CN101294945A - White edge detecting method for hot galvanizing alloying plate - Google Patents

Abstract

The invention relates to a digital image process technology and a method used for detecting haloing of a hot dipped galvanized alloying plate; the digital image process technology and the method are characterized in that two horizontal justification cameras aim at two edges of left and right of a band steel respectively and carry out shooting for the band steel; two image collecting cards are respectively connected with the two cameras by cables; an industrial control computer collects the digital image by the two image collecting cards, carries out processing for the image and calculates the haloing parameter, the calculated haloing parameter is compared with the ideal haloing parameter and the current power setting value is regulated properly; the regulated power setting value is transmitted to a programmable logic controller (PLC) and the power value of a power heater is reset by the programmable logic controller (PLC); the digital image process technology and the method can accurately measure the occurrence of the haloing, response to the change of the haloing in time and shorten the transitional length, thus reducing the quantity of waste and ungraded products and improving the yield and the quality of the products.

Description

A kind of white edge detection method that is used for the hot-dip galvanized alloy plate

Technical field

The present invention relates to a kind of digital image processing techniques, relate in particular to a kind of white edge detection method that is used for the hot-dip galvanized alloy plate.

Background technology

The white edge situation of hot-dip galvanized alloy plate is an important indicator of reflect heat galvanized alloy plate alloying effect.Operating personnel carry out the adjustment of alloying power usually according to the situation of white edge, to reach the purpose of little white edge control.Under the situation, all be in the past, and rule of thumb carried out the adjustment of alloying power, thereby reach the purpose of little white edge control by the width of operating personnel with the visual inspection white edge.But, cause the quality of product also to be not quite similar because each operating personnel's experience difference there are differences between the individuality.Again because the setting and the adjusting of alloying power rested in the manual control, control lag, transit time is long, has influenced the raising of machine set product lumber recovery and product quality.Simultaneously can not review the white edge situation of having produced volume.

Summary of the invention

The object of the present invention is to provide a kind of white edge detection method that is used for the hot-dip galvanized alloy plate, this detection method can be measured the white edge image exactly, through dialogue limit treatment of picture, calculates the white edge parameter, thereby timely response is made in the variation of white edge parameter.

The present invention is achieved in that a kind of white edge detection method that is used for the hot-dip galvanized alloy plate, be the camera of two horizontal alignments aim at respectively the band steel about two edges the band steel is taken, two image pick-up cards are connected with two cameras respectively by cable; Industrial computer is handled image after collecting digital picture by image collection card, calculates the white edge parameter, and the white edge parameter that calculates and desirable white edge parameter are compared, and suitably regulates the current power setting value; Send the set value of the power of regulating to degree controller PLC able to programme, reset the performance number of power heater by programmable controller PLC;

Image processing process after white edge detects is: carry out medium filtering earlier; Use for reference histogram method, adopt the method for iteration to determine threshold value, carry out threshold transformation to cut apart background; Adopt the white edge of Robort edge detection operator detected image again; Determine the white edge position; Last calculation of parameter white edge parameter according to camera and image pick-up card; Concrete steps are as follows:

The histogram H (D) of the first step, computed image

$H\left(D\right)=\underset{\mathrm{\ΔD}\→0}{\lim }\frac{A\left(D\right)-A(D+\mathrm{\ΔD})}{\mathrm{\ΔD}}=-\frac{d}{\mathrm{dD}}A\left(D\right)---\left(1\right)$

In the formula (1), D represents gray level, and gray-scale value is greater than the area of D in A (D) presentation video; Second step, to image f (x y) adopts the pre-service of 3x3 medium filtering,

$\left[\begin{array}{ccc}D1& D2& D3\\ D4& D5& D6\\ D7& D8& D9\end{array}\right]\&DoubleRightArrow;\mathrm{<figure-callout\; id="5"\; label="D"\; filenames="A20071004010500202.png"\; state="\{\{state\}\}">D</figure-callout>}5=\frac{\underset{i=1}{\overset{9}{\mathrm{\&Sigma;}}}\mathrm{Di}}{9}---\left(2\right)$

By formula (2), promptly in the image arbitrarily the gray-scale value of any replace by the average of this point with adjacent 8 points;

