CN104636760B - A kind of localization method of weld seam - Google Patents
A kind of localization method of weld seam Download PDFInfo
- Publication number
- CN104636760B CN104636760B CN201510106889.6A CN201510106889A CN104636760B CN 104636760 B CN104636760 B CN 104636760B CN 201510106889 A CN201510106889 A CN 201510106889A CN 104636760 B CN104636760 B CN 104636760B
- Authority
- CN
- China
- Prior art keywords
- image
- weld
- mtd
- weld seam
- msub
- 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.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30152—Solder
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Quality & Reliability (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of localization method of weld seam, including step:Step 1: weld image collection to be positioned;Step 2: weld image to be positioned is pre-processed, detailed process is:Step 201, weld image piecemeal to be positioned, step 202, weld image dimensionality reduction to be positioned, step 203, weld image to be positioned classification;Step 3: weld and HAZ, detailed process is:Step 301, the two-value classification chart picture for obtaining weld image to be positioned, step 302, obtain two-value classification chart picture horizontal direction integral projection curve, step 303, the integral projection curve tack weld according to the horizontal direction of two-value classification chart picture.The inventive method step is simple, and it is convenient to realize, strong antijamming capability, functional reliability is high, and weld and HAZ cost is low, installs and uses more convenient, and practical, using effect is good, is easy to promote the use of.
Description
Technical field
The invention belongs to technical field of image processing, and in particular to a kind of localization method of weld seam.
Background technology
In the production procedure of steel-pipe welding, have and weld preceding welding bead tracing and positioning, the cleaning of postwelding solder flux welding slag, and subsequently
The technological process such as welding quality inspection.When defective, it is necessary to weld defect position is sprayed and marked, by multiple after repairing
Inspection determines whether weld seam is qualified.Manually adjustment mechanical driving device searches weld seam mark, and positioning precision is not high, and operates numerous
It is trivial, thus recheck the stage automatic weld and HAZ just have great importance.
At present, at home and abroad there is extensive research using visible sensation method automatic identification tack weld:One type be according to
The weld image obtained according to X-ray, and using SVMs, fuzzy neural network to Welding Line Flaw Detection;Also one
Class is to obtain weld image according to high-speed motion picture camera CCD, and utilizes single stripe laser, image matching technology, structure light vision three
Point, structure light and uniform light multiple features technology carry out tack weld;Furthermore also have and carry out three-dimensional localization using ultrasonic sensing technology.
However, after being cleared up by solder flux welding slag, welded seam area tends to be smooth, and structure light just can not be fine with single stripe laser
Utilization groove obtain identification information, and metallic luster is presented in cleaning area, and poor weld scribbles mark, and also there are different journeys at edge
That spends is fuzzy, in the case of having light source, and local reflective, metal material corrosion factor itself can cause serious interference to identification,
So that being disturbed during images match larger, it is impossible to adapt to the weld seam recognition positioning in reinspection stage, X-ray technology and ultrasonic technique very well
The relatively high speed video camera cost of equipment again it is too high.
The content of the invention
In view of the above-mentioned deficiencies in the prior art, the technical problem to be solved by the present invention is that providing a kind of determining for weld seam
Position method, its method and step is simple, and it is convenient to realize, strong antijamming capability, functional reliability is high, and weld and HAZ cost is low, installs
Using more convenient, practical, using effect is good, is easy to promote the use of.
