Image gray processing is to keep R, G of each pixel in color image, B component value equal.After image gray processing can reduce
The complexity of phase image procossing, while gray level image still is able to be well reflected the entirety of image and local feature.
The image of acquisition usually contains various noises, and the presence of these noises will affect the post-processing of image, and image is gone
The main method made an uproar is filtering.Not only there is denoising effect well using median filtering, while its fog-level is more flat than linear
Filter slide is lower.
Since the filtering processing of image frequently can lead to the blurring of image border, and the edge of image usually contains target
The characteristic information of object, therefore the edge of image must be compensated, i.e., processing is sharpened to image, to enhance image border
Information.In the present embodiment, processing is sharpened to filtered image using laplacian spectral radius method.
Laplacian spectral radius method is second-order differential algorithm, the Laplace operator of image f (x, y) is defined as:
Wherein:
Laplace operator of the f (x, y) at (x, y) can be obtained by formula (1), (2), (3) are as follows:
If being g (x, y) by the image after Edge contrast, then g (x, y) can be indicated are as follows:
Wherein: α is constant.
The characteristic at the detection of Canny operator dual-threshold voltage and connection edge can handle the marginal information of image well, this
In embodiment, edge detection is carried out using Canny operator butt welding point image, Fig. 1 is Canny operator edge detection effect picture.
By first smoothed image, then seek the method for gradient magnitude realizes edge detection to Canny operator, algorithm comprising steps of
A. Gaussian filter smoothed image is used；
If f (x, y) indicates that input picture, G (x, y) indicate Gaussian function, then:
If smoothed out image is g (x, y), then:
G (x, y)=f (x, y) * G (x, y) (7)
B. gradient magnitude and the direction of image are calculated smoothed out image using first derivative operator；
The gradient magnitude M (x, y) and deflection θ (x, y) of image are respectively as follows:
Wherein, P (x, y), Q (x, y) respectively indicate g (x, y) in first derivative values both horizontally and vertically:
C. non-maxima suppression is carried out to gradient magnitude；
After the gradient magnitude and direction for obtaining image, each pixel of image is traversed, is checked around the pixel
The point for whether having gradient magnitude bigger than it in pixel identical with its gradient direction.If it does not exist, then the ladder of the pixel
Amplitude is spent to retain；Otherwise gradient magnitude zero setting.
D. edge is detected and connected with dual-threshold voltage.
Dual threshold detection method carries out dual threshold processing to the image after non-maxima suppression, and two threshold values are respectively THWith
TL, and TH≈TL.If the gradient magnitude of pixel is higher than T in image to be processedH, then it is assumed that the pixel is real edge
Point；If being less than TLThen it is rejected；Two threshold skirt image T are respectively obtained after dual threshold is handled1(x, y) and T2(x,y)。
T1(x, y) is that the false edge that handles to obtain by high threshold, therefore include is less, but there may be gaps for profile.Dual threshold
Algorithm is in T1Edge is connected into profile in (x, y), when profile is interrupted then in T2Edge is found in (x, y), until T1
No longer there is gap in the profile in (x, y).
S2, in pretreated image solder joint and interference circular hole carry out position extraction.
The chaff interferents such as solder joint on automobile door plate and circular hole are closer, and image capture device acquired image is simultaneously
There is round solder joints and interference circular hole, can use Hough (Hough) transformation and extract solder joint and interfere the center of circular hole
And radius size.When parameter space is more than the situation of two dimension, traditional Hough transform algorithm committed memory is big, time-consuming, needs
Use each pixel on original image.In the present embodiment, traditional Hough transform algorithm is improved, it will be on original image
Each pixel determine central coordinate of circle from two-dimensional map to the three-dimensional space determined by parameter a, b, r, then in three-dimensional space
And radius, a kind of improved random Hough transformation algorithm is obtained, Fig. 2 is to be welded using improved random Hough transformation algorithm
The effect picture that point and interference circular hole position are extracted.
Solder joint is extracted using improved random Hough transformation algorithm and interferes center and the radius size of circular hole, packet
Include following steps:
(1) three pictures being not arranged on the same straight line are randomly selected from all edge point set V in image to be detected
Vegetarian refreshments is used to determine the parameter of candidate circle.
Round representation are as follows:
(x-a)2+(y-b)2=r2 (12)
Wherein: (a, b) is the central coordinate of circle of circle, and r is the radius of circle.Take vi=(xi,yi), i=1,2,3, then this 3 points it is true
Fixed central coordinate of circle and radius are as follows:
It can determine the parametric equation of candidate circle by formula (13), (14), (15).