The 3rd step, use for reference histogrammic result, adopt the method for iteration to obtain best threshold value, with image binaryzation to cut apart background; Algorithm is as follows:

Utilize the result of formula (1), order:

Tmax＝Max{H(D _i)|H(D _i)＞0}

Tmin＝Min{H(D _i)|H(D _i)＞0}

1) obtain minimum in the image and maximum gray-scale value Tmax and Tmin, make initial threshold be:

${\mathrm{<figure-callout\; id="0"\; label="T"\; filenames="A20071004010500131.png,A20071004010500171.png"\; state="\{\{state\}\}">T</figure-callout>}}_0=\frac{T_{\max }+T_{\min }}{2}---\left(3\right)$

2) according to threshold value Tk (k is an iterations, is initially 0) image segmentation is become target and background two parts, obtains two-part average gray value Zf (prospect gray scale) and Zb (background gray scale):

$\mathrm{Zf}=\frac{\underset{z(i,j)\&le;T_k}{\mathrm{\&Sigma;}}Z(x,y)\&times;N(i,j)}{\underset{z(i,j)\&le;T_k}{\mathrm{\&Sigma;}}N(i,j)}---\left(4\right)$

$\mathrm{Zb}=\frac{\underset{z(i,j)\&GreaterEqual;T_k}{\mathrm{\&Sigma;}}Z(x,y)\&times;N(i,j)}{\underset{z(i,j)\&GreaterEqual;T_k}{\mathrm{\&Sigma;}}N(i,j)}---\left(5\right)$

(x y) is that ((I j) is that (I, weight coefficient j) equal the number of the pixel of same gray-scale value here to N for I, gray-scale value j) on the image to Z in the formula;

3) obtain new threshold value:

$T_{k+1}=\frac{Z_f+Z_b}{2}---\left(6\right)$

4) if Tk+1==Tk, perhaps iterations is greater than the number of times of regulation, EOP (end of program) then, otherwise K ← K+1 are changeed step 2);

5) with Tk be the threshold values changing image, outside in write down the position POS of first pixel then line by line greater than 0 point _dAs the cut-point position;

The 4th step, employing Roberts edge detection operator are that template and image are made process of convolution, with the limit portion feature of outstanding band steel, i.e. white edge;

The Roberts edge detection operator is provided by following formula:

$g(x,y)={\{{[\sqrt{f(x,y)}-\sqrt{f(x+1,y+1)}]}^2+{[\sqrt{f(x,y)}-\sqrt{f(x+1,y+1)}]}^2\}}^{1/2}---\left(7\right)$

Wherein (x y) is the input picture with integer pixel coordinate to f;

The 5th step, determine the white edge position, utilize the 4th step institute result calculated, calculate adjacent 10 pixel grey scale rate of change and, rate of gray level and maximal value promptly can be used as the separatrix of white edge and normal belt steel surface; Obtain white edge position pos by following formula:

Pos=n+5; Wherein $n=\min \left\{n\right|\left(\underset{i=n}{\overset{n+10}{\mathrm{\&Sigma;}}}\mathrm{PixelValue}\right[i\left]\right)<N\}---\left(8\right)$

If $\min \left\{n\right|\left(\underset{i=n}{\overset{n+10}{\mathrm{\&Sigma;}}}\mathrm{PixelValue}\right[i\left]\right)<N\}$ There is not the then current white edge that do not exist;

Pos is the white edge amount of being asked in the formula, and n is the point of grey scale change maximum, and N is an empirical value;

The 6th step, acquisition white edge parameter according to the ratio of single pixel value and physical length, in conjunction with the white edge position of calculating above, just can be calculated the width of white edge, and the ratio K of single pixel value and physical length obtains by following formula:

K=developed width * magnification/picture traverse (unit: pixel) (9)

White edge width w=(POS _d-pos) * K (10)

Described empirical value N is 2370.

The present invention is in band steel operational process, with the camera of two horizontal alignments aim at respectively the band steel about two edge collecting images, by image pick-up card image information is sent to industrial computer, obtain the white edge parameter after by industrial computer digital picture being handled again, export to programmable controller, the programmable controller controls external unit is realized the correction to band steel white edge.