In order to solve the above technical problems, the technical solution adopted by the present invention is:A kind of localization method of weld seam, its feature exists
Comprise the following steps in this method:
Step 1: weld image collection to be positioned:Shot using monochromatic light exposure position while welding, and using industrial camera
Weld image to be positioned and the weld image to be positioned photographed is transferred to image processor;
Step 2: weld image to be positioned is pre-processed, detailed process is:
Step 201, weld image piecemeal to be positioned:Image processor receives the weld image to be positioned and according in the ranks
The weld image to be positioned is divided into M × N number of weld seam subgraph Y to be positioned every width d and row interval width H1、Y2、…、
YM×N, each weld seam subgraph to be positioned constitutes by m × n pixel, wherein, M for weld seam subgraph to be positioned row
Number, N is the columns of weld seam subgraph to be positioned, and d, H, M, N, m and n are natural number, and d and H unit are pixel;
Step 202, weld image dimensionality reduction to be positioned:Image processor calls the drop trained in advance using PCA
Tie up matrix W and according to formula Yf'=WYfDimension-reduction treatment is carried out to M × N number of weld seam subgraph to be positioned, by M × N number of weldering to be positioned
Crack image Y1、Y2、…、YM×NBe converted to the M after dimension-reduction treatment × N number of weld seam subgraph characteristic vector Y to be positioned1′、
Y2′、…、YM×N', wherein, YfFor f-th of weld seam subgraph to be positioned andYf' be dimension-reduction treatment after f-th
Weld seam subgraph characteristic vector to be positioned, f value is 1~M × N natural number;
Step 203, weld image to be positioned classification:Image processor will carry out M × N number of weldering to be positioned after dimension-reduction treatment
In three layers of BP neural network grader that crack image feature vector input training in advance is obtained, M × N number of three layers of BP nerves are obtained
The output of network classifier, and using the output of M × N number of three layers of BP neural network grader as corresponding respectively to after dimension-reduction treatment
M × N number of weld seam subgraph characteristic vector to be positioned classification results;Wherein, three layers of BP nerve nets are output as being used for
Represent that the first kind of the weld seam sub-picture pack to be positioned containing weld seam is identified and for representing that weld seam subgraph to be positioned does not include weld seam
Equations of The Second Kind mark;
Step 3: weld and HAZ, detailed process is:
Step 301, the two-value classification chart picture for obtaining weld image to be positioned:Image processor is by M × N after dimension-reduction treatment
The classification results of individual weld seam subgraph characteristic vector to be positioned are combined, and constitute two corresponding to the weld image to be positioned
It is worth classification chart as O (r, c), the two-value classification chart is M × N number of pixel, each pixel correspondence one as O (r, c) size
Subgraph;Wherein, r be the two-value classification chart as O (r, c) row coordinate, c is the two-value classification chart as O (r, c) row are sat
Mark;
The integral projection curve of step 302, the horizontal direction of acquisition two-value classification chart picture:Image processor is to the two-value
Classification chart is integrated summation as the pixel of O (r, c) every a line, obtains the two-value classification chart as O (r, c) level side
To integral projection curve
Step 303, the integral projection curve tack weld according to the horizontal direction of two-value classification chart picture:If region [r1,r2]
The integral projection curve of horizontal direction for the two-value classification chart as O (r, c)On row coordinate points ri's
ValueExceed threshold value T maximum region, image processor calls position while welding judge module judgment formula
r2-r1>Whether D sets up, as formula r2-r1>When D is set up, it is determined as in the weld image to be positionedRegion is weld seam position region;Otherwise, as formula r2-r1>When D is invalid, it is determined as described
Weld seam is not included in weld image to be positioned;Wherein, D is the minimum developed width of weld seam to be positioned, and D unit is pixel, ri
∈[r1,r2], threshold value T unit is pixel.
A kind of localization method of above-mentioned weld seam, it is characterised in that:Image processor uses principal component analysis in step 202
Method training dimensionality reduction matrix W detailed process be:
Step 2021, according to the method for step one gather one group of weld image to be positioned as training weld image, and
Piecemeal is carried out to group weld image to be positioned according to the method for step 201;
Step 2022, q image block X of selection1、X2、…、Xq, each described image block constitutes by m × n pixel;Its
In, i-th of image block isP is the number of pixels in image block, and p and q are natural number, and i value is 1~q
Natural number;
Step 2023, the matrix that q image merged block is obtained to p × q dimensions
Step 2024, to matrix X carry out singular value decomposition obtain X=U ∑s VT, wherein, U=[u1,u2,…,up]∈Rp×p,
Rp×pThe real number matrix tieed up for p × p, V=[v1,v2,…,vq]∈Rq×q, Rq×qThe real number matrix tieed up for q × q, Σ=diag
(σ1,σ2,…,σk), k=min (p, q), uxFor matrix X x-th of left singular vector, x value is 1~p natural number, vyFor
Matrix X y-th of right singular vector, y value is 1~q natural number, σjFor matrix X j-th of singular value, j value is
1~k natural number, σ1≥σ2≥…≥σk≥0;
Step 2025, the preceding h feature constitutive characteristic vector for taking U, obtain orthogonal dimensionality reduction matrix W=[u1,u2,…,uh
]T, wherein,
A kind of localization method of above-mentioned weld seam, it is characterised in that:Image processor training in advance obtains three in step 203
Layer BP neural network grader detailed process be:
Step 2031, the s image block X of selection from the training weld image after step 2021 processing1、X2、…、
Xs, each described image block constitutes by m × n pixel;Wherein, the λ image block isP is in image block
Number of pixels, p and s are natural number, and λ value is 1~s natural number;
Step 2032, according to formula Xλ'=WXλBy s image block X1、X2、…、XsBe converted to s image block characteristics vector
X1′、X2′、…、Xs', wherein, Xλ' it is the λ image block characteristics vector;
Step 2033, with s image block characteristics vector X1′、X2′、…、Xs' as the input of three layers of BP neural network, adopt
With Levenberg-Marquardt algorithms and by the use of tansig functions as the input layer of three layers of BP neural network and hidden layer and
Neural transferring function between hidden layer and output layer, is trained and trains to three layers of BP neural network and obtain three layers of BP
Neural network classifier;Wherein, three layers of BP neural network includes an input layer, a hidden layer and an output layer, institute
The nodes for stating input layer are h, and the nodes of the hidden layer are 4, and the nodes of the output layer are 1, and frequency of training is 1000
The secondary or deconditioning when target square error is less than 0.0001.