(2) pixel other than set V is traversed, judges the pixel whether on candidate's circle；Record falls in candidate simultaneously
The number of pixel on circle.
Take other point v in V4=(x4,y4), its coordinate is brought into formula (12), radius r is acquired4If r4With the difference of r
d4Less than some threshold value δ, it may be assumed that
Then indicate point v4On candidate's circle.
(3) after the parameter for determining candidate circle, evidence accumulation is carried out, determines center location and radius size.
Take the point v in VkIf dk< δ, then count value adds 1, has traversed in V after all the points, refers to if the value counted is greater than
Determine threshold value, then it is assumed that the circle is true circle, and the corresponding marginal point of the circle is deleted in V.
(4) according to determining central coordinate of circle position and radius size interception solder joint and circular hole image, complete solder joint and
The position of circular hole is interfered to extract.
S3, butt welding point and interference circular hole carry out texture feature extraction.
Since solder joint is similar with the interference shape of circular hole, and in preprocessing process, image has carried out gray processing processing, therefore face
Color characteristic and shape feature cannot accurately distinguish the two, in the present embodiment, using line of the extraction based on gray level co-occurrence matrixes
Feature is managed, the texture feature extraction result based on gray level co-occurrence matrixes is as shown in Figure 3.
The textural characteristics based on gray level co-occurrence matrixes are extracted, comprising steps of
S31, pretreated image is divided into 16 gray levels, and the distance between capture element is 1；
S32, the gray level co-occurrence matrixes for constructing 0 °, 45 °, 90 °, 135 ° four direction；
Gray level co-occurrence matrixes description: being located at standoff distance on the direction θ is d, and gray value is respectively two pixels of i and j
The probability occurred simultaneously is denoted as P (i, j | d, θ).Gray level co-occurrence matrixes are a symmetrical matrixes, order by pixel gray scale
Series determines, for example for binary image, gray value only has 0 and 1, and matrix is second-order matrix.Gray level co-occurrence matrixes it is each
Element can be acquired by following formula:
Wherein, T (i, j | d, θ) indicates that standoff distance is d on the direction θ, and gray value is respectively a pair of of pixel of i and j
The number that point occurs；The order of N representing matrix.
S33, matrix element is configured to four characteristic quantities, respectively energy, the moment of inertia, entropy, phase using knowledge of statistics
Guan Xing, and calculate its value；
Energy, the moment of inertia, entropy, correlation definition are as follows respectively:
A. energy
The definition of energy is as follows:
B. the moment of inertia
The definition of the moment of inertia is as follows:
C. entropy
The definition of entropy is as follows:
D. correlation
The definition of correlation is as follows:
In formula (21):
Wherein, μx、μyRespectively indicate the mean value for the pixel that gray value is i and j, σx、σyRespectively indicating gray value is i and j
Pixel variance.
S34, the feature vector of the mean value and standard deviation of each characteristic quantity as textural characteristics, the dimension of feature vector are calculated
It is 8.
S4, building GA-SVM disaggregated model, are trained GA-SVM disaggregated model, the GA-SVM classification after being trained
Model.
The present invention optimizes SVM model parameter using genetic algorithm, obtains carrying out parameter optimization based on genetic algorithm
Svm classifier model (GA-SVM).It the characteristics of high efficiency having using genetic algorithm when carrying out parameter optimization, quickly looks for
To the global approximate optimal solution of system.Svm classifier model uses Gaussian radial basis function as kernel function, and uses genetic algorithm
The nuclear parameter g and penalty factor c of global optimum are searched for, using the parameter searched as the final ginseng of GA-SVM disaggregated model
Number.
In the present embodiment, GA-SVM disaggregated model is constructed, GA-SVM disaggregated model is trained, after being trained
GA-SVM disaggregated model, as shown in Figure 4, comprising:
S41, building svm classifier model, determine the value range and code length of penalty factor and nuclear parameter；
In the present embodiment, the value range of penalty factor c is set as c ∈ [0,100], the value range setting of nuclear parameter g
For g ∈ [0,100], while the binary coding length of the two variables is set as 10.The two of penalty factor c and nuclear parameter g into
Coding range processed is all are as follows: 0000000000-1111111111.
S42, binary coding is carried out to svm classifier model parameter, and N number of chromosome is randomly generated and generates initial population.
