CN111696107B - Molten pool contour image extraction method for realizing closed connected domain - Google Patents
Molten pool contour image extraction method for realizing closed connected domain Download PDFInfo
- Publication number
- CN111696107B CN111696107B CN202010776108.5A CN202010776108A CN111696107B CN 111696107 B CN111696107 B CN 111696107B CN 202010776108 A CN202010776108 A CN 202010776108A CN 111696107 B CN111696107 B CN 111696107B
- Authority
- CN
- China
- Prior art keywords
- molten pool
- image
- gradient
- gray
- edge
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000605 extraction Methods 0.000 title claims abstract description 32
- 238000003466 welding Methods 0.000 claims abstract description 53
- 230000000694 effects Effects 0.000 claims abstract description 46
- 238000000034 method Methods 0.000 claims abstract description 38
- 238000009826 distribution Methods 0.000 claims abstract description 23
- 238000007781 pre-processing Methods 0.000 claims abstract description 11
- 230000008859 change Effects 0.000 claims abstract description 9
- 238000004422 calculation algorithm Methods 0.000 claims description 44
- 230000004044 response Effects 0.000 claims description 43
- 238000004364 calculation method Methods 0.000 claims description 25
- 230000009466 transformation Effects 0.000 claims description 24
- 238000012545 processing Methods 0.000 claims description 16
- 238000001514 detection method Methods 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 13
- 238000006243 chemical reaction Methods 0.000 claims description 5
- 238000012805 post-processing Methods 0.000 claims description 4
- 230000002411 adverse Effects 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 238000011426 transformation method Methods 0.000 claims description 3
- 239000002131 composite material Substances 0.000 claims description 2
- 230000001131 transforming effect Effects 0.000 claims description 2
- 230000006872 improvement Effects 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 20
- 238000003708 edge detection Methods 0.000 description 19
- 229910001220 stainless steel Inorganic materials 0.000 description 11
- 239000010935 stainless steel Substances 0.000 description 10
- 229920006395 saturated elastomer Polymers 0.000 description 7
- 239000010963 304 stainless steel Substances 0.000 description 2
- XKRFYHLGVUSROY-UHFFFAOYSA-N Argon Chemical compound [Ar] XKRFYHLGVUSROY-UHFFFAOYSA-N 0.000 description 2
- 229910000589 SAE 304 stainless steel Inorganic materials 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000015572 biosynthetic process Effects 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000003672 processing method Methods 0.000 description 2
- 230000005855 radiation Effects 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 229910052786 argon Inorganic materials 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000001681 protective effect Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000005728 strengthening Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
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
-
- G06T5/94—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
Abstract
The invention relates to a molten pool contour image extraction method for realizing closed connected domains, which comprises the following steps: 1. extracting a molten pool image, 2, preprocessing and calculating the molten pool image, 3, calculating a gradient map based on a template of a guide operator, 4, merging the images, and 5, extracting a contour line of the molten pool image. The device and the method solve the problem that the contour line of the molten pool is difficult to extract because the brightness distribution of the front end and the rear end of the molten pool is uneven and the front end of the molten pool is affected by arc light; the problems that the brightness of the area at the rear end of the molten pool is dark and the distribution is uneven, and the gray level change is not obvious and the gradient is weak after the image is converted into a gray level image at the two end parts of the rear end of the molten pool are solved; the method has an indication effect on the forming of the welding seam and the improvement of the welding quality.
Description
Technical Field
The invention relates to a molten pool contour image extraction method for realizing closed connected domains, and belongs to the technical field of image analysis.
Background
The image characteristics of the molten pool have important guiding function on the formation of the welding line and the detection of welding quality. The brightness distribution of the front end and the rear end of the molten pool image extracted by the traditional method is uneven, wherein the front end of the molten pool is seriously affected by the arc light, the detail of the part is more abundant, and when the arc light is too serious, the situation that the arc light covers the edge information of the head part of the molten pool is likely to occur. Compared with the front end, the brightness of the rear end area of the molten pool is darker and the distribution is uneven, the brightness is brighter in the middle part of the rear end of the molten pool, and the brightness is darker from the middle part to the two ends, so that the two end parts of the rear end of the molten pool have the characteristics of insignificant gray level change and weaker gradient after the image is converted into a gray level image.
When the region is detected by the edge detection operator, the weak edge region will not respond and the detected edge is not continuous enough. Edge detection is directly carried out on a stainless steel molten pool image in the TIG welding process by using a Canny operator, and the detected contour edge line can be affected by three aspects: 1. the edge of the weak edge area at the rear end of the molten pool cannot be accurately detected; 2. the brightness saturation area formed by arc light reflection on the surface of the molten pool introduces an interference edge; 3. texture information of the formed weld area at the rear end of the molten pool introduces interference edges. Compared with the traditional image detection method, the Canny edge detection operator is an optimized edge detection operator scheme with high performance, and if the algorithm cannot achieve a satisfactory effect on a TIG welding stainless steel molten pool image, the accurate and complete contour of a molten pool area is difficult to obtain by the traditional image edge detection algorithm.
An effective molten pool contour detection and extraction method is urgently needed at present to solve the problem that contour edges detected by an edge detection operator are incomplete.