A kind of localization method of above-mentioned weld seam, it is characterised in that:The value of the m and n are 1~15 natural number.
A kind of localization method of above-mentioned weld seam, it is characterised in that:The first kind described in step 203 is designated 1, described
Two classes are designated 0.
A kind of localization method of above-mentioned weld seam, it is characterised in that:The value of threshold value T described in step 303 is 0.5N.
A kind of localization method of above-mentioned weld seam, it is characterised in that:Monochromatic light described in step one is feux rouges or blue light.
The present invention has advantages below compared with prior art:
1st, method and step of the invention is simple, and it is convenient to realize.
1st, the present invention to image to be detected texture by classifying, so that tack weld position, significantly enhances detection
Method antijamming capability, the present invention program can still be carried out when having obvious light reflection, steel pipe corrosion to weld seam
Reliably detect and position.
2nd, the present invention using industrial camera complete weld seam automatic detection and positioning, compared to based on X-ray technology with
The method of ultrasonic technique, significantly reduces cost, and equipment volume is small, installs and uses more convenient.
3rd, the present invention's is practical, and using effect is good, is easy to promote the use of.
In summary, method and step of the invention is simple, and it is convenient to realize, strong antijamming capability, functional reliability is high, weld seam
Position cost low, install and use more convenient, practical, using effect is good, is easy to promote the use of.
Below by drawings and examples, technical scheme is described in further detail.
Brief description of the drawings
Fig. 1 is method flow block diagram of the invention.
The weld image to be positioned that Fig. 2 is photographed for the present invention using industrial camera.
Fig. 3 corresponds to the two-value classification chart picture of weld image to be positioned for the present invention.
Fig. 4 is two-value classification chart of the present invention as O (r, c) integral projection curve
Embodiment
As shown in figure 1, the localization method of the weld seam of the present invention, comprises the following steps:
Step 1: weld image collection to be positioned:Shot using monochromatic light exposure position while welding, and using industrial camera
Weld image to be positioned and the weld image to be positioned photographed is transferred to image processor;
In the present embodiment, monochromatic light described in step one is feux rouges or blue light.As shown in Fig. 2 to use work in this implementation
The weld image to be positioned that industry video camera is photographed;
Step 2: weld image to be positioned is pre-processed, detailed process is:
Step 201, weld image piecemeal to be positioned:Image processor receives the weld image to be positioned and according in the ranks
The weld image to be positioned is divided into M × N number of weld seam subgraph Y to be positioned every width d and row interval width H1、Y2、…、
YM×N, each weld seam subgraph to be positioned constitutes by m × n pixel, wherein, M for weld seam subgraph to be positioned row
Number, N is the columns of weld seam subgraph to be positioned, and d, H, M, N, m and n are natural number, and d and H unit are pixel;
In the present embodiment, the value of the m and n are 1~15 natural number.
Step 202, weld image dimensionality reduction to be positioned:Image processor calls the drop trained in advance using PCA
Tie up matrix W and according to formula Yf'=WYfDimension-reduction treatment is carried out to M × N number of weld seam subgraph to be positioned, by M × N number of weldering to be positioned
Crack image Y1、Y2、…、YM×NBe converted to the M after dimension-reduction treatment × N number of weld seam subgraph characteristic vector Y to be positioned1′、
Y2′、…、YM×N', wherein, YfFor f-th of weld seam subgraph to be positioned andYf' be dimension-reduction treatment after f-th
Weld seam subgraph characteristic vector to be positioned, f value is 1~M × N natural number;
In the present embodiment, image processor trains the detailed process of dimensionality reduction matrix W using PCA in step 202
For:
Step 2021, according to the method for step one gather one group of weld image to be positioned as training weld image, and
Piecemeal is carried out to group weld image to be positioned according to the method for step 201;
Step 2022, q image block X of selection1、X2、…、Xq, each described image block constitutes by m × n pixel;Its
In, i-th of image block isP is the number of pixels in image block, and p and q are natural number, and i value is 1~q
Natural number;
In the present embodiment, the value of the m and n are 1~15 natural number.