If the binary coding string of penalty factor c is A:a1a2...a10, the binary coding string of nuclear parameter g is B:
b1b2...b10, each of binary string corresponds to a gene, then chromosome AB are as follows: a1a2...a10b1b2...b10。
S43, list is obtained using the classification accuracy rate of SVM model calculating training sample to each individual in initial population
The ideal adaptation angle value f (c, g) of chromosome, so progress n times, obtain the individual adaptation degree of each chromosome in initial population
Value.
In the present embodiment, if the fitness function of chromosome are as follows: f (c, g)=e, wherein e indicates supporting vector machine model
Classification accuracy in training sample；Classification then when svm classifier model test sample under a certain parameter combination is accurate
Rate is higher, then the fitness value of chromosome corresponding to this group of parameter is bigger.
S44, selection, intersection and mutation operation are executed, forms next-generation population, and the number of iterations of population is added 1.
A. selection operation
In the present embodiment, selection operation uses roulette algorithm, each of old population chromosome aiIt is selected to enter
Follow-on probability are as follows:
In formula (26), N indicates the scale of population, fit (aj) indicate population in each chromosome fitness value.
It is next-generation that formula (23) also indicates that the higher individual of fitness is more possible to be entered by selection.
B. crossover operation
In the present embodiment, crossover operation is by the way of linear combination.For two chromosome x1And x2, they are by handing over
Become x after fork operation1 NAnd x2 N。x1 NAnd x2 NIt can be indicated with the following formula:
x1 N=β x1+(1-β)x2 (27)
x2 N=(1- β) x1+βx2 (28)
In formula (27) and formula (28), β is the random number between 0 to 1.
C. mutation operation
In the present embodiment, realize mutation operation using basic bit mutation mode, i.e., to chromosome binary coding string with
The specified a certain position of machine carries out inversion operation.For example, the binary coding string of original chromosome is " 11001001 ", variation position is the
Three, then the chromosome binary coding after making a variation is " 11101001 ".
S45, the judgement for being iterated termination condition, algorithm stops if the number of iterations reaches the specified upper limit, by population
In optimum individual decode optimized parameter as svm classifier model；Otherwise, step S44 is gone to continue to execute；
S46, the best parameter group obtained using decoding, are inputted svm classifier model, obtain GA-SVM disaggregated model；
S47, GA-SVM disaggregated model is trained, the GA-SVM disaggregated model after being trained.
In the present embodiment, 100 weld point images and 100 circular hole images is taken to be used as training sample set.
S1-S3 step, texture feature extraction, using the textural characteristics of extraction as GA-SVM are executed to each training sample image
The input of disaggregated model is simultaneously trained, the GA-SVM disaggregated model after being trained.
S5, based on the GA-SVM disaggregated model after training, butt welding point is identified.
Solder joint identification mainly distinguishes solder joint and other interference informations, this interference mainly comes from circular hole,
It needs circular hole and solder joint will be interfered to distinguish.Specifically: based on the GA-SVM disaggregated model after training, position in S2 is extracted
Obtained solder joint and interference circular hole carries out Classification and Identification, identifies solder joint.
In the present embodiment, select 45 to automobile door plate image be used as test sample, utilize after training GA-SVM classification
Model calculates test sample classification accuracy.Experiment shows that the present invention can effectively improve solder joint recognition efficiency and accuracy.
The above embodiment is a preferred embodiment of the present invention, but embodiments of the present invention are not by above-described embodiment
Limitation, other any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present invention,
It should be equivalent substitute mode, be included within the scope of the present invention.
Claims (10)
1. a kind of solder joint recognition methods based on image procossing characterized by comprising
S1, original image is pre-processed；
S2, in pretreated image solder joint and interference circular hole carry out position extraction；
S3, butt welding point and interference circular hole carry out texture feature extraction；
S4, building GA-SVM disaggregated model, are trained GA-SVM disaggregated model, the GA-SVM classification mould after being trained
Type；
S5, based on the GA-SVM disaggregated model after training, butt welding point is identified.
2. solder joint recognition methods according to claim 1, which is characterized in that step S1 includes: that image gray processing, image are gone
It makes an uproar, image sharpening and Edge extraction.
3. solder joint recognition methods according to claim 2, which is characterized in that carry out image denoising using median filtering.
4. solder joint recognition methods according to claim 2, which is characterized in that it is sharp to carry out image using laplacian spectral radius method
Change.
5. solder joint recognition methods according to claim 2, which is characterized in that carry out image border using Canny operator and mention
It takes.