Disclosure of Invention
In order to solve the technical problems, the invention provides a molten pool contour image extraction method for realizing a closed connected domain, which has the following specific technical scheme:
the method for extracting the contour image of the molten pool for realizing the closed connected domain comprises the following steps:
step one: extracting a molten pool image: carrying out welding work on a welding workpiece by using a welding machine provided with a camera, so that the camera can acquire molten pool images of each welding line from forming to welding;
step two: preprocessing and calculating a molten pool image: the method for preprocessing and calculating the molten pool image by introducing a weak edge enhancement method based on nonlinear gray level transformation and a gray level transformation method based on sliding window gray level information comprises the following steps:
(2-1) a weak edge enhancement method based on nonlinear gray level transformation, introducing a composite power function represented by formula (1.1),
wherein x represents the gray value of a single pixel point in the image, m and k are integers larger than 1, and are adjustable parameters, and the distribution condition of the gray value after transformation is different along with the change of m and k;
(2-2) a gray level conversion method based on sliding window gray level information, the calculation flow is as follows:
(2-2 a) traversing from a pixel point at a corner of the puddle image, setting a gray value G (i, j) at the pixel point of the current traversing, and setting a gray maximum value G in a neighborhood region of 3*3 centered on the pixel point max The minimum value is set as G min ；
(2-2 b) normalizing the gray values of the pixels in the 3*3 neighborhood in the region in the formula (1.2),
(2-2 c) the pixel low gray value G (i, j) in the step (2-2 b) is determined according to the value of the pixel low gray value G (i, j) in the step (0, 1)]The interval in the gray scale range of (1.1) is substituted to obtain the gray value G of the pixel point calculated by the nonlinear gray scale transformation algorithm n (i,j)，
(2-2 d) G n (i, j) substituting the value of g (i, j) in the original formula to the left of the equal sign in the formula (1.2), namely, the gray value range in the neighborhood region is from [0,1]Stretched to the original gray value range [ G min ,G max ]As shown in the formula (1.3),
g n (i,j)＝G min +(G max -G min )G n (i,j)， (1.3)
(2-2 e) adjusting the gray distribution of the image based on the calculation method of traversing all the pixels in the image by the sliding window, suppressing the adverse effect generated by the brightness saturation region, firstly calculating the gray average value G of the pixels in the 3*3 neighborhood centered on the traversed pixel (i, j) according to the formula (1.4) mean G (i, j), i.e. the gray value of the pixel point (i, j),
G mean (i,j)＝(g(i-1,j-1)+g(i-1,j)+g(i-1,j+1)+g(i,j-1)+g(i,j)
+g(i,j+1)+g(i+1,j-1)+g(i+1,j)+g(i+1,j+1))/9， (1.4)
after obtaining the neighborhood gray average value, performing calculation transformation based on an inverse proportion function on gray values of all pixel points in the 3*3 neighborhood, as shown in formula (1.5):
g(i+k1,j+k2)＝g(i+k1,j+k2)×((M/G mean ) 2 ) k1,k2∈[-1,1]， (1.5)；
step three: based on gradient map calculation facilitating guide operator template matching: expanding Sobel operators in the initial horizontal direction and the initial vertical direction to four directions, and carrying out convolution operation on the preprocessed image by using the Sobel operators to obtain a gradient map of a molten pool image;
step four: image merging: thresholding the gradient map to obtain a gradient binary map, and splicing the rear end part of the gradient binary map with the front end part of the contour line obtained by the detection of the Canny operator to obtain a rough extraction result map of the edge of the molten pool image, wherein: traversing the original pixels of the gradient map in the third step, traversing the traversed pixels (i, j), and determining the gradient amplitude P of the corresponding pixel coordinate points after convolution operation 0° ，P ω ，P 90° P σ Maximum value P of (B) max And a minimum value P min ，
P max ＝max{P 0° ，P ω ，P 90° ，P σ }， (1.6)
P min ＝min{P 0° ，P ω ，P 90° ，p σ }， (1.7)
P in the formula ω ，P σ For two gradient magnitudes in horizontal and vertical directions, ω and σ are expansion direction angles, so as to obtain a gradient response of the pixel point in the remaining two molten pool gradient maps as P 1 ，P 2 And calculates pixel point (i, j) based on edge-oriented operator gradient value P according to formula (1.8),
step five: extracting contour lines of molten pool images: performing edge connection operation based on judging gradient strength and direction on the edge rough extraction result graph, obtaining weak edge information of an image, filling to obtain a complete connected domain area, performing edge smoothing operation based on mathematical morphology on the connected domain area, and finally obtaining a molten pool image contour line, wherein the method comprises the following steps:
(5-1) weak edge connection with angle based on gradient strength, the procedure is as follows:
(5-1 a) starting to search for strong response pixel points at the edge corner pixel points of the gradient map of the weak edge at the rear end of the molten pool and searching for the neighborhood region of the strong response pixel points 5*5, and transforming the expanded omega and sigma directions into horizontal and vertical directions through trigonometric functions, wherein the formula is shown as formula (1.9)
(5-1 b) calculating the amplitude P and the direction theta of the gradient of the pixel points in the neighborhood according to the formulas (1.8) and (1.9),
(5-1 c) judging whether the pixel points in the neighborhood are set as the pixel points with strong response, as shown in the formula (1.10),
if the pixel points in the neighborhood satisfying the formula (1.10) do not respond to the pixel points as strong response pixel points, setting the pixel points as strong response pixel points, wherein the pixel points are the strong response pixel points, and performing no additional processing;
(5-2) post-processing based on mathematical morphology, the processing flow being as follows:
(5-2 a) for removing small connected domain whose connected domain scale is smaller than the set threshold value by utilizing method of judging connected domain scale for the interference edge introduced by Canny splice in molten pool region,
(5-2 b) filling a closed hole area in the image, subtracting the filled image from the image before filling to obtain a gradient binary image of the closed hole area, and reserving a connected domain with the largest area scale;
(5-2 c) in order to make the connected domain have the characteristic of smoother edge of the molten pool area, establishing a disc operator structure with a radius of 15 pixels, and performing a closing operation and then an opening operation on the connected domain by using the disc operator structure;
and (5-2 d) superposing the obtained contour edge of the molten pool area on the original image of the molten pool image to obtain a contour line of the molten pool image.
Further, in the third step, the Sobel operator in only two directions of horizontal and vertical is extended to four directions, a first-order differential operator in two directions of ω=45° and σ=135° is added, and convolution operation is performed on the molten pool image, so as to obtain a molten pool gradient map in four directions.
In the second step, m=2 and k=2 are substituted into equation (1.1), so that gray information of a certain region in the molten pool image before and after transformation is obtained, gradient amplitude of a weak edge region at the rear end of the molten pool is increased, and noise is prevented from being introduced.
In the second step, m=90 is substituted into (1.5), the weak edge of the rear end of the molten pool image is enhanced, and the brightness saturation region generated by the welding arc light at the front end of the molten pool is restrained.
Further, the template after the Sobel operator expansion is as follows:
the (a), (b), (c) and (d) are templates of the Sobel operator in the horizontal direction, the 45-degree direction, the vertical direction and the 135-degree direction in sequence.
The working principle of the invention is as follows:
in order to strengthen the weak edge area part at the rear end of a molten pool, preprocessing calculation is carried out on the molten pool image by a weak edge strengthening method based on nonlinear gray level transformation and a gray level transformation method based on sliding window gray level information; then, expanding Sobel operators in the horizontal direction and the vertical direction to four directions, carrying out convolution operation on the preprocessed molten pool image by utilizing the Sobel operators in the four directions to judge the direction of the weak edge at the rear end, wherein each sliding window corresponds to a Sobel operator template in the strongest response direction, carrying out convolution operation on each sliding window of the image by using the Sobel operator template in the strongest response direction, obtaining a gradient map of the molten pool image, obtaining a gradient binary map after reasonable thresholding operation, and splicing the rear end part of the obtained binary map with the front end part of a contour line detected by a Canny operator to obtain an edge rough extraction result map; finally, in order to obtain a closed connected domain, edge connection operation based on judging gradient strength and direction is carried out on the edge rough extraction result graph, so that more weak edge information is obtained, a complete connected domain area is obtained through filling on the basis, edge smoothing operation based on mathematical morphology is carried out, and finally, a molten pool contour line to be extracted is obtained.