Step 2023, the matrix that q image merged block is obtained to p × q dimensions
Step 2024, to matrix X carry out singular value decomposition obtain X=U ∑s VT, wherein, U=[u1,u2,…,up]∈Rp×p,
Rp×pThe real number matrix tieed up for p × p, V=[v1,v2,…,vq]∈Rq×q, Rq×qThe real number matrix tieed up for q × q, Σ=diag
(σ1,σ2,…,σk), k=min (p, q), uxFor matrix X x-th of left singular vector, x value is 1~p natural number, vyFor
Matrix X y-th of right singular vector, y value is 1~q natural number, σjFor matrix X j-th of singular value, j value is
1~k natural number, σ1≥σ2≥…≥σk≥0;
Step 2025, the preceding h feature constitutive characteristic vector for taking U, obtain orthogonal dimensionality reduction matrix W=[u1,u2,…,uh
]T, wherein,
Step 203, weld image to be positioned classification:Image processor will carry out M × N number of weldering to be positioned after dimension-reduction treatment
In three layers of BP neural network grader that crack image feature vector input training in advance is obtained, M × N number of three layers of BP nerves are obtained
The output of network classifier, and using the output of M × N number of three layers of BP neural network grader as corresponding respectively to after dimension-reduction treatment
M × N number of weld seam subgraph characteristic vector to be positioned classification results;Wherein, three layers of BP nerve nets are output as being used for
Represent that the first kind of the weld seam sub-picture pack to be positioned containing weld seam is identified and for representing that weld seam subgraph to be positioned does not include weld seam
Equations of The Second Kind mark;
In the present embodiment, the first kind described in step 203 is designated 1, and the Equations of The Second Kind is designated 0.I.e. when will be to be positioned
In the obtained three layers of BP neural network grader of weld seam subgraph input training in advance, three layers of BP neural network grader is obtained
When being output as 1, represent that the weld seam sub-picture pack to be positioned contains weld seam;Obtained when by weld seam subgraph to be positioned input training in advance
Three layers of BP neural network grader in, when obtaining three layers of BP neural network grader and being output as 0, represent the weld seam to be positioned
Subgraph does not include weld seam.
In the present embodiment, image processor training in advance obtains the specific of three layers of BP neural network grader in step 203
Process is:
Step 2031, the s image block X of selection from the training weld image after step 2021 processing1、X2、…、
Xs, each described image block constitutes by m × n pixel;Wherein, the λ image block isP is in image block
Number of pixels, p and s are natural number, and λ value is 1~s natural number;
Step 2032, according to formula Xλ'=WXλBy s image block X1、X2、…、XsBe converted to s image block characteristics vector
X1′、X2′、…、Xs', wherein, Xλ' it is the λ image block characteristics vector;
Step 2033, with s image block characteristics vector X1′、X2′、…、Xs' as the input of three layers of BP neural network, adopt
With Levenberg-Marquardt algorithms and by the use of tansig functions as the input layer of three layers of BP neural network and hidden layer and
Neural transferring function between hidden layer and output layer, is trained and trains to three layers of BP neural network and obtain three layers of BP
Neural network classifier;Wherein, three layers of BP neural network includes an input layer, a hidden layer and an output layer, institute
The nodes for stating input layer are h, and the nodes of the hidden layer are 4, and the nodes of the output layer are 1, and frequency of training is 1000
The secondary or deconditioning when target square error is less than 0.0001.Using Levenberg-Marquardt algorithms and utilize
Tansig functions transmit letter as the neuron between the input layer and hidden layer and hidden layer and output layer of three layers of BP neural network
Number, the method being trained to three layers of BP neural network is referring to document:Yu,Hao,and Bogdan M.Wilamowski."