6. solder joint recognition methods according to any one of claims 1-5, which is characterized in that in step S2, using improvement
Random Hough transformation algorithm carry out solder joint and interference circular hole position extract, comprising the following steps:
(1) three pixels being not arranged on the same straight line are randomly selected from all edge point set V in image to be detected
For determining the parameter of candidate circle；
(2) pixel other than set V is traversed, judges the pixel whether on candidate's circle；Record is fallen on candidate circle simultaneously
Pixel number；
(3) after the parameter for determining candidate circle, evidence accumulation is carried out, determines center location and radius size；
(4) according to the image of determining central coordinate of circle position and radius size interception solder joint and circular hole, solder joint and interference are completed
It extracts the position of circular hole.
7. solder joint recognition methods according to claim 6, which is characterized in that total based on gray scale using extracting in step S3
The textural characteristics of raw matrix, comprising:
S31, pretreated image is divided into 16 gray levels, and the distance between capture element is 1；
S32, the gray level co-occurrence matrixes for constructing 0 °, 45 °, 90 °, 135 ° four direction；
S33, gray level co-occurrence matrixes element is constructed into energy, the moment of inertia, four entropy, correlation characteristic quantities using knowledge of statistics,
And calculate its value；
S34, the feature vector of the mean value and standard deviation of each characteristic quantity as textural characteristics is calculated, the dimension of feature vector is 8.
8. solder joint recognition methods described in any one of -5,7 according to claim 1, which is characterized in that step S4 includes:
S41, building svm classifier model, determine the value range and code length of penalty factor and nuclear parameter；
S42, binary coding is carried out to svm classifier model parameter, and N number of chromosome is randomly generated and generates initial population；
S43, list is obtained using the classification accuracy rate of svm classifier model calculating training sample to each individual in initial population
The ideal adaptation angle value of chromosome, so progress n times, obtain the ideal adaptation angle value of each chromosome in initial population；
S44, selection, intersection and mutation operation are executed, forms next-generation population, and the number of iterations of population is added 1；
S45, the judgement for being iterated termination condition, algorithm stops if the number of iterations reaches the specified upper limit, will be in population
Optimum individual decodes the optimized parameter as svm classifier model；Otherwise, step S44 is gone to continue to execute；
S46, the best parameter group obtained using decoding, are inputted svm classifier model, obtain GA-SVM disaggregated model；
S47, GA-SVM disaggregated model is trained, the GA-SVM disaggregated model after being trained.
9. solder joint recognition methods according to claim 8, which is characterized in that the value range of penalty factor be c ∈ [0,
100], the value range of nuclear parameter is set as g ∈ [0,100], and the binary coding length of penalty factor and nuclear parameter is 10, two
Scale coding range is all are as follows: 0000000000-1111111111.
10. solder joint recognition methods according to claim 8 or claim 9, which is characterized in that in step S44, selection operation is using wheel
Disk gambles algorithm；Crossover operation is by the way of linear combination；Mutation operation is realized using basic bit mutation mode.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910214056.XA CN110009011A (en) | 2019-03-20 | 2019-03-20 | A kind of solder joint recognition methods based on image procossing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910214056.XA CN110009011A (en) | 2019-03-20 | 2019-03-20 | A kind of solder joint recognition methods based on image procossing |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110009011A true CN110009011A (en) | 2019-07-12 |
Family
ID=67167543
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910214056.XA Pending CN110009011A (en) | 2019-03-20 | 2019-03-20 | A kind of solder joint recognition methods based on image procossing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110009011A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110582159A (en) * | 2019-08-30 | 2019-12-17 | 武汉科技大学 | FPC bonding pad surface defect processing system and method |
| CN111160479A (en) * | 2019-12-31 | 2020-05-15 | 华南理工大学 | Welding spot identification method |
| CN111914473A (en) * | 2020-06-24 | 2020-11-10 | 浙江吉利汽车研究院有限公司 | Welding parameter determination method and device for resistance spot welding, electronic equipment and storage medium |