The invention has the beneficial effects that:
the invention utilizes a camera to shoot and extract a molten pool image generated on a welding workpiece by a butt welder, obtains a complete and clear molten pool image contour line by preprocessing the molten pool image, calculating a gradient map based on a template of a convenient guiding operator, merging images, extracting contour lines of the molten pool image and other image processing methods, and solves the problem that the contour lines of the molten pool are difficult to extract due to uneven brightness distribution of the front end and the rear end of the molten pool area and the influence of arc light on the front end of the molten pool; the problems that the brightness of the area at the rear end of the molten pool is dark and the distribution is uneven, and the gray level change is not obvious and the gradient is weak after the image is converted into a gray level image at the two end parts of the rear end of the molten pool are solved; and plays an indicating role in the formation of the weld joints and the improvement of the welding quality.
Drawings
Figure 1 is a schematic flow chart of the algorithm of the present invention,
FIG. 2 is a schematic view showing the structure of a visual sensor device for molten pool according to the present invention,
figure 3 is a graph of the weak edge enhancement contrast effect of the present invention based on a nonlinear gray scale transformation,
figure 4 is a schematic view of gray scale distribution information of the present invention after enhancement of the weak edge local area,
FIG. 5 is a schematic view of pseudo edges formed by the saturated region of the surface brightness of the molten pool according to the present invention,
figure 6 is a graph of the comparative effect of the algorithmic pre-processing of the present invention,
figure 7 is a four-way Sobel operator template convolution gradient map of the present invention,
figure 8 is a graph of the gradient magnitude contrast effect of the present invention based on edge direction operator template matching,
FIG. 9 is a graph showing the splicing effect of the detection results of the rear weak edge region and the front Canny operator of the molten pool,
figure 10 is a graph of the edge connection contrast effect based on gradient magnitude and direction of the present invention,
figure 11 is a graph of interference edge elimination contrast effect based on connected domain scale size of the present invention,
FIG. 12 is a flow chart of the invention for acquiring a communicating region of a molten pool area based on a filling and differencing operation,
FIG. 13 is a graph of the smooth contrast effect of connected domains based on mathematical morphology operations of the present invention,
FIG. 14 is a drawing showing the outline extraction effect of the connected domain of the present invention,
figure 15 is a flow chart of the OTM-EDG algorithm of the present invention,
figure 16 is a comparison of puddle profile extraction results for various algorithms of the present invention,
in the figure: 1-welding machine, 2-image processor, 3-welding workpiece and 4-camera.
Detailed Description
The invention will now be described in further detail with reference to the accompanying drawings. The drawings are simplified schematic representations which merely illustrate the basic structure of the invention and therefore show only the structures which are relevant to the invention.
As shown in fig. 1, in the method for extracting the molten pool profile image for realizing the closed connected domain, a welding machine 1 provided with a camera 4 is used for performing welding work on a welding workpiece 3, so that the camera 4 can acquire the molten pool image from forming to welding of each welding seam. Wherein, the welding machine 1 is a TIG welding machine, the workpiece is a 304 stainless steel flat plate welding workpiece, and the TIG welding machine performs welding seam welding on the 304 stainless steel flat plate welding workpiece. TIG welder was welded under 100% argon using stainless steel wire 1.2mm diameter. In order to facilitate the processing of the collected molten pool image from the forming to the welding end of each welding line at the later stage, the camera 4 adopts a CCD industrial color camera, and in order to eliminate the interference of redundant band radiation during the collection, the front end of the camera 4 is provided with a 660nm optical filter and a protective glass for protecting the lens of the camera 4 from heat radiation loss. The CCD industrial color camera shooting angle is set in alignment with the weld joint area, so that exposure is carried out every 20 mu s, and a molten pool image with 1920 x 1200 resolution is acquired. Since the proportion of the molten pool area in the acquired image to the whole image is not large and the texture of the stainless steel flat plate can interfere with the post-processing, the image is cut by using the molten pool area as the center to form the ROI with the size of 400 pixels. The CCD industrial color camera is fixed on the mechanical arm of the TIG welder at a certain angle, so that the molten pool positions in hundreds of molten pool images acquired in one welding line are basically fixed at fixed positions in the images. The CCD industrial color camera is connected with an image processor 2, and the image processor 2 calculates and processes the molten pool image.