Levenberg-marquardt training."The Industrial Electronics Handbook 5(2011):1-
15。
Step 3: weld and HAZ, detailed process is:
Step 301, the two-value classification chart picture for obtaining weld image to be positioned:Image processor is by M × N after dimension-reduction treatment
The classification results of individual weld seam subgraph characteristic vector to be positioned are combined, and constitute two corresponding to the weld image to be positioned
It is worth classification chart as O (r, c), the two-value classification chart is M × N number of pixel, each pixel correspondence one as O (r, c) size
Subgraph;Wherein, r be the two-value classification chart as O (r, c) row coordinate, c is the two-value classification chart as O (r, c) row are sat
Mark;As shown in figure 3, to correspond to the two-value classification chart of the weld image to be positioned in the present embodiment as O (r, c);
The integral projection curve of step 302, the horizontal direction of acquisition two-value classification chart picture:Image processor is to the two-value
Classification chart is integrated summation as the pixel of O (r, c) every a line, obtains the two-value classification chart as O (r, c) level side
To integral projection curveAs shown in figure 4, being two-value classification chart in the present embodiment as O (r, c) integration
Drop shadow curveIn Fig. 4, abscissa represents two-value classification chart as O (r, c) line number, and ordinate represents bag
The number of subgraph containing weld seam.
Step 303, the integral projection curve tack weld according to the horizontal direction of two-value classification chart picture:If region [r1,r2]
The integral projection curve of horizontal direction for the two-value classification chart as O (r, c)On row coordinate points ri's
ValueExceed threshold value T maximum region, image processor calls position while welding judge module judgment formula
r2-r1>Whether D sets up, as formula r2-r1>When D is set up, it is determined as in the weld image to be positionedRegion is weld seam position region;Otherwise, as formula r2-r1>When D is invalid, it is determined as described
Weld seam is not included in weld image to be positioned;Wherein, D is the minimum developed width of weld seam to be positioned, and D unit is pixel, ri
∈[r1,r2], threshold value T unit is pixel.
In the present embodiment, the value of threshold value T described in step 303 is 0.5N.
It is described above, only it is presently preferred embodiments of the present invention, not the present invention is imposed any restrictions, it is every according to the present invention
Any simple modification, change and equivalent structure change that technical spirit is made to above example, still fall within skill of the present invention
In the protection domain of art scheme.
Claims (6)
1. a kind of localization method of weld seam, it is characterised in that this method comprises the following steps:
Step 1: weld image collection to be positioned:Shoot undetermined using monochromatic light exposure position while welding, and using industrial camera
The weld image to be positioned photographed and is transferred to image processor at weld image by position;
Step 2: weld image to be positioned is pre-processed, detailed process is:
Step 201, weld image piecemeal to be positioned:Image processor receives the weld image to be positioned and wide according to between-line spacing
The weld image to be positioned is divided into M × N number of weld seam subgraph Y to be positioned by degree d and row interval width H1、Y2、…、YM×N, often
The individual weld seam subgraph to be positioned is constituted by m × n pixel, wherein, M is the line number of weld seam subgraph to be positioned, and N is to treat
The columns of tack weld subgraph, d, H, M, N, m and n are natural number, and d and H unit are pixel;
Step 202, weld image dimensionality reduction to be positioned:Image processor calls the dimensionality reduction square trained in advance using PCA
Battle array W and according to formula Y 'f=WYfDimension-reduction treatment is carried out to M × N number of weld seam subgraph to be positioned, by M × N number of weld seam to be positioned
Image Y1、Y2、…、YM×NBe converted to the M after dimension-reduction treatment × N number of weld seam subgraph characteristic vector Y to be positioned1′、Y2′、…、
YM×N', wherein, YfFor f-th of weld seam subgraph to be positioned andY′fIt is to be positioned for f-th after dimension-reduction treatment
Weld seam subgraph characteristic vector, f value is 1~M × N natural number;
Step 203, weld image to be positioned classification:Image processor will carry out the M after dimension-reduction treatment × N number of weld seam to be positioned