| CN113505657A (en) * | 2021-06-18 | 2021-10-15 | 东风汽车集团股份有限公司 | Welding spot quality detection method and device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103810459A (en) * | 2012-11-07 | 2014-05-21 | 上海航天设备制造总厂 | Image recognition device and solar array welding system by using same |
| CN104942496A (en) * | 2015-06-29 | 2015-09-30 | 湖南大学 | Car body-in-white welding spot positioning method and device based on robot visual servo |
| CN105938563A (en) * | 2016-04-14 | 2016-09-14 | 北京工业大学 | Weld surface defect identification method based on image texture |
| CN106228565A (en) * | 2016-08-02 | 2016-12-14 | 电子科技大学 | A kind of oil pipeline weld defect detection method based on radioscopic image |
| CN108346137A (en) * | 2017-01-22 | 2018-07-31 | 上海金艺检测技术有限公司 | Defect inspection method for industrial x-ray weld image |
-
2019
- 2019-03-20 CN CN201910214056.XA patent/CN110009011A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103810459A (en) * | 2012-11-07 | 2014-05-21 | 上海航天设备制造总厂 | Image recognition device and solar array welding system by using same |
| CN104942496A (en) * | 2015-06-29 | 2015-09-30 | 湖南大学 | Car body-in-white welding spot positioning method and device based on robot visual servo |
| CN105938563A (en) * | 2016-04-14 | 2016-09-14 | 北京工业大学 | Weld surface defect identification method based on image texture |
| CN106228565A (en) * | 2016-08-02 | 2016-12-14 | 电子科技大学 | A kind of oil pipeline weld defect detection method based on radioscopic image |
| CN108346137A (en) * | 2017-01-22 | 2018-07-31 | 上海金艺检测技术有限公司 | Defect inspection method for industrial x-ray weld image |
Non-Patent Citations (1)
| Title |
|---|
| 刘金: "基于机器视觉的汽车门板焊点识别算法研究", 《中国优秀硕士学位论文全文库》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110582159A (en) * | 2019-08-30 | 2019-12-17 | 武汉科技大学 | FPC bonding pad surface defect processing system and method |
| CN111160479A (en) * | 2019-12-31 | 2020-05-15 | 华南理工大学 | Welding spot identification method |
| CN111914473A (en) * | 2020-06-24 | 2020-11-10 | 浙江吉利汽车研究院有限公司 | Welding parameter determination method and device for resistance spot welding, electronic equipment and storage medium |
| CN113505657A (en) * | 2021-06-18 | 2021-10-15 | 东风汽车集团股份有限公司 | Welding spot quality detection method and device |
| CN113505657B (en) * | 2021-06-18 | 2022-05-03 | 东风汽车集团股份有限公司 | Welding spot quality detection method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110009011A (en) | A kind of solder joint recognition methods based on image procossing | |
| Bergmann et al. | Improving unsupervised defect segmentation by applying structural similarity to autoencoders | |
| CN108416266B (en) | Method for rapidly identifying video behaviors by extracting moving object through optical flow | |
| CN104680519B (en) | Seven-piece puzzle recognition methods based on profile and color | |
| CN107967695B (en) | A kind of moving target detecting method based on depth light stream and morphological method | |
| CN109871938A (en) | A kind of components coding detection method based on convolutional neural networks | |
| CN110148162A (en) | A kind of heterologous image matching method based on composition operators | |
| CN106803257B (en) | Method for segmenting disease spots in crop disease leaf image | |
| CN105930854A (en) | Manipulator visual system | |
| CN105320917B (en) | A kind of pedestrian detection and tracking based on head-shoulder contour and BP neural network | |
| CN104063711B (en) | A kind of corridor end point fast algorithm of detecting based on K means methods | |
| Funk et al. | Beyond planar symmetry: Modeling human perception of reflection and rotation symmetries in the wild | |
| CN103632137B (en) | A kind of human eye iris segmentation method | |
| CN108985337A (en) | A kind of product surface scratch detection method based on picture depth study | |
| CN115797354B (en) | Method for detecting appearance defects of laser welding seam | |
| CN109598200B (en) | Intelligent image identification system and method for molten iron tank number | |
| CN116091455A (en) | Steel mesh surface defect judging method based on machine vision | |
| CN111199556A (en) | Indoor pedestrian detection and tracking method based on camera | |
| CN112258525B (en) | Image abundance statistics and population identification algorithm based on bird high-frame frequency sequence | |
| CN108647689A (en) | Magic square restored method and its device based on GoogLeNet neural networks | |
| CN116071560A (en) | Fruit identification method based on convolutional neural network | |
| CN103279944A (en) | Image division method based on biogeography optimization | |
| CN104915951A (en) | Stippled DPM two-dimensional code area positioning method | |
| CN113673621A (en) | Quasi-circular target detection method based on convolutional neural network and MAML algorithm | |
| CN113421223A (en) | Industrial product surface defect detection method based on deep learning and Gaussian mixture |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190712 |