In the TIG welding stainless steel molten pool image, the difficulty of contour extraction is that the weak edge area at the rear end of the molten pool cannot be accurately extracted and the interference caused by the saturated area of the brightness of the surface of the molten pool to the edge extraction. Since the difference of gray values at two sides of the edge in the weak edge area at the rear end of the molten pool is tiny, if the gradient amplitude obtained by directly using a differential operator to calculate is small, the gradient amplitude is insufficient to support the gradient amplitude to be detected as a real contour edge of the molten pool. To change this situation, it is desirable that the pixel gray value on the side of the higher gray value at the weak edge is pulled high, the pixel gray value on the other side of the lower gray value is pulled low, and the pixel gray values on the same side of the treated weak edge region remain at a similar level. To achieve this, a non-linear gray scale transformation algorithm based on a power function is used. The key point of realizing the effect through the nonlinear transformation algorithm is to introduce a compound power function expressed as a formula (1.1), wherein x represents gray value information of a certain pixel point in an image, m and k are integers larger than 1 and are adjustable parameters, and the distribution condition of the gray values after transformation is different along with the change of m and k. As formula (1.1):
it can be seen that the calculation mode of the formula (1.1) is to perform power function transformation on the gray value x in the value interval of [0,1], after the gray values of the pixel points are normalized and the function transformation in the formula, the gray values of the pixels with the gray value range lower than 0.5 at the two sides of the weak edge are pulled down, otherwise, the gray values of the pixels with the gray value range higher than 0.5 are pulled up, so that the difference value between the strong gray values and the weak gray values is increased. If the nonlinear gray-scale transformation algorithm based on the power function is to be applied to global processing of the image, and because the normalization of gray values is involved in the expression (1.1), the nonlinear gray-scale transformation algorithm based on the power function is realized by using a sliding window calculation mode in order to better preserve the spatial relationship between pixel points in the image. The specific calculation flow is as follows:
(a) The pixel point at the upper left corner of the molten pool image is traversed, the gray value of the pixel point currently traversed is set as G (i, j), and the gray maximum value in the neighborhood region with the size of 3*3 taking the pixel point as the center is set as G max The minimum value is set as G min ；
(b) The gray values of the pixels in the 3*3 neighborhood are normalized in the area as in (1.2) and the gray value of the center pixel is taken as an example here:
(c) The pixel gray value G (i, j) after normalization obtained in the step (b) is calculated according to the value of the pixel gray value G (i, j) in the step (0, 1)]The interval within the gray range of (1.1) is substituted into the formula to calculate, and the gray value G of the pixel point after being calculated and processed by the nonlinear gray conversion algorithm is obtained n (i,j)；
(d) Will G n (i, j) substituting the left side of the medium number in the formula (1.2) to reversely push g (i, j) in the original formula, namely, the gray value range in the neighborhood region is from [0,1]Stretched to the original gray value range [ G min ,G max ]The calculation process is shown as (1.3):
g n (i,j)＝G min +(G max -G min )G n (i,j)， (1.3)
in the first step in the procedure, gray information is calculated in a neighborhood region where the traversed pixel point takes 3*3, and the reason for selecting the neighborhood radius as 3 is because if the radius is larger, the calculation amount is increased, and the sensitivity to weak edge information may be reduced. For the selection of m and k values in the formula (1.1), the image of the molten pool is tested by using several groups of different parameters, the effect after processing is shown in fig. 3, and it can be seen that the image processing effect has obvious difference along with the difference of m and k values, wherein a proper combination is that both m and k are 2, so that the gradient amplitude of the weak edge area at the rear end of the molten pool is effectively increased, unnecessary noise is not introduced to the image like the third and fifth images from left to right in fig. 3, and the subsequent processing is burdened. The first diagram from left to right in fig. 3 is a molten pool original diagram, the second diagram is an effect diagram with m=2 and k=2, the third diagram is an effect diagram with m=2 and k=3, the fourth diagram is an effect diagram with m=3 and k=2, and the fifth diagram is an effect diagram with m=3 and k=3. Substituting m=2 and k=2 into the expression (1.1) and executing the nonlinear gray level conversion calculation operation based on the power function to obtain gray level information of a certain region in the molten pool image before and after conversion, as shown in fig. 4.
Another problem in the difficulty of extracting the contour of the molten pool is that the brightness saturation region of the surface of the molten pool caused by the reflection of welding arc light interferes with the extraction of edges, and because of the large gray level difference between the brightness saturation region and the surrounding surface region of the molten pool, a pseudo edge with a wide range is generated during edge detection, as shown by the region marked by circles in fig. 5, and the pseudo edge exists in the real contour region of the molten pool and usually exists in a communicated manner, so that it is difficult to adaptively reject the edge line of the part according to the judging edge length. In order to completely eliminate this influence, it is desirable that the gradation value variation distribution of the luminance saturated region be identical to that of the region in the vicinity thereof, and if the gradation values of the pixels in the luminance saturated region, that is, the set of pixels whose gradation value is close to 255, are directly replaced with those of the pixels around them, the gradation value distribution of the luminance saturated region is adjusted, and this does not reduce the gradient amplitude between the region and the surrounding region, a finer algorithm is required to adjust the gradation value distribution of the region. Similar to a weak edge enhancement algorithm based on sliding window nonlinear gray level transformation, the gray level distribution condition of the image is adjusted by a calculation mode of traversing all pixel points in the image based on the sliding window, so that the aim of suppressing adverse effects generated by a brightness saturation region is fulfilled. The specific calculation flow is as follows:
firstly, calculating the gray average value G of the pixel points in the 3*3 neighborhood centering on the traversed pixel point (i, j) according to the formula (1.4) mean G (i, j), i.e. the gray value of the pixel point (i, j),
G mean (i,j)＝(g(i-1,j-1)+g(i-1,j)+g(i-1,j+1)+g(i,j-1)+g(i,j)
+g(i,j+1)+g(i+1,j-1)+g(i+1,j)+g(i+1,j+1))/9， (1.4)
under the condition that the neighborhood gray average value is obtained, the gray values of all pixel points in the whole 3*3 neighborhood are calculated and transformed based on an inverse function, as shown in a formula (1.5):
g(i+k1,j+k2)＝g(i+k1,j+k2)×((M/G mean ) 2 ) k1,k2∈[-1,1]， (1.5)；
from equation (1.5), it can be seen that the adaptation of the algorithm is mainly reflected in the fact that the transform coefficient of the gray value in the sliding window is related to the gray mean value of the window, due to M/G mean Is a G mean As the inverse proportion function of the variables, the processing effect of the step operation on the image is to reduce the gray level value of the brightness saturation region and increase the gray level value of the weak edge region at the rear end of the molten pool, however, due to the existence of the sliding window operation, the distribution change of the gray level values among different regions in the adjusted image shows a slow trend, the M value is taken as 90, and the gray level adjustment effect of the molten pool image is shown in fig. 6. Wherein, fig. 6 is a molten pool original image and an effect image after algorithm pretreatment from left to right in sequence. It can be seen that the puddle image passes through a sliding window based non-lineAfter the linear gray level transformation algorithm is processed, the weak edge at the rear end is obviously enhanced, the calculation and the processing of a subsequent edge detection operator are facilitated, and the brightness saturation area generated by welding arc light on the surface of a molten pool and the front end of the molten pool is well suppressed in the step.