In three layers of BP neural network grader that image feature vector input training in advance is obtained, M × N number of three layers of BP neural network is obtained
The output of grader, and using the output of M × N number of three layers of BP neural network grader as corresponding respectively to the M after dimension-reduction treatment
The classification results of × N number of weld seam subgraph characteristic vector to be positioned;Wherein, three layers of BP nerve nets are output as being used to represent
First kind mark of the weld seam sub-picture pack to be positioned containing weld seam and for representing that weld seam subgraph to be positioned does not include the of weld seam
Two classes are identified;
Step 3: weld and HAZ, detailed process is:
Step 301, the two-value classification chart picture for obtaining weld image to be positioned:Image processor treats the M after dimension-reduction treatment × N number of
The classification results of tack weld subgraph characteristic vector are combined, and constitute the two-value point corresponding to the weld image to be positioned
Class image O (r, c), the two-value classification chart is M × N number of pixel, each pixel one subgraph of correspondence as O (r, c) size
Picture;Wherein, r be the two-value classification chart as O (r, c) row coordinate, c is the two-value classification chart as O (r, c) row coordinate;
The integral projection curve of step 302, the horizontal direction of acquisition two-value classification chart picture:Image processor is classified to the two-value
The pixel of image O (r, c) every a line is integrated summation, obtains the two-value classification chart as O (r, c) horizontal direction
Integral projection curve
Step 303, the integral projection curve tack weld according to the horizontal direction of two-value classification chart picture:If region [r1,r2] for institute
Two-value classification chart is stated as the integral projection curve of O (r, c) horizontal directionOn row coordinate points riValueExceed threshold value T maximum region, image processor calls position while welding judge module judgment formula r2-
r1>Whether D sets up, as formula r2-r1>When D is set up, it is determined as in the weld image to be positioned
Region is weld seam position region;Otherwise, as formula r2-r1>When D is invalid, it is determined as in the weld image to be positioned not
Include weld seam;Wherein, D is the minimum developed width of weld seam to be positioned, and D unit is pixel, ri∈[r1,r2], threshold value T list
Position is pixel;
In step 202 image processor use PCA train dimensionality reduction matrix W detailed process for:
Step 2021, according to the method for step one gather one group of weld image to be positioned as training weld image, and according to
The method of step 201 carries out piecemeal to group weld image to be positioned;
Step 2022, q image block X of selection1、X2、…、Xq, each described image block constitutes by m × n pixel;Wherein,
I image block beP is the number of pixels in image block, and p and q are natural number, and i value is 1~q nature
Number;
Step 2023, the matrix that q image merged block is obtained to p × q dimensions
<mrow>
<mi>X</mi>
<mo>=</mo>
<mo>&lsqb;</mo>
<msub>
<mi>X</mi>
<mn>1</mn>
</msub>
<mo>,</mo>
<msub>
<mi>X</mi>
<mn>2</mn>
</msub>
<mo>,</mo>
<mo>...</mo>
<mo>,</mo>
<msub>
<mi>X</mi>
<mi>q</mi>
</msub>
<mo>&rsqb;</mo>
<mo>=</mo>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<msub>
<mi>X</mi>
<mn>11</mn>
</msub>
</mtd>
<mtd>
<msub>
<mi>X</mi>
<mn>21</mn>
</msub>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<msub>
<mi>X</mi>
<mrow>
<mi>q</mi>
<mn>1</mn>
</mrow>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>X</mi>
<mn>12</mn>
</msub>
</mtd>
<mtd>
<msub>
<mi>X</mi>
<mn>22</mn>
</msub>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<msub>
<mi>X</mi>
<mrow>
<mi>q</mi>
<mn>2</mn>
</mrow>
</msub>
</mtd>
</mtr>
<mtr>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
</mtr>
<mtr>
<mtd>
<msub>
<mi>X</mi>
<mrow>
<mn>1</mn>
<mi>p</mi>
</mrow>
</msub>
</mtd>
<mtd>
<msub>
<mi>X</mi>
<mrow>
<mn>2</mn>
<mi>p</mi>
</mrow>
</msub>
</mtd>
<mtd>
<mo>...</mo>
</mtd>
<mtd>
<msub>
<mi>X</mi>
<mrow>
<mi>q</mi>
<mi>p</mi>
</mrow>
</msub>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>;</mo>
</mrow>
Step 2024, to matrix X carry out singular value decomposition obtain X=U ∑s VT, wherein, U=[u1,u2,…,up]∈Rp×p, Rp×p
The real number matrix tieed up for p × p, V=[v1,v2,…,vq]∈Rq×q, Rq×qThe real number matrix tieed up for q × q, Σ=diag (σ1,
σ2,…,σk), k=min (p, q), uxFor matrix X x-th of left singular vector, x value is 1~p natural number, vyFor square
Battle array X y-th of right singular vector, y value is 1~q natural number, σjFor matrix X j-th of singular value, j value is 1
~k natural number, σ1≥σ2≥…≥σk≥0;