The present invention can measure the occurrence of white edge automatically, exactly, and timely response is made in the variation of white edge parameter, shortens transition length, thereby reduces the waster amount, improves product lumber recovery and product quality.

Description of drawings

Fig. 1 is a white edge Flame Image Process process flow diagram of the present invention;

Fig. 2 is a white edge pick-up unit synoptic diagram of the present invention;

Fig. 3 is the white edge synoptic diagram;

Fig. 4 is the synoptic diagram of white edge after binaryzation;

Fig. 5 is the synoptic diagram of white edge through edge featureization;

Fig. 6 is a strip edge portion original image;

Fig. 7 is band steel original image horizontal direction pixel value curve;

Fig. 8 is band steel original image histogram;

Fig. 9 is for being with the image after the former limit of steel portion is partitioned into background.

Figure 10 is the image of strip edge portion after process of convolution and color inversion;

Figure 11 is Figure 10 horizontal direction pixel value curve;

Figure 12 is the data of Figure 11 curve correspondence;

Figure 13 is white edge position and a background position data curve among Figure 10;

Figure 14 is a white edge width value curve among Figure 10;

Figure 15 is the data of Figure 13, Figure 14 curve correspondence.

Among the figure: 1 concentric cable, 2 telecommunication cables, 31 industrial computers, 32 former industrial computers, 4 preceding induction heaters, 5 back induction heaters, 6DP coupling mechanism, 7 right cameras, 8 left cameras, 9PCI bus, 10 image pick-up cards, 11 band steel, 12 S5PLC, 13 S7PLC.

Embodiment

The invention will be further described below in conjunction with the drawings and specific embodiments.

Referring to Fig. 2, a kind of white edge detection system that is used for the hot-dip galvanized alloy plate, comprise the left and right camera 7,8, industrial computer 31, the capture card 10 that are positioned at the strip edge edge, left and right camera 7,8 connects capture card 10 through concentric cable 1, capture card 10 is through PIC bus 9 input industrial computers 31, industrial computer 31 meets S7PLC through telecommunication cable 2, and S7PLC meets S5PLC through DP coupling mechanism 6, S5PLC power controlling well heater 4,5.

Use the performance number of setting induction heater manually as need, can switch the control mode of S5PLC in original system industrial computer 32, the signal that shielding S7PLC sends here switches to original system.

A kind of white edge detection method that is used for the hot-dip galvanized alloy plate, be the camera 7,8 of two horizontal alignments aim at respectively band steel 11 about two edges the band steel is taken, two image pick-up cards 10 are connected with two cameras 7,8 respectively by cable 1; Industrial computer 31 is handled image after collecting digital picture by image collection card 10, calculates the white edge parameter, and the white edge parameter that calculates and desirable white edge parameter are compared, and suitably regulates the current power setting value; The set value of the power of regulating is write S7PLC by telecommunication cable 2; S7PLC delivers to S5PLC by DP coupling mechanism 6 with set value of the power again, is reset the performance number of power heater by S5PLC; Above process circulation is carried out, thereby control white edge parameter is near desired quantity.

Image processing process after white edge detects is: carry out medium filtering earlier; Use for reference histogram method, adopt the method for iteration to determine threshold value, carry out threshold transformation to cut apart background; Adopt the white edge of Robort edge detection operator detected image again; Determine the white edge position; Calculate the white edge width with reference to cam lens parameter and image pick-up card parameter at last; Referring to Fig. 1.Concrete steps are as follows:

1, the histogram H (D) of computed image

$H\left(D\right)=\underset{\mathrm{\&Delta;D}\&RightArrow;0}{\lim }\frac{A\left(D\right)-A(D+\mathrm{\&Delta;D})}{\mathrm{\&Delta;D}}=-\frac{d}{\mathrm{dD}}A\left(D\right)---\left(1\right)$

What histogram was described is the number that has the pixel of a certain gray-scale value in the image, and in the formula (1), D represents gray level, and gray-scale value is greater than the area of D in A (D) presentation video.The intensity profile situation is seen Fig. 8.

2, (x y) adopts the pre-service of 3x3 medium filtering, takes out noise when the characteristics of medium filtering are the protection image border to image f.Original white edge image is referring to Fig. 3.