Gradient computation based on edge-guided operator template matching. The edge detection operators such as Sobel and Canny are used for calculating the gradient amplitude in the neighborhood region of the pixel point traversed in the image by utilizing a first-order differential operator template in the horizontal direction and a first-order differential operator template in the vertical direction, the calculation method in the Canny operator is to take the root value of the square sum of the gradient amplitude in the two directions as the gradient amplitude of the pixel point, and the direction of the gradient of the point is calculated by utilizing an arctangent function of the ratio of the gradient amplitudes in the two directions in the process. The problem that may occur in the gradient calculation process is that the directions of edges in the image extend towards many different directions, and only the first-order differential operator templates in the horizontal direction and the vertical direction are used for detecting the gradient amplitude of the neighborhood, and this way is sensitive to noise in the image, and may cause missing detection and false detection of the edges. On the basis of the preprocessed molten pool image, the Sobel operator templates expanded to four directions are utilized to perform edge direction strong response detection on the molten pool image, the operator template of the edge direction corresponding to the 3*3 neighborhood window of each pixel point is determined, then the gradient value of the central pixel point of the neighborhood window is obtained in a weighted calculation mode, and the gradient map of the whole image is obtained through operation based on a sliding window. Expanding Sobel operators in the original horizontal and vertical directions to four directions, adding a first-order differential operator in the two directions of 45 DEG and 135 DEG, and the expanded Sobel operator template is shown as follows:
and (a), (b), (c) and (d) are template diagrams of the horizontal direction, the 45-degree direction, the vertical direction and the 135-degree direction of the Sobel operator in sequence. The Sobel operator templates in four directions are utilized to perform molten pool alignmentThe image is convolved to obtain four gradient maps as shown in fig. 7. It can be seen from the figure that after convolution with the operator templates in different directions, the edge information in different regions and in different directions in the puddle image produces responses of different intensities. The first drawing from left to right in fig. 7 is the original drawing; the second graph is the result of convolution with the differential operator template in the horizontal direction, and it can be seen that the lower side edge information and the upper side edge information of the molten pool produce stronger response; the third graph is the result of convolution with the 45 differential operator template, and it can be seen that the upper side edge of the puddle in this graph produces a stronger response; the fourth graph and the fifth graph are gradient amplitude graphs generated by convolution of the differential operator templates in the vertical direction and the 135 DEG direction with the original graph, and the weak edge part at the rear end of the molten pool in the two graphs generates stronger response. After four fused pool gradient amplitude diagrams generated by convolution calculation of differential operator templates in different directions and an original image are obtained, traversing row by row and column by column from the upper left corner pixel of the image on the basis of the original image, and inquiring gradient amplitude P of a pixel coordinate point corresponding to the traversed pixel point (i, j) in the four images 0° ，P 45° ，P 90° P 135° Is set between the maximum value and the minimum value of the set,
p max ＝max{P 0° ，p ω ，P 90° ，P σ }， (1.6)
P min ＝min{P 0° ，P ω ，P 90° ，P σ }， (1.7)
setting that the corresponding gradient response maximum value P of the pixel point in the four gradient amplitude diagrams is obtained max And response minimum value P min Setting the gradient response of the pixel point in the other two gradient amplitude diagrams to be P respectively 1 ，P 2 Calculating gradient amplitude P of pixel points (i, j) based on edge guide operator template matching according to the formula (1.8), wherein the M value is set as 3, the M value is set as 1/3,
the effect of the gradient amplitude map based on the edge direction operator template matching obtained in the calculation process is shown in fig. 8, the obtained gradient amplitude map is subjected to thresholding operation, the threshold coefficient is finely adjusted according to molten pool images under different process parameters, the gradient binary map shown in the third graph from left to right in fig. 8 is obtained, and it can be seen that the gradient operator passing through the algorithm can effectively generate stronger gradient response near a weak edge area in the molten pool image, and the effect cannot be achieved by using a Canny or Sobel edge detection operator for direct detection. The first graph from left to right in fig. 8 is a contrast original graph, and the second graph is a gradient amplitude gray scale graph. FIG. 8 shows that the trailing weak edge region in the puddle image produces a stronger response after the enhancement preprocessing and the gradient calculation algorithm, but it can also be seen that the contour edge at the leading end of the puddle does not produce a stronger response after the gray level transformation in the preprocessing. Finally, the detection effect of the Canny operator at the front end part of the molten pool is spliced with the weak edge area of the molten pool with the detection effect of the algorithm by taking the central line of the molten pool area as a reference, and the effect diagram is shown in figure 9.
In the post-processing part of the algorithm, a closed connected domain which is matched with the molten pool area is acquired based on the gradient map obtained by calculation, and then the connected domain is processed by an image processing algorithm based on mathematical morphology, so that the edge of the connected domain is smoother and more accurate. In fig. 9 it can be seen that the shaped region of the puddle on the right side of the image also produces a stronger response in the process of calculating the gradient because its texture information is similarly enhanced in the weak edge enhancement algorithm, but because the purpose of the present algorithm is to obtain a closed connected region rather than just a strong response to the contour edge line, the interference response produced to the shaped region of the puddle may not need to be screened out as in the conventional edge detection algorithm. In the obtained gradient magnitude map of the weak edge region at the rear end of the weld pool, even if the region is gradient-enhanced in the first step of the algorithm, the condition of the weld pool image acquired at different moments on a formed weld joint during the welding process is different, and some weak edge regions in the calculated gradient magnitude map cannot generate strong response, so that the weld pool region cannot form a closed communication hole. First, more weak edge information is obtained on the basis of the gradient map based on the magnitude and direction of the gradient. The method is characterized in that the partial thought of a Canny edge detection operator is actually used for reference, a neighborhood region of a pixel point with strong response in an image is searched 5*5, the gradient amplitude and the gradient direction of the pixel point adjacent to the pixel point are calculated, and when the difference value between the gradient amplitude and the gradient direction of the adjacent pixel and the gradient amplitude and the gradient direction of the pixel point with strong response is smaller than a set threshold value, the adjacent pixel point is also set as the strong response point, wherein the gradient amplitude is calculated according to a formula (1.8). For the calculation of the gradient direction, although the Sobel operator template expanded to four directions is utilized when the magnitude is calculated, the direction of the gradient is calculated according to the formula (1.9) because the two directions of 45 degrees and 135 degrees can be combined into the horizontal and vertical directions after the trigonometric function transformation:
the specific implementation mode is as follows:
(a) Searching the molten pool image from the pixel point at the upper left corner, setting the pixel point as a strong response pixel in the calculated gradient amplitude diagram, and traversing pixels in a 5*5-sized neighborhood region after searching the strong response pixel;
(b) For the pixel points in the neighborhood, calculating the amplitude P and the direction theta of the gradient according to the formula (1.8) and the formula (1.9):
(c) Judging whether the pixel points in the neighborhood are set as strong response pixel points according to the (1.10), and simultaneously meeting two conditions:
for the pixel points in the neighborhood meeting the formula (1.10), if the pixel points do not respond as the strong response pixel points, setting the pixel points as the strong response pixel points; if the pixel is a strong response pixel, no other processing is performed on the pixel. The effect after the above-described gradient magnitude and direction-based edge connection operation is shown in fig. 10. It can be seen from fig. 10 that the gradient amplitude response of the weak edge region at the rear end of the molten pool becomes denser from sparse after the edge connection treatment, and this operation effectively increases the connectivity probability of the broken edge region. The edge connection operation is based on quantitative comparison operation of the correlation between the weak response pixel points and the strong response pixel points, so that the edge connection operation has high reliability and does not introduce interference influence. In fig. 10, the original gradient amplitude chart and the edge connection effect chart are sequentially shown from left to right.