Step 2025, the preceding h feature constitutive characteristic vector for taking U, obtain orthogonal dimensionality reduction matrix W=[u1,u2,…,uh]T, its
In,
2. according to a kind of localization method of weld seam described in claim 1, it is characterised in that:Image processor is pre- in step 203
First training obtains the detailed processes of three layers of BP neural network grader and is:
Step 2031, the s image block X of selection from the training weld image after step 2021 processing1、X2、…、Xs, often
Individual described image block is constituted by m × n pixel;Wherein, the λ image block isP is the pixel in image block
Number, p and s are natural number, and λ value is 1~s natural number;
Step 2032, according to formula Xλ'=WXλBy s image block X1、X2、…、XsBe converted to s image block characteristics vector X1′、
X2′、…、Xs', wherein, Xλ' it is the λ image block characteristics vector;
Step 2033, with s image block characteristics vector X1′、X2′、…、Xs' as the input of three layers of BP neural network, use
Levenberg-Marquardt algorithms and by the use of tans ig functions as the input layer of three layers of BP neural network and hidden layer and
Neural transferring function between hidden layer and output layer, is trained and trains to three layers of BP neural network and obtain three layers of BP
Neural network classifier;Wherein, three layers of BP neural network includes an input layer, a hidden layer and an output layer, institute
The nodes for stating input layer are h, and the nodes of the hidden layer are 4, and the nodes of the output layer are 1, and frequency of training is 1000
The secondary or deconditioning when target square error is less than 0.0001.
3. according to a kind of localization method of weld seam described in claim 1 or 2, it is characterised in that:The value of the m and n are 1
~15 natural number.
4. according to a kind of localization method of weld seam described in claim 1 or 2, it is characterised in that:The first kind described in step 203
1 is designated, the Equations of The Second Kind is designated 0.
5. according to a kind of localization method of weld seam described in claim 1 or 2, it is characterised in that:Threshold value T described in step 303
Value be 0.5N.
6. according to a kind of localization method of weld seam described in claim 1 or 2, it is characterised in that:Monochromatic light described in step one
For feux rouges or blue light.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510106889.6A CN104636760B (en) | 2015-03-11 | 2015-03-11 | A kind of localization method of weld seam |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510106889.6A CN104636760B (en) | 2015-03-11 | 2015-03-11 | A kind of localization method of weld seam |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104636760A CN104636760A (en) | 2015-05-20 |
CN104636760B true CN104636760B (en) | 2017-09-05 |
Family
ID=53215488
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510106889.6A Expired - Fee Related CN104636760B (en) | 2015-03-11 | 2015-03-11 | A kind of localization method of weld seam |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104636760B (en) |
Families Citing this family (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106514064A (en) * | 2015-09-15 | 2017-03-22 | 苏州中启维盛机器人科技有限公司 | Weld verification method |
CN106530269A (en) * | 2015-09-15 | 2017-03-22 | 苏州中启维盛机器人科技有限公司 | Weld detection method |
CN105427295B (en) * | 2015-11-12 | 2018-03-30 | 汪月银 | A kind of image-recognizing method and image identification system based on weld seam |
CN109239081B (en) * | 2018-09-18 | 2021-01-15 | 广东省特种设备检测研究院珠海检测院 | Weld quality parameter detection method based on structured light and visual imaging |
CN109993741B (en) * | 2019-04-03 | 2022-10-04 | 南昌航空大学 | Steel rail welding seam contour automatic positioning method based on K-means clustering |
CN110276296B (en) * | 2019-06-21 | 2023-04-28 | 新疆大学 | Structured light weld joint identification method based on image processing |
CN110274960B (en) * | 2019-08-02 | 2022-07-26 | 大唐锅炉压力容器检验中心有限公司 | Steel pipe microscopic structure evaluation method and device based on nonlinear ultrasound |
CN111203676A (en) * | 2020-01-17 | 2020-05-29 | 湘潭大学 | Laser vision-based aircraft engine front support intersecting line weld joint identification method |
CN111637835A (en) * | 2020-05-25 | 2020-09-08 | 西咸新区大熊星座智能科技有限公司 | Circular tube weld joint position positioning method and device |
CN111637847A (en) * | 2020-05-25 | 2020-09-08 | 西咸新区大熊星座智能科技有限公司 | Welding seam parameter measuring method and device |
CN112362678A (en) * | 2020-12-01 | 2021-02-12 | 北京弘燕高新技术有限公司 | Intelligent welding seam recognition device |
CN112766460A (en) * | 2021-01-12 | 2021-05-07 | 上海电机学院 | Weld joint identification method based on sub-region and BP neural network |