$\left[\begin{array}{ccc}D1& D2& D3\\ D4& D5& D6\\ D7& D8& D9\end{array}\right]\&DoubleRightArrow;\mathrm{<figure-callout\; id="5"\; label="D"\; filenames="A20071004010500202.png"\; state="\{\{state\}\}">D</figure-callout>}5=\frac{\underset{i=1}{\overset{9}{\mathrm{\&Sigma;}}}\mathrm{Di}}{9}---\left(2\right)$

By formula (2), promptly in the image arbitrarily the gray-scale value of any replace by the average of this point with adjacent 8 points.

3, use for reference histogrammic result, adopt the method for iteration to obtain best threshold value, with image binaryzation to cut apart background; Algorithm is as follows:

Utilize the result of formula (1), order:

Tmax＝Max{H(D _i)|H(D _i)＞0}

Tmin＝Min{H(D _i)|H(D _i)＞0}

1) obtain minimum in the image and maximum gray-scale value Tmax and Tmin, make initial threshold be:

${\mathrm{<figure-callout\; id="0"\; label="T"\; filenames="A20071004010500131.png,A20071004010500171.png"\; state="\{\{state\}\}">T</figure-callout>}}_0=\frac{T_{\max }+T_{\min }}{2}---\left(3\right)$

2) according to threshold value Tk (k is an iterations, is initially 0) image segmentation is become target and background two parts, obtains two-part average gray value Zf (prospect gray scale) and Zb (background gray scale):

$\mathrm{Zf}=\frac{\underset{z(i,j)\&le;T_k}{\mathrm{\&Sigma;}}Z(x,y)\&times;N(i,j)}{\underset{z(i,j)\&le;T_k}{\mathrm{\&Sigma;}}N(i,j)}---\left(4\right)$

$\mathrm{Zb}=\frac{\underset{z(i,j)\&GreaterEqual;T_k}{\mathrm{\&Sigma;}}Z(x,y)\&times;N(i,j)}{\underset{z(i,j)\&GreaterEqual;T_k}{\mathrm{\&Sigma;}}N(i,j)}---\left(5\right)$

(x y) is that ((I j) is that (I, weight coefficient j) equal the number of the pixel of same gray-scale value here to N for I, gray-scale value j) on the image to Z in the formula.

3) obtain new threshold value:

$T_{k+1}=\frac{Z_f+Z_b}{2}---\left(6\right)$

4) if Tk+1==Tk, perhaps iterations is greater than the number of times of regulation, EOP (end of program) then, otherwise K ← K+1 are changeed step 2).

5) optimal threshold of obtaining by above step, with image binaryzation, the result of binaryzation is referring to Fig. 4.(outside in) writes down the position POS of first pixel greater than 0 point then line by line _dAs the cut-point position.

4, adopting the Roberts edge detection operator is that template and image are made process of convolution, with the limit portion feature of outstanding band steel, i.e. white edge; Referring to Fig. 5.

The Roberts edge detection operator has treatment effect preferably to having precipitous low noise image, is a kind of operator that utilizes local difference operator to seek the edge, and it is provided by following formula:

$g(x,y)={\{{[\sqrt{f(x,y)}-\sqrt{f(x+1,y+1)}]}^2+{[\sqrt{f(x,y)}-\sqrt{f(x+1,y+1)}]}^2\}}^{1/2}---\left(7\right)$

Wherein (x y) is the input picture with integer pixel coordinate to f, and square root calculation makes this processing be similar to the process that takes place in the human visual system.

(normal galvanizing surface) is the process of a gradual change to the strip edge edge from white (white edge) to grey, through the processing in above four steps, can obviously see the zone in this gradual change, and the rate of gray level of pixel will be higher than other zone.

5, determine the white edge position, utilize the 4th step institute result calculated, calculate adjacent 10 pixel grey scale rate of change and, rate of gray level and maximal value (being meant the gray-scale value minimum here) promptly can be used as the separatrix of white edge and normal belt steel surface.Obtain white edge position (pos) by following formula

Pos=n+5; Wherein $n=\min \left\{n\right|\left(\underset{i=n}{\overset{n+10}{\mathrm{\&Sigma;}}}\mathrm{PixelValue}\right[i\left]\right)<2370\}---\left(8\right)$

If $\min \left\{n\right|\left(\underset{i=n}{\overset{n+10}{\mathrm{\&Sigma;}}}\mathrm{PixelValue}\right[i\left]\right)<2370\}$ There is not the then current white edge that do not exist.