After a large-scale connected hole area which is matched with the molten pool area as much as possible is obtained, the image processing operation based on mathematical morphology is also needed to obtain the required molten pool profile. The specific operation flow is as follows:
(a) For the interference edges introduced by Canny splicing in the molten pool area, removing the small connected domain with the connected domain dimension smaller than the set threshold by utilizing a method for judging the connected domain dimension, wherein the effect is shown in fig. 11, and fig. 11 is a graph of an original image and an interference edge removal effect based on the connected domain dimension in sequence from left to right
(b) The closed hole area in the filling image comprises a rear end formed area, the image after filling is subtracted from the image before filling, a gradient binary image of the closed hole area is reserved, and then a connected area with the largest dimension corresponding to the molten pool area is reserved, the effect is shown as figure 12, wherein a first image from left to right in figure 12 is a comparison original image, a second image is a filling effect image, a third image is a difference effect image of the second image and the first image, and a fourth image is an effect image reserved for the connected area with the largest dimension.
(c) After the large-scale connected domain approximately matched with the molten pool area is obtained, in order to enable the connected domain to have the characteristic that the edge of the molten pool area is smooth, image processing operation based on mathematical morphology is carried out on the connected domain. The specific operation is to establish a disc operator structure with the radius of 15 pixels, and the operator structure is used for firstly performing a closing operation and then performing an opening operation on the connected domain, so that the edge of the connected domain is smoother and fuller, the effect is shown in fig. 13, wherein fig. 13 is a graph of comparing an original graph with the smooth effect of the connected domain based on mathematical morphology operation in sequence from left to right.
(d) The contour edge of the molten pool area is further obtained based on the segmentation result obtained in the step (c), and the effect of overlaying the contour edge of the molten pool area on the original image of the molten pool image is shown in fig. 14, wherein fig. 14 is a connected domain binary image, a connected domain contour edge image and an effect image of the connected domain contour overlaying the molten pool image from left to right.
It can be seen from fig. 14 that the algorithm can achieve a better effect on the TIG welding stainless steel molten pool image. In order to verify the segmentation accuracy of the algorithm, the algorithm is utilized to verify TIG welded stainless steel images under different welding process parameters, and the verification result is compared with the detection result of the traditional edge detection algorithm. The effect pairs of the three traditional edge detection algorithms and the contour extraction algorithm are shown in fig. 16, wherein fig. 16 is from left to right, the first column is an Otsu thresholding contour extraction effect diagram, the second column is a C0V active contour extraction effect diagram, the third column is a Canny edge detection operator contour extraction effect diagram, and the fourth column is an OTM-EDG algorithm contour extraction effect diagram. It can be seen that the results of the first and second columns from left to right in fig. 16 are severely affected by the gray distribution of the image, and the threshold-based image segmentation is generally suitable for the situation that the target area to be extracted in the scene has a strong contrast in gray with the background, and in the active contour model, gray information is also an important part of the model convergence criterion. The gray value difference between the molten pool area and the background area in the TIG welding stainless steel molten pool image is not large, and the welding arc light has serious influence on the gray distribution of the image, so that both algorithms can be found to be unsuitable for being applied to the molten pool profile extraction of the TIG welding stainless steel molten pool image in theory or from the result. The third column of fig. 16 shows the result of directly using the Canny edge detection operator to extract the contour of the molten pool image, and it can be seen that the Canny operator can detect accurate edges for the region with larger gradient values, but at the same time, the Canny operator also detects the bright saturated region of the molten pool surface and the gradient value of the formed weld joint region at the rear end of the molten pool as interference edges because the saturated region of the molten pool surface and the gradient value of the formed weld joint region at the rear end of the molten pool are also large. In addition, it can be seen that the Canny operator has a large limitation in detecting weak edges, and it is difficult to obtain a complete continuous contour edge which coincides with the molten pool area. In summary, the reason why the conventional edge detection algorithm cannot achieve a good effect in the TIG welding stainless steel molten pool image is classified into two points: 1. uneven image gray level distribution caused by strong welding arc light; 2. texture features on the surface of the stainless steel melt pool are not conducive to the application of conventional edge detection algorithms. The molten pool profile extraction algorithm provided by the patent combines the accurate positioning advantage of the edge detection operator template and the complete and continuous edge advantage of processing based on mathematical morphology, and as can be seen from fig. 16, compared with the algorithm of the first column and the second column in fig. 16, the molten pool profile extraction algorithm provided by the patent can more accurately position the strong edge and the weak edge of the molten pool area without being influenced by the image gray level distribution condition; compared with the algorithm in the third row in fig. 16, the algorithm can obtain a closed complete contour edge on the premise of precisely positioning the weak edge, has no interference of other redundant edge lines, and facilitates the extraction and calculation of the morphological parameters of the subsequent molten pool. In order to verify the robustness of the algorithm provided by the patent, continuous 5 frames of images in the images acquired under the four groups of different welding parameter processes in fig. 16 are randomly extracted for testing, so that the accuracy of the algorithm provided by the patent in the contour extraction of continuous frame molten pool images under fixed process parameters can be obtained, and the algorithm can be suitable for industrial welding production environments.
The invention utilizes a camera to shoot and extract a molten pool image generated on a welding workpiece by a butt welder, obtains a complete and clear molten pool image contour line by preprocessing the molten pool image, calculating a gradient map based on a template of a convenient guiding operator, merging images, extracting contour lines of the molten pool image and other image processing methods, and solves the problem that the contour lines of the molten pool are difficult to extract due to uneven brightness distribution of the front end and the rear end of the molten pool area and the influence of arc light on the front end of the molten pool; the problems that the brightness of the area at the rear end of the molten pool is dark and the distribution is uneven, and the gray level change is not obvious and the gradient is weak after the image is converted into a gray level image at the two end parts of the rear end of the molten pool are solved; the extracted molten pool profile has high accuracy, can obtain a closed and complete molten pool profile, meets the requirements of all aspects on the detection accuracy of the molten pool profile, and effectively plays an indicating role in the forming of the later welding seam and the improvement of the welding quality.
With the above-described preferred embodiments according to the present invention as an illustration, the above-described descriptions can be used by persons skilled in the relevant art to make various changes and modifications without departing from the scope of the technical idea of the present invention. The technical scope of the present invention is not limited to the description, but must be determined according to the scope of claims.