CN113042939B (en) * | 2021-03-22 | 2022-06-28 | 山东大学 | Workpiece weld joint positioning method and system based on three-dimensional visual information |
DE102021117714A1 (en) | 2021-07-08 | 2023-01-12 | Endress+Hauser SE+Co. KG | Automatic seam detection for a welding process |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1924896A (en) * | 2006-09-14 | 2007-03-07 | 上海交通大学 | Partial image processing based butt type welding seam recognition method |
CN102509108A (en) * | 2011-10-14 | 2012-06-20 | 桂林电子科技大学 | Welding spot defect identifying method |
CN103268491A (en) * | 2013-05-30 | 2013-08-28 | 哈尔滨工业大学 | Weld defect ultrasound phased array sector scanned image feature extraction method |
CN104166842A (en) * | 2014-07-25 | 2014-11-26 | 同济大学 | Three-dimensional palm print identification method based on partitioning statistical characteristic and combined expression |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
DE102004016669B3 (en) * | 2004-01-07 | 2005-10-13 | Daimlerchrysler Ag | Method for testing a laser weld seam |
US9180552B2 (en) * | 2013-05-27 | 2015-11-10 | ThinkSmart IT Solutions Private Limited | System and method for identifying defects in welds by processing X-ray images |
-
2015
- 2015-03-11 CN CN201510106889.6A patent/CN104636760B/en not_active Expired - Fee Related
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1924896A (en) * | 2006-09-14 | 2007-03-07 | 上海交通大学 | Partial image processing based butt type welding seam recognition method |
CN102509108A (en) * | 2011-10-14 | 2012-06-20 | 桂林电子科技大学 | Welding spot defect identifying method |
CN103268491A (en) * | 2013-05-30 | 2013-08-28 | 哈尔滨工业大学 | Weld defect ultrasound phased array sector scanned image feature extraction method |
CN104166842A (en) * | 2014-07-25 | 2014-11-26 | 同济大学 | Three-dimensional palm print identification method based on partitioning statistical characteristic and combined expression |
Also Published As
Publication number | Publication date |
---|---|
CN104636760A (en) | 2015-05-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104636760B (en) | A kind of localization method of weld seam | |
WO2021203618A1 (en) | Image sample generating method and system, and target detection method | |
CN104992449B (en) | Information identification and surface defect online test method based on machine vision | |
CN105352437B (en) | Board method for detecting position and device | |
CN106934813A (en) | A kind of industrial robot workpiece grabbing implementation method of view-based access control model positioning | |
CN104977313A (en) | Method and device for detecting and identifying X-ray image defects of welding seam | |
CN107808376A (en) | A kind of detection method of raising one's hand based on deep learning | |
CN106067031B (en) | Based on artificial mechanism for correcting errors and deep learning network cooperation machine vision recognition system | |
CN106934355A (en) | In-car hand detection method based on depth convolutional neural networks | |
CN110189375A (en) | A kind of images steganalysis method based on monocular vision measurement | |
CN102565103B (en) | Tracking detection method for weld defects based on X-ray image | |
Kobayashi et al. | Fine-grained action recognition in assembly work scenes by drawing attention to the hands | |
Liu et al. | Extended faster R-CNN for long distance human detection: Finding pedestrians in UAV images | |
CN112364754A (en) | Bolt defect detection method and system | |
CN110910363A (en) | Insufficient solder joint detection method, system and medium based on machine vision and deep learning | |
CN115958605A (en) | Monocular infrared thermal imaging vision manipulator object grabbing posture estimation device and method | |
CN116385758A (en) | Detection method for damage to surface of conveyor belt based on YOLOv5 network | |
Patil et al. | An autonomous technique for multi class weld imperfections detection and classification by support vector machine | |
CN111208148A (en) | Dig hole screen light leak defect detecting system | |
CN105678731A (en) | Identification method of buckling of band steel | |
CN109975307A (en) | Bearing surface defect detection system and detection method based on statistics projection training | |
Park et al. | FCN based gas leakage segmentation and improvement using transfer learning | |
CN117218457A (en) | Self-supervision industrial anomaly detection method based on double-layer two-dimensional normalized flow | |
CN112766460A (en) | Weld joint identification method based on sub-region and BP neural network | |
WO2023159927A1 (en) | Rapid object detection method based on conditional branches and expert systems |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170905 Termination date: 20180311 |
|
CF01 | Termination of patent right due to non-payment of annual fee |