Pos is the white edge amount of being asked in the formula, and n is the point (pixel value) of grey scale change maximum, and 2370 is empirical value.

6, obtain the white edge parameter, calculate the ratio K of single pixel value and physical length,, just can calculate the width of white edge in conjunction with the white edge position of calculating above.

K=developed width * magnification/picture traverse (unit: pixel) (9)

White edge width w=(POS _d-pos) * K (10)

Embodiment

1, obtains image (Fig. 6), obtain original image pixels value curve (Fig. 7)

2, the histogram H (D) of computed image

$H\left(D\right)=\underset{\mathrm{\&Delta;D}\&RightArrow;0}{\lim }\frac{A\left(D\right)-A(D+\mathrm{\&Delta;D})}{\mathrm{\&Delta;D}}=-\frac{d}{\mathrm{dD}}A\left(D\right)---\left(1\right)$

What histogram was described is the number that has the pixel of a certain gray-scale value in the image, and in the formula (1), D represents gray level, and gray-scale value is greater than the area of D in A (D) presentation video.The histogram of original image is Fig. 8.

3, (x y) adopts the pre-service of 3x3 medium filtering, takes out noise when the characteristics of medium filtering are the protection image border to image f.

4, use for reference histogrammic result, adopt the method for iteration to obtain best threshold value.In this example, try to achieve Tmax=253, Tmin=43, Tk=161 carries out binary conversion treatment with threshold value Tk=161 to image, is partitioned into band steel image (Fig. 9) from image (Fig. 6), obtains POS _d

5, adopting the Roberts edge detection operator is that template and image are made process of convolution, and carries out image inversion, with the limit portion feature of outstanding band steel, be white edge,, obtain the corresponding curve of this figure horizontal direction pixel referring to Figure 10, referring to Figure 11, and corresponding data, referring to Figure 12.

6, by formula:

K=developed width * magnification/picture traverse (unit: pixel) (9)

White edge width w=(POS _d-pos) * K (10)

Calculate the developed width of white edge, the camera collection developed width is 94 centimetres in this example, and the camera lens magnification is 0.11,640 pixels of images acquired width, single pixel fish physical length ratio K=0.01615625.Boundary position and white edge position pixel value curve are seen Figure 13 in this example, and actual white edge width curve is seen Figure 14, and Figure 15 is Figure 13 and Figure 14 value corresponding.

The present invention can measure the occurrence of white edge automatically, exactly, and timely response is made in the variation of white edge parameter, shortens transition length, thereby reduces the waster amount, improves product lumber recovery and product quality.

Claims (2)

1, a kind of white edge detection method that is used for the hot-dip galvanized alloy plate is characterized in that:

The camera of two horizontal alignments aim at respectively the band steel about two edges the band steel is taken, two image pick-up cards are connected with two cameras respectively by cable; Industrial computer is handled image after collecting digital picture by image collection card, calculates the white edge parameter, and the white edge parameter that calculates and desirable white edge parameter are compared, and suitably regulates the current power setting value; Send the set value of the power of regulating to degree controller PLC able to programme, reset the performance number of power heater by programmable controller PLC;

Image processing process after white edge detects is: carry out medium filtering earlier; Use for reference histogram method, adopt the method for iteration to determine threshold value, carry out threshold transformation; Adopt the white edge of Robort edge detection operator detected image again; Determine the white edge position; Obtain the white edge parameter at last; Concrete steps are as follows:

The histogram H (D) of the first step, computed image

$H\left(D\right)=\underset{\mathrm{\&Delta;D}\&RightArrow;0}{\lim }\frac{A\left(D\right)-A(D+\mathrm{\&Delta;D})}{\mathrm{\&Delta;D}}=-\frac{d}{\mathrm{dD}}A\left(D\right)---\left(1\right)$

In the formula (1), D represents gray level, and gray-scale value is greater than the area of D in A (D) presentation video;