Claims (5)
1. A molten pool contour image extraction method for realizing closed connected domain is characterized in that: the method comprises the following steps:
step one: extracting a molten pool image: carrying out welding work on a welding workpiece by using a welding machine provided with a camera, so that the camera can acquire molten pool images of each welding line from forming to welding;
step two: preprocessing and calculating a molten pool image: the method for preprocessing and calculating the molten pool image by introducing a weak edge enhancement method based on nonlinear gray level transformation and a gray level transformation method based on sliding window gray level information comprises the following steps:
(2-1) a weak edge enhancement method based on nonlinear gray level transformation, introducing a composite power function represented by formula (1.1),
；
wherein x represents the gray value of a single pixel point in the image, m and k are integers larger than 1, and are adjustable parameters, and the distribution condition of the gray value after transformation is different along with the change of m and k;
(2-2) a gray level conversion method based on sliding window gray level information, the calculation flow is as follows:
(2-2 a) traversing from the pixel point of one corner of the molten pool image, setting the gray value at the pixel point of the current traversing asThe maximum gray level in the neighborhood region of 3*3 centered on the pixel is set to +.>The minimum value is set to->；
(2-2 b) normalizing the gray values of the pixels in the 3*3 neighborhood in the region in the formula (1.2),
；
(2-2 c) the gray value of the pixel point in the step (2-2 b)According to its position in [0,1]]The interval in the gray scale range of (1.1) is substituted to obtain the gray value of the pixel point calculated by the nonlinear gray scale transformation algorithm +.>，
(2-2 d) toSubstituting the left side of the equal sign in the formula (1.2), and reversely pushing the +.>I.e. the gray value in the neighborhood region ranges from 0,1]Stretching to original gray value range +.>For example @1.3），
；
(2-2 e) adjusting the gray level distribution of the image based on the calculation mode of traversing all the pixel points in the image by the sliding window, suppressing the adverse effect generated by the brightness saturation region, and firstly, traversing the pixel pointsCalculating gray average value +.of pixel point in 3*3 neighborhood centered on it according to (1.4)>，/>I.e. pixel +.>Is used for the gray-scale value of (c),
；
after the neighborhood gray average value is obtained, the gray values of all pixel points in the 3*3 neighborhood are calculated and transformed based on an inverse proportion function, as shown in the formula (1.5),
；
wherein k1 and k2 represent the variation values of pixel points in 3*3 neighborhood, M is a constant, and the value is 90;
step three: based on gradient map calculation facilitating guide operator template matching: expanding Sobel operators in the initial horizontal direction and the initial vertical direction to four directions, and carrying out convolution operation on the preprocessed image by using the Sobel operators to obtain a gradient map of a molten pool image;
step four: image merging:thresholding the gradient map to obtain a gradient binary map, and splicing the rear end part of the gradient binary map with the front end part of the contour line obtained by the detection of the Canny operator to obtain a rough extraction result map of the edge of the molten pool image, wherein: traversing the original pixels of the gradient map in the third step, and traversing the traversed pixelsAnd determining the gradient amplitude ++of the corresponding pixel coordinate after the convolution operation>And +.>Maximum value of +.>And minimum value of，
；
In the middle ofFor two gradient magnitudes in horizontal and vertical directions, ω and σ are expansion direction angles, so that the gradient response of the pixel point in the remaining two molten pool gradient maps is obtained as +.>And calculating the pixel point according to the formula (1.8)Based on the gradient value P of the edge guiding operator, the gradient value P is two constants
；
In the method, in the process of the invention,M 1 andm 1 is a function of two constants, namely,M 1 the value is set to 3,m 1 a value of 1/3;
step five: extracting contour lines of molten pool images: performing edge connection operation based on judging gradient strength and direction on the edge rough extraction result graph, obtaining weak edge information of an image, filling to obtain a complete connected domain area, performing edge smoothing operation based on mathematical morphology on the connected domain area, and finally obtaining a molten pool image contour line, wherein the method comprises the following steps of:
(5-1) weak edge connection with angle based on gradient strength, the procedure is as follows:
(5-1 a) starting to search for strong response pixel points at the edge corner pixel points of the gradient map of the weak edge at the rear end of the molten pool and searching for the neighborhood region of the strong response pixel points 5*5, and transforming the expanded omega and sigma directions into horizontal and vertical directions through trigonometric functions, wherein the formula is shown as formula (1.9)
；
(5-1 b) calculating the magnitude of the gradient of the pixel points in the neighborhood according to the formulas (1.8) and (1.9)PWith the direction theta of the direction theta,
(5-1 c) judging whether the pixel points in the neighborhood are set as the pixel points with strong response, as shown in the formula (1.10),
；
if the pixel points in the neighborhood satisfying the formula (1.10) do not respond to the pixel points as strong response pixel points, setting the pixel points as strong response pixel points, wherein the pixel points are the strong response pixel points, and performing no additional processing;
(5-2) post-processing based on mathematical morphology, the processing flow being as follows:
(5-2 a) for removing small connected domain whose connected domain scale is smaller than the set threshold value by utilizing method of judging connected domain scale for the interference edge introduced by Canny splice in molten pool region,
(5-2 b) filling a closed hole area in the image, subtracting the filled image from the image before filling to obtain a gradient binary image of the closed hole area, and reserving a connected domain with the largest area scale;
(5-2 c) in order to make the connected domain have the characteristic of smoother edge of the molten pool area, establishing a disc operator structure with a radius of 15 pixels, and performing a closing operation and then an opening operation on the connected domain by using the disc operator structure;
and (5-2 d) superposing the obtained contour edge of the molten pool area on the original image of the molten pool image to obtain a contour line of the molten pool image.
2. The molten pool profile image extraction method for realizing closed connected domain according to claim 1, wherein: and in the third step, the Sobel operator in only two directions of horizontal and vertical is expanded to four directions, a first-order differential operator in two directions of omega=45 DEG and sigma=135 DEG is added, and convolution operation is carried out on the molten pool image, so that a molten pool gradient map in four directions is obtained.
3. The molten pool profile image extraction method for realizing closed connected domain according to claim 1, wherein: in the second step, m=2 and k=2 are substituted into formula (1.1), gray information conditions of a certain area in the molten pool image before and after transformation are obtained, gradient amplitude of a weak edge area at the rear end of the molten pool is increased, and noise is prevented from being introduced.
4. The molten pool profile image extraction method for realizing closed connected domain according to claim 1, wherein: in the second step, m=90 is substituted into (1.5), the weak edge at the rear end of the molten pool image is enhanced, and the brightness saturation region generated by the welding arc at the front end of the molten pool is restrained.
5. The molten pool profile image extraction method for realizing a closed connected domain according to claim 1 or 2, characterized in that: the template after the Sobel operator expansion is as follows:
；
the (a), (b), (c) and (d) are templates of the Sobel operator in the horizontal direction, the 45-degree direction, the vertical direction and the 135-degree direction in sequence.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010776108.5A CN111696107B (en) | 2020-08-05 | 2020-08-05 | Molten pool contour image extraction method for realizing closed connected domain |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010776108.5A CN111696107B (en) | 2020-08-05 | 2020-08-05 | Molten pool contour image extraction method for realizing closed connected domain |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111696107A CN111696107A (en) | 2020-09-22 |
| CN111696107B true CN111696107B (en) | 2024-01-23 |
Family
ID=72486379
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010776108.5A Active CN111696107B (en) | 2020-08-05 | 2020-08-05 | Molten pool contour image extraction method for realizing closed connected domain |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111696107B (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112033313B (en) * | 2020-10-10 | 2022-01-11 | 成都瑞拓科技有限责任公司 | Eccentricity detection method for wet blasting bead |
| CN113077423B (en) * | 2021-03-22 | 2023-06-09 | 中国人民解放军空军工程大学 | Laser selective melting pool image analysis system based on convolutional neural network |
| CN113674206B (en) * | 2021-07-21 | 2022-10-25 | 华南理工大学 | Extraction method suitable for characteristic parameters of deep-melting K-TIG welding molten pool and keyhole entrance |
| CN114187289B (en) * | 2021-12-23 | 2022-08-09 | 武汉市坤瑞塑胶模具制品有限公司 | Plastic product shrinkage pit detection method and system based on computer vision |
| CN114170228B (en) * | 2022-02-14 | 2022-04-19 | 西南石油大学 | Computer image edge detection method |
| CN114523236A (en) * | 2022-02-28 | 2022-05-24 | 柳州铁道职业技术学院 | Intelligent automatic detection platform based on machine vision |
| CN114331923B (en) * | 2022-03-11 | 2022-05-13 | 中国空气动力研究与发展中心低速空气动力研究所 | Improved Canny algorithm-based bubble profile extraction method in ice structure |
| CN114820674B (en) * | 2022-05-17 | 2024-04-05 | 中国南方电网有限责任公司超高压输电公司广州局 | Arc profile extraction method, device, computer equipment and storage medium |
| CN114841999B (en) * | 2022-07-01 | 2022-10-11 | 湖南科天健光电技术有限公司 | Method and system for adjusting monitoring image of welding area |
| CN115063422B (en) * | 2022-08-18 | 2022-11-08 | 建首(山东)钢材加工有限公司 | Intelligent detection method for container welding quality |
| CN116433700B (en) * | 2023-06-13 | 2023-08-18 | 山东金润源法兰机械有限公司 | Visual positioning method for flange part contour |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2015196616A1 (en) * | 2014-06-23 | 2015-12-30 | 京东方科技集团股份有限公司 | Image edge detection method, image target recognition method and device |
| CN108320280A (en) * | 2018-01-16 | 2018-07-24 | 南京理工大学 | The crater image method for detecting abnormality of view-based access control model clarity and contours extract |
| CN109308705A (en) * | 2018-09-27 | 2019-02-05 | 上海交通大学 | A kind of weld pool image profile real time extracting method |
-
2020
- 2020-08-05 CN CN202010776108.5A patent/CN111696107B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2015196616A1 (en) * | 2014-06-23 | 2015-12-30 | 京东方科技集团股份有限公司 | Image edge detection method, image target recognition method and device |
| CN108320280A (en) * | 2018-01-16 | 2018-07-24 | 南京理工大学 | The crater image method for detecting abnormality of view-based access control model clarity and contours extract |
| CN109308705A (en) * | 2018-09-27 | 2019-02-05 | 上海交通大学 | A kind of weld pool image profile real time extracting method |
Non-Patent Citations (1)
| Title |
|---|
| jing han等.Weld Reinforcement Analysis Based on Long-Term Prediction of Molten Pool Image in Additive Manufacturing.IEEE Access.2020,第66908-66918页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111696107A (en) | 2020-09-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111696107B (en) | Molten pool contour image extraction method for realizing closed connected domain | |
| CN105913415B (en) | A kind of image sub-pixel edge extracting method with extensive adaptability | |
| CN114757949B (en) | Wire and cable defect detection method and system based on computer vision | |
| CN107808378B (en) | Method for detecting potential defects of complex-structure casting based on vertical longitudinal and transverse line profile features | |
| CN110033447B (en) | High-speed rail heavy rail surface defect detection method based on point cloud method | |
| US11580647B1 (en) | Global and local binary pattern image crack segmentation method based on robot vision | |
| He et al. | Weld seam profile detection and feature point extraction for multi-pass route planning based on visual attention model | |
| CN107610111B (en) | deep learning-based welding spot image detection method | |
| CN113362326A (en) | Method and device for detecting welding spot defects of battery | |
| CN111127402A (en) | Visual detection method for welding quality of robot | |
| CN109472788B (en) | Method for detecting flaw on surface of airplane rivet | |
| CN112053376B (en) | Workpiece weld joint identification method based on depth information | |
| CN112001906B (en) | Steel plate weld image detection method combined with non-maximum suppression | |
| CN110717872A (en) | Method and system for extracting characteristic points of V-shaped welding seam image under laser-assisted positioning | |
| CN114118144A (en) | Anti-interference accurate aerial remote sensing image shadow detection method | |
| CN109829858B (en) | Ship-borne radar image oil spill monitoring method based on local adaptive threshold | |
| CN117095004B (en) | Excavator walking frame main body welding deformation detection method based on computer vision | |
| CN112862794A (en) | Fillet weld appearance detection method based on structured light | |
| CN111667470A (en) | Industrial pipeline flaw detection inner wall detection method based on digital image | |
| CN116660286A (en) | Wire harness head peeling measurement and defect detection method and system based on image segmentation | |
| CN108492306A (en) | A kind of X-type Angular Point Extracting Method based on image outline | |
| CN114332081A (en) | Textile surface abnormity determination method based on image processing | |
| CN116630321B (en) | Intelligent bridge health monitoring system based on artificial intelligence | |
| CN111784722A (en) | Improved Canny lane line edge detection algorithm | |
| CN109741311B (en) | Aluminum alloy fusion welding back face fusion width detection method with false edge |
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 |