Second step, to image f (x y) adopts 3 * 3 medium filtering pre-service,

$\left[\begin{array}{ccc}D1& D2& D3\\ D4& D5& D6\\ D7& D8& D9\end{array}\right]\&DoubleRightArrow;D5=\frac{\underset{i=1}{\overset{9}{\mathrm{\&Sigma;}}}\mathrm{Di}}{9}---\left(2\right)$

By formula (2), promptly in the image arbitrarily the gray-scale value of any replace by the average of this point with adjacent 8 points;

The 3rd goes on foot, uses for reference histogrammic result, adopts the method for iteration to obtain best threshold value, and to cut apart background, algorithm is as follows with image binaryzation:

Utilize the result of formula (1), order:

Tmax＝Max{H(D _i)|H(D _i)＞0}

Tmin＝Min{H(D _i)|H(D _i)＞0}

1) obtain minimum in the image and maximum gray-scale value Tmax and Tmin, make initial threshold be:

$T_0=\frac{T_{\max }+T_{\min }}{2}---\left(3\right)$

2) according to threshold value Tk (k is an iterations, is initially 0) image segmentation is become target and background two parts, obtains two-part average gray value Zf (prospect gray scale) and Zb (background gray scale):

$\mathrm{Zf}=\frac{\underset{z(i,j)\&le;T_k}{\mathrm{\&Sigma;}}Z(x,y)\&times;N(i,j)}{\underset{z(i,j)\&le;T_k}{\mathrm{\&Sigma;}}N(i,j)}---\left(4\right)$

$\mathrm{Zb}=\frac{\underset{z(i,j)\&le;T_k}{\mathrm{\&Sigma;}}Z(x,y)\&times;N(i,j)}{\underset{z(i,j)\&le;T_k}{\mathrm{\&Sigma;}}N(i,j)}---\left(5\right)$

(x y) is that ((I j) is that (I, weight coefficient j) equal the number of the pixel of same gray-scale value here to N for I, gray-scale value j) on the image to Z in the formula;

3) obtain new threshold value:

$T_{k+1}=\frac{Z_f+Z_b}{2}---\left(6\right)$

4) if Tk+1==Tk, perhaps iterations is greater than the number of times of regulation, EOP (end of program) then, otherwise K ← K+1 are changeed step 2);

5) with Tk be the threshold values changing image, outside in write down the position POS of first pixel then line by line greater than 0 point _dAs the cut-point position;

The 4th step, employing Roberts edge detection operator are that template and image are made process of convolution, with the limit portion feature of outstanding band steel, i.e. white edge;

The Roberts edge detection operator is provided by following formula:

$g(x,y)={\{{[\sqrt{f(x,y)}-\sqrt{f(x+1,y+1)}]}^2+{[\sqrt{f(x,y)}-\sqrt{f(x+1,y+1)}]}^2\}}^{1/2}---\left(7\right)$

Wherein (x y) is the input picture with integer pixel coordinate to f;

The 5th step, determine the white edge position, utilize the 4th step institute result calculated, calculate adjacent 10 pixel grey scale rate of change and, rate of gray level and maximal value promptly can be used as the separatrix of white edge and normal belt steel surface; Obtain white edge position pos by following formula:

Pos=n+5; Wherein $n=\min \left\{n\right|\left(\underset{i=n}{\overset{n+10}{\mathrm{\&Sigma;}}}\mathrm{PixelValue}\right[i\left]\right)<N\}---\left(8\right)$

If $\min \left\{n\right|\left(\underset{i=n}{\overset{n+10}{\mathrm{\&Sigma;}}}\mathrm{PixelValue}\right[i\left]\right)<N\}$ There is not the then current white edge that do not exist;

In the formula: pos is the white edge amount of being asked, and n is the point of grey scale change maximum, and N is an empirical value;

The 6th step, acquisition white edge parameter according to the ratio of single pixel value and physical length, in conjunction with the white edge position of calculating above, just can be calculated the width of white edge, and the ratio K of single pixel value and physical length obtains by following formula:

K=developed width * magnification/picture traverse (unit: pixel) (9)

White edge width w=(POS _d-pos) * K (10)

2, the white edge detection method that is used for the hot-dip galvanized alloy plate according to claim 1, it is characterized in that: empirical value N is 2370.

Images